This report contains the collective views of an international group of
    experts and does not necessarily represent the decisions or the stated
    policy of the United Nations Environment Programme, the International
    Labour Organisation, or the World Health Organization.

    Published under the joint sponsorship of
    the United Nations Environment Programme,
    the International Labour Organisation,
    and the World Health Organization

    World Health Orgnization
    Geneva, 1983

         The International Programme on Chemical Safety (IPCS) is a
    joint venture of the United Nations Environment Programme, the
    International Labour Organisation, and the World Health
    Organization. The main objective of the IPCS is to carry out and
    disseminate evaluations of the effects of chemicals on human health
    and the quality of the environment. Supporting activities include
    the development of epidemiological, experimental laboratory, and
    risk-assessment methods that could produce internationally
    comparable results, and the development of manpower in the field of
    toxicology. Other activities carried out by the IPCS include the
    development of know-how for coping with chemical accidents,
    coordination of laboratory testing and epidemiological studies, and
    promotion of research on the mechanisms of the biological action of

        ISBN 92 4 154087 7 

         The World Health Organization welcomes requests for permission
    to reproduce or translate its publications, in part or in full.
    Applications and enquiries should be addressed to the Office of
    Publications, World Health Organization, Geneva, Switzerland, which
    will be glad to provide the latest information on any changes made
    to the text, plans for new editions, and reprints and translations
    already available.

    (c) World Health Organization 1983

         Publications of the World Health Organization enjoy copyright
    protection in accordance with the provisions of Protocol 2 of the
    Universal Copyright Convention. All rights reserved.

         The designations employed and the presentation of the material
    in this publication do not imply the expression of any opinion
    whatsoever on the part of the Secretariat of the World Health
    Organization concerning the legal status of any country, territory,
    city or area or of its authorities, or concerning the delimitation
    of its frontiers or boundaries.

         The mention of specific companies or of certain manufacturers'
    products does not imply that they are endorsed or recommended by the
    World Health Organization in preference to others of a similar
    nature that are not mentioned. Errors and omissions excepted, the
    names of proprietary products are distinguished by initial capital




   1.1. Interrelationships with toxicological studies
   1.2. Design
   1.3. Environmental agents and assessment of exposures
   1.4. Effects on health
   1.5. Organization and conduct
   1.6. Analysis and interpretation of results
   1.7. Uses of epidemiological information



   2.1. Introduction
   2.2. Preliminary review of state of knowledge
   2.3. Descriptive studies and use of existing records
        2.3.1. Mortality statistics
        2.3.2. Morbidity statistics
        2.3.3. Populations at risk
        2.3.4. Geographical differences in mortality and morbidity
        2.3.5. Time trends
        2.3.6. Associations with environmental indices
        2.3.7. Case registers
        2.3.8. General surveys
   2.4. Formulation of hypotheses
   2.5. Cross-sectional studies
   2.6. Prospective and follow-up studies
   2.7. Retrospective cohort studies
   2.8. Time-series studies
   2.9. Case-control studies
   2.10. Controlled exposure studies
   2.11. Monitoring and surveillance



   3.1. Introduction
   3.2. Exposure and dose
        3.2.1. Systematic agents
        3.2.2. Local exposure
        3.2.3. Physical factors
   3.3. Combined exposure, physical and chemical
        3.3.1. Same agent, various sources
        3.3.2. Various agents, same source
        3.3.3. Various agents, various sources
        3.3.4. Impurities
        3.3.5. Interactions
   3.4. Qualitative assessment of exposure
   3.5. Environmental assessment of exposure
        3.5.1. Quality of data
        3.5.2. Monitoring strategy for air pollutants
      What to sample, how long, how frequently?
        3.5.3. Monitoring of pollutants in food and water
      Overall assessment of dietary
                       intake of toxic elements
      Indirect assessment of intake
      Direct assessment of intake
        3.5.4. Monitoring of physical factors
      Ionizing radiation
      Non-ionizing radiation
   3.6. Personal sampling
   3.7. Biological assessment of exposure
        3.7.1. Advantages, disadvantages, limitations
        3.7.2. Collection for future reference
        3.7.3. Index specimens for various pollutants
        3.7.4. Example of environmental versus biological
              assessment of exposure: inorganic lead
      Lead in blood (Pb-B)
      Lead in urine (Pb-U)
      Lead in faeces (Pb-F)
      Lead in deciduous teeth (Pb-T)
   3.8. Assessment of the subjective environment
        3.8.1. Assessment of odour
        3.8.2. Assessment of taste
        3.8.3. Example of sensory assessment of drinking-water
   3.9. Interindividual and intergroup variability in
        exposure: population at risk
   3.10. Outdoor/indoor exposure
   3.11. Time-weighted exposure



   4.1. Introduction
        4.1.1. General comments on effects
        4.1.2. General comments on measurements of effects
       Inter- and intrainstrument variation
       Inter- and intralaboratory differences
       Inter- and intraobserver variations
   4.2. Mortality and morbidity statistics
        4.2.1. Mortality statistics
        4.2.2. Routine morbidity statistics
   4.3. Cancer
        4.3.1. Cancer and environmental factors
        4.3.2. Measurements of cancer
       Incidence and mortality rate
       Variations of incidence with age
       Geographical differences
       Cancer and life-style
       Cancer in migrants
       Time trends
       Correlation studies
       Hospital data
       Cancer and occupation
      Case reports
      Epidemiological uses of pathological findings
   4.4. Respiratory and cardiovascular effects
        4.4.1. Symptom questionnaires
        4.4.2. Tests of system function
        4.4.3. Standardization of methods
        4.4.4. Radiographic measurements
        4.4.5. Hypersensitivity measurements
        4.4.6. Example: Effects of manganese on
               respiratory and cardiovascular systems
   4.5. Effects on nervous system and organs of sense
        4.5.1. Central and peripheral nervous systems
        4.5.2. Ear: Effects of sound
        4.5.3. Eye and vision
   4.6. Behavioural effects
        4.6.1. Effects of environmental exposure
        4.6.2. Indicators and measurements of effects
        4.6.3. Interpretation of data
   4.7. Haemopoietic effects
        4.7.1. Environmental agents inducing direct toxic
               effects in the haematological system
        4.7.2. Environmental agents inducing indirect
               toxic effects in the haematological system
        4.7.3. Measurements and their interpretation
        4.7.4. Example: Effects of low lead
               concentrations on workers' health

   4.8. Effects on the musculoskeletal system and growth
        4.8.1. Effects of environmental exposure
        4.8.2. Identification of effects
        4.8.3. Intrinsic liability
        4.8.4. Extraneous influences
        4.8.5. Development states
        4.8.6. Example: Endemic fluorosis
   4.9. Effects on skin
        4.9.1. Environmentally caused skin diseases
        4.9.2. Epidemiological methods of study
   4.10. Reproductive effects
        4.10.1. Effects on reproductive organs
        4.10.2. Genetic effects
      Assessment of genetic risks
        4.10.3. Fetotoxic effects
      Measurement of fetotoxic effects
        4.10.4. Registries of genetic diseases and
        4.10.5. Example: EEC study of congenital
   4.11. Effects on other major internal organs
        4.11.1. Renal system
      Detection of renal diseases
        4.11.2. Bladder
        4.11.3. Gastrointestinal tract
       Stomach and duodenum
        4.11.4. Liver
        4.11.5. Pancreas



   5.1. Introduction
   5.2. Study protocol
        5.2.1. Description of problems and hypothesis formulation
        5.2.2. Description of methods
        5.2.3. Evaluation of institutional-based data sources
        5.2.4. Analysis and reporting of data
        5.2.5. Resources required
        5.2.6. Studies in developing countries
   5.3. Ethical and legal considerations
        5.3.1. Medical confidentiality
   5.4. Time schedule of study
        5.4.1. Preparatory phase
        5.4.2. Pilot study
        5.4.3. Main study
   5.5. Composition of the study team
        5.5.1. Team leadership and epidemiology
        5.5.2. Clinical specialist
        5.5.3. Statistical expertise
        5.5.4. Environmental scientists
        5.5.5. Interviewers and technicians
        5.5.6. Support staff
        5.5.7. Special considerations for developing countries
        5.5.8. Example: Study teams of Itai-Itai disease
               and chronic cadmium poisoning
   5.6. Implementation of study
        5.6.1. Arrangements with local authorities and study population
        5.6.2. Picking samples
       Example: Sampling procedures
        5.6.3. Designing recording forms and questionnaires
        5.6.4. Planning for control of data and computer programming
        5.6.5. Training of personnel
        5.6.6. Pilot study
       Example: Testing of spirometers
                        and assessment of observer error
       Example: Assessment of X-ray observer error
        5.6.7. Main study
       Advance contact
       Interview studies
       Medical and laboratory examinations
       Environmental measurements
       Linkage and evaluation of data
       Reporting of results
        5.6.8. Examples of cohort studies
       Michigan polybrominated biphenyls study
       Study on air pollution and
                        adverse health effects in Bombay
       Tucson chronic obstructive lung disease study
       The Tecumseh community health study
       Late effects of atomic bomb radiation

   5.7. International collaborative studies
        5.7.1. Study protocol and timetable
        5.7.2. Organizational and sampling procedures
        5.7.3. Questionnaires
        5.7.4. Standardization of measurement instruments
               and methods and quality assurance
        5.7.5. Reporting forms



   6.1. Introduction
   6.2. Data preparation
        6.2.1. Coding
        6.2.2. Key punching
        6.2.3. Data monitoring and editing
   6.3. Data description (or reduction)
        6.3.1. Purpose
        6.3.2. Frequency distributions and histograms
        6.3.3. Bivariate distributions and scattergrams
        6.3.4. Discrete variables and contingency tables
        6.3.5. Independent and related data
        6.3.6. General points on tables and graphs
        6.3.7. Summary statistics and indices
       Scatter (or dispersion)
       Morbidity and mortality indices
       Proportional mortality
       Relative risk and attributable risk
       Concluding remarks about summary
                        statistics and indices
   6.4. Analysis and interpretation
        6.4.1. Statistical ideas about the interpretation of data
        6.4.2. Errors
        6.4.3. Analysing results from cross-sectional studies
       Qualitative data
       Quantitative data: response and
                        explanatory variables
       Statisticians and computers
       Analysis of variance
       Multiple regression
       Additive linear models
       More complicated models
       Dummy variables
      Selection of variables
      Evaluating "goodness of fit"
      Evaluating the stability of models
      Predicted normal values
      Other methods for studying multivariate data

        6.4.4. Analysis of data from prospective and
               follow-up studies
       Time as a measured variable
       Person-years method
       Modified life-table method
       Overlap of exposure and observation periods
       Lagged exposures
       Measures of latency
       Some analytical techniques
        6.4.5. Analysis of data from case-control studies
       Relative and absolute risks
       Relation between prospective and
                        case-control studies
       Analysis of stratified samples
       Analysis of matched samples
       Effect of ignoring the matching
       Alternative methods of analysis
        6.4.6. Drawing conclusions from analyses
   6.5. Reporting
        6.5.1. The variety of epidemiological reports
        6.5.2. Main scientific report
        6.5.3. Non-technical reports



   7.1. Introduction
   7.2. Communication with the public
   7.3. Important features and limitations of epidemiological information
   7.4. Standard setting
        7.4.1. Factors in standard setting
        7.4.2. Interim standards
   7.5. Assessment of effectiveness of control measures taken
   7.6. Policy of openness



    While every effort has been made to present information in
the criteria documents as accurately as possible without
unduly delaying their publication, mistakes might have
occurred and are likely to occur in the future.  In the
interest of all users of the environmental health criteria
documents, readers are kindly requested to communicate any
errors found to the Manager of the International Programme on
Chemical Safety, World Health Organization, Geneva,
Switzerland, in order that they may be included in corrigenda
which will appear in subsequent volumes.

    In addition, experts in any particular field dealt with in
the criteria documents are kindly requested to make available
to the WHO Secretariat any important published information
that may have inadvertently been omitted and which may change
the evaluation of health risks from exposure to the
environmental agent under examination, so that the information
may be considered in the event of updating and re-evaluation
of the conclusions contained in the criteria documents.


    For more than a century, epidemiological studies have
played an important part in investigations on the ways in
which infectious diseases spread through the community.  At
the same time, intensive experimental work has resulted, in
many cases, in the identification of the bacterial, viral, or
other biological agents involved and therapeutic, preventive,
and control procedures have been introduced.  Epidemiological
studies of biological agents and the effective control of
infectious diseases have paved the way to the use of
epidemiological methods in studying effects of non-biological
agents present in the environment.  However, it is generally
more difficult to reach unequivocal conclusions in studies
involving physical and chemical agents than in those involving
biological ones.  This is because the various factors involved
in studies of non-biological agents and their interactions are
usually more complex.

    The relationship between environmental hazards and the
health of human communities is of growing concern and
increasing interest to governmental and public health
administrators, politicians, and the public.  Despite the
substantial efforts made during the past two decades to expand
epidemiological studies on the effects of environmental
agents, there is a paucity of good studies that are useful in
establishing health criteria and a lack of adequate guidance
for the design and execution of studies and for the evaluation
of results.  There are numerous papers and published reports
scattered throughout the literature.  However, perhaps because
of the rapid development of the subject or of the pressure for
action once a new problem has been recognized, a large
proportion of the studies published to date have suffered from
deficiencies in design, analysis, or interpretation.  There is
an urgent need, therefore, for a publication that sets out
appropriate methodology for such studies, which would help
Member States, relevant scientific institutions and individual
research workers to conduct epidemiological studies in a more
correct manner.

    The concept of a monograph on the "Guidelines on Studies
in Environmental Epidemiology" was born at the meeting of the
WHO Study Group on Epidemiological Methods for Assessment of
the Effects of Environmental Agents on Human Health convened
in Geneva from 7-13 October 1975.  The Study Group considered
that the underlying principles of environmental epidemiology,
particularly with regard to biological agents, had been
sufficiently well established, and that the main problem was
their proper application to the study of the health effects of
physical and chemical agents present in the environment.

    The Study Group therefore recommended, among other
proposals, the preparation of a WHO monograph to provide
guidelines on epidemiological methods for assessing the
effects of environmental agents on human health.  This
recommendation was unanimously endorsed by the Executive
Board of WHO at its fifty-eighth session in 1976.

    In order to fulfil this recommendation, the first meeting
of the editorial group to prepare the monograph was held at
the Medical Research Council Toxicology Unit, St. Bartholomew's
Hospital Medical College, London, on 30 January - 1 February
1978.a  The group agreed on the outline of the monograph,
on the tentative list of over 60 contributors or chapter
coordinators, and on the time schedule for preparation of

    The second meeting of the editorial group was held in
London on 26-29 February l980.  The group reviewed the first
draft of chapters prepared by coordinators, based on
contributions received, and made suggestions for further
editorial work.  It was decided to make clear that the chapters
were not the work of individual coordinators, as the basic
material was being prepared by numerous contributors and
some material was being transposed from one chapter to another
during the course of editing.  The chapter coordinators had a
dual role, compiling the chapters from individual contributions,
and also serving both as editors of individual chapters and
members of the editorial group for the monograph as a whole.
The members of the editorial group and all contributors are
shown on pages 14 - 18, respectively.  Individual contributions
do not generally appear in their original form, as they have
been merged into various chapters; however, some of them have
been more extensively quoted and appear almost as originally

    The editorial group last met for two days at the Institute
of Occupational Medicine in Edinburgh from 20-2l August 1981.
The members of the editorial group were able to consider and
comment on the structure of the individual chapters and on the
volume as a whole.  The arrangements for the joint IEA/WHO
workshop on the monograph, which was to take place the
following week during the IXth Scientific Meeting of the IEA,
were also discussed and the work plan for the final editing of
the monograph was agreed upon.

    Thirty-eight environmental health scientists, including
six members of the editorial group, attended the joint IEA/WHO
workshop, coming from 16 countries and from the Commission of
a   Professor J. Kostrzewski, then the President of the
    International Epidemiological Association (IEA),
    participated in this meeting and was elected chairman of
    the editorial group.  He emphasized the preparedness of
    the IEA to offer full technical support to the project.

European Communities.  The workshop was held in Edinburgh on
24 August 1981.  The members reviewed the drafts of individual
chapters and the monograph as a whole, making valuable comments
and suggestions for improvement.  A list of the participants  
appears on pages 19 and 21.

    The final discussion on the draft of the monograph was held
in Moscow from 30 November - 3 December 1981, attended by a group
of international experts from all WHO regions, five members of
the editorial group, and two WHO staff members.  The participants
are shown on pages 22 and 23.  This meeting was  convened with
the financial assistance of the United Nations Environment
Programme (UNEP) and was hosted by the Centre for International
Projects of the USSR State Committee on Science and Technology
and the A.N. Sysin Institute of General and Communal Hygiene of
the USSR, Moscow, which is also a WHO Collaborating Centre for
for Environmental Health Effects.  Comments and proposals made by
this group greatly helped in making the monograph a more cohesive

    Thus, a total of 102 experts from 26 Member States were
involved in the preparation of the monograph.

    In 1980, WHO, together with UNEP and ILO, launched an 
International Programme on Chemical Safety (IPCS) and the present 
project has become one of the priority projects within the frame-
work of this new international programme. 

    Although efforts have been made to avoid inconsistency in
terminology, uniformity has not been possible;  indeed, this is
something beyond the scope of the present monograph.  However,
within the IPCS, a project is under way to compile internationally
agreed definitions for the terms most frequently used in toxicological 
and epidemiological studies.  At present, therefore, it is important 
to understand that some terms may have various meanings and 
implications in different countries or in different scientific 
circles and that it may be highly misleading to employ them outside 
the national pattern of use or outside the context of a specialized 
field, without precise definition.  The reader is therefore warned 
to be wary of the uncritical transfer of technical terms from one 
set of circumstances to another.

    The present monograph should serve as a useful guide to
the conduct of epidemiological studies on the effects of
non-biological agents on the health of human communities.  An
epidemiological investigation would involve not only
epidemiologists but also other experts (e.g., clinicians,
statisticians, engineers, chemists, and nurses), and the
monograph is intended to address such a broad audience.  The

editorial group believes that this publication will pave the
way for future studies.a

    The final editing of the monograph was carried out during
1982 by M. Lebowitz and R. Waller, assisted by J. Kostrzewski,
M. Jacobsen, and Y. Hasegawa.

    A special tribute should be paid to the late Dr F. Sawicki,
in recognition of his valuable contribution to this monograph,
and his extensive work on environmental factors in relation to
respiratory diseases.

                               The Editorial Group

                              * * *

    Partial financial support for the development of this
criteria document was kindly provided by the Department of
Health and Human Services through a contract from the National
Institute of Environmental Health Sciences, Research Triangle
Park, North Carolina, USA - an IPCS Lead Institution.

a   For instance, collaborative epidemiological studies on the
    health effects of several chemicals being initiated under
    the coordination of the WHO Regional Office for Europe
    will apply the principles set out in this publication.
    The editorial group was aware that a manual on epidemiology
    for occupational health was being prepared jointly by the
    World Health Organization in Geneva and the WHO Regional 
    Office for Europe; although the present Monograph has cited 
    several occupational health studies, it is addressed to 
    problems in the general population rather than among 
    specific occupational groups. 


 Editorial Groupa

Dr K.P. Duncan, Health and Safety Executive, London, England
   (Coordinator for Chapter 7)

Dr M. Greenberg, Health and Safety Executive, London, England
   (Coordinator for Chapter 4)

Dr Y. Hasegawa, Environmental Hazards and Food Protection,
   Division of Environmental Health, World Health
   Organization, Geneva, Switzerland  (Secretary)

Professor I.T.T. Higgins, School of Public Health, University
   of Michigan, Ann Arbor, Michigan, USA (Coordinator for
   Chapter 2)

Dr M. Jacobsen, Institute of Occupational Medicine, Edinburgh,
   Scotland (Coordinator for Chapter 6)

Professor J. Kostrzewski, Department of Epidemiology, National
   Institute of Hygiene, Warsaw, Poland  (Chairman)

Professor P.J. Lawther, MRC Toxicology Unit, Clinical Section,
   St. Bartholomew's Hospital Medical College, London, England

Professor M.D. Lebowitz, Division of Respiratory Sciences,
   Health Sciences Center, University of Arizona, Tucson,
   Arizona, USA (Coordinator for Chapter 5)

Dr I. Shigematsu, Radiation Effects Research Foundation,
   Hiroshima, Japan

Mr R. E. Waller, Toxicology and Environmental Protection,
   Department of Health and Social Security, London, England
   (Coordinator for Chapter 1)

Professor R.L. Zielhuis, University of Amsterdam, Coronel
    Laboratorium, Netherlands (Coordinator for Chapter 3)

a   Professor W.W. Holland and Dr C. du V. Florey, St.
    Thomas's Hospital Medical School, London, were initially
    the Coordinators for Chapter 4.  As they were unable to
    continue the work it has been taken over by Dr Greenberg.
    The Secretariat wishes to thank Professor Holland and Dr
    Florey for their efforts.



Professor M. Alderson, Institute of Cancer Research, Surrey,
   England (contributor to section 4.2)

Professor K. Biersteker, Agricultural University, Wageningen,
   Netherlands (contributor to Chapter 2)

Professor N.P. Bochkov, Institute of Medical Genetics, Moscow,
   USSR (contributor to section 4.10)

Professor D.S. Borgaonkar, North Texas State University,
   Denton, Texas, USA (contributor to section 4.10)

Dr N. Breslow, International Agency for Research on Cancer,
   Lyon, France (contributor to Chapter 6)

Dr D.G. Clegg, Food Directorate, Department of National Health
   and Welfare, Ottawa, Canada (contributor to section 3.5)

Dr J.H. Cummings, Dunn Clinical Nutrition Centre, Addenbrookes
   Hospital, Cambridge, England (contributor to section 4.11)

Professor W.J. Eylenbosch, University of Antwerp, Wilrijk,
   Belgium (contributor to sections 4.10 and 4.11)

Dr C. Favaretti, Institute of Hygiene, University of Padua,
   Italy (contributor to Chapter 7)

Dr A.J. Fox, The City University, London, England (contributor
   to Chapter 6)

Mrs M. Fugas, Institute of Medical Research and Occupational
   Health, Zagreb, Yugoslavia (contributor to Chapter 3)

Professor J.R. Goldsmith, Ben Gurion University, Beer Sheva,
   Israel (contributor to Chapter 3)

Professor B.D. Goldstein, New York University Medical Center,
   New York, USA (contributor to section 4.7)

Professor I.F. Goldstein, Columbia University, School of
   Public Health, New York, USA (contributor to Chapter 6)

Professor M. Hashimoto, Graduate School of Environmental
   Sciences, Tsukuba University, Ibaraki-ken, Japan
   (contributor to Chapter 7)

Professor M.W. Higgins, School of Public Health, University of
   Michigan, Ann Arbor, Michigan, USA (contributor to section

 Contributors (contd.)

Dr A.W. Hubbard, Ministry of Agriculture, Fisheries and Food,
   London, England (contributor to section 3.5)

Dr A. Jablensky, Division of Mental Health, World Health
   Organization, Geneva, Switzerland (contributor to section

Dr A. Jakubowski, Institute of Occupational Medicine and Rural
   Hygiene, Lublin, Poland (contributor to section 3.5)

Dr H. P. Jammet, Centre for Nuclear Studies,
   Fontenay-aux-Roses, France (contributor to section 3.5)

Professor F. Kaloyanova, Institute of Hygiene and Occupational
   Health, Sofia, Bulgaria (contributor to Chapter 3)

Professor S.R. Kamat, Department of Chest Medicine, K.E.M.
   Hospital, Bombay, India (contributor to sections 4.8 and

Dr H. Kato, Radiation Effects Research Foundation, Hiroshima,
   Japan (contributor to section 5.6)

Professor L.T. Kurland, Mayo Clinic, Rochester, Minnesota, USA
   (contributor to section 4.5)

Dr J.F. Kurtzke, Veterans Administration Hospital, Washington,
   DC, USA (contributor to section 4.5)

Dr P.J. Landrigan, National Institute for Occupational Safety
   and Health Cincinnati, Ohio, USA (contributor to Chapter 5)

Professor M.F. Lechat, Catholic University of Louvain,
   Brussels, Belgium (contributor to section 4.10)

Professor S.R. Leeder, University of Newcastle, New South
   Wales, Australia (contributor to section 4.4)

Dr D.T. Mage, Health Effects Research Laboratory,
   Environmental Protection Agency, Research Triangle Park,
   North Carolina, USA (contributor to Chapter 3)

Dr P.B. Meijer, TNO Research Institute for Environmental
   Hygiene, Delft, Netherlands (contributor to Chapter 3)

Dr W.E. Miall, MRC Epidemiology and Medical Care Unit, Harrow,
   England (contributor to Chapter 5)

Dr C.S. Muir, International Agency for Research on Cancer,
   Lyons, France (contributor to Chapter 2 and section 4.3)

Profesor M. Nikonorow, National Institute of Hygiene, Warsaw,
   Poland (contributor to section 3.5)

 Contributors (contd.)

Dr B.R. Ordo]ez, Autonomous Metropolitan University, Mexico
   City, Mexico (contributor to Chapter 7)

Professor B. Paccagnella, Institute of Hygiene, University of
   Padua, Italy (contributor to Chapter 7)

Dr R.F. Packham, Water Research Centre, Medmenham, England
   (contributor to section 3.5)

Professor B.S. Pasternack, New York University Medical Center,
   New York, USA (contributor to Chapter 6)

Professor W.O. Phoon, University of Singapore, Singapore
   (contributor to Chapter 5)

Dr I. Purcell, University of Newcastle, New South Wales,
   Australia (contributor to section 4.4)

Professor A.V. Roscin, Central Institute for Advanced Medical
   Training, Moscow, USSR (contributor to section 4.7)

Dr M. Saric, Institute for Medical Research and Occupational
   Health, Zagreb, Yugoslavia (contributor to section 4.4)

Dr F. Sawicki, National Institute of Hygiene, Warsaw, Poland
   (contributor to Chapter 5)

Dr M.A. Schneiderman, National Cancer Institute, Bethesda,
   Maryland, USA (contributor to Chapter 2)

Dr I. Shigematsu, Radiation Effects Research Foundation,
   Hiroshima, Japan (contributor to section 5.5 and Chapter 7)

Dr C. Silverman, Bureau of Radiological Health, US Food and
   Drug Administration, Rockville, Maryland, USA (contributor
   to section 3.5)

Professor F.H. Sobels, University of Leiden, Leiden,
   Netherlands (contributor to section 4.10)

Dr E. Somers, Environmental Health Directorate, Department of
   National Health and Welfare, Ottawa, Canada (contributor
   to Chapter 7)

Dr J. H. Stebbings, Jr., Los Alamos Scientific Laboratory, Los
   Alamos, New Mexico, USA (contributor to Chapter 6)

Dr A. H. Suter, Occupational Safety and Health Administration,
   Department of Labor, Washington, DC, USA (contributor to
   section 3.5)

Dr H. Tamashiro, Institute of Public Health, Tokyo, Japan
   (contributor to section 5.5 and Chapter 7)

 Contributors (contd.)

Dr M.P. van Sprundel, University of Antwerp, Wilrijk, Belgium
   (contributor to sections 4.10 and 4.11)

Dr M. Violaki-Paraskeva, Ministry of Social Services, Athens,
   Greece (contributor to Chapter 7)

Professor D. Wassermann, The Hebrew University Hadassah
   Medical School, Jerusalem, Israel (contributor to Chapter

Professor M. Wassermann, The Hebrew University Hadassah
   Medical School, Jerusalem, Israel (contributor to Chapter

Professor W.E. Waters, Community Medicine, Southampton General
   Hospital, Southampton, England (contributor to section 5.3)

Dr J.A.C. Weatherall, Office of Population Censuses and
   Surveys, London, England (contributor to section 4.10)

Professor P.H.N. Wood, The Arthritis and Rheumatism Council
   Epidemiology Research Unit, University of Manchester,
   Manchester, England (contributor to section 4.8)

Dr M. Zaphiropoulos, Ministry of Social Services, Athens,
   Greece (contributor to Chapter 7)



Dr E. Bennet, Health and Safety Directorate, Commission of the
   European Communities, Luxembourg

Dr S. Beresford, Royal Free Hospital School of Medicine,
   London, England

Dr G. W. Brebe, Clinical Epidemiology, National Cancer
   Institute, National Institutes of Health, Bethesda,
   Maryland, USA

Professor L. Breslow, School of Public Health, University of
   California Los Angeles, California, USA

Dr D. Brille, Studies and Research Mission, Ministry of the
   Environment, Paris, France

Dr P.G.J. Burney, Department of Community Medicine, St Thomas'
   Hospital Medical School, London, England

Miss M. Deane, Epidemiological Studies Section, California State
   Department of Health Services, Berkeley, California, USA

Dr A.R. Eltom, Faculty of Medicine, Khartoum, Sudan

Professor W.S. Eylenbosch, Department of Epidemiology and
   Social Medicine, University of Antwerp, Wilrijk, Belgium

Professor G.M. Fara, Institute of Hygiene, Milan, Italy

Dr J.J. Feldman, Analysis and Epidemiology, National Center
   for Health Statistics, Hyattsville, Maryland, USA

Dr I.F. Goldstein, Environmental Epidemiology Research Unit,
   Columbia University, School of Public Health, New York, USA

Dr Y. Hasegawa, Environmental Hazards and Food Protection,
   Division of Environmental Health, World Health
   Organization, Geneva, Switzerland

Dr N. M. Hanis, Epidemiology Unit, Research and Environmental
   Division, Medical Department, Exxon Corporation, E.
   Millstone, New Jersey, USA and Cornell University Medical
   School, New York, USA

Professor I.T.T. Higgins, School of Public Health, University
   of Michigan, Ann Arbor, Michigan, USA

Professor M.W. Higgins, Department of Epidemiology, School of
   Public Health, University of Michigan, Ann Arbor,
   Michigan, USA

 Members (contd.)

Dr M. Hitosugi, Department of Public Health, School of
   Medicine, Kitasato University, Sagami-hara City,
   Kanagawa-Ken, Japan

Professor A.C. Irwin, Department of Preventive Medicine,
   Dalhousie University, Halifax, Nova Scotia, Canada

Dr M. Jacobsen, Institute of Occupational Medicine, Edinburgh,

Professor H. Kasuga, Department of Public Health, School of
   Medicine, Tokai University, Isehara-Shi, Kanagawa-Ken,

Dr M. Khogali, Faculty of Medicine, Kuwait University, Safat,

Professor M.A. Klinberg, Department of Preventive and Social
   Medicine, Tel-Aviv University School of Medicine, Ramat
   Aviv, Israel

Professor J. Kostrzewski, Department of Epidemiology, National
   Institute of Hygiene, Warsaw, Poland

Professor L.T. Kurland, Department of Medical Statistics and
   Epidemiology, Mayo Clinic, Rochester, Minnesota, USA

Professor R.A. Kurtz, Faculty of Medicine, Kuwait University,
   Safat, Kuwait

Professor M. Lebowitz, Division of Respiratory Sciences,
   Health Sciences Center, University of Arizona, Tucson,
   Arizona, USA

Dr S. Mazumdar, Department of Biostatistics, University of
   Pittsburgh, Pittsburgh, Pennsylvania, USA

Dr U. G. Oleru, College of Medicine, University of Lagos,
   Lagos, Nigeria

Professor B.S. Pasternack, Department of Environmental
   Medicine, New York University, Medical Center, New York,

Professor M.R. Pandey, Thapathali, Katmandu, Nepal

Professor W.O. Phoon, Department of Social Medicine and Public
   Health, National University of Singapore, Singapore

Dr M.P. Sprundel, Department of Epidemiology and Social
   Medicine, University of Antwerp, Wilrijk, Belgium

 Members (contd.)

Professor R. Steele, Department of Community Health and
   Epidemiology, Queen's University, Kingston, Ontario, Canada

Dr H. Tamashiro, National Institute for Minamata Disease,
   Minamata City, Kumamoto-Ken, Japan

Professor K.W. Tietze, Federal Health Office, Berlin (West)

Dr M. Wahdan, Regional Adviser on Epidemiology, WHO Regional
   Office for Eastern Mediterranean, Alexandria, Egypt

Mr R.E. Waller, Toxicology and Environmental Protection,
   Department of Health and Social Security, London, England

Professor W. Winkelstein, School of Public Health, University
   of California, Berkeley, California, USA



Dr L.K.A. Derban, Medical Officer, Volta River Authority,
   Accra, Ghana

Dr C. Favaretti, Institute of Hygiene, University of Padua,

Dr M. Jacobsen, Institute of Occupational Medicine, Edinburgh,

Professor S.R. Kamat, Department of Chest Medicine, K.E.M.
   Hospital, Bombay, India

Professor J. Kostrzewski, Department of Epidemiology, National
   Intitute of Hygiene, Warsaw, Poland  (Chairman)

Professor M.D. Lebowitz, Division of Respiratory Sciences,
   Health Sciences Center, University of Arizona, Tucson,
   Arizona, USA  (Co-rapporteur)

Professor A. Massoud, Department of Community, Industrial and
   Environmental Medicine, Ain Shams University, Cairo, Egypt

Dr B.R. Ordonez, Environmental Health Programme, Autonomous
   Metropolitan University, Mexico City, Mexico

Professor A.V. Roscin, Central Institute for Advanced Medical
   Training, Moscow, USSR

Dr I. Shigematsu, Radiation Effects Research Foundation,
   Hiroshima, Japan

Academician G.J. Sidorenko, A.N. Sysin Institute of General
   and Communal Hygiene, Academy of Medical Sciences of the
   USSR, Moscow, USSR  (Vice Chairman)

Mr R.E. Waller, Toxicology and Environmental Protection,
   Department of Health and Social Security, London, England

 WHO Secretariat

Dr I. Farkas, Promotion of Environmental Health, WHO Regional
   Office for Europe, Copenhagen, Denmark

Dr Y. Hasegawa, Medical Officer, Environmental Hazards and
   Food Protection, Division of Environmental Health, WHO,
   Geneva, Switzerland  (Secretary)

 Other participants

Dr I.R. Golubev, Department of Public Health, USSR State
   Committee for Science and Technology, Moscow, USSR

Dr Z.P. Grigorievskaya, A.N. Sysin Institute of General and
   Communal Hygiene, Moscow, USSR

Dr Y.E. Korneyev, Laboratory of Epidemiological Methods of
   Study, A.N. Sysin Institute of General and Communal
   Hygiene, Moscow, USSR  (Co-rapporteur)

Dr N.N. Litvinov, A.N. Sysin Institute of General and Communal
   Hygiene, Moscow, USSR

Dr Y.I. Prokopenko, Department of the Influence of
   Environmental Factors of Public Health, A.N. Sysin
   Institute of General and Communal Hygiene, Moscow, USSR

Dr Ya.I. Zvinjackovskij, Laboratory of the Influence of
   Environmental Factors of Public Health, Marseev Institute
   of General and Communical Hygiene, Kiev, USSR

                           * * *


Dr A. David, Office of Occupational Health, Division of
   Noncommunicable Diseases, WHO, Geneva, Switzerland

Mr J. Duppenthaler, Division of Epidemiological Surveillance
   and Health Situation and Trend Assessment, WHO, Geneva,

Dr K. Hemminki, Institute of Occupational Health, Helsinki,

Dr J. Stjernswärd, Cancer, Division of Noncommunicable
   Diseases, WHO, Geneva, Switzerland

Dr C. Xintaras, Office of Occupational Health, Division of
   Noncommunicable Diseases, WHO, Geneva, Switzerland


1.1.  Interrelationships with Toxicological Studies

    In some respects the present volume is intended to complement 
an earlier publication in the Environmental Health Criteria series, 
 "Principles and methods for evaluating the toxicity of chemicals - 
 Part I",  which dealt with experimental work using mainly animals
and other biological assay systems (WHO, 1978).  There are some 
parallels between such laboratory studies and epidemiological  
investigations of the effects of hazardous substances on human 
populations.  The object, in each case, is to compare the effects 
on groups subjected to different levels of the suspect agent, 
always ensuring that the groups are matched as far as possible in 
respect of other relevant factors (which may include sex, age, 
temperature etc).  Much experimental work is indeed done on human 
subjects, restricted to doses that will evoke only relatively minor 
physiological or biochemical responses that are readily reversible.  
The borderline between laboratory experimentation and epidemological 
work is not clearly defined.  For the present purposes, however, 
straight-forward toxicological studies on human beings, in which 
the effects of specified doses of suspect agents administered 
to small groups of subjects in the laboratory are examined, will 
not be considered. 

    There are extensions of this approach which form a bridge 
between laboratory work and that in the general environment, and 
some mention should be made of these.  Environmental chambers have 
been constructed by some research groups, where small numbers of 
subjects may spend periods of hours, days, or weeks under closely 
controlled conditions.  These have application in studies on acute 
effects, and have been used, for example, to investigate the 
effects of exposure to polluted urban air, drawn in from the 
general atmosphere and carrying out control experiments with clean 
air (Kerr, 1973).  In isolated instances, this approach can be taken 
a little further, for example, studies on the effects of lead 
intake can be done by controlling the air and/or diet of groups 
such as prisoners living in confined conditions (Cole & Lynam, 
1973).  More generally, however, the investigator cannot control 
either the exposure of the subjects or their activities.  Advantage 
must then be taken of existing contrasts in environmental exposures 
to obtain evidence on effects on health.  In many cases, both 
toxicological and epidemiological data are essential in 
establishing sound health criteria and they are complementary to 
each other. 

1.2.  Design

    Perhaps as an over-reaction to a number of environmental 
"disasters" that have occurred around the world, there has been a 
tendency in recent years to carry out epidemiological imvestigations 
without first posing any specific questions.  There is indeed a 
place for exploratory studies, often based  on existing routinely 
collected data on mortality or morbidity together with general 
observations on environmental factors, but further studies 

need to be carefully designed to test specific hypotheses.  Then 
one has to ask: 

     Who  should be studied?  Are particular subgroups of the
population at risk?  How should control groups be selected?

     What  should be measured?  Can specific agents be
identified?  Is there a single pathway (for example, via
inhalation) or have several ways of entry to be considered
simultaneously?  How are effects on health to be assessed?

     Where  has the study to take place?  Should geographical
position, altitude, meteorology, etc., be taken into account
in selecting a locality?  Are there existing monitoring
stations or sets of data relating to the environmental factors
in question?

     When  should the study be carried out?  Are seasonal
effects likely to be important?  Is the available time-span
long enough to provide a satisfactory estimate of long-term
exposures?  Should exposures be averaged over months or years,
or are short-term peaks relevant in some cases?

    In designing a special investigation or survey in the field of 
environmental health, the objects of the exercise must first be 
considered carefully.  Without advocating a strict cost/benefit 
approach to such studies, the question of the amount of time and 
money spent in relation to the probable yield of information must 
obviously be of importance.  At one end of the spectrum, one might 
consider monitoring the health records of the whole population, and 
linking the information with as many data on environmental factors 
as possible.  Certainly, the monitoring of national death 
statistics and of some aspects of morbidity records is possible, 
looking particularly for the emergence of new trends or patterns of 
distribution in congenital abnormalities and relatively rare 
diseases.  The need for this was underlined by the thalidomide 
episode and, although the use of therapeutic drugs is not being 
considered specifically in the present context, there are many 
parallels between present-day enquiries into the safety of drugs 
and the conduct of epidemiological studies on environmental agents.  
However, to go beyond a broad surveillance such as this, with 
enquiries into "health and habits" on a national scale, might be 
regarded as an intrusion on personal privacy, apart from the 
prohibitive cost.  Even so, there is evidence that careful scanning 
of linked records maintained on a regional, if not national, scale 
can reveal new problems, as in the case of the occurrence of nasal 
cancer among furniture makers (Acheson et al., 1967). 

    In this particular example, suspicions had been aroused by 
clinical investigations on a few cases; however often, with a 
relatively rare disease the "clustering" of a few cases in one area 
or within one small subgroup of the population is sufficient to       
give a positive lead on a new environmental hazard.  In general, 
when there is some indication of adverse effects of a particular 
agent, the most effective way to conduct further epidemiological 

studies is to concentrate attention on groups of people considered 
to be particularly at risk.  An example of this for a physical 
agent - noise - is the investigation of exposure to "pop" music, 
conducted among young college students (Hanson & Fearn, 1975).  In 
this study, dose-response relationships were examined within the 
group selected, but, in general, it may be necessary to include 
appropriate control groups, not exposed to the suspect agent. 

    An alternative approach, still directed towards high-risk 
subgroups, is to consider a specific disease or effect, and to 
compare the available information on exposures to environmental 
agents with those of a control group.  This is the "case-control" 
type of study that was so successful in the early stages of the 
investigation of the role of environmental factors in the 
development of lung cancer (Doll & Hill, 1950; Wynder & Graham, 

    On wider issues, where the interrelationships between agents 
and effects are more diffuse or more tenuous, relatively expensive 
general community surveys may be needed, based on random samples of 
the population concerned, or of particular age or occupational 
groups.  This technique has been particularly valuable in studies 
on the role of environmental factors in the development of 

    The types of survey that have proved to be of value in the 
study of the effects of environmental agents are described in 
Chapter 2.  The dividing lines between them are not always clearly 
defined, and there may be advantages in combining several 
approaches within a single survey.  The choice will depend on the 
objects of the study and on the resources available. 

1.3.  Environmental Agents and Assessment of Exposures

    As indicated in the preface, epidemiological methods were 
developed initially to investigate the distribution and determinants 
of communicable diseases, but their scope has now been widened to 
include all aspects of health and wellbeing in relation to 
biological or non-biological agents.  Much of the discussion that 
follows in later chapters on the design, conduct, and analysis of 
epidemiological studies could apply to any field of interest, but 
the prime concern here is with effects of chemical and physical 
agents.  The interaction of bacteria, viruses, fungi, yeasts, 
protozoa, and higher animal agents or vectors with non-biological 
agents is, however, recognized as contributing to human disease. 

    The term "agent" is a neutral one with no intrinsic implication 
of "beneficial" or "adverse" characteristics.  Most agents have the 
potential for one or other or both of  these effects, varying with 
the precise nature of the agent, the level and duration of 
exposure, and the state of nutrition and other acquired or 
inherited characteristics of the subject.  Thus, for example, the 
chemical agents constituting vitamins and their analogues, that may 
serve as essential food factors, are claimed to offer protection 
against certain diseases (e.g., vitamin A against carcinogenesis), 

or to have severe toxic effects, according to dose, to the state of 
nutrition and acquired characteristics of the subject, and to other 
agents operating coincidentally. 

    When studying these chemical and physical agents, it is 
necessary to characterize them and to determine their absorption, 
concentration in air, water, etc. with careful attention to 
precision.  For example, when describing a mineral, it is not 
enough to give the name, chemical formula, and dose.  An adequate 
description involves specifying its contaminants, its physical form 
(amorphous, crystalline, discrete particulate or fibrous), and 
particle size distribution, and sometimes its physicochemical 
surface properties.  Due consideration has to be given to demo-
graphic and sociocultural factors that may affect the degree of 
exposure or uptake as well as to special host characteristics, 
including immunological status, before extrapolating the experience 
in one population to that of another. 

     Chemical agents involved in environmental considerations have 
been characterized as natural and manufactured organic (but not 
living) and inorganic substances occurring in food, air, water, 
soil, and other media.  While living materials are excluded from 
this category, their products are widely distributed in the 
environment, in the form of metabolites, cell bodies, or bio-
chemical extracts.  Thus, many foodstuffs are infested by, or 
require for their synthesis, micro-organisms that are also found in 
the wild and may contaminate the general environment. 

     Physical agents that impinge on man may occur naturally or be 
man-made or man-intensified.  They include ionizing and non-
ionizing radiation, the latter ranging from ultraviolet through 
visible light and infrared to microwave, radio frequency and 
extremely low frequency electromagnetic fields.  Climatic 
conditions of temperature and humidity play important direct and 
indirect roles in environmental health.  Noise and vibration at the 
intensities experienced occupationally are associated with 
objective evidence of damage; lesser intensities occurring outside 
occupational environments, apart from affecting amenity, are a 
source of concern in case they present health hazards. 

    The assessment of exposures is the most difficult aspect of 
epidemiological research on environmental agents, and the one that 
requires most careful thought, if any attempt is to be made to 
establish "exposure/effect" relationships.  For mixtures of 
pollutants that are found under actual environmental conditions, 
some integrated approach would be required for adequate exposure 
assessment.  However, such an approach has still to be developed. 
    The commonest practice is to monitor concentrations or 
intensities (in air, water, etc.) at fixed points in order to make 
estimates of the exposure of the community being investigated.  
Measurements may have to be specially made for each investigation, 
but advantage can often be taken of existing monitoring networks.  
There has been a rapid expansion of monitoring activities 
throughout the world in recent years:  many reports have been 

written about national and international programmes (Munn, 1973; 
Department of the Environment, 1974), and a computer-based record 
of current work is maintained in the United States of America 
(Whitman, 1975).  Although it is possible to monitor environmental 
variables continuously at a large number of sites, it is impossible 
to use that information in an undigested form in epidemiological 
studies in which the health indices are generally crude.  Provision 
must therefore be made for statistical analyses of the data, once 
collected, and, in some schemes operating with automatic 
instruments, statistical analysers are incorporated, or the 
instruments are "on line" to a central computer.  A scheme of this 
type in the field of air pollution monitoring has been described by 
Lauer & Benson (1974).  There is a risk however of becoming 
overwhelmed with data from such complex networks and in most 
epidemiological studies, it is more important to consider what is 
the minimum requirement for a reasonable assessment of exposures 
than to collect a vast array of data from which to select a few 

    "Personal" monitors may sometimes be applicable.  This is 
particularly true in assessing exposures to ionizing radiation, as 
simple integrating devices (such as film badges) are available.  It 
is more difficult to measure most other environmental agents with 
light portable equipment, but personal samplers for air pollutants, 
such as suspended particulate matter and sulfur dioxide, are 
available.  Even so, the initial cost and maintenance problems 
associated with these are at present deterrents to their use in 
large-scale epidemiological studies. 

    For multimedia and non- or less-degradable pollutants, such as 
metals and many organochlorine compounds, the biological monitoring 
method, namely, the measurement of levels of polllutants in 
tissues and fluids, has proved to be a useful tool for exposure 

    Procedures for the assessment of exposures and for their 
quality control are described in detail in Chapter 3:  in general 
the principles involved are meant to apply to any type of 
environmental agent, though measurement methods are specific to 
the one in question. 

1.4.  Effects on Health                          
    Physical and chemical agents generated by man's activities   
may have various effects on human being.  Some substances may not 
produce any adverse effects, while others, may be liable, if     
exposures are sufficient, to affect such basic phenomena as      
growth and development.  Sometimes, environmental exposures may 
affect host susceptibility or resistance, or produce functional or 
prepathological changes.  Behaviour may be modified by exposure, 
especially to physical agents such as noise, light, and heat.  A 
wide range of pathological states in different organs may be 
induced by exposure to environmental agents, and even death may be 
caused or hastened by such exposures. 

    The starting point for many studies on the effects of 
environmental agents has been the examination of existing records 
of mortality or morbidity.  The interpretation of findings from 
these may itself be hazardous, but to determine which effects 
should be studied, this retrospective approach is often considered 

    In most countries, there are well-established systems of 
registration of deaths, in which the cause of death is reported 
(with varying accuracy) along with the age, date and place of 
death, place of usual residence, marital status, occupation, and, 
in some cases, additional information that may allow links to be 
established with birth registration or other particulars of the 
same individual. 

    Examination of long-term trends in death rates, or of 
differences between countries, can occasionally give leads on 
suspect environmental agents, but the most fruitful analyses in the 
past have been those of local and regional differences in death 
rates from specific diseases, within single countries.  Thus, an 
excess of cancer of the oesophagus could be seen in certain areas 
of France, or of bronchitis in the industrial towns of the United 
Kingdom, and, in these and many other examples, the findings have 
been confirmed and investigated further in carefully designed 
epidemiological surveys. 

    Occupational mortality studies can also be very valuable, but 
those based on nationally-collected statistics are difficult to 
interpret, since "occupation" may be inadequately described by the 
relatives who have to give the information entered on death 
certificates.  Adelstein (1972) has also drawn attention to the 
difficulty in distinguishing occupational risks from those of 
"social" origin (notably tobacco smoking) in the more recent 
records, and occupational studies are now better done as special 
surveys within industries. 

    Short-term changes in death rates can provide information on a 
limited range of agents that are subject to large variations in 
intensity over periods of months, weeks, or days, and are 
potentially lethal to some sections of the community.  Many 
diseases spread by bacterial and viral agents fall into this 
category, but the main examples among physical and chemical agents 
are air pollution and climatic conditions.  Tabulations of deaths 
on a monthly or weekly basis may be of some value in seeking any 
evidence of acute effects of these factors, but ideally daily 
tabulations are required, for selected areas containing large 

    Death is a crude but clearly defined index of response and it 
is the one that has been most widely used in studies of the effects 
of environmental agents.  Where a small number of otherwise healthy 
people die suddenly in one incident, perhaps as the result of an 
accidental release of toxic materials in industry, cause and effect 
relationships are easily established, but, in the general 
community, associations are usually far more tenuous.  It is 
commonly the weakest sections of the community that are most 

sensitive to potentially lethal effects of environmental agents:  
the very old, the chronic sick, and the very young.  In studies of 
acute effects, it may be sufficient to study changes in the total 
number of deaths in a given area, but specificity can often be 
improved by considering deaths within limited age-ranges or for 
certain causes only.  Surprisingly, even the effects of major 
insults to health may not be immediately obvious, if they impinge 
mainly on the very old, among whom death rates are normally 
relatively high.  In the London fog of December 1952, there were 
general indications of an exceptional death rate, such as a 
shortage of coffins and flowers for funerals, but it was not until 
all the returns of deaths from local registrars were collected 
together and scrutinized that it was realized that the number of 
deaths during and just after the fog was about 4000 more than would 
normally have been expected (Ministry of Health, 1954). 

    Illness, as defined in various ways in routinely collected 
morbidity statistics, can be regarded as a further index of 
response.  The more it is qualified in terms of age-range and 
disease category the better it is, but there are many hazards in 
accepting information collected largely for administrative 
purposes, because of possible biases.  Morbidity data are far more 
subject to interference from social factors than mortality data; 
weekends and holidays, for example, have little effect on death 
rates, but they have a profound effect on consultation rates with 
general practitioners and on hospital admissions.  Provided these 
reservations are borne in mind, it is still possible to make use of 
some existing information for epidemiological studies. 

    The range of indices of effects on health available for use in 
specially designed epidemiological surveys is very wide and covers 
all organ systems.  It is described in detail in Chapter 4 and it 
can include death, the onset or prevalence of specific illnesses, 
measurements of developmental, behavioural, functional, and 
prepathological changes, and biochemical indices.  There are, 
however, limitations of costs, usage, and acceptability of some of 
the tests. 

1.5.  Organization and Conduct

    There are many practical problems to consider in the 
organization and conduct of epidemiological studies and the 
recommended procedures are described in Chapter 5.  Both the level
of study to be conducted (simple to complex) and the resources 
required to do it must be considered.  As in the rest of the 
monograph, studies are described which can be conducted in various 
settings in the world.  There is often preliminary work to do in 
contacting organizations that may be able to provide, or help in 
collection of, the health and environmental data, or may merely 
need to be made aware of the aims and existence of the survey.  
Some advance publicity may be desirable to gain the cooperation of 
subjects and, where occupational groups are concerned, discussions 
with managements and unions are essential. 

    Having selected the population required for the study, initial 
contacts with individuals may need to be made by letter, prior to 
any interview or examination.  One of the major difficulties is to 
obtain an adequate response from the population selected.  Particu-
larly in studies of chronic effects, where contrasts are being 
sought between people living in different areas, it is essential to 
ensure that failure to contact or to follow up some of the subjects 
does not bias the result.  The possibility of observer bias must 
also be considered.  Where a number of observers are engaged in 
interviewing or examining subjects in different localities, joint 
training sessions are required to ensure uniformity of approach and 
it may be necessary to interchange the teams to reduce risk of 
bias.  Even where objective assessments are being made, for example 
with peak flow measurements in surveys of respiratory disease, it 
is important to ensure uniformity of procedure and to check 
regularly the performance or calibration of the instruments being 
used.  Careful standardization of methods of measuring the related 
environmental agents is also required and, if biological indices of 
effects are being used, it may be necessary to ensure that all 
these measurements are made in a single laboratory. 

    The conduct of the field work itself will depend on the nature 
of the survey.  Surveys can be simple or complex.  Subjects may be 
seen only by field workers, but preferably by those who know the 
community and its culture.  Subjects may be asked to come to one of 
several bases that might be set up in the areas near where the 
subjects live or to a central laboratory or clinic.  In surveys 
requiring instruments for the measurement of lung function, etc., 
mobile laboratories are sometimes used.  Where the subjects are 
grouped together, for example in selected schools, offices, or 
factories, the survey team will normally visit them there, by prior 
arrangement with the authorities concerned.  The most labour-
intensive survey, but often the most satisfactory, where the 
effects of common environmental agents are being studied on 
samples of the general population, is where the field workers 
visit the subjects in their own homes. 

    Ethical problems sometimes arise; for example, if some of the 
tests involved are regarded as intrusive.  In surveys of exposure 
to lead, blood samples may be required and, although there is 
relatively little difficulty in taking these with the prior 
permission of the subject in the case of adults, problems arise 
with children, for parents and others cannot properly give 
permission for samples to be taken in this way, if it is not for 
the benefit of the child.  Apart from this, confidentiality of 
all information obtained in surveys must be maintained at all 
times, hence it is common practice to exclude names and addresses 
at all stages of the preparation and analysis of results, beyond 
the original survey form. 

1.6.  Analysis and Interpretation of Results

    In some surveys of modest size and quite often in the case of 
studies on acute effects of environmental agents, the findings may 
be tabulated manually and/or presented graphically in a straight-

forward manner.  More generally, however, the data will be 
transcribed on to punched cards, paper tape, or magnetic tape for 
analysis by computer.  Procedures for the preparation of the data, 
and for the analysis and interpretation of findings are described 
in detail in Chapter 6, in which the need for the close involvement 
of statistical staff throughout the study is stressed. 

    The statistical analysis of epidemiological studies has been 
revolutionized by the application of "package" programmes.  These 
have been written as general purpose statistical routines and 
survey analyses and, although they may be handled by people with 
relatively little statistical and computing experience, it is 
essential to have expert advice and guidance to avoid misapplying 
the techniques or misinterpreting the findings.  The more complex 
the technique, the more necessary it is to pause to consider the 
relevance of the data, and, if possible, to try to provide some 
visual presentation of the main features, for example, in the form 
of a graph that may be displayed on a screen linked with the 
computer, or plotted out on microfilm or by line-printer. 

    A fundamental point in relation to the control of 
environmental pollutants is whether there is any kind of level of 
exposure, below which effects of an environmental agent are not 
detectable (with the techniques used), but beyond which effects 
increase gradually in a defined relationship.  A feature such as 
this may be extremely difficult to establish, since the effects of 
very low levels of exposure cannot be assessed with a degree of 
precision great enough to allow much discrimination between 
alternative hypotheses. 

    The greatest risks of mistaken interpretation occur in multiple 
regression analysis where attempts are made to assess the extent to 
which each of several variables affects some index of health; for 
example, the prevalence of respiratory symptoms in a number of 
communities may be studied in relation to several different 
measures of air pollution, to climatic factors, and to the levels 
of cigarette smoking.  In such cases, there is a need to consider 
whether a linear relationship is appropriate for each of the 
variables, but beyond that, if some of those variables included are 
correlated with one another (as is likely with measures of air 
pollution and climate) then the regression coefficients cannot be 
determined with any satisfactory degree of precision, and there is 
a serious risk of overestimating the effect of one variable at the 
expense of another (McDonald & Schwing, 1973). 

    Even when a significant correlation is found between an index 
of health and one or more environmental factors, the relevance of 
this must be considered carefully, for example in terms of 
biological plausibility.  If the number of observations in a study 
is large, a correlation coefficient as low as 0.2 may be 
statistically significant, but it would account for only 4% of the 
variance in the health index, leaving 96% to be explained some 
other way, perhaps in part by environmental factors that were not 
measured.a  In such cases, it may be necessary to consider 
whether the assessments of environmental exposures were adequate, 
or whether the overall effect of any environmental factors may have 
been trivial in relation to that of other determinants. 

    Above all, the fact that correlation does not necessarily imply 
causation must be recognised.  Many unrelated factors exhibit 
similar time trends or geographical distributions, and much 
supporting evidence is required before there can be any presumption 
of causation. 

1.7.  Uses of Epidemiological Information        
    The problem of considering whether a statistical association,
observed between indices of health and various chemical and      
physical agents in the environment, suggests any cause and effect
relationship, is much more difficult than in the case of classical 
epidemiological studies concerned with communicable diseases.  The 
basic difficulty is that few of the non-biological agents have     
unique effects on health, and conversely the effects considered may
often be related to a wide range of factors.  Thus, when decisions 
have to be made about the need for control of suspect agents,    
within industry or in the community at large, many aspects of the
situation may have to be taken into account, such as the strength
and consistency of associations seen in epidemiological studies, 
related toxicological and clinical findings, and economic or social 
implications of control measures.                                   

    Clearly many different disciplines become involved at this     
stage and a full discussion is beyond the scope of the present     
monograph, but this very important facet, which should involve the 
scientist as well as the administrator, is introduced in Chapter 7.

a   In general terms, the proportion of variance explained by
    a regression is r2, where r is the correlation
    coefficient; hence r = 1 (perfect agreement), all the
    variance is explained: for r = 0.2, r2 = 0.04 (i.e. 4%).
    The standard error of a correlation coefficient is
    approximately -- where n is the number of observations.
                         1     1
    Hence for n = 10000, -- = ---  = 0.01 and a coefficient r
                         n  100
    in excess of 0.02 would be significant at the 5% level.


ACHESON, E.D., HADFIELD, E.H., & MACBETH, R.G.  (1967)  Carci-
noma of the nasal cavity and accessory sinuses in wood-
workers.   Lancet,  1: 311-312.

ADELSTEIN, A.M.  (1972)  Occupational mortality: cancer.   Ann.
 occup. Hyg.,  15: 53-57.

COLE, J.F. & LYNAM, D.R.  (1973)  ILZRO's research to define
lead's impact on man.  In:  Environmental aspects of lead,
Luxembourg, Commission of the European Communities, pp.

DEPARTMENT OF THE ENVIRONMENT  (1974)   The monitoring of the
 environment in the United Kingdom.  London, Her Majesty's
Stationery Office.

DOLL, R. & HILL, A.B.  (1950)  Smoking and carcinoma of the
lung.   Br. med. J.,  2: 739.

HANSON, D.R. & FEARN, R.W.  (1975)  Hearing acuity in young
people exposed to pop music and other noise.   Lancet,  2:

KERR, H.D.  (1973)  Diurnal variation of respiratory function
independent of air quality.  Experience with an environ-
mentally controlled exposure chamber for human subjects.
 Arch. environ. Health,  26: 144-152.

LAUER, G. & BENSON, F.B.  (1974)  The CHAMP air quality moni-
toring program.  In:  Proceedings of the International
 Symposium, Recent Advances in the Assessment of the Health
 Effects of Environmental Pollution (Paris).  Luxembourg,
Commission of the European Communities.

MCDONALD, G.C. & SCHWING, R.C.  (1973)  Instabilities of
regression estimates relating air pollution to mortality.
 Technometrics,  15: 463-481.

MINISTRY OF HEALTH (1954)   Mortality and morbidity during the
 London fog of December 1952.  London, Her Majesty's Stationery

MUNN, R.E. (1973)   Global environmental monitoring systems.
Toronto (SCOPE Report 3).

WHITMAN, J.  (1975)  More on monitoring.   Environ. Sci.
 Technol.,  9: 611.

WYNDER, E.L. & GRAHAM, E.A.  (1950)  Tobacco smoking as a
possible etiologic factor in bronchiogenic carcinoma.   J. Am.
 med. Assoc.,  143: 329.

WHO  (1978)   Environmental Health Criteria 6: Principles and
 methods for evaluating the toxicity of chemicals.  Part I.
Geneva, World Health Organization.


2.1.  Introduction

    This chapter is concerned with the type of approach to be used 
in an epidemiological study, starting with exploratory investigations, 
which may be based on existing mortality or morbidity records, on 
general health surveys, or sometimes on quite small-scale clinical 
observations, and are aimed at seeking indications of the role of 
environmental factors in a particular disease or condition.  Such 
investigations may be of value in formulating hypotheses that can 
be followed up by studies designed specially to test them and, 
where appropriate, to try to assess relationships between exposure 
and effect in a quantitative manner. 

    Generally, it is an unusual distribution of disease in a 
locality or a particular population that prompts the enquiry (which 
could be regarded then as "effect-oriented"), though sometimes 
concern arises because of some characteristic of the environment 
that is thought, either on toxicological or more general grounds, 
to have adverse effects on health ("agent-oriented").  In the 
former category, an example is the recent epidemic of a severe 
respiratory and generally debilitating disease in Spain (Tabuenca, 
1981; Aldridge & Connors, 1982).  This affected people over a wide 
range of ages in several parts of the country, and it was at first 
thought to be due to a respiratory infection.  Astute clinical 
enquiries concentrating attention on infants in the first instance, 
because of their more closely confined environments and more 
readily specified diets, revealed that each case was related to the 
use of a particular supply of cooking oil that proved to be 
chemically contaminated.  These initial enquiries constituted the 
exploratory study that generated a hypothesis capable of being 
tested by both toxicological and epidemiological techniques. 

    The investigation of long-term effects of exposure to ionizing 
radiation, following the 1945 atomic bomb explosions in Japan, 
could be regarded as falling in the agent-oriented category.  While 
immediate effects were disastrous and there was every indication 
that survivors would be liable to develop further radiation-induced 
illnesses over the years, the exact nature of the effects and the 
form of exposure/effect relationships were unknown.  A longitudinal 
study, in which defined populations were to be followed through to 
death, was designed to examine these questions, and this is 
referred to in detail in section 

    In the following sections, some of the more commonly used types 
of design in epidemiological studies are described, but it is 
essential to stress that they are not alternatives that can be 
chosen freely for any given situation.  The choice of design 
depends primarily on the questions being asked (the objectives of 
the study) and on constraints imposed by factors such as resources 
available, the time limit within which at least provisional answers 
are required, accessibility of the population to be studied, and 
ethical considerations.  It is vital that a sensible hypothesis, 
supported wherever possible by toxicological evidence, is 

formulated first and the art of good survey design is to reconcile 
conflicts between the ideal and what is possible in a way that will 
maximize the acquisition of useful data. 

2.2.  Preliminary Review of State of Knowledge

    The available literature on the clinical features and natural 
history of the disease or condition being considered, on what is 
known of its causes and distribution in the population, and on 
trends with time, should be critically reviewed.  Often, there are 
conflicting findings between different published studies in the 
field of environmental epidemiology and it is important to try to 
establish which findings can be regarded as reasonably well-

    At the same time, a review is required of information on all 
the relevant environmental factors, including physical and chemical 
properties, possible interactions with other agents, and anything 
known about their spatial and temporal distribution.  Any data 
available on toxicological properties from animal experiments or 
other biological testing procedures also needs to be examined 

    In some instances, where new problems are encountered suddenly 
and immediate action is required, as in the Spanish cooking-oil 
problem cited above, or the Seveso accident in which dioxin was 
dispersed in the vicinity of a chemical works (see section 7.3), 
there may be little prior information on the agents concerned or 
their effects, and, in any case, little time to study it.  Even so, 
it remains vitally important to consider carefully the types of 
epidemiological studies that could and should be undertaken.  A 
false move in the beginning could completely undermine the chances 
of yielding results that would contribute to the identification of 
causal agents, and to the specification of exposure/effect 

2.3.  Descriptive Studies and Use of Existing Records

    Investigations of the general distribution of disease and of 
possible environmental determinants on the basis of existing 
records are referred to as descriptive studies:  they describe the 
situation as it exists in the community, without special efforts to 
investigate symptoms, physiological functions, or exposures to 
particular agents in defined groups.  They may be included among 
the exploratory investigations mentioned above, but they can 
nonetheless be major undertakings in their own right, as in the 
case of the construction of the detailed atlases of cancer 
mortality that have now been prepared in a number of countries 
(Mason et al., 1975; Editorial Committee for the Atlas of Cancer 
Mortality in the People's Republic of China, 1979; Japan Health 
Promotion Foundation, 1981). 

    Although past records frequently suffer from lack of 
reliability, they also have certain advantages and have been used 
not only for descriptive studies but for other types of 

epidemiological studies including retrospective studies and case-
control studies (sections 2.7 and 2.9).  For example, many of the 
diseases and conditions of importance in environmental health 
studies, as in the case of a number of cancers, occur many years 
after significant exposure has taken place.  In these circumtances, 
it is usually wise to consider using information about the effects 
of past exposures as the basis for providing answers to the 
questions of interest. 

    Another advantage of existing records is economics.  In most 
situations, it will be found that the length of time required to 
gather relevant new data would justify some initial investment of 
effort in the study of past records. 

    There are two further reasons why such an approach should 
always be considered.  First, environmental hygiene changes with 
time; recent exposures are generally at lower levels than those in 
the more distant past.  The effects of exposure are likely to be 
more evident in people exposed to higher levels than in those 
exposed to lower levels.  If, therefore, the aim is to seek an 
answer to a preliminary question as to whether or not there is a 
real association between the hazard and the suspected environmental 
agent, then attention must be focused initially on so-called "high-
risk" groups who are most likely to demonstrate an effect, if it 
exists.  The second reason is based on ethical considerations.  
Knowing that a group of people has been exposed to a certain toxic 
substance, it seems incumbent on society to assess the possible 
health effects from such exposures in order to take preventive 

2.3.1.  Mortality statistics

    The routine collection of national mortality data commenced in 
a number of countries in the mid-nineteenth century; for example, 
since 1837, material has been collected for virtually every death 
occurring in the United Kingdom.  The World Health Organization 
(WHO) has been responsible for sponsoring and encouraging the 
collection of accurate mortality statistics throughout the world, 
and the majority of developing countries now have some system for 
the recording, collection, processing, and production of mortality 

    All sets of routine data have disadvantages however.  The 
majority of deaths are certified by the practitioner attending the 
patients, or sometimes by an official responsible for investigations 
in cases of doubt or of violent or unnatural death, which may 
include occupationally-associated disease.  Though many systems 
suffer from delay in data collection, legal requirements to 
register the death and the establishment of registrars responsible 
for handling this material usually result in a steady flow of data 
into the central processing system.  Insofar as autopsy contributes 
to accurate diagnosis, varying rates will affect the validity of 
comparisons between different countries and different periods 
(Moriyama et al., 1966).  Diagnostic vogues and differing vigilance 
may also introduce bias. 

    Waldron & Vickerstaff (1977) have reviewed the subject of the 
accuracy of diagnoses of fatal conditions and the quality of 
certification.  Although a clinician may be clear in his own mind 
about the diagnosis, he does not always record it on the death 
certificate in a way that can be appropriately coded.  For a number 
of years, it has been recognized that death is commonly the result 
of a complex of diseases, and the international system for the 
derivation of a single underlying cause of death from a full death 
certificate can produce unrealistic statistics.  This issue has 
been discussed by a number of authors, for example, Alderson 
(1976).  For all its imperfections, the International Statistical 
Classification of Diseases, Injuries and Causes of Death (WHO, 
1977) is of great value.  It contains definitions and recommendations 
together with rules for medical certification, for the clerical 
coding of primary causes of death and for quality control. 

   If death certificates themselves are used for epidemiological 
purposes rather than the officially published statistics, then the 
person undertaking the coding of cause of death should check his 
performance against that of national coding staff.  It is possible 
to undertake analyses of morbid conditions mentioned on death 
certificates apart from the primary cause of death.  These can be 
of value in studying health service requirements as well as their 
relationship with environmental hazards.  In some countries (e.g., 
Scotland, Sweden, and the USA) it is considered worth coding all 
the conditions mentioned on the death certificates. 

2.3.2.  Morbidity statistics

    A wide range of routine morbidity statistics is now available 
in many developed countries.  These may include data on abortion, 
cancer, congenital abnormalities, hospital inpatients, infectious 
diseases, school health, and sickness absence, including accidents 
at work and occupational diseases. 

    WHO plays a major role in the standardization of morbidity 
statistics.  Various contributions to the World Health Statistics 
Quarterly have discussed aspects of the methods required to 
collect, analyse, and present material on all aspects of health 
care.  A general review of this topic has been published by WHO 
(1965).  Wagner (1976) reviewed 91 projects in 25 European 
countries, concerning processed data on patients discharged from 
hospital in-patient care.  This report provides detailed 
information about the capture, coding, and processing of the data 
but limited indication of how the output from these systems was 
used.  A conference of the Commission of the European Community 
discussed the relationship between health interview surveys, health       
examination surveys, and routinely processed data on hospital 
inpatient discharge records; Armitage (l977) indicated the 
possibilities of international collaboration and the topics for 
which this seemed feasible. 

    Despite the extensive data base on morbidity in a number of 
countries, much care is generally required in using this type of 
information, even for exploratory studies in environmental 
epidemiology.  The records may not provide complete coverage of the 
population and there may be many in-built biases, particularly in 
relation to socioeconomic class.  Thus official sickness/absence 
records show large variations in the apparent extent of illness 
between different occupations, but these are often connected with 
social factors or the amount of physical or mental effort required 
in the job rather than with specific hazards precipitating illness. 

    Data assembled at cancer registries can, however, provide a 
valuable supplement to those obtained from mortality records.  Each 
newly diagnosed case of cancer enters the system and near-complete 
coverage of the population has been achieved in many countries.  
The techniques involved have been reviewed by McLennan and co-
workers (1978).  While both cancer registry and mortality data 
suffer from differences in diagnostic standards and practice that 
make international comparisons difficult, the former avoids some of 
the problems introduced by different treatment regimes in the 
interpretation of mortality statistics, and they are particularly 
valuable for studies on conditions such as skin cancer that have a 
low fatality rate. 

2.3.3.  Populations at risk

    Occasionally, the absolute numbers of deaths or cases of a 
particular disease can be of value in establishing relationships 
with environmental factors without reference to the size or age 
structure of the population at risk.  This is particularly true of 
rare conditions:  for example, the identification of just a few 
cases of angiosarcoma of the blood vessels of the liver was 
sufficient, coupled with experimental animal studies, to 
demonstrate a clear link with occupational exposure to vinyl 
chloride.  Similarly, clusters of cases of mesothelioma of the 
pleura demonstrated links with particular types of fibres 
(crocidolite asbestos, among occupational groups in South Africa 
and elsewhere, and a local volcanic rock with an unusual fibrous 
structure in the case of a village community in Turkey).  Also, the 
proportion of deaths attributed to a certain cause among all deaths 
in a defined group can provide useful clues about environmental 
factors, providing that basic data on sex and age are taken into 
account (section 

    More generally, however, detailed information on the size, sex, 
and age structure of the population at risk is required for the 
proper interpretation of mortality and morbidity statistics.  The 
calculation of appropriate rates is discussed further in section 
6.3.7, and it is necessary here only to stress the importance of 
obtaining adequate information on the denominators (the populations 
at risk) as well as on the numerators (the numbers of deaths, or 
cases of disease). 

    In most countries, complete censuses of the population are done 
at intervals of the order of 10 years, and estimates of changes in 
the intervening period are made from records of births, deaths, and 
migration.  Such records are capable of providing a detailed break-
down by sex and age, not only on the national scale but also for 
individual towns and smaller communities within them.  Even so, 
much care is required in studies confined to small local areas and 
it may be necessary to check or supplement the official data, even 
to the extent of carrying out an unofficial census.  This type of 
approach may, in any case, be necessary in countries where census 
data are incomplete or where internal migration rates are high. 

2.3.4.  Geographical differences in mortality and morbidity

    Contrasts in appropriately standardized mortality and morbidity 
rates (section can be made between countries, or within 
countries between groups characterized by their area of residence 
or any other qualifier (such as ethnic group) that may be included 
on the official records.  These characteristics may, to a limited 
extent, provide a qualitative guide to exposures to environmental 
agents, thus allowing some exploratory studies to be done.  
International comparisons are, however, fraught with difficulties, 
due to differences in diagnostic practice or other factors.  For 
example, in the 1950s, mortality from bronchitis was about 25 times 
higher in Scandinavia than in the United Kingdom.  It was suspected 
that this was partly an artifact of definition, and it led to 
studies on variations between countries on the certification of 
bronchitis and emphysema on death certificates (Fletcher et al., 
1965).  In this particular case, it appeared that while differences 
in terminology and in rules for assigning cause of death explained 
quite a large part of the difference in mortality between the 
United Kingdom and other countries, environmental factors probably 
also contributed.  To pursue this question further, however, it was 
necessary to set up specially designed studies (Holland et al., 

    In general, geographical contrasts between areas within a 
single country are likely to be less than those between countries, 
but they can be more revealing in relation to environmental 
influences.  Possibly, one of the most exciting intracountry 
variations hitherto uncovered is the 30-fold difference in 
oesophageal cancer risk for women in different areas along the 
Caspian Littoral of Iran where, in the high incidence areas, this 
form of cancer, generally rare in females (Kmet & Mahboubi, 1972), 
is two to three times commoner than the relatively high incidence 
of breast cancer in North American and European women. 

    It is not only in developing countries that such variations are 
to be found.  In England, stomach cancer is 50% commoner in 
Liverpool than in Oxford.  While some of the differences 
demonstrated in the recently published maps of cancer morality, 
referred to at the beginning of section 2.3, will turn out, when 
examined closely, to be due to artefacts, others will prove to be 
real and suitable for study. 

    Some of these contrasts can be linked with differences in 
social class distribution between areas, implying effects of broad 
environmental factors related to lifestyle, to concentrations of 
recent immigrants or ethnic groups or to the selective migration of 
relatively fit members of the community in or out of the areas 

    Studies based on routinely collected mortality and morbidity 
data usually have to be confined to comparisons based on area of 
residence at the time of death or of occurrence of the illness in 
question, and this is a limiting factor in studies on chronic 
diseases, particularly in countries with high internal migration 
rates.  However, in the case of migrants between countries, 
official records of country of origin are often maintained, and it 
is possible to compare the experience of migrants with that of 
their compatriots in both the country of origin and that of 
subsequent residence.  This sheds some light on the relative roles 
of environmental and genetic factors in the development of disease. 

    The migrant exchanges one environment and its associated 
exposures for another.  If the international differences in various 
disease risks observed are due to genetic factors, then incidence 
should not be influenced by migration.  Yet, as the pioneer studies 
of Haenszel & Kurihara (1968) have shown, cancer morbidity and 
mortality rates in migrant populations gradually come to approximate 
those of the host country. 

2.3.5.  Time trends

    Long-term trends with time in the mortality or morbidity rates 
for specific diseases can be of value in indicating possible 
effects of environmental factors, though interpretation is 
complicated by the effects of improvement in therapeutic treatment, 
etc.  There has, for example, been a massive decline in mortality 
from pulmonary tuberculosis in most developed countries during the 
present century.  It is difficult however to separate out all the 
factors responsible:  much of the decline occurred before the 
really effective treatment by antibiotics became available, and, to 
some extent, it can be attributed to environmental factors in the 
broadest sense, i.e., to improved housing and social conditions and 
to better medical care generally. 

    In many countries, the incidence of cancer of the breast, lung, 
pancreas, and prostate is rising.  It has been suggested, particularly 
for lung cancer, that these increases are artefactual, being due to 
better diagnosis, changes in classification, etc. (Percy et al., 
1974).  While such factors probably have had some influence, it is 
very difficult to believe that for an organ as accessible as the 
breast they explain more than a small proportion of the observed 
increase.  The increase in malignant melanoma of the skin, a very 
accessible cancer, has been carefully investigated by Magnus (1973) 
and others who conclude that the rise is real.  In the United 
States of America, cancer of the oesophagus has doubled in persons 
of Negroid origin, since 1935.  Nonetheless, it is worth while 
remembering that were it not for tobacco-caused lung cancers, the 

overall cancer mortality in the USA for Caucasian males would be 
falling and that for Caucasian females, the overall cancer 
incidence is falling slowly (Devesa & Silverman, 1978). 

    When examining trends over an extended period, it is always 
important to ensure either that sex/age specific rates are used or 
that the data are standardized with respect to age (section, since there have been considerable changes in the age-
structure of the population in most countries during recent 
decades.  Sometimes, contrasts in trends between men and women can 
provide clues about the factors responsible, as in the case of lung 
cancer, for which death rates began to increase sharply sooner in 
men than in women (consistent with an effect of cigarette smoking). 

2.3.6.  Associations with environmental indices

    Apart from the general guidance that can be obtained from the 
examination of geographical differences and trends in mortality and 
morbidity, it is often possible to use observations on dietary 
factors, or on air or water pollution, etc. to carry out further 
descriptive studies. 

    For example, a large number of studies concerned with 
associations between mortality and routine observations of urban 
air pollution have been reviewed by Holland and co-workers (1979).  
While most of these indicate positive correlations with measurements 
of pollutants such as smoke, total suspended particulates or sulfur 
dioxide, there is probably an interaction with other confounding 
factorsa not taken into account, notably tobacco smoking.  These 
initial studies were however valuable as exploratory ones, leading 
to the development of studies designed specifically to test the 
hypothesis that exposure to urban air pollution contributes to the 
development of chronic respiratory disease. 

2.3.7.  Case registers                        
    As mentioned in section 2.3.3, it is sometimes possible to  
identify environmental agents related to the development of     
relatively rare conditions, simply from the clustering of a few 
cases in local areas or in particular occupations.  It is seldom 
possible to recognize associations between common exposures and 
common conditions in this way, but one effect-oriented approach is 
to establish case registers through hospitals and/or general  
practitioners for selected conditions for which there is already 
some indication (e.g., an irregular geographical distribution)   
that environmental factors may play a part.  It may then be      
possible, through careful enquiry into domestic and occupational 
histories, to identify some common factor that can be followed up 
further with additional epidemiological and toxicological        

a  Defined in section

    In developing countries, careful appraisal of a wider range of 
cases and their associated histories can, however, help to provide 
background information in the absence of comprehensive official 
statistics.  Even so, with the 3000-5000 people for whom a single 
primary health care worker may be responsible, the wide random 
fluctuations in morbidity or mortality rates that would be likely 
to occur, would have little real meaning, and it would probably be 
necessary to assemble information at a district or provincial level 
in order to seek evidence of unusual local patterns of disease 
(WHO, 1982). 

2.3.8.  General surveys

    While survey techniques, considered in greater detail in 
subsequent sections of this chapter, form an essential part of most 
of the study designs, in many countries, regular surveys of the 
population, made for administrative purposes, can be of value as 
exploratory studies in relation to environmental factors.  Thus, in 
the United Kingdom, there is a General Household Survey that 
enquires into family expenditure on foods, etc., and within this, 
questions are asked on recent illnesses.  There are possibilities 
of adding additional questions on matters that may affect health 
and, in this way, data have been obtained that could be used in 
conjunction with mortality records to demonstrate strong 
interrelationships between smoking and occupation and, in turn, 
with lung cancer mortality (Office of Population Censuses and 
Surveys, 1978).  The application of information from other types of 
surveys is discussed further in section 4.2.2. 

2.4.  Formulation of Hypotheses

    Studies are most likely to be productive if they are based on 
clearly stated hypotheses.  These can be developed from the results 
of various descriptive studies, as discussed above.  Basically, 
this is to try to demonstrate an association between carefully 
specified effects on health and assessments of exposure to 
specified environmental agents.  Epidemiological studies cannot by 
themselves prove that a particular agent causes a particular health 
effect; they may, however, demonstrate quantitatively the strength 
of an association between the presence of the agent and the 
occurrence of the hypothesized effect.  Appropriate statistical 
analyses may in turn determine the probability that an association 
as strong as that observed might have occurred by chance (section 
6.4.1).  Whether the correct agent has been identified or whether 
the apparent association has arisen artefactually, because of 
correlations with exposure to other agents or factors that were not 
studied, is a question requiring further epidemiological studies 
and, where possible, also toxicological work. 

    Most investigations in the field of environmental epidemiology 
are necessarily of an observational nature, that is, they are 
observations based on existing situations.  Associations can be 
demonstrated most clearly if it is possible to compare groups 
exposed to several levels of the agent in question, but, in the 

last resort, hypotheses about the exact form of exposure/effect 
relationships can be tested effectively in experimental situations, 
where the research worker has some control over exposures. 

    While the working hypothesis must be as simple as possible, it 
has to be recognized that causes of ill-health are commonly multi-
factorial, and that the environment, though it comprises many 
individual components, acts as an entity, having effects liable to 
be greater than the total of those of the components.  It may be 
that, in the subsequent statistical analysis, a complex variable 
can be developed to describe the combined effects of exposures to a 
range of different agents as measured within the study (Cassell & 
Lebowitz, 1976), but such ideas are difficult to incorporate into 
the initial hypotheses. 

    It may be helpful to view the formulation and testing of 
hypotheses in environmental epidemiology as an example of the 
essentially iterative process of science, which comprises an 
initial (or crude) hypothesis, assembling of data from available 
sources or from planned investigations, testing of the validity of 
the hypothesis, rejection of the hypothesis leading to its revision 
or refinement, and the further assembling of data to test a revised 

    The main types of study designs in environmental epidemiology 
and some of the salient features of each are presented in Table 
2.1, and described further in the sections below. 

2.5.  Cross-sectional Studies

    Cross-sectional studies, sometimes called prevalence studies, 
provide information on disease frequency (prevalence) at a given 
time.  Estimates of exposures, and measurements of personal 
characteristics and biological effects may be made at the same time 
or may be derived from existing records.  Thus, for example, an 
investigator might pose the question:  Are small opacities on a 
chest radiograph more often found in welders than in other men (of 
the same age)?  He might attempt to answer this question by 
obtaining 1000 chest radiographs of welders and 1000 chest radio-
graphs of other men.  After mixing the films to ensure blinding 
with respect to which film was of a welder and which of a non-
welder, the 2000 films would be examined and categorized by two 
independent readers and then the changes observed would be 
compared, between welders and the non-welders, within 5- or 10-year 
age groups.  This would be a pure cross-sectional study.  In 
practice, it is seldom that a cross-sectional study is so precisely 
limited with respect to time. 

    Usually, historical information is collected so that a 
retrospective component is included in the study.  Thus, information 
would be collected on past as well as current smoking habits, an 
occupational history would be taken, comprising details of all jobs 
held since leaving school, and often residential details of each 
community in which the subject had lived, and dietary information 
and data on any other present and past exposures of potential 

significance would be obtained.  On the disease side of the 
equation, attempts are often made to establish the time of onset, 
mode of development and course of disease, and any relevant 
antecedent conditions.  Thus, although the information may be 
collected at one time, it often refers to events that may have 
taken place over a period of years.  Hospital records, information 
from physicians about past episodes of disease, and any measurements 
that may have been made on relevant environmental factors may be 
used, if they are likely to contribute useful information to the 

(a) Choice of population

    Cross-sectional studies are often designed to compare the 
prevalence of disease in different places and in different groups 
of people according to their measured, assessed, or surmised 
exposures.  The two most common population types that need to be 
considered are:  the general population, comprising the whole 
community or some segment of it, based on age, sex, and race; and 
the occupational group.  The former will usually be more 
appropriate to the investigation of wider community exposures (air 
pollution, water quality and contaminants, effects of hot or cold 
weather, neighbourhood pollution from some plant or factory).  
Sometimes, the families of workers may be exposed to pollutants of 
industrial origin not only because of local emissions, but also 
through dust being brought home on the workers' clothes.  Many 
studies concentrate on the health of children (for example, Golubev 
et al., 1979; Dantov et al., 1980); apart from the importance of 
this topic in its own right, where concern is primarily with 
general environmental agents, the confounding effects of 
occupational exposures and of smoking can be minimized in this way.  
Among adults, a single occupational group may be chosen for the 
investigation of community problems, in order to avoid interference 
from specific occupational factors. 

Table 2.1.  Major features of various study designs in environmental epidemiology
Study      Population      Exposure     Health effect   Confounders    Problems           Advantages               
design                                                  are:                                             
Descrip-   Various         Records      Mortality and   Difficult      Hard to establish  Cheap, useful            
tive       sub-            of past      morbidity       to sort        cause-result and   to formulate             
study      populations     measure-     statistics,     out            exposure-effect    hypothesis               
                           ments        case regist-                   relationships                               
                                        ries, etc.                                                       
Cross-     Community       Current      Current         Usually easy   Hard to establish  Can be done              
sectional  or special                                   to measure     cause-relation-    quickly; can             
study      groups;                                                     ship; current      use large popu-          
           exposed vs.                                                 exposure may be    lations; can              
           non-exposed                                                 irrelevant to      estimate extent
           groups                                                      current disease    of problem                    
Prospec-   Community       Defined at   To be deter-    Usually easy   Expensive and      Can estimate             
tive       or special      outset of    mined during    to measure     time consuming;    incidence and  
study      groups;         study (can   course of                      exposure cate-     relative risk;           
           exposed vs.     change dur-  study                          gories can         can study many 
           non-exposed     ing course                                  change; high       diseases; can            
           groups          of study)                                   dropout rate       infer cause-             
                                                                                          result rela-             

Retro-     Special         Occurred in  Occurred in     Often          Changes in         Less expensive           
spective   groups such     past - need  past - need     difficult      exposure/effect    and quicker      
cohort     as occupa-      records      records of      to measure     over time of       than cohort  
study      tional groups,  of past      past diagnosis  because of     study; need to     prospective          
           patients,       measure-     and measure-    retrospective  rely on records    study giving     
           and insured     ments        ments           nature (e.g.,  that may not be    similar                        
           persons                                      past smoking   accurate enough    response, if   
                                                        habits)                           sufficient             
                                                                                          past records          
                                                                                          are available

Table 2.1.  (contd.)
Study      Population      Exposure     Health effect   Confounders    Problems           Advantages
design                                                  are:
Time-      Large com-      Current,     Current,        Often          Many confounding   Useful for     
series     munity with     e.g., daily  e.g., daily     difficult      factors, often     studies on   
study      several mil-    changes in   variations in   to sort out,   difficult to       acute effects
           lion people;    exposure     mortality       e.g., effects  measure
           susceptible                                  of influenza
           groups such
           as asthmatics

Case-      Usually small   Occurred     Known at        Possible to    Difficult to       Relatively     
control    groups;         in past      start           eliminate      generalize         cheap and
study      diseased        and deter-   of study        by matching    due to small       quick; useful 
           (cases) vs.     mined by                     for them       study group;       for studying 
           non-diseased    records or                                  some incor-        rare diseases
           (controls)      interview                                   porated biases

Experi-    Community       Controlled/  To be measured  Can be         Expensive;         Well accepted 
mental     or special      known        during course   measured;      ethical            results; strong
(inter-    groups                       of study        can be         consider-          evidence for
vention)                                                controlled by  ation study        causality
study                                                   randomization  subjects'  
                                                        of subjects    compliance 

Monitor-   Community       Current      Current         Difficult      Difficult to       Cheap when 
ing and    or special                                   to sort out    relate exposure    using existing 
surveil-   groups                                                      data with          monitoring and
lance                                                                  effects            surveillance    
(b) Assessment of exposure and effects on health

    The index of occurrence of disease in a cross-sectional study 
is prevalence, or the prevalence rate, i.e., the number of persons 
in the group who are affected, expressed as a proportion of the 
total number in the group.  For physiological or biochemical 
variables, the average and the distribution are the parameters of 
interest (section  However, as in the case of exposure, 
some assessment of the onset, development, and progression of the 
effect may be obtained from judicious questioning; available 
information may also be sought from records. 

(c) Confounding variables

    It is not possible to list all the confounding factors that 
need to be considered.  These will vary from study to study 
according to the condition under investigation and, in many cases, 
it may not be possible to avoid confounding factors entirely.  
However, it is necessary to ensure that potential confounding 
variables are identified at the design stage and that all the 
available information on them is recorded.  Unless a single sex/age 
group is being examined, it may be necessary to ensure that the 
contrasting groups that are being selected for study have similar 
age and sex distributions, by stratified random sampling.  For age, 
10-year groups are sufficient for most purposes.  Smoking has been 
found to be such an important factor in so many of the effects 
likely to be investigated, that it should always be recorded.  Some 
index of social circumstances, number of years of education, 
occupation, type and quality of housing, degree of overcrowding and 
so on should often be included.  Other factors will need to be 
considered in certain studies though not necessarily in all.  In 
short, the appropriate attention to confounding factors can only be 
given if the epidemiological and other knowledge about the causation 
of the effect of interest is carefully reviewed before and during 
the design stage. 

(d) Analysis

    In many parts of the world, only limited and non-specialized          
statistical help may be available for research workers.  The 
absence of elaborate statistical facilities should not deter would-
be researchers from undertaking prevalence surveys.  Full exploitation 
of results from such studies may require the application of fairly 
complex methods, but important new knowledge about relationships 
between environmental factors and indices of health can be 
established without sophisticated statistics.  The essential 
requirements are:  attention to the principles of study design 
mentioned above; conscientious adherence to protocols and survey 
methods (chapter 5); and careful description of the results, as 
discussed in section 6.3. 

(e) Advantages and disadvantages

    A cross-sectional study may provide the answers to many 
questions.  Thus, this method has been extensively used to compare 
the prevalence of respiratory symptoms and levels of lung function 
in different groups of people, living in different places and 
working in different jobs with various potential levels of 
exposure.  Prevalence studies have been used to study such diverse 
chronic conditions as rheumatoid arthritis, asymptomatic bacteriuria, 
diabetes mellitus, hypertension, peptic ulcer, stroke, and coronary 
disease.  In the occupational setting, cross-sectional studies of 
exposure to chemicals, dusts, fumes, and gases have often provided 
valuable information to guide decisions on permissible levels of 
different substances in the workplace.  The threshold limit value 
for mercury in the workplace, for example, was initially based on a 
cross-sectional study (Neal et al., 1937) and, in the absence of 
new relevant data, remained unchanged for 25 years.  Cross-
sectional studies were also the basis for standards of cotton dust 
in the workplace (Roach & Schilling, 1960). 

    Thus, despite some difficulties of interpretation, as discussed 
below, determination of the prevalence of a disease in groups at a 
particular time may give important information required for 
preventive action.  In any case, a cross-sectional study is a 
necessary prerequisite for any longitudinal or prospective study.  
Thus, if the incidence of a disease (i.e., the rate of occurrence 
of new cases) is to be measured, it is essential to identify persons 
who already have the disease in question. 

    Difficulties may arise because of selection within groups.  
Much publicity has been given to the so-called "healthy worker 
effect" in occupational health studies, but there is a danger that 
this will lead to the underestimation of risk in some cases.  How-
ever, this is only one of several population-selection artefacts 
that may occur (Fox & Collier, 1976). 

    It should be noted that certain jobs preferentially attract 
persons who may be less fit than the average. In the 1950s, the 
attraction to the boot and shoe industry of the tuberculous worker 
was noted by Stewart & Hughes (1951).  Selection may occur within 
occupations.  In coalmining, fitter men may work in dustier jobs 
where the pay is higher, disabled miners may leave the coalface and 
work on haulage or eventually take up lighter jobs on the surface.  
Disabled workers may, of course, also leave the industry altogether 
and consequently will not be included in a prevalence study.  The 
impact of these movements may be hard to detect in a cross-
sectional study.  Thus, an early study of lung cancer in relation 
to chromate manufacture, based on a cross-sectional study (Bidstrup 
& Case, 1956), failed to reveal any increased risk of cancer in 
relation to chromate exposure, whereas a subsequent prospective 
survey revealed an increased risk. 

    There are important selective factors within local communities 
that also have a bearing on the design of cross-sectional studies.  
Apart from the "polarization" of different social classes into 
different parts of a town, there is a tendency for the less fit to 
be left behind in the less favoured areas as others move out.  In 
the rapidly growing cities in developing countries, new residents 
may gather in particular areas, and it has been noted that migrants 
into cities are affected more by urban pollution than are the 
earlier residents, who may have become adapted to it. 

2.6.  Prospective and Follow-up Studies

    These two types of study may be considered together, though 
conceptually they differ to some extent.  In prospective studies, 
study subjects are observed over a period of time according to the 
study protocols that are set out at the start of a study.  In a 
follow-up of a cross-sectional study, the original findings may be 
analysed in greater depth using additional information that has 
become available.  However, in a follow-up study, unlike a 
prospective study that is planned as such from the start, it may 
not be possible to follow all of the procedures used during the 
cross-sectional study itself.  In the discussions that follow, 
reference is made only to "typical" prospective studies. 

    Prospective studies permit the investigator to measure the 
rate of development (incidence), the rate of deterioration 
(progression or complications), the rate of improvement 
(remission), and the rate of mortality of the disease.  Repeated 
measurements of functions of various organs will reveal how these 
are changing over time.  Studies of this kind have been carried 
out, for example, on chronic respiratory diseases such as chronic 
bronchitis, emphysema, and pneumoconiosis, and on hypertension 
with particular reference to the factors influencing the level of 
blood pressure and its change over time. 

(a) Choice of population

    Prospective studies can be carried out on the general community 
or some special subpopulations.  Examples include the Framingham 
Heart Study in Massachusetts (Gordon & Kannel, 1970), the Tecumseh 
Community Health Study in Michigan (section, the Atomic 
Bomb Casualty Commission's study of survivors in Hiroshima and 
Nagasaki (section, the investigation of a number of 
diseases by Cochrane and his colleagues in the Rhondda fach and 
Vale of Glamorgan (Cochrane, 1960), the studies of air pollution in 
New Hampshire by Ferris and his colleagues (1973), in the 
Netherlands by Van der Lende and co-workers (1973) and Douglas & 
Waller (1966) and the studies of respiratory disease in Arizona 

    For reasons of economy, prospective studies have often 
exploited the potential opportunities of data from occupational or 
insured groups, or from rosters of patients who have been treated 
in some manner that may possibly raise questions about untoward 
side-effects later on.  Examples of prospective studies using

occupational groups are the study on British doctors of smoking in 
relation to respiratory cancer and other causes of death (Doll & 
Peto, 1976), the studies of coronary heart disease such as those of 
Stamler and co-workers (1975) and Doyle and co-workers (1957), and 
the studies of cardiovascular and respiratory diseases by Fletcher 
& Tinker (1961).  Prospective studies focusing on patients include 
the studies of cancer in children treated by thymus irradiation, 
and leukaemia in persons with ankylosing spondylitis treated with 

    The British Pneumoconiosis Field Research on coalminers 
provides one of the best illustrations of a prospective study 
designed to investigate the influence of occupational exposures on 
various respiratory conditions in coal workers (Jacobsen, 1981).  
Briefly, a sample of 24 collieries in England, Scotland and Wales 
were selected for the study.  All the men employed in these 
collieries were examined by a respiratory-symptoms questionnaire, 
spirometry, anthropometry, and chest radiography on several 
occasions over 20-year periods.  Dust sampling was carried out in 
the coalmines in order to be able to estimate a cumulative dust 
exposure for each man.  These dust measurements were related to 
various indices of disease derived both from the initial cross-
sectional data and from the longitudinal findings.  In this way, 
the most accurate estimate was made of the influence of coalmine 
dust exposure on respiratory conditions (bronchitis, lung function, 
pneumoconiosis, and mortality) that is ever likely to be attempted.  
The paper by Jacobsen illustrates many of the more interesting 
features of this work, and his summary of the way that the study 
developed is reproduced in Table 2.2. 

(b) Choice of controls (or comparison group)

    For prospective studies, either external or internal controls 
may be chosen.  The general population or a particular segment of 
it is often used as an  external control.  The mortality or
morbidity experienced by members of the population (usually 
specific for age, sex, and race) over the period of observation 
becomes the standard to which the  observed  mortality or morbidity 
of the cohort is compared.  In prospective studies of occupational 
groups, the use of the general population as a control group 
introduces a bias commonly known as the "healthy worker" effect.  
This selection bias appears to be higher for long-term chronic 
conditions, such as hypertension and rheumatic heart disease, than 
for diseases having a fairly short duration and no early warning 
signs, but the effect is detectable also for malignancies, 
including respiratory cancer (Fox & Collier, 1976).  The ideal 
controls would be individuals similar in every respect to the group 
under study, except for exposure to the agent of interest.  For 
example, workers in the same industry or factory, who are not 
exposed to the agent in question, often serve as  internal controls  
for a cohort of workers who have been exposed to the agent. 
Measurements of different cumulative exposures for individuals or 
subgroups in a cohort constitute the most effective internal 
control and lead directly to estimates of exposure/effect 

(c) Assessment of exposure

    In a carefully planned prospective study, exposure is measured 
at the start and periodically afterwards.  The most appropriate 
methods can be used and checks to ensure good quality control can 
be incorporated into the design. 

(d) Assessment of effects

    Since, in prospective studies, the decision on diagnostic 
criteria is taken at the start of a study, the investigator has 
ample opportunity to specify these with precision and to take due 
precautions to ensure that they are applied in a uniform and 
standardized manner.  Any manifestations that may indicate an 
earlier stage in the development of the condition of interest can 
also be recorded.  Identification and categorization of persons 
with disease in a prospective study takes place after they have 
been categorized with respect to exposure but the time varies.  It 
is clearly desirable that, as far as possible, investigators 
categorizing the population with respect to disease should not be 
aware of the particular exposure category of any subject. 

Table 2.2.  Progress and development of the pneumoconiosis field researcha
1953               1958               1963               1968               1973
1st surveys        2nd surveys        3rd surveys        4th surveys        5th surveys

24 collieries      24 collieries      24 collieries      10 collieries      16 collieriesb

31 629 miners      21 849 (69%)       14 888 (47%)       4 077 (13%)        5 709 (18%)
                   of original group  of original group  of original group  of original groupb
(+477 others from  (+8 463 others)    (+11 649 others)   (+6 311 others)    (+5 755 others)
 a 25th colliery)
a   From: Jacobsen (1981).
b   Including some ex-miners seen in the "Follow-up" surveys.  Complete radiological and dust 
    exposure data available for 2 600 (8%) of the original group at 10 callieries.

Note:   1.  Radiography and interviews on previous occupational history at all surveys.
        2.  Records of attendance in occupational groups kept throughout.
        3.  Spirometry, anthropometry, and questionnaire on respiratory symptoms and
            smoking habits at 2nd and subsequent surveys.
        4.  More complex lung function measurements in sample at 4th and 5th surveys.
        5.   Dust sampling in occupational groups:
            1952 With Thermal Precipator.
            1965 With Gravimetric Sampler.
        6.  1971 Study of mortality in a (56%) sample of men seen at the 1st surveys.
        7.  1974 Start of follow-up surveys of survivors in the same sample (miners and ex-miners).
        8.  1977 Extension of mortality study to include all (31 629) miners seen at 1st surveys.

(e) Confounding factors

    The important point is to consider and record necessary 
information on any confounding factors.  A review of the 
etiological factors should be carried out before starting the study 
and a thorough check of the protocol should be made to ensure that 
information on important potential confounding factors has not been 

    One particular problem in prospective studies, liable to affect 
the assessment of both exposure and effects, is the tendency for 
methods to change as technology progresses.  Changes may have to be 
resisted if bias is to be avoided.  At least the effects of such 
changes must be investigated in carefully designed comparative 

(f) Exposure/effect

    With a carefully performed prospective study it will be 
possible to establish relationships between exposure and effect.  
If measurements are made early enough in life, a study of this kind 
provides perhaps the best estimates of risk based on lifetime 
exposures.  The study of effects of air pollution on the health of 
children carried out by Douglas & Waller (1966) is a good example.  
Had this been directed initially at air pollution instead of

ingeniously exploiting a set of data as an afterthought, the 
exposures might have been better measured. 

(g) Advantages

    Prospective studies, if properly conducted, may provide 
measures of incidence, estimates of relative risk and inference 
about cause/effect relationships with greater confidence than most 
other types of epidemiological investigations. 

(h) Disadvantages

    Prospective studies are usually very expensive and time-
consuming.  Loss of study participants in a follow-up is another 
serious problem.  A follow-up of persons who left an industry can 
usually be done only with considerable effort.  Changes in the 
quantity and quality of exposure over time have to be taken into 

2.7.  Retrospective Cohort Studies

    When data are available from observations and/or measurements 
that have been made in the past, it may be possible to design a 
study that avoids the long waiting time of a prospective study.  
This is often the case in industry, where records may have been 
kept of all the departments in which employees have worked and also 
of the actual job held since the worker was recruited into the 
industry.  Examples include the studies of cancer of the urinary 
bladder in chemical and rubber workers (Case et al., 1954), cancer 
of the lung in smelter workers (Lee & Fraumeni, 1969), cancer of 
the respiratory system in chromate workers (Bidstrup & Case, 1956), 
and mortality from all causes in miners and millers of asbestos in 
Quebec (McDonald et al., 1971).  Insured persons often provide a 
good opportunity for studies of this kind. 

    The same principle has been applied for epidemiological studies 
concerning side-effects of therapies and diagnostic procedures in 
groups of patients.  For example, the relation between radiation 
and breast cancer has been studied in patients with pulmonary 
tuberculosis; patients with tuberculosis, who had been treated with 
isoniazid, and mental hospital patients, who had received pheno-
barbital, have both constituted cohorts for the study of possible 
relationships between the use of these drugs and the incidence of 
bladder cancer. 

    Sometimes, material collected during the course of a prospective 
study may be stored for future analyses, should a hypothesis that 
was not included in the original plan subsequently appear worth 
investigation.  Materials may also be stored for subsequent testing 
in the interest of economy.  In a study of viral infections in 
pregnancy in relation to subsequent congenital malformations, Evans 
& Brown (1963) collected and stored sera during pregnancy.  
Virological tests were carried out later, if the child was born 
with a congenital malformation.  Similar methods have been used for 
the storage of blood samples for subsequent analysis, should 

questions become relevant later in continuing studies of coronary 
heart disease.  Other samples such as food may also be stored for 
subsequent analysis. 

(a) Assessment of exposure

    Assessment of exposure in a retrospective study is dependent on 
the subject's memory and reliable past records.  For those who have 
died, some information will have to be obtained from a proxy.  Its 
quality will inevitably be more questionable than that obtained 
from the subject himself, and means of checking the validity of 
such proxy information should be incorporated into the study 

(b) Assessment of effects

    Usually reliance will be placed on mortality.  Valid morbidity 
data were seldom available in the past with a few exceptions from 
occupational health studies such as Morris and his colleagues' 
study of coronary disease in the transport industry in the l950s 
and 1960s (Morris et al., 1966).  Many industries are now 
collecting morbidity information in a way that should provide 
usable diagnostic data (Pell et al., 1978) and, as discussed in 
section 2.3.2, various morbidity statistics may be available in 
more developed countries. 

(c) Confounding factors

    Information on factors such as smoking and social class is 
often not available from existing records.  Sometimes, it is 
possible to remedy the gap, but the effort required is time-
consuming, and the reliability of proxy information about those 
who have died may be questionable. 

(d) Advantages/disadvantages

    This approach is generally much less expensive and quicker than 
a prospective cohort study.  However, as mentioned above, a 
retrospective study relies entirely on past records, which usually 
do not provide precise information.  It is therefore seldom 
possible to extract a valid quantitative exposure/effect relation-
ship.  Methods may have changed so that past and present exposures 
may be hard to combine.  Usually a qualitative relationship is all 
that is possible (Lee & Fraumeni, 1969).  A notable exception is 
the study of asbestos miners and millers in Quebec (McDonald & 
McDonald, 1971), though even here the study shows the problems 
introduced by changes in measurement methods for the asbestos 
exposures (Health and Safety Executive, 1979). 

2.8.  Time-series Studies

    When exposure to some environmental hazard varies substantially
over short periods, it may be particularly useful to observe how 
this variation affects some biological effect.  Ambient temperature 
varies from day to day.  Does this have any effect on mortality or 

morbidity?  Does it affect symptomatology or functional capacity?  
Thus, a study in which daily temperatures and daily changes in the 
number of deaths or cases of illness, or in the values of some 
physiological function are compared, might be envisaged.  Such 
investigations have been used most effectively to study the acute 
effects of exposure to air pollution.  For example, daily mortality 
and hospital admissions data were related to daily concentrations 
of smoke and sulfur dioxide and to weather by Martin & Bradley 
(1960) and Martin (1964).  This type of study is effective only 
when it involves large communities of several million people, 
presumably because the contribution of air pollution to day-to-day 
variations in mortality is relatively small compared with that of 
the other factors that determine death or would lead to hospitalization.  
Simple procedures for collecting self-recorded information on the 
health of bronchitic patients using pocket diaries have also proved 
valuable in establishing relationships between exacerbations of 
their illness and air pollution (Lawther et al., 1970).  There are 
advantages in concentrating attention on particularly sensitive 
groups in studies of this kind, as mentioned above. 

(a) Confounding factors

    Many factors influence daily mortality and morbidity.  For 
example, in studies of the effects of air pollution, temperature, 
humidity, and other climatic variables are important as they affect 
both air pollution levels and health indices.  Either extremely 
high or low temperatures may be lethal, thus posing considerable 
problems in the analysis and interpretation of effects of pollu-
tion.  Epidemics of communicable diseases such as influenza could 
be troublesome confounding factors.  Ethnic group or sex, major 
confounding factors in most epidemiological studies, would not be 
great problems in time-series studies, since day-to-day changes in 
the relative distributions of these variables among subgroups under 
study are likely to be small.  However, problems may arise if 
studies persist over many years, because the effect of differential 
migration may then be considerable. 

2.9.  Case-control Studies

    The focus of a case-control study is on a disease or on some 
other condition of health that has already developed.  The 
questions asked relate to personal characteristics and antecedent 
exposures which may be responsible for the condition studied.  In 
particular, the investigator wishes to determine if the environmental 
exposures of those who have the condition of interest differ from 
those of persons who do not. 

    Such studies are relatively cheap and quick, but they depend on 
the ability of cases and controls to recall information on past 
habits and exposures, often in a quantitative manner, or on the 
availability of relevant records. 

    When the accumulation of cases and controls extends over a 
lengthy period, then the data available for study may include a 
variety of genetic, immunological, biochemical, virological, and 
serological measurements, but apparent differences between cases 
and controls may be due to the presence of the disease and cannot 
be interpreted as indicating a causal relationship. 

    Case-control studies can be indicative and economical when the 
suspect agent is distributed in say 50-70% of the population, and 
the hypothesized effect is relatively rare.  On the other hand, if 
cases occur frequently in the population being studied and the 
suspected agent is only one of several causal factors, then it may 
be difficult to establish an association using the case-control 
approach.  In general, apparent associations in case-control 
studies need to be confirmed, in the same as well as in different 
settings, before they can be interpreted as indicating a causal 
relationship (see also the discussion in section 6.5.6 and Crombie, 

(a) Population for study

    By definition a case-control study involves two populations - 
cases and controls.  The problem is to ensure that the particular 
cases and controls that are studied are representative and unbiased 
samples from these populations.  The majority of case-control 
studies have been based on patients in or attending hospitals.  For 
diseases where most patients have to undergo diagnosis at hospital, 
this is obviously a suitable method for identifying cases.  It has 
been used effectively for studies of many cancers and for other 
serious conditions such as cirrhosis of the liver, lupus erythematosis, 
and congestive heart failure.  However, if most patients do not 
have to go to hospital as in the case of, for example, chronic 
bronchitis, maturity-onset diabetes, hypertension, etc., then 
focusing on hospitalized patients will bias any conclusions. 

    If patients are to be obtained from hospitals, then all 
hospitals in a geographically defined area should be included, so 
that comprehensive and unbiased coverage is ensured, for many 
hospitals cater for particular segments of the population. 

    Should all patients with the disease be included or should the 
focus be on newly-diagnosed cases?  The answer to this may depend 
on the condition under study.  Chronic long-term disease can 
perhaps be adequately studied by considering all cases, but it is 
usually recommended to take newly diagnosed cases; their recall is 
better and their exposure history is less altered by the presence 
of disease. 

    Patients with conditions of interest may be obtained from other 
sources.  For example, cancer cases can be drawn from a cancer 
registry, birth defect cases from a malformation registry, etc.  
Such sources are often more likely to be representative than 
patients obtained from a sample of hospitals.  Cases of rare fatal 
disease have sometimes been identified by writing to all pathologists 
in a particular area.  Studies on mesothelioma, for example, have 
been made in this way (McDonald & McDonald, 1971). 

(b) Source of controls

    Hospital controls, matched for relevant characteristics, have 
often been used.  In their early study of smoking and lung cancer, 
Doll & Hill (1952) used persons with other cancers as one control 
group.  They also included a group of hospital patients with 
diseases other than cancer who were matched for age, sex, and 
hospital as a second control group.  Hospital controls are 
particularly useful to obtain initial information quickly and 
relatively cheaply.  Hospital sources of cases and controls, do, 
however, introduce considerable difficulties with regard to the 
representativeness of all patients with the disease of interest, 
and in terms of the controls, the degree to which they are 
representative of the general community.  Furthermore, response 
rates are liable to differ between cases and controls, especially 
in those from hospitals.  A random or stratified (age, sex) sample 
of persons living in the area covered by the hospitals is perhaps 
the best source for a control group.  There are various ways of 
obtaining such a group.  A sample might be drawn using city 
directory data, tax or electoral rolls, etc.  One theoretically 
simple, if taxing, way is to draw a domiciliary matched 
("neighbourhood") sample.  Here, a house is selected in the 
neighbourhood of the patient's home and a search is made in a 
systematic way, from house to house until a suitable control is 
found.  In a recent study of bladder cancer, conducted by the US 
National Cancer Institute and sponsored by the Food and Drug 
Administration, dialing of telephone numbers chosen at random was 
used to identify one control group (Hoover & Strasser, 1980). 

(c) Measurement of exposure

    In most case-control studies, much reliance is usually placed 
on past information elicited in a comparable manner from cases and 
controls.  Occasionally, measurements or records of past exposures 
may be available, but, in most cases, it is unlikely that these 
will be of comparable quality for cases and controls. 

(d) Confounding factors

    In a case-control study, these can be dealt with initially by 
matching cases and controls in terms of major confounding factors.  
"Matching" may refer to pairing individual controls with particular 
cases according to the matching factors ("matched pairs"), or it 
may refer to arranging that the distributions of the matching 
factors among all controls are similar to those found among the 
cases, without pairing individual controls with cases.  These 
design strategies need to be distinguished, because they attract 
different approaches in the statistical analyses of results. 

    It is usually desirable to match for several potentially 
confounding characteristics such as age, sex, ethnic groups, and 
socioeconomic circumstances.  In view of unavoidable differences 
in diagnostic precision and entry characteristics, it is also 
desirable to match for hospital, in hospital-based studies.  
However, it is not possible to study the importance of a 

potentially confounding factor in relation to the occurrence of 
cases, if that factor has been matched in cases and controls.  For 
instance, data from a case-control study in which controls were 
matched with cases with respect to hospitals, as described above, 
would not provide information about the suspected differences 
between the hospitals in diagnostic precision or entry characteristics.  
It follows therefore that factors which are the  subject of 
 investigation (including so-called "confounders") must never be 
matched. Their relative importance and co-associations with 
recurrence of cases may however be studied using appropriate 
analytical methods, if (unmatched) controls are selected randomly 
and provided that correlations between these factors themselves are 
not too high.  For further details, see section 

(e) Advantages and disadvantages

    Case-control studies of hospital groups can be carried out 
fairly quickly and cheaply.  As a first approach to many diseases 
about which causation is obscure, such studies are very valuable 
for identifying hazards and suggesting hypotheses for more rigorous 
testing.  The method is particularly useful in studying rare 
diseases.  The main disadvantages are that bias may be incorporated 
into any comparisons, because of greater preoccupation by the 
cases than by the controls about the disease under study.  Bias can 
also arise rather easily because of preconceived ideas on the part 
of the investigators.  As a case-control study normally deals with 
a small group, the wider application of its results has to be made 
with caution.  Temporal relations as to whether the disease 
preceded or followed the exposure may at times be hard to establish 
in a case-control study.  In a prospective case-control study, loss 
of study subjects from the case group may also be a problem.  
Furthermore, a case-control study gives only an approximation of 
relative or attributable risk. 

2.10.  Controlled Exposure Studies

    The demonstration of prevention of some effect by a well-
designed controlled human exposure study is perhaps the most 
convincing way of showing a relationship between cause and effect.  
Unfortunately, "experimental" studies often raised insuperable 
ethical and practical problems in the past.  It has to be 
emphasized that any controlled exposure study should be safe, that 
any adverse biological changes that may be induced should be 
reversible, and that no discomfort (or at most only minimal 
discomfort) should be produced.  There is also general agreement 
about the desirability of informed consent, which may imply 
understanding by participants of the study design in some cases 
(section 5.3). 

    Studies of "natural experiments", such as those on 
environmental accidents and adverse effects on health that have 
ensued, have been a recognized epidemiological approach for a long 
time.  These include the studies conducted in London, after the 
1952 December smog (Ministry of Health, 1954) and those performed 
in Hiroshima and Nagasaki on survivors from the atomic bombs 
(section  These examples have provided a great deal of   
useful information on the acute effects of air pollution and on the
health effects of ionizing radiation, respectively.                

    On a more limited scale, studies of workers before and after 
the working shift have provided useful information on the possible 
hazards of exposure of the respiratory tract to vegetable and 
mineral dusts and various toxic gases.  Studying populations before 
and after a pollutant has been removed is a reasonable approach to 
design, especially when a latency period is part of the design;  
certainly an improvement in health would be expected if the 
pollutant were causing adverse effects.  Sometimes, the 
deterioration of the environment following industrial development 
may be foreseen and observations may be made to exploit such an 
opportunity in the most effective manner.  One example of this type 
is the pre- and post-studies in relation to the siting of a new 
power plant.  The possibility that the use of high sulfur fuel has 
increased sulfur oxide emissions in some cities, but not in others, 
should stimulate the collection of appropriate data in cities where 
such changes are anticipated and in control cities where they are 

(a) Choice of population

    Controlled exposure studies can be based on the general 
community or some particular subgroups, such as a specific age 
group or occupational group.  Such subgroups may be studied by 
exploiting some fortuitous change that has divided the population 
into the treated and control groups that are needed to test some 
hypothesis.  One classical example is John Snow's admirable 
epidemiological analysis on the natural experiment of cholera 
outbreaks in London in the nineteenth century (Snow, 1855).  The 
study by Harrington and his colleagues on cancer in relation to 
asbestos fibres in drinking-water supplied by asbestos cement pipes 
to half the households in Connecticut is another model example of 
this approach (Harrington et al., 1978). 

(b) Exposure, effects, and confounding factors

    In controlled exposure studies, the levels of exposure are 
known by the investigators.  Effects are measured in the course of 
study and confounding variables can be identified and controlled. 

(c) Advantages/disadvantages

    Cause/result and precise exposure/effect relationships can be 
obtained.  However, a study of this type tends to be costly.  The 
drop-out rate may be high.  As already mentioned, great care must 
be given to the ethical problems and the consent of the participants 
is required. 

2.11.  Monitoring and Surveillance

    As the network of monitoring stations to measure environmental 
pollutants, in particular air pollutants, expands in many countries, 
data from these monitoring activities are being increasingly used 
for epidemiological studies.  However, such monitoring being 
primarily for the purpose of pollution control, the data do not 
necessarily provide exposure information that is adequate to relate 
to the health status of the study population.  The use of routine 
data for establishing exposure/effect relationships must be made 
with great caution (section 3.5). 

    Assessment of exposure by personal monitoring and biological 
monitoring would provide more precise exposure data (sections 3.6 
and 3.7), but these methods tend to be expensive. 

    To relate data from routine monitoring activities to the 
information on health effects from a variety of surveillance work, 
would need the development of some means of linking records from 
different sources. 


ALDERSON, M.R.  (1976)   An introduction to epidemiology.  Lon-
don, Macmillan.

ALDRIDGE, W.N. & CONNORS, T.A.  (1982)  Rapid Communication:
Toxic oil syndrome in Spain.   Food Chem. Toxicol.,  20: 989-992.

ARMITAGE, P.  (1977)   National health survey systems in the
 European Economic Community.  Luxembourg, Commission of the
European Community.

BIDSTRUP, P.L. & CASE, R.A.M.  (1956)  Carcinoma of the lung
in workmen in the bichromates-producing industry in Great
Britain.   Br. J. ind. Med.,  13: 260-264.

(1954)  Tumours of the urinary bladder in workmen engaged in
the manufacture and use of certain dyestuff intermediates in
the British chemical industry.  Part I.  The role of aniline,
benzidine, alpha-naphthylamine and beta-napthylamine.   Br. J.
 ind. Med.,  11: 75-104.

CASSELL, E.J. & LEBOWITZ, M.D.  (1976)  The utility of the
multiplex variable in understanding casuality.   Perspect.
 Biol. Med.,  19: 338-341

COCHRANE, A.L.  (1960)  Epidemiology of coalworker's pneumo-
coniosis, Chapter l8.  In : King, E.J. & Fletcher, C.M., ed.
 Industrial pulmonary diseases.  London, Churchill.

CROMBIE, I.K.  (1981)  The limitations of case control studies
in the detection of environmental carcinogens.   J. Epidemiol.
 commun. Health,  35: 281-287.

Yu.N.  (1980)  [The incidence of diseases in children living
in urban areas with different atmospheric air pollution
levels.]  Gig. i Sanit.,  N.II: 3-4 (in Russian).

DEVESA, S.S. & SILVERMAN, D.T.  (1978)  Cancer incidence and
mortality trends in the United States, 1935-74.   J. Natl
 Cancer Inst.,  60: 545-571.

DOLL, R. & HILL, A.B.  (1952)  A study of the aetiology of
carcinoma of the lung.   Br. med. J.,  2: 1271-1286.

DOLL, R. & PETO, R.  (1976)  Mortality in relation to smoking:
20 years observation on male British doctors.   Br. med. J.,  2:

DOUGLAS, J.W.B. & WALLER, R.E.  (1966)  Air pollution and res-
piratory infection in children.   Br. J. prev. soc. Med.,  20:

KORNS, R.F.  (1957)  A prospective study of degenerative
cardiovascular disease in Albany:  Report of 3 years' ex-
perience.  I.  Ischaemic heart disease.   Am. J. pub. Health, 
47 (Suppl.): 25-32.

PEOPLE'S REPUBLIC OF CHINA (1979)   Atlas of cancer mortality
 in the People's Republic of China,  Beijing, China Map Press.

EVANS, T.N. & BROWN, G.C.  (1963)  Congenital anomalies and
virus infections.   Am. J. Obstet. Gynecol.,  87: 749.

J.M.  (1973)  Chronic nonspecific respiratory disease, Berlin,
New Hampshire 1961-67.  A follow-up study.   Am. Rev. respir.
 Dis.,  107: 110-122.

(1965)  American emphysema and British bronchitis.  A
standardized comparative study.  Am. Rev. respir. Dis.,  40: 112.

FLETCHER, C.M. & TINKER, C.M.  (1961)  Chronic bronchitis:  a
further study of simple diagnostic methods in a working popu-
lation.   Br. med. J.,  1: 1491.

FOX, A.J. & COLLIER, P.F.  (1976)  Low mortality rates in in-
dustrial cohort studies due to selection for work and survival
in the industry.   Br. J. prev. soc. Med.,  30: 225-230.

GOLUBEV, I.R., BALATSKY, O.F., & CHUPIC, A.B.  (1979)  [On the
quantitative evaluation of the effects of atmospheric air pol-
lution on the incidence of diseases in children.]   Gig. i
 Sanit.,  N6: 50-53 (in Russian).

GORDON, T. & KANNEL, W.B.  (1970)  The Framingham,
Massachusetts study twenty years later.  In:  Kessler, I.I. &
Levin, M.L., ed.   The community as an epidemiological labora-
 tory.  Baltimore, John Hopkins Press.

FLANNERY, J.T., & WOODHILL, R.S.  (1978)  An investigation of
the use of asbestos cement pipes for public water supply and
the incidence of gastrointestinal cancer in Connecticut,
1935-1973.   Am. J. Epidemiol.,  107(3): 96-103.

HEALTH & SAFETY EXECUTIVE  (1979)   Asbestos report.  Volumes 1
 and 2, Final reports of Health & Safety Commission.  London,
HM Stationery Office, pp. 100 and 103.

HEANSZEL, W. & KURIHARA, M.  (1968)  Studies of Japanese
migrants. I. Mortality from cancer and other diseases among
Japanese in the United States.   J. Natl Cancer Inst.,  40:

(1979)  Health effects of particulate pollution: reappraising
the evidence.   Am. J. Epidemiol.,  110: 527-659.

HOLLAND, W.W., REID, D.D., & SELTSER, R.  (1965)  Respiratory
disease in England and the United States.  Studies of compara-
tive prevalence.  Arch. environ. Health,  10: 338-343.

HOOVER, R. & STRASSER, P.H.  (1980)  Artificial sweetners and
human bladder cancer:  Preliminary results.  Lancet,  1:

JACOBSEN, M.  (1981)  The importance of epidemiology in
research on pneumoconiosis.  In:   Proceedings of Seminar on
 Epidemiology and Technical and Medical Prevention of Coal
 Miner's Pneumoconiosis.  Luxembourg, Commission of the Euro-
pean Communities, pp. 5-25 (Industrial Health and Safety, EUR

 major disease mortalities for cities, towns and villages in
 Japan 1969-1978.  Tokyo, Research Committee on Geographical
Distribution of Diseases.

KMET, J. & MAHBOUBI, E.  (1972)  Oesophageal cancer studies in
the Caspian Littoral of Iran.  Initial observations.   Science,
175: 846-853.

LAWTHER, P.J., WALLER, R.E., & HENDERSON, M.M.  (1970)  Air
pollution and exacerbations of bronchitis.   Thorax,  25:

LEE, A.M. & FRAUMENI, J.F., Jr  (1969)  Arsenic and respira-
tory cancer in man: an occupational study.   J. Natl Cancer
 Inst.,  42: 1045-1052.

MAGNUS, K.  (1973)  Incidence of malignant melanoma of skin in
Norway, 1955-70.  Variations in time and space and solar
radiation.   Cancer,  32: 1275-1286.

MARTIN, A.E.  (1964)  Mortality and morbidity statistics and
air pollution.   Proc. Royal Soc. Med.,  57: 969-975.

MARTIN, A.E. & BRADLEY, W.H.  (1960)  Mortality, fog and
atmospheric pollution - an investigation during the winter of
1958-59.   Monthly Bull. Min. Health Public Lab. Service,  19:

F.R., Jr  (1975)   Atlas of Cancer Mortality for US counties,
 1950-69.  Bethesda, Md, US Dept of Health, Education and
Welfare,  National Institutes of Health.

MCDONALD, A.D. & MCDONALD, J.C.  (1971)  Epidemiologic sur-
veillance of malignant mesothelioma in Canada.   Can. Med.
 Assoc. J.,  109: 359-362.

& ROSSITER, C.E.  (1971)  Mortality in the chrysotite asbestos
mines and mills of Quebec.   Arch. environ. Health,  22: 677-686.

 Cancer registration and its techniques.  Lyons, IARC (IARC
Sci. Publ. No.2l).

MINISTRY OF HEALTH  (1954)  Mortality and morbidity during the
 London fog of December l952.  London, HM Stationery Office.

MORIYAMA, I.M., DAWBER, T.R., & KANNEL, W.B.  (1966)  Evalua-
tion of diagnostic information supporting medical certifica-
tion of deaths from cardiovascular disease.  In:  Haenszel,
W., ed.  Epidemiological approaches to the study of cancer and
 other chronic disease,  Washington, DC, US Dept of Health,
Education and Welfare (Health Service Monograph 19).

RAFFLE, P.A.B.  (1966)  Incidence and prediction of ischaemic
heart disease in London busmen.   Lancet,  2: 553-559.

& EDWARDS, T.I.  (1937)   Study of chronic mercurialism in the
 hatters fur-cutting industry.  Washington DC, USPHS (Public
Health Bulletin No. 234).

 mortality decennial supplement England and Wales, 1970-72.
London, HM Stationery Office (Series DS No. 1).

(1974)  Apparent changes in cancer mortality 1968:  A result
of the introduction of the Eighth Revision of the Inter-
national Classification of Diseases.   Pub. Health Rep.,  89:

PELL, S., O'BERG, M.T., & KARRH, B.W.  (1978)  Cancer epi-
demiologic surveillance in the Du Pont Company.   J. occup.
 Med.,  20: 725.

ROACH, S.A. & SCHILLING, R.S.F.  (1960)  A clinical and
environmental study of byssinosis in the Lancashire cotton
industry.   Br. J. ind. Med.,  17: 1-9.

SNOW, J.  (1855)   On the Mode of Communication of Cholera,
2nd. ed.,  London, Churchill.  Reproduced in  Snow on Cholera.
New York, Commonwealth Fund, 1936.  Reprinted by Hafner, New
York, 1965.

Multivariate analysis of the relationship of seven variables
to blood pressure:  Findings of the Chicago Health Association
Detection Project in Industry, 1967-72.   J. Chronic Dis.,
28(10): 527-548.

STEWART, A. & HUGHES, J.P.W.  (1951)  Mass radiography find-
ings in the Northamptonshire boot and shoe industry, 1945-6.
 Br. med. J.,  1: 899-906.

TABUENCA, J.M.  (1981)  Toxic-allergic syndrome caused by
ingestion of rapeseed oil denatured with aniline.   Lancet,  2:

G.J., DE VRIES, K., & ORIE, N.G.M. (1973)  Epidemiological
investigations in the Netherlands into the influence of smok-
ing and atmospheric pollution on respiratory symptoms and lung
function disturbances.  Pneumologie,  149: 119-126.

WAGNER, G.  (1976)   Uses of hospital discharge summary forms
 in the European Region.  Copenhagen, World Health Organiza-
tion Regional Office for Europe (ICP/SHS 029).

WALDRON, H.A. & VICKERSTAFF, L.  (1977)   Accuracy of diagnosis
 of fatal conditions and quality of certification.  London,
Nuffield Provincial Hospitals Trust.

WHO  (1965)   Trends in the study of morbidity and mortality.
Geneva, World Health Organization (Public Health Papers No.27).

WHO  (1977)   Manual of the international statistical classifi-
 cation of diseases, injuries, and causes of death.  Ninth
Revision.  Volume 1, pp. 779 and Volume 2, pp. 659., Geneva,
World Health Organization.

WHO  (1982)   The place of epidemiology in local health work;
 the experience of a group of developing countries.  Geneva,
World Health Organization (Offset Publication No. 70).


3.1.  Introduction

    The validity of studies in the field of environmental 
epidemiology depends both on the assessment of exposure and of the 
effects on health.  Each of these aspects is liable to present 
difficulties and uncertainties.  Thus, it is important that every-
one involved in the design and conduct of investigations and in the 
interpretation of results, has a complete understanding of the 
problems.  It is the purpose of this chapter to discuss basic 
aspects of exposure assessment, in order to improve the quality of 
epidemiological studies, and consequently the scientific basis for 
control measures.  The emphasis is on general population studies, 
but exposure assessment is also of major importance in occupational 
health studies.  The general approach is similar:  much of what is 
practised in population studies has been markedly influenced by the 
practice of exposure assessment in workers.  Moreover, for many 
environmental agents, occupational exposure may contribute 
substantially to the total exposure in some subgroup of the general 

    The environment may be divided into two types with regard to 
exposure assessment: (a) the  objective environment, which means
the actual physical, chemical, and social environment as described
by objective measurements such as noise levels in decibels (dB)
and concentration of air polluting: and (b) the  subjective 
(perceived) environment, as it is perceived by persons who live in
it, e.g., annoyance caused by air pollution or noise, or pleasure 
arising from good housing conditions.  In this chapter most 
sections deal with the objective assessment of exposure; in section 
3.8, however, special emphasis will be laid on the assessment of 
subjective exposure. 

    Epidemiological studies may be concerned with scattered 
individuals, with groups living or working together, or with 
populations in defined areas or countries; in each case appropriate 
exposure assessments have to be made.  For the present purpose the 
environments in which people operate can be considered at the four 
following levels: 

(a) The domestic or "micro" environment, concerned with the 
    subject in the home.  Exposure may be determined by
    personal or family eating habits, cooking facilities,
    hobbies, other personal habits (e.g., smoking or
    drinking), use of therapeutics, drugs, or cosmetics,
    pesticides applied in the home and garden, etc.

(b) The occupational environment.  The subject may spend a
    large part of his/her life in occupational environments
    such as coal mines, steel works, etc., where there may be
    specific environmental problems.  Periods spent in schools
    or other educational establishments might also be 
    considered under this heading.

(c) The local or community environment.  In the immediate area 
    in which the subject lives he/she may be exposed for example 
    to ambient air pollution, aircraft and traffic noise, or 
    drinking water containing particular constituents. 

(d) The regional environment.  The subject lives in a
    particular climatic zone, at a certain geographical
    longitude, latitude, and altitude, etc.

    A few examples of exposure to the same environmental factor at 
various levels of operation are given in Table 3.1.
Table 3.1.  Examples of exposure to environmental factors at various levels 
of exposure
Level of   Carbon      UV          Noise         Solvents      Ionizing
operation  monoxide    radiation                               radiation    
Micro      smoking,    therapeu-   music,        cleaning,     medical
or         cooking,    tics,       hammering,    hobbies       diagnosis
domestic   heating     gardening,  noise from                  and therapy,
                       sunbathing  neighbours                  emissions
                                                               from struc-

Occupa-    traffic     laboratory  construction  workers in    x-ray
tional     policemen,  workers,    workers,      solvents      techni-
           metallur-   agricul-    military      manufac-      cians;
           gical       tural       service       turing,       workers
           workers     workers                   painters,     in nuclear
                                                 dry cleaners  plants

Local      traffic     sunlight    aircraft,     emissions     tubercu-
           exhaust                 town          from          losis mass
                                   traffic       industry      screening

Regional   -           high        storm,        -             fallout
                       altitude,   hurricane                   from atomic
                       tropics                                 weapons
    In assessing individual and group exposure to specific agents, 
the contribution from each of these four environmental levels to 
the total exposure has to be taken into account; the intensity and 
duration of exposure and the coexistence of other hazardous factors 
may differ (section 3.3). 

3.2.  Exposure and Dose

    In pharmacological and toxicological studies, the term,  dose 
is used to indicate the amount administered, and  dose-rate to 
indicate the dose per unit of time.  The unit quantity, and the  
frequency and duration of administration determine the total dose 
received over a day, a week, or a year.  In epidemiology, one often 
hesitates to use the term dose, because generally it is only 
possible to make an estimate of the actual dose received.  There-
fore, the terms,  exposure, instead of dose, and  exposure/effect 
relationships rather than dose/effect relationships are preferred.
The exposure may often be assessed by measuring the concentration 
of a substance in air, water etc., or the intensity in the case of 
sound or radiation, and some effects may be determined more by the 
instantaneous concentration or intensity than by the total dose. 

3.2.1.  Systemic agents

    There are four indices of exposure in the case of agents that 
exert an effect after being absorbed into the body: 

     External exposure in a general sense.  This is the 
concentration that is present in, for example, food, drinking-water,
or air, in relation to frequency and duration of exposure.

     External exposure in a narrow sense  - intake.  Often the
only data available are concerned with the concentrations of 
agents (mg/kg in food, mg/litre in water, mg/m3 in air) and not
the amounts of food, drinking water, and air, to which man is 
exposed per unit of time.  In medicine, however, the dose 
administered is never expressed as the concentration, but as the 
amount ingested, injected, or inhaled.  In work and sports 
physiology, energy consumption is not calculated in concentrations 
of oxygen in inhaled and exhaled air, but as the difference between 
the amount of oxygen inhaled and exhaled.  Therefore, in exposure 
assessment, an effort should also be made to measure the 
concentration of the agent in its vehicle and the amount of food, 
water, and air, consumed by an individual, i.e., the intake.  In 
most studies reported so far, no endeavour has even been made to 
estimate respiratory volume or actual food and water intake.  The 
oxygen consumption for an adult man (70 kg) at rest is about 0.3 
litre/min; the uptake of 1 litre of oxygen requires an intake of 
about 25 litres of air; therefore, the respiratory volume/h, at 
rest, is about 0.5m3; in moderately heavy work, which can be
sustained during a 8-h working day, the respiratory volume/8 h will 
be 8-10m3; for 24 h, the respiratory volume will be 15-20m3.
The energy requirement for a child of 1-3 years is about 420 kJ/kg 
body weight, for an adult about 170 kJ/kg body weight; the relative 
exposure to a food contaminant per unit of body weight, therefore, 
may be higher in children than in adults by a factor of 2-3.  The 
intake of drinking-water may vary considerably from subject to 
subject, consequently the amounts of pollutants ingested through 
drinking-water will differ greatly among the subjects. 

    For particulates in inhaled air, the particle size 
distribution determines the fraction that reaches various parts of 
the airways, and thus the possibility of local action or pulmonary 
absorption will also be determined.  Particles with a diameter
> 5 µm tend to be deposited in the nasopharyngotracheal region.  
The chemical composition may vary with particle size:  carbon, 
lead, and sulfates, for example, occur mainly in very fine 
particles, generally < 1 µm diameter.  The particle size 
distribution in occupational exposure may differ greatly from that 
in ambient exposure.  Fibres of materials such as asbestos, with 
very small diameters, tend to follow the air-flow through the 
respiratory system and even ones up to some 200 µm in length may 
penetrate into the deeper airways. 

    Highly water-soluble gases, for example sulfur dioxide and 
formaldehyde, are trapped by the moist environment of the upper 
airways, whereas the less soluble nitrogen dioxide or phosgene 
penetrate into the bronchiolar and aveolar regions.  Agents in food 
also differ in the degree of absorption according to their chemical 
composition.  The presence of vegetable fibres may produce bulky 
gastrointestinal content and increase the speed of passage; the 
decreased exposure time might be one of the reasons why the fibre 
content of food could have a preventive effect on colonic tumours.  
In South Africa, bowel cancer is much rarer in the Bantu peoples 
than in the Caucasians; even among the Bantu, intestinal transit 
times have been found to be markedly different, probably because of 
differences in the fibre content of food (Walker, 1978).  Hardness 
of drinking water may determine whether elements are leached from 
vegetables during cooking or whether their concentration is 
increased (Moore et al., 1979). 

    These examples show that the true intake may differ considerably 
from the levels of exposure calculated from concentrations in ambient 
air, food, or drinking-water.  

 Internal exposure - uptake.  The agents available for absorption 
are usually only partially absorbed into the body:  uptake = intake 
x (fractional) absorption rate.  The degree of absorption varies 
widely, for example, in the gastrointestinal tract, methylmercury 
is absorbed almost completely, whereas metallic mercury is hardly 
absorbed at all.  Absorption of lead is higher in an empty stomach 
than in a full one, and it is probably higher in children than in 

    In the case of inhaled gases or vapours, the concentrations 
in both inhaled (Ci) and exhaled (Ce) air must be measured
and multiplied by the respiratory minute volume (V).  The uptake 
will be (Ci - Ce) x V x t (where t = time).  As soon as an 
equilibrium has been achieved between uptake and elimination 
(such as by biotransformation and excretion), the level of uptake 
becomes constant at constant Ci and V.  During physical activity,
V increases and equilibrium is achieved earlier than at rest.  
Carbon monoxide provides a good example:  toxic levels in blood 
are achieved earlier during physical activity than at rest, and 
sooner in children than in adults. 

     Exposure at the target organs.  In epidemiological studies,
it is usually not possible to measure the concentrations (or 
amounts) of agents present at the target organs, for example, 
liver, brain, etc., although it is true that determination of the 
concentrations (or amount) of cadmium in liver and kidney is 
possible by neutron activation analysis (Ellis et al., 1981).  The 
Task Group on Metal Toxicity (Nordberg, 1976) presented a few 
definitions, which not only can be used in metal toxicity studies, 
but are also applicable in the study of many other environmental 

     Critical concentration for a cell.  This is the concentration
at which an adverse functional change, reversible or irreversible, 
occurs in the cell. 

     Critical organ concentration.  This is the mean concentration
in the organ at the time when the most sensitive types of cell 
reach the critical concentration. 

     Critical organ.  This term is used for the particular organ
that first attains the critical concentration under specified 
circumstances or exposure and for a given population. 

    Assessment of exposure through biological monitoring or 
analysis of samples from specimen banks (section 3.7) may provide 
data that approximate the relevant exposure at the target organs 
much better than those obtained through environmental monitoring 
(section 3.5). 

3.2.2.  Local exposure

    Some agents act on the surface linings of eyes and airways or 
on the skin.  Oxidants, such as peroxyacetylnitrate (PAN), exert an 
irritant effect  on the eyes as a function of the number of oxidant 
molecules that are absorbed in the eye fluids per unit of time.  
Exposure is a function of the ambient concentration of PAN and of 
the physical properties of the fluid, such as solubility and 
diffusion coefficient. Because the physical properties may be 
assumed to be constant, the intensity of exposure will be 
determined by the concentration in ambient air and the frequency 
and duration of exposure. 

    Some agents may penetrate the skin; this depends on physio-
chemical properties of the agent, properties of the skin (variable 
at different sites in one individual, and variable between 
individuals), environmental temperature and humidity, presence of 
skin disease, etc. 

3.2.3.  Physical factors

    The considerations under sections 3.2.1. and 3.2.2. apply 
mainly to chemical agents, but also apply to compounds with 
radioactive properties.  However, in the case of physical factors, 
for example, noise, vibration, and ultraviolet radiation, the 

actual exposure of the subjects has to be assessed as carefully as 
possible, using measurements of intensity, frequency, and duration 
(section 3.5.4). 

3.3.  Combined Exposure, Physical and Chemical Interactions

    Health effects due to environmental factors are manifested in 
various ways (Chapter 4).  However, the range of effects is limited 
compared with the large variety of chemical and physical factors 
that may produce them.  To a large extent, health effects are non-
specific; the causative agents can seldom be identified from the 
effects manifested.  This is the main crux of exposure/health 
effect studies. 

    Simultaneous or consecutive exposure to several agents may 
modify risks to health.  Nelson (1976) summarized existing data on 
the role of the interactions of environmental agents that may 
modify biological activity, distinguishing synergism (potentiation), 
antagonism, or merely additive effects.  Potentiation and 
antagonism may be due either to modified toxicokinetics (affecting 
internal exposure) or modified toxicodynamics (relating to health 

3.3.1.  Same agent, various sources

    A well-known example is exposure to noise.  In a study in 
Japan, Kono and his coworkers (1982) measured total noise exposure 
per day as the summation of exposure during work, in the domestic 
environment, and while travelling.  For housewives, the equivalent 
level over 24-h periods (Leq 24) (section was 70.2 dB(A)a
in an industrial area and 67.4 dB(A) in a residential area.  As
regards noise exposure in the home, the Leq 24 was higher in 
housewives of less than 40 years of age, than in older age groups, 
because of different patterns of activity. 

3.3.2.  Various agents, same source        
    It is well known that air, food and water carry mixtures of 
many environmental agents.  In the air pollution situation the 
general population may be exposed to a mixture of sulfur dioxide, 
sulfuric acid, smoke, sulfates, ozone, oxides of nitrogen, 
peroxyacetylnitrate, hydrocarbons, aldehydes, etc.  Assessment of 
exposure to indicator agents is a valid procedure, provided that 
the composition of the pollutants is well known.  However, there 
has been a considerable change in the composition of pollutants 
in urban air and in water supplies in the past few decades, 
making it difficult to use any one component as an indicator in 
long-term studies.     

a   The expression dB(A) is commonly used to refer to the
    A-filter frequency weighting, which usually provides the
    highest correlation between physical measurements and
    subjective evaluations of the loudness of noise, by
    modifying the effects of the low and high frequencies with
    respect to the medium frequencies (WHO, l980b).

    Food may contain a wide range of trace metals (e.g., cobalt, 
copper, iron, manganese, selenium, and zinc); however the 
proportions may differ from place to place and from time to time.  
If only one factor is to be selected for the assessment of 
exposure, at least approximate data concerning the composition of 
the mixture must be obtained. 

    In elucidating the so-called "soft water story", i.e., the 
observed inverse relation between water hardness and cardio-
vascular mortality, the sum of the calcium and magnesium content of 
drinking-water had, until recently, been relied on.  However, in 
recent years, there have been indications that the magnesium 
content might be more relevant than calcium.  Generally, with 
increasing hardness, the corrosiveness of the water decreases;  
however, the ability of hard water to dissolve metals from pipes 
is not always less than that of the soft water.  The "natural" 
relationship between metal concentration and softness of water has 
disappeared in the Netherlands in recent decades (Zielhuis & 
Haring, 1981).  Vos et al. (1978) found higher lead, cadmium, and 
zinc (but not copper) concentrations in hard than in soft tap-water 
in two adjoining communities; higher lead, cadmium, and zinc levels 
in blood were also found in the hard-water town.  In addition, hard 
water often contains higher concentrations of silicon and lithium.  
A valid epidemiological study, therefore, should assess exposure to 
a multitude of agents in tap water, which may vary with geographical 
areas and with water distribution systems. 

    In occupational health studies, a relation has been established 
between the incidence of lung cancer and exposure to nickel and 
chromium compounds and there is evidence that certain medium- 
or slightly-soluble compounds of both nickel and chromium are 
carcinogenic.  If only the total nickel and/or chromium contents of 
workroom air are measured, not taking into account individual 
compounds, an overestimation of health risk may occur. 

3.3.3.  Various agents, various sources

    The most important example under this heading concerns inter-
actions between tobacco smoking and exposure to environmental 
pollutants (particularly by inhalation).  For example, it has 
been established that the risk of lung cancer in asbestos workers 
or uranium miners who smoke is much higher than that in smokers who 
are not exposed to asbestos or uranium, or in non-smoking workers; 
the risk is not additive, but is more or less multiplicative.  In 
foundry workers, silicon dioxide dust affects the condition of the 
airways and smoking increases its health risk (Kärävä et al., 
1976).  Not only the chemical factors should be considered.  The 
high risk of skin tumours in road-tar workers exposed to the 
ultraviolet rays of sunlight is well known. 

3.3.4.  Impurities

    In industry and in chemical applications in the environment, 
compounds of commercial quality that may contain up to several 
percent of impurities are often used.  If trace amounts of such 

impurities are responsible for the health risk, then the 
exposure/effect relationship of the parent compound is not 
representative of the true one.  A well known example is the 
herbicide 2,4,5-trichlorophenoxyacetic acid (2,4,5-T), which 
contains trace amounts (< 0.1 mg/kg) of the extremely toxic 
2,3,7,8-tetrachlorodibenzo- p-dioxin (TCDD).  Due to dilution
during formulation, the process of application, etc., the final 
levels of TCDD in food are usually not detectable, but exposure of 
workers may be high enough to cause health effects. 

    Nitrosamine formation may take place during the production of 
some pesticides (e.g., trifluorolin and dinitramine); the 
nitrosamine levels in the technical products may occasionally 
exceed even 100 mg/kg and therefore become detectable in crops.  In 
addition, nitrosamines may also be formed owing to reactions 
between the pesticides and naturally occurring amines (chemical 
interaction, section 3.3.5). 

    In exposure assessment, therefore, due attention should be paid 
to the presence of impurities that may be more toxic than the 
parent compounds. 

3.3.5.  Interactions

    Three types of interaction can be distinguished, leading to a 
change in the composition of chemical compounds between the point 
of emission and the target organs: 

-   change in the chemical composition and/or physical form
    within the environment;

-   physical interaction between chemical agents and particulates
    within the environment; and

-   change in physical and/or chemical composition within the
    human body.

    Such interactions may essentially change the nature of exposure 
and, consequently, the health risk. 

    Some examples of changes in chemical composition in the 
environment are; formation of alkylmercury compounds in sediments 
from inorganic mercury compounds; secondary oxidation of sulfur 
dioxide to sulfuric acid and sulfates; the build-up of photo-
chemical smog in ambient air.  As the chemical reactions in air are 
time-dependent, the resultant composition of the mixture may change 
over a large distance because of air movements.  Furthermore, the 
source of emission may also affect the ultimate composition.  In a 
study from the Netherlands, the ratio of ozone to peroxyacetyl-
nitrate was higher when the main source was automotive exhaust than 
when it was the petrochemical industry (Guicherit, 1979). 

    The Proceedings of the International Workshop on Factors 
influencing Metabolism and Toxicity of Metals (Nordberg, 1978) 
summarized the present state of knowledge on the interaction of 

metals.  Among toxic metals, mercury provides a well-known example 
of transformation into a more toxic compound, methylmercury, in the 
environment.  On the contrary, methylation of inorganic arsenic 
probably leads to non-toxic organic arsenic compounds, present in 
marine food.  Within the human body, in the intestines after 
ingestion, and in organs after absorption, interaction also may 
take place between metals and nutritional factors, either 
increasing or decreasing the health risk.  At present, only a few 
human data are available.  However, a number of animal studies 
indicate that such interactions might possibly influence human 
health risks.  Physical agents, e.g., ultraviolet radiation, may 
induce changes in the body that affect the subsequent action of 
chemical agents. 

    A low intake of calcium and vitamin D in patients with Itai-
Itai disease (section 5.5.8) may have contributed to high 
accumulation of cadmium and to the development of bone changes 
associated with high cadmium exposure.  Increases in the cadmium/
zinc ratios in blood (or kidney) at higher cadmium exposure may 
constitute a more relevant index of exposure than cadmium levels 
as such. 

    In non-occupationally exposed groups, effects of the inter-
action of lead and iron are most often seen in children:  iron 
deficiency is associated with increased lead levels in the blood, 
probably because of increased enteric absorption of lead; more-
over, both iron deficiency and lead overexposure may induce the 
same effect - an increase in porphyrin in erythrocytes.  A low 
nutritional intake of calcium and proteins also may increase lead 
absorption.  Very probably the intake of selenium counteracts the 
toxicity of mercury.  In miners exposed to inorganic mercury, a 
parallel increase in mercury and selenium levels in the blood has 
been demonstrated, suggesting a biological interaction. 

    Within the body, biotransformation of many organic compounds 
takes place, usually mediated by enzymes.  Exposure may increase 
the production of enzymes (enzyme induction), and thus internal 
exposure to the original agent may be changed; in not a few cases, 
the parent compound is transformed into metabolites that constitute 
the true toxic agents.  Assessment of exposure by means of 
biological monitoring (section 3.7) also aims at measuring these 
relevant metabolites in biological specimens. 

    Simultaneous exposure to pharmaceuticals may affect the 
metabolism of environmental chemicals.  In epileptic workers 
exposed to DDT and receiving anti-epileptic drugs, the DDT level in 
adipose tissue was found to be considerably lower than in non-
epileptic fellow workers.  Consecutive exposure to trichloro-
ethylene and alcoholic drinks (after work) may cause skin flushing, 
probably because of interference with the metabolic transformation 
of ethanol.  Industrial exposures may influence the therapeutic 
effects of drugs.  For example, the anticoagulatory effect of 
warfarin may be decreased during exposure to chlorinated 

    In the gastrointestinal tract, nitrites from food may interact 
with secondary amines, and carcinogenic nitrosamines may be formed.  
High dietary fat may increase the concentration of bile acids in 
the large bowel with subsequent metabolism by bacterial flora to 
carcinogens or cocarcinogens; research workers at the International 
Agency for Research on Cancer (IARC, 1977) observed differences in 
faecal flora between two Scandinavian populations with low and high 
risk from carcinoma of the colon. 

    There also exists physical interaction.  A well-known example 
is the absorption of gases or vapours on to particulates in air, 
thus increasing exposure of the lower airways to agents that might 
otherwise have been trapped in the higher airways. 

    These examples only serve as illustrations and certainly do not 
present an exhaustive review.  Both in environmental assessment 
(section 3.5) and in biological assessment (section 3.7) of 
exposure, account must always be taken of the possibility of chemical 
or physical interaction, because it may change the nature of health 
effects, both qualitatively and quantitatively. 

3.4.  Qualitative Assessment of Exposure

    While the ultimate aim in an epidemiological study should be 
the assessment of exposure in quantitative terms, to allow the 
derivation of dose/effect relationships, there is a place for 
qualitative assessment within exploratory studies, or for the 
formulation of hypotheses, as has been discussed in Chapter 2. 

    In chronic disease studies, it is usually necessary to assess 
exposure retrospectively, and since quantitative data are seldom 
available for periods extending back for some 40 years or more, 
qualitative indices of exposure may have to be used as the 
independent variable.  In occupational studies, these may be 
provided by job histories, together with information on the types 
of materials that people in each job might be exposed to, and on 
the degree of control that existed in the past.  In community 
studies, area of residence, information on migration and on ethnic 
and racial characteristics may be used as indices, and personal 
habits such as smoking, alcohol intake, betelnut chewing, 
sunbathing, etc. may provide indicators of exposure to agents of 
interest in their own right or as factors interacting with other 
environmental pollutants. 

3.5.  Environmental Assessment of Exposure

    The most general method of assessing exposure in quantitative 
terms is referred to as  environmental monitoring.  The definition
of  monitoring  adopted by the 1974 Intergovernmental Meeting on
Monitoring convened by the UNEP was "the system of continued 
observation, measurement, and evaluation for defined purposes" 
(WHO, 1975).  An International Workshop cosponsored by the 
Commission of the European Community (CEC), the US Environmental 
Protection Agency (USEPA), and WHO defined the term  "environmental 
 monitoring"  as "the systematic collection of environmental samples

for analysis of pollutant concentrations" (Berlin et al., 1979).  
In epidemiological studies, the observations must be made in a way 
that relates as closely as possible to the exposure of the 
population being considered, but they need not necessarily be of a 
repetitive or continuous form as required for some other monitoring 

    In designing a monitoring programme, general questions of the 
type posed in Chapter 1 have to be considered again, namely: 

-    what agents need to be studied?
-    how long and how often should samples be taken?
-    where should samples be drawn from, or instruments located?
-    what quality of data is needed?
-    which instruments or analytical techniques should be used?

    In practice, it is not always possible to meet all these 
requirements in a faultless manner because of, for example, 
budgetary or technological limitations.  However, it should be 
emphasized that, if the quality of exposure assessment is below a 
certain minimum, the data obtained may be valueless. Many 
epidemiological studies, both in occupational and public health, 
lack even an adequate qualitative assessment of exposure. 

    It should be realized, that environmental monitoring, under-
taken to determine whether ambient levels meet legal quality 
standards for ambient air, water, occupational environment etc., 
usually does not provide adequate data on exposure for use in 
exposure/health-effect studies. 

3.5.1.  Quality of data

    To describe the quality of data, a number of concepts are used, 
for example: 

-    Repeatability:  the difference between measurements
    carried out at a given time with the same instrument, by
    the same person, determining the same property of the same

-    Reproducibility:  the difference between measurements,
    carried out at different times, with different instruments
    usually of the same type, by different persons determining
    the same property of the same material.

-    Precision:  the magnitude of the deviations of a series of
    measurements, usually expressed as the coefficient of
    variation (standard deviation as a percentage of the mean).

-    Accuracy:  the difference between the measured value and
    the true value.

-    Resolution:  the smallest difference of the measured
    property which can still be quantitatively distinguished.

-    Time constant and band width:  the way an instrument
    follows sudden changes in magnitude of the property to be
    measured, to be derived from its response to a step

-    Detection limit:  the smallest measured quantity that can
    be distinguished from zero.

    The quality is determined both by sampling and by analytical 
procedures.  In recent years, developments in sampling instruments 
and in analytical techniques have been substantial and the quality 
of data has improved considerably.  Many exposure data, used in 
epidemiological studies a few years ago, however, were of a 
comparatively low quality.  Ferris (1978) presented several 
examples of measurement errors in monitoring concentrations of air 
pollutants.  There have been interferences with measurement:  for 
bubblers there may be thermal effects (reaction does not take place 
if the vehicle is too cold; or there is decay or evaporation, if 
it is too hot).  Most ambient air monitoring systems for 
particulates in Europe have used the standard smoke method - a non-
gravimetric method using light reflectance from a stained filter 
paper - the reflectance is calibrated and expressed in terms of 
equivalent concentrations of standard smoke.  The results cannot 
however be taken to be equivalent to those obtained by a high 
volume sampler with direct weighing, since refectance/weight 
relationships vary widely with the composition of the particulates.  
In measuring photochemical oxidants, the quality of the potassium 
iodide method, previously used in Los Angeles County, USA and 
elsewhere, has been seriously questioned, which places many air 
quality data in doubt. 
    Another recent example is the development in sampling and 
analytical techniques for the determination of asbestos fibres.  
Before 1964, in the United Kingdom, the commonest instrument was 
the thermal precipitator, while in Canada and the USA, most data 
were derived from midget impingers.  After 1964, membrane filters 
were used in the United Kingdom; these allow fibres to be counted 
specifically, whereas impingers give general particle counts.  
Comparability between particle counts and fibre counts is poor.  
Since 1970, personal sampling (section 3.6) has, to a large extent, 
replaced static sampling.  In 1969, a new method of fibre counting 
(eye-piece graticule) was introduced, which increased the fibre 
count by a factor of 2-3.  This change in sampling and analytical 
techniques resulted in 5 times larger fibre counts in 1979 compared 
with those in 1970, for the same levels of exposure to chrysotile 
fibre (Health & Safety Executive, 1979). 

    Particularly in health-effects studies over long durations of 
exposure, special attention has to be paid to possible changes in 
sampling and analytical methods, that may invalidate the 
comparability of data; the same applies to the comparison of data 
published in the literature. 

3.5.2.  Monitoring strategy for air pollutants

    Reviews on monitoring and/or instrumentation have been 
presented by WHO (1976), Stern, ed. (l976), WHO (l977b), the 
American Conference of governmental Industrial Hygienists (ACGIH) 
(l978), Atherley (l978), NATO/CCMS (l979), and Katz (l980).  Before 
developing a strategy for the assessment of exposure to ambient air 
pollutants, it is important first to evaluate published quantita-
tive or semi-quantitative studies to determine whether there is any 
evidence at all of adverse effects on health.  Such an exploratory 
evaluation may save unnecessary expenditure of time and money, and 
moreover, may be essential for a valid design or exposure 
assessment.  What to sample, how long, how frequently?

    In the case of exposure to chemicals, differences in expected 
health effects require differences in sampling strategy: 

-    Irritants:  Sampling has to be carried out with high time
    resolution:  the frequency of peak concentrations may be
    more relevant than time-weighted average concentrations.

-    Narcotic agents:  Sampling may also have to be carried out
    with high time resolution, particularly in assessing
    occupational exposure at high concentration levels.

-    Systemic agents, including teratogens:  These agents exert
    a toxic action after being absorbed and may cause effects
    in the liver, haematopoetic system, kidney, nervous system,
    etc.  The time resolution of sampling should be geared to
    the biological half-livesa of the agent (or its 
    metabolites) at the target organs (Roach, 1977).  For
    teratogens, the time of exposure during pregnancy may be

-    Carcinogens, mutagens:  The latent period before health  
    effects become manifest may have to be counted in years, 
    or even decades.  In most epidemiological studies, the   
    assessment of exposure is performed retrospectively, and 
    consequently the assessment is only qualitative or       
    semi-quantitative.  Information on peak concentrations,  
    however, should not be neglected, because - at least for 
    some agents - temporary overloading of biological        
    detoxication systems may open up deviant metabolic       
    pathways resulting in carcinogenic/mutagenic metabolites.

a   Biological half-life or half-time is the "time required          
    for the amount of a particular substance in a biological         
    system to be reduced to one-half of its value by
    biological processes when the rate of removal is                 
    approximately exponential" (ISO, l972).                          

-    Agents that may cause pneumoconiosis:  Long-term local
    deposition of certain chemical compounds in the lungs
    results in silicosis, asbestosis, talcosis, etc., in
    workers exposed.  Average concentrations over months or
    over years are particularly relevant for the assessment of
    exposure to these compounds.

-    Agents that cause asthma, chronic bronchitis, or emphysema
    through local action in the airways are usually sampled in
    order to obtain the time-weighted average exposures for a
    working day (8 h) or for the whole day (ambient exposure,
    24 h).  However, peak concentrations may be relevant in
    some cases, particularly in occupational exposures.

    Some agents may exert two or more effects.  For instance, 
benzene acts as a narcotic agent at high concentrations, and as a 
carcinogen, probably at much lower levels; cadmium oxide acts  
directly  on  the  airways and is also a systemic kidney poison; 
inorganic mercury at high concentrations acts on the airways and, 
after absorption, on the brain, whereas, in long-term exposure to 
low concentrations effects may be found only on the brain; toluene 
diisocyanate and formaldehyde act as irritants in short-term 
exposures to high concentrations and as sensitizers in long-term 
exposures to low concentrations. 

    The time resolution has to be adapted to the technological  
process in the case of occupational exposures, in order to      
characterize those at different phases of production.  The basic
considerations, therefore, concern the agent as such, the health
effects under study, and technology.  For occupational agents   
causing pneumoconiosis, rules for sampling frequency have been  
derived, on the basis of the assumption that fluctuations are   
caused by stationary stochastic processes (Coenen, 1976, 1977). 

    Concentrations in ambient air not only depend on the intensity 
of emission (often related to season), but also on meteorology.  
This variability in concentration should be taken into account, 
particularly for agents that exert immediate effects on eyes or 
airways, and for those that induce both short-term and long-
term effects.  Therefore, both the distribution of concentrations 
over time and the toxicodynamics should determine the strategy of 
exposure assessment.  Larsen (1970) observed that, for many ambient 
air pollutants (carbon monoxide, hydrocarbons, nitric oxide, 
nitrogen dioxide, oxidants, and sulfur dioxide), the concentration 
can be described by a mathematical model with the following 
-   concentrations are approximately log-normally distributed
    for all pollutants in all cities for all averaging times;

-   the median concentration (50 percentile for all averaging
    times) is proportional to averaging time raised to an
    exponent; and

-   maximum concentrations are approximately inversely
    proportional to averaging time raised to an exponent.

    Two parameters may adequately describe exposure over a period 
of, say, one year:  for example, 50 percentile and 95 or 98 
percentile for 1 h or 24 h concentrations.  On log-probability 
paper, the percentiles follow a straight line.  This method of 
presentation allows easier interpretation of data than the 
corresponding use of geometric average and standard deviation.  
This subject has been discussed in greater detail in WHO (1980a).  
Ideally, assessment of exposure to these pollutants should be based 
on the percentile distributions of averages over 24 h or less, but, 
in practice, arithmetic means over months or whole years are often 
used as indicators of long-term exposures.  Whether the basic 
sampling period should be 1 h, 8 h or 24 h depends on the type of 
health effects studied.  In epidemiological studies, data with 
high time resolution, but moderate precision, may be more valuable 
than those with high precision, but low time resolution.  Representativeness

    It is essential to obtain exposure data that are representative 
of the exposure of the population at risk (section 3.9).  Although 
this statement may appear to be self-evident, it still needs to be 
re-emphasized.  Many studies are based on estimated exposure from 
data obtained at monitoring sites selected for regulatory purposes 
rather than for estimating the exposure of the population.  More-
over, sites tend to be selected at which relatively high 
concentrations are expected.  Sampling points are often placed at a 
much higher level than the human breathing zone.  Many sampling 
stations are erected at, or near, research institutes for the sake 
of convenience.  A single site may be assumed to represent a large 
area and the number of sites is often limited because of budgetary 
restrictions.  Modelling techniques are only a partial answer to 
the measurement of actual exposure. 

    If data corresponding to the true exposures are to be obtained, 
a monitoring system must be established that is especially designed 
for the study.  With static sampling, it is possible to measure air 
quality at fixed sites.  However, even in occupational settings, 
people move about and therefore are exposed at various work sites; 
exposure may occur in corridors, canteens, offices, or even in the 
vicinity of the industry.  Non-occupational indoor monitoring has 
seldom been carried out.  Thus, total exposure is often either 
underestimated or the indoor exposure is missed entirely, as for 
example in the case of formaldehyde (National Academy of Sciences, 
1981).  It is an enormous task to derive true time-weighted 
exposures by means of static sampling (section 3.11).  Two methods 
are available for approximating the true exposure more adequately: 
personal sampling (section 3.6) and biological monitoring (section 

3.5.3.  Monitoring of pollutants in food and water

    The principles discussed in section 3.5.2 for monitoring air 
apply equally to the assessment of exposure by ingestion of food 
and water.  However, the variability in actual exposure is likely 
to be much larger in the case of ingestion than in the case of 
respiratory exposure, because, in the same ambient or occupational 
environment, subjects inhale the same air, but the intake of food, 
water, and beverages is purely a personal matter.  Consumption of 
contaminated food or beverages constitutes a variable part of total 
food and water consumption.  In addition, cleaning, washing, and 
cooking may change the concentration of contaminants considerably. 
Therefore, assessment of exposure to contaminants in food and water 
has to take into account the individual habits in food preparation 
and in the choice of various foods and beverages.  Furthermore, 
people may consume contaminated food and water that have been 
brought from outside, even though local food and water may be 
clean.  "Ready food" may constitute a mixture from various sources. 

    Water monitoring can be simple, through the frequent evaluation 
of coliform organisms, or more complex, as for trace metals and 
organic compounds such as halothane, polychlorinated biphenyls 
(PCBs), ketones etc.  Some chemical analyses can be performed in 
routine laboratories, while others require more specialized 
instrumentation, e.g., gas-liquid chromatography/mass spectrometry 
equipment (WHO, 1983). 

    In practice, reliable and representative data are difficult to 
obtain, particularly in areas where the population consumes a 
heterogeneous diet, where family units are not very uniform, or 
where the same element may be distributed throughout many items of 
the diet.  Within a population, cultural habits and availability 
largely affect the choice of food and beverages.  Consequently, 
many approaches have been adopted with wide differences in accuracy 
and representativeness.  Overall exposure is a function of 
concentration, amount, frequency of intake, and duration.  All 
dietary studies should be devised to enable such data to be 
obtained.  In food consumption, the emphasis is placed on long-term 
exposure.  In such a case, the time resolution and frequency of 
sampling may be less important. 

    Food and beverages may contain chemicals which as such have no 
nutritional value: 

-    intended additives:  added to obtain or change certain
    qualities, e.g., colouring agents, emulsifiers,
    sweeteners; these regulated chemicals will not be

-    accidental additives:  entering food, water, or beverages
    from containers, transportation accidents, etc.; and

-    incidental additives:  present in original raw food or in
    water; pesticides, fertilizers, fungi (e.g., aflatoxins),
    naturally-occurring chemicals, fall-out, etc.  In the case
    of breast milk the mammary excretion of pollutants, such
    as polychlorinated biphenyls, has to be considered.

    Various approaches are being followed in the assessment of 
exposure.  Overall assessment of dietary intake of toxic elements

    Information is collected about the types and quantities of 
food/beverages consumed, so that it is representative of national 
consumption patterns or those of subgroups within the population.  
National data have been obtained by surveys based on: 

-   The total amount available per person, as an annual
    average, from information on the amounts of food and
    beverages produced, after adjustment for imports and
    exports (FAO, 1971; OECD, 1973).  Correction is necessary
    for food wastage, subsistence food production, use of food
    for animal production, and non-food uses.

-   The purchase of food or the amount of food entering a
    representative sample of homes in a given week, during
    each season of the year (FAO, 1962).  This again only
    allows calculation of average food purchases, and requires
    a knowledge of the wasted amount associated with culinary
    preparation and the amount of left-over food.

-   Questionnaires, interviewing, or weighing all the food,
    beverages, and water being consumed over several days
    (Marr, 1971; Haring et al., 1979).  This is the only
    approach by which information on individual consumption
    can be obtained and which therefore provides the most
    accurate account.  Exposure to specific chemicals can then
    be assessed, if data are available on the overall
    concentration of the substance under discussion in
    foodstuffs and beverages.

    In their review of studies on the relationship between organic 
chemical contamination of drinking water and cancers, Wilkins et 
al. (1979) discussed the pitfalls to avoid in such studies.  
Amongst the pitfalls are:  the chosen indicator agents (e.g., 
chloroform), which may not necessarily represent potential 
carcinogens; non-uniform distribution of contaminants in place and 
in time; no data available on levels in past years; routinely-kept 
records are not adequate for generating "ideal" exposure data; 
aggregate migration data may be an inadequate index of mobility; 
classification of individuals on the basis of residence may be 
inaccurate with respect to total exposure; considerable individual 
variation in water consumption; consumption of bottled water; and 
differences between administrative districts and water-distribution 

areas.  All these pitfalls point to one main difficulty - the 
linking of individual exposures to all potential drinking-water 
carcinogens over a long period, with medical histories. 

    In epidemiological studies on relationships between the 
hardness of drinking-water and cardiovascular mortality or 
morbidity, the composition of mains-water has generally been used 
as the indicator of exposure.  However, an area may be served by 
several water sources, resulting in variable composition.  The 
Water Research Centre in the United Kingdom has worked out various 
mathematical procedures to achieve an estimate of the weighted mean 
for the population under study over a certain period (Lacey & 
Powell, 1976).  Formulae can be applied to a group of important 
water determinants, or simply element by element.  It may show that 
homogeneity exists for one element, but not for another.  For the 
study of long-term exposure, it would be preferable to define 
areas, served by one water plant, with water of a reasonably stable 
composition over at least 10 years.  Also, changes in water 
composition may be taken as a criterion for change in both exposure 
and health hazard, as has been done by Crawford et al. (1971). 

    There is a large variation in water composition according to 
source.  In the United Kingdom, for instance, the pH was shown to 
vary from 3.6 to 7.9 (waters with low pH being more liable to 
dissolve material from metal pipes); total hardness (CaCO3) ranged 
from 16 to 270 mg/litre, total dissolved solids from 77 to 600 
mg/litre and nitrates from 0.5 to 4.0 mg/litre (Packham, 1978).  In 
recent years, more account has been taken of the composition of 
tap-water, particularly the levels of cadmium, copper, lead, and 
zinc, which may change between the water mains and the tap, 
depending on pH and type and length of the home distribution 
system.  Haring (1978) has developed a proportional sampler for 
tap-water; 5% of the water flowing through the tap is sampled over 
a whole week.  The consumers are instructed to turn on the sampler 
only when water is taken for preparation of food and drinks; this 
yields the average intake of water (and pollutants) per household 
per week. 

    In comparing the mortality or morbidity of population groups 
from different towns, weighted intakes representative for those 
respective populations have to be estimated.  This calls for a 
number of random samples in each town, increasing with the size of 
the population.  In the Netherlands, it has been estimated that for 
towns of 20 000-50 000 inhabitants, tap-water in about 200 homes 
should be sampled to achieve a representative weighted concentration 
for metals that are liable to increase in concentration from source 
to tap.  For pollutants that do not change in concentration, sample 
water at the outlet of the water treatment plant can be used 
(Zielhuis & Haring, 1981).  Indirect assessment of intake

(a)  Total diet or market basket studies (composite technique)

    Food samples are prepared that are composed of the main 
constituents such as cereals, meat, root vegetables etc., based on 
national consumption data.  They are analysed after normal 
preparation and cooking.  The mean concentration of toxic elements 
in each constituent is measured and an average daily intake can be 
calculated for each constituent and for the diet as a whole.  Such 
studies are repeated for different seasons and in different regions 
to reflect local variations in the diet.  A number of countries 
have based their initial assessment of population exposure on data 
obtained from such studies (Ushio & Doguchi, 1977; Dick et al., 
1978).  These studies are particularly valuable when elements 
(e.g., lead, cadmium) are widely distributed amongst all major food 
items, or, as is the case with mercury and arsenic, where bio-
concentration occurs almost exclusively in fish and shellfish. 

(b)  Selective studies on individual foodstuffs

    By measuring concentrations in representative samples of staple 
foods, it is possible to use the modal levels found, together with 
food consumption data, to calculate average daily intakes.  Such an 
approach is particularly useful, if the intake is predominantly 
influenced by one or two items of food, and where food monitoring 
programmes have established an average concentration in a commodity, 
e.g., DDT in cereals.  The following three groups of consumers may 
require special attention (section 3.9):  those who have different 
patterns of food consumption from ordinary adults (e.g., infants or 
the elderly); those whose metabolism is different from ordinary 
adults (e.g., infants who normally absorb lead from the gastro-
intestinal tract at a higher rate than adults); and those who are 
exposed to an above-average concentration of toxic chemicals in the 
diet (e.g., fishermen on tuna boats - methylmercury). 

(c)  Habit survey (nutrition table method)

    A sample is selected within a critical population to obtain 
information about the food consumption of extreme consumers.  In 
the United Kingdom, such an approach has been used to determine the 
consumption habits of a critical population by interview; from this 
a reference level of consumption of the most extreme consumers is 
calculated.  This reference level is given as the arithmetic mean 
of the consumption rates of a fixed number (usually 30) of the 
people at the upper end of the distribution of consumption rates. 
Such a reference level has been shown to reflect reasonably the 
time-weighted average consumption levels of the most extreme 
consumers (Shepherd, 1975), though recent duplicate diet studies 
(section have shown that consumption rates determined by 
interview are usually an overestimate of actual consumption rates 
(Haxton et al., 1979).  However, the use of this method enables 
consumers to be identified, who are subject to unacceptable 
exposure because of their patterns of food consumption, and/or to 

increased metabolic susceptibility to the toxic element (section 
3.9).  In these cases, further direct assessment of exposure must 
be undertaken.  Direct assessment of intake

    The external dose, in a narrower sense, can only be obtained by 
the weighing and analysis of a duplicate sample of meals actually 
consumed by an individual, including those consumed outside his 
home (duplicate portion technique). Practical constraints invariably 
limit the application of this approach to the collection of meals 
for one or a few weeks from a limited number of subjects.  Although 
the exercise can be repeated, the demands on individual participants 
are quite high and, ideally, the survey should be supervised.  
Consequently, such studies are only undertaken if the information 
from indirect exposure assessment techniques is such that; 

 -  an average intake is not appreciably lower than the 
    acceptable or tolerable intake;

-   there is a well-defined critical group within the
    community; and

-   the community is subject to an atypical level of
    contamination in the area, in which they live, or in their

    In a sense this approach is comparable with the methods of 
personal sampling described in section 3.6. 

3.5.4.  Monitoring of physical factors  Noise

    Reviews on the assessment of exposure to noise have been 
published by Broch (1971), Burns (1973), Persons & Bennett (1974), 
Peterson & Gross (1974), Lipscomb (1978), and WHO (l980b). 

    Noise (i.e., unwanted sound) does not leave a residue; it 
dissipates as soon as the vibrating source is discontinued.  It is, 
however, very pervasive.  Environmental assessment of exposure is 
possible but not biological assessment. 

    Noise occurs almost everywhere in a highly-mechanized society.  
Occupational exposure to noise is widespread in the production 
industry, in transportation, construction, mining, and even in 
agriculture.  Non-occupational exposure to noise occurs more 
extensively in urban than in rural environments.  Aircraft are 
often regarded as the most annoying source of community noise.  In 
addition, there is increasing exposure during recreation, e.g., 
from discotheques, shooting, or motorcycling.  In remote undeveloped 
areas, noise levels are much lower than those in industralized 

    Noise is characterized by three basic parameters:   frequency, 
 level  (intensity), and  duration.  The frequency is the number of
oscillations per unit of time, stated in terms of cycles per
second (Hertz).  Most common noises contain a complex combination 
of frequencies.  High frequency noise, in which the energy is 
concentrated in a relatively narrow band, is generally more 
annoying and damaging to the ear than low frequency broad-band 
noise.  Noise level, measured in decibels (dB), is perceived as 
loudness; the noise level and duration of exposure are closely 
correlated with adverse health effects.  In exposure assessment, it 
is necessary to quantify level (dB) and duration, and to some 
extent frequency. 

    The type of noise should also be distinguished according to the 
way it occurs in time.   Continuous noise maintains a fairly
steady level over time.   Intermittent noise is caused, for
example, by vehicles or aircraft passing by.   Impulse or impact 
noise (high level, short duration) is generated by two objects
striking together or by a sudden, forceful release of air pressure: 
e.g., gunfire, sonic boom, explosions, hammering.  The time 
resolution of sampling should be geared to these different 

    One of the easiest, though not the most precise method, is to 
use one's own ears.  The ear is extemely sensitive and capable of 
interpreting sound over a broad range of intensity.  In industry 
the following estimate is used as a "rule of thumb", on the basis 
of being able to understand the spoken language:  if loud speech 
can be understood at 0.8, 0.45, 0.25, 0.14, or 0.08 m, the noise 
level is 65, 70, 75, 80, or 85 dB(A), respectively (footnote to 
section 3.3.1). This is an example of assessment of perceived 

    However, instrumentation, such as sound level meters, impulse 
noise meters and personal dosimeters (section 3.6) are needed in 
order to quantify exposure.  Accurate measurements are essential.  
All measurements should be conducted and calibrations performed 
according to the accepted standards, such as those of the 
International Organization for Standardization (ISO).  Since 
frequency and duration are also essential parameters in the 
assessment of exposure, equipment has been devised that incorporates 
these characteristics.  Some equipment gives direct measurement, as 
in the case of frequency weighting networks built into sound level 
meters; with other equipment, calculations are required from a 
knowledge of the time pattern. 

    In occupational studies, in addition to a sound level meter, a 
work study is needed to calculate the duration of exposure; this 
demands a large number of measurements and a close follow-up of the 
worker over the entire workshift.  However, the recovery of the 
temporary auditory threshold shift, caused by exposure to noise
> 80 dB(A), will depend on the noise exposure during commuting, in 
the home, or during recreation.  In epidemiological studies, the 
overall exposure per day should be assessed.  More sophisticated 
equipment is available that makes use of tape recorders, which can 
be taken to a site (workplace, community site); the tapes can 

afterwards be analysed in a laboratory, so that a complete history 
of noise can be obtained on the site, showing noise level and 
spectral characteristics at various times.  The noise can then be 
described statistically as the amount of time that it exceeds a 
certain level (for example, L10 means that the level is exceeded
for 10% of time). 

    In epidemiological studies on the effects of aircraft noise, 
contour noise level lines, computed for areas around an airport can 
be used; exposure of the general population can then be expressed 
in terms of location of homes (and communities) on this contour 
line map.  Vibration

    Reviews of this topic have been prepared by Dupuis (1969), 
Coerman (1970), Guignard & King (1972) (particularly vibration in 
aeroplanes) and Wasserman & Taylor (1977). 

    Vibration is a series of reversals of velocity:  both 
displacements and accelerations take place.  Vibration may be 
defined as any sustained oscillating disturbance that is perceived 
by the senses (Guignard & King, 1972).  Distinction can be made 
between deterministic (i.e., the variation can be predicted), non-
deterministic (random), and transient vibration (short-term). 

    Contact with vibration can be regarded as:

-    intended, e.g., during work (or at home), medical
    treatment, and nursing; or

-    unintended, e.g., as a passenger in cars, aeroplanes,
    etc., living in a house situated near factories or busy
    traffic routes.

    The health effects of vibration on workers are related to the 
duration and to the intensity of exposure: 

-   vibration transferred to the body from tools or machines
    through the upper limbs or other parts:  local vibration;

-   vibration transferred from the base, e.g., vibrating
    platforms, through muscles and pelvic bones:  whole-body

    From the physical point of view, vibration is a complex 
oscillatory movement of a particle or a body with respect to a 
given reference point; the movement is transferred in transverse 
and longitudinal waves. 

    The least complicated form is simple vibration, also known as 
harmonic vibration, mathematically represented as a sinusoidal 
curve.  The maximum deflection of a particle from its state of 
equilibrium is called the amplitude, measured in cm, mm, or µm.  
The overall deflection in both directions, performed by oscillatory 

particles in a given time, is called the vibration cycle.  The 
number of cycles of full vibration per second is called frequency 
and is measured in Hertz (Hz). 

    In practice, vibration is a complex of periodic movements 
composed of many sinusoidal curves.  Therefore, the amplitude 
diagram as a function of time is not sufficient for describing the 
number, character, and frequency of its components.  The value that 
best characterizes vibration is the root-mean-square value (RMS), 
because it accounts for both the time and the magnitude of the 

    In assessing human exposure to vibration, four basic physical 
parameters have to be considered, i.e.,  intensity, frequency, 
 direction, and  duration.  In the case of local vibration, the
quantity and direction of forces employed by the operator in 
touching the tools or working materials, the position of upper 
limbs or the position of the whole body, the type of vibrating 
tool, climatic conditions, work methods, and energy consumption 
should also be taken into account.  In the case of whole-body 
vibration, it is necessary to take into account the position of the 
body, the direction of vibration, and microclimatic conditions. 

    Modern apparatus for the measurement of vibration is equipped 
with an electronic integration system enabling measurement of 
acceleration, velocity, and deflection.  Experiments have shown 
that the value of RMS of the amplitude is the best characteristic 
of vibration in the range of 10-1000 Hz.  In order to examine 
individual components of a wide-band signal, it is necessary to 
perform the analysis of frequency in one-third of an octave. 

    The measurement of the direction of the penetration of 
vibration is of great importance in the case of whole-body 
vibration.  It is known that the human body is sensitive to 
vibration directed parallel to the long-body axis. 

    In epidemiological studies, it is not necessary to make a 
detailed spectral analysis of vibration emitted from various 
sources.  It is possible to conduct the assessment by measuring a 
single parameter - the frequency weighting value of acceleration. 

    Guignard & King (1972) have presented a review of the 
subjective assessment of exposure to vibration.  Ionizing radiation

    The ionizing radiations to which man is exposed can be electro-
magnetic such as X- or gamma-rays or corpuscular radiation such as
alpha- or beta-rays.  Methods for the assessment of exposure have
to be extremely sensitive, because, for ionizing radiation, it is
assumed that a no-adverse-effect-level does not exist. 

    These radiations can be emitted by radioactive elements 
(radionuclides), the presence of which in the environment of man is 
likely to lead to exposure.  Exposure may occur within all levels 

of the environment (domestic, occupational, local, or regional) 
(section 3.1).  Radon progeny generation from soil, rocks, and 
building materials is an example.  Thus, it is necessary to assess 
the total exposure to the various types of ionizing radiation. 

    A number of radionuclides are of natural origin and are always 
found in the environment; they lead, together with contributions 
from cosmic rays, to background radiation.  The total background 
radiation varies according to altitude and longitude.  Other 
radionuclides are man-made, especially those derived from the use 
of fission reaction as a source of energy, but exposures to 
ionizing radiation may occur also from diagnostic and therapeutic 
appliances and from some home equipment such as luminous watches. 

    As with a stable element, a radionuclide is characterized first 
by a number of chemical properties, by the physical or physico-
chemical form in which it is found, and finally by its behaviour in 
biological media, chiefly its metabolic properties.  It also has 
particular nuclear characteristics, namely, its disintegration rate 
(represented by its radioactive half-life corresponding to the time 
necessary for the disintegration of half the atoms present) and the 
nature and energy of the emitted radiation. 

    The human body or some of its tissues or organs may be exposed 
to radionuclides in two different ways, namely, externally or 

    External exposure may result from radionuclides present in the 
environment outside the body.  This is especially the case when 
irradiation results directly from the source itself such as a 
nuclear plant.  This kind of exposure is almost exclusively 
occupational.  In addition, there may be external exposure from 
diagnostic or therapeutic X-rays.  It chiefly involves X- and 
gamma-rays, which are penetrating radiations and therefore likely 
to reach tissues lying at a distance from the point of emission. 

    Internal exposure occurs when radionuclides have been absorbed 
by inhalation, ingestion, or percutaneous transfer.  It involves 
both penetrating gamma-radiation and the much less penetrating 
alpha- and beta-radiations.  After uptake, the radionuclides may 
remain local (e.g., dust inhaled in mines) or may be distributed 
throughout the body, according to the normal kinetics of the 
element.  It is essential to distinguish between the two modes of 
exposure, since different methods of measurement and means of 
protection apply. 

    Tissue damage is measured by the energy absorbed at the level 
of the tissue, taking into account the type of radiation involved;  
it is called  "dose equivalent" and used to be expressed in "rem"
but this was replaced in l975 by the joule per kilogram (l rem = 
10-2J/kg); more recently the sievert (1 rem = 10m Sv) has come
into use. 

    In the case of external exposure, many techniques make it 
possible to measure the maximum dose equivalent received by the 

organism directly from the radiation source itself.  This 
measurement can be carried out using well-developed and sensitive 
techniques:  ionization chambers, scintillation counters, thermo- 
or photoluminescent dosimeters, etc.  Workers likely to be exposed 
can be equipped with direct reading personal samplers (but with-out 
subsequent chemical analysis) (section 3.6). 

    In the case of internal exposure, dose equivalents cannot be 
measured directly.  The methods used consist of assessing exposure 
from the evaluation of incorporated activity, derived from the 
direct measurement of radioactivity in air, drinking-water, and 
various foodstuffs.  Many methods are available to determine such 
radioactivity either directly on a sample, or after its physical or 
chemical treatment, with sensitivities and accuracies that can 
seldom be achieved when measuring other hazards.  Non-ionizing radiation

    Non-ionizing radiation refers to all radiation in the electro-
magnetic spectrum exclusive of the ionizing range.  It includes 
the various forms of light waves, microwaves, and radiowaves.  
It is part of the natural atmospheric background radiation to 
which all living things are exposed to a varying degree.  As a 
result of technological advances in recent decades, man-made 
electronic sources have added greatly to the environmental levels 
of some forms of non-ionizing radiation in parts of the world.  
The health significance of such exposures is related to the 
physical characteristics of the radiation, the conditions and 
duration of exposure, and the characteristics of the persons at 
risk.  Reviews on exposure assessment have been prepared by Czerski 
and collaborators (1974), Scotto and co-workers (1976), WHO (1979), 
and WHO (1981). 

     Light radiation includes the ultraviolet, visible and
infrared wavelengths of the electromagnetic spectrum.  All are 
found in various proportions in sunlight.  These wavelengths may 
also be emitted by man-made products or processes; in the case of 
laser devices, the emissions are coherent monochromatic beams of 

    Exposure is measured as radiant energy.  Biological indicators 
of human exposure are changes in the eye and skin, the principal 
organs that absorb light.  Ultraviolet radiation (UVR) is most 
important from the standpoint of human health hazards.  The UV 
spectral range includes three regions (UV-A) extending from near to 
far UVR that are referred to as:  UV-A (black light region), UV-B 
(erythemal range that is believed to be instrumental in producing 
skin cancer), and UV-C (germicidal region).  UVR is an important 
factor essential for the normal functioning of the body:  a 
deficiency not only leads to specific adverse effects such as 
disturbance of the phosphorus/calcium metabolism and rickets, but 
there is evidence that it also reduces resistance to chemical 
substances (Zabalyeva et al., 1973; Prokopenko, et al., 1981). 

    By making UVR measurements at specified times in several 
locations, quantitative information can be obtained on the 
association between solar UVR exposure and cancer of the skin.  
This requires assessment of exposure intensities and durations.  
Meteorological data combined with interviews on personal habits 
(sunbathing, home-treatment, gardening, vitamin consumption, etc.) 
may give an approximate estimate of exposure.  Ethnic groups may 
differ in susceptibility, which is greater in light-skinned races 
than in dark-skinned; hereditary predisposition also exists 
(xeroderma pigmentosum).  A review on ultraviolet radiation has 
been published by WHO (1979a). 

    Visible light from various types of lamps is used for photo-
therapy in hyperbilirubinaemia of the newborn.  Combined drug/ 
light therapies have been developed for certain skin discorders. 
Standardized conditions for such therapeutic exposures, including 
specific wavelength limits, have not yet been achieved and 
exposures are uncertain.  Combined exposure to volatile tar 
products and sunlight (e.g., among road-workers) has been shown to 
lead to synergistic skin effects. 

    The principal sources of  microwave (MW) and  radiofrequency
(RF) radiation are electronic devices that generate and transmit 
these frequencies.  Electromagnetic pollution is becoming world-
wide because of the universal use of radar, heating techniques, 
telecommunications, and broadcasting systems.  Exposure to MW/RF 
fields is generally assessed by the the measurement of average 
power density under specified conditions.  In the case of some 
sources, such as radar, peak power density may also have to be 
measured.  The principles of measurement, together with a review of 
effects on man, have been discussed in a recent WHO publication 
(WHO, 1981).  Dosimetry is complex and international standardization 
of measurement techniques has not yet been reached. 

     Ultrasound is conventionally included in many non-ionizing
radiation programmes.  Ultrasound-emitting devices are used for 
therapy and most extensively for diagnostic imaging among various 
medical and industrial applications, and in consumer products.  
Average output power is measured for two types of ultrasonic 
exposures:  continuous wave and pulsed.  Little is known about the 
distribution or absorption of energy in man (WHO, l982b). 

    Another physical agent that might be considered are power 
frequency electric fields (i.e., fields around power cables 
operating at high voltage with frequencies in the range 50-60 
Hertz).  The physical implications of exposures of human beings to 
such fields have been discussed in a recent review by Bridges & 
Preache (1981). 

3.6.  Personal Sampling

    Environmental monitoring of air (section 3.5.2) has many 
drawbacks in procuring true and representative exposure data for 
groups of subjects.  In recent decades, particularly for the 
assessment of occupational exposures, an alternative method has 

been introduced:  the subject carries a sampling instrument during 
the whole (or part) of the working day with the sampling head in 
the breathing zone.  Within certain limits, sampling time and 
frequency during a working day (8 h) can be adjusted to suit the 
specific situation; however, monitoring with a high time resolution 
is not feasible.  Although the air thus monitored is more 
representative of the air inhaled, the sampling rate does not 
depend on the respiratory volume of the subject; therefore personal 
sampling only gives an approximation of the respiratory intake. 

    The rapid development of personal air sampling instrumentation, 
in recent years, has made it possible to monitor a large variety of 
workroom air pollutants:  metals, solvents, vinyl chloride, etc.  
One of the first epidemiological studies in industry based on 
personal sampling was performed by Williams et al. (1969) on the 
assessment of exposure to lead in an electric accumulator factory.  
Personal sampling methods in occupational health were revised by 
Meyer (1975) and a review related more particularly to the 
assessment of pollutants in the general environment has been 
published by the US Environmental Protection Agency (1979). 

    There are considerable difficulties in using the personal 
sampling technique in community health studies.  These include:  
the large number of subjects involved; at-risk groups may consist 
of young children or the elderly; personal contact between the 
subjects and the research worker is not as close as it would be in 
a factory, and cooperation may not easily be achieved.  Azar et al. 
(1973) performed a study on lead exposure in taxi-drivers in 
various cities in the USA, which showed that, notwithstanding the 
difficulties encountered in assessment of exposure to ambient air 
pollutants, the use of the personal sampling technique appeared to 
be feasible in specific situations.  In planning limited scale 
epidemiological studies, use of this technique should always be 
considered.  A review on progress in this field has been published 
by the US National Academy of Sciences (1981). 

    The personal sampling technique not only applies to the 
monitoring of air pollution, but a similar approach is used in 
assessing exposure to noise (section and ionizing 
radiation (section  A comparable approach is followed in 
the duplicate portion technique of assessing exposure to chemicals 
in food (section 

3.7.  Biological Assessment of Exposure

    The biological assessment of exposure is defined as the 
systematic collection of human specimens for the determination of 
pollutant concentrations (or metabolites).  The joint CEC-WHO-EPA 
Workshop in l977 distinguished between  biological monitoring, 
i.e., "a systematic collection for immediate application; analysis
and evaluation would be performed within a period of weeks after 
collection" and  collection for future reference,  i.e., "a
systematic collection and respository of samples for deferred

examination; analysis and evaluation generally will be deferred for 
a period of years or even decades following collection" (Berlin et 
al., 1979). 

    Sampling and analysis of faeces may be regarded as a 
combination of environmental and biological assessment:  the amount 
excreted per day reflects ingestion (minus absorption) and 
excretion into the gastrointestinal tract, if fractional gastro-
intestinal absorption is low; for some agents (e.g., heavy metals) 
both assessment possibilities exist simultaneously.  Sampling of 
breast milk assesses both the internal exposure of mothers and the 
external exposure of infants. 

    Environmental and biological monitoring are not competitive, 
but, depending on the objective of the study, on the environmental 
factor(s) under consideration, and on the available expertise and 
methods; either or both forms of monitoring may be preferred.  
Sometimes a combined approach is required. 

    A pilot project on the assessment of human exposures to cadmium 
and lead by means of biological monitoring has been undertaken as 
part of the UNEP/WHO GEMS programmea.  The final report (Vahter
et al., 1982) stresses the importance of quality assurance 
procedures in this collaborative study.  With these established, it 
was possible to make valid comparisons of the blood levels of lead 
and cadmium among residents in a number of cities around the world.  
To avoid occupational factors that might affect the results, 
observations were limited to a single group (schoolteachers).  
While this project was not in itself part of an epidemiological 
study, it provides a valuable example of procedures to be observed 
in the biological assesment of exposure. 

    The method for the assessment of exposure to radiation is quite 
different from that for chemicals.  Many gamma-emitting radonuclides 
can be determined directly by  in vivo counting of subjects in a 
wholebody counter; with this method it is possible to identify the 
organs most affected.  Moreover, in most cases chemical contamination 
of reagents and apparatus does not influence the measurement.  In 
addition, unlike chemical contaminants, the elements concerned do 
not undergo changes during metabolism, which may alter their 
analytical behaviour. 

    Excreta (notably urine, faeces, and in a few cases, exhaled 
air) or tissues taken from autopsy are also used for the biological 
assessment of exposure to radiation.  Fall-out surveys include 
analysis of bone samples for strontium-90  and plutonium and 
thyroid samples for iodine-131 and measurement of caesium-137 in 
the total body by  in vivo counting (UNSCEAR, 1972).  Age has to 
be taken into account, because both the metabolism of the particular 
element and the expected exposure/response relationship may vary 
with age.  In contrast to the biological assessment of exposure to 
chemicals, it is usually necessary to examine large samples, up to 
several hundred grams of tissue or excreta, over several days.  

a  GEMS - Global Environmental Monitoring System. 

Because of their rapid decay, short-lived radionuclides must be determined 
quickly and storage of samples is not possible in such cases.  A 
review (with many literature sources) on analytical procedures for 
biological exposure to radiation has been given by Harley (1979). 

3.7.1.  Advantages, disadvantages, limitations

    In biological monitoring, parameters of internal exposure 
(uptake) are measured as the result of external exposure through 
ambient air, workroom air, smoking, indoor air pollution, use of 
cosmetics, contaminated food and water pollution, etc.  Total 
exposure, irrespective of the source of pollution, is measured 
indirectly.  In environmental monitoring for total exposure, on the 
other hand, it is necessary to measure simultaneously all sources 
of external exposure, together with the duration; this is seldom 
possible and may require an enormous input of manpower and money. 

    Internal exposure is the result of external exposure and of the  
characteristics of the subjects exposed.  In the case of respiratory 
exposure, only concentrations in air are assessed in environmental 
monitoring and not respiratory volume, which depends on physical 
activity, whereas parameters of internal exposure may increase with 
higher physical activity.  The health-relevant exposure is much 
better assessed in biological monitoring, because the impact on 
internal exposure of personal behaviour, choice of foodstuffs, 
biological characteristics such as age, sex, interindividual 
differences in absorption and metabolism, disease states, and 
anthropometry are taken into account.  Biological monitoring pin-
points the groups and individuals actually at risk (section 3.9) 
and studies may be possible of much larger numbers of subjects 
than, for example, with personal exposure monitoring. 

    In examining biological specimens, measurements can sometimes 
be made (in addition to parameters of internal exposure) of 
possible health effects, such as changes in blood cells, enzymes, 
proteins, lipids, kidney or liver function.  In addition, it may be 
possible to measure several exposure indices simultaneously, for 
example, various metals in blood, hair, urine, or different 
solvents in exhaled air. 

    In recent years, by analogy with environmental quality guides 
for air, water, or food, much emphasis has been placed on the 
development of  biological exposure limits, namely, acceptable 
concentrations of agents or metabolites in exhaled air, blood, 
urine, etc.  In order to fulfil the requirements for establishing 
such biological exposure limits, it is necessary to conduct 
epidemiological studies on the relationships between environmental 
exposure, biological exposure, and health effects. 

    However, biological monitoring has limitations in comparison 
with environmental monitoring.  The main drawback is inconvenience 
to the subjects; ethical aspects must be considered carefully 
before starting such a programme.  These may include questions of 
inconvenience, health risk, confidentiality of information, and 
freedom to refuse participation.  Moreover, biological monitoring 

can be applied only in the case of compounds that are taken up by 
the body.  It cannot be applied in the case of several highly 
important environmental pollutants that exert their effects 
primarily at the point of absorption (for example, sulfur dioxide, 
nitrogen dioxide, ozone, and oxidants), or in the case of noise or 
ionizing radiation.  In general, biological monitoring is not 
suitable for registering highly variable exposure, which may 
constitute a limitation of this method of monitoring, if systemic 
health effects depend on internal peak exposures (very short 
biological half-time).  In addition, for many chemical agents, not 
enough basic data are available to design a biological monitoring 

    To sum up, exposure can be estimated in two ways:  (a) by 
assessing the external (environmental and/or occupational) 
exposure, which only approximates the actual external dose; and 
(b) by assessing the internal exposure, which approximates the 
actual dose at target organs and provides a better estimate than 

    The recent publication from the l977 CEC-WHO-EPA Workshop 
(Berlin et al., l979), together with those of Zielhuis (1973), and 
Aitio and coeditors (1981) present a considerable amount of 
information on instrumental, analytical and organizational aspects 
of programmes for the assessment of environmental and occupational 
exposure by means of biological monitoring. 

3.7.2.  Collection for future reference

    This new approach was also extensively discussed in the CEC-
WHO-EPA Workshop (Berlin et al., 1979) and by Leupke (1979).  With 
repositories of samples, retrospective studies may be possible to 
ascertain whether a pollutant observed in body tissues (or in 
environmental samples) at a certain time, already existed in 
earlier years, before its occurrence was investigated.  Moreover, 
with improvements in analytical techniques, it may be possible to 
determine concentrations too low to be determined previously.  In 
some countries such as the United States of America and the Federal 
Republic of Germany, pilot schemes are being developed to organize 
such repositories.  Many problems regarding organization, costs, 
storage of various types of specimens, analysis, and evaluation 
have still to be solved, before the data can be incorporated in 
epidemiological studies.  Some work has been done on the freezing 
of aliquot diets for the subsequent determination of contaminants, 
such as aflatoxins that may have, or later may be shown to have, 
carcinogenic properties.  In general, however, it should be said 
that there is little point in "banking" all kinds of specimens for 
the future, unless there is some reasonably well-defined object in 

3.7.3.  Index specimens for various pollutants

    The CEC-WHO-EPA Workshop (Berlin et al., 1979) prepared a 
comprehensive table showing major environmental pollutants of 
concern from the public health point of view and various human 
tissues, organs, and fluids, which might be considered for 

collection, either for biological monitoring or for collection for 
future reference.  A summary is presented in Table 3.2, covering 
pollutants and specimens of major importance.  The Workshop 
concluded that the most important pollutants for which biological 
monitoring programmes could be implemented, immediately, in order 
to assess exposure of the general population, were:  arsenic: 
blood, urine, hair;  cadmium:  blood, urine, faeces, kidney,
liver, and sometimes placenta;  chromium:  urine;  lead:  blood,
urine, hair, faeces, kidney, liver, bone, and sometimes placenta; 
 inorganic mercury:  blood, urine, kidney, brain;  methylmercury: 
blood, brain, hair;  organochlorine pesticides:  adipose tissue,
blood, milk;  pentachlorophenol:  urine,  polychlorinated biphenyls: 
adipose tissue, milk, blood;  chlorinated solvents:  blood, expired
air, and sometimes urine;  benzene:  blood, expired air;  carbon
 monoxide:  blood, expired air. 

3.7.4.  An example of environmental versus biological assessment
of exposure:  inorganic lead

    The data are mainly based on reviews by Zielhuis (1975), 
Nordberg (1976), and WHO (1977a). 

    The potential contribution of airborne lead to total lead 
intake is presented in Fig. 3.1 (WHO 1977a):  6 of the 7 pathways 
to man point to ingestion.  This clearly indicates that in 
epidemiological studies, environmental assessment of exposure will 
require an enormous and complicated effort to assess lead levels 
through ingestion of dust and soil (through pica) and through 
consumption of water, beverages, vegetables, and animal food, as 
well as by direct inhalation.  Moreover, there are some other 
sources that are not included in Fig. 3.1, for example, the use of 
certain cosmetics containing lead and the use of lead-glazed ceramics. 


Table 3.2.  Index specimens to be used in the biological assessment of exposurea
            As Cd Cr F Pb Hg MHgb DDT and  Phenoxy PCB,  Chlori-  Fluori- Non-substi- Alcohols Organo- CO     
                                  organo-  herbi-  PBBc  nated    nated   tuted aliph.         phos-           
                                  chlorine cides         solvents prope-  arom. vola-          phorus          
                                  pesti-                          llants  tile hydro-          esters          
                                  cides                                   carbons                              
Adipose                           +                +                                                   0      
Blood       +  +       +  +  +                           +        +       +           +        +       +      
Bone        0     0  + +          0        0       0     0        0       0           0        0       0      
Brain                     +  +                                                                         0      
Expired     0  0  0  0 0  0  0    0        0       0     +        +       +           +        0       +      
Faeces         +       +                                          0       0                            0      
Hair,       +          +     +                                            0           0                0      
Kidney         +       +  +                                                                            0      
Liver          +             +                                                                         0      
Milk           0                  +                +                                                   0      
Placenta       +       +                                                                               0      
Teeth       0        + +     0    0                0     0        0       0           0        0       0       
 cord blood            +  +  +    +                +                                                   +      
Urine       +  +  +  + +  +  0             +             +                +           +                0      
a From: Berlin et al. (1979).
b methylmercury.
c polychlorinated and polybrominated biphenyls.
+ pollutants and specimens for which there exists sufficient information
  to suggest a valid biological monitoring approach.
0 such an approach is not (yet) feasible.
    Where lead is used in petrol, it is generally the main 
contributor to lead concentrations found in air, which, 
consequently, vary sharply with the distance from busy streets and 
the amount of automotive traffic on them.  Concentrations may also 
be elevated around smelters or other local sources (such as scrap-
metal yards). 

    Lead levels in drinking-water are usually low, except when soft 
and/or acidic water flows through lead pipes; this may constitute 
a serious health hazard in relation to private water wells, and 
also in some public distribution systems.  Studies in Glasgow 
(Scotland) and in Verviers (Belgium) reported levels of lead up to 
2 mg/litre.  In exposure assessment, it is important to note water-
drinking habits:  water left standing overnight in the pipes 
usually contains relatively high lead levels and infant formulae, 
morning coffee or tea made with this may be highly contaminated. 

    Many studies have been carried out measuring intake from food 
according to methods discussed in section 3.5.3.  Using duplicate 
portion techniques, a daily intake of about 80-350 µg/day have been 
established for adults (Finland, United Kingdom, USA); higher 
levels (up to 500-600 µg/day) have usually been indicated using 
composite techniques (Federal Republic of Germany, Italy, Japan).  
The washing and processing of food may considerably reduce existing 
lead contamination.  On the other hand, vegetables cooked in lead-
containing water may take up some from that source.  If it is 
assumed that foodstuffs with a lead content below detection limits 
contain a zero lead level, then the total exposure may be 

    On the basis of the assumption that 90-95% of orally-ingested 
lead is not absorbed in the intestinal tract, intake can also be 
measured indirectly from lead levels in the faeces.  This method 
has been used by Tepper & Levin (1972) who established an intake of 
90-150 µg/day in adult females in the United States of America. 

    A matter for concern may be the lead content of milk, 
particularly for infants; human breast milk has been reported to 
contain < 5-12 µg/litre, cow's milk, 9 µg/litre; and processed 
cow's milk, higher levels than fresh milk.  Canning and packaging 
may lead to contamination of foods and beverages.  Another source 
may be wine:  it may become a substantial source of exposure in 
countries in which the habit is to drink 1-2 litres of wine daily, 
as occurs in France and Italy, because of contamination from lead-
containing caps.  There are further risks from wines prepared at 
home in lead-glazed vessels. 

    Tobacco smoking has been shown to increase exposure, probably 
because of the use of lead-containing pesticides, but, because the 
practice of using such pesticides has been abandoned to a large 
extent, this source may become less important.  A crude estimate of 
uptake is 1-5 µg/day from smoking 20 cigarettes/day.  However, in 
occupational exposure, smoking during work may considerably 
increase oral intake, through transfer from contaminated fingers. 

    Children undoubtedly constitute a group at high risk (section 
3.9).  Exposure to contaminated soil and dust and to lead-based 
paints has been responsible for large epidemics of lead poisoning 
in young children, particularly in the USA.  This is related mainly 
to indoor paint, but may also occur from outdoor paints and dust.  
The hands of inner-city children are often more contaminated than 
those of suburban children.  The presence of lead in the home 
environment is related to the socioeconomic and sociocultural 
status of the family, and the behaviour of children also affects 
exposure through personal hygiene (cleansing hands) or pica 
(pathological mouthing).  Another source of exposure for young 
children is coloured newsprint; coloured pages have been found to 
contain lead concentrations of 1140-3170 mg/kg.  In section 3.10, 
it is shown that occupational exposure of parents may also increase 
the exposure of children. 

    Many epidemiological studies carried out so far have been 
concerned with the assessment of exposure of, and possible effects 
in, schoolchildren (e.g., 6-14 years of age), apparently because 
they can more easily be approached and because the neuropsychological 
tests required can be carried out more readily with this age group 
than with younger age groups.  However, it has been shown repeatedly 
that young preschool children (2-4 years of age) have the highest 
exposure, because of their behaviour.  In Asian populations (and in 
those migrating to other countries), cultural habits may increase 
exposures, for example, through the use of lead-containing cosmetics. 

    This short review indicates that, in epidemiological studies on 
segments of the general population, environmental assessment of 
exposure to lead is an almost impossible task, if it is desired to 
achieve a valid estimate of total dose.  Even in occupational 
studies, in which respiratory exposure is usually predominant, 
workroom levels are poorly-related to lead levels in blood.  
However, particularly in the last two decades, biological 
assessment of exposure (section 3.7) has proved to be of great 
value, both in the general and in the occupational environment.  Lead in blood (Pb-B)

    For long-term steady exposures, blood-lead is a valid indicator 
of total exposure over the previous few months.  In blood, 90-95% 
of lead is located in the erythrocytes.  In the case of anaemia, 
whole-blood lead levels may underestimate actual exposure; this can 
be corrected by means of the haematocrit. 

    Pb-B levels do not provide direct information on the existence 
of health effects.  However, they can be used as an estimate of 
exposure in exposure/effect relationships.  Blood  per se is not
the target organ, but, at equilibrium, there is a relationship
between total external exposure, total uptake, and levels in whole 
blood and in target organs (the central and peripheral nervous 
system, the haemopoetic system, and the kidneys).  The duration of 
exposure in adults does not affect levels in the blood or in the 
target organs.  Only the levels in bone (the main site for 
deposition) and the aorta increase with duration of exposure and 

    One method of estimating the lead level in target organs 
indirectly is the "provocation test":  administration of calcium 
disodium edetate (Ca-EDTA) or penicillamine enhances urinary 
excretion of lead and, in this way, an estimate can be obtained of 
biologically available tissue lead that is not deposited in bone.  
Chisolm et al. (1976) established a linear relationship in children 
between Pb-B and the logarithmic values of mobile chelatable lead 
levels in 24 h urine after administration of lead at 25 mg/kg body 
weight.  Data from both preschool children and adolescents gave a 
common regression line.  This important measure of assessment of 
internal dose can only be carried out in hospitals and, in any 
case, careful consideration must be given to ethical aspects.  Lead in urine (Pb-U)

    Pb-U levels are also indicative of total exposure, but a normal 
rate of lead excretion does not serve as a reliable means of 
excluding excessive exposure.  Moreover, there are risks of 
contamination from clothes, collection vessels, etc.  To minimize 
the influence of diuresis on the Pb-U level, 24 h samples are 
preferred.  However, this again limits application to 
institutionalized subjects.  Measurement of Pb-U levels, therefore, 
does not provide a practical method for assessing exposure.  Lead in faeces (Pb-F)

    Pb-F levels, on the other hand, offer a good estimate of total 
oral exposure in adults.  The unknown, and probably higher rate of 
absorption of lead in young children makes this method less 
suitable for exposure assessment in this most important group.  Lead in deciduous teeth (Pb-T)

    Pb-T and lead levels in hair (Pb-H) are being used increasingly 
for estimating integrated long-term exposure.  They have the 
advantage that samples are easy to procure.  Pb-T levels even 
indicate exposures of previous years, offering a means of 
estimating the history of exposure in children.  Deciduous teeth
have been used in studies of the neuropsychological effects of lead 
on children (Needleman et al., 1979).  Two developments may make it 
possible to measure Pb-T levels  in situ:  the chemical measurement 
of lead in enamel biopsies (Brudevold et al., 1977) and X-ray 
fluorescence analysis (Shapiro et al., 1978). 

    Surface contamination of hair has to be removed by careful 
washing; it may be very difficult to ensure that measured Pb-H 
levels are really due to increased body burden.  Routine 
application of these methods in exposure assessment has still to 
await further research. 

    Thus, in the present state of the art,  biological assessment
 of exposure has first of all to be based on measurement of lead
levels in blood.  The most reliable method of sampling is by 
venepuncture; this demands highly trained personnel, experienced in 
taking blood from young children.  In some studies, reliance has 

been placed on analysis of blood taken by finger-prick.  These Pb-B 
levels tend to be higher than those measured in venous blood, 
partly through contamination from the skin.  Only when extreme care 
is taken, can similar Pb-B levels be obtained, although even then 
capillary samples give higher levels than venous ones (Elwood et 
al., 1977). 

    In the last few years, a simple method has been developed to 
identify individuals with probable overexposure to lead.  This is 
the measurement of zincprotoporphyrin (ZPP) in blood obtained from 
a finger- or ear-prick by means of an automated haematofluorimeter.  
If, in the case of long-term lead exposure, ZPP is not increased in 
comparison with an unexposed control group of the same age and sex, 
there is no need to examine for Pb-B.  This quick screening method 
may save a lot of time and expense. 

3.8.  Assessment of the Subjective Environment

    Two types of environment are distinguished:  the "objective" 
and the "subjective" (perceived) environment (section 3.1).  Where 
possible, objective (mostly instrumental) methods should be applied 
in the assessment of exposures, but there are some exposures that 
may defy objective assessment.  Assessment of perceived exposure 
differs fundamentally from biological assessment of exposure 
(section 3.7), because, in the latter, parameters of internal 
exposure are examined, while in the former information is 
systematically collected on the subjective response (e.g., to odour 
or taste).  Because subjective response usually shows wide 
interindividual differences in intensity and quality of perception, 
exposure assessment has to be based on the response of carefully-
selected groups of subjects, under controlled test conditions. 

    Although in the case of exposure to noise (section 
annoyance reaction may constitute the most important response, 
objective methods are widely available for exposure assessment.  
The same is true for exposure to oxidants (section, that 
may have irritation effects. 

3.8.1.  Assessment of odour

    This short survey is based mainly on reviews by Sullivan 
(l969), Turk et al. (1974), and the National Academy of Sciences 
(l979a).  Odorants are chemical compounds; even with sensitive 
analytical methods (chromatography, mass spectrometry, etc.) 
complete determination and identification of these substances are 
often not possible.  Physical and chemical determinants of odour 
are not yet fully understood.  Odours are sensations that have to 
be measured as a perceptive human response. 

    The subjective properties of odour include  intensity, 
 detectability, quality (character), and  hedonic tone (pleasant 
or unpleasant).  Intensity (magnitude of sensation) can be 
described in ordinal categorization:  faint, moderate, strong, or 
possibly with a numerical assignment of magnitude. 

    Some properties of olfaction can introduce errors into 
subjective assessment;  adaptation:  a rapid decrease of odour 
intensity during continuous exposure;  recovery:  restoration of 
olfaction when exposure is removed;  habituation:  getting used to 
the odours, which operates over much larger periods than adaptation 
and recovery.  Quiet breathing allows only about 3% of odorants to 
enter the nose and to contact the olfactory epithelium, whereas 
sniffing brings more odorants into contact with these perceptors.  
Human sensory responses to individual compounds vary widely and, in 
some cases, it may be possible to detect 1 mol of odorant per 109 
mol of air.  All these characteristics of olfaction require a very 
rigid design for the assessment of exposure. 

    Up to now, a correlation has not been established between a 
predicted or measured ambient odour intensity and community odour 
annoyance mainly because of:  (a) difficulty in obtaining an 
unbiased measurement of community odour annoyance; (b) difficulty 
in defining an ambient odour intensity level through diffusion 
modelling of source odour intensity measurements; and (c) 
difficulty in measuring ambient odour intensity, because of 
variability in meteorological conditions (Franz, 1980). 

3.8.2.  Assessment of taste

    In general, principles similar to those for odour assessment 
apply (Zoeteman, 1978), though it is not the volatile compounds 
that are concerned, but the substances dissolved in saliva, which 
enter the pores of the taste buds, located on the tongue.  The 
sense of taste is much less sensitive than that of smell.  There 
are four classic tastes: sour, salty, bitter, and sweet.  Inter-
individual sensitivity may vary up to a thousandfold.  Sensitivity 
for bitter tastes tends to decrease with age and with smoking.  
Many sensations, commonly attributed to taste, are in fact a 
combination of taste and odour. 

3.8.3.  Example of sensory assessment of drinking-water

    Zoeteman (1978) conducted a study of sensory assessment of the 
chemical composition of drinking-water in the Netherlands.  The 
purpose was to investigate the suitability of sensory assessment of 
water quality as an indicator for the presence of chemical contaminants.  
At first, an inquiry was held among a sample of the Dutch population 
(n = 3073, 18 years of age and over).  In 3.2% and 6.9% of the 
subjects the water quality was rated as "offensive" or worse for 
odour and for taste, respectively.  Water taste proved to be the 
main factor in assessment of the sensory quality. 

    In order to identify the compounds causing bad taste and odour, 
20 types of drinking-water (8 of ground water, 5 of drinking-water 
from dune filtration, 7 from reservoirs) were collected; a panel of 
52 subjects assessed the quality.  Because the taste proved to be 
more noticeable than odour, sensory assessment was restricted to 
taste assessment.  The average taste scales clearly differed 
between the various types of water.  The measured levels of sodium, 

calcium, and magnesium salts could not explain the large differences 
in taste between various waters.  Therefore, organic contaminants 
seemed likely to be the cause of the observed differences. 

    In the 20 types of water, 280 organic substances were detected 
(gas chromatographic-mass spectrometer-computer system), but nearly 
100 could not be chemically identified.  Nearly twice as many 
organic compounds were found in drinking-water derived from surface 
water, compared with that from ground water.  In water derived from 
surface water, several taste-impairing compounds could be 

    This example merely illustrates the sensory approach in 
assessing exposure to organic compounds that have unpleasant 

3.9.  Interindividual and Intergroup Variability in Exposure:
Population at Risk

    Individuals vary greatly in exposure and susceptibility to 
environmental pollutants.  Therefore they should not be treated 
like homogeneous groups of experimental animals.  Close attention 
should be paid to the frequency distribution of exposures, since 
this will affect procedures in the statistical analysis (Chapter 
6).  A common form of distribution is the log-normal.  Blood-lead 
values, for example, generally follow this pattern, and the median 
may then be more appropriate than the arithmetic mean as a central 
value for a group.  When several groups are being compared in 
respect of exposures to environmental pollutants and the 
corresponding effects on health, it is important to be able to look 
at interindividual as well as intergroup variations in exposures. 

    As already stated, preschool children are liable to be at 
greater risk of exposure to lead than adults living in the same 
environment.  Furthermore, a mother may act as an external source 
of exposure for her child during pregnancy, because of the 
transplacental passage of lead and methylmercury, or, during 
lactation, more particularly for fat-soluble substances such as 
organochlorine pesticides. 

    Subjects with specific food habits, for instance those 
consuming merely macrobiotic foods such as seaweed, or those 
consuming marine shellfish, tend to have a high intake of arsenic; 
fish eaters may be exposed to a higher level of methylmercury.  The 
presence of moulds in foodstuffs, producing the highly toxic, 
carcinogenic aflatoxin, may also lead to defined groups at risk. 

    Various groups at risk can also be distinguished with regard to 
physical factors, for example, with regard to exposure to ultra-
violet radiation, those with light skin (in comparison with those 
with dark skin), and subjects who spend much time outdoors 
(fishermen, farmers, etc.). 

3.10.  Outdoor/Indoor Exposure

    In investigating exposure-response relationships in the general 
population exposed to air pollutants, the studies have usually been 
designed to relate health effects with concentrations in the out-
side air.  However, human beings usually spend about 80% of their 
time indoors; those particularly at risk (young children, the 
elderly, and the chronic sick) spend even more time indoors.  
Concentrations of pollutants in the home, at the workplace, or in 
public buildings etc., can be quite different from those outdoors. 
In recent years, increased attention has been given to pollution 
indoors, either in relation to the penetration of pollution from 
outside, or to that from sources within the home itself (as from 
smoking, heating, or cooking).  Reviews have been prepared by 
Benson et al. (1972), Henderson et al. (1973), Halpern (1978), WHO 
(1979b), and the National Academy of Sciences (1981). 

    Biersteker (1966) measured the ratio between the concentrations 
of sulfur dioxide (SO2) and smoke indoors and outdoors in 60 houses 
in Rotterdam, for periods of at least one week.  In the average 
home, the ratio for SO2 was 0.20 and that for smoke, 0.80.  In a 
few homes, indoor pollution greatly exceeded outdoor pollution, 
apparently because of faulty stoves and chimneys.  Biersteker also 
established a relation between windspeed (< 1 m/s versus > 6 m/s) 
and mortality, which he believed might be due to accidental high 
indoor carbon monoxide (CO) concentrations on days of low wind-
speed.  Extreme examples of such problems are seen through the use 
of coal, wood, or "non-standard" fuels such as dried cow dung for 
heating or cooking in poorly constructed dwellings without proper 
chimneys.  Measurements of smoke and carbon monoxide in such homes 
in Nigeria were reported by Sofoluwe (1968), who related cases of 
bronchopneumonia among infants to exposure to pollution while on 
their mothers' backs or laps during cooking.  Fuel burning indoors 
has also been demonstrated to lead to chronic bronchitis in Papua, 
New Guinea, and Nepal (Anderson, 1979; Pandey et al., 1981). 
Clearly, in such circumstances, exposure to pollution indoors is 
liable to be vastly greater than that outdoors.  Also, exposures to 
transient peaks, while close to the fire, are likely to be more 
important than those averaged over long periods, but it is 
extremely difficult to obtain any proper assessment of them. 

    Another source of indoor pollution is para-occupational 
exposure:  workers take pollutants attached to their skin, hair, 
clothes, and shoes into their homes.  The increased incidence of 
mesothelioma in female members of the family who have cleaned 
asbestos-polluted workclothes is well known. Watson et al. (1978) 
examined 1-6-year-old children of workers at a storage battery 
plant, and compared them with as many controls.  The levels of lead 
in the blood and free erythrocyte porphyrin were higher in the 
exposed group, and the workers' homes had much higher lead levels 
in the domestic dust. 

    Jacobson et al. (1978) observed a peculiar source of indoor 
pollution with radionuclides.  They used thermoluminiscent 
dosimeters (TLD) placed in wristbands and worn by members of 

families, in each of which one family member had been treated with 
iodine-131; in addition they monitored radioactivity in the air at 
home.  Adults and children received much higher direct exposure to 
radiation through their skin than their thyroid; external doses 
ranged from 6 to 2220 mrems (60 µSV to 22.2 mSv) and thyroid dose 
equivalents from 4 to 1330 mrems (40 µSV to 13.3 mSv).  In these 
families, childhood exposure could double the risk of developing 
thyroid malignancies. 

    Exposures to radiation of natural origin is generally greater 
indoors than outdoors, primarily owing to the emission of the gas 
radon from the soil below and from building materials such as 
stone, brick, or concrete.  This radionuclide decays to solid 
materials (generally referred to as radon daughter products) that 
become attached to other fine particulate matter in the air, and 
concentrations indoors are largely a function of ventilation. 

    Another pollutant that may be liable to be at higher 
concentration indoors than outdoors is formaldehyde, which is 
emitted from the urea-formaldehyde resin used in chipboard 
furniture, in fabrics, and in the insulating material sometimes 
applied to the cavity walls of houses. 

    Particulates may be at similar levels indoors and outdoors, 
though indoor particulates can be quite different in composition, 
with contributions from smoking, house dust, aero-allergens, human, 
and animal dandruff shedding, consumer products, and furnishings 
(National Academy of Sciences, 1981). 

    The examples above indicate clearly that for a number of air 
pollutants, overall human exposures are determined more by 
conditions inside individual homes than by those outside.  There 
are major differences in this respect related to lifestyle, social 
conditions, and the structure of buildings, and it is important to 
consider the local situation very carefully before embarking on 
studies requiring detailed assessments of air pollution exposures.  
This topic has been discussed further in a recent WHO document 
(l982) and much information on indoor pollution can be found in the 
report of an international symposium held in the USA (Spengler et 
al., 1982). 

3.11.  Time-weighted Exposure

    While, for some purposes, it may be sufficient to assess the 
exposures of defined groups by using average values for the 
locality, where more detailed information is required, personal 
sampling might be used (section 3.6) or biological monitoring may 
be applicable in some cases (section 3.7).  An intermediate 
approach, however, is to calculate time-weighted averages by noting 
the amount of time spent by individuals in different types of 
environment and then relating this to concentrations measured in 
those environments, using a combination of fixed site or personal 
samplers as required. 

    Fugas (1976, 1977) calculated the time-weighted exposure of a 
group of urban dwellers by using time spent and average concentration 
in air at various places.  She estimated the weighted weekly exposure 
(WWE) of urban dwellers living and working in a combination of 
situations as shown in Table 3.3. 

Table 3.3.  Time-weighted exposure (TWE) in urban dwellers
                   t(h per    SO2         Pb          Mn
Type of exposure   week)    Ca   Ct    C     Ct    C     Ct
Home               110      89   9790  2.5   275   0.04  4.4
Occupation,        42       8    336   0.3   12.6  0.02  0.84
Street F           10       600  6000  6.0   60    0.80  8.0
Street B           4        180  720   3.5   14    0.12  0.48
Countryside        2        25   50    0.1   0.2   0.01  0.02
Total              168           16896       361.2       13.74
WWEb                             101         2.2         0.08
a  C = concentration in µg/m3.
b  WWE = weighted weekly exposure in µg/m3.

    Five urban dwellers spent an average of 14 h/week outdoors and 
2 h/week out of town.  The individual weighted weekly exposures (in 
µg/m3) depending on the individual exposures in the home and in
the street (no. 5 being a traffic policeman), are given in Table 
Table 3.4.  Individual weighted 
weekly exposures (µg/m3) of five
urban dwellers spent an average of 
14 h/wk outdoors and 2 h/wk out 
of town
subject  sulfur   lead  manganese
1        33       1.0   0.05
2        101      2.2   0.08
3        108      1.5   0.10
4        55       1.0   0.06
5a       177      2.6   0.25
a Subject number 5 was a traffic 

    This example shows the large interindividual variations in 
exposure that may occur.  Moreover, the time spent in the polluted 
outdoor environment (streets) was only an average of 14 h in a week 
(168 h).  These data refer solely to concentrations in air, mostly 
measured by means of personal sampling, without taking account of 
variations in respiratory volume, in absorption, or in exposure 

through food (for lead and manganese).  Therefore, although the 
time-weighted average concentration through inhalation is better 
estimated in this example than in most exposure assessment, it only 
improves the measurement of respiratory exposure in the general 
sense; the data are still far from giving a measurement of the 
actual dose, as such.  Biological monitoring of lead and manganese 
would have provided indices of total exposure in a far easier way, 
including those through food and water. 

    No fixed rules can be laid down for computing time-weighted 
exposures, since the relative importance of the different 
contributions varies with the pollutant under consideration, and it 
may change also from place to place or from time to time. 


AITIO, A., RIIHIMAKI, V. & VAINIO, H., ed.  (1983)   Biological
 monitoring and surveillance of workers exposed to chemicals.
Washington DC, Hemisphere Publishing.

(l978)   Air sampling instruments,  5th ed., Washington DC,

ANDERSON, H.R.  (1979)  Respiratory abnormalities, smoking
habits and ventilatory capacity in a highland community in
Papua New Guinea:  prevalence and effect on mortality.   Int.
 J. Epidemiol.,  8 (2): 127-135.

ATHERLEY, G.R.C.  (1978)   Occupational health and safety
 concepts; chemical and processing hazards.  London, Applied
Science Publishers.

AZAR, F., SNEL, R.D., & HABIBI, K.  (1973)  Relationship of
community levels of air lead and indices of lead absorption.
In:   Proceedings of the International Symposium; Environmental
 Health Aspects of Lead, Amsterdam, Oct. 1972,  Luxembourg,
CEC-Directorate-General for Dissemination of Knowledge (EUR
5004 d-e-f), pp. 581-593.

BENSON, F.B., HENDERSON, J.J., & CALDWELL, D.E. (l972)  Indoor/
 outdoor air pollution relationships.  A literature review.
Washington DC, A.P.-112 (Superintendent Docum., Aug. 1972).

BERLIN, A., WOLFF, A.H., & HASEGAWA, Y., ed.  (1979)  The use
of biological specimens for the assessment of human exposure
to environmental pollutants.  In:  Proceedings of the
 International Workshop CEC-WHO-EPA, 1977,  The Hague, Nijhoff,
368 pp.

BIERSTEKER, K.  (1966)  [Polluted air.]  Thesis, University of
Amsterdam (in Dutch).

BRIDGES, J.E. & PREACHE, M.  (1981)  Biological influences of
power frequency electric fields.  A tutorial review from a
physical and experimental viewpoint.   Proc. Inst. Elect.
 Electron. Eng. Inc.,  69: 1092-1119.

BROCH, J.T.  (1971)   Acoustic noise measurements.  2nd ed.,
Naerum, Denmark, Bruel & Kjaer.

(1977)  Lead in enamel biopsies for measuring past exposure to
lead.   J. dent. Res.,  56: 1165-1171.

BURNS, W.  (1973)   Noise and man.  2nd ed., London, John

Interrelationships among blood lead concentrations,
quantitative daily ALA-U and urinary lead output following
calcium EDTA.  In: Nordberg, G.F., ed.   Effects and
 dose-response relationships of toxic metals,  Amsterdam,
Elsevier Publishing, pp.416-433.

COENEN. W.  (1976)  [Description of the time-related changes
of harmful substance concentrations by means of a continuous
Markow process.]   Staub-Reinhalt. Luft,  36: 240-248 (in

COENEN, W.  (1977)  [Description of the time-related changes
of harmful substance concentrations by means of a continuous
Markow process I.],  Staub-Reinhalt. Luft,  37: 271-273 (in

COERMAN, R.  (1970)  Mechanical vibrations.   Occupational
 Safety & Health Series,  Geneva, International Labour Office,
21, p. 17.

CRAWFORD, M.D., GARDNER, M.J., & MORRIS, J.N.  (1971)  Changes
in water hardness and local death-rates.   Lancet,  2: 327-329.

CZERSKI, P., OSTROWSKI, K., & SILVERMAN, C., ed.  (1974)
 Proceedings of an International Symposium on Biologic Effects
 and Health Hazards of Microwave Radiation, Warsaw, Poland,
 15-18 October 1973.  Warsaw, Polish Medical Publishers.

(1978)  Survey of trace elements and pesticide residues in the
New Zealand diet.   N.Z. J. Sci.,  21: 57-69.

DUPUIS, H.  (1969)  [The physiological strain on human beings
resulting from mechanical vibrations.]   Düsseldorf Ver. Dtsch.
 Ing.,  Nr.7, (in German).

D., & COHN, S.H.  (1981)  Critical concentrations (or amount)
of cadmium in human renal cortex.  Dose-effect studies in
cadmium smelter workers.   J. Toxicol. environ. Health,  7:

ELWOOD, U.J., CLAYTON, B.E., & COX, R.A.  (1977)  Lead in
human blood and in the environment near a battery factory.
 Br. J. prevent. soc. Med.,  31: 154-163.

FERRIS, B.G.  (1978)  Health effects of exposure to low levels
of regulated air pollutants.  A critical review.   J. Air
 Pollut. Control Assoc.,  28: 482-495.

FAO  (1962)   Manual on household food consumption surveys.
Rome, Food and Agricultural Organization of the United Nations
(FAO Nutritional Studies, Series No.18).

FAO  (1971)   Food balance sheets, l944-l966 average.  Rome,
Food and Agricultural Organization of the United Nations.

FRANZ, J.J.  (1980)  Comments on the NAS-Report.   J. Air
 Pollut. Control Assoc.,  30: 173-174.

FUGAS, M.  (1976)  Assessment of total exposure to an air
pollutant.  In:  Proceedings of the International Conference on
 Environmental Sensing and Assessment, Las Vegas, 1975. Vol. 2,
pp. 1-3,  New York, Inst. Elect. Electron Eng. Inc.

FUGAS, M.  (1977)   Biological significance of some metals as
 air pollutants; Part I:  Lead.  Research Triangle Park, NC,  US
 Environmental Protection Agency  (No. EPA-600/1-77-041).

GUICHERIT, R., ed.  (1979)   Photochemical smog formation in
 the Netherlands.  Delft, The Netherlands, 1978-1979,  Delft,
Institute of Environmental Hygiene and Sanitary

GUIGNARD, J.C. & KING, P.K.  (1972)   Aeromedical aspects of
 vibration and noise,  Brussels, North Atlantic Treaty
Organization  (AGARD graph No. 151).

HALPERN, M.  (1978)  Indoor/outdoor air pollution exposure
continuity relationships.   J. Air Pollut. Control Assoc.,  28:

HARING, B.J.A.  (1978)  Human exposure to metals released from
water distribution systems, with particular reference to water
consumption patterns.   Trib. Cebedeau,  31: 349-355.

B.C.J.  (1979)  [An investigation of the drinking water
consumption in the Netherlands.]  H2O, 12: 213-216 (in Dutch).

HARLEY, J. (1979)  Biological specimen collection for analysis
of radioactivity.  In: Berlin, A., Wolff, A.H., & Hasegawa,
Y., ed.   Proceedings of the International Workshop by
 CEC-WHO-EPA, April l977,  The Hague, Nijhoff, pp. 201-213.

(1979)  Duplicate diet studies on fishing communities in the
U.K. Mercury exposure in a critical group.   Environ. Res.,  18:

HEALTH & SAFETY EXECUTIVE  (1979)  Asbestos, Volumes 1 & 2.
In:   Final Reports of the Advisory Committee, Health & Safety
 Commission.  London, Her Majesty's Stationery Office, pp. 100
and 103.

 Indoor/outdoor air pollution relationships.  An annotated
 bibliography,  Springfield, VA National Technical Information
Service (EPA-112 B).

Intestinal microecology group.  Dietary fibre, transit-time,
faecal bacteria, steroids and colon cancer in two Scandinavian
populations.   Lancet,  8031: 207-211.

ISO  (l972)   Nuclear energy glossary,  Geneva, International
Organization for Standardization (Ref. No. ISO 92l).

JACOBSON, A.P., PLATO, P.A., & TOEROEK, D.  (1978)  Contamina-
tion of the home environment by patients treated with
iodine131: Initial results.   Am. J. Public Health,  68: 225-230.

Prevalence of penumoconiosis and chronic bronchitis in foundry
workers.   Scand. J. Work environ. Health,  22 (Suppl. 1): 64-72.

KONO, S., SONE, T., & NIMURA, T.  (1982)  Personal reaction to
daily noise exposure.   Noise Control Eng.,  19: 4-16.

KATZ, M.  (1980)  Advances in the analysis of air
contaminants.   J. Air Pollut. Control Assoc.,  30: 528-557.

LACEY, R.F. & POWELL, P.  (1976)   Water quality statistics for
 the MRC-WRC study,  Marlow, Bucks, Water Research Centre.

LARSEN, R.J.  (1970)  Relating air pollutant effects to
concentration and control.   J. Air Pollut. Control Assoc.,  20:

LEUPKE, N.P.  (1979)   Monitoring environmental materials and
 specimen banking,  The Hague, Nijhoff.

LIPSCOMB, D.M., ed.  (1978)   Noise and audiology, Baltimore,
University Park Press, pp.488.

MARR, J.W.  (1971)  Individual dietary surveys; purpose and
methods.   World Rev. Nutrit. Diet.,  13: 105-164.

MEYER, P.B.  (1975)  The use of personal sampling methods in
the assessment of the toxicological risk at the working
place.  In:   Transactions of the FEICRO Conference on
 Technology and Working Conditions, Rotterdam, Sept. 1975,
Delft, TNO Research Institute for Environmental Hygiene.

MOORE, M.R., HUGHES, M.A., & GOLDBERG, D.J.  (1979)  Lead
absorption in man from dietary sources.  The effect of cooking
upon lead concentrations of certain foods and beverages.   Int.
 Arch. occup. environ. Health,   44: 81-90.

NATIONAL ACADEMY OF SCIENCES  (1979)   Odors from stationary
 and mobile sources.  Washington, DC., National Research

NATIONAL ACADEMY OF SCIENCES  (1981)   Indoor pollutants,
Washington, DC, National Research Council.

NATO/CCMS  (1979)   Air pollution assessment methodology and
 modelling,  Brussels, NATO Committee on the challenges of
modern society.

H., MAHER, C., &  BARRETT, P.  (1979)   Deficits in
psychologic and classroom performance of children with
elevated dentine lead levels.   New England J. Med.,  300:

NELSON, N.  (1976)  The role of interaction of environmental
agents in modifying their biological activity.  In: Mehlman,
M.S., Shapiro, R.E., & Blumenthal, H., ed.   New concepts in
 safety evaluation,  London, Wiley, pp.3-10.

NORDBERG, G.F., ed.  (1976)   Effects and dose-response
 relationships of toxic metals,  Amsterdam, Elsevier Publishing.

NORDBERG, G.F., ed.  (1978)  Factors influencing metabolism
and toxicity of metals:  a consensus report.   Environ. Health
 Perspect.,  25: 3-41.

OECD  (1973)   Food consumption statistics, 1955-1971,  Paris,
Organisation for Economic Co-operation and Development.

PACKHAM, R.F.  (1978)   Water Uses, Sources and Chemical
 Composition,  Symposium, January, 1978,  Marlow, Bucks, Water
Research Centre.

NEUPANE, R.P.  (1981)  Domestic smoke pollution and chronic
bronchitis and corpulmonale - an epidemiological study in a
rural community of Nepal.  In:  Proceedings of the 9th
 Scientific Meeting of the International Epidemiological
 Association, Edinburgh, August 1981.

PERSONS, K.S. & BENNETT, R.L.  (1974)   Handbook of noise
 ratings.  Washington DC, National Aeronautics and Space
Administration (NASA Constructor Report CR-2376).

PETERSON, A.P.G. & GROSS, E.E.  (1974)   Handbook of noise
 measurements.  7th ed., Concord, Massachusetts, Genrad.

PROKOPENKO, Ju, I., ILYIN, V.P., & ZURKOV, V.S.  (1981)  [The
effect of long-wave UV radition on the mutagenic and
embryotoxic effects of exposure to chemical substances.]   Gig.
 i Sanit., 7: 13-15 (in Russian).

ROACH, S.A.  (1977)  A most rational basis for air sampling
programs.   Ann. occup. Hyg.,  20: 65-84.

SCOTTO, J., FEARS, T.R., & GORI, G.B.  (1976)   Measurements of
 ultraviolet radiation in the United States and comparisons
 with skin cancer data,  Washington, US National Cancer
Institute (DHEW No. (NIH) 76-1029).

X-ray fluorescence analysis of lead in teeth on urban children
in situ:  correlation between the tooth lead level and the
concentration of blood lead and free erythroporphyrins.
 Environ. Res.,  17: 46-52.

SHEPHARD, J.G.  (1975)   The application of the critical group
 concept.  Ministry of Agriculture, Fish, and Food.  Lowestoft,
UK,  Technical Note RL/4/75.

ed.  (1982)  Indoor air pollution.  Amherst, 1981.   Environ.
 Int.,  8: 1-534.

STERN, A.C., ed.  (1976)   Air pollution, Vol. III.  Measuring,
 monitoring and surveillance of air pollution,  New York,
Academic Press.

SOFOLUWE, G.D.  (1968)  Smoke pollution in dwellings of
infants with broncho-pneumonia.   Arch. environ. Health,  16:

SULLIVAN, R.J.  (1969)   Preliminary air pollution review of
 odorous compounds, a literature review.  Raleigh, US DHEW,
Public Health Service, NAPCA.

TEPPER, L.B. & LEVIN, L.S.  (l972)   A survey of air and
 population lead levels in selected American communities.
(Final report to the US Environmental Protection Agency.)

TURK, A., JOHNSTON, J.W., & MOULTON, D.G.  (1974)   Human
 responses to environmental odors,  New York, Academic Press.

UNSCEAR.  (1972)   Ionizing radiation:  levels and effects.
New York, United Nations  (Report of the UN Scientific
Committee on the Effects of Atomic Radiation).

 Symposium on the Development and Usage of Personal Monitors
 for Exposure and Health Effect Studies.  Research Triangle
Park, NC, US EPA (EPA-600/9-79-032).

USHIO, F. & DOGUCHI, M.  (1977)  Dietary intakes of some
chlorinated hydrocarbons and heavy metals on experimentally
prepared diets.   Bull. Environ. Contam. & Toxicol.,  17:

VAHTER, M., ed.  (1982)   Assessment of human exposure to
 cadmium and lead through biological monitoring.  Stockholm,
Karolinska Institute, pp. 136  (Prepared for United National
Environment Programme and World Health Organization by
National Swedish Institute of Environmental Medicine and
Department of Environmental Hygiene, Karolinska Institute,

HERBER, R.F.M., & CASTILHO, P.  (1978)  [An explorative study
of the relations between water and blood metal levels in two
communities.]   Tijdschr. Soc. Geneeskd.,  56: 110-114 (in

WALKER, A.R.P.  (1978)  The relationship between bowel cancer
and fiber content in the diet.   Am. J. clin. Nutr., 31
(Suppl.): S 248 - S 251.

Increased lead absorption in children of workers in a lead
storage battery plant.   J. occup. Med.,  20: 759-761.

WASSERMAN, D. & TAYLOR, W.  (1977)   Proceedings of the
 International Occupational Hand-arm Vibration Conference,
 l977, Cincinnati, US DHEW, Public Health Service, NIOSH.

WILLIAMS, M.K., KING, E., & WALFORD, J.  (1969)  An
investigation of lead absorption in an electric accumulator
factory with the use of personal sampler.   Br. J. ind. Med.,
26: 202-216.

WILKINS, J.R., REITHES, N.A., & KRUSE, C.W.  (1979)  Organic
chemical contaminants in drinking water and cancer.   Am. J.
 Epidemiol.,  110: 421-448.

WHO  (1975)   WHO Environmental Health Monitoring Programme.
Report of a WHO Meeting. Geneva, World Health Organization
unpublished document (EHE/75.1).

WHO  (l976)   Selected methods for measuring air pollutants.
Geneva, World Health Organization  (WHO Offset Publication No.

WHO  (1977a)   Environmental Health Criteria, 3:  Lead.
Geneva, World Health Organization.

WHO  (1977b)   Air monitoring programme design for urban and
 industrial areas.  Geneva, World Health Organization (WHO
Offset Publication No. 33).

WHO (l979a)   Environmental Health Criteria, l4: Ultraviolet
 radiation.  Geneva, World Health Organization.

WHO  (l979b)   Environmental Health Criteria, 8: Sulfur oxides
 and suspended particulate matter.  Geneva, World Health

WHO  (l980a)   Analysing and interpreting air monitoring data.
Geneva, World Health Organization (WHO Offset Publication No.

WHO  (l980b)   Environmental Health Criteria, l2: Noise.
Geneva, World Health Organization.

WHO  (l98l)   Environmental Health Criteria, 16: Radiofrequency
 and microwaves.   Geneva, World Health Organization.

WHO  (l982a)   Estimating human exposure to air pollutants.
Geneva, World Health Organization (WHO Offset Publication No.

WHO  (l982b)   Environmental Health Criteria, 22: Ultrasound.
Geneva, World Health Organization.

WHO  (l983)   Guidelines for drinking water quality.  Vol. I.
Geneva, World Health Organization.

[On the mechanism of adaptogenic effect of UV radiation.]
 Vestn. AMN SSSR.,  3: 23-26 (in Russian).

ZIELHUIS, R.L.  (1975)  Dose-response relationships for
inorganic lead.  I. Biochemical and haematological responses.
II. Subjective and functional responses.  Chronic sequela.  No
response levels.   Int. Arch. occup. Health,  35: 1-18, 19-35.

ZIELHUIS, R.L.  (1978)  Biological monitoring.   Scand. J. Work
 environ. Health,  4: 1-l8.

ZIELHUIS, R.L. & HARING, B.J.A.  (l98l)   Water quality and
 mortality in the Netherlands.  Published in:   Water Supply and
 Health,  Amsterdam,  Elsevier, p. 397.

ZOETEMAN, B.C.J.  (1978)   Sensory assessment and chemical
 composition of drinking water.  Thesis, University of Utrecht.


4.1.  Introduction

    Historically, much of the information on the effects of 
environmental agents on health has come from rather crude counts of 
deaths or of clinically-recognized cases of disease, but, with the 
passage of time, assessment of symptoms, of specific pathological 
conditions, or of biochemical, physiological or neurological 
dysfunction has progressed, providing a wealth of tools for the 
epidemiologist.  Caution is required, however, in proceeding along 
such lines for, if the only effect of an agent at a given intensity 
is a small change in function, well within the normal physiological 
range of variation in an individual, then its importance in 
comparison with other factors affecting health must be weighed 
carefully.  Judgement is required taking into account the duration 
of the effects, the number of persons likely to be affected, and 
the relative importance of immediate but relatively minor effects 
and of long-delayed but more serious ones.  In a sense, the 
population's health could be an integral index of environmental 
quality, and the effects of various environmental agents are 
multifactorial (Sidorenko, 1978).  Some effects are specific to an 
agent, but most are non-specific (Bustueva & Sluchanko, 1979). 

    In this chapter, the measurement of effects in terms of the 
relatively crude, but widely available mortality statistics and 
routinely-collected morbidity statistics are discussed in some 
detail, followed by a section devoted to one important disease 
group (cancer).  Techniques available for the more objective 
measurement of effects of environmental agents have been arranged 
on the basis of anatomical systems, starting with the respiratory 
and cardiovascular systems followed by the nervous system and a 
section on behavioural effects; though the latter are not strictly 
effects on an anatomical system, they are closely associated.  
While it is probably true that more measurement techniques have 
been developed for the first two systems than for others, the order 
in which the systems are discussed does not otherwise imply any 
order of their individual importance. 

    In the study of adverse health effects, the effects under study 
must be clearly defined in advance and the terminology employed 
must not be confusing. 

4.1.1.  General comments on effects

    A biological effect may be a subjective or objective phenomenon 
experienced or measured in the short term or over a long term.  
Such phenomena may be ranked in some order or measured on a scale, 
if they are graded effects, or simply be registered as "present" or 
"absent", as for example with death or cancer morbidity (quantal).  
Each measure of health effects depends on the definition of what is 
a biological effect.  It is risky to generalize beyond any given 

    Effects may be described as either "stochastic" or "non-
stochastic".  A stochastic effect is one for which the probability 
of  occurrence, rather than the severity, depends on the absorbed
dose and there may be no threshold.  Carcinogenesis asssociated 
with ionizing radiation is placed in this category, although it is 
possible that a threshold exists for certain other cancer-producing 
processes.  The non-stochastic effect is one where the  severity 
varies with the exposure level and for which there may be a 
threshold.  Damage to the lens of the eye from electromagnetic 
radiation constitutes such an effect. 

    Clinically, an effect may be an  acute effect,  for example,
chemical pneumonitis following shortly after exposure to a 
substantial amount of an irritant, or a  chronic effect  such as
progressive interstitial pulmonary fibrosis following repeated 
exposure to fine particulate allergenic agents, or after intense 
exposure to certain fibrogenic dusts, or protracted deposits of 
such dusts in the lungs of workers. 

    However, clinically acute effects may also result following 
long-term exposures, for example, epileptic convulsions after long-
term exposure to dieldrin, myocardial infarction in workers 
chronically exposed to carbon disulphide, or convulsions and acute 
abdominal colic following long-term exposure to lead.  On the other 
hand, following short-term exposures to asphyxiants such as carbon 
monoxide, or histotoxic agents such as nitrogen dioxide, chronic 
after-effects may be observed.  A special category relates to 
sensitizing agents, where repeated exposure to levels that are non-
irritant will be uneventful in a clinical sense, until a degree of 
sensitization occurs after which the next dose initiates an acute 
clinical response:  during the non-symptomatic (prepathological) 
phase, the only objective sign may be elicited by serological 
studies, though physiological function tests may reveal deficits 
that are not yet appreciated by the exposed person. 

    An agent or combination of agents may induce an effect that can 
be readily accepted as being harmful in the short term.  Where a 
physiological change occurs with no overt clinical benefit or 
detriment, then there can be grounds for considerable disagreement 
as to the significance to be attached to it.  For example, it might 
have to be considered whether minor departures from the normal 
function might affect the ability to respond to other stresses and 
diminish life expectation. 

    In any population, there will be a range of responses to an 
agent with, at the two extremes of distribution, resistant and 
susceptible persons; arbitrary cut-off points may define these 
sectors.  There is usually a number of subsets in a population in 
which differing exposure/effect relationships are seen. 

    Because of differences in susceptibility of populations and 
selection factors, exposure/effect relationships derived from one 
group should only be used with reservation for other groups. 

    The terms "susceptible", "vulnerable", "hyperreactive", 
"hypersensitive", and "high risk" are often used indiscriminately.  
The following are some of the working definitions for these terms: 

     Susceptibility (vulnerability):  The state of being readily
affected or acted upon.  In hypersusceptible persons, "normal, 
expected" effects occur, but with a lower exposure than in the 
majority of the population.  Vulnerability can be used inter-
changeably with susceptibility. 

     Hyperreactivity:  In hyperreactive persons, the effects of
the agent are qualitatively those expected, but quantitatively 

     Hypersensitivity:  To react with "allergic" effects following
reexposure to a certain substance (allergen) after having been 
exposed to the same substance. 

     High risk:  The term "risk" can be defined as the expected
frequency of undesirable effects arising from a given exposure to a 
pollutant.  Thus, the populations at risk are those who have been 
exposed specifically to a defined pollutant that may produce a 
particular adverse effect. 

    Susceptibility may be based on physiological factors.  For 
example, the very young and the very old are relatively vulnerable 
to exposure to temperature extremes.  Another variety of 
susceptibility is associated with a genetically determined 
reduction or a virtual absence of important enzymes involved in the 
detoxication of compounds, or the repair or reversal of their 
effects.  Thus glucose-6-phosphate dehydrogenase deficiency renders 
persons susceptible to the development of clinical methaemoglobinaemia 
following exposure to a range of compounds occurring occupationally 
and therapeutically.  Exposure to low intensities of environmental 
agents may not produce disease but may reduce host resistance 
(Litvinov & Prokopenko, 1981). 

4.1.2.  General comments on measurements of effects

    The effect should be defined and measured in as standardized a 
manner as possible, whether it is measured using a physical test 
(e.g., skin testing and radiography), or is physiologically or 
biochemically measured, or some other index of morbidity or 
mortality is used. 

    A number of aspects of measurement need to be considered before 
embarking on a study or attempting to evaluate published data.  The 
results from measurement techniques, such as questionnaires and 
function tests, are used with stated criteria to define the health 
status of people in the study.  The consistency or comparability of 
these tools and of the criteria, from area to area or from study to 
study, determine whether results can be compared, either as 
specific rates or as trends.  This, in turn, determines the 
replicability of studies and the broader generalizations from 
studies. Instruments from one study must be compared with those of 

previous studies to maintain standardization.  Errors and biases, 
which may occur in all techniques, can be minimized with proper 
usage, otherwise they may seriously affect the results. 

    The advantages or disadvantages of techniques or instruments 
have to be judged on the basis of (i) acceptability by the study 
population; (ii) the accuracy and reliability of the results 
obtained using them; and (iii) their ease of use and the 
availability of technicians or other persons who can use them.  
Instruments for field studies must be simple and robust, with the  
necessary supplies and, if possible, electric sources available in 
the field. 

    The  sensitivity  of the test must be determined, i.e., the
proportion of all persons with a particular characteristic
such as a disease or variation in function that is detected by
the test.  Similarly, the  specificity  needs to be determined,
meaning the proportion of persons lacking a particular 
characteristic who are correctly identified by the test.  The 
purposes of the study will determine the acceptability of the 
orders of insensitivity (rate of false positives) and non-
specificity (rate of false negatives).  The determination of what 
constitutes a true or false positive or negative will depend on the 
standard employed, which itself may have been previously validated.  
The two measures are related to one another; usually an increase in 
sensitivity will mean a decrease in specificity.  The degree of 
sensitivity and specificity required depends on the objectives of 
the study.  Each objective will have different needs in terms of 
the permissible rate of false positives and false negatives. 

    Instruments for measurement, whether electronic or mechanical 
or in the nature of questionnaires or radiographs, have a number of 
characteristics that need to be appreciated, if comparisons are to 
be made sequentially within the same study or with other studies.  
Some of the characteristics, such as  accuracy, precision, 
 repeatability, and  reproducibility, have been explained in
section 3.5.1.  Other major characteristics are discussed below. 

     Reliability is the measure of consistency or reproducibility
and is dependent in particular on accuracy.  The  validity of a
measurement made by an instrument is determined by the extent to
which it relates to the effect that it is intended to measure.  The 
reliability of an instrument can be determined by frequent tests in 
which everything is the same except the time (test-retest).  Some 
instruments are or can be set up to obtain measurements of 
duplicate samples and results at the same time (split-half 
testing); this method is often used in questionnaires where 
"identical" questions are used in different parts of the 
questionnaire (sections 4.4.1 and 5.6.3).  Validity often depends 
on the criteria of what is a true characteristic of disease.  It 
also depends on what is used as the standard.  The validity of a 
test procedure can often be checked by applying it, along with 
other assessments, to a well-defined population. 

    Some conditions or physiological functions display distinct 
cyclical (e.g., diurnal or seasonal) variations, and these must be 
taken into account when making measurements related to them.  
Effects themselves might also follow a cyclical pattern, as in some 
occupational examples in which adverse effects can be more severe 
at the beginning of the day's shift or the working week, so that 
again the time at which measurements are made becomes critical.  Inter- and intrainstrument variation

    All instruments have a certain degree of variation in addition 
to limitations of accuracy.  The instrument may have variations 
over time or from place to place due to different circumstances, 
such as environmental factors or electrical current, etc.  
Questionnaires are also perceived differently in different social 
settings.  Each instrument has to be evaluated before use and in 
any given setting.  Changes in barometric pressure, temperature, or 
humidity may also affect the functioning of physiological measuring 
instruments.  Any moveable part of an instrument, or any component 
of an instrument may develop defects or change its characteristics 
through a variety of causes, which might influence the readings 
obtained at different times.  The use of standardized protocols and 
standardized instruments helps to minimize these problems.  
Calibrations and frequent checks are required to examine 
intramachine differences:  if necessary, adjustments can be made or 
correction factors can be determined and applied.  The behaviour of 
the instrument, such as its consistency or linearity has to be 
determined (e.g., section  Interinstrument differences 
need to be studied carefully, bearing in mind how far such 
differences might introduce bias in the particular study in 
question.  It may sometimes be necessary to set up a special 
randomized experiment in which a selected set of subjects 
(conveniently research workers) are tested with all the instruments 
to be used in a survey and the instruments should be interchanged 
with one another during the course of the study in a randomized 
manner.  Inter- and intralaboratory differences

    Quality assurance procedures should be established within a 
laboratory and between it and a reference laboratory.  Such 
procedures include not only analytical quality assurance, but also, 
if the study involves biological material, quality assurance during 
the sampling and storage of the material.  Quality assurance checks 
should be carried out before the start of the study as well as 
during it. 

    The use of any instrument will, in part, depend on whether 
there are  reference values  with which results can be compared.
"Normal" will always have to be defined in each study as to whether 
that means either an ideal, average, or standard value, or any one 
of a dozen other possible definitions.  Inter- and intraobserver variations

    In any investigation in which there is more than one observer, 
technician, or inteviewer, there may be differences between them.  
These differences may be due to a large number of factors, 
including the way in which tests are applied, information recorded, 
or findings interpreted.  Interobserver variation must be tested 
regularly in a systematic fashion (section  An observer 
varies in performing and interpreting tests, from time to time and 
place to place.  Sometimes, this variation can only be measured by 
having the observer read a standard test or perform a standard test 
at different times, or by evaluating random aliquots of an 
observer's work to see if there are differences from time to time 
and place to place. 

    In interviews, though a question may have been asked and an 
answer given, the resulting information cannot be immediately 
accepted as accurate.  Even when independent checks have been 
carried out and estimates of error rates produced, caution must be 
exercised.  There are likely to be errors from:  random sampling; 
bias from failure to respond or incomplete response; misunder-
standing; memory; inapproriate attempts to quantitate vague or 
imprecise notions; deliberate distortion; and recording, coding, 
analysis, or retrieval of the results.  The sources of error in 
interview surveys have been reviewed by Moser & Kalton (l97l), 
Bennett & Ritchie (l975), Alderson (l977), and Abramson (l979). 

4.2.  Mortality and Morbidity Statistics

    All sources of data have their advantages and disadvantages, 
and their cost/benefit ratios.  Existing sources of data are often 
the easiest to use, the least costly, and require the least 
recourse to subjects.  On the other hand, they often lack the 
information needed for a thorough examination of the objectives.  
Usually, studies, in which existing sources of data are used, are 
descriptive in nature, but are retrospective.  Such information has 
been used in many studies of geographical differences in the 
distribution of disease. 

4.2.1.  Mortality statistics

    Some mortality data adequately reflect certain medical 
problems.  Some conditions have a high mortality rate and death 
occurs quickly; thus mortality data for malignant melanoma can 
provide fairly accurate quantification of this disease, but such 
data would be quite inadequate for quantifying squamous carcinoma 
of the skin.  Diseases that have a long natural history may be 
susceptible to treatment or have a variable mortality:  the longer 
the natural history, the less indication mortality statistics can 
provide about the causative factors in the disease.  Though many 
conditions create considerable discomfort for patients, this rarely 
figures in the mortality data.  An extreme example of this is the 
common cold which is responsible for virtually no deaths.  A more 
severe disease is rheumatoid arthritis; again the mortality is so 
low that routine mortality data are of limited value in its study.  

Many routine data collection systems are relatively inflexible and 
there is little facility for introducing new items into the system 
and processing these alongside the original data.  Flexibility can 
only be provided by the special study of additional material 
collected independently of the routine health statistics system. 

    Because of differences in death certification procedures, in 
diagnoses, and in coding causes of death, international comparisons 
are subject to possible biases.  Furthermore, death certificates 
usually indicate only the place of residence and the place of 
occurrence of the death, thus precluding any estimate of lifelong 
exposure that is so critical in the examination of mortality due to 
chronic diseases.  Death certificates do not contain information on 
tobacco smoking, occupational exposures, ethnic groups, and other 
host and environmental factors.  Without multiple causes of death 
coding and autopsies, the cause of death information that is coded 
from the death certificate may often be misleading (Moriyama et 
al., 1958).  Co-morbidity data are rarely available, unless all 
causes mentioned on the certificate have been specially coded 
(immediate and contributory, as well as underlying).  However, the 
mortality data for a country can still show very important trends. 

    An additional dimension is added to mortality statistics when 
the data are tabulated according to occupation.  In many countries, 
census questions have included the occupations of individuals 
listed in each household and this can provide a denominator for the 
calculation of occupational mortality rates.  However, there are 
some reservations about this approach:  the onset of an 
occupationally-induced disease might be associated with a decline 
in ability to work and this could result in an individual changing 
his job.  Since mortality rates are calculated from the terminal 
occupation, the true etiology of the disease may not then be 
revealed.  Occupational mortality statistics, however, provide a 
very useful background for the study of occupational disease, 
despite doubts about the validity of the data (Alderson, 1972; 
Mason et al., l975; Fox, l977). 

4.2.2.  Routine morbidity statistics

    Two indices are used to describe morbidity: incidence and 
prevalence.   Incidence is the number of newly diagnosed cases of a 
 disease occurring in a defined population in a defined period of 
 time.  The incidence rate is often expressed as the number of
newly diagnosed cases occurring in 100 000 population in one year.  
The second measure,  prevalence, is the number of people with a 
 given disease alive at a point in time (point prevalence) or
over a period of time, say one year  (period prevalence). 
Again, the prevalence can be expressed as a fraction of the number 
of persons in the population at that time:   prevalence rate. 

    In general, morbidity is harder to ascertain than mortality, 
but is generally a more sensitive indicator of the health effects 
of pollutants.  Aspects of the collection of morbidity statistics 
have been discussed in various statistical reports (WHO, 1965). 

    Morbidity data are available in some countries where the 
national or local health authorities regularly conduct a health 
interview survey or a health examination survey. 

(a)  Health interview survey

    This method involves interviewing a sample of individuals and 
asking them questions about their social setting, their recognition 
of signs and symptoms of disease, their attitude to sickness and 
health, and their contact with health services in the past.  It may 
be applied to a representative sample of the total population, a 
sample drawn from carefully selected locations throughout the 
country, or subpopulations chosen by locality, age, occupation, or 
other characteristics.  The virtue of a health interview survey is 
that data from a fairly large sample of respondents may be obtained 
with limited expenditure of resources (compared with the use of 
medical and other staff to investigate the subjects).  In certain 
circumstances, this indication of the knowledge, attitudes, and 
practices of members of a population may provide a more appropriate 
picture of their health care than that derived from other 
approaches.  In particular, the respondents' answers indicate how 
much ill health they perceive, their reactions to this, and the 
reported incapacity from the ailment.  Perceived ill health may 
also be studied in relation to such variables as age, sex, 
socioeconomic status, occupational class, smoking and drinking 
habits, and ethnic origin.  Data may be obtained from subjects by a 
variety of methods, such as the use of self-completion 
questionnaires, direct interviews, group interviews, or diaries;  
as has been mentioned in section 4.1.2, it is important to consider 
the reliability and validity of data obtained from such approaches. 

(b)  Health examination survey

    In this type of survey, data are collected by examination and 
investigation of the respondents in a sample.  Some data will have 
to be collected by questioning the subject, but the aim is to cover 
quite different issues by direct examination, or to complement any 
responses with observations and investigations.  It is essential 
that the original test results, whether in the form of electro-
cardiographic (ECG) tracings, blood pressure observations, 
ventilatory function indices, flow volume curves, or skinfold 
thickness measurements, be preserved for study in addition to 
diagnoses or judgements based on these results.  Certain problems 
that may be encountered in this type of survey include the 
magnitude of resources required to carry out such a survey and the 
fact that gossip about the study may have a marked effect upon the 
response rate.  By identifying variations from the physiological 
and psychological norms, the investigator may be able to quantify 
the "morbidity" in a population, which is not recognized by the 
subjects themselves or by their families. Even with measurements 
such as blood pressure, there is a grey area between normal levels 
and those at which treatment is definitely justified. 

(c)  General practice morbidity data

    In the absence of a national or local health interview survey 
or health examination survey, it may be appropriate to use 
morbidity data from general or family practice as a source of 
relevant material.  An extensive amount of literature is available 
on the organization of medical records in general practice, the 
coding of such material, and the use of such data to identify 
morbidity in the population; this has been reviewed by Alderson & 
Dowie (1979).  In addition to considering problems involved in the 
measurement of morbidity, the appropriateness of the available 
denominator should be considered on the basis of practice size, 
which is an essential component in the calculation of morbidity 
rates by age and sex.  Papers discussing the validity of data 
recorded by general practitioners include those by Dawes (l972), 
Hannay (l972), Morrell (1972), and Munro & Ratoff (1973). 

(d)  Hospital morbidity data

    Such data are valuable in reference to malignant diseases 
(section  For many other chronic diseases, hospital 
discharge statistics are less useful; for example, where patients 
suffer from an acute stroke, there may be strong pressure from the 
patient or the family to nurse the patient at home.  Hospital 
morbidity data may thus not provide an accurate indication of the 
load of disease in the community; if an appreciable proportion of 
patients are cared for outside the hospital, the statistics will be 
unreliable as measures of disease prevalence.  Morbidity statistics 
from hospital and general practice are biased, because they reflect 
the use made of health care rather than the prevalence of 
morbidity.  Titmuss (1968) commented that the higher income groups 
know how to make better use of the health service; they tend to 
receive more specialist attention, occupy more beds in better-
equipped and better-staffed hospitals, receive more effective 
surgery and better maternal care, and are more likely to get 
psychiatric help and psychotherapy than low-income groups.  Forsyth 
& Logan (1960) indicated a close relationship between the use of 
hospitals and the availability of hospital facilities - the more 
acute beds in a district, the higher the admission rate and the 
longer the length of stay.  Vaananen (1970) suggested that 
emergency admissions are a reflection of population structure, 
whilst planned hospital admissions vary in relation to the 
facilities available.  Changes in hospital morbidity data may 
indicate changes in facilities, rather than any underlying 
alteration in the distribution of the disease in the population. 

(e)  Record linkage studies

    The development of a record linkage study has been described by 
Acheson (1967) and by Baldwin (1972).  Record linkage requires the 
construction of a cumulative file of events occurring in the lives 
of individual patients.  The longitudinal study described by the 
Office of Population Censuses and Surveys of England and Wales 
(Office of Population Censuses and Surveys, 1973) links birth 

registration, domestic migration, overseas emigration, census of 
population, notification of cancer, and death registration, for a 
l% sample of the total population. 

    For linking an individual with his health and other relevant 
records, a unique identifying marker is required.  Some countries 
attempt to use unique identification numbers for all or some of 
their populations, when operating social security or health service 
systems.  The use of these numbers can be of considerable help, 
even when they are limited to certain categories of persons, for 
example, those in active employment. 

     Absenteeism.   Records of leave from work or school due to
specific morbidity may be used to determine the effects of 
pollutants on such morbidity.  Generally, there is difficulty in 
ascertaining the real causes of the absenteeism.  This is more 
directly related to the day of the week, the season, to epidemics 
of some communicable diseases, such as influenza, to some social 
event or to behavioural factors. 

(f)  Occupational morbidity studies

    Data on number and duration of episodes of incapacity, in 
relation to age, sex and cause of incapacity, have been published 
regularly in the United Kingdom.  These statistics can be used as 
an indication of morbidity in the working population, though 
incapacity may be influenced by:  occupation; the worker's domestic 
environment and access to medical care; selection 'into' and 'out 
of' a particular occupation; the financial and social consequences 
of declared illness; the completeness of notification; the 
unemployment situation; industrial morale; and other subcultural 
factors (Alderson, 1967).  Specific statistics are produced on 
industrial injuries and workers who develop prescribed industrial 
diseases, but the general problems of routine statistics apply to 
these data. 

4.3.  Cancer

4.3.1.  Cancer and enviromental factors

    Within a very broad definition, it may be true that some 80-90% 
of cancers can be related to environmental influences, as has been 
frequently stated (section  In fact, however, only a 
relatively small proportion of cancers have as yet been related to 
specific agents (Doll & Peto, 1981).  The most important factor 
identified so far is tobacco smoking, related not only to lung 
cancer but also to cancer of other sites, such as the larynx, 
buccal cavity, pancreas, and bladder.  Excess consumption of 
alcohol, when combined with the use of tobacco, increases the risk 
of oesophageal and oropharyngeal cancers - frequently in a 
multiplicative fashion.  Occupational exposures to agents such as 
asbestos, tars, radioactive materials, chromates, and products 
associated with the refining of nickel also contribute to the 
incidence of lung cancer, although their influence is small 
compared with that of smoking. 

    Evidence from migrant and other studies shows that dietary 
factors are likely to be of major importance for cancers of the 
digestive tract and reproductive organs.  Sunlight is an important 
cause of malignant melanoma and other skin cancer in the less-
pigmented races.  Small numbers of cancers are attributable to 
ionizing radiation and chemotherapeutic agents. 

4.3.2.  Measurements of cancer  Incidence and mortality rate

    The burden of cancer is measured by the determination of 
incidence and mortality rates.  When such data are not available, 
the relative frequency of the various organs affected as a 
proportion of all diagnosed cancers may provide useful information.  
Prevalence rate is rarely used.  The various sites of cancer are 
divided into broad groups in the International Classification of 
Diseases (ICD) (WHO, l977) and in the International Classification 
of Diseases for Oncology (ICD-0) (WHO, l976). 

    Incidence data of good quality are available for relatively few 
areas in the world (Waterhouse et al., 1982), and few cancer 
registries have been in existence for longer than 20 years.  The 
World Health Organization maintains a data bank of mortality data 
of cancer which are periodically analysed (Segi & Kurihara, 1972; 
Segi, 1978). 

    Incidence data on cancer are preferable to the more widely 
available mortality figures because mortality is influenced by the 
rates of cure, which vary from centre to centre.  Nonetheless, 
relative frequency data can, if possible sources of bias are 
assessed, indicate the likely cancer distribution in a region.  The 
high levels of nasopharynx cancer, seen in the population of 
southern Chinese origin in Singapore, were known from relative 
frequency studies long before cancer registration became 
established.  Variations of incidence with age

    Any theory of carcinogenesis must be consistent with observed 
age patterns.  Cook and collaborators (1969) examined the shape of 
the age-incidence curve for many different tumours using data from 
eleven different cancer registries; they concluded that epithelial 
tumours probably arise from a similar process, part of which would 
be continuous exposure to an environmental agent, and that, even if 
there is a variation in susceptibility among the population, there 
is no indication of a reduction in the relative size of the pool of 
susceptibles with age. 

    Not all cancers behave in the same way and a series of such 
age-incidence curves are shown in Fig. 4.1.  These different curves 
must be explicable in terms of causal factors.  The ICD often 
aggregates neoplasms with quite different age-distributions, e.g., 
osteosarcoma and chondrosarcoma, hence histology-specific and site-
specific incidence curves may be more informative. 

FIGURE 4.1  Geographical differences

    Comparison of data from different parts of the world is subject 
to bias.  The age-adjusted cancer incidence figures contained in 
the Cancer Incidence in Five Continents monographs (UICC, 1970; 
Waterhouse et al., 1982) show that the reported incidence of 
cancer, taken as a whole, varies between countries by a factor of 
around three; when separate anatomical sites are considered, the 
difference may be as much as 100-fold.  While such extremes are 
unusual, for many common sites such as the breast, stomach, and 
cervix uteri, risk ratios of 10-30 are observed (Muir, 1975). 

    Such differences in the geographical ditribution of cancer were 
often ascribed to ethnic or genetic factors.  Kennaway (1944) 
studied primary liver cancer in Bantu in South Africa and Negroes 
in the USA, and concluded that "the very high incidence of primary 
cancer of the liver found among African negroes does not appear in 
US negroes and is therefore not a purely racial character.  Hence 
the prevalence of this form of cancer in Africa may be due to some 
extrinsic factor which should be studied". 

    Higginson (1960) concluded that these differences were due to 
environmental factors.  Extending this concept, he examined the 
differences in cancer incidence by site reported in the first 
volume of Cancer Incidence in Five Continents.  Assuming that the 
smallest recorded rate represented a level that should be 
considered as due to genetic factors, he postulated that the 
difference between this rate and the highest observed rate probably 
represented those cancers due to exogenous factors.  Doll (1967), 
Boyland (1967), and Higginson & Muir (1976) have presented further 
supportive evidence that most human cancers are due to 
environmental causes. 

    These studies do not imply that genetic factors may not have a 
role to play.  Skin cancer is clearly influenced by a genetically 
determined skin pigmentation, being particularly frequent in 
northern Europeans with lightly pigmented skins who have emigrated 
to Western Australia.  However, objective measurements for 
genetically determined susceptibility do not exist for most 
cancers.  Cancer and lifestyle

    Lifestyle is difficult to measure in an objective manner. 
Ethnic effects may be involved indirectly through postulated 
dietary influence on hormone levels, promoters, and inhibitors.  
Some factors, such as age at first pregnancy or age at first 
coitus, which are often socially determined, are clearly linked 
with breast and cervix uteri cancer risk. 

    Some religious groups have cancer rates that differ 
substantially from those of the general population.  Wynder and 
co-workers (1959) first reported a lower cancer mortality in a 
population of Seventh Day Adventists who neither smoke tobacco nor 
drink alcohol and who follow an ovo-lacto-vegetarian diet.  
Phillips (1975) has confirmed these findings showing that cancer 
death rates in lifetime Adventists, in California, USA, are 50-70% 
of those of the general population.  Cancer in migrants

    It has been reported that the very high incidence of gastric 
cancer seen in Japan slowly decreases in Japanese migrants moving 
to the USA, but that the incidence of large intestine cancer 
increases much more rapidly reaching a level close to that found 
among those born in the USA (Haenszel & Kurihara, 1968).  Buell 
(1973) has also shown that breast cancer morbidity rates in US-
born Japanese approximate to those of the non-Japanese population 
of the USA, which could be interpreted as indicating that the 
environment - perhaps the diet in the USA - may have manifested its 
effect in successive generations to influence hormonal levels and 
cancer risk.  Time trends

    The incidence rates of some cancers are rising, but those of 
others are falling over a period of years.  The increasing 
incidence of cancer of several sites, e.g., the pancreas, has been 
interpreted as being due to the introduction of new environmental 
agents that are largely unknown at the moment.  It has been 
relatively easy to link the rise in lung cancer to the increase in 
the number of persons smoking cigarettes.  Such a link has been 
confirmed, for example, by the fact that the decline in the 
proportion of physicians in the United Kingdom who smoke has been 
followed by a fall in the lung cancer rates in the professional 
group.  Correlation studies

    Correlation techniques based on descriptive data have brought 
out statistically significant associations between, for example, 
cigarette smoking and lung cancer, spirit consumption and 
oesophageal cancer, beer intake and large bowel cancer (Breslow & 
Enstrom, 1974).  However, by the laws of probability, a certain 
number of the correlations that emerge will be due to chance alone 
and others may be 'indirect', i.e., linked in some ways to the true 
cause.  A correlation between coronary heart disease and lung 
cancer would merely reflect the fact that both are strongly linked 
to cigarette smoking.  As, in such studies, the pooled experience 
of several populations is contrasted, the effect of intense 
exposure within a small segment of a population, say an industry, 
would be diluted out (Muir et al., 1976). 

    However, sizeable international differences in cancer risk are 
more likely to be due to the existence of a widespread exposure in 
one population compared with another, than to the presence of a 
small group at a very high risk.  The environmental data available 
for use in the correlations are usually of poorer quality than the 
cancer incidence figures, and because of the long latent period for 
cancer, the correlations should be made using the environmental 
data for 10, 20, 30, or 40 years ago and present day cancer 
figures.  Such historical information on the environment is rarely 

    Results of correlation studies should never be accepted without 
further testing.  Failure to demonstrate a correlation probably 
indicates that no association exists, though these may fail to 
emerge by chance alone.  Hospital data

    Only a small proportion of patients, thought by the family 
doctor to have a malignant disease, are not referred, either on 
medical grounds or because of refusal of the patient to attend 
hospital.  For example, Alderson (1966) found that only 1.1% of 540 
patients, dying from malignant disease in a defined population, had 
not been referred to hospital.  Identification of the presence of 
malignant disease is still a problem, but the progress of the 

disease is such that deterioration in a patient will usually lead 
to hospital referral.  Although Heasman & Lipworth (1966) have 
demonstrated appreciable discrepancies between clinical and 
postmortem diagnoses of patients dying in hospital, hospital 
morbidity data may in general provide a good indication of the 
distribution of cancer in the population.  Cancer and occupation

    The Office of Population Census and Surveys of England and 
Wales has published data on cancer risk by occupation since the 
turn of the century (Office of Population Census and Surveys, 

    The frequency of a given occupation - as assessed at the 
national census - is compared with the frequency of that occupation 
for a given disease on death certificates and a standardized 
mortality ratio (SMR) that takes age into account is computed.  
Woodworkers, for example, have an excess risk of lung cancer (SMR 
113) and of bladder cancer (SMR 145) and teachers a relatively low 
risk of lung cancer (SMR 32).  The interpretation of such findings 
is complex (Gaffey, 1976).  Fox & Adelstein (1978) estimated that 
perhaps 12% of the excess cancer risk is due to carcinogenic 
exposures at work, the remainder to the lifestyle that is 
associated with an employment.  They base their argument on the 
fact that standardization for social class and cigarette smoking 
removes, or substantially reduces, the difference in risk and on 
the finding that the spouses of those in certain high risk 
occupations, for example miners, also have very high risks for 
stomach cancer.  Thus, while analyses of the type carried out by 
the Office of Population Census and Surveys can suggest where 
workplace exposure to carcinogens may exist, the presence of such 
carcinogens must be established by other methods.  Case reports

    From time to time, reports are published of a patient or a 
small group of patients, who have an unusual cancer and an out-of-
the-way ocupation or exposure:  adenocarcinoma of the nasal passages 
in furniture manufacturers and bootmakers (Acheson et al., 1968, 
1970); adenocarcinoma of the vagina in the daughters of women given 
diethylstilboestrol for miscarriage (Herbst et al., 1971); 
mesothelioma of the pleura in shipyard workers using asbestos 
(Stumphius, 1971).  This type of evidence, correlation-based, 
usually uncovered by alert clinicians, is unlikely to uncover risks 
due to common exposures or to those causing common cancers.  Never-  
theless, it has resulted in the discovery of human carcinogens.  Epidemiological uses of pathological findings

    The lack of accuracy of histopathological diagnosis in the 
population studied is frequently such as to understate the true 
burden of disease.  For most tumours, there is a spectrum of 
microscopic appearances ranging from marked anaplasia through 
various degrees of development and organization, sometimes 

differentiating into several patterns.  These variations may appear 
in different persons or in one block of tumour from a single case.  
Attempts have been made, employing concensus diagnosis, to conduct 
expert panel readings of sections from patients with mesothelial 
tumours meeting agreed criteria (Greenberg & Lloyd-Davies, 1974).  
More sophisticated methods have been employed for the diagnosis of 
angiosarcoma, using mixed sections, read blind, and recycled for 
the study of intraobserver variation (Baxter et al., 1980); this 
has lead to reduced interobserver variation, too.  Improving 
diagnostic accuracy for a special tumour with its attendant 
publicity may increase vigilance among non-panel pathologists, so 
that an increase in reported cases may include an element of 
increased recognition that has to be taken into account in 
attempting to determine an increase in true incidence over time.  
Further discussions on the contributions of pathology to 
epidemiological knowledge are found in the review by Muir (1982). 

4.4.  Respiratory and Cardiovascular Effects

    There has been much confusion in the past over the definition 
of "bronchitis" or conditions known variously as chronic non-
specific respiratory disease or chronic obstructive lung disease.  
Standardized questionnaires, together with lung function tests, 
have played a vital role in establishing a common definition and in 
allowing studies of prevalence to be undertaken among occupational 
or general population groups in a comparable way throughout the 
world.  In this way, it has been shown that the dominant factor in 
the development of the disease is tobacco smoking.  Beyond this, 
there are associations between the development of symptoms in adult 
life and the earlier occurrence of acute respiratory illnesses.  
The effects of environmental agents have been demonstrated in terms 
of exposure to urban air pollution (notably by the sulfur 
dioxide/particulates complex) and to a wide range of dusts and 
fumes in industry. 

    Equally, investigations of the etiology of cardiovascular 
diseases have been greatly aided by the use of questionnaires 
together with electrocardiograms (ECGs) and other objective 
assessments.  In this case, the role of specific environmental 
agents is not very clear, but many studies have indicated the 
adverse effects of smoking, obesity, and dietary factors, and the 
possibly protective effects of exercise on the development of 
cardiovascular disease. 

    In this section, an account is provided of the ways in which 
the occurrence of respiratory and cardiovascular disease can be 
investigated in order to explore associations with environmental 
agents, examining indices that have been developed, methods of 
measurement, and interpretation of results. 

4.4.1.  Symptom questionnaires

    One method of determining the environmental agents important in 
the development of respiratory and cardiovascular disease has been 
the use of standardized questionnaires.  This approach, by obtaining 

details of respiratory and cardiovascular symptoms, makes it 
possible to compare symptom prevalence in groups of individuals 
exposed and unexposed to different environmental agents. 

    The assessment of clinical symptoms is an important technique 
in epidemiological surveys, because it can increase the yield of 
positive cases of respiratory and cardiovascular disease and act as 
an index of disease, measurement errors being different from those 
of other methods, such as ECG and lung function tests. 

    In the construction of symptom questionnaires, it is essential 
to formulate precise questions to reduce variations that may result 
when different observers ask people about their respiratory or 
cardiovascular symptoms.  It is necessary to use identical, or at 
least very similar, symptom questions to compare data from 
different studies.  An important advance was the publication, by 
the British Medical Research Council's Committee on the Aetiology 
of Chronic Bronchitis, of recommended questionnaires for recording 
respiratory symptoms, together with instructions for their use 
(Medical Research Council's Committee on the Aetiology of Chronic 
Bronchitis, 1960; Medical Research Council, 1966, 1976).  A similar 
advance occurred with the development of the London School of 
Hygiene Standardised Cardiovascular Questionnaire (Rose, 1962, 
1965).  These respiratory and cardiovascular symptom questionnaires 
have been translated into different languages and used for surveys 
in different countries (Higgins, 1974; Rose et al., 1982).  A 
recent development was the construction and testing of a standardized 
questionnaire for use in respiratory epidemiology by the American 
Thoracic Society and the Division of Lung Diseases of the United 
States National Heart and Lung Institute (Ferris, 1978). 

    Self-completed versions of the symptom questionnaires have been 
developed to avoid the problem of observer variation, to be used 
when personal contact is not practicable, and because they are more 
economical than personal interviewing.  Epidemiologists have used 
this method with various degrees of success to collect information 
on respiratory and cardiovascular symptom prevalence.  Fletcher & 
Tinker (1961) noted that answers on cough, phlegm, dyspnoea, and 
smoking habits on a self-completed questionnaire did not always 
correspond with those on an interviewer-administered questionnaire.  
Furthermore, the self-completed questionnaire was not returned or 
was incomplete, in about 25% of the cases under study.  This error 
rate was only 7% in a group of post office clerks and the 
investigators concluded that the self-completed questionnaire might 
be particularly useful in persons with a clerical background.  
Sharp and coworkers (1965) obtained satisfactory agreement between 
the self-completed and the interviewer-administered questionnaires 
on respiratory symptoms in an industrial population in Chicago.  
Higgins & Keller (1970) used both self-completed and interviewer-
administered respiratory questionnaires successfully in Tecumseh, 
Michigan but self-completed questionnaires were not satisfactory in 
a survey by Higgins and coworkers (1968) in mining communities in 
Marion County, West Virginia. 

    Lebowitz & Burrows (1976) compared the interviewer-administered 
British Medical Research Council and the US National Heart and Lung 
Institute respiratory symptom questionnaires with each other and 
with a self-administered questionnaire, of their own design, in 
Tucson, Arizona.  There was a basic 10% agreement between responses 
of any two questionnaires for all questions that asked about 
symptoms, but less disagreement for more factual questions, such as 
those concerning smoking.  For questions with similar wording, the 
British and USA questionnaires yielded very similar results in 
terms of prevalence of responses, relationship to answers on an 
independent questionnaire, and interrelationships of positive 
responses.  The Tucson self-completed questionnaire was a 
satisfactory instrument in the population surveyed, detecting more 
abnormalities and better delineating cough and phlegm "syndromes" 
than the interviewer-administered versions. 

    Zeiner-Henriksen (1976) sent the London School of Hygiene 
cardiovascular questionnaire, by post, to random national samples 
in Norway with response rates of around 80%.  There were relatively 
few missing or incorrect answers and the estimates of mortality 
prediction were broadly similar to those in Rose's (1971) follow-up 
study of the interviewer-administered version.  The evaluation of 
these cardiovascular questionnaires by Rose and coworkers (1977) 
found the yield of positives for "angina" and "history of possible 
infarction" was about twice as high with interviewers than self-
administration, but the positive groups obtained by the two 
techniques differed little in their association with electrocardio-
graphic findings or their ability to predict five-year coronary 
mortality risk.  This suggests self-completion does not produce any 
major loss of specificity or dilution with less severe cases. 

    There is other research that suggests that self-completed 
questionnaires can be used successfully to examine the effects of 
different environments on the cardiovascular and respiratory 
symptoms in migrants and people born in the USA (Krueger et al., 
1970; Reid et al., (1966), in twins in Sweden (Cederlöf et al.,
1966a,b) and in a sample of 37 to 67-year-old people in the United 
Kingdom (Dear et al., 1978). 

    Validity and reproducibility are important criteria in the 
assessment of the symptom questionnaires.  Reproducibility can be 
affected by the changing disease status of individuals or 
measurement variability.  Interobserver measurement variability in 
a study can result in the indiscriminate pooling of heterogeneous 
results.  It can also produce systematic differences between 
studies, so that measurement differences could be mistaken for 
differences between populations.  Consequently, the method of 
administering the questionnaire should be standardized and the 
interviewers comparably trained to reduce these systematic 

    Checks can be incorporated in the survey to examine inter-
observer reliability.  Each observer may be allocated to a randomly 
chosen group of subjects with each observer's results analysed 
separately for means and standard deviations or their prevalence 

estimates.  Where practicable, subjects may be examined more than 
once, each time by a different observer.  Interviewing techniques 
can be examined by the playing back of tape recordings. 

    The reproducibility of the answers to questions about symptoms 
for the respiratory and cardiovascular questionnaires has been 
studied (Fairbairn et al., 1959; Fletcher et al., 1959; Holland et 
al., 1966; Rose, 1968; Zeiner-Henriksen, 1972; Lebowitz & Burrows, 
1976).  Despite the considerable reproducibility of symptoms on 
re-examination, the use of estimates of prevalence based on single 
interviews only can cause problems in the interpretation of 
results.  There is often a substantial proportion of subjects 
initially reporting particular symptoms, who do not report these 
characteristics on subsequent checks.  Therefore, regular 
questioning would make it possible to grade the subjects on the 
basis of the number of times the subject has been classed as 

    To ensure the validity of the symptom questionnaires, the 
respiratory and cardiovascular diseases being assessed must be 
defined accurately and the symptoms described should be 
manifestations of these disease entities.  A problem in the 
development of stricter diagnostic criteria is that improvements 
in specificity (i.e., the yield of a few false positives) may 
reduce the sensitivity (the yield of a few false negatives).  To 
compare the amounts of disease in different populations, it is 
essential that the levels of sensitivity and specificity do not 
vary from one population to another. 

    The problem of determining the exact number of false positives 
and negatives for symptom questionnaires is complicated by the lack 
of a perfect reference test.  Therefore, validation must be based 
on correlations with different indices of disease, e.g., ECG, FEV, 
mortality, each of which is an indirect measure.  Holland and co-
workers (1966) have shown answers to questions about respiratory 
symptoms to discriminate among persons, categorized on the basis of 
more objective measures such as FEV and 1-h sputum volume.  Rose 
(1971) has examined the relationship between cardiovascular 
symptoms, electrocardiographic findings, and coronary heart disease.  
ECG findings predicted a higher proportion of cases of coronary 
heart disease than symptoms.  However, because of a measure of 
independence between the ECG and symptom findings, a combination was 
more effective than either alone. 

    Generally, the symptom questionnaires have resulted in an 
increased standardization and comparability of results from 
different surveys.  From an epidemiological perspective they are a 
useful technique, because they amplify information from other tests 
and may allow the identification of high-risk individuals.  
However, in using symptoms questionnaires to compare groups from 
different cultures or countries, it is a precaution to ensure the 
findings are consistent with other measures of disease such as FEV, 
ECG, and mortality. 

4.4.2.  Tests of system function

    There are measurement techniques to assess the effect of 
environmental agents on the cardiovascular and respiratory systems.  
The major longitudinal studies on the incidence and prevalence of 
coronary heart disease such as the Framingham Heart Disease 
Epidemiological Study (Kannel, 1976) and the Tecumseh Health Study 
(Epstein et al., 1965) have used ECG, blood pressure measurement 
and various serum cholesterol determinations.  Rose and his 
collaborators (l982) describe these epidemiological methods and 
other techniques such as chest radiography and the measurement of 
heart size. 

    Holland and coworkers (1979) have reviewed the functional tests 
used in the epidemiological study of respiratory disease.  These 
include the tests that assess airway function during an expiratory 
manoeuvre, those that measure airway resistance using a body 
plethysmograph, and the closing volume and the frequency dependence 
of compliance tests of small airway function.  Higgins (1974) 
reviewed other tests that can be used to determine the impact of 
environmental factors on respiratory function, including cough as 
an objective measure, the collection, measurement, and 
categorization of sputum, the measurement of morphological changes, 
and chest radiography.  Lebowitz (1981) reviewed other techniques 
used to study both acute and chronic effects on health of air 

    There are important criteria to be considered in the selection 
and interpretation of a particular test to determine the effect of 
the environment on cardiovascular and respiratory function.  These 
include the following: 

(a) The test should be appropriate to the problem under study.
    Some tests are able to determine functional abnormality
    while others are more able to assess the specific site of
    functional disturbance.  If a study can be done only with
    expensive and complex equipment, the investigators must
    balance the relative importance of the problem against the
    cost factors.

(b) Does the test measure one or more aspects of system 
    functioning, e.g., does it examine a specific physio-
    logical quantity or a combination of functions?  For 
    example, Bouhuys (1971) states that, to test the 
    hypothesis that the early stages of sarcoidosis are 
    characterized by increased stiffness of the lungs, the 
    static recoil curves of the lungs should be measured.  A 
    less specific test would be to measure lung compliance, as 
    factors other than lung stiffness are involved in the 

(c) To what extent is the test able to distinguish normal from
    abnormal functioning?  Tests that depend on a forced
    exhalation are the most frequently used in respiratory
    epidemiology.  PEFR (peak expiratory flow rate), FEV

    (forced expiratory volume), and airway resistance can be
    used to assess impaired lung function.  However, these
    tests cannot be used to determine specific diseases, and
    may not be sensitive to small changes that can start in
    the peripheral airways.  The other indices of respiratory
    function that involve reductions in expiratory flow rates
    at 50% and 25% of vital capacity are more sensitive
    indicators of early airways disease (Ingram & O'Cain,
    1971).  McFadden & Linden (1972) found that measurements
    of mid-expiratory flow rate were reduced in heavy smokers
    in whom FEV, airway resistance, and the maximum expiratory
    flow rate were normal.  Leeder and coworkers (1974) have
    demonstrated that changes in maximum expiratory flow rates
    at low lung volumes show greater differences between
    normal and asthmatic children than FEV or PEFR.  The tests
    of small airway function are important since expiratory
    flow rates at high lung volumes may be normal, but
    measures of closing volume and frequency dependence of
    compliance may reveal early functional abnormality.  In a
    study by Buist and coworkers (1973), an estimate of
    nitrogen closing volume was found to be a more sensitive
    test for distinguishing normal from abnormal individuals
    than FEV, or expiratory flow rates.  However, the closing
    volume test involves a problem with reproducibility
    (Martin et al., 1973).  The PEFR is considered a useful
    adjunct to clinical studies, but is not recommended
    otherwise (Ferris, 1978).  Ferris (1978) then concluded
    that tests other than spirometry and, occasionally, in
    occupational studies, the diffusion capacity/total lung
    capacity ratio (DLco), are impractical and unnecessary in
    epidemiological studies.

(d) The degree to which the test has been standardized, the
    observer and instrument measurement variability assessed,
    the relationship studied of the measure to variables such
    as age, sex, and height and whether it can be administered
    to large numbers of people, are further important
    considerations in the selection of an epidemiological
    test.  Some of these points are discussed below in section

4.4.3.  Standardization of methods

    The standardization of methods and criteria for defining 
disease is essential to facilitate comparisons between different 
studies.  The pooling of data increases confidence concerning 
the relationship of environmental risk factors in the development 
of cardiovascular and respiratory diseases.  The Epidemiology 
Standardization Project (Ferris, 1978) was a major undertaking to 
standardize tests of pulmonary function and chest radiographs for 
epidemiological use in addition to a questionnaire on respiratory 
symptoms.  Evidence of the advantage of standardization was 
reported by the Pooling Project Research Group (1978) who pooled 
data from a number of major independent longitudinal studies of 

risk factors such as serum cholesterol, blood pressure, smoking 
habits, relative weight, and ECG abnormalities in the incidence of 
major coronary illness. 

    Blackburn (1965) and Rose and his collaborators (1982) have 
described a classification system, now in a revised form and known 
as the Minnesota Code 1982, from which it is possible to evaluate 
ECG measurements according to exact dimensional criteria, and which 
has been used extensively in the earlier form in epidemiological 
surveys.  Schwartz & Hill (1972) examined the problem of 
standardization for cholesterol analysis, as different laboratories 
use different methods of extraction of cholesterol from serum and 
of isolation of cholesterol, and different types of colour 

    Comparability of results is affected if different investigators 
use different methods.  Problems occur if different investigators 
use different numbers of practice and test trials and examine 
different quantitative aspects of the measures of cardiovascular 
and respiratory function.  The British Medical Research Council 
(Medical Research Council, 1966) recommended the use of three 
technically satisfactory exhalations after two practice trials in 
the forced expiratory manoeuvre.  Tager and coworkers (1976) have 
compared the three largest and the three last of five forced 
expiratory manoeuvres and recommended that five forced exhalations 
should be made and the three largest recorded.  Epidemiological 
studies based on a single measurement of blood pressure may give an 
erroneous representation of the prevalence of hypertension (Armitage 
et al., 1966; Carey et al., 1976; Hart, 1970).  The mean value 
derived from single measurements taken at relatively lengthy 
intervals corresponds more closely to the subject's general level of 
blood pressure than do single readings (Armitage et al., 1966; 
Armitage & Rose, 1966). 

    The ability of the cardiovascular or respiratory test to detect 
functional abnormality can be influenced by changes in diagnostic 
criteria.  For example, Rose & Blackburn (1968) state that the 
definition of myocardial infarction to persons with Minnesota Code 
1:1 (extensive Q/QS changes) may reduce the prevalence to well below 
1%, even in countries where the incidence is high. 

    In blood pressure measurement using a sphygmomanometer, potential 
sources of variability include variable size of cuff and deflation
speeds and observer preferences for certain terminal digits, 
usually 0 or 5 (Rose et al., 1964).  There have been several 
studies (Blackburn, 1965; Higgins et al., 1965) of observer 
variations in the coding of electrocardio-grams by the Minnesota 
Code.  If the interobserver and interinstrument variations are 
substantial, the small difference observed between the 
subpopulations being studied may lie within this range.  To 
eliminate these influences, which can confound the interpretation 
of results, it is advisable to use random-zero sphygmomanometers 
and standardization of the sound at which the blood pressure levels 
are recorded. 

    Intrasubject variability can be affected by various factors 
including age, sex, seasonal and diurnal variation, stress, genetic 
characteristics, and drugs.  Blood pressure readings can be 
influenced as well by the time of day, amount of rest, physical 
strain, and pain or excitement that precedes the measurement (Bevan 
et al., 1969).  Rose and his coworkers (1982) state that meals, 
glucose administration, smoking, and heavy physical exercise in the 
two hours preceding the ECG recording can influence the measurement. 

    The PEFR, FEV, and FVC (maximum expiratory volume with maximum 
effort to full inspiration) can vary with season (Morgan et al., 
1964) and time of day (Guberan et al., 1969).  Green and coworkers 
(1974), in a study of the variability of maximum expiratory flow 
volume curves found that flows above 70% of vital capacity varied 
substantially between individuals, which was attributed to the 
degree of individual efforts.  There is some variability in 
measures taken at low lung volumes owing to failure to reach the 
same minimal lung volume on repeated efforts (Black et al., 1974). 

    The precise criteria for distinguishing normal from abnormal 
functioning are complicated by the relationship of cardiovascular 
and respiratory variables to age, sex, and other factors.  Techniques 
have been developed to overcome the problem of controlling for these 
confounding effects.  Ferris (1978) discussed such techniques for 
respiratory tests.  For blood pressure measurement (Tyroler, 1977), 
these included the use of standardized blood pressure scores referred 
to a common age and of age-specific standard deviations from the 
means for that stratum, which is the most comonly used and reported 
method for the adjustment of blood pressure for major age, sex, and 
ethnic origin effects.  Black and collaborators (1974) and Knudson 
and coworkers (1976a,b) have determined "normal" values for the 
expiratory flow volume curve at selected lung volumes.  Although 
they provide prediction equations based on height, age, and sex, 
intrasubject variability in performance can reduce the ability of 
the measure of expiratory volume to distinguish normal from abnormal 

4.4.4.  Radiographic measurements

    The epidemiological use of radiography has made considerable 
advances, especially in developing methods for studying dust 
diseases of the lung.  Historically, schemes devised were related 
to the diagnosis and classification of severity of disability for a 
few specific diseases.  The current concept is that observers 
should describe and quantify the opacities observed in the chest 
radiograph, rather than interpret these findings.  Although changes 
are considered to lie on a continuum, the ILO International 
Classification of Radiographs of pneumoconioses provides a means 
for the systematic recording and ranking, in a simple reproducible 
way, of radiographic changes in the chest produced by dust (ILO, 
1980).  It provides a text and a set of standard films that define 
the limits of normality and guide the film reader in the 
classification and quantification of radiographic features.  A full 
plate posteroanterior film is required and the technical desiderata 
for radiographic technique and reading conditions are specified.  

Each film has to be read by several trained and quality controlled 
readers.  The derivation of a final score for the film is still a 
matter for discussion (Fox, 1975), as is the sequential study of a 
series of films (Reger et al., 1973).  Nevertheless, valuable use 
of the scheme has been made in epidemiological studies of coal-
miners for dust standard setting (Jacobsen, 1972) and it has been 
used for studies in other industries as for example by the 
Employment Medical Advisory Service, 1973 (asbestos), by Lloyd-
Davies, 1971 (foundries), and by Fox and co-workers, 1975 
(potteries).  It has been possible to detect interaction between 
the occupational environment and the cigarette smoking habit. 

    On the other hand, radiographic signs of obstructive lung 
diseases and their relation to morphology have not been very useful 
nor have they been standardized (Higgins, 1974).  Chest radiographs 
can be used to obtain total lung capacity (TLC), but TLC is not a 
critical epidemiological measurement (Ferris, 1978).  Chest radio-
graphs are still a major tool for appraising the possibility of 
lung cancer. 

4.4.5.  Hypersensitivity measurements

    Immunological reactions are functional changes that can be 
environmentally induced (Litvinov & Prokopenko, 1981).  Immediate 
hypersensitivity (IgE mediated) responses may be associated 
directly or indirectly with air pollutants or smoking (Lebowitz, 
1981).  It is hypothesized that particulates of the size that 
impact on the nasal pharyngeal area, or some gases such as sulfur 
dioxide, may release mediators through a variety of pathways.  
These mediators may lead to an asthmatic type reaction, that is 
generally acute, but may be associated with chronic effects.  There 
are various skin tests for immediate hypersensitivity, including 
those that use histamine or non-specific antigens.  There are also 
serological tests for IgE.  Responses to skin tests have been used 
as an intervening variate in the study of air pollution effects 
(van der Lende, 1969).  The tests are easy to administer and 
measure, the standard protocols have been developed and studies 
have been performed examining immediate hypersensitivity responses 
to various environmental antigens (Pepys, 1968).  Various B and T 
cell immune mechanisms may be appropriately studied in chronic 
obstructive pulmonary disease responses to environmental agents. 

    Bronchial challenge is another method by which the role of 
immediate hypersensitivity and bronchoconstriction is assessed.  
Standard protocols for challenges with spirometric measurements 
before and after challenge have been formulated, such as for 
histamine (van der Lende et al., 1973). 

4.4.6.  Example:  Effects of manganese on the respiratory and 
cardiovascular systemsa

    Saric and colleagues (1975, 1977a, 1977b) studied the effects 
of manganese aerosols in and around a ferro-maganese alloy smelter 
in Yugoslavia that has been operating since before 1940.  The 
hypotheses were that the exposed workers and the community 

population experienced more acute respiratory diseases, especially 
pneumonia and bronchitis, and the exposed workers would show some 
neurological signs and increased blood pressure.  Unexposed control 
workers and populations were used for comparisons.  Ambient sulfur 
dioxide, sulfate, and respirable manganese concentrations were 
sampled within the factory, at five sites around the factory, and 
at a control point, 25 km distant.  Sulfur dioxide was low every-
where with an annual mean of 13-27 µg/m3 and a maximum of 47-122 
µg/m3, and sulfate was about 9.9-13.9 µg/m3.  Mean concentrations 
of manganese were 0.3-20.4 µg/m3 in the plant (400 workers) and 
0.002-0.302 µg/m3 in the control plants (about 800 workers). 
Zones around the plant had annual mean manganese concentrations of 
0.236-0.39 µg/m3 (8 700 people), 0.164-0.243 µg/m3 (17 100 people),
0.042-0.099 µg/m3 (5 300 people) and the manganese levels at the   
control point were 0.024-0.04l µg/m3.                              

    A retrospective study of work absenteeism due to pneumonia and 
bronchitis from workers' medical files showed an increase in       
incidence rates correlated with exposure.  The British Medical     
Research Council questionnaire (Medical Research Council, 1966), a 
neurological questionnaire, spirometry, and blood pressure         
measurements were used in the cross-sectional study of workers.    
There was more chronic respiratory disease in exposed smokers   
compared with unexposed smokers, but not in exposed non-smokers. 

    Spirometric results were lower in those exposed for more than  
ten years.  Neurological signs occurred in 16.8% of exposed workers 
and less than 6% of controls, but clinical manganism was not       
present in any group.  Diastolic blood pressure was also higher in 
exposed workers than controls.  A prospective study of town        
inhabitants using data available from the local chest clinic,      
showed increased incidences of acute bronchitis and peribronchitis, 
but not of pneumonia, related to the zone of residence.  Children   
under age 4 years were especially affected.  School children and    
their families were studied with spirometry and acute disease       
questionnaires (as carried out by Shy et al., 1970).  Those in the  
exposed town showed a tendency towards lower spirometric values and 
had a higher incidence of acute respiratory disease.                

4.5.  Effects on Nervous System and Organs of Sense

4.5.1.  Central and peripheral nervous systems

    Disorders of the nervous system are mediated by alteration in 
structure or function of the various components of the central 
nervous system, the motor and sensory portions of the peripheral 
nervous system as well as functional and organic disorders of the 
autonomic system.  Environmental agents may act directly on the 
nervous system or the injury may be mediated by circulatory 
disturbance or by vascular accident. 

a   Based on the contribution from Dr M. Saric, Institute of
    Medical Research and Occupational Health, Zagreb, Yugoslavia.

    Some signs may be observed clinically including alterations in 
aim and sensation.  Disorders of the autonomic system may be 
manifested as functional disturbances of the cardiovascular system 
(e.g., cardiac arrhythmia and vasospasticity).  Vasospasticity can 
be expressed clinically by signs of skin pallor and coldness.  In 
practice, while it is occasionally possible to observe the effects 
mediated by a single lesion of a particular part of the central 
nervous system, complex disorders may occur involving symptomatic 
and behavioural changes that are gross enough to be detectable on 
clinical examination or more subtle changes that require 
electrophysiological examination and sophisticated behavioural 
investigation for their detection. 

    Friedlander & Hearne (1980) reviewed available neurological 
examination methods used for epidemiological studies including:  
 electroencephalography (EEG) for studies on styrene and mixed
solvents;  nerve conduction velocity measurements for styrene and
mixed solvents;  sensory nerve conduction velocity measurements 
for mixed solvents;  measurements of slow nerve fibres conduction 
 velocity for lead and trichloroethane;  electromyography (EMG)
for trichloroethane, 2-hexanone (methylbutyl ketone), mixed 
solvents and lead;  electroneuromyography for mixed solvents; 
 specific questionnaires for chlordecone (Kepone) (tremor, 
nervousness), methylmercury (paraesthesia), trichloroethane 
(headache, nervousness) and maganese (tremor and other symptoms). 

    The neurological methods used for an epidemiological study of 
employed workers exposed to lead and of controls (Baloh et al., 
1979) included clinical neurological examinations, oculomotor 
function tests, nerve conduction studies and auditory measurements.  
The clinical neurological examination, though known not to be 
sensitive for detecting the early effects of increased lead 
absorption, was carried out primarily to exclude confounding 
conditions.  The neurologist was required to pay special attention 
to early signs of peripheral neuropathy.  Oculomotor function tests 
were carried out to make precise measurements of the extraocular 
muscles and their brain control system.  The test battery included 
tests of saccadic and smooth pursuit and optokinetic nystagmus.  
Nerve conduction studies included fast and slow motor conduction 
velocities in ulnar and peroneal nerves and sensory latencies of 
ulnar and sural nerves.  Environmental and skin temperatures were 
carefully controlled during these examinations.  Under standard 
acoustic conditions, a battery of tests was carried out to 
determine the magnitude of hearing loss and to differentiate the 
sites of lesion. 

    In a survey of shoe and leather workers exposed to solvents, a 
very high prevalence of polyneuropathy was observed in persons, 
supposedly normal according to clinical examination of muscle tone, 
tendon reflexes, muscle wasting and normal sensation, when compared 
with a control population (Buiatti et al., 1978).  Electromyography 
was carried out using needle electrodes and motor conduction 
velocity was measured in the median and lateral popliteal nerves. 
Conduction velocity was considered to be in the pathological range, 
when it was lower than the 95% confidence limits for values in the 

normal control population of the same age.  For an electromyo-
graphical diagnosis of polyneuropathy, the presence of spontaneous 
activity, polyphasia, and irregular potentials and a reduced 
interference pattern were considered, as well as alteration in the 
size of motor responses and sensory action potentials.  Examining 
motor nerve maximum conduction velocity, the authors observed not 
only that maximum conduction velocity fell with age, but that 
exposure to solvents increased the physiological lowering with age. 
When analysing the decrease in maximum conduction velocity as a 
function of age in the group of workers not considered to have 
polyneuropathy, it was possible to differentiate between them and 
the normal control population.  However, it was not a reliable 
criterion for the diagnosis of polyneuropathy, when taken in 
isolation.  These methods are not always considered suitable for 
field work.  The use of surface electrodes is more socially 
advantageous than needle electrodes. 

    EMG and EEG have also been used for epidemiological studies of 
exposures to organophosphorus compounds (Roberts, 1979; Duffy & 
Burchfield, 1980). 

    With regard to the effects of physical factors, some 
neurological tests have been used.  In an epidemiological study of 
vibration white finger (Pelmear et al., 1975), the objective tests 
used included depth test aesthesiometry, two point discrimination 
and the vibrotactile threshold.  In another epidemiological 
study on effects of vibration, Vaskevich (1978) employed 
electroaesthesiometry to measure impairment of electrotactile 
sensation and elevation of pain threshold as well as reflex 
response times.  He discussed other methods of measurement and how 
they might be used for discriminating between the various sites in 
the nervous system for the functional lesion.  A range of electro-
physiological tests exists for sensory motor nerve conduction and 
for the study of motor power and physiological and adventitious 
movement of eyes and limbs.  However, electroneurophysiologists 
will not always agree on the battery of tests required or on their 

4.5.2.  Ear:  Effects of sound

    The study of the prevalence and degree of hearing loss lend 
themselves to field studies employing screening audiometers or 
diagnostic audiometers and the provision of transportable sound-
insulated booths.  National and international standards have been 
set for virtually all aspects of hearing monitoring.  These 
standards range from the specification for the design and operation 
of audiometers and the practice of audiometry, to calibration and 
testing of calibration, designs for headphones and their testing, 
and for the design and performance of acoustic booths.  Over the 
past decade, there have been a number of changes in these standards.  
Therefore, before embarking on an audiometric study, it would be 
wise to refer to the appropriate national institute responsible for 
standard setting or to the International Organization for 
Standardization (ISO).  Taking such standards into consideration, 
it is possible to design methods suitable for epidemiological audio-

metric studies (Health and Safety Executive, 1978).  Where 
appropriate, further laboratory tests may be employed to pinpoint 
the site of the lesion and to quantify the order of disability 
(Baloh et al., 1979). 

    An example employing mobile facilities on a large scale, but 
with relatively unsophisticated means of analysis, is given in an 
occupational noise surveillance study in Austria involving 165 000 
tests (Raber, 1973).  Facilities have now been developed for the 
direct recording of audiograms on tape or disc systems and for 
their filing in a minicomputer for easier data handling and 
subsequent analysis. 

    Apart from sound energy, neurotoxic agents may affect 
perceptive hearing and balance, including lead and carbon monoxide, 
as may barotrauma and electrical energy.  Agents that produce 
obstructive chronic inflamatory lesions in the nasopharynx may lead 
to conductive disorders of hearing and disorders of balance.  The 
investigation of non-conductive deafness is a laboratory activity 
as is the investigation of disorders of balance. 

4.5.3.  Eye and vision

    Environmental and occupational eye diseases include:  (a) 
irritation of the cornea and conjunctiva (acute or chronic) from a 
variety of gases, fumes, and dusts (e.g., bromine, chlorine 
dioxide, hydrogen sulfide) leading to discomfort and temporary 
visual impairment from coloured haloes; (b) corneal dystrophy, for 
example, from occupational exposure to coaltar pitch, leading to 
deformation of the cornea, keratoconus, and progressive 
astigmatism; (c) staining of the cornea, by quinones and other 
organic compounds, which may be intense enough to impair vision and 
affect colour vision; (d) lens changes because of deposition of 
metal or alteration of the lens material producing a range of 
opacities (e.g., due to ultraviolet and infrared light) from 
asymptomatic to blinding cataract formation; (e) retinal injury 
which may be asymptomatic or, where the fovea is affected, lead to 
a loss of central vision and fine discriminations; (f) optic 
neuritis (e.g., due to alcohol or tobacco), with effects ranging 
from peripheral field loss to total blindness; (g) visual cortical 
atrophy, from alkylmercurial compounds, with various degrees of 
visual impairment; (h) derangement of accommodation, due to some 
organic compounds; (i) diplopia, due to carbon monoxide, methyl 
chloride, or alkyltin compounds; (j) visual field constriction, 
associated with exposure to carbon disulphide, carbon monoxide, or 
ethyl glycol; and (k) nystagmus, due to poor illumination. 

    One of the earliest effects on the eyes demonstrated on the 
victims from the atomic bomb explosions in Hiroshima and Nagasaki, 
was the occurrence of lenticular opacities.  A small number of 
well-developed cataracts have been observed, but for the most part, 
the lesions have consisted of a posterior lenticular sheen or of 
small subcapsular plaques that do not interfere with vision (Finch 
& Moriyama, 1980). 

    Eye strain, in numbers of persons affected, is the most 
prominent condition.  Even where refractive errors are absent or 
have been corrected, it may occur under circumstances because of 
lighting of inadequate intensity, poor contrast, glare, imperfect 
visual presentation often compounded by psychological factors 
including management deficiencies.  Although it does not threaten 
vision, it makes demands on nervous energy and may lead to 
symptomatic complaints of headaches of various degrees of 
intensity; there may also be signs of conjunctival irritation. 

    The study of organic lesions of the eye necessitates the 
services of a clinical ophthalmologist:  the apparatus required and 
the conditions of examination do not lend themselves readily to 
field study.  Simple near and distant vision tests may be used, 
which are designed to determine the ability to discriminate objects 
that subtend particular angles at the eye.  They have been designed 
to deal with the literate and the non-literate, but it is necessary 
to standardize the lighting and other conditions under which they 
are carried out.  Relatively easy tests exist for stereoscopic 
vision and colour appreciation that commend themselves for field 
use; however, their execution and interpretation may be difficult.  
Transportable and portable instruments have been designed to 
provide a battery of tests of visual functions under standardized 
conditions for screening purposes.  The study of visual fields has 
been a time-consuming exercise, and current developments are 
restricted to the clinic. 

    Any attempt to measure the effect of environment on functional 
or organic disease in particular populations, unless the excesses 
are extreme or the lesions are peculiar, requires that their 
incidence or prevalence rate in a control population be determined.  
For example, with the limited evidence on human exposure to the 
electromagnetic spectrum, it is not possible to state an exposure/
response relationship for the appearance of lenticular opacities 
from cataracts, nor to determine whether the disease of the eye is 
commoner in exposed populations.  However, contributions to the 
determination of the prevalence of common eye disorders have been 
made in a study, known as the HANES study, of a population of some 
10 000 persons, using a symptom questionnaire and a standardized 
ophthalmological examination (US National Health and Nutrition 
Examination Survey, 1972; US Department of Health, Education and 
Welfare, 1973).  To test the hypothesis that radiant energy (sun-
light) was an important causal factor in the development of 
cataracts, Hiller and collaborators (1977) used data from the HANES 
study and from a group of blindness registries in the USA, and 
related prevalences to average annual sunlight hours in each 
geographical area, taking non-cataract disease as controls.  A 
technique for studying exposed and controlled persons employing two 
ophthalmologists to minimize observer bias is given by Elofsson and 
co-workers (1980) together with a protocol for ophthalmological 

4.6.  Behavioural Effects

4.6.1.  Effects of environmental exposure

    The effects on mental health of environmental agents fall into 
three broad categories.  The first includes effects that are 
directly attributable to structural or functional damage to the 
central nervous system (CNS), such as those resulting from carbon 
monoxide or carbon disulfide poisoning.  The second category 
includes effects that arise as a generalized behavioural (or 
psychosocial) response of the individual to a physiological 
impairment caused by a noxious factor, for example, the syndrome of 
irritability, depression, and loss of interest in a person who has 
developed a chronic lung disease following long-term exposure to 
industrial dusts. 

    A classical epidemiological precedent was the study of the 
etiology of pellagra (Goldberger, 1914), in which environmental 
causes of what had been previously considered an endogenous 
disease, were revealed by epidemiological mapping of the cases of 
pellagra psychosis and the distribution of dietary patterns in the 

    For the best results in applying epidemiological methods it is 
necessary to be familiar with the clinical manifestations of CNS 
responses to exogenous insults, and with the individual's ways of 
coping with impairment.  These responses, constituting the first of 
the above categories, have been described as 'exogenous reaction 
types' (Bonhoeffer, 1909), or as 'psychoorganic syndromes' 
(Bleuler, 1951), and can be presented as shown in Table 4.1. 

Table 4.1.  Clinical manifestations of organic damage to the 
central nervous system (CNS)

 CNS response   Generalized                 Focal
Acute          Confused states1            Epileptic seizures,
                                           Other neurological
Chronic        Korsakov-type psychosis2    Frontal lobe syndrome3
               Dementia4                   Temporal lobe syndrome5
                                           Parietal lobe syndrome6
1 Disorientation, excitement or stupor, incoherent speech,
  hallucinations, acute anxiety or euphoria.
2 Memory disturbance, confabulations.
3 Personality change, loss of control over own behaviour.
4 Loss of learning ability, intellectual deterioration, apathy,
  social withdrawal.
5 Language difficulties, apraxia, emotional instability.
6 Reading difficulties, arithmetic difficulties, disturbance of
  body image.

    The scheme in Table 4.1 does not exhaust the great variety of 
clinical manifestations of organic damage to the central nervous 
system, but the conditions listed are characteristic examples.  
Distinctions between acute/chronic, and generalized/focal, are 
never quite clearcut, and many transitional phenomena may occur 
between them. 

    The second category of mental health effects, as mentioned 
above, comprises a wide variety of behavioural responses of a 
predominantly neurotic or emotional type, often accompanied by 
characteristic physiological dysfunctions (psychosomatic 
reactions).  Such responses may arise, either under the influence 
of unpleasant or stressful environmental stimuli (e.g., noise), or 
as a behavioural (symbolic) overlay of physical impairment.  In 
both cases, they are mediated by the previous experience, 
attitudes, and coping skills of the personality, as well as by the 
social environment. 

    Until recently, knowledge about the mental health effects of 
environmental agents was derived mainly from clinical studies, 
following short- or long-term exposure to agents, such as lead, 
mercury, organic mercury compounds, manganese, thallium, methyl-
bromide, tetraethyl lead, and carbon monoxide.  A synthesis of such 
knowledge has been provided by Lishman (1978).  Many behavioural 
toxicology studies have been focused on the effects of heavy metals 
or chemicals that are common in industry, such as petroleum 
distillates, jet fuel (Knave et al., 1978), organic solvents 
(Hänninen et al., 1976; Lindström, 1973), and carbon disulfide 
(Hänninen, 1971).  A review of recent research in such areas is 
provided by Horvath (1976).  Environmental pollutants of a physical 
nature have also been the subject of studies, for example, the 
investigations of the mental health effects of aircraft noise 
around airports (Gattoni & Tarnopolsky, 1973; Jenkins et al., 1979; 
Shepherd, 1974). 

4.6.2.  Indicators and measurements of effects

    The epidemiological approach requires suitable indicators and 
the application of some standardized measurement of effects, though 
cruder methods based on clinical records and hospital admission 
data have been of use as well (Edmonds et al., 1979).  The indicators 
of behavioural effects of noxious environment agents fall into two 
broad groups as shown in Table 4.2. 

    Psychological tests have proved to be effective in the dectection 
and assessment of organic brain damage, and relatively simple 
techniques, such as Raven's progressive matrices and vocabulary and 
memory tests, can be both reliable and practical in field studies 
involving the screening of a large number of individuals.  A 
concise guide to the most widely-used psychometric tests is included 
in Lishman's review (1978). 

Table 4.2.  Indicators of behavioural effects of noxious 
environmental agents
Indicators                     Examples of methods of measurement
1.  Measures of psychological  Psychological and psychophysiological
    and psychophysiological    tests (e.g., performance, verbal,
    functioning                learning tests; skin conductance
                               changes in response to standard

2.  Measures of mental state   Psychological and psychiatric
    and behaviour              screening questionnaires;
                               standardized psychiatric interviews
Note:  Neurological effects must be clarified first, using methods
       previously described (section 4.5)                             

    Instruments, used in the standardized assessment of mental 
state and behaviour, fall into two groups: 

(a)  Screening instruments

    Most of these are relatively short questionnaires that can be 
self-administered or used as an interview by a research assistant, 
for example, the General Health Questionnaire (GHQ) developed by 
Goldberg (1972), which is now available in several languages.  In a 
modified form, this has been used in several WHO coordinated cross-
cultural psychiatric studies.  The scoring of the "yes/no" responses 
of the subject to a number of questions is simple, and cut-off 
points are provided for the sorting of respondents into a group of 
likely cases of psychological disorder and a group of likely non-
cases.  The GHQ and other similar instruments are not diagnostic 
tests, in the sense that they do not lead to a subclassification of 
the detected cases into diagnostic categories.  Therefore, their 
great usefulness is as first-stage screening tools for the 
selection of affected individuals for more detailed investigation. 

(b)  Mental state and psychosocial functioning assessment tools

    These include the Present State Examination (PSE), which has 
been the main assessment tool in major cross-cultural studies of 
functional psychoses (WHO, 1974; WHO, 1979a) and is available in 19 
different languages.  The PSE is a structured interview guide based 
on clinical concepts, and has been designed for use by psychiatrists, 
who receive brief special training before applying the instrument 
in research projects.  It can be used to elicit and record 
information about the presence or absence of 140 clinical symptoms 
the operational definitions of which are provided in a glossary 
(Wing et al., 1974). This information can be processed by a 
computer diagnostic program (CATEGO) which produces a standard 
diagnostic classification of cases.  The PSE exists also in a 
shorter version, the administration of which in an interview can 
take 10-45 min (depending on the number of symptoms present) and 
can be applied by interviewers who are not psychiatrists, provided 

that they receive the special training required.  Although the PSE 
provides systematic coverage of all major areas of psychopathology, 
it was not specifically designed for the study of organic brain 
syndromes.  It is necessary to combine the PSE interview with the 
application of some simple tests for organic brain damage, if such 
pathology is suspected among the population studied. 

    Among other instruments available for epidemiological research 
is the Disability Assessment Schedule (DAS) developed by the World 
Health Organization for standardized recording of information about 
disturbances in social behaviour and adjustment.  This is a semi-
structured interview that can be used as a complement to the PSE 
and can be applied by a social worker or a research assistant. 

4.6.3.  Interpretation of data

    Adequate consideration should be given to procedures necessary 
for achieving and sustaining a high level of interobserver and 
intraobserver reliability of the assessment of psychological and 
behavioural variables (section 4.1.2). 

    Interaction effects are often the rule in behavioural research, 
and these should be taken into account, both in the design of the 
study and in data analyses (Cooper & Morgan, 1973).  The mental 
health effects of environmental agents will be influenced not only 
by the length and intensity of exposure, but also by factors such 
as age, sex, previous personality, learning, and social experience, 
and various individual levels of susceptibility.  Therefore, the 
question of the selection of a control population is a more complex 
one when behavioural effects are studied.  However, the recognition 
of many interacting factors must not lead to an unnecessary 
proliferation of items in the research instruments as this would 
make a meaningful analysis of the data collected impossible. 

4.7.  Haemopoietic Effects

    The haematological system is affected by many environmental 
agents, both chemical and physical.  These environmental agents can 
be loosely grouped under the following two headings, reflecting 
their relative involvement in the haematological system in 
toxicological action. 

4.7.1.  Environmental agents inducing direct toxic effects in the
haematological system

    These include agents such as benzene and ionizing radiation 
affecting haemopoietic precursor cells.  In a recent retrospective 
evaluation of the occupational history of patients with acute non-
lymphoblastic leukaemia in conjunction with cytogenetic findings, 
it was suggested that perhaps 50% of such leukaemias had been 
caused by environmental agents (Mitelman et al., 1978); though this 
claim awaits confirmation.  A number of environmental agents 
directly affect circulating red cells.  Inhalation of relatively 
low levels of arsine (AsH3) induces acute haemolysis, which is 
frequently fatal. 

    Methaemoglobinaemia is produced by certain aromatic hydro-
carbons, including aniline dyes and sulfonamide derivatives.  
Alteration in the ability of haemoglobin to bind or release oxygen 
is a potential mechanism of toxicity of environmental agents.  
Chemicals with oxidizing properties can induce Heinz body 
haemolytic anaemias.  There are a number of inherited abnormalities 
of red cell function that increase individual susceptibility to 
environmental agents, producing methaemoglobinaemia or Heinz body 
haemolytic anaemias.  These include unstable haemoglobinopathies, 
such as methaemoglobin reductase deficiency and glucose-6-phosphate 
dehydrogenase (G-6-PD) deficiency.  The latter is relatively common 
in a mild form, possibly associated with systemic protection 
against malaria.  There are, however, more severe variants of G-6-
PD deficiency in which exacerbations due to environmental agents 
could have potentially grave consequences. 

    Many environmental agents also affect immunoglobulins 
circulating in the blood that are important in sensitization, 
allergic reactivity, and general host resistance. 

4.7.2.  Environmental agents inducing indirect toxic effects in the
haematological system

    The haemopoietic system may be indirectly affected by almost 
any chronic disease state.  For instance, environmental processes 
producing chronic lung disease may lead to secondary polycythaemia, 
while those causing chronic renal disease will generally result in 
anaemia.  An elevated granulocyte count is the usual response to 
acute injury of any organ system.  There are also agents that have 
direct effects on the haematological system, but exert their 
principal toxicity in other organs.  An example is lead:  the 
anaemia and relatively specific changes in haem metabolism are of 
diagnostic value and provide insight into certain mechanisms of 
effect.  However, the more important signs of dysfunction occur in 
the central and peripheral nervous system, varying with age at 
exposure and the nature and exposure levels of the lead compounds. 

4.7.3.  Measurements and their interpretation

    Blood cells are one of the most accessible human tissues, and 
therefore relatively more is known about these cells, both in terms 
of understanding basic biomolecular processes as well as associating 
changes in this system with disease processes.  The information 
obtainable from laboratory evaluation of blood cells is pertinent 
both to disorders of the haematological system and to diseases 
primarily affecting many other organs in which blood cell changes 
occur as secondary manifestations.  Its accessibility, at no 
significant risk to the subject, makes it possible to include 
fairly simple examination of mature blood cells in population 
studies.  Recent advances in medical technology have resulted in 
improved ability to perform routine haematological studies that are 
reproducible, rapid, and inexpensive.  Thus, many formerly highly-
specialized laboratory procedures may be used now in epidemiological 

    The obvious haematological laboratory tests for use in 
epidemiological studies are the standard procedures usually covered 
by the term "complete blood count".  There is a wide range of 
"normal" values for most blood elements.  The counts cited as 
abnormal also vary greatly among laboratories.  Unless particular 
care, such as the analytical quality assurance procedures, is 
taken, the use of routine blood counts can be an insensitive means 
of detecting small but real differences between populations due to 
differences in exposure to environmental agents. 

    For population studies of circulating erythrocytes, where 
complex automated equipment is not available, the microhaematocrit 
performed on approximately 0.1 ml of blood in a capillary 
centrifuge is probably preferable to the spectrophotometric 
determination of haemoglobin levels and certainly preferable to the 
manual red cell count.  One of the more recent advances in 
automated cell counting is the use of machinery to perform 
differential white cell counts.  Though it is still too soon to 
gauge the relative effectiveness of this procedure, accurate 
differentials, in conjunction with a total leukocyte count, may be 
of value in studying populations for the effects of environmental 
pollutants.  For instance, lymphocytopenia is known to be a 
consequence of benzene exposure and there are suggestions that 
lymphocyte levels may be affected by such diverse situations as 
polybrominated biphenyl toxicity and zinc deficiency. 

    Modern instrumentation has also been developed for the rapid 
evaluation of other specific haematological parameters of more 
limited application.  Two excellent recent examples are instruments 
of use in the study of lead absorption and neonatal 
hyperbilirubinaemia.  A single drop of blood placed on a filter 
paper, is then automatically inserted into a compact light-weight 
fixed florometer providing an immediate digital read out of zinc 
protoporphyrin, free erythrocyte protoporphyrin, and erythrocyte 
protoporphyrin levels.  These species of protoporphyrin are 
increased following lead absorption and should therefore be an 
excellent tool for population studies of inorganic lead effects 
(Eisinger et al., 1978).  For the interpretation of the results, a 
knowledge of the chemobiokinetics of lead is essential.  An 
instrument, which depends on specific fluorescence, measures free 
bilirubin in minute amounts of neonatal blood.  A widely-adopted, 
relatively simple and reproducible assay might make possible the 
type of comparison studies that would lead to identification of 
environmental factors involved in neonatal hyperbilirubinaemia. 

4.7.4.  Example:  Effects of low lead concentrations on workers, healtha

    Studies of low-intensity exposures to lead have been conducted 
in the USSR on male and female solderers and on unexposed female 
controls.  The solderers were generally (64%) exposed to levels of 
lead below 0.01 mg/m3 (the USSR maximum permissible level);  95%

a   Based on the contribution from Professor A.V. Roscin,
    Order of Lenin Central Institute for Advanced Medical
    Training, Moscow, USSR.

of samples were below 0.02 mg/m3.  Lead on the skin ranged from
0.043 mg/100 cm2 (before work) to 0.057 mg/100 cm2 (after work),
and lead on the clothes averaged 0.22 mg/100 cm2.  It was estimated
that intake by ingestion and absorption through any external 
surface was less than that by inhalation and all three intake modes 
at work were less than lead intake from food (0.3 mg/day). 

    To determine the possible effects of the exposure to lead, 
various haematological and biochemical indices were measured.  The 
results showed disturbances of porphyrin metabolism and changes in 
enzyme activity, protein fractions of blood serum, liver function, 
and blood cell production.  This indicates that exposure to 
relatively low concentrations of lead, i.e., levels less than the 
established health standard, produced a complex of haematomorpho-
logical and biochemical changes, which must be regarded as early 
signs of effects of lead. 

    After termination of the lead exposure, the previous bio-
chemical and haematological deviations in the women workers tended 
to return to normal, and the porphyrin metabolism and other indices 
investigated showed normal values. 

4.8.  Effects on the Musculoskeletal System and Growth
4.8.1.  Effects of environmental exposure                     
    Only rarely do musculoskeletal disorders have a fatal out-                 
come and so virtually all studies are of morbidity, in terms of                 
disturbance of, or interference with, normal structure and                     
function.  Apart from certain occupational disorders, the most                  
important target to be considered is bone.  Physical development 
may be affected by exposures to some chemicals.                     

    Some of the reported environmental effects on the musculo-             
skeletal system are summarized as follows (in most cases the 
association is related to extreme occupational or accidental       
exposures rather than to those that would normally be encountered  
in the general environment):   ionizing radiation, specifically
bone-seeking isotopes, may lead to bone necrosis or bone sarcoma
(as with strontium);  ultraviolet radiation may precipitate
systemic lupus, through activation of lysosomal enzymes;  
 electrical shock, in which trauma may lead to cervical disc 
degeneration;  ultrasonics, which may lead to bone necrosis;  
 extremes of barometric pressure may, due to gas embolism, lead
to aseptic necrosis (head of the humerus and femur) in caisson
disease and related conditions;  thermal sensitivity has been
associated with Raynaud's syndrome and aggravation of rheumatic
syndromes;  vibration may lead to Raynaud's syndrome, carpal 
decalcification, occasional soft tissue injury (bursitis, muscle 
atrophy, Dupuytren's contacture), or arthrosis (especially in 
elbow);  fluoride exposure, skeletal fluorosis;  iron exposure, 
siderosis progressing to spinal osteoporosis, destructive lesions,
and arthropathy (especially in hands) as seen in Bantus;  lead
 exposure, gout associated with lead poisoning;  arsenic exposure, 
osteoarthropathy;  cadmium exposure, secondary osteomalacia

resulting from renal damage;  vinyl chloride exposure, osteolysis;
 asbestos exposure, hypertrophic osteoarthropathy with pulmonary
disease;  phosphorus exposure, bone necrosis (phossy jaw);  poly-
 chlorinated biphenyls (PCBs) exposure, diminished growth in boys
who had consumed rice oil contaminated with PCBs (Yoshimura, 1971) 
and smaller babies than normal, born to mothers with this disease 
(Yamaguchi et al., 1971). 

4.8.2.  Identification of effects

    Many musculoskeletal disorders are diagnostically vague; they 
usually lack a specific feature or diagnostic test and may as a 
result be heterogeneous, with similar effects resulting from 
different causes.  The methods for detection have not been very 
well developed.  It is necessary to be alert to the array of 
syndromes that may be encountered, but systematic searches are 
cumbersome.  By adopting a screening approach, it is possible to 
limit consideration to three features, though each unfortunately 
poses its own problems.  The first, pain and weakness, can be 
elicited by questionnaires, but with all the attendant difficulties 
of behavioural phenomena related to subjective experience.  The 
second, functional changes, can be elicited by physical 
examinations and functional tests, many of which are tests of other 
systems, as musculoskeletal changes may be secondary (see above). 

    The third, structural changes particularly in bone, if they 
call for radiographic detection, raise ethical problems about 
radiation exposure as well as requiring technological sophistication 
(i.e., X-ray equipment, film processing facilities, etc.) and 
greater expense.  For example, epiphysial deposition of lead (lead 
line) may be detected in children by radiography.  The epidemiological 
use of radiography in the study of bone pathology, where small areas 
of rarefaction or reaction to necrosis feature, merits the same 
scrupulous attention to the establishment of reading standards, the 
training of quality control readers, and the improvement of film 
techniques, as in the case of chest radiography.  Deformities such 
as lordosis and kyphosis of the spine and limitation of movement of 
joints may be measured objectively in a standard manner (Russe & 
Gerhardt, 1975). 

    Indirect measures, such as incidence of disability or absence 
statistics, lack specificity as they are associated with multiple 
factors, as are population prevalence rates (Bennett & Burch, 
1968a,b).  Standardized epidemiological methods, diagnostic 
techniques, and serological studies for rheumatic diseases have 
been discussed by Bennett & Wood (1968). 

4.8.3.  Intrinsic liability

    Biological and genetical factors contribute to variation 
between individuals in their susceptibility to outside influences.  
Differences in disease experience related to age and sex are very 
evident, though most of these have still not been accounted for.  
Human leukocyte antigens (HLA) and haemoglobinopathies of SS and SC 
genotypes have been associated with several musculoskeletal 

disorders in recent years.  Other conditions such as spondylolis-
thesis may predispose individuals to the development of severe 
musculoskeletal changes like an incapacitating back symptom (Wood, 

4.8.4.  Extraneous influences

    The influence of very general and non-specific aspects of 
the individual's surroundings and highly particular disturbances 
of those circumstances may be confounding.  This is very evident 
with geographical variations;  uncertainty arises about the 
relative importance of ethnicity (cultural or genetic), lifestyle, 
and specific agents in particular environments, such as minerals 
in the water supply.  The ubiquity of many rheumatic disorders 
also gives rise to problems.  Thus, the suggestion arises that 
the frequency of a well-recognized existing disorder may be 
increased in certain environmental circumstances.  As in other 
situations, the occurrence of graded variation rather than 
discontinuous experience tends to blunt the precision of analysis 
and to make establishment of a causal relationship more difficult. 

4.8.5.  Development states

    In the case of studies of development, subjects may be 
categorized in terms of weight/height relationships.  Some 
population studies are facilitated by not requiring persons 
to remove footwear, in which case allowance requires to be made 
for this artefact.  Posture during measurement also requires to 
be standardized.  Alternatively, skin thickness may be measured 
at standard sites using spring-loaded standardized calipers 
(Billewicz et al., 1962). 

    Biological age standards and sexual development indices have 
been determined for clinical use:  they may be adapted for 
epidemiological purposes.  Studies of childhood physical 
development as influenced by the environment have been conducted in 
the USSR (Melekhina & Bustueva, 1979) and in Poland (Pilawska, 
1979).  In studying the effects of pollutants on growth and 
nutrition, ethnic and cultural factors should be carefully taken 
into account (Chandra, 1981). 

4.8.6.  Example : Endemic fluorosisa

    For a number of years, signs and symptoms of endemic fluorosis 
had been noted to occur in residents of several areas in India 
(e.g., Punjab, Andhra Pradesh, Karnataka).  Several epidemiological 
studies were carried out in clusters of villages where the 
population was affected.  There was a high incidence of dental 
mottling (50-70%) in those who had skeletal fluorosis.  These 
subjects had joint pains, muscle wasting, and developed severe bone 
deformities, including sclerosis, kyphosis, and calcification, seen 
on X-ray.  Nerve compression, with signs of radiculomyelopathy, was 
found on examination.  The epidemiological studies showed higher 
rates in males than in females, in areas with sandy soil and where 
the summer water source was well water.  Blood and urine samples 

yielded high levels of fluoride - 6 mg/litre and up to 20 mg/litre 
respectively (normal levelsb are only traces in blood and less than 1 
mg/litre in urine).  Some bone specimens had levels up to 7 g/litre 
(the normal rangeb was less than 300 mg/litre).  The water samples 
obtained from the community had levels up to 14 mg/litre (Siddiqui, 
1955; Singh et al., 1961; Singh & Jolly, 1962). 

4.9.  Effects on Skin               
4.9.1.  Environmentally-caused skin diseases

    Diseases of the skin, except skin cancer, are rarely life-     
threatening, though they can be a considerable annoyance, either in
terms of effects on an individual or in terms of the number of     
persons that may be affected.  The effects observed may result from
the direct local action of the agent or may be mediated as part of 
a systemic disorder.  In their causation, host factors such as     
idiosyncracy, hyperreactivity, and hypersensitivity may play a     

    Adverse effects observed following exposure to physical and   
chemical agents include the following:  unwanted pigmentation or  
loss of pigmentation; premature ageing with alteration of        
subepithelial connective tissue; inflammation, necrosis and      
atrophy; eczematous dermatitis, photoactinic sensitization; skin 
cancer (basal cell carcinoma, epithelioma and malignant melanoma), 
precancerous skin conditions and similar conditions of the mucose 
of the buccal cavity; acne; drying; maceration; hair loss or      
dystrophy of scalp hair and alteration of body hair; and disorders 
of the nails.                                                     

    Infections with microorganisms may complicate these conditions 
by aggravating a local skin lesion or by inducing adverse effects  
mediated by immunological mechanisms at distant sites.             

    The agents associated with the disorders may be part of the 
general environment or may be found in foods, drugs, cosmetics, 
other consumer products, or occupationally.  In the general 
environment, ultraviolet light is responsible for skin cancers 
among lightly pigmented inhabitants of sunny climates (section  Other portions of the electromagnetic spectrum, found 
within the general environment, do not play a significant role in 
the causation of skin disorders, unless there is a personal 
idiosyncracy or sensitization by chemicals.  Low relative humidity 
and cold, again in association with personal susceptibility are 
responsible for dry scaly erythematous lesions on exposed areas 
(chapping).  Cold on its own can produce injury on fingers and toes 
(chilblains) and other exposed parts. 

a Based on the contribution of Professor S.R. Kamat, K.E.M.
  Hospital, Bombay, India.
b Provided by the National Institute of Occupational Health,
  Ahmedabad, India.

    Food additives, drugs, and cosmetics have been responsible for 
skin eruptions produced by relatively simple local sensitization as 
well as photo- and actinosensitization.  Additives that enhance the 
keeping quality, or other performance, of foodstuffs have been 
responsible for outbreaks of dermatitis.  Colouring agents such as 
tartrazine, used in drug formulae and foodstuffs, have produced 
sensitization. Several materials used for cosmetic purposes may be 
allergens, including orris root, bergamot, wool alcohol, parabens, 
and eosin dyes. 

    Through occupation, a person may be exposed to a range of 
irritant, sensitizing, and carcinogenic agents.  For example, 
coaltar materials have the potential for all three effects, with the 
sensitization also extending to photosensitivity.  Among the more 
interesting recently-observed materials is vinyl chloride which, in 
addition to its other systemic effects, is associated with scleroderma-
like skin lesions accompanied by micro-vascular changes (Maricq et 
al., 1976). Excessive exposure to ionizing radiation is associated 
with acute inflammatory effects, which may be followed by atrophic 
changes and skin tumours.  Ultraviolet light in substantial dosages, 
which may be incidental to a process like welding or constitute the 
essence of a process for the polymerization of resins, can be 
phototoxic or photoallergic (WHO, 1979b).  Formaldedehyde is also a 
skin sensitizer through industrial as well as domestic exposures 
(Gupta et al., 1982). 

    Predisposing conditions that render persons hypersusceptible 
to environmental agents include the atopic diathesis, which 
predisposes to sensitization, and inherited conditions.  Xeroderma 
pigmentosum, a recessive autosomal disease in which there is 
absence of enzymes involved in DNA repair, renders persons 
excessively sensitive to ultraviolet light and leads to a very high 
frequency of skin cancer.  Linking skin eruptions with systemic 
disease is a hereditary form of photosensitivity reported among 
North and South American Indians.  This appears to be an autosomal 
dominant state leading to pleomorphic light eruptions, which become 
secondarily infected with nephritogenic organisms to form a hazard 
for the individual. 

    Errors of porphyrin metabolism, where crises may be provoked by 
drugs or ultraviolet light, are also important conditions in 
certain populations and individuals. Malnutrition commonly 
presenting with multiple deficiencies is associated with cutaneous 
and mucosal dystrophy. 

4.9.2.  Epidemiological methods of study

    Dermatological examinations are essentially clinical, but they 
are susceptible to a standardized approach with a protocol designed 
for ease of subsequent analysis.  Thus, for example, a substantial 
number of persons inadvertently exposed to polybrominated biphenyls 
(PBBs) were subject to a range of investigations including a 
detailed scrutiny of finger nails and toenails, scalp hair, body 
hair, general skin lesions, acne, and lesions of the oral cavity 
(Selikoff & Anderson, 1979).  Studies on skin cancer have been 
discussed in section 4.3.2. 

    The use of patch testing is primarily a clinical procedure; for 
practical reasons, it may have limited application in field 
studies.  In population studies, it is common to discover a higher 
prevalence of disease than is reported by spontaneous complaint.  
Thus, while the use of general practice records and hospital 
outpatient records may be of value for the study of skin cancer, an 
assessment of the full burden of other skin lesions depends on a 
systematic study of the population at risk and a carefully matched 
control population. 

4.10.  Reproductive Effects

4.10.1.  Effects on reproductive organs

    A wide variety of environmental factors act directly on the 
gonads, or indirectly by interfering with the complex regulatory 
mechanisms of sexual and reproductive functions.  Physical agents 
most often mentioned in relation to genetic disorders include 
ionizing radiation, non-ionizing electromagnetic waves, vibrations, 
and high temperatures.  The chemicals most likely to produce 
genital disorders are heavy metals and organic solvents. 

(a)  Female

    The only common and easily detectable index in women, that can 
be asked for by questionnaire, is the occurrence of menstrual 
disorders such as dysmenorrhoea, oligomenorrhoea, or amenorrhoea.  
Other more complex assessments are not suitable for epidemiological 

(b)  Male

    Symptoms of decreased libido and functional disorders can be 
revealed by simple questionning, but are not specific enough to be 
of much value.  Testosterone blood levels yield more information 
about hormonal production.  Routine spermiograms for the early 
assessment of the influence of environmental agents on reproductive 
function are not suitable for environmental studies. 

4.10.2.  Genetic effects

    Environmental factors such as ionizing radiation and some 
chemical compounds may induce changes in human germ and somatic 
cells.  Evaluation of mutagenic effects in these cells should be 
made separately, as methods of study for the two types of cells 
differ significantly. 

    Mutagenic agents may induce different kinds of damage in the 
genetic material.  Methods for the detection of chemical mutagens 
have been described by Hollaender (1971-1976), Hollaender & de 
Serres (1978), and Kilbey and coworkers (1977).  The time between 
the origin of a mutation and its manifestation depends on its mode 
of inheritance. 

    Mutations may be responsible for a sizeable fraction of 
spontaneous abortions, congenital malformations, and mental and 
physical defects, and it has been advised that sentinel diseases 
known to be genetically determined or due to a mutation should be 
monitored in human populations.  Those recognizable at birth will 
probably be picked up by a birth-defect recording system (section 
4.10.4).  Others, which develop later, will need to be detected by 
other means - possibly notification, or the detection of "new" 
cases, when they start to attend medical institutions - primary 
care centres or hospitals. 

    The evaluation of mutagenic effects in germ cells under the 
influence of environmental factors involves a comparison of 
frequencies of gene and chromosome diseases in the control and 
exposed groups.  The most complete investigations of the genetic 
effects of ionizing radiation have been carried out on the 
population of Hiroshima and Nagasaki, exposed during atomic 
bombing (Neel et al., 1974).  In the progeny of persons 
surviving after the explosion of atomic bombs, there were no 
noticeable changes in the proportion of sexes (recessive lethal 
mutations in X-chromosome), in the frequency of chromosome 
diseases, or in the mortality rate (section  In the 
USSR, the frequency of spontaneous abortions is regarded as a 
major index of mutational impairments (Bochkov, 1971).  Shandala 
& Zvinjackovskij (1981) reported an increase in the frequency of 
spontaneous abortions in relation to the level of ambient air 

    Many chemicals may induce chromosome aberrations in somatic 
cells.  These include vinyl chloride monomer (Funes-Cravioto et 
al., 1975; Purchase et al., 1978), and a number of other industrial 
chemicals and drugs (Evans & Lloyd, 1978).  Assessment of genetic risks

    Few methods are at present available to assess the presence of 
mutagenic agents in the human body (Sobels, 1977).  Ehrenberg and 
co-workers (1977 a,b) have developed a method to estimate the 
frequency of induced mutations by determining the degree of electro-
philic substitution of proteins as haemoglobin in the exposed 
persons.  In a study by Strauss & Albertini (1977), an autoradio-
graphic method was reported for the detection of 6-thioguanine-
resistant lymphocytes.  The method should have the advantage of 
being capable of detecting somatic cell mutations  in vivo. 

    Another approach to assessing the presence of mutagenic agents 
in the human body consists of testing samples of blood or urine 
with sensitive microbial assay systems (Legator et al., 1978).  
Mutagenic activity has also been assessed using human faeces and 
breast milk.  Indications for chromosome breakage activity can be 
obtained in short-term lymphocyte cultures from peripheral blood 
samples.  These aberration yields in somatic cells cannot be 
correlated, however, with the frequencies of translocations to be 
expected in the germ cells.  Other indicators for genetic activity 
concern sister-chromatid exchanges in peripheral blood cells and 

morphological abnormalities of spermatozoa (for the latter, see 
Wyrobeck & Bruce, 1978).  Thus, an increase in the proportion of 
abnormal sperm atozoa has been observed in direct relation to the 
degree of cigarette smoking (Viczian, 1969).  Epidemiological 
surveys relating heritable damage in man to exposure to chemical 
mutagens have not yielded statistically convincing results, except, 
perhaps, in the case of cigarette smoking (Mau & Netter, 1974). 

4.10.3.  Fetotoxic effects

    Some substances absorbed by the mother pass across the 
placenta, but others do not.  The substances transported into the 
fetus are not necessarily distributed within the fetal tissues in a 
similar way to that in the mother's tissues.  It is possible that a 
substance administered to the mother or entering her blood stream 
is not of a sufficient dosage itself to harm the fetus, but 
metabolites developed during the elimination of a substance from 
the mother may pass across the placenta and be harmful to the fetus 
(Longo, 1980).  Adverse effects on the fetus must be distinguished 
from adverse effects on the germ cells, before fertilization.  A 
symposium, reported by Boué (1976), reviewed the knowledge, up to 
that date, on this subject. 

    In order to interpret data about fetal toxicity, it is 
desirable to measure the reproductive efficiency of couples 
(Levine et al., 1980) and the number of spontaneous miscarriages as 
distinct from abortions.  The difficulties in the field of 
reproductive epidemiology are well reviewed by Buffler (1978) and 
Erickson (1978) and available methods have been reviewed by Hemminki 
et al. (1983) and Leck (1978).  All show the necessity of collecting 
reliable data about exposure to substances that might be toxic to 
the fetus.  Measurement of fetotoxic effects

    To make a quantitative study of fetal loss, the pregnant women 
must first be identified at an early stage of her pregnancy. 

    Loss of fetuses in the first trimester is difficult to quantify 
because, in many women, irregularity of the menses may confuse the 
identification of early pregnancy.  Loss in the third month is more 
easily recognised, because of the menstrual bleeding periods that 
will have been missed, and it is more likely that the pregnancy has 
been reported to a doctor.  However, the chance that miscarriage 
can occur without the women noting the event or receiving medical 
care is high.  It is likely that women observe and report 
spontaneous abortions very individually, which makes interview 
studies on these abortions liable to bias.  Even though notes on 
early spontaneous abortions are not found in medical records, it is 
advisable, for confirmation of data, to consult such sources in 
studies on spontaneous abortions (Hemminki et al., 1983). 

    Legal abortions may obscure attempts to measure spontaneous 
fetal loss and means for counting the effect of this group should 
be provided if early fetal loss is to be studied accurately.  The 
loss of a fetus may not be a matter of the mother's problems alone.  
An abnormal fetus is frequently aborted (Alberman, 1976).  Thus, to 
measure abnormalities of aborted fetuses involves obtaining the 
fetus for subsequent examination.  Loss rates and the proportions 
with specified abnormalities may be measured and analysed by 
various factors, such as drug usage, substances used in employment, 
food, water, and other environmental influences that may affect the 
health of the fetus. 

    During the second trimester, there are three main types of 
fetal loss:  spontaneous loss, abortions (legal and quasi-legal) of 
an unwanted child, and legal termination after the diagnosis of an 
abnormal child.  Again, discrimination among these three groups is 
essential, if the toxic effects are to be distinguished from the 
chromosomal effects.  Although very difficult, attempts should be 
made to examine dead fetuses from all three sources in order to 
determine the number malformed, so that those in which genetic 
damage is present may be distinguished from those where toxic 
damage to the fetus has occurred  in utero. 

    In the third trimester (after about 24 weeks), premature labour 
or legal or quasi-legal termination is quite likely to result in 
the delivery of a live baby and problems arise in many countries as 
to whether the fetus from a termination is to be registered as a 
live baby.  Dead fetuses must be examined to distinguish between 
those with chromosomal anomalies and those with other anomalies.  
The latter should also be distinguished from those who have been 
injured during the birth process or during antenatal examinations. 
Occasionally, a fetus damaged during the intrauterine period may 
not show damage till later in childhood. 

    If there are sufficiently rare malformations, retrospective 
examination of the environmental factors prevailing during 
pregnancy may help to identify a causal agent (Bakketeig, 1978).  
This was done successfully in demonstrating the association of 
thalidomide in pregnancy with gross limb malformations in the 
offspring (Lenz, 1962). 

    Longitudinal studies of pregnancies with detailed case 
histories give a good picture of toxic effects, though many years 
may have passed by the time the data have been collected and 
processed.  Often, such studies indicate 'significant' effects but 
lack replicate observation, thus necessitating other investigations 
(Rumeau-Rouquette et al., 1978).  To overcome this problem, in a 
longitudinal study of l4 774 women who gave birth in 2l clinics in 
the Fereral Republic of Germany between 1964 and 1972, data was 
analysed in two sets so that the second set could be used to 
corroborate or refute the findings in the first set (Deutsche 
Forschungsgemein-schaft, 1977).  The study involved recording many 
details about normal pregnancies in order to obtain data on the 
relatively few pregnancies that ended in spontaneous abortion, or 
where the baby was born with anomalies. 

    An alternative strategy involves collecting data from mothers 
of abnormal babies and from a control mother who has had a normal 
baby (Saxen et al., 1974; Greenberg et al., 1977).  If data are 
recorded routinely for all pregnancies, the records can be examined 
for mothers of abnormal babies and for a matched control mother; 
such a procedure would substantially reduce the risk of bias being 
introduced by the outcome of the pregnancy. 

    Transplacental carcinogenesis studies in human beings have to 
date only conclusively established one such process, involving the 
administration of diethylstilbestrol in high doses to mothers in 
the first trimester whose daughters presented with the rare tumour 
of adenocarcinoma of the vagina at 14-22 years of age (Herbst & 
Scully, 1970). 

    The testing of a hypothesis that a given environmental factor 
is causing malformations is probably demonstrated most convincingly 
by a study in which the factor in question is excluded from some of 
the mothers, i.e., a selective avoidance trial.  When the incidence 
of an abnormality is less than 5%, such a trial would need to be 
extensive before a significant difference between mothers excluded 
from, and mothers exposed to, the factor is demonstrated.  However, 
when women in such a study are at a known and high risk of having 
abnormal babies, an avoidance trial can be used without using 
control mothers, as was done by Nevin & Merrett (1975).  They 
studied mothers who had already given birth to infants with central 
nervous system anomalies and who abstained from eating potatoes 
during subsequent pregnancies and found that the avoidance of 
potato eating did not reduce the risk of giving birth to infants 
with these anomalies. 

4.10.4.  Registries of genetic diseases and malformations

    Registration of spontaneous abortions, birth defects, and 
perinatal deaths might not always reflect genetic effects, because 
these events take place as a result of changes in the hereditary 
structures in gametes as well as from a number of other non-genetic 

    Only a few programmes have so far developed genetic registries 
on a wide scale with a defined population.  For instance, genetic 
disorders in the population are included in the Birth Defects 
Registry (established in 1952) developed in British Columbia, 
Canada.  Originally concerned with the delivery of medical 
services, it includes the provision of incidence and prevalence 
statistics on handicapping illnesses in all age groups, and 
provides a basis for surveillance, and genetic counselling.  It has 
attempted to ascertain and document all relevant cases in British 
Columbia, though registration is not mandatory.  Some data are 
provided on certain genetic conditions such as cleft lip and per
palate, clubfoot, and Down's Syndrome.  There were 1.42 liveborn 
1000 live births with Down's Syndrome. 

    Borgaonkar and his co-workers at North Texas State University, 
USA, established an International Registry of Abnormal Karyotypes - 
later called Repository of Chromosomal Variants and Anomalies 
(Borgaonkar, 1980; Borgaonkar et al., 1982).  They contacted all 
established cytogenetic laboratories around the world and have an 
open-door recruitment policy.  With the support of the World Health 
Organization, they distributed several cumulative listings of the 
Repository.  The most recent, the Ninth Listing, has data from 140 
contributors on about 200 000 cases.  The modes of ascertainment of 
cases and the total number of cases studied are included in the 
report.  All types of variations and anomalies are systematically 
arranged in a format used earlier in preparing a catalogue of 
chromosomal variants and anomalies - Chromosomal Variation in Man.  
By analysing data in the Repository and the catalogue, it has been 
possible to draw some conclusions about the origin of certain 
chromosomal disorders. 

    Some specific types of new chromosomal mutations are almost 
always "environmentally" induced.  The ring chromosomes and 
isochromosomes have been reported more than 600 times in the 
Repository and there are about 500 more published cases.  An 
examination of the origin of these cases shows that with few 
exceptions almost all the cases are new mutations; that is, the 
parents, when examined, have been found to have normal chromosomes.  
Most of the examples are also "genetic lethals" in that they do not 
reproduce.  Early death does not seem to be a characteristic of 
these individuals.  Almost all the cases come to attention because 
of the medical problems that the individual encounters, including 
developmental and maturational anomalies.  Very few cases are 
detected in general population surveys.  Use of the Repository in 
developing uniform growth patterns and syndrome delineation has 
been well documented (Mulcahy, 1978). 

    The use of cytogenetic approaches in the monitoring of 
industrial workers has been defined, presumably with systematic 
development of registries.  Records prior to and during employment 
can provide data for assessment of the genetic effect of the 
occupational exposure (Kilian et al., 1975). 

    Registries of birth defects exist in a number of countries.  
For example, in Finland, there is a registry of all congenital 
malformations reported from all hospital deliveries, with 
registration of the various environmental factors involved.  This 
data base has been used for a study of the relationships between 
solvent exposure of fathers and mothers and congenital abnormalities 
of the nervous system (Holmberg & Nurminen, 1980). 

    The following European Economic Community (EEC) study of 
congenital malformations provides an example of an international 
collaborative study on their registration. 

4.10.5.  Example:  EEC study of congenital malformationsa

    In 1974, the Committee of Medical and Public Health Research of 
the EEC decided to promote, as its concerted action, an international 
cooperative study on the registration of congenital malformations. 

    After a feasibility study conducted in 1975 and 1976, the 
Concerted Action on Congenital Abnormalities and Multiple Births 
was established in February 1978.  The study is supervized by a 
steering committee whose members are nominated by the participating 
member countries.  Initially, 15 study areas were proposed within 9 
countries (Belgium, Denmark, France, Federal Republic of Germany, 
Italy, Ireland, Luxembourg, Netherlands, and the United Kingdom).  
In 1979, the participation of Greece with an additional area was 
approved.  The Concerted Action Project in Registration of 
Congenital Abormalities and Twins has received the acronym -

    The long-term objective of this study is to test the 
feasibility of carrying out epidemiological surveillance in the 
countries of the EEC, taking surveillance of congenital 
abnormalities and multiple births as an example. 

    The specific objectives of the study are:

(a)  To set up, within each selected area in each country, a
      population-based register of congenital abnormalities and
     multiple deliveries.  In order to achieve full recording,
     ideally the outcome of each conception in women resident in
     the defined area should be known.  This involves searching for
     congenital malformations, and biochemical and chromosomal
     anomalies in aborted fetuses, in live and stillborn infants,
     in dead children, and during childhood.  Babies from multiple
     deliveries should be recorded at birth.

(b)  To study the methods of data collection in each centre,
     to evaluate the effects of these in biasing the data
     collected, and to propose and test ways to circumvent these

(c)  To monitor the incidence rates reported in different
     population groups at different times in order to identify
     possible etiological factors.

(d)  To create, in each country, an area where reporting is
     reliable, so that base line rates are available for use in
     calibrating any national warning system established for the
     detection of adverse environmental influences by allowing the
     interpretation of a reported increasing rate.

a   Based on the contribution from Professor M.F. Lechat,
    Catholic University of Louvain, Brussels, Belgium, with
    the help of Dr J.A.C. Weatherall, Office of Population
    Censuses and Surveys, London, England.

(e)  To evaluate the effectiveness and efficiency of screening
     programmes and preventive measures.

(f)  To provide a well-documented set of individuals recorded
     in a defined population for further specific studies, such as
     follow-up studies of cases with specified malformations, in
     order to compare the results of different treatment regimens.

(g)  To establish the means by which multiple births can be
     efficiently registered at birth and how information can be
     collected that will make possible the reliable and cheap
     recording of zygosity.

    A feasibility study showed that considerable variations existed 
in the recording of malformations in different countries and in the 
collection of morbidity and mortality statistics and of other 
relevant epidemiological data concerning children, both in the 
definitions used and in the methods of processing of the data.  To 
ensure valid comparability of the data, it was decided to concentrate 
first on well-defined geographical areas in each country where 
studies could be performed. 

    The study has concentrated, at the start, on recording malformed 
infants at birth.  Special studies are being undertaken to measure 
the efficiency of the reporting of cases among the births to women 
living in each area.  As soon as good birth coverage is achieved, 
the observations will be extended to the recording of all the 
abnormalities found in the children born in the area during their 
childhood.  Those discovered in spontaneously- and legally-aborted 
fetuses will be recorded as well.  By 1980, the recording of 
multiple births was being carried out in only a few areas, but 
other areas will start to record multiple births, when the methods 
for recording zygosity have been established in each area. 

4.11.  Effects on Other Major Internal Organs

    Environmental factors may have beneficial or harmful effects on 
internal organs.  The gastrointestinal system, especially, receives 
beneficial essential metals and other minerals from the environment.  
On the other hand, some metals and many other chemical compounds 
are hazardous to these organs, if concentrations are sufficiently 

4.11.1.  Renal system

    Renal damage can be caused by many chemical compounds or 
physical factors.  Depending on the kind and concentration of 
noxious agents, and the intensity and duration of exposure, an be 
caused by nephrotoxic products such as mercury, chromium, arsenic, 
and ethylene glycol. 

    Subacute and chronic renal diseases are caused by a wide 
variety of environmental factors and can generally be related to 
either glomerular or tubular injury.  Nephrotoxic agents may lead 
to a quantitative alteration in the filtration rate or to 

qualitative changes in the filtration pattern by influencing 
glomerular permeability.  Inorganic mercury, cadmium, potassium 
perchlorate, and different chelating agents increase glomerular 
permeability.  Hydrocarbon solvents and pertroleum products may 
produce antiglomerular basement-membrane-mediated glomerulone-
phritis (Van Der Laan, 1980). 

    The most common type of chronic renal impairment of toxic 
origin is tubular injury with suppression of tubular reabsorption.  
Proximal tubular damage is produced by all the nephrotoxic heavy 
metals such as lead, mercury, cadmium, uranium, and bismuth.  X-
radiation produces renal disorders, mainly of the tubular type.  Detection of renal diseases

    Most tests suitable for epidemiological studies do not yield 
much information about the specific causes of renal dysfunction, 
but they provide a crude measure of the degree of renal damage.  
The early stages of renal damage are seldom accompanied by 
symptoms.  Thus, questionnaires are useless in the early detection 
of renal impairment, and laboratory tests are imperative. 

(a)  Functional change

    One of the simplest tests is the assessment of renal 
concentrating capacity by measuring urinary specific gravity after 
a period of restricted fluid intake.  Results can be biased by the 
presence of glucose, proteins or other substances in the urine or 
by extrarenal factors such as hypertension or a low-protein diet. 

(b)  Test for urinary sediment

    Analysis of urinary sediment may indicate generalized kidney 
damage as for instance the number of epithelial cells excreted in 
the urine.  The presence of even microscopic haematuria must evoke 
the possibility of cancer of the urinary tract, especially in high-
risk groups. 

(c)  Test for glomerular function

    Except under certain physiological conditions characterized by 
a temporarily increased output of proteins, normal urine contains 
only small amounts of proteins.  Significant proteinuria is always 
pathological and generally reflects glomerular dysfunction 
characterized by a high relative molecular mass protein output.  
The appearance of low relative molecular mass proteins in the urine 
must be interpreted as a sign of tubular damage.  Proteinuria is 
considered an early sign of renal injury, preceding other signs 
such as aminoaciduria or glycosuria.  These tests can be used in 
epidemiological surveys. 

(d)  Test for tubular function

    All substances excreted or reabsorbed in the tubular area can 
be used as indices of tubular function.  Normal phenosulfon-
phthalein (PSP) and para-aminohippurate secretion tests indicate 
tubular functional integrity.  In general the performance of 
clearance tests requires a clinical setting and therefore they are 
not useful in masssurveys for renal dysfunction.  However, if 
comparison is made with urinary creatinine, ambulant clearance 
testing can be carried out, as was done for example by Nogawa et 
al. (l980). 

    A Fanconi-like syndrome of glucosuria, phosphaturia, and 
aminoaciduria is precipitated by most heavy metal intoxications.  
Ammonium ion excretion after acid loading may serve as an indicator 
of distal tubular function.  Renal tubular damage in persons with 
excessive cadmium intake is accompanied by an increase in urinary 
excretion of beta-2-microglobulin.  Nowadays, quantitative 
immunological methods for the measurement of beta-2-microglobulin 
(Evrin et al., l97l) and retinol-binding proteins (Bernard et al., 
l982) in urine are available and these methods facilitate the 
detection of tubular dysfunction.  The methods of diagnosing 
cadmium-induced proteinuria have been reviewed by Piscator (l982).  
Quantitative protein excretion should be relied on in preference to 
simple paper tests (Lauwerys et al., 1979; Roels et al., 1981).  
Various other factors involved in the effects of cadmium on renal 
function have been reported (Friberg et al., 1974; Tsuchiya, 1979; 
Commission of the European Communities, 1982). 

(e)  Tests for enzymuria

    Disruption of kidney cells by nephrotoxic agents results in the 
release of specific renal enzymes in the lumen of the nephron.   
Enzymes present in both serum and kidney tissue can often be 
distinguished from each other as isoenzymes and separated by 
electrophoresis.  Since the enzyme patterns of the different parts 
of the nephron are well characterized, study of enzymuria often 
makes it possible to localize the injury (for example, a rise in 
urinary acid phosphatase (EC indicates glomerular lesions, 
while an increase in alkaline phosphatase (EC suggests 
proximal tubular damage). 

    Distal tubular injury gives rise to the appearance of lactate 
dehydrogenase (EC or carbonic anhydrase (EC in 
the urine.  Other enzymes, not found in serum, appear in urine 
after toxic renal damage as, for instance, beta- N-acetylgucosamin-
idase (EC, glycine amidinotransferase (EC, etc.  
Aminopeptidase-activity (EC increase occurs in many 
pathological conditions of the kidney, especially in the case of 
tubular lesions induced by many chemicals.  Enzymuria is a highly 
sensitive and specific criterion for the early assessment of renal 
damage and precedes any other symptom either functional or 
morphological.  The usefulness of the assessment of enzymuria in 
epidemiological surveys is limited by its high cost and the need 
for specialized laboratories. 

4.11.2.  Bladder

    Bladder cancer is a known hazard in many industries resulting 
principally from exposure to carcinogenic amines.  Screening for 
early tumour development may be done by the determination of 
urinary beta-glucuronidase (EC, or by cytodiagnosis of 
tumour cells exfoliated in the urine.  The second method, however, 
requires a high degree of skill and experience for reliable 
diagnosis, but it has been accepted as a good screening method. 

4.11.3.  Gastrointestinal tract

    The gastrointestinal tract is particularly susceptible to 
environmentally-determined disease, because it is the first system 
in contact with chemicals contained in food and drink. In addition, 
through the liver and biliary system, the gut provides a route for 
the excretion of toxic chemicals, drugs, and products of metabolism. 

    There are many tests available for the detection of existing 
gastrointestinal diseases and some for the identification of 
persons at risk of them.  Not all of the tests are suitable for use 
on an epidemiological scale, because they involve sophisticated 
equipment, significant doses of radiation, or are so labour-
intensive as to be uneconomic.  The methods mentioned below are the 
minimum necessary for the identification of these diseases.  All 
are fully described in standard texts (Russell, 1978; Sleisinger & 
Fordtran, 1978; Bateson & Bouchier, 1982).  These text books also 
contain information concerning other more detailed tests, suitable 
for use in smaller groups of people, provided adequate resources are 
available.  Oesophagus

    Cancer is the main environmentally-determined disease of the 
oesophagus.  It can be detected readily by a combination of a 
clinical history and either X-ray or fibroptic endoscopy.  
Endoscopy has been found acceptable, on a population basis, in Iran 
(Crespi et al., 1979), where precursor inflammatory changes were 
detected in the oesophagus, often in quite young subjects.  In 
China, X-ray examination is widely used to screen populations at 
risk (Coordinating Group, 1975).  Exfoliative cytology of the 
oesophagus is also a valuable screening test for oesophageal 
neoplasms.  Only a simple apparatus is needed to obtain the sample, 
although a trained cytologist must look at the specimen.  The test 
is positive in 70-94% of cases with 1-2% false positives.  Stomach and duodenum

    Gastric cancer is amongst the world's commonest fatal 
malignancies.  It may be detected best by X-ray, which is the most 
accurate simple screening test for established disease.  Fibroptic 
endoscopy is also a useful diagnostic procedure and may be 
essential in distinguishing large gastric ulcers from malignant 
ulcers.  In these situations, multiple biopsies must be made;

80-90% of malignancies are detected in this way.  Much effort, 
especially in Japan, has gone into the detection of early stages of 
gastric malignancy in an effort to prevent this disease.  Population 
screening by X-ray has been widely used (Nagayo & Yokoyama, 1974). 

    Exfoliative cytology is also useful and can be performed by a 
medical assistant.  Gastric lavage is carried out, usually with 
chymotrypsin, on fasting patients and the whole test requires only 
a small stomach tube, syringe, and centrifuge.  Accuracy of 
diagnosis of 90% has been claimed for proved tumours with only 1-2% 
false positive (Brandborg, 1978).  Intestines

    The small bowel plays a vital role in digestion and absorption, 
but few environmental causes of small bowel diseases are known.  On 
the other hand, in many industralized countries the large bowel is 
one of the major sites of cancer. 

    The simplest and most widely applicable test for large bowel 
disease is examination of the faeces.  Depending on the cooperation 
of the population under study, anything from a random sample of 
stool to a full 5-day collection can be made.  Stool samples have 
been collected from randomly selected members of the public in 
studies of the etiology of large bowel cancer (International Agency 
for Research on Cancer; Intestinal Microecology Group, 1977).  
These were used for the measurement of faecal bile acid 
concentrations and faecal microflora studies.  Also valuable is 
the stool test for occult blood.  In the early detection of bowel 
cancer, this test, if done under properly controlled dietary 
conditions, can be organized for a very large number of people. 

4.11.4.  Liver

    Hepatic tumours, particularly primary tumours, often produce no 
change in standard function tests.  They may be demonstrated by any 
of a number of radioisotopic scans now available, but all involve 
considerable doses of isotope.  Hepatic ultrasonography offers a 
useful non-invasive alternative and computerized axial tomography 
scanning may also prove to be a valuable alternative, although 
expensive.  Serum alpha 1-fetoprotein appears in the blood of 
patients with primary hepatic tumours.  The proportion of positives 
varies from 30-80% depending on the area under investigation. 

    A great number of tests of hepatic function are available and 
should be tailored to the particular objective of any epidemiological 
study.  Standard texts on the subjects should be consulted (Schiff, 
1975; Sherlock, 1975).  Some of these tests are simple and accurate 
while others require great resources and are inherently dangerous. 

    Examination of the urine can be most useful in liver disease.  
The presence of conjugated bilirubin or urobilinogen is often an 
early index of disease.  Faecal examination is much less useful.  

Serum tests of liver function are widely available and easy to 
perform.  These include bilirubin, aspartate (EC and other 
aminotransferases (EC 2.6.1), gamma-glutamyltransferase (EC, alkaline phosphatase (EC, 5'-nucleotidase (EC, serum proteins, blood cholesterol and ammonia. 

4.11.5.  Pancreas

    Pancreatic cancer and pancreatitis have been often implicated 
to be related to environmental factors such as smoking and intake 
of alcohol and coffee (Wynder et al, 1973; Lin & Kessler, 1981; 
MacMahon et al., 1981). 

    However, the pancreas is one of the least accessible internal 
organs and is thus difficult to investigate.  No simple tests for 
epidemiological study are available, though some pancreatic 
function tests may be used (Mottaleb et al., 1973; Mitchell et al., 
1977) and indirect evidence of pancreatic disease may also be 
obtained from determination of serum amylase and lipase levels. 


ABRAMSON, J.H.  (1979)   Survey methods in community medicine,
2nd ed., Edinburgh, Churchill Livingstone, 229 pp.

ACHESON, E.D.  (1967)   Medical record linkage,  London, Oxford
University Press, for Nuffield Provincial Hospital Trust, 213

ACHESON, E.D., HADFIELD, E., & MACBETH, R.G.  (1968)  Nasal
cancer in woodworkers in the furniture industry.   Br. med. J.,
2: 587.

ACHESON, E.D., COWDELL, R.H., & JOLLES, B.  (1970)  Nasal
cancer in the Northamptonshire boot and shoe industry.   Br.
 med. J.,  l: 385.

ALBERMAN, E.  (1976)  The epidemiology of spontaneous abortion
and their chromosome constitution.  In: Boué, A. & Thibault,
C., ed.   Chromosomal errors in relation to reproductive
 failure.  Paris, INSERM.

ALDERSON, M.R.  (1966)  Referral to hospital amongst a repre-
sentative sample of adults who die.   Proc. R. Soc. Med.,  59:

ALDERSON, M.R.  (1967)  Data on sickness absence in some
recent publications of the Ministry of Pensions and National
Insurance.   Br. J. prev. soc. Med.,  21: 1-6.

ALDERSON, M.R.  (1972)  Some sources of error in British
occupational mortality data.   Br. J. ind. Med.,  29: 245-254.

ALDERSON, M.R.  (1977)   An introduction to epidemiology,
London, Macmillan, 226 pp.

ALDERSON, M.R. & DOWIE, R.  (1979)   Health surveys and related
 studies,  Oxford, Pergamon, 356 pp.

ARMITAGE, P. & ROSE, G.A.  (1966)  The variability of measure-
ments of casual blood pressure:  I.  A laboratory study.
 Clin. Sci.,  30: 325-335.

ARMITAGE, P., FOX, W., ROSE, G.A., & TINKER, C.M.  (1966)  The
variability of measurements of casual blood pressure:  II.
Survey experience.   Clin. Sci.,  30: 337-344.

BAKKETEIG, L.S.  (1978)  Detection of teratogens by monitoring
human births.   Contr. Epidem. Biostat. (Basle),  1: 53-56.

BALDWIN, J.A.  (1972)  Linked record health data systems.
 Statistician,  21: 325-331.

F.J., & CULVER, B.D.  (1979)  Subclinical effects of chronic
increased lead absorption - A prospective study.   J. occup.
 Med.,  21: 490-496.

BATESON, M.C. & BOUCHIER, I.A.D.  (1982)   Clinical
 investigation of gastrointestinal function,  2nd ed., pp.
1-232, Blackwell Scientific Publications, Oxford.

P.J.  (l980)  Agiosarcoma of the liver:  Annual occurrence and
aetiology in Great Britian.   Br. J. ind. Med.,  37: 213-221.

BENNETT, A.E. & RITCHIE, K.  (1975)   Questionnaires in medi-
 cine:  A guide to their design and use,  London, Oxford
University Press, 110 pp.

BENNETT, P.H. & BURCH, T.A.  (1968a)  The epidemiology of
rheumatoid arthritis.   Med. Clin. N. Am.,  52: 479-91.

BENNETT, P.H. & BURCH, T.A.  (1968b)  The distribution of
rheumatoid factor and rheumatoid arthritis in the families of
Blackfeet and Pima Indians.   Arth. Rheum.,  11: 546-53.

BENNETT, P.H. & WOOD, P.H.N.  (ed.)  (1968)  Population
studies of rheumatic diseases.  In:   Proceedings of the Third
 International Symposium,  Amsterdam, Exerpta Medica Foundation.

Comparison of retinol-binding protein and beta2-microglobulin
determination in urine for the early detection of tubular
proteinuria.   Clin. Chim. Acta,  l26: l-7.

BEVAN, A.T., HONOUR, A.J., & SCOTT, F.H.  (1969)  Direct
arterial pressure recording in unrestricted man.   Clin. Sci.,
36: 329-344.

Indices of adiposity.   Br. J. prev. soc. Med.,  16: 183-188.

BLACK, L.F., OFFORD, K., & HYATT, R.E.  (1974)  Variability in
the maximal expiratory flow volume curve in asymptomatic
smokers and non-smokers.   Am. Rev. resp. Dis.,  110: 282-292.

BLACKBURN, H.  (1965)  The electrocardiogram in cardiovascular
epidemiology: Problems in standardized applications.   Ann. N.
 Y. Acad. Sci.,  126: 882-905.

BLEULER, M.  (1951)  Psychiatry of cerebral diseases.   Br.
 med. J.,  2: 1233-1238.

BOCKOV, N.P.  (1971)   [Human chromosomes and radiation.]
Moscow,  Atomizdat, 180 pp (in Russian).

BONHOEFFER, K.  (1909)  Exogenous psychoses.  In: Hirsch, S.R.
& Shepherd, M., ed.   Themes and variations in European
 psychiatry,  pp. 499-505,  Bristol, J. Wright.

BORGAONKAR, D.S.  (1980)   Chromosomal variation in Man.  A
 catalog of chromosomal variants and anomalies,  3rd ed., New
York,  Alan R. Liss, 7l4 pp.

LEVITSKY, K., & SOKOLOFF, B.  (l982)   Repository of chromoso-
 mal variants and anomalies in man - An international registry
 of abnormal karyotypes.  Ninth Listing.  Philadelphia, Thomas
Jefferson University Press, 5l3 pp.

BOUE, A., ed.  (1976)   Prenatal diagnosis.  Paris, INSERM, 329
pp (Colloques, Vol. 61).

BOUHUYS, A.  (1971)  Pulmonary function measurements in epi-
demiological studies.   Bull. Physiopathol. Respir.,  6: 561-578.

BOYLAND, E.  (1967)  A chemist's view of cancer prevention.
 Proc. R. Soc. Med.,  60: 93-99.

BRANDBORG, L.L.  (1978)  Polyps, tumours and cancer of the
stomach.  In:  Sleisenger, M.H. & Fordtran, J.S., ed.  Gastro-
 intestinal disease,  2nd ed., Philadelphia, USA, W.B. Saunders,
pp. 752-776.

BRESLOW, N.E. & ENSTROM. J.E. (1974)  Geographic correlations
between cancer mortality rates and alcohol-tobacco consumption
in the United States.   J. Natl Cancer Inst., 53: 631-639.

BUELL, P.  (1973)  Changing incidence of breast cancer in
Japanese-American women.   J. Natl Cancer Inst.,  51: 1479-1483.

BUFFLER, P.A.  (1978)  Some problems involved in recognising
teratogens used in industry.   Contr. Epidemiol. Biostat.
 (Basle),  1: 118-137.

BULGARELLI, G.  (1978)  Relationship between clinical and
electromyographic findings and exposure to solvents, in shoe
and leather workers.   Br. J. ind. Med.,  35 (2): 168-173.

BUIST, A.S., VAN FLEET, D.L., & ROSS, B.B.  (1973)  A compari-
son of conventional spirometric tests and the test of closing
volume in an emphysema screening centre.   Am. Rev. respir.
 Dis.,  107: 735-743.

BUSTUEVA, K.A. & SLUCANKO, I.S.  (1979)   [Methods and criteria
 of the evaluation of the health of the population in relation
 to environmental pollution.]  Moscow, Meditsyna (in Russian).

W.L., III, & VAUGHAN, E.D., Jr  (1976)  The Charlottesville
blood pressure survey.  Value of repeated blood pressure
measurements.   J. Am. Med. Soc.,  236: 847-851.

Respiratory symptoms and "angina pectoris" in twins with
reference to smoking habits.  An epidemiological study with
mailed questionnaires.   Arch. environ. Health,  13: 743-748.

CEDERLOF, R., JONSSON, E., & LUNDMAN, T.  (1966b)  On the
validity of mailed questionnaires in diagnosing "angina
pectoris" and "bronchitis".   Arch. environ. Health,  13:

CHANDRA, R.K.  (1981)  Immuno-complex as a functional index of
nutritional status.   Br. med. Bull.,  37: 89.

 of cadmium.  Report presented by R.R. Lauwerys to the
toxicology section of the Scientific Advisory Committee,
Commission of European Communities (EUR 7649 EN), Luxembourg,
82 pp.

COOK, P.J., DOLL, R., & FELLINGHAM, S.A.  (1969)  A mathema-
tical model for the age distribution of cancer in man.   Int.
 J. Cancer,  4: 93-112.

COOPER, B. & MORGAN, G.  (1973)   Epidemiological psychiatry.
Springfield, Illinois, Thomas, 2ll pp.

CANCER IN NORTH CHINA  (1975)  The epidemiology and etiology
of oesophageal cancer in China:  A preliminary report.
 Chinese med. J.,  1 (3): 167-183.

MOJTABAI, A., & CASALE, V.  (1979)  Oesophageal lesions in
Northern Iran:  a premalignant condition?   Lancet,  2: 217-220.

DAWES, K.S.  (1972)  Survey of general practice records.  Br.
med. J., 3: 219-223.

(1978)  Factors related to respiratory and cardiovascular
symptoms in the United Kingdom.  J. Epidemiol. community
 Health,  32: 82-96.

 pregnancy and the development of the child.]  Harald Boldt.
Verlag KG, Boppard (in German).

DOLL, R.  (1967)   Prevention of cancer - pointers from epi-
 demiology.  The Rock Carling Fellowship 1967 - Nuffield
Provincial Hospitals Trust.  London, Whitefriars Press, l44 pp.

DOLL, R. & PETO, R.  (1981)   The causes of cancer,  Oxford,
Oxford University Press, 116 pp.

DUFFY, F.H. & BURCHFIELD, J.L.  (1980)  Long-term effects of
the organophosphate sarin in EEG's in monkeys and humans.
 Neurotoxicology,  l: 667-689.

EDMONDS, L.D., LAYDE, P.M., & ERICKSON, J.D.  (1979)  Airport
noise and teratogenesis.   Arch. environ. Health,  34: 243-247.

EHRENBERG, L. & OSTERMAN-GOLKAR, D.  (1977a)  Reaction kine-
tics of chemical pollutants as a basis of risk estimates in
terms of rad-equivalence.  In: Chanet, R., ed.   Radiological
 protection.  First European Symposium on rad-equivalence.
Luxembourg. Commission of the European Communities.  pp.
199-205 (Eur. 5725e).

K., & CALLEMAN, C.J.  (1977b)  Evaluation of the genetic risks
of alkylating agents.  III.  Alkylation of haemoglobin after
metabolic conversion of ethene to ethene oxide  in vivo.
 Mutat. Res.,  45: 175-184.

SELIKOFF, I.J.  (1978)  Zinc protoporphyrin in blood as a bio-
logical indicator of chronic lead intoxication.   J. environ.
 Pathol. Toxicol.,  1: 897-910.

E., WENNBERG, A., & WIDEN, L.  (1980)  Exposure to organic
solvents.   Scand. J. Work Environ. Health,  6: 239-273.

3.   A study of asbestos workers.  London, Department of
Employment, 2l pp.

HAYNER, N.S., CELLER, J.B., & FRANCIS, R., Jr  (1965)
Epidemiological studies of cardiovascular disease in a total
community - Tecumseh, Michigan,  Ann. intern. Med.,  62:

Parental occupation and birth defects.   Contr. Epidemiol.
 Biostat. (Basle),  1: 107-117.

EVANS, H.J. & LLOYD, D.C., ed.  (1978)   Mutagen-induced
 chromosome damage in man.  Edinburgh, Edinburgh University
Press,  355 pp.

Radioimmunoassay of beta-2-microglobulin in human biological
fluids,  Scand. J. clin. lab. Invest.,  28: 439.

Variability in answers to a questionnaire on respiratory
symptoms.   Br. J. prev. soc. Med.,  13: 175-193.

FERRIS, B.G.  (1978)  Epidemiology standardization project.
 Am. Rev. Res. Dis.,  118 (6): l-l20.

FINCH, S.C. & MORIYAMA, I.M.  (1980)   The delayed effects of
 radiation exposure among atomic bomb survivors, Hiroshima and
 Nagasaki, 1945-79:  A brief summary.  pp. 16-78  (Radiation
Effects Research Foundation Technical Report).

FLETCHER, C.M. & TINKER, C.M.  (1961)  Chronic bronchitis.  A
further study of simple diagnostic methods in a working
population.   Br. med. J.,  1: 1481-1498.

(1959)  The significance of respiratory symptoms and the
diagnosis of chronic bronchitis in a working population.   Br.
 med. J.,  2: 257-266.

FORSYTH, G. & LOGAN, R.F.L.  (1960)   The demand for medical
 care: a study of the case load in the Barrow-in-Furness group
 of hospitals.  London, Oxford University Press for Nuffield
Provincial Hospital Trust, 153 pp.

FOX, A.J.  (1975)  Classification of radiological appearance
and the derivation of a numerical score.   Br. J. ind. Med.,
32: 273-282.

FOX, A.J. & ADELSTEIN, A.M.  (1978)  Occupational mortality:
work or way of life?   Epidemiol. community Health,  32: 73-78.

R.N.J.  (1975)   A survey of respiratory disease in the pottery
 industry,  London, HMSO,  20 pp.

FOX, J.  (1977)  Occupational mortality:  a new study.   Popu-
 lation Trends,  9: 8-15.

(1974)   Cadmium in the environment,  2nd ed., Cleveland, CRC

FRIEDLANDER, B.R. & HEARNE, F.T.  (1980)  Epidemiologic
consideration in studying neurotoxic disorders.  In: Spencer,
P.S. & Schaumburg, H.H., ed.  Experimental and clinical
 neurotoxicology,  Baltimore, Williams & Wilkins, pp. 650-662.

NATARAJAN, A.T., & OSTERMAN-GOLKAR, S.  (1975)  Chromosome
aberrations in workers exposed to vinyl chloride.   Lancet,  1:

GAFFEY, W.R.  (1976)  A critique of the standardized mortality
ratio.   J. occup. Med.,  18: 157-160.

GATTONI, F. & TARNOPOLSKY, A.  (1973)  Aircraft noise and
psychiatric morbidity.   Psychol. Med.,  3: 516.

GOLDBERG, D.  (1972)   The detection of psychiatric illness by
 questionnaire.  London, Maudsley Monograph, 156 pp.

GOLDBERGER, J.  (1914)  The etiology of pellagra.   Public
 Health Reports,  29: 26.

GREEN, M., MEAD, J., & TURNER, J.M.  (1974)  Variability of
maximum expiratory flow-volume and curves.   J. appl. Physiol.,
26 : 121-125.

A.M., & HASKEY, J.C.  (1977)  Maternal drug history and
congenital abnormalities.   Br. med. J.,  11: 853-856.

GREENBERG, M. & LLOYD-DAVIES, T.A.  (1974)  Mesothelioma
Register, 1967-68.   Br. J. ind. Med.,  31: 91-104.

GUBERAN, E., WILLIAMS, M.K., & WALFORD, J.  (1969)  Circadian
variation of FEV in shift workers,  Br. J. ind. Med.,  26:

GUPTA, K.C., ULSAMER, A.G., & PREUSS, P.W.  (1982)
Formaldehyde in indoor air; sources and toxicity.   Environ.
 Int.,  8: 349-358.

HAENSZEL, W. & KURIHARA, M.  (1968)  Studies of Japanese
migrants.  I.  Mortality from cancer and other diseases among
Japanese in the United States.   J. Natl Cancer Inst.,  40:

HANNAY, D.R.  (1972)  Accuracy of health centre records.
 Lancet,  2: 371-373.

HANNINEN, H.  (1971)  Psychological picture of manifest and
latent carbon disulfide poisoning.   Br. J. ind. Med.,  28: 374.

(1976)  Behavioural effects of long-term exposure to a mixture
of organic solvents.   Scand. J. Work environ. Health.,  2:

HART, J.T.  (1970)  Semicontinuous screening of a whole com-
munity for hypertension.   Lancet,  2: 233-266.

HEALTH AND SAFETY EXECUTIVE (1978)   Audiometry in industry:
 Report of the HSE Working Group on Audiometry,  London, HMSO,
18 pp.

HEASMAN, M.A. & LIPWORTH, L.  (1966)   Accuracy of
 certification of cause of death.   London, HMSO, 133 pp
(General Register Officer Studies on Medical Population
Subjects, No.20).

Assessment of methods and results of reproductive occupational
epidemiology:  Spontaneous abortions and malformations in the
offspring of working women.   Am. J. ind. Med.,  4: 293-307.

HERBST, A.L. & SCULLY, R.E.  (1970)  Adenocarcinoma of the
vagina in adolescence.  A report of 7 cases including 6
clear-cell carcinomas.   Cancer,  25: 745-757.

Adenocarcinoma of the vagina.  Association of maternal
stilbestrol therapy with tumour appearance in young women.
 New Engl. J. Med.,  284: 878.

HIGGINS, I.T.T.  (1974)   Epidemiology of chronic respiratory
 disease; a literature review.  Research Triangle Park, NC,
Environmental Protection Agency,  129 pp (Environmental Health
Effects Research Series EPA - 650/1 - 74-007).

HIGGINS, I.T.T. & KELLER, J.B.  (1970)  Predictions of
mortality in the adult population of Tecumseh;  Respiratory
symptoms, chronic respiratory disease and ventilatory lung
function.   Arch. environ. Health,  21: 418-424.

(1968)  Chronic respiratory disease in mining communities in
Marion County, West Virginia.   Br. J. ind. Med.,  25: 165-176.

HIGGINS, I.T.T., KANNEL, W.B., & DAWBER, T.R.  (1965)  The
electrocardiogram in epidemiological studies:  reproducibility, 
validity and international comparison.  Br. J.  prev. soc. Med.,  
19: 53-68.

HIGGINSON, J.  (1960)  Population studies in cancer.   Acta.
 Um. Int. Cancrum,  16: 1667-1670.

HIGGINSON, J. & MUIR, C.S.  (1976)  The role of epidemiology
in elucidating the importance of environmental factors in
human cancer.   Cancer Prev. Det.,  1: 79-105.

HILLER, R., GIACOMETTI, L., & YUEN, K.  (1977)  Sunlight and
cataract: An epidemiological investigation.   Am. J. 
 Epidemiol.,  105: 450-459.

HOLLAENDER, A.  (1971-1976)   Chemical mutagens.  Principles
 and methods for their detection,  New York, Plenum Press.
Vol. 1-2, 610 pp, Vol. 3, 304 pp, Vol. 4., 364 pp.

HOLLAENDER, A. & DE SERRES, F.J.  (1978)   Chemical mutagens.
 Principles and methods for their detection,  New York, Plenum
Press, Vol. 5, 348 pp.

PEARSON, N.J.  (1966)  A comparison of two respiratory symptom
questionnaires. I.  Methodology and observer variation.   Br.
 J. prev. soc. Med.,  20: 76-96.

(1979)  7.  Tests of lung function.  Special issue on
particulate air pollution.   Am. J. Epidemiol.,  110: 635-650.

HOLMBERG, P.C. & NURMINEN, M.  (1980)  Congenital effects of
the central nervous system and occupational factors during
pregnancy.   Am. J. ind. Health,  l: 167-176.

HORVATH, M., ed.  (1976)   Adverse effects of environmental
 chemicals and psychotropic drugs.  Amsterdam, Elsevier, 334 pp.

INGRAM, R.H. & O'CAIN, C.F.  (1971)  Frequency dependence of
compliance in apparently healthy smokers versus non-smokers.
 Bull. Physiopathol. Respir.,  7: 195-210.

MICROECOLOGY GROUP  (1977)  Dietary fibre, transit-time faecal
bacteria, steroids, and colon cancer in two Scandinavian
populations.   Lancet,  2: 207-212.

ILO  (1980)    International Classification of Radiographs of
 Penumoconiosis.  Geneva, International Labour Organisation, 48
pp (Revised Occupational Health and Safety Series, No. 22).

JACOBSEN, M.  (1972)  The basis for the new coal dust
standards.   Mining Engineer,  131: 269-279.

(1979)  Comparison of three studies of aircraft noise and
psychiatric hospital admissions conducted in the same area.
 Psychol. Med.,  9: 681-693.

KANNEL, W.B.,  (1976)  Some lessons in cardiovascular epi-
demiology from Farmingham.   Am. J. Cardiol.,  37: 269-282.

KENNAWAY, E.L.  (1944)  Cancer of the liver in the negro in
Africa and America.   Cancer Res.,  4: 571-577.

 Handbook of mutagenicity test procedures,  Amsterdam, Elsevier,
485 pp.

Industrial monitoring: a cytogenetic approach.   Ann. N.Y.
 Acad. Sci.,  269: 4-11.

WESTERHOLM, P.  (1978)  Long-term exposure to jet fuel.  II.
A cross-sectional epidemiologic investigation on occupationally
exposed industrial workers with special reference to the
nervous system.   Scand. J. Work environ. Health,  4: 19-45.

KNUDSON, R.J., BURRROWS, B., & LEBOWITZ, M.D.  (1976a)  The
maximum expiratory flow-volume curve:  Its use in the detec-
tion of ventilatory abnormalities in a population study.   Am.
 Rev. respir. Dis.,  114: 871-879.

(1976b)  The maximum expiratory flow-volume curve.  Normal
standards, variability, and effects of age.   Am. Rev. respir.
 Dis.,  113: 587-600.

(1970)  The predictive value of a postal questionnaire on
cardio-respiratory symptoms.   J. chronic. Dis.,  23: 411-421.

STANESCU, D.  (1979)  Investigations on the lung and kidney
functions in workers exposed to cadmium.   Environ. health
 Perspect.,  28: 137-145.

LEBOWITZ, M.D. & BURROWS, B.  (1976)  Comparison of
questionnaires:  The BMRC and NHLI Respiratory Questionnaires
and a new self-completion questionnaire.   Am. Rev. respir.
 Dis.,  113: 627-635.

LEBOWITZ, M.D.  (1981)  Respiratory indicators.   Environ.
 Res.,  25: 225-235.

LECK, I.  (1978)  Teratogenic risks of disease and therapy.
 Contr. Epidem. Biostat. (Basle),  1: 23-34.

(1974)  Assessment of ventilatory function in an
epidemiological study of Sydney schoolchildren.   Bull.
 Physiopathol. Respir.,  10: 635-641.

LEGATOR, M., TRUONG, L., & CONNOR, T.H. (1978)  Analysis of
body fluids including alkylation of macromolecules for
detection of mutagenic agents.  In: Holleander, A. & De
Serres, F.J., ed.   Chemical mutagens.  Principles and methods
 for their detection.  New York, Plenum Press, Vol. 5, pp. 1-23.

LENZ, W.  (1962)  Thalidomide and congenital abnormalities
(Letter).   Lancet,  1: 7219: 45-46.

KASWANDIK, N.T., & GENTILE, J.W.  (1980)  A method for
monitoring the fertility of workers.  J. occup. Med.,  32:

LIN, R.S. & KESSLER, I.I.  (1981)  A multifactorial model for
pancreatic cancer in man: epidemiologic evidence    J. Am. Med.
 Assoc.,  245: 147-52.

LINDSTRÖM, K.  (1973)  Psychological performance of workers
exposed to various solvents.   Work Environ. Health,  10:

LISHMAN, W.A.  (1978)   Organic psychiatry.  The psychological
 consequences of cerebral disorder.  Oxford, Blackwell, 996 pp.

LITVINOV, N.N. & PROKOPENKO, Ju.I.  (1981)  [Problem of
evaluating the degree of human health hazards caused by
environmental factors.]   Gig. i Sanit.,  10: 71-74 (in Russian).

LLOYD-DAVIES, T.A. (1971)   Respiratory disease in foundrymen,
London, HMSO, 73 pp.

LONGO, L.D.  (1980)  Environmental pollution and pregnancy;
Risks and uncertainties for the fetus and infant.   Am. J.
 Obstet. Gynecol.,  137: 162-73.

G.  (1981)  Coffee and cancer of the pancreas.   New England J.
 Med.,  304: 630-633.

(1976)  Capillary abnormalities in polyvinyl chloride
production workers.   J. Am. Med. Assoc.,  236 (12): 1368-1371.

(1973)  Measurement of closing volume:  Application and
limitations.   Bull. Physiopathol. Respir.,  9: 979-995.

J.F., Jr  (1975)   Atlas of cancer mortality for U.S.
 countries:  1950-1969.  Bethesda, US Department of Health,
Education and Welfare, National Institutes of Health, l03 pp.

MAU, G. & NETTER, P.  (1974)  [The effects of paternal
cigarette smoking on perinatal mortality and the frequency of
malformations.]   Dtsch Med. Wochenschr.,  99: 1113-1118 (in

MCFADDEN, E.R., Jr & LINDEN, D.A.  (1972)  A reduction in
maximum mid-expiratory flow rate.  A spirographic manifesta-
tion of small airway disease.   Am. J. Med., 52: 725-737.

MEDICAL RESEARCH COUNCIL  (1966)   Questionnaire on respiratory
 symptoms and instructions for its use (1966),  London, Medical
Research Council, 30 pp.

MEDICAL RESEARCH COUNCIL  (1976)   Questionnaire on respiratory
 symptoms and instructions for its use (1976),  London, Medical
Research Council, 6 pp.

MEDICAL RESEARCH COUNCIL'S Committee on the Aetiology of
Chronic Bronchitis  (1960)  Standardised questionnaire on
respiratory symptoms.   Br. med. J.,  2: 1665.

MELEHINA, V.P. & BUSTUEVA, K.A.  (1979)  [The informativeness
of individual indices of physical development in studies
concerned with the health effects of air pollution in
children.]   Gig. i Sanit.,  9: 8-l0 (in Russian).

(1977)  Diagnostic value of the oral pancreatic function test
(PFT).   Gut,  19: A979-A980.

MITELMAN, F., BRANDT, L., & NILSSON, P.G.  (1978)  Relation
among occupational exposure to potential mutagenic/carcino-
genic agents, clinical findings, and bone marrow chromosomes
in acute nonlymphocytic leukaemia.   Blood,  52: l229-l237.

MORGAN, D.C., PASQUAL, R.S.H., & ASHFORD, J.R.  (1964)
Seasonal variations in the measurement of ventilatory capacity
and in the answers of working coal miners to a respiratory
symptoms questionnaire.   Br. J. prev. soc. Med.,  18: 88-97.

(1958)  Inquiry into diagnostic evidence supporting medical
certification of death.   Am. J. public Health,  48: 1376-87.

MORRELL, D.C.  (1972)  Symptom interpretation in general
practice.   J.R. Coll. gen. Pract.,  22: 297-309.

MOSER, C.A. & KALTON, G.  (1971)   Survey methods in social
 investigation.  London, Heinemann, 549 pp.

WIGGINS, H.S., & WALLER, S.L.  (1973)  The Lundh test in the
diagnosis of pancreatic disease:  A review of five years' ex-
perience.   Gut,  14: 835-841.

MUIR, C.S.  (1975)  International variation in high-risk
populations.  In:  Fraumeni, J.F., ed.   Persons at high risk
 for cancer:  An approach to cancer etiology and control.  New
York, Academic Press. pp. 293-305.

MUIR, C.S.  (1982)  The pathologist's role in cancer epidemio-
logy.  In:  Grundmann, E., ed.  Cancer campaign, Vol. 6, Cancer
 epidemiology,  New York, Gustav Fischer Verlag, pp. 259-273.

(1976)  Feasibility of monitoring populations to detect
environmental factors,  In:  Rosenfield, C. & Davis, N., ed.
 Environmental pollution and carcinogenic risk.  IARC
Scientific Publications No.13: INSERM Symposia Series, 52:
Paris, Institut national de la Santé et de la Recherche
médicale, pp. 279-293.

MULCAHY, M.T.  (1978)  Another case of 9P-syndrome.   Ann.
 Genet.,  2l: 47-49.

MUNRO, J.E. & RATOFF, L.  (l973)  The accuracy of general
practice records.   J. R. Coll. gen. Pract.,  23: 82l-826.

NAGAYO, T. & YOKOYAMA, H.  (1974)  Cancer of gastrointestinal
tract.  Early phases and diagnostic features.   J. Am. Med.
 Assoc.,  228: 888.

NEEL, J.V., KATO, H., & SCHULL, W.L.  (1974)  Mortality in the
children of atomic bomb survivors and controls.   Genetics,  76:

NEVIN, N. C. & MERRETT, J.D.  (1975)  Potato avoidance during
pregnancy in women with a previous infant with either
anencephaly and/or spina bifida.   Br. J. prev. soc. Med.,  29:

S., & MATSUDA, H.  (1980)  Renal dysfunctions of inhabitants
in a cadmium-polluted area.   Environ. Res.,  23: 13-23.

 studies, new developments.  Studies on Medical and Population
 Subjects No.25: London, HMSO, 13 pp.

 Occupational mortality.  The Registrar General's Decennial
 Supplement for England and Wales, 1970-1972.  London, Her
Majesty's Stationery Office, 224 pp (Series DS No.1).

PELMEAR, P.L., TAYLOR, W., & PEARSON, J.C.G.  (1975)  Clinical
objective tests for vibration white finger.  In: Taylor, W. &
Pelmear, P.L., ed.   Vibration white finger in industry,
London, Academic Press, pp.53-81.

PEPYS, J.  (1968)  Immunological mechanisms in allergic
diseases of the lungs,  J. clin. Pathol.,  2l: 127-131.

PHILLIPS, R.L.  (1975)  Role of life-style and dietary habits
in risk of cancer among Seventh-Day Adventists.   Cancer Res.,
35: 3513-3522.

PILAWSKA, H.  (1979)  Physical development of children in
Szizecin, Poland.  In:  Holland, W., Ibsen, J., & Kostrzewski,
J., ed.   Measurement of levels of health,  Copenhagen, World
Health Organization Regional Office for Europe, pp. 373-379
(European Series, No. 7).

PISCATOR, M.  (l982)  Cadmium and cancer of the prostate.
In:   Cadmium 8l, Proceedings of the Third International
 Cadmium Conference, Miami, Februry l98l,  London, Cadmium
Association, pp. l35-l37.

POOLING PROJECT RESEARCH GROUP  (1978)  Relationship of blood
pressure, serum cholesterol, smoking habit, relative weight
and ECG abnormalities to incidence of major coronary events:
Final report of the Pooling Project.   J. chron. Dis.,  31:

G.M., & ADAMS, W.G.F.  (1978)  Chromosomal analysis in vinyl
chloride-exposed workers.   Mutat. Res.,  57: 325-334.

RABER, A.  (1973)  The incidence of impaired hearing in rela-
tion to years of exposure and continuous sound level (pre-
liminary analysis of 26,179 cases).   Proceedings of an Inter-
 national Congress on Noise,  Dubrovnik, pp. 115-138.

REGER, R.B., BUTCHER, D.F., & MORGAN, W.K.C. (1973)  Assessing
changes in the pneumoconioses using serial radiographs.   Am.
 J. Epidemiol.,  98: 243-254.

(1966)  Studies of disease among migrants and native
populations, Great Britain, Norway and the United Stated.
III.  Prevalence of cardiorespiratory symptoms among migrants
and native born in the United States.   Natl Cancer Inst. Mon.,
19: 321-346.

ROBERTS, D.V.  (1979)  A longitudinal electromyographic study
of six men occupationally exposed to organophosphorus
compounds.   Int. Arch. occup. environ. Health,  38: 221-229.

(1981)  Environmental exposure to cadmium and renal function
of aged women in three areas of Belgium.   Environ. Res.,  24:

ROSE, G.A.  (1962)  The diagnosis of ischaemic heart pain and
intermittent claudication in field surveys.   Bull. WHO,  27:

ROSE, G.A.  (1965)  Ischaemic heart disease chest pain.   Mill-
 bank Mem. Fund. Q.,  43:  32-39.

ROSE, G.A.  (1968)  Variability of angina.  Some implications
for epidemiology.   Br. J. prev. soc. Med.,  22: 12-15.

ROSE, G.A.  (1971)  Predicting coronary heart disease from
minor symptoms and electrocardiographic findings.   Br. J.
 prev. soc. Med.,  25: 94-97.

ROSE, G.A. & BLACKBURN, H. (l968)   Cardiovascular survey
 methods.  Geneva, World Health Organization, l88 pp (Monograph
Series No. 56).

(1982)   Cardiovascular survey methods.  2nd ed., Geneva, World
Health Organization, l78 pp (Monograph Series, No. 56).

ROSE, G.A., HOLLAND, W.W., & CROWLEY, E.A.  (1964)  A
spygamomanometer for epidemiologists.   Lancet,  1: 296.

ROSE, G.A., MCCARTNEY, T., & REID, D.D. (1977)  Self
administration of a questionnaire on chest pain and inter-
mittent claudication.   Br. J. prev. soc. Med.,  32: 42-48.

ed.  (1978)   Malformations congenitales.  Risques perinatals,
Paris, INSERM, 491 pp.

RUSSE, O.A. & GERHARDT, J.J. (1975)   International S.F.T.R.
 method of measuring and recording joint motion.  Berne, Hans
Huber, 8l pp.

RUSSEL, R.I.  (1978)   Clinics in gastroenterology.  Vol. 7,
Investigative tests and techniques, pp 1-552, New York,

SANDALA, M.G. & ZVINJACKOVSKIJ, Y.I.  (1981)  [Identification
of the role of separate factors in the complex exposure of
human population to environmental factors.]   Gig. i Sanit.,
pp. 4-6 (in Russian).

SARIC, M. & HRUSTIC, O.  (1975)  Exposure to airborne manga-
nese and arterial blood pressure.   Environ. Res.,  10: 314-318.

SARIC, M., MARKICEVIC, A., & HRUSTIC, O. (1977a)  Occupational
exposure to manganese.  Br. J. ind. Med.,  34: 114-116.

SARIC, M., OFNER, E., & HOLETIC, A.  (1977b)  Acute respira-
tory diseases in a manganese contaminated area.  Proceedings
of the International Conference on Heavy Metals in Environ-
ment, Toronto 1975, Institute of Environmental Studies,
University of Toronto,  Symp. Proc.,  III: 389-398.

SAXEN, L., KLEMETTI, A., & HARO, A.S.  (1974)  A matched-pair
register for studies of selected congenital defects.   Am. J.
 Epidemiol.,  100: 297-306.

SCHIFF, L.  (1975)   Disease of the liver,  4th ed., pp 1-1461.
Philadelphia, USA, J.B. Lippincott Co.

SCHWARTZ, M.K. & HILL, P.  (1972)  Problems in the interpre-
tation of serum cholesterol values.   Prev. Med.,  1: 167-177.

SEGI, M.  (1978)   Age-adjusted rates for cancer for selected
 sites (A-classification) in 52 countries in 1973,  Segi
Institute of Cancer Epidemiology, Nagoya, Japan, 30 pp.

SEGI, M. & KURIHARA, M. (1972)   Cancer mortality for selected
 sites in 24 countries.  No.6 (1966-1967),  Segi Institute of
Cancer Epidemiology, Nagoya, Japan, l37 pp.

SELIKOFF, I.J. & ANDERSON, H.A.  (1979)   A survey of the
 general population of Michigan for health effects of poly-
 brominated biphenyl exposure.  Report to the Michigan Depart-
ment of Public Health.  Environmental Sciences Laboratory, Mt.
Sinai School of Medicine, N.Y. 264 pages.

SHARP, J.T., PAUL, O., & LEPPER, M.H.  (1965)  Prevalence of
chronic bronchitis in an American male urban industrial
population.   Am. Rev. respir. Dis.,  91: 510-520.

SHEPHERD, M.  (1974)  Pollution and mental health, with
particular reference to the problem of noise.  Psychiatr.
 Clin.,  7: 226-236.

SHERLOCK, S.  (1975)   Diseases of the liver.  5th ed., Oxford,
Blackwell, pp 1-821.

BENSON, F.B.  (1970)  The Chattanooga School Children Study:
Effects of community exposure to nitrogen dioxide, I. Methods,
description of pollutant exposure, and results of ventilatory
function testing,  J. Air. Pollut. Control Assoc., 20:
539-545.  II.  Incidence of acute respiratory illness.   J.
 Air. Pollut. Control Assoc.,  20: 582-588.

SIDDIQUI, A.H.  (1955)  Endemic fluorosis.   Br. med. J.,  2:

SIDORENKO, G.I.  (1978)  [Modern problems of environmental
health science.]   Gig. i Sanit.,  10: 9-15 (in Russian).

SINGH, A. & JOLLY, S.S.  (1962)  Endemic fluorosis.  Ind. J.
 med. Res.,  50: 387.

SINGH, A., JOLLY, S.S., & BANSAL, C.C.  (1961)  Skeletal
fluorosis and its neurological complications.   Lancet,  1: 197.

SLEISENGER, M.H. & FORDTRAN, J.S.  (1978)   Gastrointestinal
 disease.  2nd ed., pp 1-1977, Philadelphia, USA, W.B. Saunders.

SOBELS, F.H.  (1977)  Some problems associated with the
testing for environmental mutagens and a perspective for
studies in "Comparative Mutagenesis".   Mutat. Res.,  46:

STRAUSS, G.H. & ALBERTINI, R.J.  (1977)  6-thioguanine-
resistant lymphocytes in human peripheral blood.  In: Scott,
D., Bridges, B., & Sobels, F.H., ed.   Progress in genetic
 toxicology,  Amsterdam, Elsevier/North-Holland, pp. 327-334.

STUMPHIUS, J.  (1971)  Epidemiology of mesothelioma on
Walchern Island.   Br. J. ind. Med.,  28: 59-66.

TAGER, I., SPEIZER, F.E., ROSUER, B., & PRAIG, G.  (1976)  A
comparison between the three largest and three last of five
forced expiratory manoeuvres in a population study.   Am. Rev.
 respir. Dis.,  114: 1201-1203.

TITMUSS, R.M.  (1968)   Commitment to welfare.  London, Allen &
Unwin, 272 pp.

TSUCHIYA, K., ed.  (1978)   Cadmium studies in Japan: A review.
Amsterdam Elsevier/North Holland, 376 pp.

TYROLER, H.A.  (1977)  The Detroit Project Studies of Blood
Pressure.  A prologue and review of related studies and
epidemiological issues.   J. chronic. Dis.,  30: 613-624.

UICC (1970)   Cancer incidence in five continents,  Vol. II.
Berlin, Springer-Verlag, 388 pp.

 and operation of the health and nutrition examination survey.
 United States 1971-1973  (DHEW, PHS Publ. No. (HSM) 73-1310,
Series l, No.10a, 1973, and No. 10b, 1973).

 Ophthalmic examination protocol:  B. Disease definition.
 National Center for Disease Statistics, DHEW, PHS, Revised
 l972.  Washington DC, US Government Printing Office.

VAANANEN, I.  (1970)  The role of the Medical record in hospi-
tal planning systems.  In: Anderson, J. & Forsythe, J.M., ed.
 Information processing of medical records,  Amsterdam, North
Holland, pp. l60-168.

VAN DER LAAN, G.  (1980)  Chronic glomerulonephritis and
organic solvents, a case control study.   Int. Arch. occup.
 environ. Health,  47: 1-8.

VAN DER LENDE, R.  (1969)   Epidemiology of chronic non-speci-
 fic lung disease (chronic brochitis),  Thesis, University of
Groningen, Assen, Van Gorcum.

G.J., DE VRIES, K., & ORIE, N.G.M.  (1973)  Epidemiological
investigations in the Netherlands into the influence of
smoking and atmospheric pollution on respiratory symptoms and
lung function disturbances.   Pneumologie,  149: 119-126.

VASKEVIC, N.N.  (1978)  [Electroesthesiometry in vibration
disease.]   Gig. Trud. Prof. Zab.,  8: 52-54 (in Russian).

VICZIAN, M.  (1969)  [Results of the examination of the sperm
of cigarette smokers.]   Zschr. Haut-Geschl - Krck., Berlin,
44: l83-l87 (in German).

(1982)   Cancer incidence in five continents,  Vol. IV, Lyons,
International Agency for Research on Cancer, 8ll pp, (IARC
Scientific Publications No.42).

WING, J.K., COOPER, J.E., & SARTORIUS, N.  (1974)   Measurement
 and classification of psychiatric symptoms,  Cambridge,
University Press, 233 pp.

WOOD, P.H.N. (1972) Radiology in the diagnosis of arthritis
and rehumatism.  Trans. Soc. Occ. Med.,  22: 69-73.

WHO  (1965)   Trends in the study of morbidity and mortality.
Geneva, World Health Organization, l96 pp (Public Health
Papers, No. 27).

WHO  (1974)   International Pilot Study of Schizophrenia,
Geneva, World Health Organization, Vol. l, 427 pp (WHO Offset
Publication No. 2).

WHO (l976)  International Classification of Diseases for
 Oncology (ICD-O).  Geneva, World Health Organization, l3l pp.

WHO  (1977)   Manual of the International Statistical Classifi-
 cation of Diseases, Injuries, and Causes of Death,  (Ninth
Revision)   Geneva, World Health Organization, Volumes I & II.
779 and 659 pp.

WORLD HEALTH ORGANIZATION  (1979a)   Schizophrenia.  An inter-
 national follow-up study,  Chichester, Wiley 436 pp.

WHO  (1979b)  Environmental Health Criteria 14,  Ultraviolet
 radiation,  Geneva, World Health Organization, pp. 76-78.

WYNDER, E.L., LEMON, F.R., & BROSS, I.J.  (1959)  Cancer and
coronary artery disease among Seventh-Day Adventists.   Cancer,
12: 1016-1028.

(1973)  Epidemiology of cancer of the pancreas.   J. Natl
 Cancer Inst.,  50: 645-67.

WYROBECK, A.J. & BRUCE, W.R.  (1978)  The induction of sperm-
shaped abnormalities in mice and humans.  In: Hollaender, A. &
de Serres, F.J., ed.   Chemical mutagens.  Principles and
 methods for their detection.  New York, Plenum Press, Vol. 5,
pp. 257-285.

YAGAMUCHI, A., YOSHIMURA, T., & KURATSUNE, M.  (1971)  [A sur-
vey on pregnant women having consumed rice oil contaminated
with chlorobiphenyls and their babies.]   Fukuoka Acta Med.,
62: 112-117 (in Japanese).

YOSHIMURA, T.  (1971)  [Epidemiological analysis of "Yosho"
patients with special reference to sex, age, clinical grades
and oil consumption.]   Fukuoka Acta Med.,  62: 109-116 (in

ZEINER-HENRIKSEN, T.  (1972)  The repeatability at interview
of symptoms of angina and possible infarction.   J. chronic.
 Dis.,  25: 407-414.

ZEINER-HENRIKSEN, T.  (1976)  Six year mortality related to
cardio-respiratory symptoms and environmental risk factors in
a sample of the Norwegian population.   J. chronic. Dis.,  29:


5.1.  Introduction

    The plan, organization, and conduct of epidemiological studies 
on health effects of environmental pollution depend on the 
objectives and types of studies.  In geographical comparison 
studies, mortality and morbidity rates are compared for areas with 
different environmental risks (e.g., pollution levels).  Existing 
data may be used in this type of study, as in case-control studies 
(Chapter 2).  They are useful in initial risk assessment.  However, 
many studies, in which the long-term effects of environmental risks 
are examined, are prospective (cohort) studies (section 2.6).  This 
chapter concentrates more on prospective studies and various 
examples are provided.  These types of studies would require the 
cooperation of governmental authorities and institutions, some 
professional organizations such as a local medical association, and 
the populations concerned. 

5.2.  Study Protocol

    The study protocol is a formal document prepared by the leader 
of the study team, who is usually an epidemiologist, in consultation 
with the members of his team and with any outside experts who can 
provide pertinent information and advice.  Preparation of the 
protocol serves several purposes; (a) it helps the investigator to 
focus on the critical issues of the proposed study; (b) it 
delineates objectives, hypotheses, study design, study populations, 
methods of measurement, ethical and legal matters involved, methods 
for the analysis of data, and expected results; and (c) it is often 
used as a prospectus to be presented to funding agencies. 

    The study protocol should also include a detailed description 
of all activities to be performed during the preparatory phase, the 
pilot study, and the main study, in order to achieve the established 
study objectives.  Thus, the points of protocols should include: 

    -   subject;

    -   objectives;

    -   background, past work;

    -   detailed time-table of the whole study including time
        schedules of each phase of the study;

    -   methods for measurement of exposure and effects
        including specifications of measurements to be
        performed with the indication of place and time of
        measurements, questionnaires, recording forms and

    -   characteristics and size of study populations;

    -   methods for selection of study samples;

    -   list of all anticipated activities and specifications
        (including post-descriptions) of all members of the
        study team; plan of recruitment and training of the
        field workers; plan of testing of instruments and

    -   details of the required resources including premises,
        equipment, materials, administrative services, etc.;

    -   plan of arrangements with the local authorities as
        well as other relevant organizations such as the
        local medical association.

    The decision concerning the conduct of a study should be 
preceded by a thorough review of the relevant literature and an 
awareness of related studies that have been completed or are in 
progress (Zvinjackovskij et al., 1981). 

5.2.1.  Description of problems and hypothesis formulation

    The first task in preparing a protocol is to describe the 
problem to be examined, to describe previous studies of the 
problem, and to state the object of the proposed study.  The object 
is best stated in the form of an hypothesis - i.e., of a question 
posed, the answer to which, it is hoped, will shed light on the 
etiology, prevention, or treatment of the disease problem under 

    The second task is to ensure that the study protocol indicates 
the generic type of epidemiological study planned (Chapter 2) and 
describes the details of the study design. The investigator must 
weigh the advantages and disadvantages of different designs 
applicable to a particular study (Merkov, 1979; Litvinov & 
Prokopenko, 1981). 

5.2.2. Description of methods

    The protocol should describe in full detail the type and size 
of study samples and the methods for the collection of data, 
including the sampling scheme, questionnaires and instruments to be 
used, procedures for measurements of environmental exposure and 
effects, and methods for the laboratory analysis of specimens 
obtained.  Procedures for quality assurance checks should also be 
mentioned.  An effort must be made to specify and standardize the 
methods to be used.  It would be useful, when relevant, to indicate 
the influence that the pilot study experience has had on the 
inclusion or modification of a particular procedure in the final 
study plan.  It is also useful, for planning purposes, to indicate 
the length of time that will be required for the collection of each 
set of data on each subject. 

    The measuring instruments required should be known when the 
study design and variables to be measured are decided.  The testing 

of the instruments is essential and instructions should indicate 
any difficulties in use and inadequacies in the precision and 
accuracy of measurements. 

    Questionnaires should be employed as frequently as the study 
objectives warrant.  They provide (subjective) perceptions of 
health status by the study subjects.  Designing questionnaires is 
more difficult than those without experience may imagine and the 
value of testing of the questionnaire before starting a study 
cannot be overstressed.  An explanation should be provided as to 
why each proposed question is to be asked.  Special arrangements 
must be made, when studying illiterate subjects. 

5.2.3.  Evaluation of institutional-based data sources

    In some epidemiological studies, institutional-based data may 
be used, routinely collected in hospitals, outpatient settings, 
other health departments, and in environmental departments.  In 
such circumstances, a pretext is necessary, consisting of visits to 
the institution to see whether the information is recorded in such 
a way that it will be possible to gather data in accordance with 
previously-established study objectives.  Often, routine health 
data, as with routine environmental monitoring data, are of only 
limited use for epidemiological studies (sections 2.3 and 2.1l).  
On the other hand, they are often useful for generating hypotheses 
(Litvinov, 1978). 

5.2.4.  Analysis and reporting of data

    The methods of computerizing the data base, where possible, 
should be determined prior to the study.  Thus, plans must be made 
as to how the data are to be put into the computer, how it is to be 
edited and checked on the computer, and how data reports are to be 
generated from the computer. 

    The most useful procedure is to make specific plans at the 
beginning of the study for carrying out the final tabulations.  The 
tabulations flow directly from the hypotheses and from the 
selection of a particular type of study design.  It may be 
desirable to construct "dummy" tables and figures, fully labelled, 
which will portray the final results.  The methods of analysis and 
presentation can be tested in the pilot study, if one is performed. 

    The protocol should state specifically the format and manner in 
which results will be reported publicly as well as to individual 
participants.  It should state also the clearances to be sought, 
before release of the data (sections and 6.5.1).  If any 
adverse effects have been found in the study participants, these 
conditions should be referred to their physicians for further 
clinical observation, examination, or treatment.  Agreement on 
these arrangements must be reached with the clinicians in advance 
of the study, as far as this is feasible. 

5.2.5.  Resources required

    It is essential that detailed estimates of the personnel, 
equipment, and financial resources required be presented in the 
protocol.  Such information is required for internal planning as 
well as for presentation to funding agencies.  Preparation of a 
detailed budget for a large-scale study requires a good deal of 
skill and experience.  There is no worse eventuality in a study 
than to exhaust allocated funds, three-quarters of the way through 
the project or, because of inadequate planning, to have the project 
last twice as long as anticipated.  The importance of a limited, 
but full-dress pilot study (where appropriate) as a means of 
testing the estimated costs of a proposed study will be of great 
benefit in helping to refine those estimates. 

5.2.6.  Studies in developing countries

    In 1979, health workers from 7 developing countriesa from 5 WHO 
regions exchanged their national experiences on the need and 
feasibility of using simple epidemiological techniques at the 
periphery of the health system (WHO, 1982).  There was a consensus 
that the most innovative and useful aspect of this approach was the 
development of appropriate epidemiological thinking by the 
community health workers themselves. 

    The general epidemiological thinking suggested by them is shown 
in Fig. 5.1 and the use of epidemiology at the periphery of health 
care is shown in Fig. 5.2. 

    It is expected that the use of appropriate epidemiology would 
enable workers at the periphery to attain a degree of self-reliance 
in the guidance of their own work by their own scientific 
interpretation of the local reality.  This would help them to give  
better support to the community for its informed participation, and 
to manage more effectively the local health services. 

    The conclusions drawn from the epidemiological interpretation  
of data by the peripheral health workers themselves may be used,   
first, to guide their practical day-to-day activities in trying to 
solve local health problems and health care demands.  Second, the  
conclusions may be presented in such a way that they can be        
conveyed directly to community members and local government        
officials.  Such presentations will also be useful in communicating
with other levels of the health services and with other 
governmental sectors that operate at the periphery, such as 
education, agriculture, community development, and public works.              

a   Burma, Botswana, Ecuador, the Islamic Republic of Iran,              
    Malaysia, Niger, and Thailand.                                       


5.3.  Ethical and Legal Considerations        
    Social and legal requirements vary from country to country.  In
recent years, they have become an increasing issue of concern.     
Many academic and research institutions and government agencies use
independent bodies to review research proposals on human subjects  
with regard to the ethics of the work proposed.  In a number of    
countries, the law protects the rights of individuals to privacy   
and requires "informed consent" by study participants in medical   
research.  On the other hand, requirements may be less strict in 
some countries.  In all  cases, where scientific research involving 
human subjects is concerned, the ethical codes developed 
internationally, such as those of the Council for International 
Organizations of Medical Sciences (1982) should be taken into 
    Informed consent is a procedure whereby each potential 
participant must be carefully informed as to the overall nature of 
the study and all its component procedures, must not be pressured 
in any way to participate, must be given every opportunity to ask 
questions about the study, and must be given the opportunity to 
withdraw, at any point, without prejudice.  A written consent form, 
though not an absolute requirement, has become a vehicle for 
conveying the information about a study to each participant.  The 
information on the form may be read aloud to illiterate 
participants, and may be printed in several languages, if the study 
population is multilingual.  Where relevant, the community and the 
authorities should also give their consent to the study. 

    Scientists have a social responsibility to provide benefits to 
the communities and to people in general.  The study should be 
justifiable from the point of view of the community involved and 
their needs.  Benefits of the study to the participants or to 
society in general should far outweigh the risks.  Benefit/risk 
determinations should include sociocultural considerations, such as 
traditions, as well as considerations of the relevance and 
importance of the research. 

    In conducting a study, the technique used should not produce 
any harmful effects on the subjects.  In addition to ensuring the 
safety of the study subjects, the team leader is responsible also 
for the safety of the study team.  The members of the team should 
be protected from, or insured against, legal action, when doing 
their work. 

    Each participant should also be informed in detail of the 
individual study results and their interpretation.  These 
individual results must be held strictly in confidence and cannot 
be released, even to a family physician, except on the 
authorization of the participant.  The participant should have the 
right to be informed of any adverse medical conditions that are 
discovered in the course of the study.  Sometimes, the individual 
patient or subject may benefit directly by the detection of a 
previously undiagnosed disease or susceptibility.  The subject may 
also be reassured, if no abnormality is found.  On the other hand, 
there may be negative effects from informing people that they have 
diseases that cannot be treated effectively and also, where no 
abnormality has been found, the transient and limited value of a 
negative examination may not be appreciated by a patient.  These 
factors have been taken into account when screening for diseases.  
In some cases, job or insurance opportunities could be denied to 
subjects in whom abnormalities are found and reported.  Even if the 
research procedures are beneficial, on the average, some 
individuals may lose more than they gain in terms of peace of mind 
or physical or psychological discomfort.  Furthermore, even if the 
investigator has exercised every caution to protect the subjects 
and has informed them honestly of all possible risks and benefits, 
the ethical concerns are not over; they pervade all phases of the 
study from its design to the publication of results and include 
matters of scientific honesty as well as humanity. 

5.3.1.  Medical confidentialitya

    One of the major difficulties in epidemiological research is 
the problem of confidentiality.  In the more usual use of this 
term, it is concerned with the identification of individual 
patients and the disclosure of medical information to other 
individuals about these patients.  In many clinical investigations, 
the patient's permission for the disclosure of this information can 
easily be discussed with the patient.  However, in some 
epidemiological studies it may be desirable to look at the case 
notes of large groups of patients, but no direct contact with the 
individual patients is made.  In these circumstances it may be 
difficult, and sometimes impossible, to contact the individual 
patients to seek their permission.  To do this would also increase 
very considerably the cost and complexity of the study, especially 
if large numbers of individuals were involved.  The response rate 
may then be much lower through many individuals not being contacted.  
The question arises as to whether it is sufficient to obtain the
agreement of the appropriate hospital doctors for the use of 
this information, if other reasonable safeguards are arranged.  
Obviously, the type of information that is to be extracted from 
the patients' case notes is relevant, but, even if relatively 
non-controversial items are being examined (e.g., blood pressure 
or haemoglobin level), those extracting the data might see other
more sensitive information (e.g., psychiatric history, tests for 

a   Based on the contribution from Professor W. E. Waters,
    Southampton General Hospital, England.

veneral disease).  There are further questions regarding reasonable  
safeguards.  How many individuals could have access to this 
information?  Under what conditions will this information be 
stored?  For example, will it be stored on a computer and, if on 
paper, will all the papers be kept in locked filing cabinets at all 

    However, confidentiality of the sort of information that the 
epidemiologist may use may also involve other units than individual 
patients.  For example, it may sometimes be necessary to avoid 
precise identification of small groups of individuals, such as 
those who live in a defined area or certain minorities.  It may 
also be necessary to protect groups of doctors and nurses, who are 
involved in the care of these patients, and sometimes even of 
medical institutions or the region served by them, depending on the 
particular results of the study. 

    Problems about confidentiality are sometimes very difficult and 
it has to be accepted that the needs of society as a whole are 
sometimes in conflict with the individual needs of their members.  
An excessive concern about confidentiality may sometimes prevent 
the use of some clinical information on individual patients, even 
where the identities of patients and doctors have been removed, 
because the information was originally collected for a different 
purpose and explicit consent for the particular study was not 

    An alternative view of this particular problem is that of the 
doctor who feels that it is justifiable to obtain the use, in a 
non-identifiable way, of such information, if there are reasonable 
safeguards during the investigation and if the information may be 
used for the common good. 

    Many doctors may take a view somewhere between these two 
statements.  Who should decide for any particular proposal?  
Perhaps there should be more explicit recognition that, under some 
circumstances, some information can be used for the common good 
even without the specific approval of each individual.  It should 
be remembered that such studies may often involve hundreds, if not 
thousands, of individuals and the difficulty of obtaining such 
permission is almost beyond the bounds of any reasonable 

    Concern about confidentiality may often extend outside the 
records of the health service.  For example, in studies of 
occupational diseases, the payrolls of various factories have 
often been used.  Such records may be kept by many firms for a long 
period of time and the information can enable a cohort study to be 
done in a retrospective way.  Obviously this payroll sampling 
frame, which is of such great value to the epidemiologist, was 
originally set up without the original employees being informed 
that it might be used, perhaps long after they were dead, for a 
study on the health effects of their particular occupation.  Yet, 
if the use of such payroll lists were restricted in any way by the 
epidemiologist, it would delay for many years, perhaps forever, 
information about the health risks of particular occupations. 

    The question of confidentiality of medical information, that is 
stored on information systems, is giving rise to more ethical 
questions and the problems increase the more complicated and 
important the information systems become.  First, there is the 
legal question as to who "owns" the medical records - the doctor, 
the patient, or the health authority?  There is also the question 
of whether the ownership applies to the paper on which the record 
is written or to the record itself.  There are the further 
difficult questions such as, who has the legal right of access and 
who has the legal right to prevent access to this information? 

    Although much medical information is now stored, it is often 
when this information is used for research or linked with other 
information about individual patients that ethical problems appear 
to rise.  There is a fear of invasion of privacy and, in many 
countries, this has become a politically sensitive issue.  The fear 
is that such information could now be linked together in precise 
terms by computers and other means, whereas, in the past, society 
was safeguarded by the reluctance of many clinicians to divulge 
this information and by the less sophisticated methods of handling 
the information. 

5.4.  Time Schedule of Study

    The total period of time provided for the preparation and 
execution of a study may be divided into three parts, i.e., the 
preparatory phase, the pilot study, and the main study. 

5.4.1.  Preparatory phase

    The preparation of a study should start with:  (a) reviewing 
available information; (b) determining the specific aims of the 
study; and (c) designing the study and developing the study 

    After the study protocol has been prepared, the practical 
logistic steps that lead to the actual conduct of the field study 
can begin.  Although there are a number of steps to be undertaken, 
it is well to bear in mind that work can be proceeding on several 
of these at the same time.  Such steps can be portrayed by means of 
a flow diagram, that can be of great aid to study organization in 
several respects:  (a) the diagram will help the team leader to 
organize and assign the myriad tasks that must be performed 
simultaneously in the preparation for, and conduct of, a study; (b) 
the diagram can serve as a combination of a calendar and a check-
list and this will enable the leader to see at a glance whether or 
not the various activities are proceeding as planned; (c) the 
preparation of the flow diagram will help the team leader to 
identify in advance potential bottlenecks and points of 
obstruction, thus, the leader will have an opportunity to adjust 
schedules, to redistribute assignments more equitably and avoid 
delays; and (d) the preparation of a flow diagram will enable the 
leader to identify the rate-limiting sequence of events that 
determines the overall timing of the study, when all tasks are 
performed at maximum efficiency. 

    The practical steps to be undertaken in the preparatory phase 

    -   negotiations with local authorities, community
        leaders, local professional associations, etc., as
    -   advance contact with study subjects;
    -   recruitment and training of members of study team;
    -   pretest of questionnaires;
    -   preparation of all indispensable intructions for
        field workers and recording forms including coding
    -   testing of instruments and observers;
    -   purchase of equipment and other materials as required;
    -   renting premises for study as required; and
    -   preparation of basic computer programs for analysis
        of data to be collected.

5.4.2.  Pilot study

    The pilot study should be an effective device for judging the 
overall adequacy, feasibility, and appropriateness of the proposed 
study, and for checking the accuracy of cost and time estimates.  
While usually limited in scope to no more than 2 or 3 days' work in 
a single location, the pilot study should be a full-dress operation 
on a similar population.  An effort should be made to capture the 
tempo and spirit of the actual study.  At the conclusion, the team 
leader  must either abort the main study, if it appears to be 
irremediably impractical, or amend and adjust it as required.  It 
is important that sufficient time be allowed between the pilot 
study and the main study to allow for any adjustments.  The pilot 
study should also provide an opportunity to test the adequacy of 
training under controlled field conditions (see section 5.6.5 for 
further discussion). 

5.4.3.  Main study

    Two concepts must be central in the planning for, and conduct 
of, the main study:  (a) everything and every person involved must 
be on site at the proper time; and (b) nothing can be changed.  It 
may also be useful, on occasion, to construct an additional flow 
diagram for the main study detailing the timetable for the 
examinations of subjects as well as the collection of interview 
data and environmental information.  Attention must be given to the 
times at which the study can be carried out in terms of hours 
during the day and week when the subjects are available and the 
seasons during which field studies are possible. 

    The flow diagram is required to include the timetable for the 
analysis of the data collected, the thorough discussions on how to 
interpret and draw conclusions from the results obtained, the 
reporting of results to relevant parties (section and the 
publication of the study. 

5.5.  Composition of the Study Team

    The composition of a study team will vary with the design and 
scope of the study and with the resources available. Study teams 
may be pre-existing, for example, in public health institutions, 
newly-formed for a particular study, or a combination of the two.  
Some studies, especially preliminary studies to generate an 
hypothesis, often require only a principal investigator 
(epidemiologist).  Analytical studies of existing data may require 
the addition of one or two specialists, for example, in statistics 
and computer sciences. 

5.5.1.  Team leadership and epidemiology

    A team for environmental epidemiology studies should 
successfully combine the talents of several scientific disciplines.  
In small-scale studies, the epidemiologist must be familiar with 
these disciplines.  In general, an epidemiologist should be the 
team leader, for it is he or she who is most likely to have the 
best overall view of the project and its goals and to be at least, 
familiar with, if not actually competent in the other component 
disciplines.  The team leader is responsible for the overall 
planning and conduct of a study, for maintaining team discipline 
and, at the conclusion of the study, for the analysis, 
interpretation, and reporting of study data. 

5.5.2.  Clinical specialist

    In some studies, the performance of medical examinations or 
clinical measurements requires clinicians on the study team, even 
if the epidemiologist is medically qualified.  This is particularly 
true, when the clinical examinations to be performed are of a 
highly specialized nature.  It may be necessary to bring clinical 
specialists, as well as necessary equipment, to the field or to 
take all or some of the study subjects to clinicians in, for 
example, a hospital.  The recent development of portable, 
miniaturized equipment for many clinical examinations has made it 
easier to conduct a number of clinical tests in the field.  It is 
important to establish, at the outset, that clinical examinations 
performed in a study must be done according to a protocol that is 
standard (the same) for each examining doctor and each subject. 

5.5.3.  Statistical expertise

    Statisticians have key roles in study planning, computerization, 
and data analysis.  Even in small-scale studies it is recommended 
that statistical advice be sought.  At the planning stage, the 
statistician will work closely with the team leader in establishing 
the design; in determining procedures for the selection of study 

subjects; in designing questionnaires and other survey material for 
the collection of data in a standardized, processable manner; and, 
most importantly, in helping to formulate the study hypotheses in 
quantitative terms, to the extent that the crucial final 
tabulations can be portrayed in blank tabular form, long before the 
start of any field work.  During the study, the statistician should 
review the original data and computer files and indicate any 
omissions, inconsistencies, or errors in the data.  The 
statistician would assess the quality of the data by various 
comparisons of data from different observers and coders and would 
assist the team leader in the analysis and interpretation of the 

5.5.4.  Environmental scientists

    Specialists in environmental sampling and measurements are also 
important members of a study team in environmental epidemiology.  
Prior to the start of field work, their function is to assist the 
team leader to plan strategies for environmental monitoring and to 
develop liaisons with one or more laboratories to ensure that they 
are able to process and analyse the environmental samples to be 
collected.  In the field, the environmental specialist, aided 
perhaps by one or more assistants or technicians in larger studies, 
will have responsibility for collecting, labelling, and properly 
storing environmental samples, and will be responsible for 
maintaining the calibration of equipment and for conducting quality 
assurance procedures. 

    When the use of complex or sophisticated equipment is planned 
during a field study, it may be useful to have a specialist in the 
repair and maintenance of such equipment attached to the study 

5.5.5.  Interviewers and technicians

    In some studies, interviewers are needed to obtain questionnaire 
data and technicians are needed to perform various clinical tests.  
In occupational health studies, industrial hygienists are often 
required.  Adequately trained interviewers or technicians are often 
unavailable for a field study and it is frequently necessary to 
recruit less experienced persons and train them for specific 
duties.  A study on the characteristics of successful interviewers 
has not substantiated the idea that either inborn talent or 
inherent knowledge determines the quality of interviewing (Kahn & 
Cannell, 1965).  It seems that the success of interviewing depends 
rather on the respondents perceiving the interviewer as being one 
with whom they can communicate.  A business-like manner is needed 
as well as "social sensitivity".  These qualities may be developed 
through both training and experience.  Special interviewers may be 
necessary to interview disabled (e.g., blind or deaf) or illiterate 
subjects, or those who speak a different language. 

    The number of interviewers and technicians to be employed 
depends on the type of study, the size of the study population, the 
place where the respondents will be studied (at home, at work, or 

in a clinic), the anticipated availability of the respondents, and 
the period of time scheduled for the field work.  Because of 
unavoidable observer bias, which may affect the study results, it 
seems advisable either to employ only one observer or to have 
several whose interobserver differences can be checked and 
controlled.  In addition, random allocation of interviewers/
technicians to respondents makes it possible to diminish and assess 
interviewer bias. 

    In a large-scale field study, senior interviewers and 
technicians may be necessary in order to supervise and control the 
work of the other interviewers and technicians, who will be 
organized in several groups, each of which should be headed by a 
supervisor.  Supervisors should be experienced and responsible 
staff.  They must ensure that questionnaires have been returned and 
completed, and that tests have been performed correctly. 

5.5.6.  Support staff

    There are other professionals who may be helpful in the 
development of questionnaires (e.g., sociologists, psychologists), 
tests (e.g., physiologists, toxicologists, biochemists), and flow 
diagrams (managerial experts).  When a sampling frame is available, 
it will usually be necessary to employ part-time workers 
responsible for archiving the existing lists or files.  Under the 
direction of an experienced statistician and coworkers and provided 
with minute written instructions, such people are usually able to 
select and to list appropriate sampling units from the sampling 

    When usable sampling frames are unavailable and the area 
concerned has to be sampled, a group of people need to be employed 
who will be able to list all dwellings.  These people must be fully 
acquainted with their duties and provided with detailed written 
instructions as well as with maps of areas or blocks of dwellings.  
Their work must be very carefully supervised, because the 
completeness and validity of the sampling frame depends on this 

    People such as nurses, nursing aides, social workers, and 
community volunteers, who are assigned to, or live in, the area 
where a study is to be conducted, may often possess important 
information about the health status, the social mores, and other 
important aspects of life in the community - information that will 
aid greatly in planning a study.  They would also be very effective 
recruiters of potential subjects into a study. 

    When a large-scale field study is being carried out, it is 
necessary to provide administrative offices for field workers.  In 
many studies, even when self-coding questionnaires and other 
recording forms are used, it is necessary to check the answers to 
questions and possibly to do supplementary coding, using full- or 
part-time people.  In smaller studies, these clerks may be the only 
support staff necessary.  Other support staff for large-scale 
studies may include secretaries, coders, laboratory technicians, 
dieticians, receptionists, and hygienists. 

5.5.7.  Special considerations for developing countries

    Epidemiological studies may be conducted by primary health care 
workers, under the supervision of an experienced epidemiologist, 
especially in developing countries (WHO, 1982).  Such workers may 
include the following types singly or together:  welfare workers, 
nurses, auxiliary nurses, midwives, health visitors, family 
planning educators, sanitary inspectors, and sanitarians.  They 
frequently constitute the essential components of primary health 
care and may have valuable knowledge of the local health situation.  
They will work on the day-to-day collection and reporting of 
epidemiological data.  Basic epidemiological knowledge and skills 
are required by these primary health care workers. 

5.5.8.  Example: Study teams of Itai-Itai disease and chronic cadmium 

    An outbreak of a disease, characterized by osteomalacia with 
severe pain, occurred in a rural area in the north-west part of 
Japan.  Because of the characteristic pain, the disease was named 
the Itai-itai (ouch-ouch) disease by a local clinician who first 
reported it.  The clinician and his associates claimed that cadmium 
in the rice eaten by the patients was responsible for the disease.  
Cadmium was considered to have been discharged from a zinc mine 
that was situated upstream, thus contaminating rice fields 

    Increasing social concern resulted in the organization of a 
study team on the Itai-Itai disease, in 1963, with a grant from the 
national government which covered a 3-year study.  The study 
included clinical and pathological examination of the patients, an 
epidemiological survey employing the case-control method and 
surveys on levels of cadmium in the environment (rice and other 
crops, river water, well water, paddy-field soils, etc.) as well as 
on the source of cadmium.  Experimental studies with cadmium were 
also performed.  In January 1967, a joint report was prepared and 
it was concluded that cadmium was most strongly suspected of being 
responsible for the disease.                   

    In the mean time, a large health survey of inhabitants in    
cadmium-polluted areas throughout Japan was launched in 1965 by a
study team newly organized by the Japan Public Health Association
with a grant-in-aid from the Ministry of Health and Welfare, 
because it had been ascertained that there were a number of other 
areas polluted by cadmium in addition to the endemic area of Itai-
Itai disease.  These cadmium-polluted areas were designated by the 
Ministry of Health and Welfare as "Areas Requiring Observation for 
Environmental Pollution by Cadmium" in 1969 and subsequent years.

a   Based on the contribution of Dr I. Shigematsu, Radiation              
    Effects Research Foundation, Hiroshima, Japan.                        

    When the Environment Agency of Japan was established in 1971, 
various research groups on the effects of cadmium, supported by the 
Ministry of Health and Welfare, were integrated into a comprehensive 
research team, the composition of which is illustrated in Fig. 5.3 
(Shigematsu, 1978).  This research team has undertaken the following 
studies, which have been coordinated by an epidemiologist as the 
team leader: 

      i.    experimental studies on effects of chronic cadmium

     ii.    clinical studies on renal tubular dysfunction;

    iii.    follow-up studies on Itai-Itai disease;

     iv.    pathological studies on Itai-Itai disease and chronic
            cadmium poisoning; and

      v.    mortality studies in cadmium polluted areas.


5.6.  Implementation of Study

5.6.1.  Arrangements with local authorities and study population

    The local authorities should be informed about the study 
objectives and about the details of the organizational aspects of 
the study.  Such information should convince them that the proposed 
study will be useful for improving the health of the local people 
involved and should assure them that the study methods are safe. 

    When the general population is studied, the local administrative 
authorities may be helpful in obtaining the sampling frame and the
environmental data, routinely collected in the study area.  They 
may also be helpful in the organization of a field study, for 
example, in lending premises for performing examinations on the 
subjects.  When a working population is studied, the managerial 
board (and local trade union groups, where appropriate) may 
facilitate the organization of interviewing or other examinations 
in the enterprise.  Where appropriate, meetings should be organized 
with the relevant officials, community leaders, representatives of 
relevant professional asociations, etc., in order to provide 
detailed information about the study and to obtain their 

    The way of informing the study subjects as to the nature and 
purposes of the proposed study, which is important for their active 
cooperation, will vary with the study designs.  If a high 
proportion of the population is to be asked to participate, then 
available mass media (radio, newspapers) or communitywide meetings 
would be most efficient.  If the sampling fraction is small, 
individual contact would be more appropriate, either by letter or 
home visits. 

5.6.2.  Picking samples

    Preliminary evaluation of a chosen sampling frame, that is, the 
population from which sample is to be chosen, should be performed 
by an experienced investigator during the preparatory phase.  Kish 
(1965) summarized various types of problems of sampling frames as 
follows:  an incomplete sampling frame, the appearance of clusters 
of elements as a single element in the list, the appearance of 
blanks or foreign sampling units in the list, and duplicate listing 
of some sampling units in the list.  A final evaluation of the 
sampling frame may be made during the pilot study.  If deficiencies 
in the sampling frame are judged to be small in comparison with 
other errors inherent in the study and, if it is costly to correct 
the frame, the usual practice is to disregard the problem and use 
the frame as it is (Sagen, 1970).  When the sampling frame is 
chosen, it is necessary to establish the number of sampling units 
(individuals or dwellings) to be selected and to prepare the plan 
of sample selection before the start of the field study.   Example: Sampling procedures

    The main purpose of the epidemiological study of chronic, non-
specific respiratory disease in Cracow was to estimate the 
previously unknown prevalence of this disease in the adult urban 
population in Poland (Collective Work, 1969; Sawicki, 1969a, 1977).  
As it was impossible and unjustified to examine the entire target 
population, it was decided to draw a random sample from this 
population.  The target population included the non-institutional 
population of permanent inhabitants of the City of Cracow, who were 
born between 1898 and 1949.  The main study was scheduled for 1968.  
During the preparatory phase, which started at the end of 1965, the 
available sampling frames were explored.  It appeared that existing 
voting lists were out-of-date.  However, the files of dwelling 
cards in each of the six District Councils were available.  These 
cards contained the addresses of dwellings and the names of persons 
permanently living in each dwelling. 

    It was decided to use the existing file of dwelling cards as 
the sampling frame.  The sampling units were the dwellings, and all 
permanent inhabitants born between 1898 and 1949 were the units of 
inquiry.  As the files within each district were kept separately in 
the subdistricts, into which each of the districts was divided, it 
was decided to treat these subdistricts as the strata and to select 
dwellings separately from each stratum subdistrict.  Therefore, the 
design was a staged, stratified, cluster sampling.  The sampling 
design was prepared with professional advice. 

    The first pilot study was performed in May 1965 in one of the 
districts.  Out of the total number of dwelling cards that were in 
the files of this district, 200 were selected.  The interviewers 
received addresses of the selected dwellings (street, street 
number, number of the apartment) and the names of the permanent 
inhabitants of the dwellings.  In addition, interviewers received 
the addresses of the dwelling that was next in the file after the 
one selected to the sample.  They received also the names of people 
who lived in this additional dwelling.  These additional addresses 
were given to interviewers in order to check whether there were 
additional dwellings, between the one selected and the next one on 
file, and to test completeness of the sampling frame.  Using this 
technique, known as the "half-open interval" (Sagen, 1970), it is 
possible to assess the proportion of missing elements in the 
sampling frame. 

    The experience obtained in the pilot study revealed that the 
lists of names of inhabitants placed in the dwelling cards were 
inaccurate and partly out-of-date.  Therefore, it was decided that, 
in the main study, interviewers would only be given addresses of 
the selected dwellings without the names of the inhabitants, with 
the instructions to interview all permanent residents of these 
dwellings, born between 1898 and 1949.  In case of doubt related to 
the permanent residence, persons who reported that they had slept 
in the dwelling every or nearly every night (at least four nights 
during a week) during six months preceding the interview were to be 
considered as permanent residents. 

    The check of completeness of the sampling frame revealed that 
the existing number of missing elements in the frame was 
negligible.  However, it appeared that there was a significant 
number of apartments marked with the same sequential number and the 
subsequent letters, e.g., 2a, 2b, or 16a, 16b l6c etc., as a result 
of subdivision of a building or by an addition to the building.  
Furthermore, it was found that some apartments marked with the same 
number and different letters were listed together on one dwelling 
card and some had separate cards in the file.  Taking into account 
various possible solutions and their consequences, it was finally 
decided to link in files, before the sample selection, all cards 
for apartments marked with the same number and different letters 
and to treat them as one apartment.  The interviewers were 
instructed to examine all apartments marked with the selected 
number and different letters. 

    A second small pilot study performed in December 1960 confirmed 
the usefulness and pertinence of the above procedure. 

    Analysis of the data collected during the pilot study 
determined the sample size for the main study, the estimation of 
the size of sampling error, and the assessment of the effects of 
clustering and stratification.  The applied sampling design did not 
affect the size of sampling error.  An intraclass correlation did 
not exist within clusters (dwellings).  Thus, the analysis of 
collected data, it was possible to apply simple statistical 
methods, adequate for an unbiased random sampling design. 

    Although the applied stratification did not decrease the 
sampling error, it was decided to maintain the basic design for the 
main study, because it was easy to draw the sample of addresses 
(dwellings) from each subdistrict-stratum, separately. 

    Before the main study started, the random-number tables were 
chosen.  As there were 39 subdistrict-strata in the city, 39 places 
were selected at random in these tables.  These places (numbers) 
indicated the starting points for the random selection of numbers 
of each stratum.  Detailed instructions describing the selection of 
numbers were prepared. 

    In all 39 subdistricts, the number of dwelling cards was 
calculated simultaneously.  During the calculation, each set of 50 
dwelling cards was separated with a small stick (below).  According 
to the previously established sampling size and taking into account 
the estimated average number of adult inhabitants per dwelling, on 
the basis of results obtained in the pilot study, it was decided to 
select 1930 dwellings in the whole city.  Given the total number of 
dwellings in the city, a sampling fraction was calculated.  Then, 
using this fraction, the number of dwellings to be drawn in each 
subdistrict-stratum was calculated and checked.  The appropriate 
numbers within each stratum were selected from the prepared tables 
of random numbers, according to instructions.  Correctness of the 
selection was checked and the randomly-selected numbers were 
recorded on the lists prepared separately for each subdistrict.  
These lists were transferred to the offices in each subdistrict, 

where the clerks wrote down the appropriate addresses of 
dwellings.  The small sticks inserted between batches of 50 
dwelling cards facilitated finding the sequential numbers of 
dwellings in the files.  This simple technique avoided laborious 
and time-consuming enumeration of all dwelling cards.  All these 
procedures were performed in three days.  Selection of random 
numbers was done by three persons.  The calculation of number 
of dwellings in each subdistrict and selection and listing of 
addresses according to the selected random numbers was made by 39 
clerks, under the supervision of six persons, at the subdistrict 
offices, where the files were kept.  Each of these 45 persons was 
provided with the detailed instructions. 

5.6.3.  Designing recording forms and questionnaires

    All relevant information from epidemiological studies has to be 
recorded at first on recording or questionnaire forms.  A good form 
design is essential for the adequate presentation of the data 
obtained.  The forms would be required for records of measurements 
and interviews (questionnaire), results from laboratory tests, and 
any observations to be recorded (for instance, comments by an 
observer on the reliability of data recorded). 

    The forms should be easy to complete, easy to read, and as 
short as possible.  The layout should be arranged so that gaps 
in the information recorded are conspicuous.  The ease with 
which data may be extracted for tabulation, coding, or direct 
entry into a computer should be tested in order to determine the 
final format.  It is desirable to minimize the need for recoding 
and copying information from forms completed at the survey or in 
the laboratory.  Each such operation consumes resources and 
introduces the possibility of new errors.  The size, material 
(paper or card), colour, and typographical style of forms also 
need to be considered carefully and will be affected by the 
following questions.  How are the forms to be stored:  in filing 
cabinets, boxes, or card-index cabinets?  Who will complete them:  
nurses, technicians, physicians, clerks, or the subjects being 
surveyed?  Where will they be completed:  in a clinic, a mobile 
unit, at home, or in the open-air? 

    Essential procedures for the preparation of forms include the 

(a) list all items of information that are required on the

(b) are they all relevant?  If not, delete the superfluous

(c) order the items in a sequence corresponding to the
    anticipated flow of information and prepare a first draft
    of the layout;

(d) seek comments and criticisms from others in the team,
    particularly those who will have to complete the form and
    process the data, and amend as needed;

(e) decide on size, material, colour, and typography, and
    produce a prototype;

(f) test-use the prototype in a realistic pilot-study

(g) test the ability to process data entered on the form
    (coding; checking, transcription; entry into a computer);

(h) amend as needed.

    The questionnaire should include the name and address of the 
respondent or the address of a selected dwelling, the name and 
identification number of the interviewer, a place for the 
interviewer's notes, e.g., the dates of the visits, the first and 
return ones, the period of time spent on interviewing, reasons for 
non-interviewing (refusals, no one at home, persons temporarily 
absent, impossible to establish contact with the respondent because 
of mental illness, etc.), and a list of all household members, as 

    Nevertheless, all questionnaires should be as short and as easy 
to complete as possible and should be constructed in a way that 
facilitates the checking of completeness and correctness of the 
records as well as the data processing. 

    If a follow-up study is planned, it is useful to insert the 
information about changes of residence or changes in the names of 
respondents on the forms. 

5.6.4.  Planning for control of data and computer programming

    During the course of a study, the data may be generated from a 
variety of source points (perhaps geographically-distant places) 
and over various periods of time, extending sometimes into years.  
The flow of the data from the source to the place where they are to 
be analysed and stored must be planned and controlled.  The plan 
may consist simply of a systematic list of the detailed steps that 
are required; or it might be a formal flow-chart prepared by a 
system analyst.  In a large-scale study, this may require a 
specific data control unit which would receive data from the points 
of collection and inspect them prior to their transfer to the next 
stage in their route. 

    The control procedures are primarily to ensure that no material 
is lost.  Inadvertent misplacement or destruction of manuscripts in 
a busy clinic, laboratory, or mobile unit are hazards that must not 
be ignored.  On receipt of a batch of data at the control point, 
the number of items (forms, cards, etc.) in the batch are counted.  
The type of item and the number received are noted in a receipt 
book and this information is compared with that on the data 
transfer cover note, which should have been completed at the source 
point.  Discrepancies are noted and queried immediately.  It may be 
desirable for the control point to issue a receipt to the source 

    As mentioned in section 5.4.1, it is worthwhile to start the 
preparation of relevant computer programmes before the pilot field 
study starts.  The programmes for file creation and manipulation, 
for checking errors, and for checking the consistency of 
information "within" each subject in the study, may be prepared in 
advance.  New programmes can be written or suitable programmes 
chosen from existing statistical packages.  The suitability and 
validity of the prepared programmes may be checked on a set of 
special-prepared dummy documents.  It is useful to prepare dummy 
data with intentional errors in order to find whether the prepared 
"debugging" programmes comply with the established requirements. 

5.6.5.  Training of personnel

    It is not easy to recruit, train, and maintain a staff of 
competent professional and other workers in productivity for a 
protracted period, but a corps of experienced collaborators and 
staff is the greatest asset that the team leader can have.  It is 
wise to recruit and train staff at the very beginning of a research 
project and to impress on them the importance of their remaining 
with the study until its completion, so that observer variation as 
well as training costs can be kept to a minimum. 

    The training of team staff should, preferably, be performed by 
professional training experts or, at least, experienced senior 
staff, according to a well-prepared programme.  At the beginning of 
the training, all staff should be given complete sets of 
instructions and forms to be used in the study.  The objectives and 
organization of the study, as well as the investigative methods, 
are then explained to all staff.  Training should normally be 
carried out in a group, because experience has shown that group 
training is more efficient and economical in the teaching of new 

    The interviewers play a major role in epidemiological studies 
and require extensive training.  The aims of training interviewers 
are to help them obtain an adequate knowledge of the subject 
matter, such as the objectives and organization of the study, make 
them well aware of sensitive human relationships at the interview 
and help them develop adequate interviewing techniques, for 
example, how to motivate the respondents (Kahn & Cannell, 1965).  
The interviewers should be able to explain the objectives of the 
study to the respondents and to convince them of their important 
role in the study.  Interviewers should be taught about the 
significant effects that the  interviewer's behaviour, language, 
and even attire may have on respondents.  The questionnaires to be 
used should be explained in depth. Interviewers should be provided 
with detailed written instructions.  A good example of such 
instructions is that of the Epidemiology Standardization Project of 
the American Thoracic Society (Ferris, 1978). 

    One of the specific interviewer-training methods is "role 
playing", when interviewers play alternately the role of respondent 
and interviewer, using the questionnaire that is to be used in the 
actual study.  Training of interviewers should be further conducted 

in a pilot study.  Their performance should be critically assessed 
and supplementary individual training may be performed as required. 

    Statistical analysis of the type and direction of interviewer 
error could be done in the pilot study as well as in the main 
study, if the interviewers were randomly allocated to the 
respondents (Ury, 1965; Sawicki, 1969b, 1977). 

5.6.6.  Pilot study

    It is highly desirable to conduct a pilot study (as stated in 
section 5.4.2) in order to check the adequacy of various components 
to be used in the main study, including the study protocol, the 
sample size, the method for sampling interviewers, the laboratory 
work, questionnaires, instructions for field workers, and methods 
for the statistical analysis of the data collected including the 
computer programmes. 

    Furthermore, it is useful that the questionnaires and tests may 
be subjected to a small-scale pretest before the pilot study.  This 
is especially important when a particular questionnaire or test has 
never been used before or when the questionnaire to be used in the 
main study has been translated from another language or used in a 
different sociocultural population.  Questionnaires, in other 
words, must be relevant and specific to local situations.  Never-
theless, questionnaires should remain as standardized as possible. 

    All measurement methods should be tested before conducting the 
main study to ensure that comparable results will be obtained from 
all instruments and all observers during the study.  This is 
usually done by measuring the reliability (reproducibility) of the 
measurements.  It is frequently difficult to assess the validity of 
measurements because of the lack of criteria of validity, such as 
knowledge of the "true" values, or the unavailability of specific 
reference measurement methods.  Instruments operated by observers 
(e.g., blood pressure or spirometric measurements) involve errors 
from both instruments and observers (see example in section, whereas reading X-ray films gives rise only to observer 
errors (see example in section  Assessment of variations 
in the first case requires careful statistical designs for analysis 
in order to be able to distinguish the variations attributable to 
the instruments from those of the observers.  The characteristics 
of the instruments (accuracy, precision, sensitivity, specificity) 
should be known in advance of the main study. 

    There are some other problems that may occur and that 
would produce errors.  For instance, if electrical instruments 
are used, problems from breaks in the electricity supply or 
voltage fluctuations, which may frequently occur in developing 
countries, have to be resolved.  Simpler instruments for the 
field use such as "mini" X-rays or function test instruments 
may be subject to greater errors. 

    Other preparatory activities, indispensable for the proper 
conduct of the main study, include the recruitment and training of 
any supplementary personnel that has been found necessary, 
preparation of final documentation of study procedures, purchase of 
materials such as reagents, and renting and preparation of the 
necessary premises.  Example: Testing of spirometers and assessment of
observer error

    In an epidemiological study on the long-term effects on health 
of air pollution in Poland (Rudnik et al., 1978), the ventilatory 
capacity of children was measured by means of Wright peak-flow-
meters and LODE D-53 spirometers. 

    During the preparatory phase before the pilot study, the 
measurement equipment was tested in combination with the testing 
and training of the technicians (observers). 

    Five peak-flow-meters (PFMs) were randomly labelled with 
letters, and numbers were allocated to two observers.  There were 
ten "treatments" altogether:  in the experimental design a 10x10 
Latin square was used.  Ten children participated in the 
experiment, each child performing the test ten times, in turn.  
Results of three technically satisfactory blows were recorded.  
Maximum and mean peak expiratory flow rate (PEFR) values were 
analysed by the method of analysis of variance.  The analysis 
evealed that, apart from expected biological variation between the 
children, there was also significant variation between the PFMs.  
Two flow-meters were the source of systematic error, and were 

    Similarly, two LODE D-53 spirometers were labelled with letters 
and two observers, with numbers.  In the experimental design, a 4x4 
Latin square with three replications was used.  Twelve children 
participated in the study.  Maximum results from three recordings 
were used for calculation of the forced expiratory volume in three-
quarter second (FEV0.75) and forced vital capacity (FVC) values.  
Analysis of variance of the results obtained revealed that the 
examined children were the only source of variability.  There were 
no significant differences between the spirometers or observers. 

    Using one spirometer and one peak-flow-meter, the various 
combinations of the sitting versus the standing position and use of 
a noseclip were studied.  In the experimental design an 8x8 Latin 
square with three replications was used.  Twenty-four children 
participated in the study.  Maximum values of PEFR and FVC were 
recorded and analysed.  Analysis of variance revealed that the only 
source of variation was biological variability between the 
children.  There were no significant differences between results 
obtained in a sitting or a standing position, with or without a 
noseclip, either in regard to PEFR or FVC values.   Example: Assessment of X-ray observer error

    The plan of an epidemiological study of chronic non-specific 
respiratory disease in Cracaw (Sawicki et al., 1969) included 
examination by small X-ray films (70x70mm); if lesions were 
observed, large X-ray films were also to be taken.  In the pilot 
study, the influence of observer error on the interpretation of
X-ray films was studied to establish a method of minimizing the 
influence of this error on the expected results. 

    Small X-ray films were made on 363 subjects.  For control 
purposes, large X-ray films were also made on every eighth person.  
The number of control large films was 45.  After the small films 
were read, 3l persons, in whose films changes had been noted, had 
large films made.  The small films were interpreted by three 
readers, identified by the code letters A, B, and C, who each 
inspected the films twice independently, at an interval of several 
weeks.  Before the second reading, the films were mixed and read in 
a different order from that at the first reading.  Next, the films 
were read jointly by the three readers simultaneously; they were 
aware of their previous two readings.  The large films were also 
interpreted independently by each observer, A, B, and C, and then 
by all three observers jointly.  Differences of opinion of the 
readers were decided by the fourth observer, acting as arbiter.  
For the purposes of statistical analysis, the results were divided 
into four groups:  "without changes", "changes in the cardiac 
silhouette", "tuberculous lesions" and "other changes".  If the 
films were classified as in both the 2nd and 3rd, or 3rd and 4th 
groups, they were placed in the 3rd group.  If they were classified 
in both the 2nd and 4th groups, they were included in the 2nd 

    Statistical analysis was carried out by means of two-way 
analysis of variance.  Differences between observer pairs and pairs 
of readings by the same observer were assessed by the chi-square 
test for paired variables.  Agreement between results was studied 
by calculating percentages of agreement and coefficients of 
agreement (Robinson, 1957), the latter being mainly taken into 

    The analyses demonstrated marked "interobserver" differences in 
the reading of small and large films.  The greatest discrepancy 
concerned evaluation of films with no changes and changes in the 
cardiac silhouette.  The intraobserver error was much smaller than 
the interobserver error.  After detailed statistical analyses, it 
was decided that only a single reading of each small X-ray film 
would be performed in the main study, and that observers A and C 
should be employed for this purpose. 

5.6.7.  Main study  Advance contact

    The use of a letter in advance to the subjects of the study, in 
population studies, is strongly recommended.  The letter should 

inform them about the objectives of the study, the procedures, time 
schedules, and other details as appropriate.  Depending on the type 
of study, an advance letter should motivate the subjects and 
stimulate them to meticulously complete the questionnaires, to 
cooperate with the interviewer, and to attend the medical 
examinations.  Where working with illiterate populations, a 
community meeting or personnal contact must fulfill the same 
function.  In all cases, the protection of confidentiality and 
privacy must be indicated to all those involved.  Such a letter 
should be signed by the team leader or a person, who is known to be 
reliable and trustworthy by the study subjects, and may be sent by 
mail or messenger from the study office.  Interview studies

    In the case of self-administered questionnaires, these may be 
sent together with the advance letter, or separately.  Records 
should be kept of questionnaires sent out and returned and the 
correctness of responses should be checked.  The appropriate 
procedurefor non-responses must be worked out.  It is possible to 
send one or more reminders or it may be necessary to make telephone 
calls or to visit non-responsive subjects at home.  It is also 
necessary to decide what to do with missing answers to part of the 
questions.  This may be solved by writing further to the respondent, 
by interrogation over the telephone or by a home-visit. 

    When an interviewer-administered questionnaire is used, the 
preparatory work includes the random allocation of interviewers to 
the selected subjects or dwellings, the schedule of visits for each 
interviewer, and data management.  A letter requesting participation 
in the study should be delivered prior to the interviewer's visit, 
with the information about the approximate time of the visit and 
the name of the interviewer. 

    Other work of the study office during the main study includes: 
(a) provision of questionnaires, instructions, and the list of 
addresses of the respondents or dwellings to interviewers and 
arrangements for their visits to subjects; and (b) recording of 
results of the work of each interviewer on special files (number of
completed interviews, refusals, persons who were inaccessible or
unavailable).  Medical and laboratory examinations

    For medical and other examinations, special premises must be 
arranged.  It is often easier to obtain appropriate premises for 
studies in the workplace or in schools.  The preparation of 
appropriate premises is more difficult, when the study covers a 
sample of a general population.  If some infectious disease, such 
as influenza, is prevailing, it would be prudent to avoid using 
clinics, hospital premises, etc., which are otherwise convenient 
for these examinations.  The location of the premises should be 
easily accessible to the study population by public transportation; 
the subjects may be provided with bus or train tickets in order to 
stimulate them to attend the examinations.  Adequate car parking 

space should be available where necessary.  Premises must be 
equipped with such facilities as reception and waiting rooms, 
interviewing or examination rooms, toilets, etc. 

    There are various methods of inviting the subjects to the 
examinations.  When medical or laboratory examinations follow an 
interview, they may be invited by the interviewers.  In other 
circumstances, they may be invited by mail, but this method is 
usually not very successful.  An appropriate procedure with 
refusals and with difficult subjects should be prepared.  Sending 
interviewers, recall letters, or telephoning are the usual methods.  
In order to increase the response rate, it may be possible to 
perform some measurements in the home of the study subjects, if 
portable instruments are available.  Consideration should be given 
to providing some people, especially the disabled and aged, with 
transportation to the examination premises. 

    If a number of observers and instruments are involved in 
the study, each observer and each instrument should have an 
identification number.  The numbers should be recorded on each 
subject's record form.  The method of measuring instrument or 
observer bias, described in section 5.6.6 may also be used for 
the analysis of results obtained in the main field study.  When 
the main study has been completed, it is certainly too late to 
improve the quality of measurements, but a knowledge of the 
sources and size of the bias that affected measurement results 
would avoid misinterpretation of the results and improve the 
quality and pertinence of the final conclusion.   Environmental measurements

    A main organizational problem in any study involving 
environmental measurements is the distribution of sampling sites 
according to previously-established objectives of the study and 
sampling design.  There may be a number of difficulties to over-
come.  For example, according to the study design, the inlets of 
the instruments for measuring pollutants in the ambient air should 
be situated at a certain uniform height above ground level and at a 
distance from busy streets and chimney stacks.  It may be difficult 
to find appropriate places for setting up all the sampling sites 
according to all above requirements.  The consent of the local 
people for the installation of a sampling site may be required, 
especially when the instruments are noisy. 

    A plan of collection and transportation of samples to the 
laboratories should be prepared.  The laboratories should be 
properly equipped and should have competent analysts.  Recording 
forms for the results should be prepared and the identification of 
samples, according to the place where the samples were taken and the 
date when they were taken, should be recorded.  Adequate control 
procedures to assure the quality of the measurements should be 

    It is necessary to foresee appropriate solutions for unexpected 
events, such as damage to apparatus.  Linkage and evaluation of data

    Because subjects in a study are usually submitted to various 
procedures, such as interviewing, medical examinations, laboratory 
tests, etc., there is a need to link the information related to the 
same subject, recorded in different places and on various forms.  
Therefore, the same identification numbers or symbols of individuals 
should be used on all forms.  In follow-up studies, when the same 
subjects are re-examined one or more times, each form or the group 
of forms used in the subsequent round of the study should be marked 
with the same identification symbol or number as that used 
previously to ensure linkage. 

    Nowadays, linkage problems in the analysis of data are usually 
solved with the help of computers.  Individual questionnaires and 
recording forms include information about the date and place of the 
measurements.  Each sample taken from air, water, food, human 
tissues, etc., is identified by place and time on the recording 
forms.  This facilitates the linkage of the data from different 

    When all data are in the computer file, the next task is to 
obtain a total print-out of the file and review aberrant values and 
perform any further clean-up, as indicated.  Then, the team leader 
and epidemiologist/statistician, aided by the data processor, must 
obtain simple numerical and demographic outputs, e.g., the number of 
participants; the overall participation rate; characterization of 
participation by such variables as age, sex, place of residence, or 
occupation; and participation rates in each of the subgroups.  If a 
proportion of the non-respondents has been surveyed, it will be 
desirable to compare and contrast the salient features of 
respondents and non-respondents.  When these necessary preliminaries 
have been completed, it will be possible to proceed to the crucial 
final tabulation and to the conduct of such other calculations as 
are suggested by the results or as are required to separate out the 
possible influences of potentially confounding variables (sections 
2.5,, and 6.4.6).  Reporting of results

    A feed-back of the study results to the participants is 
essential; there is little hope that the study team will ever be 
invited back to do a follow-up investigation, if the results are not 
made known to the subjects.  However, release of unverified results 
must be avoided, and the unduly urgent demands of the press and 
public officials must not be allowed to take precedence over the 
need for scientific accuracy.  If public inquiries are considered to 
be likely, it is advisable to appoint in advance one person, 
generally the team leader, as spokesman and to instruct all other 
team members to keep a prudent silence.  Some have found that the 
release of study results is best handled by the simultaneous public 
announcement of summary results to all parties, followed immediately 

by the personal notification to each subject by letter or by home 
visit of his individual results and their meaning.  In a study that 
takes months or years to complete, participants should be informed 
of their own results from time to time and referral for further 
clinical examination or treatment should be suggested and expedited, 
if necessary. 

    Reports of study results to the community and to policy makers 
would frequently provide the basis for the assessment and evaluation 
of environmental health risks, in a particular local situation.  The 
study team may then be responsible for providing further advice to 
the community and policy makers for the control of the hazards and 
for the prevention of disease.  Additional discussions on the 
subject will be found in section 6.5 and in Chapter 7. 

5.6.8.  Examples of cohort studies  Michigan polybrominated biphenyls studya

    In prospective cohort studies, great care must be exercised to 
maintain sufficiently close contact with the cohort to ensure that 
its size is not appreciably diminished, over time, as the result of 
cumulative refusals, or simply as the result of a slow loss of 
interest.  In a long-term prospective cohort study of the health 
status of persons exposed to polybrominated biphenyls (PBBs) in 
Michigan, USA, the Centers for Disease Control and the Michigan 
Department of Public Health found that a combination of the 
following techniques was useful and effective in maintaining contact 
with the exposed cohort:  first, a detailed explanation of the 
proposed study was sent by mail to all prospective participants and 
to the physicians in the area; second, a field office was established 
in the centre of the severely-affected area and participants were 
told that they would always be welcome with questions, comments, or 
complaints about the study or about the chemical exposure situation 
generally; and third, all participants were visited in their homes 
and again the study was explained to them.  If a prospective 
participant indicated at this point that he wished to join the 
study cohort, he was asked to read and sign a detailed consent 
form.  An admission interview was then conducted and a venous blood 
sample was collected for analysis for PBBs.      

    In each subsequent year, every participant was sent a postal    
card to ascertain his current place of residence and to inquire     
about the occurrence of any major illnesses in the preceding year.  
Those who did not reply to the card were visited personally.  Every 
two to four years, each subject was revisited at his home and a     
brief follow-up interview conducted.  This interview was intended to
supplement the necessarily limited data obtained by the postal      
interview.  Most importantly, one to three months after each 
interview and blood-collection, all participants were sent a 

a   Based on the contribution of Dr P. J. Landrigan, National              
    Institute for Occupational Safety and Health, Cincinnati,              
    Ohio, USA.                                                             

detailed letter giving a summary of the data obtained on them and 
an explanation of its significance; if the subject so desired, a       
similar letter was sent to the family physician.  As a supplement to
these formal letters, informal newsletters, which described the     
progress of the study in general terms, were sent regularly to all  

    While these procedures for maintenance of a cohort are obviously 
expensive and time-consuming, many are of the opinion that a 
decision to embark upon a cohort study of an environmental health 
problem should not be undertaken unless the principal investigator 
and his team are willing to commit themselves to carrying out 
procedures such as these.                                                  Study on air pollution and adverse health effects in Bombaya

    The experimental evidence for the biological effects of air 
pollutants is well accepted, but further epidemiological evidence is 
needed regarding the relationship between ambient air pollution and 
its long-term effects.  This involves a study of interaction between 
other environmental factors, such as tobacco smoking, occupational 
exposure, and indoor air pollution.  Other factors such as under-
nutrition, contaminated food or water supplies and poor sanitation, 
which exist in many developing countries, may lower resistance to 
infections and may complicate interpretation of air pollution 

    Several communities in Bombay, India, had been attributing human 
morbidity there to prevailing levels of air pollutants, such as a    
mean sulfur dioxide level of 50-130 µg/m3 over 24 h.  An             
epidemiological study was initiated to elucidate the claimed         
relationship, taking into account the effects of a tropical humid    
climate and the poor nutritional and sanitary conditions of many of  
the inhabitants.                                                     

    After a pilot survey of prevailing levels of sulfur dioxide   
(SO2), suspended particulate matter (SPM), and oxides of nitrogen 
(NOx) in Bombay at 10 sites for 3 years, three areas in the city  
with different levels of pollution, namely, high, moderate, and   
low, were chosen.  The last area was to serve as a control, but it
showed significant SPM and nitrogen dioxide (NO2) pollution.      
Therefore, a rural control area situated 40 km southeast of the   
city was added.                                                   

(a) Composition of study team                                        
    A team to study health effects was set up consisting of doctors, 
social workers, statisticians, technicians, health visitors,         
dieticians, and administrative support.  An environmental study team 
included engineers, chemists, meteorologists, field assistants, and  
a Based on the contribution from Professor S. R. Kamat, Department of 
  Chest Medicine, K.E.M. Hospital, Bombay, India. 

(b) Study areas and populations                                      
    In any large city, localities usually do not grow 
simultaneously or similarly.  By natural selection, different 
areas may have distinctive profiles because of differences in 
housing, ethnic, income, and other factors.  In order to reduce as 
much confounding as possible by other health effects compared with 
those of air pollution, employee groups living in a cluster of 
buildings were chosen for the study subjects.  They were more 
stable in residence and had their own welfare and health 
activities, which made it easier to get their cooperation and 
involvement in the study.        

    A full census of four chosen communities located in central 
Bombay (Lalbaug), an eastern suburb (Chembur), a western suburb 
(Khar), and a rural area (Poynad), was undertaken in December 1976.  
Lalbaug had various different industries that had been in operation 
for up to 100 years; while Khar did not have any large industries 
and Chembur was a new suburb developed in the last 25 years with 
fertilizer and petrochemical industries.  The rural area had only 
two rice mills, but had poor sanitation with 39% of the population 
living in temporary housing. 

    During the census, data concerning age (grouped as 1-9, 10-19, 
20-44, 45+ years), sex, family income, duration of residence (up to 
5, 6-10 years, and over), occupation, smoking, and housing were 
collected.  In the four areas, information was obtained on the 
subjects from 1060, 456, 605, and 393 families, respectively, and 
41, 27, 50, and 4 families were not covered.  Of the 122 families 
not covered, 28 refused to cooperate while others were only 
temporary residents. 

    In order to reduce differences among study areas the above 
factors (age, sex, etc.) were matched on a computer.  In each study 
area, 200-250 families were chosen, with a 20% excess in case of 

(c) Measurements of pollutants

    One (or two, if the communities were spaced more than l km 
apart) monitoring station in each area was set up.  In each of 3 
areas, the stations monitored SO2, NO2 and SPM, every fifth day
for a full 24 h (at a height of 12-18 m), thus covering all working 
days, once every 4 weeks.  In the rural area, as pollutant levels 
were low, the measurements were restricted to 7 week days, once 
every 4 months.  Though about 8 months were needed to set up all the 
stations, an 80-90% coverage for monitoring schedules was 
subsequently achieved over 3 years. 

    For deriving readings for SO2, NO2, and SPM, the standard US 
Environmental Protection Agency (USEPA) methods were followed.  
Daily, monthly, and yearly mean readings for each area were derived.  
These levels were correlated with measures of morbidity by clinical 
examinations, lung function test, and daily health diaries. 

(d) Assessment of health effects

    During the summer of 1977, a laboratory was set up for 4-6 
weeks, in each study area, in turn.  A coded form with details of 
occupation, housing, smoking, and clinical history was devised.  
Clinical examination, blood count, urinary sugar test, and lung 
function tests (FVC, FEV1, maximum expiratory flow rate, and peak
expiratory flow) were carried out.  Most adult subjects were 
subjected to a 70 mm X-ray on another day.  All urban subjects were 
re-examined six times, and the rural subjects four times, over three 
years.  Daily health diaries were maintained for common colds, 
cough, breathlessness, diarrhoea, medical treatment, and absence 
from work. 

(e) Cooperation of study subjects

    The initial cooperation of study subjects was obtained by 
discussing with the subjects themselves, administrative personnel, 
social workers, and community doctors, through small meetings.  The 
investigators promised confidentiality, care, and non-interference 
in local affairs, and avoided reference to political matters. 

    For each medical examination, about 25% and particularly the 
younger subjects submitted to tests promptly and 30% cooperated 
after frequent visits.  In many cases, habitual lack of punctuality 
contributed to delays.  About 20% of subjects persistently refused 
and about 25-30% came provided that examinations were performed 
during the evening. 

    In the more polluted areas (Lalbaug and Chembur), cooperation 
was greater.  In the rural area, despite care, an impending local 
election and local feuds and rivalry resulted in a poorer coverage, 
though this situation improved slightly when the team stayed in the 
villages for the period of the follow-up.  In the rural area, the 
habit of families to move out to farms in the summer reduced the 
success of the follow-up during summer examinations.  Overall, 35% 
of the subjects were lost to the study in 3 years. 

    Health diaries were maintained initially by 670-850 urban 
subjects in each of 3 areas and 250 rural subjects.  The cooperation 
at one year dropped to about 600 in each urban area and 125 subjects 
in the rural area.  At 2 years, because of certain doubts about the 
reliability of some of the records and cards, the diaries were 
continued by only 328 to 465 urban subjects from each of 3 areas and 
100 in the rural area. 

    The causes for non-participation in the urban study areas were 
refusals (30-80%), temporary absence (7-30%), moving away (2-30%), 
deaths (1-2%), and physical disability (3%).  In the rural area, 
non-participation was due to refusal (82%) and temporary absence 
(12%).  The main reasons for refusals included lack of 
communication, ignorance about the nature of the study and 

(f) Results of medical examinations

    The initial results suggested a relation between the air 
pollution levels and several health abnormalities.  Generally, the 
areas of high and moderate pollution showed a high morbidity; the 
area of low pollution had the best health status. 

    Radiographs were done on 55% of the 4129 subjects.  Of these, 
87-90% were normal, 0.7-1.0% showed evidence of old or recent 
tuberculosis, 5.7% showed cardiac problems, and 3.9% postinfective 

    As it was known that 20-40% of urban subjects had recurrent 
nasal problems and postnasal discharge, sputum samples were studied 
in 149 subjects (63, 41, 31, and 14 in the respective areas).  In 
86-94% subjects, the specimens revealed upper respiratory epthelial 
cells, suggesting that this prevailing morbidity in Bombay seemed to 
originate in the upper respiratory tract. 

(g) Results of other studies

    The smokers (mostly cigarette smokers in urban areas) amounted 
to about 17% of the subjects, with 1-6% of ex-smokers and 5-9% of 
tobacco chewers.  In females, there were 10% tobacco chewers and 
only 0.4% smokers. 

    There were major differences in housing:  in the rural area, 39% 
of the houses were temporary structures with bamboo walls and 
thatched grass roofs.  The majority of the urban subjects lived in 
small flats with a "poor" environment. However, this situation still 
represented the better aspects of the city's housing, because 30% of 
the population in Bombay live in slums with unhygienic sanitary 

    The use of kitchen fuel showed large differences, as 96% of 
rural families used wood in poorly-ventilated kitchens, compared 
with 6-12% in the city.  These differences, along with poorer 
quality of water and sanitation in the rural area, may explain the 
urban/rural differences in the morbidity observed. 

    A full diet and nutritional survey was carried out in all areas 
with the help of two nutritionists.  The procedure was to complete 
on a form the family's consumption of all food commodities, over a 
week, and the quantities eaten by each subject, daily, for 7 days.  
The results indicated that poor nutrition was a significant factor 
in producing increased morbidity, particularly in the rural area.   Tucson chronic obstructive lung disease studya

    To illustrate techniques in population studies of chronic 
diseases and the environment, a multidisciplinary study in Tucson is 
described (Lebowitz et al., 1975).  The study team had an 
a Based on the contribution from Professor M.D. Lebowitz, University 
  of Arizona, Tucson, USA.                               

epidemiologist/statistician and a clinician as principal 
investigators.  Co-investigators included physiologists, 
immunologists, and other clinicians.  Other staff included nurses, 
technicians, programmers, statisticians, key punchers, clerks, and 

    The major objectives of the Tucson study were the etiology, 
natural history, and early detection of chronic obstructive lung 
disease.  The general hypotheses included: the influence of various 
environmental and social factors on the development of the asthma-
chronic bronchitis-emphysema syndrome, including the importance of 
familial factors and of childhood respiratory illnesses in the 
development of chronic airway obstruction.  This was a longitudinal 
study of a large multi-stage, stratified geographical cluster 
population sample.  It was endeavoured to keep to a minimum the 
prestudy self-selection as well as withdrawal or loss of 
participants in order to avoid demographic and health biases. 

    Before starting the study, the study protocols and the 
consent form for participants were reviewed by the institutional 
review board.  It was felt that it was easier to keep track of 
families than of individuals.  Micro-environmental characteristics 
could be determined in this manner.  As in most chronic disease, 
age, sex, social status, and ethnic groups are all highly 
significant variables in the study of chronic obstructive lung 
disease.  Therefore, stratification was made on all these variables 
except sex (since families were the study units).  To ensure 
adequate geographical representation, the sample was a two-stage 
stratified cluster sample, using the 1970 census block statistics 
for the Tucson area. 

    Samples were selected from almost all of the blocks that met 
the criteria for the older-age strata and from most of the blocks 
that met the criteria for the middle-age strata.  The blocks were 
picked randomly within clusters in the strata.  Within each block, 
households were systematically sampled at a 1 to 6 ratio, starting 
at a random corner and going clockwise.  Before participation, 
subjects were informed about the general nature of the study and 
the benefits and risks.  The participants, willing to participate, 
signed consent forms.  Potential bias in both demographic and 
health characteristics was minimized, since it was confirmed that 
the refusal households were not different from the consenting 

    Extensive training was given to nurses to qualify them as survey 
interviewers and technicians.  Their training enabled them to 
administer the questionnaires, answer questions on the self-
completion questionnaire, and carry out objective testing.  They 
were instructed to respond to questions in a standard manner.  A 
pilot study was conducted for further training and pretesting of the 
various techniques. 

    As bias was likely to be introduced through the way in which the 
questionnaire was administered by different interviewers, a self-
completion questionnaire was devised.  This was pretested, compared 
to the original standardized questionnaires, and revised as required 
(Lebowitz & Burrows, 1976). 

    Interobserver variability measurements were made regularly on 
the spirometry (Knudson et al., 1976) and on the reading of the 
allergy tests (Barbee et al., 1976), in order to make appropriate 
corrections.  Quality control procedures were carried out on the 
laboratory determinations and on the strip chart readings from the 
spirometry.  All information was recorded on preprinted forms ready 
for computerization.  Standard quality control techniques were also 
used in the coding, key-punching and computer edit-checking of the 
data.  Confidentiality was maintained by means of limiting access 
to original data by staff, elimination of personal names or 
addresses on data files, and the use of a pass word in computer 

    The main study has been running since the beginning of 1972 with 
funding by US National Institutes of Health.  It has been shown that 
weekly respiratory symptoms were strongly correlated with weekly 
levels of air pollution and pollen, when controlling for climatic 
conditions (Lebowitz, 1977), and that the micro-environment is 
important.   The Tecumseh community health studya

    The Tecumseh community health study is a comprehensive, 
prospective epidemiological investigation of health and disease in 
the population of a geographically defined community (Higgins, et 
al., 1967a,b; Higgins & Keller, 1975; Higgins, 1977).  The purpose 
of the investigation was to detect the characteristics of man and 
the environment related to health, to resistance and susceptibility 
to diseases, such as coronary heart disease, hypertension, chronic 
obstructive lung disease, diabetes mellitus, and arthritis, and to 
the onset and course of the diseases. 

    The community to be chosen had to be stable, well-defined, and 
with a variety of occupations and living conditions; it should not 
have large seasonal fluctuations in population or be a suburb or 
dormitory of a larger city.  Other desirable features were:  the 
presence of a hospital and the cooperation of the medical 
profession; a history of community interest in health affairs; and 
the support of local mass media and community organizations.  The 
most critical element that would determine the success or failure 
of a long-term study was, of course, the willingness of the people 
themselves to take part. Tecumseh was chosen from all the possible 
cities, towns, and villages because it satisfied most of these 

a Based on the contribution from Professor M.W. Higgins, School of 
  Public Health, University of Michigan, Michigan, USA.  

    The physical boundaries of the study area were drawn to include 
the population that used Tecumseh as the centre for social and 
economic services and activities.  A map was constructed with 
reference to fixed boundaries, such as administrative subdivisions, 
school and postal districts, and utility service areas.  Information 
was collected on shopping habits and on patterns of membership in 
churches and other local organizations.  Almost all persons living 
within the study area of 145 km2 were members of the Tecumseh 
comunity.  About two-thirds of the population lived within the city 
limits and one-third in the surrounding rural area. 

    In 1957, door-to-door canvassing was conducted by trained     
interviewers who completed household and kindred listings and left
forms for residents to report chronic conditions and physical     
impairments as well as a monthly record of illness, injury, and  
disability.  There were about 8800 residents in the study area,  
living in 2400 households and belonging to 3400 kindreds or blood
lines.  It appeared that there would be enough cases of the diseases 
of major interest and that the population was cooperative.  The 
decision was made to  proceed and funds were secured for the major 
study.  Questionnaires were developed, examination procedures 
selected, and a system to determine the sequence of contacts was 
established.  The study area was divided into five strata based on 
geographical and socioeconomic considerations:  a sixth stratum was 
provided for newly-constructed housing.  Ten percent of the 
households in each stratum were selected at random and combined to 
form one representative sample.  This was repeated until the entire 
population was assigned to one of the 10% representative samples.  
Each sample was therefore a cross-section of the whole community, 
which made findings referable to the whole community. 

    For medical examinations, appointments were made for attendance 
at a special clinic where the staff physicians from the University 
of Michigan reviewed information from questionnaire surveys, 
collected additional medical information, and carried out physical 
examinations.  They diagnosed diseases present with two degrees of 
certainty, probable and suspect.  Nurses and trained technicians 
performed clinical measurements and tests.  The staff physicians 
reviewed all laboratory results and prepared reports for the 
subjects and their physicians.  Agreement was reached with the 
local physicians about which abnormalities should be referred to 
them.  No treatment was provided by the Tecumseh study staff.  
Reviewing physicians also completed diagnostic summaries on which 
they indicated whether diseases were absent or present at probable 
or suspect levels.  Diagnostic criteria were developed for diseases 
of major concern including coronary heart disease, diabetes 
mellitus, chronic bronchitis, asthma, and arthritis.  Death 
certificates were obtained for all the deceased. 

    In addition to the information collected from subjects about 
aspects of their own current and past environments, a number of 
studies were made of the physical, biological and social conditions 
existing in the study area.  These studies included measurements of 
meteorological conditions, air pollution, water hardness and purity, 
radioactive content of milk and water, identification of infectious 

agents prevalent in the community, characterization of the animal 
population of the area, and descriptions of social stratification 
and organizations in the community.  The community had had little 
air pollution.  The drinking-water had high concentrations of 
calcium, magnesium, and iron and was hard by usual standards. 

    A variety of cross-sectional, retrospective, and prospective 
studies have been carried out over a period of 20 years.  During 
this period, there has been a good deal of movement into and out of 
Tecumseh and the subjects examined in recent years have no longer 
constituted a geographically-defined population.  There have also 
been births and deaths, and the best estimate is that less than 
half of the current residents within the study area ever took part 
in the study.  Lack of resources precluded continuing the study of 
the entire community.  Late effects of atomic bomb radiationa

    There have been basically three longitudinal (prospective) 
studies in progress to determine the effects of radiation exposure 
(gamma and neutron) on the cohorts who were present or  in utero in 
Hiroshima, Nagasaki, and their environs at the time of the atomic 
bomb explosions in 1945.  A cohort of about 110 000 individuals was 
established in 1950; about 51 000 were exposed at less than 2000 m 
from the hypocentre, 32 000 were controls in the city, but more 
than 2500 m from the hypocentre, and 27 000 were unexposed controls 
(controls matched to the exposed by age and sex).  The mortality of 
these survivors and non-survivors has been studied since 1950, many 
with autopsies (20-40%) (Zeldis & Matsumoto, 1961) and findings in 
both groups have been published (UNSCEAR, 1977).  Since 1958, 
twice-yearly examinations have been conducted on 20% of the cohort 
(the Adult Health Study), and cancer mortality and incidence have 
been followed in all.  The  in utero exposed (2800) and matched 
controls represent another longitudinal study to determine 
mortality and morbidity after birth (the  In Utero Exposure Study).  
A large-scale genetic study based on pregnancy registration (1948-
1954) was conducted, starting in 1958, on a cohort of 54 000 
children, whose parents include the exposed and unexposed groups, 
for cytogenetic and biochemical studies as well as mortality (Neel 
& Schull, 1956; Kato & Schull, 1960) (the Genetic Study).  No 
genetic defects have been noted up to now from the records of birth 
defects or mortality, or from the results of the chromosomal 
aberrations study. 

    The mortality follow-up study relies on keeping track of the 
people; where and when they die, making autopsies if possible, and 
obtaining death certificates (Ishida & Beebe, 1959).  It has 
demonstrated increased relative risks for some cancers (leukaemia, 
lung and stomach cancers); confirmation rates were 70-80% for these 
cancers, but less than 50% for some others (pancreas and liver 
cancers) (Yamamoto et al., 1978).  The Adult Health Study (JNIH-
a Based on the contribution from Dr H. Kato, Radiation Effects 
  Research Foundation, Hiroshima, Japan.                  

ABCC, 1962) contacts subjects by telephone or in person to arrange 
the examination; participation was initially 85% and is now 75%, 
though decreases in sample size have occurred through death and 
migration.  Medical records on the cohort are studied  between 
examinations.  There is a record linkage with the tumour registry,  
from which incidences of cancers (i.e., breast and thyroid cancers) 
have been determined in relation to radiation exposure.  Information
on other carcinogenic etiological factors is obtained from the      
cohort through interviews, mail surveys, and record linkage with    
census data.  Information is obtained on the following:  smoking,   
occupation, history of mental illness, family history, dietary      
habits, exposure to medical X-rays.  So far multiple-risk models    
have shown additive effects (but no synergism of the radiation) with
smoking for lung cancer and with various risk factors for breast    
cancer (Nakamura et al., 1977).                                     

    Data are updated and analysed continuously.  Other results 
indicate:  eye problems (section 4.5.3); increase in chromosomal 
aberration with dose, similar to leukaemia (Awa et al., 1971) though 
its clinical meaning for carcinogenesis or immune abnormality is 
unknown; no lowering of the immune function has been observed; no 
shortening of life span (except by cancer mortality) has been found 
(Finch & Beebe, 1975); the frequency of mental retardation increased 
among  in utero exposed children (Blot & Miller, 1973).  Leukaemia 
began appearing 2 years after exposure, reached its peak in 5-7 
years (depending on the age at exposure) and now has decreased 
almost to control level (Ichimaru et al., 1978).  Other cancers with 
longer latency periods appeared at ages when such cancers normally 
occur and increased proportionally to age-specific population rates. 

    As to exposure/effect relationships, some uncertainty continues 
to surround both the quantity and quality of the radiation released 
by these two nuclear devices, particularly the Hiroshima bomb.  Only 
one weapon of the latter type has ever been detonated and thus its 
yield has had to be reconstructed.  Different reconstructions have 
led to different estimates of the gamma and neutron exposures.  A 
recent reassessment suggests that the gamma estimates used in the 
1965 calculations might have been too low and the neutron estimates 
too high, and that total kinetic energy released (kerma) may have 
been greater than previously supposed (International Committee on 
Radiation Protection, 1977)a. 

    Given the uncertainties, attention here is restricted to 
exposure expressed as total kerma (tissue), since this metric 
changes least, relatively, for exposures of 0.1 Gy or more when 
these assessments are contrasted with the radiation dose calculated 
in 1965.  Unfortunately, the newer calculations are still not 
complete enough to form the basis of a meaningful dose-response 
analysis based on individual exposure assessments. 

a Another review of the dosimetry (Beebe et al., 1978) differs in 
  some of these particulars.                                  

5.7.  International Collaborative Studies      
    The principles of planning and execution of an epidemiological  
study are similar for both national and international studies.  When
a national study is performed simultaneously in various areas within
a country, the problems of coordination and standardization of study
methods are similar to those that have to be solved in collaborative
international studies (Acheson, 1965).                              

    Although it may be necessary to have various expertises in a    
study team in participating countries, practical solutions should be
left to the local team leaders.  For example, social workers may be 
employed as interviewers in one country and professional inter-     
viewers used in another.  However, interviewers should be trained in
a uniform way, in accordance with the protocols set.  It is highly  
desirable that an experienced epidemiologist or interviewer conducts
the training in a participating country after having received joint 
training at the coordinating centre of the study.  The same 
principles of uniform training methods should be applied to all 
other field workers.  Detailed instructions should be prepared for 
each group of field workers and should be carefully translated for 
use in different countries. 

5.7.1.  Study protocol and timetable

    The study protocol, as well as the plan of the whole study, 
should be identical for all countries and areas. Initially, a draft 
of such a protocol, as complete as possible, should be prepared by 
the coordinator of the international study.  The draft should be 
sent to the groups in the participating countries, prior to the 
meeting at which the protocol will be discussed, corrected if 
necessary, and approved. 

    Although the study protocol should be identical for all 
participating countries, there may be some problems that require a 
different solution in different countries.  However, the solutions 
should be designed and realized in a way that will assure obtaining 
comparable results. 

    The timetable for each study group should be centrally 
coordinated.  However, it does not seem necessary to perform the 
particular stages of a study at exactly the same time, in various 
countries.  Sometimes, when possible climatic effects are to be 
taken into account, it may be necessary to perform the main field 
study in different months in different countries, in order to assure 
that the results will be obtained under similar climatic conditions. 

5.7.2.  Organizational and sampling procedures

    Although it is essential to secure comparable results, 
different procedures in the study organization may be followed 
in different participating countries.  For example, the contents 
of an advance letter may differ according to local customs and 
other conditions.  The letter may be signed by different persons 
in various countries (e.g., the mayor of the community in one and 

the president of a university or chief medical officer in another).  
In some countries, additional inquiries may be made by telephone, 
in others, only a few respondents may have a telephone at home and 
people may be unaccustomed to discuss matters by this means.  All 
local solutions and adaptations should be mentioned in the local 
study protocol. 

    Problems related to sampling procedures, which may be faced in 
the international studies, are mainly related to the availability of 
sampling frames, especially where general population surveys are to 
be performed.  The use of various sampling frames in different 
countries implies the use of different methods of sample section,
which may lead to an increase in the extent of sampling error.  
Consequently, it may be necessary to increase the sample size in the 
respective areas leading to an increase in costs.  In addition, some 
methods of sample selection (e.g., selection of clusters) may 
necessitate the application of specific methods of statistical 
analysis.  When a unique sampling frame is not available for all 
participating countries, the best solution is to emphasize "ends" 
rather than "means" and to require each country to prepare a 
sampling design in accordance with accepted general principles (Kohn 
& White, 1976). 

5.7.3.  Questionnaires

    One of the main problems to be solved in an international 
collaborative study is the correct translation of the questionnaire 
into different languages in order to obtain comparable results among 
participating countries.  The main task of translation of the 
questionnaire is to assure the semantic equivalence of the contents 
of questions. Translators must be aware of differences in the 
colloquial usage of specific phrases or words.  Some phrases or 
words may even have a different meaning in countries using the same 
language, or in various areas within the same country.  For example, 
it is not always possible to find an adequate literal translation 
from the English of the question "Does the weather affect your 

    In addition to the purely linguistic problems, it may happen 
that the study population in various countries or in various areas 
in the same country, may be different from the point of view of 
literacy, education level, cultural background, etc.  These problems 
should also be considered when the questionnaire is translated.  
Before the final approval of the translated questionnaire, it is 
usually necessary to perform several pretests of parts of, or of the 
whole questionnaire. 

    The translation of the questionnaire must be carefully done, 
preferably by a person who is bilingual and specialized in that 
particular field.  This text should then be retranslated into the 
original language, if possible by a professional translator 
unfamiliar with the subject study.  A broad description of problems 
related to the translation of the original questionnaire into 
several languages is given by Kohn & White (1976). 

    Another non-linguistic problem, related to the comparability of 
a questionnaire used in different countries, may arise when the it 
includes questions designed for the measurement of various social 
characteristics.  Even when such a seemingly simple variable as, for 
example, the education level of a respondent is to be measured, 
questionnaires in different countries should be devised taking the 
different schooling systems into account, in order to assure 
comparable measurement of the education level in the participating 
countries.  Similar considerations may be required when studying 
family income, coverage by social insurance, disability, sickness, 
use of health services, and other similar variables. 

5.7.4.  Standardization of measurement instruments and methods and 
quality assurance

    In order to maintain uniformity of measurement methods and 
comparability of results, it is best to furnish all study groups 
with the same measuring instruments and equipment.  However, this may 
be impractical and prohibitive in cost and various instruments and 
consequently different methods may have to be used.  Under such 
circumstances, it is necessary to check the comparability, 
reliability, and validity of measurements by these instruments and 

    The importance of, and detailed procedures for analytical 
quality assurance in an international collaborative study, in which 
ten countries including six developing countries participated, have 
been described by Hasegawa (1983).  The essential components of the 
quality assurance programme in this international study were:  (a) 
preanalytical quality assurance - use of the same equipment, 
reagents, etc. with known contents of contaminants in question, and 
ultra-care for avoidance of any contamination during the collection, 
transportation, and storage of samples; (b) establishment of 
criteria for acceptance of analytical results; (c) repetition of 
quality assurance exercises until all the analytical results by the 
participants have met the criteria for acceptance; and (d) quality 
assurance checks during the analyses of samples from an actual study 
- the analytical results were to be adopted only when the checks met 
the criteria for acceptance. 

5.7.5.  Reporting forms

    The need for uniform recording forms depends mainly on the 
arrangements for data processing and analysis.  If all data are to 
be processed and analysed together in one computing centre, 
standardized reporting forms and coding sheets must be used in each 

    If the study design provides for separate processing and 
analysis of data for each country, the design of reporting forms and 
coding sheets must be unified to an extent that the same analysis of 
data and production of comparable tables and indices are possible.  
These tables and indices will be necessary for the preparation of a 
common final report for all participating countries. 


ACHESON, R.M. (ed)  (1965)  Comparability in international
epidemiology. Selected papers from the Fourth Scientific
Conference of the International Epidemiological Association,
Princeton, 1964.  Milbank Memorial Fund Quarterly,  43(2): (part
2), 432 pp.

MATSUI, T.  (1971)  Chromosome aberration frequency in
cultured blood-cells in relation to radiation dose of A-bomb
survivors.  Lancet,  2: 903-905.

(1976)  Immediate skin-test reactivity in a general population
sample.  Ann. Int. Med.,  84(2): 129-133.

BEEBE, G.W., KATO, H., & LAND, C.E.  (1978)  Studies of the
mortality of A-bomb survirors. 6. Mortality and radiation
dose, 1950-74.  Radiat. Res.,  75: 138-201.

BLOT, W.J. & MILLER, R.W.  (1973)  Mental retardation
following in utero exposure to the atomic bomb of Hiroshima
and Nagasaki.  Radiology,  106: 617-619.

COLLECTIVE WORK  (1969)  Chronic non-specific respiratory
diseases in the city of Cracow. XI. The cross-section study.
 Epidemiol. Rev.,  23: 242-252.

(1982)  Proposed International Guidelines for Biomedical
Research involving Human Subjects. In:  Human Experimentation
 and Medical Ethics, Proceedings of the XVth CIOMS Round Table
 Conference, Manila, 13-16 September 1981, Geneva, CIOMS/WHO.

FERRIS, B.G.  (1978)  Epidemiology standardization project
(American Thoracis Society).  Am. Rev. Respir. Dis.,  118 (No.6,
Part 2): 1-120.

FINCH, S.C. & BEEBE, G.W.  (1975)  Review of thirty years
study of Hiroshima and Nagasaki atomic bomb survivors, F.
Aging.  J. Radiat. Res.,  16 (Suppl.): 108-121.

HASEGAWA, Y.  (1983)  "Normal" levels of cadmium in blood and
kidney-cortex in the general population. In:  Proceedings of
 International Workshop on Biological Indicators of Cadmium
 Exposure and Diagnostic and Analytical Reliability, jointly
 organized by the Commission of the European Communities and
 the International Union of Pure and Applied Chemistry,
 Luxembourg, 7-9 July 1982  (in press).

Characteristics of smokers and non-smokers in Tecumseh,
Michigan. I. The distribution of smoking habits in persons and
families and their relationship to social characteristics.  Am.
 J. Epdemiol.,  86(1): 45-59.

Characteristics of smokers and non-smokers in Tecumseh,
Michigan. II. The distribution of selected physical
measurements and physiologic variables and the prevalence of
certain diseases in smokers and non-smokers.  Am. J.
 Epidemiol.,  86(1): 60-77.

HIGGINS, M.W. & KELLER, J.B.  (1975)  Familial occurrence of
chronic respiratory disease and familial resemblance in
ventilatory capacity.   J. Chronic Dis.,  28: 239-251.

HIGGINS, M.W.  (1977)  Epidemiology of chronic bronchitis and
emphysema at the Symposium on Pulmonary Disease Defense
Mechanisms and Populations at Risk.  Proceedings of the Tobacco
 and Health Research Institute - 2,  12-14 April 1977.
Lexington, University of Kentucky.

ICHIMARU, M., ISHIMARU, T., & BELSKY, J.L.  (1978)  Incidence
of leukemia in atomic bomb survivors belonging to a fixed
cohort in Hiroshima and Nagasaki, 1950-71.  J. Radiat. Res.,
19: 262-282.

 Recommendations of the International Commission on
 Radiological Protection.  ICRP Publication 26.

ISHIDA, M. & BEEBE, G.W.  (1959)  Joint JNIH-ABCC study of
life span in atomic bomb survivors. Research plan.  ABCC
 Technical Report,  4-59.

JNIH-ABCC  (1962)  Research plan for joint JNIH-ABCC adult
health study in Hiroshima and Nagasaki.  ABCC Technical Report,

KAHN, R.L. & CANNELL, C.F.  (1965)   The dynamics of
 interviewing.  New York, John Wiley, 368 pp.

KATO, H. & SCHULL, W.J.  (1960)  Joint JNIH-ABCC life span
study of children born to atomic bomb survivors. Research
plan.  ABCC Technical Report,  04-60.

KISH, L.  (1965)   Survey sampling.  New York, John Wiley,
643 pp.

KOHN, R. & WHITE, K.L., ed.  (1976)   Health care. An
 international study.  London, Oxford University Press, 557 pp.

(1976)  The maximum expiratory flow-volume curve: normal
standards, variability, and effects of age.  Am. Rev. Respir.
 Dis.,  113: 587-600.

LEBOWITZ, M.D., KNUDSON, R.J., & BURROWS, B.  (1975)  The
Tucson epidemiology study of chronic obstructive lung disease.
I. Methodology and prevalence of disease.  Am. J. Epidemiol.,
102: 137-152.

LEBOWITZ, M.D. & BURROWS, B.  (1976)  Comparison of
questionnaires: the BMRC and NHLI respiratory questionnaires
and a new self-completion questionnaire.  Am. Rev. Respir.
 Dis.,  113: 627-635.

LEBOWITZ, M.D.  (1977)  Temporal analysis of acute respiratory
symptoms, aero-allergens and air pollution. In:  8th
 International Scientific Meeting of the International
 Epidemiological Association (Abstracts of scientific papers),
 1977, San Juan, Puerto Rico,  p.194.

LITVINOV, N.N.  (1978)  [Approaches to the evaluation of
environmental factors as causes of diseases and the conditions
for their development.]  Proceedings of the 2nd Republican
 Conference on Medical Geography, Kiev,  1978, pp. 60-61 (in

LITVINOV, N.N. & PROKOPENKO, Yu. I.  (1981)  [To the problem
of evaluating the degree of hazards of environmental factors
for the health of population.]   Gig. i Sanit.,  10: 71-73 (in

MERKOV, A.M.  (1979)  [The health of population and methods
for its study.] Selected papers by Professor M.S. Bedny.
Moscow,  Statistika,  334 pp (in Russian).

(1977)  Epidemiologic study of breast cancer in A-bomb
survirors.  RERF Technical Report,  9-77.

NEEL, J.V. & SCHULL, W.J.  (1956)   The effect of exposure to
 the atomic bombs on pregnancy termination in Hiroshima and
 Nagasaki,  Washington, D.C., NAS-NRC Publication No. 461.

ROBINSON, W.S.  (1957)  The statistical measurement of
agreement.  Am. Sociol. Rev.,  22: 17.

RUDNIK, J., SAWICKI, F., & KLYS, J.  (1978)  [Epidemiological
study on long-term effects on health of air pollution. Study
in Poland.]  Problemy Medycyny Wieku Rozwojowego, Development
 Period Medicine,  National Research Institute for Mother and
Child, Research Reports, Supplement 7a, 1977, Warsaw, PZWL,
183 pp (in Polish).

SAGEN, O.K.  (1970)  Problems in sampling practice. In:
 Proceedings of the 5th International Scientific Meeting of the
 International Epidemiological Association. Primosten,
 Yugoslavia, 25th-31st August, 1968,  Belgrade, Savremena
Administracija, pp. 345-352.

SAWICKI, F.  (1969a)  Chronic non-specific respiratory
diseases in the city of Cracow. IV. Statistical evaluation of
the sampling methods.  Epidemiol. Rev.,  23: 56-64.

SAWICKI, F.  (1969b)  Chronic non-specific respiratory
diseases in the city of Cracow. VI. Assessment of the
interviewer error.  Epidemiol. Rev.,  23: 135-145.

SAWICKI, F.  (1977)  Regression, persistence and incidence of
chronic bronchitis in a sample of an urban population
followed-up for 5 years. Paper presented at the  8th
 International Scientific Meeting of the International
 Epidemiological Association, Puerto Rico,  pp.25.

(1969)  Chronic non-specific respiratory diseases in the city
of Cracow. VIII. Assessment of the x-ray observer error.
 Epidemiol. Rev.,  23: 146-158.

SHIGEMATSU, I.  (1978)  Epidemiological studies on cadmium
pollution in Japan. In:  Cadmium 77, Proceedings of the First
 International Cadmium Conference, San Francisco.  London, Metal
Bulletin, Ltd., pp. 213-217.

UNSCEAR  (1977)  A report of the United Nations Scientific
Committee on the Effects of Atomic Radiation to the General
Assembly,  Sources and Effects of ionizing radiation,   New York,
United Nations, 725 pp.

URY, H.K.  (1965)  A rank order test for interviewer
agreement.  Arch. Environ. Health,  10: 373-379.

WHO  (1982)   The place of epidemiology in local health work;
 the experience of a group of developing countries,  WHO Offset
Publication No.70, Geneva, World Health Organization, 43 pp.

(1978)  RERF pathology studies, Hiroshima and Nagasaki, Report
4. The autopsy program and the life span study, January
1960-December 1975.  RERF Technical Report,  pp. 18-78.

ZELDIS, L.J. & MATSUMOTO, Y.S.  (1961)  JNIH-ABCC pathology
studies, Hiroshima-Nagasaki. Provisional Research Plan 1.
Description and scope of program.  ABCC Technical Report,  11-62.

& MARGOLINA, R.B.  (1981)  [Indices of the population's health
as a basis for developing a system of measures on the
protection and improvement of the environment in a big city.]
In:  Proceedings of the Xth Congress of the Ukranian
 Hygienists.  Kiev, Tezisi dokl., pp. 28-30 (in Russian).


6.1.  Introduction

    Methods of assimilating and reporting results of epidemiological 
studies are discussed in this chapter.  Guidance is offered on how 
to arrange, analyse, and present the information, and the principles 
underlying a statistical approach to data are outlined.  But this is 
not a text on statistical methods; it is addressed to all members of 
an epidemiological study team including clinicians, epidemiologists, 
environmental scientists, computer programmers, and statisticians.  
Statisticians are expected to play a key role at this stage of the 
work and, in many cases, their contribution will be essential.  Yet 
the effectiveness of a statistical analysis depends as much on 
informed medical and environmental expertise being brought to bear 
on the early results, as they emerge from the data, as it does on 
the professional ingenuity and mathematical sophistication of the 
statistician.  The statistician should already have been involved 
in the early planning stage.  He or she should have considered the 
implications of the design that was adopted and an effort should 
have been made to understand the essence of the technical problems 
that may lie behind the research questions being asked.  Conversely, 
the other scientists involved in the work should join in the 
exciting task of unravelling the complexities of the data.  Their 
critical appraisal of interim results, backed by an understanding 
of the broad principles, if not the details, of the statistical 
methods used, will expedite the formulation of sensible conclusions 
and may reduce pointless expenditure of time and money in pursuit 
of unrewarding statistical work. 

    The aim of this chapter is to encourage such cross-fertilization 
of expertise during the analysis of data.  Data have to be prepared, 
described, analysed, and interpreted.  Finally they have to be 
reported.   Data preparation is the systematic arrangement of the 
material prior to summarization and analysis.   Data description 
involves distillation of large quantities of numerical information
into tabular or graphical summaries that are comprehensible and 
relevant to the research questions.   Analysis and interpretation 
require the application of a mathematical probability theory, with
the aim of answering the research questions embodied in the study 
objectives and design.  These matters are discussed in section 6.4, 
and some of the material at that point presupposes familiarity with 
the theory of applied statistics.   Reporting (section 6.5) covers 
all stages of communicating results from the study and should 
involve the various disciplines contributing to the work. 

6.2.  Data Preparation

6.2.1.  Coding

    Prudent form design (section 5.6.3) will have reduced the 
necessity to code data that can easily be recorded numerically at 
the point where they are captured.  But, in many cases, it will be 
necessary to translate some of the material received into 
numerically (or alphabetically) labelled categories ("codes").  For 

instance, a detailed occupational or clinical history may have been 
taken in a semi-narrative form.  It is not possible to know in 
advance which occupations, industries, diseases, or conditions will 
be mentioned.  Distinguishing codes may have to be allocated after 
the survey to aid manual tabulation or entry of the data into a 
computer.  Which individual items should be given separate codes?  
How should similar items be grouped?  How should ambiguous items be 
classified?  Discussion and decisions on these questions will 
involve various members of the study team, certainly not just the 
statisticians or computer programmers.  In general, it is better to 
designate too many codes, rather than too few.  Symbols that are 
judged later to be redundant, for instance because they represent 
synonyms for a particular job, can easily be merged during analysis.  
But if items are grouped prematurely, during coding, they cannot be 
separated easily later, when second thoughts or the pattern of 
results may suggest that a more detailed analysis would be 

    The accuracy of coding should be verified by an independent 
check on, at least, random samples of the material.  The sampling 
should be arranged so that it is representative of the different 
types of data, times of collection, and observers who recorded the 
data.  Any errors found should be regarded either as justifying a 
complete duplication of the coding operation, or at least as 
indicating the necessity for more intensive sampling.  The accuracy 
of manual transcription should be controlled by reading back each 
item on the transcript to a person other than the one who did the 
transcription.  Squared (quadrille) paper is recommended for all 
tables (and subsequent calculations) using pencil rather than ink, 
so that errors can be erased.  Data that have been checked should be 
ticked, initialled, and dated. 

6.2.2.  Key punching

    Data preparation for a computer usually involves the use of a 
key-punch or teletype.  There are several potential sources of error 
in this operation (e.g., misreading of symbols on source document, 
depression of the wrong key, omission of lines);  the accuracy of 
the transcription to the computer medium must be verified.  This is 
achieved by an independent repetition of the data entry operations.  
Various automatic methods are then available to identify discrepancies, 
depending on the machine and the computer medium being used. 

6.2.3.  Data monitoring and editing

    "Are the data what they purport to be?"  Finney (1975) reminds 
statisticians and others concerned with the interpretation of data, 
that they have an obligation to satisfy themselves on this matter 
before they proceed to summarization and analysis.  When data have 
been prepared, they need to be inspected so that errors and 
irregularities can be at least identified, and sometimes corrected.  
Finney refers to this as "monitoring".  It is closely allied to the 
concept of "validity checking" in computer operations, which often 
precedes an "edit" on a computer data file, that is, deletion or 
other alteration of information on the file prior to further 

processing and summarization.  But the term "monitoring" applies 
equally to manually prepared data in tabular form.  A deliberate 
effort must be made to identify anomalous items of data that appear 
to be implausible.  Their occurrence should be queried and 
investigated.  They should not be "corrected", unless the cause of 
the error (if it is an error) is discovered or is suspected with a 
high degree of confidence.  For instance, a datum which reads 44.5 
and purports to be a measurement of forced expiratory volume in one 
second (FEV1), in litres, may be regarded reasonably as a
decimal-point transposition error of what probably was 4.45 litres.  
This will be very plausible, if all other measurements in the set 
are recorded to the nearest 10 ml. If, however, the datum being 
queried is a FEV1 of say 0.45 litres, then "corrections" or 
deletions are not in order, unless good independent evidence is 
available that the value recorded is in error.  This can only be 
established by making an effort to trace the record back to its 
source.  In practice, this may be very difficult, and sometimes 
impossible.  This is why emphasis was placed in section 5.6.4 on the 
importance of checking crude data, soon after collection. 

    If the data are on a computer file, programmes should be run 
that seek strange data, contradictions, and impossible data.  These 
programmes should not be restricted to a search for logic errors or 
impermissible symbols.  They should include also procedures that 
identify values that lie outside plausible limits.  The specification 
of such limits is the responsibility of the physicians and other 
experts who are familiar with the measurements or observations 
concerned and with the circumstances in which they have been 
generated.  The values being queried should be listed.  They should 
not be "corrected" automatically, but should be discussed individually 
with the members of the survey team, who specified the limits 
incorporated in the edit program. Decisions on how the "errors" are 
dealt with should be documented and, ultimately, reported. 

6.3.  Data Description (or Reduction)

6.3.1.  Purpose

    The essence of the epidemiological approach is to attempt to 
apply generalizations from the individual items of data that have 
been gathered, to the group or "population" to which the individual 
items belong.  Paradoxically, therefore, the first step in the 
"analysis" of results is a synthesis of individual data into summary 
tables and diagrams that reconstitute, in outline at least, the 
patterns and fluctuations of the variables that have been observed.  
These descriptions provide a first overall view of what has been 
achieved in the study.  The main purpose is to begin to answer the 
research questions, but the tables and graphs may also reveal 
anomalies, indicative of possible errors (in survey procedures, data 
transcription, or preparation) that were not obvious at earlier 
stages of the work.  It is essential that any such suspicions are 
investigated thoroughly, before proceeding with further statistical 
analysis and reporting.  In reality, good data description provides 
the factual basis required to justify any more detailed exploration 
of results. 

6.3.2.  Frequency distributions and histograms

    A fundamental and usually indispensable summary of measurements 
on a continuous scale (e.g., age, height, blood pressure) is to 
determine the frequency distribution:  that is, a statement of the 
numbers of observations that fall into a series of contingent 
intervals within the range of the data (Table 6.1).  There are no 
hard-and-fast rules for choosing the size of the intervals.  They 
have to be wide enough to include sensible numbers of observations, 
but not so wide that they hide what may be interesting variations 
in the density of values within the intervals.  A useful rough 
guide is to determine the range of the observations (the highest 
value - the lowest value) and divide this into six to twelve 
convenient intervals, depending on the amount of material available 
and also on whether a particular subrange is of special interest.  
It is worthwhile specifying precisely where the interval begins and 
where it ends:  avoid labelling tables or graphs for, say, age 
distributions as 15-20, 20-25, ... etc., because this notation is 
ambiguous about which interval contains the number of persons who 
were aged precisely 20 years.  It is better to write:  15-19, 
20-24, ...  This is the convention now widely adopted in the 
epidemiological literature to aid comparison of results from 
different studies. 

Table 6.1.  A frequency distribution.  Estimates of cumulative exposures
to respirable coalmine dust to time of survey; 2600 miners from 10
coalmines (Data from the British National Coal Board's Pneumoconiosis
Field Research)a
             C u m u l a t i v e   d u s t   e x p o s u r e
           (gram-hours per cubic metre of air samples, gh/m3)

             0      80     120    160    200    240    280    360+   All
             -79    -119   -159   -199   -239   -279   -359
Frequency    481    374    372    355    298    261    291    168    2600

% Frequency  18.5   14.4   14.3   13.7   11.5   10.0   11.2   6.5    100
a   From: Hurley et al. (1982).

Notes:  1.  The grouping intervals differ in width at the extremes of the 
            distribution, in order to provide sensible numbers in each 
            interval for subsequent studies of the effect of exposure.
        2.  The width of the last interval is represented as open-ended 
            (360+).  This draws attention to the "tail" of the 
            distribution (section
        3.  In fact, the highest exposure recorded was less than 500 
            gh/m3.  For a graphical representation of these data see Fig. 
            6.1a and 6.1b.


    The number of observations in a particular interval, expressed 
as a fraction of the total number in all intervals, is referred to 
as the  relative frequency.  Usually, it is the relative frequencies
of observations, rather than the absolute numbers that provide the 
easiest to assimilate picture of the shape of the distribution.  
Note that the prevalence of a disease in a group is the number of 
people in the group with the disease.  Usually, however, it is the 
relative frequency of occurrence that is quoted, the  prevalence 
 rate.  This is the number of persons with the disease divided by
the total number of persons examined, for a given time and place, 
conventionally expressed as a percentage.  Relative frequencies may 
be portrayed graphically in the form of a histogram (Fig. 6.1b).  
If the grouping intervals chosen are of equal width, then the areas 
represented by the separate columns in the histogram are directly 
proportional to the relative frequencies.  This is not true, if the 
grouping intervals are of unequal width.  In this case, the 
appearance of the relative areas may give a misleading impression 
of the real relative frequencies (Fig. 6.1a).  A convenient way to 
overcome the problem is to draw the heights of columns so that 

                         relative frequency
they are proportional to:------------------
                         width of interval.

The areas are then proportional to the relative frequencies, and the 
sum of the areas always equals unity, or 100%.  A pattern of 
alternate high and low relative frequencies is usually an 
indication of a subjective bias in the purported precision of the 
recorded data.  Graphical data description can "smooth" such 
spurious periodicity by using a wider, more realistic grouping 


6.3.3.  Bivariate distributions and scattergrams

    Two variables relating to the same individuals can be summarized 
jointly by tabulating the bivariate frequency distribution (Table 
6.2).  This is usually more interesting than two separate frequency 
distributions, because the two-way tabulation shows how the 
frequency of one variable varies depending on the value of the 
other.  A systematic pattern in such variation indicates an 
 association between the variables, which may be important.  The
association (or its absence) is representable graphically in a 
scattergram.  This may record each individual pair of data-points 
(Fig. 6.2) (rather than relative frequencies) or it may show average 
values, or relative frequencies, of one variable in suitably-sized 
and mutually exclusive subgroups of the data.  Many computer 
packages have the facility for producing such scattergrams easily.  

Note that the density of points in various regions of the 
scattergram corresponds to the height of the rectangles in the 
univariate histogram.  The sums of the number of points in the 
horizontal and vertical strips of the graph show the  marginal 
frequency distribution for each of the two variables respectively.
An important area of applied statistics deals with the quantitative 
study of associations between variables and with the relationships, 
including exposure/effect relationships, that they may imply.  In 
later sections of this chapter, some aspects are discussed of the 
formidable body of statistical theory and methods that is available 
to tackle these problems.  But it is worthwhile emphasizing, at this 
point, that the plausibility of any regression or multivariate 
models, which might be postulated, should always be considered 
before formal analysis, by careful examination of visual patterns of 
associations on the scattergrams. 


    The same principles of tabular and graphical multivariate data 
description can be applied, when more than two variables are 
recorded for the same person, but both tables and, in particular, 
the graphs are then more difficult to create and also to interpret.  
Computer programs may produce the enumerations required with 
relative ease; their condensation into a form that conveys the 
pattern of the results is more difficult.  Simultaneous 
representation of three or four covariates is usually the maximum 

that can be digested easily.  For data description purposes, it is 
generally sensible to select sets of three variables in different, 
but interesting combinations, and present these in tabular form.  
Admittedly, such tables will not display all the possible 
covariation and interactions in the data.  Exploration and 
summarization of other complexities may be pursued using multi-
variate methods of statistical analysis (section 6.4). 

    A two-dimensional scattergram may be exploited to convey also 
covariation with a third variable - by using different symbols 
(dots, crosses, etc.) to represent different levels of the third 
variable (Fig. 6.3).  Three-dimensional (perspective) graphs are 
difficult to draw, though they may look attractive if well executed, 
but they rarely add much to an understanding of the material. 


Table 6.2.   A trivariate frequency distribution
Number of men in ranges of cumulative dust exposure and hours worked.  
Also shown (in parentheses) are averages of 5 physicians' assessments of 
the percentages of men in each subgroup whose chest radiographs showed 
simple pneumoconiosis amounting to category 2 or 3 on the International 
Labour Office's scale.  (Data from the British National Coal Board's 
Pneumoconiosis Field Research)a
            C U M U L A T I V E   D U S T   E X P O S U R E  (gh/m3)

Cumulative  0-     80-    l20-   160-   200-   240-   280-   360+   ALL
hours       79     119    159    199    239    279    359
(in 1000s)

0-39        73     72     84     42     19     10     3      3      306
            (0.0)  (0.0)  (0.0)  (0.0)  (0.0)  (0.0)  (0.0)  (0.0)  (0.0)

40-48       97     81     87     106    45     26     20     3      465
            (0.0)  (0.0)  (2.3)  (2.3)  (0.9)  (0.0)  (3.0)  (0.0)  (1.2)

49-56       91     65     68     60     79     45     41     7      456
            (0.0)  (0.0)  (2.4)  (0.0)  (3.3)  (2.7)  (1.0)  (0.0)  (1.3)

57-64       70     51     53     58     67     81     64     37     481
            (0.0)  (0.0)  (1.5)  (1.4)  (3.9)  (8.9)  (6.3)  (11.4) (4.1)

65-72       75     48     36     36     42     46     70     33     386
            (0.0)  (0.0)  (0.6)  (6.7)  (6.2)  (4.8)  (10.6) (7.3)  (4.5)

73-80       42     35     29     35     29     38     68     64     340
            (0.0)  (0.0)  (0.0)  (5.1)  (4.1)  (13.2) (11.8) (13.8) (7.3)

80+         33     22     15     18     17     15     25     21     166
            (0.6)  (0.0)  (0.0)  (0.0)  (7.1)  (6.7)  (8.8)  (9.5)  (4.0)
ALL         481    374    372    355    298    261    291    168    2600
            (0.0)  (0.0)  (1.2)  (2.1)  (3.6)  (6.4)  (7.8)  (10.4) 
a From: Hurley et al. (1982).

Notes:  1.  The  marginal distribution for dust exposure is the same as
            shown in Table 6.1.
        2.  The relatively few numbers in the top-right and bottom-left
            sections of the table demonstrate the  correlation between
            cumulative exposure and hours worked.  These variables are
             associated, by definition.
        3.  The marginal distributions of percentages of men with
            pneumoconiosis indicate a positive relationship with both
            associated explanatory variables.  See Fig. 6.3 for a 
            graphical representation of these data.
6.3.4   Discrete variables and contingency tables

    Replies to a question such as "do you smoke" may be either "yes" 
or "no".  The  discrete nature of such variation ("yes" or "no") is
distinguished from variations on a continuous scale (for instance 
the weights of tobacco consumed per week by pipe-smokers).  Discrete 
variables commonly encountered in epidemiological work include sex, 
ethnic group, occupation, smoking habit, geographical location, and 
responses to questions on symptoms.  Adequate description often 
requires tabulation of frequencies and relative frequencies 
corresponding to such classifications.  Graphical representation of 
discrete variable distributions should be distinguished from their 
continuous variable analogues by representing the frequencies as 
proportional to the heights of vertical lines or clearly separated 
rectangles, rather than by adjacent rectangles depicting grouped 
data on a continuous scale. 

    Tables of multivariate discrete distributions of frequencies 
are known as  contingency tables.  They may be simple, perhaps 
involving only two variables (say sex and current smoking habit) 
with each divided into only two levels (male/female; smoker/non-
smoker).  Or they may be complicated, hierarchial (or "nested") 
arrangements of   frequencies according to several discrete 
variables, with each at two or more levels.  The statistical 
analysis of such material has generated a large literature of its 
own and the most important methods are described in standard text-
books.  For the purpose of data  description, the essence of an 
effective presentation is to arrange the tables so that the 
variables and levels of variables included will reveal the presence 
or absence of associations that are pertinent to the research 
questions.  Test-statistics calculated from such tables may 
indicate probabilities of chance occurrence of apparent 
associations, but references in reports to values of X2 (Chi 
squared) or probability levels are not adequate substitutes for 
systematic documentation of the observed frequencies themselves.     

6.3.5.  Independent and related data            
    Epidemiological investigations often involve repeated 
observations on the same individuals at different times (e.g., 
replicated measurements of lung function in a cross-sectional 
survey; assessments of acute effects of temporary exposure to 
pollutants; follow-up surveys in longitudinal studies).  Whatever 
the interval between the measurements, be it minutes, days, or 
years, it is usually quite unrealistic to suppose that the sets 
of observations corresponding to the different times are 
"statistically independent"a.  By definition, the sets of data 
a Two events are statistically independent if the occurrence of one 
  of them does not affect the probability that the other may occur.  
  Many widely-used statistical procedures are based on the 
  assumption that the individual items of data under examination 
  are independent.  Conclusions based on such methods, when the 
  assumption is not justified, can be seriously in error.

are related, and, in practice, they often show clear patterns of 
association.  The corresponding frequency distributions should 
therefore be presented in tabular or graphical form, using methods 
appropriate for associated data (section 6.3.2). 

    The same distinction between independent and related data is 
relevant to many case-control (retrospective) studies, if the 
individual cases are "matched" with controls.  Results referring to 
individuals within a matched pair (for instance, the levels of 
exposure to a pollutant) are then "related" in a statistical sense, 
by definition, and it is often helpful to reflect this design-
determined fact in the tables that describe the results. 

6.3.6.  General points on tables and graphs

    Choice of statistical methods for the analysis of data is 
affected in an important way according to whether the variables are 
discrete or continuous, and whether the data are related or 
statistically independent.  It is desirable, therefore, that the 
format of data-description aids such as tables and graphs, reflect 
both these factors.  Otherwise, perusal of the summarized material 
may confuse the issue and invite false conclusions.  Ideally, 
graphs and tables should be interpretable without reference to any 
accompanying text.  This requires that care is taken to ensure that 
captions and legends are informative and unambiguous in describing 
what the data displayed represent.  Axes of graphs must be labelled 
clearly giving the correct units.  Explanatory footnotes may be 
added, if necessary, to aid easy assimilation of the results 
displayed.  However, the addition of a narrative verbal text 
describing the material will usually be necessary, and this should 
include references to summary statistics and indices of morbidity 
and mortality. 

6.3.7.  Summary statistics and indicesa          
    Care should be taken to select the most appropriate measure of 
average tendency to indicate the approximate location of the       
observations on the scale of measurement used.  The  arithmetic mean
(or just simply the "mean") is usually the most informative 
statistic for this purpose, if the observations are on a continuous 
scale, and if they are distributed more or less symmetrically on 
either side of the mean.  Happily (for statisticians), this is 
frequently true, but whether or not it is so in any particular 
instance can only be determined by examining the observed 
distribution itself, preferably in a graphical form. 

a Most books on statistics, including some of those mentioned in 
  the list of references, give formal definitions and derivations 
  of various summary statistics mentioned in this section.  These 
  are not necessarily reproduced in this short guide on when and 
  how the statistics should be used.                               

    Marked lack of symmetry, for instance, a long "tail" on one side 
of the distribution, would justify supplementing a reference to the 
mean by quoting also the  median, i.e., the value on the scale that
splits the number of observations contributing to the distribution  
into two equal halves.  A "bump" in the tail, be it large or 
small, is a warning to re-examine how the data were collected and 
processed before proceeding with description.  Such  bimodality may
indicate that the observations refer to two or more fundamentally 
different types of situation, and it is usually wise to try at least 
to explain or, if possible, to disentangle the mixture. 

    Distributions of measurements of atmospheric pollution are 
often  positively skewed:  most of the observations fall, perhaps 
symmetrically, within a relatively small range, but a proportion 
are distributed with decreasing frequency at considerably higher 
values (Fig. 6.1b).  In this situation, the median will give a good 
idea of about where on the scale most of the data are located, 
while the higher value of the mean signals both the presence of 
assymetry and its direction. 

    An alternative useful index of central tendency for positively-
skewed distribution is the  geometric meana:  the nth-root of
the products of all (n) observations.  Reference to a geometric mean 
should alert the reader to a likely positive skew, and this measure 
will then provide a better indication of where most of the 
observations are located than would the arithmetic mean. 

    When describing averages of discrete distributions (for 
instance, the number of persons, or the number of smokers in each of 
a series of households) reference to the mean ("the mean number of 
smokers was 2.7...") may be misleading if it is quoted on its own.  
The most frequently occurring number of persons observed (the  mode) 
is usually more acceptable and at least as informative a guide to    
the facts, provided that it is supplemented with some other          
information to indicate how typical the mode really is.              

a The geometric mean is numerically equivalent to the antilogarithm 
  of the arithmetic mean of the logarithms of the observations.  
  The logarithmic transformation of positively-skewed data is often 
  used to achieve approximate symmetry prior to the use of 
  statistical procedures which appeal, in part, to arguments based 
  on assumed symmetry, or to represent what is in essence a                 
  curvilinear relationship between variables by an equation for a 
  straight line.  If logarithmic scales are used in data 
  description to illustrate such features (for instance, in graphs 
  of cumulative frequency distributions or in scattergrams), then 
  particular care should be taken to draw attention conspicuously 
  to the transformation that has been used.  Otherwise, the table 
  or graph may give a totally misleading impression of the pattern 
  of the raw data.                                                   

                                                                Scatter (or dispersion)

    Questions concerning the range of the observations, such as 
whether there was a sizeable proportion of households where the 
number seen was very different from the mode (or mean), can be 
anticipated and answered by mentioning an appropriate percentile of 
the distribution.  The pth percentile is the value that is
exceeded by p% of the observations (so that the 50th percentile is 
the same as the median). 

    Measures of dispersion, including the range and percentiles, 
are equally important when describing distributions on a continuous 
scale.  The statistical properties of the standard deviation of the 
Gaussian ("Normal") frequency function are well documenteda.  
This fact determines that reference to an estimate of the Standard 
Deviation (SD) from a sample of data, which are distributed 
approximately normally, provides a powerful and easily 
interpretable measure of dispersion.  If the observed data do mimic 
the pattern defined by the "Normal" frequency curve, more or less, 
then about two-thirds of all the observations will have occurred in 
a range stretching from one SD below the mean to one SD above it.  
Moreover, the mean will then be very close to the median and to the 
mode.  However, the ease with which pocket calculators (to say 
nothing of computers) can generate SDs from large amounts of data 
must not be allowed to obscure the fact that all or some of the 
useful properties described above may be absent, if the observed 
distribution deviates seriously from the normal function.  Summary 
statistics supplement but never substitute for careful tabulation 
and graphical description of data. 

    Sometimes, the scatter within grouped subsets of data increases 
systematically with the mean values for the different subsets.  For 
instance, the variability of somatic effects of a pollutant will 
generally be higher in groups exposed to high levels of the 
pollutant than in groups exposed to low levels.  If the standard 
deviation is approximately proportional to the mean then the ratio 
of these two measures (SD/mean), called the  coefficient of 
 variation, which is usually expressed as a percentage, is a useful 
summary statistic, because it will be more or less constant over the 
whole range of the data.  Conversely, reference to coefficients of 
variation on their own, for different sets of data, when the SD  
is not proportional to the mean can be misleading; one value may 
be relatively high, either because the SD is relatively high, or 
because the mean is relatively low (or both).  Unlike the SD, the 
coefficient of variation is a dimensionless quantity.  It can 
therefore be used to compare variability in sets of data measured 
in different units (e.g., particle-count and mass concentrations of 
particulate matter). 

a A frequency function is a mathematical specification of a curve 
  that describes a theoretical distribution of frequencies.  The 
  Gaussian "Normal" equation generates the familiar, symmetrical, 
  bell-shaped Gaussian "Normal" curve.      Morbidity and mortality indices

    The earliest indices of community health were derived using 
death certificates originally introduced to serve as legal 
documents.  As the limitations of these basic records became 
recognized and interest developed in disease and disability, 
epidemiologists turned their attention to alternative sources of 
information such as hospital and general practice records as well as 
morbidity surveys and morbidity registers (e.g., the National Cancer 
Register in the United Kingdom).  This change was not straight-
forward.  Mortality indices described absolute events occurring
at single points in time; morbidity indices on the other hand, were 
needed to summarize periods of ill-health or disability of varying 
severity.  Some of these conditions might result in deaths while 
others would be followed by complete recovery. 

     Mortality measures the numbers of people dying in a defined
 population in a defined period of time.  In most circumstances
mortality can be expressed simply as the number dying per 100 000 
population in one year.  Actuaries have used this index, referred to 
as the central death rate, to calculate the probability of surviving 
(or dying) from one age to the next. 

    Two indices used to describe morbidity; incidence and 
prevalence have been defined in section 4.2.2. 

    When a number of different morbidity indicators (symptoms, 
signs, lung function, radiological changes, etc.) are used, some 
individuals will demonstrate morbidity on more than one scale.  Care 
should therefore be taken to specify exactly what is being measured 
and how the various aspects of morbidity overlap. 

    In most studies of mortality and morbidity, measures of the 
burden (the numbers of deaths or cases of disease and disability) 
can be converted to rates by relating them to independent estimates 
of the size of the population to which the burden refers.  Thus, the 
total number of deaths in a country in a particular year may be 
divided by an estimate of the mid-year population to give a  crude 
 death rate for the country.  As record linkage has developed and
epidemiologists have started to monitor the experiences of 
individuals in well-defined groups, more precise measures of the 
population at risk have been defined by taking into account the 
changing ages of individuals as they are observed over time.  This 
new denominator, the  person-years-at-risk, is more useful than an 
estimate of the mid-year population, particularly when the age 
structure of the population may be changing rapidly with time.  
Such is the case, for example, in studies of cancer survival, when 
death rates in the first few months after diagnosis are particularly 
high (section  Standardization

    The principal determinant of mortality is age; at the ages of 85 
years and over, death rates are more than 100 times those at ages 
35-44 years.  It is essential, therefore, that comparisons between 

populations should take into account differences in the age-
structure of the populations.  The most effective approach is to 
make comparisons within age-groups by considering  age-specific death 
 rates; the narrower the age-group the more precise the comparison. 

    However, in many situations, particularly when a number of 
comparisons are required, it is not practicable to compare age-
specific death rates.  A summary statistic is needed.  A useful form 
of summarization is by age-adjustment (or age-standardization).  The 
two most common approaches are indirect and direct standardization.  
The corresponding mortality indices derived from these techniques 
are  Standardized Mortality Ratios (SMR) and  Comparative Mortality
 Figures (CMF) respectively, but the same principles can be used
for the construction of standardized morbidity indices.  For 
instance, McLintock et al. (1971) used indirect standardization, to 
adjust for differences in age and in the profusion of small 
opacities on chest radiographs, for a study on regional variations 
in the attack rate of progressive massive lung fibrosis in British 
coalminers.  Fleiss (1981) reviews both these and other methods for 
standardization and discusses some further examples. 

    Indirect standardization of mortality data answers the question 
"how many deaths would be expected if the study population were 
subject to some standard death rates"?  "Expected deaths" are 
calculated as: 
Expected deaths = sigma  [ (study population) x (standard death rates) ]
                   age   [ ( in age group i )   (  in age group i    ) ]

and the SMR is defined as:

    observed deaths
    ---------------  x 100
    expected deaths

    Direct standardization answers the question "what would be the 
death rate of the standard population, if it had experienced the 
study population's age-specific death rate?"  The approach is to 
calculate first the total equivalent deaths in the standard 

Total equi-  = sigma  [ (standard population) x (death rate for study) ]
valent deaths   age   [ (   in age group i  )   (group in age group i) ]

The age-adjusted death rate is then defined as:

     total equivalent deaths
    total standard population

and the CMF as:

    age-adjusted death rate for study group
    ----------------------------------------   x 100
    crude death rate for standard population

    total equivalent deaths for study group
    ---------------------------------------    x 100
    total deaths for standard population

    In general, the SMR and CMF are similar, numerically.  The two 
factors that contribute to differences between them are the age-
specific mortality ratio and the age-specific population 
distribution.  For the SMR and CMF to differ appreciably, the 
mortality rates must vary with age, and the population distribution 
by age for the two groups must also differ materially. 

    One advantage of the SMR compared with the CMF is that 
calculation of an SMR does not require knowledge of ages at death 
in the study group; all that is required is the age distribution of 
those at risk of death, and, of course, the corresponding standard 
death rates.  Calculation of SMRs by hand for a series of study 
groups or subgroups, but using the same standard death rates, is also 
easier than calculation of the corresponding CMFs, and this may be an 
important consideration, if electronic computing aids are not 
readily available. 

    In practice, the SMR is used mainly in occupational or other 
cohort studies, when the number of deaths observed are small 
relative to the size of the population being studied.  The CMF is 
helpful, when comparing national and regional statistics and trends 
over time.  Some statisticians argue that the CMF may be preferred 
also for summarizing data from prospective studies, if the main aim 
is to make comparisons between subgroups within the population being 
investigated, or if the length of follow-up results in relatively 
high observed age-specific death rates.  Proportional mortality

    Often the main interest in an epidemiological investigation is 
the suspected prominence of a particular cause of death, rather than 
overall mortality.  A disproportionately high number of deaths 
attributed to a particular cause may then be summarized as, the 
ratio of the fraction of all deaths attributed to that cause in the 
study group to a similar fraction in the control group or standard 
population, with which it is desired to make the comparison.  The 
resulting index, the  Proportional Mortality Ratio (PMR), has the 
advantage of simplicity:  in its crudest form neither the age 
distribution of those at risk nor the ages at death are required to 
calculate it.  But great care is required when trying to interpret 
the significance of such a ratio, particularly if the age range of 
those being studied is wide.  An unusually high proportion of 
deaths from a particular cause may be because of an unusually high 
number of people at risk in an age-group, where the cause-specific 
death rate is high in any case, even in the standard population. 

    One way in which gross anomalies of this kind can be avoided, 
even when the age distribution of those "at risk" is not known, is 
to use the distribution of ages at death to calculate the number of 
deaths due to a particular cause C that might be "expected", if the 
age-specific fractions of deaths due to C had been the same as in 

the standard (referent) population.  Note that the definition of 
"expected" here differs from that used to calculate the SMR.  The 
"expected" number of deaths due to cause C is: 

EC  =   sigma  [ ( fraction of deaths )   ( number of deaths   ) ]
        age at [ (due to C in referent) x ( from all causes in ) ]
        death  [ ( population at age i)   (study group at age i) ]

The age-standardized proportional mortality ratio for cause C
is then taken as the total number of deaths attributed to C
that have been observed, expressed as a percentage of those

                         (observed deaths due to C)
    Standardized PMRC  = --------------------------  x 100

    An even more useful way of studying proportional mortality
is to compare the observed fraction of deaths attributed to a
cause C with the fraction that would be expected if the age-
and cause-specific death rates of the referent population had
been experienced:

         (observed deaths due to C)
           (all observed deaths)
    -------------------------------------  x 100
       (expected deaths due to cause C)
      (expected deaths from all causes)

    This proportional analogue to the SMR has been referred to as a 
Relative Standardized Mortality Ratio (RSMR), because of its 
numerical equivalence to the SMR for the cause of interest divided 
by the SMR for all deaths: 

                SMR for cause C
     RSMRC   = ------------------  x 100
               SMR for all causes

    The importance of the RSMR is that it may exceed 100%, 
indicating an excess proportion of deaths due to C in the study 
population, even when the cause-specific SMR (SMRC) is similar to 
or perhaps less severe than in the referent population.  This would 
imply that the SMR for all causes is well below 100%, indicating a 
favourable overall mortality compared with the referent population.  
This situation is met frequently in occupational health studies, 
because of the selection effects common in such studies.  Kupper et 
al. (1978) discussed the theoretical relationship between the RSMR 
and the standardized PMR, and showed that, in practice, the latter 
may be a good approximation to the RSMR.  Relative risk and attributable risk

    The ratio of comparative mortality figures (CMFs) from two 
groups (when both CMFs are based on the same reference population) 
is equivalent algebraically to the ratio of the corresponding age-
adjusted death rates.  The quotient is therefore a direct measure of 
the relative risks of death in the two groups.  Similarly, the 
relative risks of disease in two groups may be expressed as the 
ratio of the appropriate (directly) age-standardized disease 
incidence rates. 

    But incidence rates can only be measured in follow-up studies.  
Prevalence rates, from cross-sectional surveys of large groups, are 
only indirect reflections of disease risks, because prevalence 
depends not only on the incidence of disease over a period of time, 
but also on how long those with disease remain in the group being 
studied.  The prevalence rate of cases (with disease) in a group 
defined for a case-control study does not provide any measure of
the disease risk for that group, because the magnitude of such a 
prevalence rate depends on an arbitrary choice of how many controls 
are included in the study. 

    Nevertheless, an approximate measure of relative risk for two 
groups, A and B, can be obtained, both from cross-sectional and from 
case-control studies, by comparing the odds favouring the occurrence 
of the disease in the two groups: 

                  (number with disease in group A)
                 (number without disease in group A)
Odds ratio  = ----------------------------------------
                  (number with disease in group B)
                 (number without disease in group B)

    This quotient is a useful index of the extent to which the 
occurrence of disease is associated with membership of one or other 
group.  If the true incidence rates of the disease concerned are 
small in both groups, then the  odds ratio approximates closely to
the ratio of the incidence rates themselves - the true relative
risk.  Of course, differences in the age distributions of the two
groups may seriously distort such a comparison, just as they would
when considering any other crude average.  Techniques analogous to
age-standardization (or standardization for any other variable,
such as smoking) can be used to calculate an adjusted  Summary 
 Relative Risk from data relating to appropriate subgroups (Mantel &
Haenszel, 1959).  These ideas are developed further in section 6.4.5. 

    Sometimes, it is necessary to make a quantitative assessment of 
the likely impact of preventive measures on the future incidence of 
disease.  It then becomes important to try to estimate how much of 
the total incidence of the disease in a community is attributable
to a risk factor under consideration.  If the incidence rates for 
the whole community and for that part of it which has been exposed 
to risk are known, then the difference between these rates provides 
an obvious measure of attributable risk.  This may be expressed as
a proportion of the total risk in the community, that is, a 

population  Attributable Risk Ratio.  But, as with relative risks, 
diffiulties arise when incidence rates are not known.  However, if 
at least a reliable estimate of the proportion of those in the 
community who are exposed to the risk factor (chi), and a reliable 
estimate (from the odds ratio) of the risk for those exposed 
relative to those not exposed (r) are available, then the 
population Attributable Risk Ratio may be approximated by 
[chi(r-1)]/[1+chi(r-1)].  For further details of these and other 
approximations see Walter (1976) and Leung & Kupper (1981).  Concluding remarks about summary statistics and indices

    It is important to recall that summary statistics were 
introduced in section 6.3.6 as supplements to data description, not 
as substitutes for that activity.  Any one summary statistic cannot 
do more than reflect one aspect of the results, and reference to 
that single aspect is unlikely to provide a convincing answer to 
even the simplest research question. 

    The morbidity and mortality indices that have been mentioned are 
measures of average tendencies and, as such, references to them 
should generally be qualified by an indication of the dispersion of 
the results on which they are based.  This may be achieved by 
tabular or graphical representations as discussed above, or by 
quoting statistics that summarize the scatter (section  A 
further extremely important way of describing the scatter associated 
with an average is to derive a measure of the variability that would 
be expected in that average, if the experiment, or survey giving 
rise to it were repeated a large number of times, that is, the 
 Standard Error (SE) of the average concerned.  Calculation of the
SE from the data leads directly to statistical inference including 
the formal testing of hypotheses, and these problems are discussed 
in the sections that follow. 

6.4.  Analysis and Interpretation

6.4.1.  Statistical ideas about the interpretation of data

    In some situations, a careful description of results from an 
epidemiological study may be enough, or almost enough, to satisfy 
the main research objectives.  More usually, perusal of the data 
descriptions leads to questions.  Is it reasonable to conclude that 
the apparent differences, trends, associations, or correlations, 
which have been observed, really reflect the effects of the 
explanatory variables hypothesized?  How easily could the results 
have arisen by chance?  What is the best estimate of the likely 
effect of a particular noxious agent, other things (smoking habits, 
age, physique, social factors, etc.) being equal?  What degree of 
confidence can be placed in the estimate? 

    These questions reflect the uncertainty associated with many 
experimental and observational settings, particularly those 
involving living organisms.  The uncertainty stems from the 
multiplicity of factors that may influence a particular outcome, 
such as the onset or cure of disease in an individual.  The outcome 

for the individual is not precisely predictable; yet a pattern is 
thought to exist and may be discernible, if enough observations are 
made on a number of persons and over a sufficiently long period.  
This type of situation is described by statisticians as a  random 

    Applied statisticians study data generated from random systems 
with the aim of quantifying the inherent uncertainty in the 
observations and teasing-out patterns that may not be immediately 
obvious.  The procedure, quite generally, is to construct an 
idealized mathematical representation (or model) of the system being 
studied and then to examine the degree to which the model conforms 
to the observed data.  Statistical models are distinguished from 
other mathematical constructions used in science by the fact that 
they always include a term, explicit or implicit, to symbolize 
variations that are not due to the factors being studied, that is 
the randomness in the system.  The aim is to characterize the 
pattern of random variation, to quantify it, and to use the 
estimated magnitude of the so-called "random error" to qualify 
statements about the factors and effects being studied.  The effects 
(e.g., disease incidence, mortality, symptom prevalence) associated 
with particular factors (e.g., exposure to a pollutant, membership 
of a social group) are estimated (so-called "point estimates").  The 
probable ranges within which the estimate might be expected to fall, 
if the study were repeated a large number of times, may also be 
determined ("interval estimation").  The estimated random error is 
also used frequently as the basis for making judgements about the 
 statistical significance of effects, apparently associated with
factors in the model. 

    The percentage level of statistical significance is an inverse 
index of how likely it is that an observed effect is due to chance.  
Reference to a 5% level of significance, for instance, is 
equivalent to stating that the probability of observing a result as 
extreme or more extreme than that recorded, purely by chance, is 
less than five in one hundred ( P < 0.05).  This particular level 
of significance is usually interpreted as some evidence that the 
effect concerned is not due to chance.  A lower significance level,
for instance,  P < 0.01, might be described as fairly strong 
evidence that the result is not due to chance.  Most people would 
regard  P < 0.001 as overwhelming evidence against chance 

    The ubiquity of significance testing in epidemiology justifies 
a brief restatement here of three important riders; 

(a) Significance testing cannot prove that an effect is real.
    A one-in-thousand chance does occur occasionally - by
    chance (about once in a thousand times, in fact).

(b) If the probability of chance occurrence is, say, less than
    1%, this does not mean that the probability that the
    effect is real is 99% or more.

(c) "Not significant" is not to be misinterpreted as "not
    real" or "not important".  The absence of a statistically
    significant result means only that the data concerned are
    consistent with chance variation.

6.4.2.  Errors

    The riders mentioned above draw attention to different types of 
error that may occur, when accepting results from significance 
tests.  The situation described in rider (a) - a chance event 
occurs in practice, although the odds are fairly heavily against it 
occurring - implies that an investigator who accepts the evidence 
from the test at its face value is accepting an error:  the so-
called  Type I error.  No statistical procedure can determine
whether or not the error has been made; but, if the test suggests
significance at, say, the alpha % level, then the probability of
making the Type I error is quantified:  it is less than alpha %.
If the results are really due to chance, and the statistical test
also suggests that they are not significant at the alpha % level
(P > alpha), then the probability that an investigator is right in
accepting the implications of the test is at least (1-alpha); i.e.,
(1-alpha) is the probability of not making the Type I error.  
Neither the level of significance (alpha) nor its complement (1-
alpha) measures the probability that an observed effect is real (see 
rider (b). 

    On the other hand, it is possible that a real effect does exist 
and yet the survey results may not provide any convincing evidence 
in support of the reality, possibly because the effect is small, or 
because not enough data are available.  The absence of a 
statistically significant result in this situation may persuade the 
investigator to accept the  Type II error:  accepting the 
hypothesis of "no effect" (the "null hypothesis") when in fact it 
is not true.  Again, no statistical procedure can determine whether 
or not such an error has been made, but the probability of making 
it (beta) can be controlled by adequate design of the study.  
Fleiss (1981) provides useful tables to help answer the familiar 
question:  "how many observations are required ...?" to achieve the 
required statistical  power (1-beta) consistent with various 
significance levels (alpha) and specified differences in 

    It is the randomness in the system that gives rise to the 
existence of Type I and Type II errors; the possibility of making 
them cannot be avoided, but the probability of their occurrence can 
be measured.  A third more serious type of error is to ignore the 
existence of alpha or beta or both, and to rely entirely on 
intuition when interpreting epidemiological data; this error is 

    Results from epidemiological surveys usually provide a large 
number of possible contrasts and comparisons.  For instance, the 
data may be divided into subsets corresponding to the two sexes, to 
exposure types or categories, etc.  A well thought-out survey plan, 
designed to answer specific questions, will have anticipated the 

main contrasts that are likely to be of interest.  Data collection 
and data description will have been organized in such a way that 
these contrasts can be studied sensibly.  But an alert eye must be 
kept for the unusual and the unexpected, and tests of significance 
in this situation are not necessarily straightforward. 

    In particular, unplanned division of the data into subsets, 
with no specific research question in mind, sometimes leads to the 
question of which, if any, of the observed differences between 
subgroups are unlikely to be due to chance.  The answer cannot be 
determined simply by repeated tests of significance on all possible 
combinations of subgroups, or by selecting for test those contrasts 
that appear to be large.  Apparent significance levels from such 
repeated tests are not sound estimates of probabilities of making 
Type I errorsa.   Multiple comparisons of this kind require special 
procedures that are described in many statistics texts, usually in 
the context of analysis of variance (section  The first 
step in these procedures is to put the question in a more general 
form, not specific to any particular subgroup, i.e., whether 
evidence from the data that the dispersion between the grouped 
results is anything but random.  Only if the answer is that there 
is indeed evidence that the dispersion is not entirely random, is 
it necessary to proceed with further modified tests to determine 
which of the contrasts show significant differences.  For instance, 
the Registrar General of England and Wales (Registrar General, 
1978) discussed multiple comparisons of SMRs in large-scale 
mortality studies. 

    The instruments used in an epidemiological survey may generate 
errors in the observations that are analogous to the Type I and    
Type II errors familiar from the theory of significance testing.   
Suppose that circumstances dictate that it is not possible to use  
sophisticated clinical laboratory equipment in the field, and that 
a simpler, but necessarily cruder, instrument (or questionnaire) is
used to classify subjects into the dichotomy diseased/not-diseased.
Suppose also that a pilot study has been conducted on a sample 
consisting of N subjects, all of whom have been examined by both 
methods, and that the results are as shown in Table 6.3. 

    A fraction [b/(b+d)] of those with no disease is classified      
wrongly as having the disease ("false positives", compare the        
probability of making the Type I error, alpha). [1-b/(b+d)] = 
d/(b+d) is known as the  specificity of the simpler test.             

a For instance, 15 subgroups would generate 105 possible contrasts 
  between pairs of subgroups.  It would not be surprising to find 
  that about five of these contrasts appear to be "statistically 
  significant at the 5% level" even if the 15 subsets of results 
  were effectively random samples from one homogeneous set of data 
  (i.e., there are no real differences between subgroups - only 
  chance variations).                                                           

Table 6.3.  T R U E   S T A T U S                          
(based on full clinical investigation)                                   
                        With disease  Without disease  Total          
               With     a             b                a + B          
Result         disease                                                    
from simpler   ----------------------------------------------------
test           Without  c             d                c + d          
               Total    a+c           b+d              N = a+b+c+d    

    Another fraction [c/(a+c)] is classified wrongly as having no 
disease ("false negatives" compare the probability of making the 
Type II error, beta).  [1-c/(a+c)] = a/(a+c) is known as the 
 sensitivity of the simpler test. 

    The hypothetical pilot study referred to above might be 
extended to help distinguish between two other types of error that 
may be incurred in the survey procedures.  Suppose that not just 
one, but repeated measurements are made with the simpler (survey) 
instrument on all N subjects involved in the pilot study.  The 
results from any one subject may be very variable (a high 
coefficient of variation perhaps) but they may average out to a 
value very close to the "true" value for that subject, as 
determined from the measurement on the more sophisticated 
equipment.  In this event, measurements using the simpler 
instrument are said to be very imprecise  but unbiased. 

    However, it may be that the simpler instrument gives very 
precise results for any one subject (a low coefficient or 
variation), but averages of results referring to individual 
subjects may differ systematically from the true values of the 
individuals.  The instrument and the measurements are said to be 
biased, although they are precise.  It is important to distinguish 
between lack of precision and bias.  In principle, the precision of 
an average of repeated measurements can always be increased (at a 
cost) by making more measurements.  Bias however is usually more 
difficult to detect or to correct.  Properly conducted pilot 
studies should include efforts to quantify these possible errors 
(section 5.6.6). 

    The various terms concerning the characteristics of results of 
measurements, such as precision, accuracy, and validity, have been 
discussed in sections 3.5.1 and 4.1.2.  It is clear that errors 
arising from the way that observations are made, from the 
limitations of the instruments themselves, and from the inherent 
variability in the (random) biological system being studied, all 
contribute to the total uncertainty that is associated with 
epidemiological results.  It is worth recalling, therefore, the 
references in Chapters 2 and 5 to the important link between the 
design and conduct of a study and the subsequent analysis of 
results.  The more carefully a study has been designed, the easier 

it is to postulate realistic models and to derive sensible 
conclusions from the observations.  Conversely, it is generally 
true that a poorly designed or conducted study is unlikely to yield 
reliable results, however ingenious the statistical analysis. 

    The remainder of this section will refer to some of the models 
that may be applicable to results from some of the study designs 
discussed in Chapter 2.  The statistical principles involved are 
emphasized, without details of statistical theory and methods.  
Recent developments in statistical thinking that are particularly 
relevant to the analysis of epidemiological data are discussed, 
although some of the ideas mentioned are still controversial. 

6.4.3.  Analysing results from cross-sectional studies  Qualitative data

     Cross-sectional studies generate results that refer to a
single point in time (or to a relatively short interval).  The 
effects observed in such a study are, of course, influenced by time 
to a major extent (e.g., the ages of those studied, the length of 
prior exposure to a pollutant, the latency period for the occurrence 
of disease), but technical complications, which arise when 
considering repeated measurements on the same individuals at 
different times, are absent.  Cross-sectional survey results are 
therefore a convenient starting point for a discussion on how to 
interpret data that have been described previously by the methods 
outlined in section 6.3. 

    In the first instance the qualitative data are considered, that 
is, the classified frequencies with respect to discrete variables in 
the contingency tables that were mentioned in section 6.3.4.  The 
variations in numbers recorded in the corresponding cells of the 
data description tables will generate questions.  For instance, is 
the relative frequency of persons with disease, in the sample 
studied, a useful indication of the true prevalence in the 
"population" from which the sample has been drawn? 

    The above enclosure in quotation marks of the word "population" 
is to draw attention to a difference between the use of the word in 
a statistical context, and the more general meaning of the word 
population as used also by epidemiologists.  For instance, a survey 
with a 100% response-rate of all people in a village exposed to a 
particular pollutant may provide an authoritative picture of disease 
prevalence in that community, or population in the sense that the 
word is used by epidemiologists; this is important in its own right.  
However, the most important aspect of the results from the survey is 
likely to be the information that it provides on the probable 
prevalence of the disease in other groups who may be exposed to the 
same pollutant.  From this point of view, the group studied in one 
village is but a  sample from a larger statistical population of
similar communities that have not been studied.  And it is this 
population of villages, or groups, that has not been surveyed, (and 
may not even have been identified), which probably lies at the 
centre of the research question that stimulated the survey in the 
first place. 

    How good then is the sample estimate of the population 
prevalence?  A quantitative answer to this question will require an 
appeal to statistical theory about binomial responses.  It will 
depend partly on the size of the sample that has been studied, and 
it may be expressed either as a  standard error of the sample
estimate of prevalence, or as  confidence limits for the estimate.
In many situations, these measures are directly related, in that an 
interval measuring approximately twice the standard error on either 
side of the estimate will approximate closely to the 95% confidence 
interval.  This is the range of figures likely to embrace 95% of the 
prevalence results in a hypothetical long series of repetitions of 
surveys, of the kind that have produced the sample estimate.  
Provided that the group studied is representative of the population 
concerned, then the best available (point) estimate of the true 
population prevalence will be that found in the survey.  The 
variability of the estimate can be quantified using probability 

    Similar principles will be used, when exploring hypothesized 
differences between groups that apparently exhibit different levels 
of prevalence.  To what extent are those differences attributable to 
the fact that the individual results are only sample estimates of 
the true corresponding population prevalences?  How likely is it 
that the differences observed are due to chance?  The extension of 
these ideas to analyses of results with more than two levels of a 
discrete variable (i.e., not just disease/no disease, the so-called 
binary responses, but possibly disease A/disease B/no disease), and 
to the investigation of possible associations between discrete 
variables (e.g., occurrence of disease and residence in a particular 
area) is often pursued using the well-known x2 test of association.  
The principle of this test is to divide the data into appropriate 
subsets corresponding to cells in a contingency table (section 
6.3.4).  The numbers of observations that might be expected in each 
cell, under the (null) hypothesis of no association, are calculated 
in proportion to the observed frequencies in the marginal 
distributions.  The sum of terms (one for each cell) each of the 

[(number observed - number expected)2/(number expected)]

constitutes the test statistic probability distribution of which, 
under the null hypothesis, is approximately the same as that of a 
mathematical function known as x2.  The x2 distribution varies, 
depending on the number of independenta terms (degrees of freedom) 
contributing to it, and this number is always less than the number 
of cells in the corresponding contingency table.  A table 
consisting of  r rows and  c columns contributes ( r-1) x ( c-1) 
degrees of freedom to the test statistic.  Note that the test 
statistic must be constructed by operating on numbers of 
observations (observed and expected), not on proportions, 
percentages, or measurements on a continuous scale.  The text by 
Maxwell (1961) provides a particularly helpful and concise guide to 
these methods of analysis.  The statistical study of rates and 
a See footnote to section 6.3.5.                                 

proportions is dealt with in a systematic and comprehensive way by 
Fleiss (1981) in a book that includes many examples of 
epidemiological applications. 

    Different categories of a discrete variable are sometimes 
arranged naturally into a distinct order; for instance, low, 
medium, and high exposure to a pollutant that has not been measured 
precisely; or categories representing the increasing severity of a 
symptom.  Generation of such  ordered categorical data is often
characterized by a subjective element, which usually implies a 
relatively high level of variability between different observers in 
their assessments of the same phenomena - observer error.  The data 
themselves are quasi-continuous, in that changes from one category  
to the next reflect an underlying continuum from low to high, or    
small to large; but the difficulty of determining precisely where on
the continuum a particular observation should be placed determines  
that the measurements are expressed as discrete categories.  Some of
the methods available for analysing data arranged in ordered     
categories have been reviewed by Jacobsen (1975a), who gives     
examples from classifications of an increasing profusion of small
shadows on chest radiographs.                                    

    Methods for investigating associations between binary responses  
and several other variables, and for estimating the separate effects 
attributable to these variables, are reviewed, explained, and        
extended in a classical monograph by Cox (1970).  Some of the ideas  
used by Cox are analogous to those applied to the study of continuous
variables that generate quantitative data.                     Quantitative data: response and explanatory variables

    Physiology, chemistry, and biochemistry, quantitative pathology, 
and anthropometry may furnish epidemiologists with biological end-   
points to relate to environmental factors.  These endpoints of       
function, concentration, or size are normally recorded as            
continuous variables, yielding "quantitative data".  Moreover, even  
when the data recorded are qualitative, for instance the numbers of  
deaths occurring in different groups, their analysis may sometimes   
be effected more efficiently by expressing the numbers as fractions  
or rates (for instance the death rates in the various groups).       
Either the fractions themselves, or one of several possible       
mathematical transformations of the fractions, may then be treated
as quantitative data, as if they were measurements on a continuous

    Statistical models involve two broad types of variables, both 
of which may include qualitative and quantitative data.  The first 
type refers to variations that are posited as the consequence of 
some other event or events; these are the so-called response or 
dependent variables, which are also referred to as regressands in 
the context of regression analysis.  Then there are the so-called 
explanatory, independent, or regressor variables, namely, those 
that are represented in the model as being responsible for some 

part of the variation in the response variablesa.  Mortality and 
different indices of morbidity including, for example, the level of 
lung function, the blood level of a pollutant, the appearance of a 
chest radiograph, or the occurrence of symptoms, are typical 
response variables in an epidemiological setting.  Age, nutrition, 
social conditions, smoking habits, and the intensity and length of 
exposure to a pollutant, are examples of observations that are 
usually regarded as explanatory variables, because the way in which 
they vary helps to explain the variation in the response variable 
of interest. 
    It will be clear immediately that what may be regarded         
reasonably as an explanatory variable in one situation may be      
considered as a response variable in another.  For instance, the   
concentration of pollutant in the ambient air is certainly an      
explanatory variable in an analysis of the effect on the health of 
the people exposed to the pollutant.  But the very same levels of  
pollutant may be regarded as responses to variations in climatic   
conditions or the intensity of production of some local factory, or
variations in types of fuel used in the area.                      

    Moreover it is commonplace that some explanatory variables are 
themselves associated, that is, one of them may vary depending on  
the level of another (Table 6.2; section 6.3.3).  Age for instance 
is often correlated with the level of cumulative exposure to a     
pollutant.  Indeed, in the absence of an independent measure of    
exposure, age, or some simple function of age is sometimes used as 
a crude, indirect index of cumulative exposure.                    

    A further complication in the statistical nomenclature for     
variables arises in  case-control or restrospective studies.  The   
cases and controls are defined and identified at the start in      
relation to the presence and absence, respectively, of some morbid 
condition.  The research question is:  do these two groups differ  
also with respect to some antecedent factor, such as exposure to a 
pollutant?  From the statistical point of view, the exposure is now
a response variable; the occurrence of a case or control is the    
explanatory variable.  This apparently curious reversal of the     
labels attached to the same variable is a reflection of the fact   
that the question presented in the case-control study can be      
formulated as follows:  "given these (independently) defined      
groups, cases, and controls, what is the difference in their prior
exposures?".  The corresponding  cross-sectional or prospective    
 design to examine the same epidemiological problem would lead to  
the question:  "given different levels of exposure to a pollutant, 
what is the difference in the prevalence or incidence of the 
conditions of interest?".  The design-determined difference in the 
way that the statistical question is posed, is often reflected in 
the methods used for the statistical analysis.  In general, a 

a In this chapter it is convenient to distinguish between these 
  broad classes of variables by referring to them as response and 
  explanatory variables, respectively.                              

decision as to which is the explanatory and which the response 
variable in a particular setting depends on which data are given at 
the outset; the given information refers to the explanatory variable.  Statisticians and computers

    Most epidemiological studies are observational, not 
experimental.  Normally, the effects of several known or suspected 
explanatory variables must be considered before useful inferences 
can be made concerning relationships of the environment to the 
biological indicators of interest.  Multiple variables must be 
included in the analysis:  hence  multivariate analysis.  Analysis
of variance, and correlation and regression analysis, described in 
any good basic statistics text (Dixon & Massey, 1957, for example) 
are the best known of the multivariate techniques, and the most 
generally useful in environmental epidemiology.  Choice of the 
appropriate multivariate approach and practical implementation of 
the analysis are tasks requiring the professional skills of a 
statistician, who should be qualified and experienced in applying 
mathematical probability concepts to data generated from 
observations on a human population. 

    For large sets of data, particularly when many variables are 
being considered, full exploitation of the methods discussed here 
will usually require access to a computer.  Generally, it will be 
desirable to inspect results from several alternative statistical 
models of the data before attempting to draw conclusions.  When a 
computer is used, it is wise to select a subportion of data for 
manual analysis, using perhaps three variables on 50 subjects, and 
compare results with those on the computer output.  This will serve 
to highlight possible errors in the computer programs themselves or 
in the choice of statistical options of the user provided in general 
purpose packaged software (preprogrammed statistical operations).  
In any case, the additional insight into the data, which will be 
gained by adopting this discipline, is a valuable aid to intelligent 
inspection and interpretation of the computer output. 

    The following discussion concerns multivariate methods in 
environmental epidemiology from three points of view:  first, the 
utility of various techniques; second, general considerations in 
formulating models; and finally, how to evaluate the effectiveness 
of particular models.  Analysis of variance

    Analyses of variance are used to establish the presence of 
statistically significant effects in designed experiments and to 
estimate the effects with appropriate confidence limits.  The 
usefulness of this powerful statistical tool depends largely on the 
ability to control the sequence in which specified combinations of 
different levels of explanatory variables are allowed to occur, and 
this is determined by the experimental, as distinct from 
epidemiological, design.  The method is therefore the mainstay of 
experimentalists, but it has relatively few applications in 

observational sciences (King, 1969)a.  Nevertheless, the analysis 
of variance is used in environmental epidemiology in certain 
circumstances, for instance, to study multiple repeated 
measurements in near-experimental study designs.  An example 
referring to daily changes in lung function over several days may 
be found in the papers by Carey et al. (1967, 1968).  An analysis 
of multiple repeated measurements over three years by McKerrow & 
Rossiter (1968) is of particular interest, yielding evidence of 
linear trends over the study period for individuals. 

    The analysis of variance can be regarded as a special case of   
the more general family of multiple regression models that are used 
extensively in observational studies and that are considered further
below.  The formulation and interpretation of such models in        
environmental epidemiology requires knowledge concerning factors    
affecting fluctuations in both response and explanatory variables.  
These problems may be studied by considering the "components of     
variance" in a set of observations.  Total variability may be       
estimated by combining independent estimates of the components,     
using analysis of variance methods, or by measuring the observed    
(total) variance directly.  Duncan (1959) discussed this important  
topic in detail.  In cross-sectional studies, it is usually         
necessary to have reasonable estimates of what proportion of the    
total variance in a measurement can be ascribed to between-person   
variability, individual or within-person variability, instrument,  
or laboratory variability, when relevant, and residual variance.    

    Another epidemiological situation, in which the idea of         
components of variance may be important, is when limited sampling   
equipment and resources have to be allocated to measure the         
concentration of a pollutant (e.g., in air, water, or food supply). 
The problem is how to plan the distribution of the available        
resources to different locations and times so as to maximize the    
precision of exposure estimates that are required for the 
subsequent epidemiological survey.  A solution to the problem may 
usually be found from a pilot study designed to yield estimates of 
the appropriate components of variance (between general locations,      
specific places within the general locations, different time        
intervals, instruments, staff using the instruments, etc.).  Such a 
preliminary study may have quasi-experimental features suited to a  
formal analysis of variance, or, otherwise, the components of       
variance may be estimated, but less efficiently, using more general 
regression models.           

    Analysis of single pairs of differences, such as changes in 
ventilatory function before and after a work shift (Lapp et al., 
1972) may be based on paired  t-tests.  "Before" and "after"
study designs are extremely useful in environmental epidemiology, 
because short-term decrements in function in response to an 

a The constraints on materials and methods in epidemiology affect 
  not only the statistical approach to the data but also the 
  conclusions that can be drawn, particularly in exposure/effect 
  studies (section 6.4.6).                               

environmental agent may suggest long-term effects.  In general, 
analyses of pairs of observations should be carried out on 
differences in averages of the observations; using the observations 
singly may generate statistical artefacts (due to regression towards 
individual means) and the spurious correlations can be misleading. 
Oldham (1962) and Gardner & Heady (1973) have discussed the problem 
in detail.  Effects of regression toward individual means in 
analyses of results from treatment of groups of patients have been 
studied by Deniston & Rosenstock (1973); their experience is 
relevant in environmental epidemiology when panels of ill subjects 
are being selected for repeated observations over a period of time 
(time series). 

    Within-individual variability often depends on the time interval 
over which the repeated measurements are made.  For instance, 
systematic reduction of lung function within individuals over 
several years has been studied longitudinally (Fletcher et al., 
1976; Love & Miller, 1982) but the variation between individuals in 
the rates of reduction with time is much higher than the variations 
between individuals of the same age, at the same time.  This fact 
points to a very substantial non-systematic (random) within-
individual component of variance over long time periods.  In 
general, variability within individuals from one year to the next is 
greater than that from one month to the next, and so on down to the 
time between immediately repeated tests (Berry, 1974; Lebowitz et 
al., 1982).  Cross-sectional evaluation of any physiological test in 
terms of within-subject variability requires explicit consideration 
of the duration over which variability is of interest.  Correlations

    A  correlation coefficient is an index of the degree of linear 
association between two variables.  It is a dimensionless decimal 
fraction that may vary between -1.0 and +1.0 depending on whether 
one of the variables decreases or whether it increases as the other 
increases.  The danger of misinterpreting numerical estimates of 
such coefficients from samples of data can be reduced by making it a 
rule always to study their significance in the context of the 
corresponding scattergrams (section 6.3.3).  Studying the scatter-
grams will often reveal features in the data that may have 
artefactually inflated or diminished the calculated values. 

    One fairly common and potentially misleading situation is the 
occurrence of a high value of the index in bivariate data consisting 
of well-separated clusters of points that happen to fall on a 
straight line drawn through them (Fig. 6.4a).  The separate clusters 
may correspond to quite different situations, each characterized by 
relatively slight variability about different mean values of one 
variable (say, blood-lead levels as determined at different 
laboratories).  Even quite small but systematic differences in the 
laboratory techniques may then generate superficially impressive but 
 spurious correlation with any explanatory variable, the mean 
levels of which also happen to differ systematically between the 
laboratories in the same (or in the opposite) direction (say some 
aspect of dietary intake among persons resident close to the 

laboratories).  A scattergram will quickly reveal the clustering and 
will also show whether there is any suggestion of a linear trend 
within the clusters.  If there is no such trend within clusters (and 
in the same direction as the trend between clusters), then the 
apparently high value of the correlation coefficient is not to be 
interpreted as an index of linear trend between the variables;  it 
reflects only a linear trend of mean values between clusters.  Good 
data description should have identified the clustering before the 
correlation coefficient was calculated, and the possible reasons why 
it occurred should have been investigated before formal analysis 
(section 6.3.1). 


    Similar spurious correlations may occur, if most of the data are 
clustered quite randomly in one region of the scattergram (say, in 
the bottom-left corner) with a few points in the opposite corner 
(Fig. 6.4b).  A high value of the correlation coefficient in this 
situation does not reveal anything more than is obvious from 
elementary geometry; the shortest distance between two points 
placed in the centroids of the two clusters is a straight line.  In 
observational studies, generally, high coefficients of correlation 
may be interpretable as reflecting real linearity in the 
relationship between the variables considered,  only if the plotted 
 points on the scattergram fall into portions of the graph that are 
 reasonably representative of the ranges of the data concerned. 

    Equal caution is required when considering low, not significant 
values of the coefficient.  The low figure indicates only the 
absence of a linear trend in the data.  But the shape of the 
relationship between two variables may be a curve rather than a 
straight line.  If so, the pattern will often be apparent from the 
scattergram.  Data transformations and regression methods may then 
be used to quantify and study such relationships (section 


    Any set of multivariate data can be used to generate a 
 correlation matrix:  an array of correlation coefficients that 
reflect the degree of linear associations between the variables 
taken in pairs.  Inspection of such a matrix, together with the 
corresponding scattergrams and frequency distributions, is usually 
an indispensable part of the process of searching for the most 
appropriate approach to building statistical models.  But it may be 
seriously misleading to use the magnitudes of such correlations, or 
partial correlations, as indicators of what are the most important 
relationships.  This is because the observed correlations depend 
not only on the underlying true relationship between the variables; 
they depend also on the particular variance structures as observed 
in the study.  The latter are frequently not characteristic of 
other groups or situations.  Using observed correlation 
coefficients to compare relationships is like comparing two overall 
death rates without age-adjustment. 

    These cautions are particularly relevant to a class of 
epidemiological investigations, sometimes referred to as 
"correlation" or "ecologic" studies.  The idea is to examine 
hypothesized associations by considering correlations between 
measures of average tendency (e.g., disease incidence rates, 
mortality incidences, average levels of air contaminants) from 
different communities; but the variability of these features within 
the different communities is not taken into consideration.  Valid 
interpretation of apparent correlations is then extremely difficult 
if not impossible, because variance structures of possible 
explanatory variables (air pollution, weather data, dietary, 
smoking or drinking habits) may differ widely from year-to-year and 
place-to-place.  At best, any correlations that are observed should 
be regarded as stimuli for further research, rather than as a basis 
for conclusions.  A useful discussion of the effect of measurement 
errors in correlation analyses is given by Fleiss & Strout (1977).  Multiple regression

    The following discussion of multiple regression includes some 
points that are also relevant to other multivariate techniques. 

    The essence of a multiple regression model is to hypothesize 
some mathematical function of the explanatory variables which, 
together with a variable representing random fluctuations, will 
provide a useful estimate of the response corresponding to a 
particular combination of values of the explanatory variables.  For 
example, it is known from many studies that, in general, a person's 
FEV1 depends in part on the individual's height (H), age (A), and 
smoking habit (S).  The following simple function usually provides 
a good approximation of the likely value for an individual, 
provided that the coefficients bo, b1 and b3 are known for the 
population of which the individual is a member. 

    FEV1 = bo + b1H + b2A + b3S

The variable S, symbolizing smoking habits, might be expressed in 
various ways:  perhaps the number of cigarettes being smoked per 
day or per week at the time of survey, or some estimate of 
cumulative exposure to tobacco smoke measured in cigarette-pack-
years.  The constant bo can be thought of as some underlying number 
for the population being studied which is modified by the addition 
of terms consisting of the explanatory variables each of which have 
first to be multiplied by appropriate constants. 

    The above equation is a deterministic mathematical model.  The  
corresponding statistical multiple regression model would alter the 
left-hand side of the equation to read "E (FEV1/H,A,S)", 
symbolizing the statement "the expected value of FEV1 given some 
particular values of H, A, and S".  This restatement of the model 
now incorporates the essential statistical concept of 
"expectation", meaning that given enough individual data from the 
same population, all of whom are characterized by identical values 
of H, A, and S, then the equation provides an estimate of the mean 
value of FEV1 that might be observed for those individuals.  It 
follows immediately that the statistical model does not purport to 
predict a precise value of FEV1 for any one individual; it provides 
an estimate of the likely value and it admits that the value 
observed for an individual may vary from that estimate, up or down, 
because of random fluctuations.  This idea can be expressed 
directly by the following alternative way of writing the 
statistical model:         

   Observed FEV1   =  bo + b1H + b2A + b3S + R

This indicates that the result observed for an individual can be 
expressed as the sum of terms corresponding to estimates 
(symbolized by the underscoring) of effects associated with the 
explanatory variables, plus an additional term R representing 
random fluctuations.  The precise value of R, negative or positive, 
is not predictable for an individual, but the regression model 

stipulates that, for a sufficiently large number of individuals, 
the average of the Rs will be zeroa.  The technical problem in a 
multiple regression analysis is to estimate the regression 
coefficients (the bs) and to estimate the variance of R from data.  Additive linear models

    Now, consider how long-term exposure to a pollutant affecting 
the respiratory system might require the model to be modified.  The 
simplest, and sometimes an effective, way of incorporating this new 
explanatory variable is to add a further term, E (symbolizing a 
continuous measure of exposure) and to attempt to estimate the 
corresponding regression coefficient b4.  Rogan et al. (1973) used 
a model of this kind to show that cumulative exposure to respirable 
coalmine dust affected the FEV1 of miners irrespective of whether 
they had simple pneumoconiosis. 

    If the exposure to the pollutant is current, and includes an 
acute reversible effect, then the coefficient bo might be altered, 
reflecting a change in the underlying average level of FEV1 for the 
subjects exposed compared with those not exposed.  Nevertheless, 
the equation as modified remains a so-called  additive linear model.  
It is "additive" because the terms involving different explanatory 
variables are added to each other, rather than being multiplied by 
each other.  It is "linear" because the effect of any one variable 
on the response is represented in the model as changes 
corresponding to a straight-line graph of FEV1 plotted against that 
variable on its own.  More complicated models

    In many situations, an additive linear model may be sufficient 
for practical purposes, despite the fact that it is probably a 
simplification of reality; the existence of statistically 
significant effects may be established and a fair idea may often be 
obtained of the order of magnitude of the change in the response 
variable that might follow changes in the explanatory variables.  
Sometimes, however, it will be convenient, advantageous, or even 
essential to work with more complicated models.  For instance, Cole 
(1975, 1977) suggested that FEV1 might be better represented as a 
power function of height rather than as a linear function.  Jacobsen 
(1975b) argued that a non-additive model incorporating a term that   
represents the product of age and height (AH), in addition to linear 
terms for each of these variables separately, might equally reflect 
the apparent non-linearity of FEV1 with respect to height.  A 
possible interaction between the effect of an environmental 
pollutant and smoking is often important  and may be reflected in a 
regression model by also including a new variable defined as the 
product (ES).  

a Some authors prefer to use the simpler deterministic form of the 
  equation to describe multiple regression models, omitting the 
  random term R, which is then implied by the statistical context 
  of the report.                                     

    In general, an interaction between two or more explanatory 
variables implies that the magnitude of the effect of any one of 
them on the response will itself vary, depending on the level of 
the other(s).  No single regression coefficient can then summarize 
the effect concerned and a simple additive model that does not 
reflect these complications can be misleading.  For instance, in an 
analysis of racial differences in FEV1, Stebbings (1973) found that 
blacks had lower mean values than whites; among non-smokers the 
coefficient for age did not differ between blacks and whites; but 
black smokers had a less rapid decline in FEV1 than white smokers 
for the same amount smoked.  A simple linear model, representing 
race and smoking as additive effects would have been grossly 
misleading.  Dummy variables

    Discrete explanatory variables can be included on the right-
hand side of the regression equation as follows.  Suppose that the 
only information available about the smoking habits of individuals 
surveyed is whether or not they were smokers or non-smokers at time 
of survey.  The variable S in the model described might then be 
defined as taking the value 1, if an individual is a smoker, and 
zero for a non-smoker.  S is now a "dummy variable"a.  The 
regression coefficient (b3) is then to be interpreted as a measure 
of the change in the response depending upon whether or not a 
person is a smoker or non-smoker.                                           
    The same idea can be extended to discrete variables having more 
than two levels.  In general, a discrete variable with n mutually 
exclusive categories can be represented in the model by a system 
consisting of (n - 1) dummy variables. 

    Interactions between dummy variables, or between one dummy      
variable and another continuous variable, can also be accommodated  
in regression models.  An alternative way of accomplishing the same 
objective is to fit separate, but identical, models to data 
corresponding to the different levels of the discrete variable 
(e.g., smokers and non-smokers).  Possible differences in the 
effect of the other explanatory variables can then be examined 
directly.  The usefulness of dummy variables in observational 
studies has been discussed by various authors (Suits, 1957; 
Johnston, 1963; Cohen, 1968; Wesolowsky, 1976) and examples of 
their application in environmental epidemiology can be found in 
several papers by Stebbings (1971a, 1971b, 1972, 1973, 1974).    

a The term "dummy variable" is used also in a different sense, in 
  the context of case-control studies.  Information judged,  a 
   priori, as very unlikely to explain the occurrence of cases is 
  gathered for both cases and controls.  The aim is to monitor the 
  absence of a possible difference in the rigour and effort used to 
  obtain all the information relating to cases and to controls.  An
  apparent association between such a "dummy variable" and the 
  occurrence of cases will alert the investigator to a possible 
  bias in the research methods.                           Selection of variables

    Which explanatory variables should be included in a regression 
model?  In what order should they be introduced into the sequence 
of computations, required to estimate the regression coefficients?  
Which, if any, criteria should be used to decide that a particular 
variable should be omitted from an equation?  These questions have 
been debated at length in the statistical literature and the topic 
remains controversial to some extent.  The issues involved have 
been reviewed by Cox & Snell (1974) and by Hocking (1976).  The 
following non-technical discussion draws attention to the main 

    In a well-designed experiment, the explanatory variables of 
interest are chosen before the experimental work begins and the 
observations are made in a way that ensures that correlations 
between explanatory variables in the experimental data are exactly 
zero.  Generally, no such arrangements are possible for 
observational studies.  For instance, the older members of a 
community being surveyed are likely to have had longer exposure to 
a pollutant under investigation.  The fact that at least part of 
their cumulative exposure may have occurred in earlier years, may 
mean that they received higher doses per unit time early in their 
exposure histories.  Their breathing capacities, as measured, say, 
by the FEV1, will certainly be lower than those of younger persons 
in the same community.  How much of the average decrement observed 
is reasonably attributable to the higher doses received at an early 
age?  How much is due to the higher cumulative exposure?  How much 
is due simply to age? 

    The non-zero correlations between the potential explanatory 
variables complicate interpretation of results, even from a 
multiple regression analysis, because any one regresion coefficient 
that emerges is an estimate of the effect of the variable 
concerned, given that the variability associated with the others 
previously included in the equation has already been taken into 
consideration.  Thus, the inclusion or omission of any one 
variable, and the sequence in which the variables are introduced 
affect the numerical values of the estimates that are made, 
sometimes substantially, and may also affect their apparent 
statistical significance. 

    An intuitively reasonable approach is to ensure, in the first 
place, that the model includes all variables, which are known to 
affect the response.  In the example of FEV1 above, these might be 
height (an intrinsic factor), age (also an intrinsic factor but 
likely to be correlated with other extrinsic factors), instrument 
effects, technician effects, and smoking. 

    Other variables, in which the reality of the effect is 
uncertain and under investigation, would be added to the equation 
at that point.  They might include  concomitant variables (or 
intervening variables) whose influence on results are suspected but 
not fully understood (social class for instance) or variables of 
direct interest such as air pollution, home-heating method, or 

    Some statisticians recommend the use of so-called stepwise 
procedures to select a subset of test variables for inclusion in 
the final fitted equation.  The principle is that if two, or more, 
potential explanatory variables happen to be highly correlated in 
the set of data being considered (say employment in a particular 
industry and use of a particular type of domestic fuel), and 
inclusion of either one of them on its own into the equation helps 
to explain a significant portion of the varitions in the response, 
then inclusion of the other may not appear to have any significant 
effect.  The stepwise procedure then determines which of the two 
explains more of the variance of the response, when they are 
entered into the equation at the same point, and it thus provides a 
criterion for excluding one and including the other variable from 
the fitted equation. 

    Used in this way, stepwise regression is essentially a decision 
procedure, and it presupposes a logical selection of the level of 
statistical significance that is required to justify inclusion of a 
variable, and also the level of change in the estimate of response 
(per unit change in the explanatory variable) that is regarded as 
important enough to merit inclusion of one or other or both of the 
competing candidates.  Useful reviews of stepwise selection 
procedures are included in the books by Wesolowsky (1976) and 
Draper & Smith (1966). 

    Other research workers feel that, because the selection of 
constraints for the decision procedure are a matter of subjective 
judgement, the apparent objectivity of variable exclusion criteria 
may be misleading.  It is argued that, if these methods are used at 
all, then all results should be reported, perhaps with some comment 
on which particular formulation of the model produced the best 
fitting equation.  Readers of the report may then use their own 
judgements as to whether results based on the statistical criteria 
adopted coincide with their general expectations based on common-
sense and on familiarity with similar data from other sources.  An 
unconstrained exploratory approach to the data, backed by 
informative scattergrams, is often the most helpful way of trying 
to understand results from epidemiological surveys. 

    If two variables are very highly correlated in the data then no 
statistical procedure can determine which of the two is more 
important;  but presumptive considerations, independent of the data 
themselves, may suggest which is of greater interest. 

    It can happen also that an explanatory variable known, in 
general, to influence the response (say height on FEV1) happens to 
be not statistically significant at some arbitrarily chosen level in 
a particular analysis, perhaps because the range of heights observed 
was too small to affect the response significantly, relative to the 
residual variability in that set of data.  This does not mean that 
the variable concerned can be excluded from the equation with 
impunity; it may nevertheless affect the estimates of one or more 
of the other variables under investigation.  If there are sound 
reasons for supposing that a particular variable influences the 
results (e.g., smoking on lung function, or the use of different 

technicians in cooperation-dependent or partly subjective tests) 
then it is wise to include those variables in the model, 
irrespective of their statistical significance.  This is equivalent 
to standardizing results; it places assessments of other effects on 
a common base-line and helps to avoid artefacts and biases.  Evaluating "goodness of fit"

    Evaluation of the model that has been chosen can be considered 
under two broad headings; goodness of fit and stability (section  First, how well does the fitted equation explain the 
variability in the particular set of data that has been observed? 
("Goodness of fit"). 

    A  residual is the difference between a single observed
response and that which may be estimated from the fitted equation.  
The pattern of residual variability in the data, that is the 
variation unexplained by the equation, can give valuable clues to 
deficiencies in the fitted model.  Graphical and tabular analyses of 
residuals may reveal trends indicating that the functional form of 
the equation is inadequate, or that there are interactions between 
explanatory variables that have been included only as additive 
terms, or that assumptions that are made in significance testing 
about the shape of the distribution of residual variability are 
inappropriate.  Draper & Smith (1966) devote an entire chapter of 
their book to this important topic and explain how graphical 
analyses may be complemented by formal tests based on probability.  
Cox (1970) also includes a discussion of how to examine residuals in 
models where the response is a transformation of a binary variable. 

    Corrective action to improve the fit can take various forms 
depending on the deficiency identified.  Nonlinear relationships may 
often be studied using regression methods either by mathematical 
transformations of the variables into a linear form or by the 
addition of polynomial terms into an additive model.  Draper & Smith 
(1966) give a useful introduction to the subject; Williams (1959) 
discusses transformations and nonlinear regressions, and the 
comparison of alternative forms of regression.  Frequently, in 
environmental epidemiology, there is no clear choice between linear 
and nonlinear regressions; both may then be tried.  Sometimes, when 
there is a large residual variance, neither form has a clearly 
demonstratable advantage.  Non-normality in the residuals or a 
changing variance of the residuals over the range of the data 
(heteroscedasticity) do not bias the estimates of the regression 
coefficients, but serious deviations from these properties may 
affect the validity of tests of significance.  Methods for dealing 
with these problems are discussed by Wesolowsky (1976). 

    Dependence of a residual on preceding residuals, when the 
observations are arranged in some natural order, indicates 
autocorrelation, and this may occur in studies involving variations 
in space or time.  Johnston (1963) and Wesolowsky (1976) discuss 
autocorrelation in the temporal case; King (1969) discusses it in 
geographical correlations. 

    Graphical analyses of residuals may draw attention to so-called 
outliers that may or may not have been revealed in the earlier data 
preparation and description.  These are values of the observed 
responses that lie well outside the range of the rest of the data, 
the plausibility of which is suspected.  The controversial question 
then arises:  is it sensible to include such unorthodox results in 
the analysis?  Tietjen & Moore (1972) described some methods for 
detecting outliers and for deciding whether or not they should be 
removed from the analysis to improve the goodness of fit.  When 
either the response or explanatory variables are highly skewed, as 
may be the case in a time-series of air pollution measurements or 
in studies of trace-substances in the body, Scott (1964) suggested 
serial deletion of outliers, and examination of a plot of 
regression coefficients to determine when they stabilize.  It is 
clearly important to be particularly cautious  in drawing 
conclusions about the importance of a pollutant, if deletion of one 
or several observations from dozens or hundreds drastically changes 
the inference that would be made without the deletion. 

    A crude overall index of "goodness of fit" of the model to the 
data is the  coefficient of determination (R2) which is a measure 
of the proportion of the total variability that has been explained 
by the particular formulation of explanatory variables appearing in 
the equation.  On its own, this index is only of limited value; it 
may be quite low, even when statistically significant effects have 
been detected. 

    Some measurements have a high proportion of between-individual 
variability to total variance.  The FEV1 is an example for which a 
review of variance components is available (Stebbings, 1971a).  
Others, like the fractional carbon monoxide uptake (Stebbings, 
1974), have a very high variability component within individuals 
over several months and a low component between individuals.  Data 
from cross-sectional studies provide the basis for studying 
variations between individuals in large human populations; studies 
involving repeated measurements in relatively small groups may be 
used to estimate within-individual variability.  But, in general, 
either type of study, on its own, describes adequately the total 
variability in large populations.  This underlines the importance of 
trying to identify and estimate the components of total variability 
in responses, rather than relying on correlation coefficients (r) or 
coefficients of determination (R2) to assess the viability of 
particular hypothesized relationships. 

    If it is known from the literature or from pilot studies that a 
particular response measurement has a relatively high within-
individual variance component, then a cross-sectional study may be 
strengthened by arranging for repeated measurements on the same 
subjects.  The analysis of results may then separate the between- 
and within-individual variance components and this will make more 
sensitive statistical tests possible for what may be small but real 
differences between groups of individuals, i.e., the statistical 
power of the procedure is increased.  Evaluating the stability of models

    If a fitted regression equation appears to explain the observed 
data moderately well, it will be desirable to consider how likely it 
is that the fitted relationship will be reproducible, at least 
approximately, in other similar surveys.  Replication of findings in 
different studies is the ultimate evidence of the reliability of 
conclusions.  Cox (1968) noted that stability might mean that, when 
the survey or experiment was repeated under different conditions:  
either (a) the same regression equation would hold, even though 
other aspects of the data changed; (b) parallel regression 
equations would be obtained; or (c) satisfactory regression lines 
would always be obtained, but with different positions and slopes. 

    The ideal situation (a) is rarely achievable in epidemiology 
because of well-known demographic variations in many physiological 
and biological indicators.  Type (b) stability is very important 
when considering estimates of the effects of a pollutant if the 
results from surveys are to be used for decisions on environmental 
hygiene standards.  Even when such stability is demonstrated with 
respect to the effect of the pollutant, the effects of other 
concomitant variables may vary in different human populations. 

    Type (c) stability amounts to convincing confirmation that 
assertions of statistically significant effects in one survey are 
not artefacts.  Cox suggests that variations in parameters indicated 
by situations (b) and (c) may be investigated by further regression 
analyses, in which the differing estimates of regression coefficients 
are treated as responses, and the new explanatory variables now 
characterize the different populations that have been studied. 

    All this presupposes that comparable studies have been reported.  
An indirect approach to the problem in "ecologic" studies is to 
investigate a number of test variables which are not expected to 
correlate with the health indicators.  A selection of such variables 
is generally available from published statistics.  They should not 
correlate with health as readily as do the suspect agents, if the 
apparent effects of the latter are real (see footnote on dummy 
variables in section  Predicted normal values

    Estimates of regression equations based on large samples of 
various populations have been published and provide so-called 
predicted normal values, particularly for various measures of lung 
function.  These predictions may be used to aid clinical diagnosis 
of disease or respiratory impairment in an individual.  Murphy and 
Abbey (1967) and Oldham (1970) discuss some of the difficulties 
involved and errors that may occur. 

    However, it will be clear from the foregoing discussion of the 
variety of regression models that can be postulated reasonably, and 
from the differences in coefficients that may be relevant in 
different studies, great care must be taken if such predictions are 
used to assess deviations from "normality" in an epidemiological 

survey.  If data are available to estimate the appropriate 
coefficients from the sample being studied, some researchers feel, 
when attempting to make inferences in epidemiology, that predicted 
normal values should never be relied on.  For FEV1, Cole (1977) 
suggested relatively simple formulae for standardizing this measure 
of lung function for age and height, while avoiding the 
computational complexities of having to fit a multiple regression 
equation.  Other methods for studying multivariate data

    Sophisticated instrumentation is frequently used in 
environmental epidemiology, for measuring both the environment and 
physiological responses.  Random error in such measurements is 
often a major contribution to the total variability in the 
experimental setting.  Great care should therefore be taken in 
using conventional regression on correlation models to analyse such 
data, since an assumption that the explanatory variables have been 
measured without error is usually not even approximately satisfied.  
This problem and some solutions are discussed by Williams (1959) 
under the title "functional relations".  One approach is orthogonal 
regression or principal component analysis.  Hartwell et al. (1974) 
presented a detailed discussion of the problem in a report on nine 
methods for monitoring nitrogen dioxide in ambient air.  In their 
numerous examples, they compared results from principal components 
with those obtained from conventional regression analyses.  A 
similar idea has been used to reduce a complex of interconnected 
variables, reflecting pollution of urban environments, to a smaller 
number of orthogonal factors with only weak correlations between 
them (Cassel et al., 1969; Zvinjackovskij, 1979; Shandala & 
Zvinjackovskij, 1981).  Such preliminary condensation of the data 
provides new explanatory variables, representing groups of 
environmental factors with only low correlations between groups, 
which may then be related to morbidity patterns in the exposed 

    One usage of the term "multivariate analysis" restricts it to 
methods involving multiple response variables.  These techniques 
(discriminant analysis, canonical correlation analysis, factor 
analysis, etc.) are used in epidemiology for the identification of 
disease entities and for the classification of patients (Kasap & 
Corkhill, 1973; Maxwell, 1970).  Application of these methods in 
environmental epidemiology is relatively rare, possibly because the 
health indicators and environmental variables measured in such 
studies are frequently of intrinsic interest individually.  
Attempts to condense several variables into one or two composite 
measures of response may then tend to obscure rather than clarify 
the relationships that are being sought.  Examples of the use of 
such methods in occupational health studies include Liddell's 
(1972) application of canonical correlations to data on the 
appearances of coal-miners' chest radiographs in life in relation 
to the weight and composition of dust found in their lungs  post 
 mortem; and a study of coal-miners' mortality in relation to 
canonical variates derived from a battery of correlated  in vivo 
lung function tests (Oldham & Rossiter, 1965). 

    Another interesting application of these ideas, relevant to 
environmental epidemiology, is in the stratification of study 
subjects by the multivariate confounder score of Miettinen (1976).  
This makes it possible to adjust for several disturbing variables at 
the same time, thereby avoiding the need for multiple cross-
classification of the data as the prelude to standardization. 

6.4.4.  Analysis of data from prospective and follow-up studies  Nomenclature

    In this section, statistical approaches to results from studies 
involving a prospective health follow-up of a defined group, or 
cohort are discussed.  This type of investigation is referred to 
variously as a prospective, longitudinal, follow-up or cohort study, 
but we use the terms prospective and follow-up as explained in 
section 2.6.  Time as a measured variable

    Most of the statistical concepts and methods discussed in the 
previous sections are also widely used in the analyses of results 
from prospective and follow-up studies.  But the fact that a 
prospective or a follow-up study involves observations at least at 
two points in time adds a new dimension to the analysis.  It is 
possible to consider the  incidence of disease, not just prevalence;  
to measure the rate of change of a condition (for instance, 
deterioration in lung function, radiological progression); and to 
monitor environmental conditions over the period of follow-up.  
Exposure/effect relationships and disease latency periods may be 
studied with greater precision than would be possible otherwise.  
But new methodological problems arise in the statistical analysis.  
Comparisons between groups over a period of time are also 
complicated by the changing age structure of those at risk of 
disease or death as the study proceeds. 

    The results of prospective and follow-up studies can be 
expressed as estimates of absolute risks (exposure-specific 
incidence rates, prevalence rates, or mortality rates) or as 
relative risks (relative that is, to some unexposed or comparison 
group; the comparative mortality figure (for instance). 

    The following discussion refers to some of these additional 
features.  Studies of cancer mortality are used to illustrate the 
main points, although many of the methods are equally applicable to 
investigations of morbidity or other causes of death.  Person-years method

    As noted earlier (section, the force of mortality in a 
population over an extended period of time may be summarized by an 
average annual death rate, that is, the total number of deaths 
divided by the  person-years-at-risk (PYR).  The same principle is 
often applied and extended when calculating denominators for 
 Standardized Mortality Ratios (SMR) in prospective and follow-up 

studies.  The number of deaths expected over the whole follow-up 
under the hypothesis of no difference in mortality between the group 
being studied and a standard control group is then: 

                    t     s   [ (survivors up to)   (standard death) ]
Expected deaths = sigma sigma [ (start of year i) x (rate in year i) ]
                   i=1   j=1  [ ( in stratum j  )   ( for stratum j) ]

    If an external control groupa is being used for the comparison, 
then the s strata are likely to refer to specified values of 
variables such as sex, race, and attained age.  Cause-specific 
standard death rates corresponding to the strata for different 
calendar years are obtained from published statistics for the 
geographical region concerned.  Note that if such statistics are 
used as standard rates, then it is improper to correct death 
certificates for the study group by the use of supplementary 
information (hospital records, pathology reports, etc.) since such 
corrections are not made for the death certificates used for the 
compilation of population death rates.  Preferential correction of 
data for the study group would tend to bias cause-specific 

    When internal controlsb are used, further stratification is 
possible to adjust for other factors (e.g., smoking history).  
Thus, by proper stratification it can be inferred that the 
difference in cancer mortality between the exposed group and 
controls is due to exposure to the agent, and not to differences in 
the composition of the comparison group such as age, race, and sex. 

    Pasternack & Shore (1977) pointed out that in studies of 
radiation carcinogenesis an assumption is often made that the 
carcinogenic risk is additive and therefore expressible in absolute 
terms.  The usual terminology expresses absolute (excess) risk as 
"cases/(106 "years" rem)".  This implies that the radiation-
induction of some types of cancer is independent of the spontaneous 
rate of that cancer; an important assumption, because the 
spontaneous incidence of most forms of cancer varies markedly with 
age.  If the assumption is not true, then the absolute risk 
estimates are interpretable, only if they are expressed specific to 
age.  In some cases, the ratio of the absolute risk to the 
spontaneous risk will be constant at different ages, suggesting the 
use of a relative risk model.  Taken as a whole, the available 
epidemiological data on radiation carcinogenesis do not clearly 
support either the absolute risk or the relative risk models.  
Therefore, the nature of the relationship between age and radiation 
risk should be examined carefully in each study. 

    Because age has such a powerful influence on the rates of 
tumour induction and on many other diseases, it is important to 
make careful adjustments for age variations in prospective and 
follow-up studies.  Ideally stratification should be by single

a See section 2.6.                                                       
b See section 2.6.                                                       

year of age, if the sample size were large enough, although 5-year 
intervals are satisfactory for most purposes.  Occasionally, it 
may be necessary to use 10-year intervals, when there are only 
small numbers in subgroups, defined by age and other factors 
simultaneously.  Studies using 20- or 30-year age groupings may 
fail to adjust for a substantial fraction of the bias introduced 
by age differences.  Even if it is desired to report data grouped 
by wide age ranges,  such as this, it is desirable to stratify by 
narrower ranges in the analysis, and then summarize results in 
broader groups.  Modified life-table method

    If information on exposure history is available, then more 
informative and sensitive analyses become possible by stratifying 
the data according to years of exposure.  With elapsed time since 
first exposure, it is possible to consider latency, as well as the 
possibility that the carcinogenic risk (the incidence rate) may 
change with time from the initial exposure.  Ignoring latency 
usually leads to an underestimate of carcinogenic effect.  Even 
when the exposures of individuals to the suspected pollutant cannot 
be measured precisely, it may still be possible to construct 
approximate exposure categories based on the relative magnitudes of 
exposure (taking duration, intensity, and timing into account).  Any 
evidence of an exposure/effect relationship would further strengthen 
inferences concerning the presence of an effect.  Overlap of exposure and observation periods

    If the exposure period and the follow-up period do not overlap 
(as might be the case when studying groups of retired workers), then 
those at risk in the exposed group may be stratified according to 
appropriately defined exposure categories.  But, if the exposure and 
follow-up periods overlap, as is usually the case, difficulties 
occur.  As noted by Enterline (1976), high exposure and death tend 
to be incompatible states.  For example, in a retrospective follow-
up of employees entering a study in 1938 say, with a follow-up 
period 1938-64 and exposure measured as time worked during the 
interval 1938-64, the maximum exposure period (27 years) could only 
be attained for a worker who survived the entire follow-up period.  
Early cancer death during the observation period had to correspond 
to lower exposure.  Thus, there is a built-in bias that tends to 
generate a spurious negative correlation between exposure and 

    When overlap occurs, modified techniques have to be used. 
Pasternack & Shore (1977) suggested the following method for 
dealing with the problem.  Each PYR for a worker is assigned to 
an exposure category that reflects the total exposure experience 
(or score), up to the time point concerned.  This implies that 
any one person may contribute PYRs to several exposure categories.  
If each PYR is also simultaneously stratified according to yearly 
interval since first exposure (eliminating early years, say the 
first five years, that is, the minimal latent period), as well 
as attained age and any other factors to be controlled, expected 

deaths can be obtained for exposure categories with equivalent 
times since first exposure and with adjustment for age and other 
confounding factors.  This method was used by Pasternack et al. 
(1977) in their retrospective cohort mortality study of occupational 
exposure to chloromethyl ethers - resulting in improved estimates of 
the exposure/response relationship.  Breslow (1977) suggested an 
alternative approach using time-dependent covariates, based on Cox's 
(1972) regression model for survival data.  Lagged exposures

    Assignment of a current cumulative exposure score to each PYR, 
as suggested above, may not be appropriate for studying cancer 
mortality, because recent exposures are unlikely to affect the 
current cancer risk.  It was therefore suggested by Pasternack & 
Shore (1977) that  lagged exposures might be used to predict risk.  
For example, the exposure/response relationship could be tested when 
the PYRs were assigned to exposure categories reflecting the 
person's cumulative exposure from, say, five years prior to the 
given PYR.  Ideally, if a model of tumour growth-rates or other 
temporal factors in tumour induction is available, the lagged years 
could be weighted in correspondence with this model.  Without such a 
model, an exploratory series of analyses might be performed with 
lags of one year, two years, three years, etc., to determine the 
degree of lag that results in the most sensitive statistical 
analysis for detecting a carcinogenic effect.  Measures of latency

    The end of a follow-up period in a study of environmental 
cancer usually occurs before all those at risk have died.  A 
precise survival time is known for those who died during the 
follow-up; but, for the remainder of the cohort, survival time is 
less definite.  Such data are said to be  censored up to the end of 
the follow-up.  As the censoring increases, the latency estimates 
derived from considering only the cancer cases will be increasingly 
too small, since the PYRs decrease as the interval from exposure 
onset increases.  This would not be of great concern if the degree 
of bias were equal in all the groups being compared.  However, 
according to the differences in the amount of censoring differs 
among the groups, the amount of bias will also differ, so a measure 
of latency is needed that takes into account the size of the 
population at risk at each interval.  A median latency derived from 
a life-table analysis provides such a measure.  Pasternack & Shore 
(1977) provide details, as do Robinson & Upton (1978) who describe 
alternative methods in the context of animal carcinogenesis 

    It should be noted that if a median or other quantile value is 
to be interpreted in an absolute sense, then it is necessary to have 
observed at least part of the sample until the end of the period of 
tumour induction.  However, even when the end of the induction 
period has not been reached, it is still possible to compare the 
median, etc., in two or more sets of data having the same length of 
observation, and to determine if there are relative differences in 
the time to tumour appearance. 

    The goal in studying latency is to examine the distribution of 
only those cancers attributable to the environmental exposure under 
study.  If the available data suggest that the exposure being 
studied accounts for virtually all the cancers that occur, then the 
method described by Pasternack & Shore (1977) may accomplish this 
goal.  Usually, however, such data are found mainly with tumours, 
which quite rarely occur spontaneously, or when the relative risk 
is very high, so that the bulk of the tumour occurrences are 
carcinogen-induced.  With a more common type of cancer and a fairly 
low relative risk, the fraction of cancers attributable to the 
carcinogen is small:  thus any calculation of latency needs to be 
corrected for the spontaneous incidence.  Some analytical techniques

    A modification of the procedure described by Mantel & Haenszel 
(1959) can be used when two sets of life-table data are to be 
compared (Mantel, 1966; Hankey & Myers, 1971).  The Mantel-Haenszel 
(1959) relative-odds measure, or Miettinen's (1972) standardized-
risk ratio provides estimates of adjusted relative risk (see also 
Thomas, 1975). 

    The Mantel-Haenszel (1959) technique can be generalized to three 
or more groups (Mantel & Byar, 1974).  When the groups are defined 
by exposure levels, exposure/response relationships can be tested 
for linearity and nonlinearity by methods developed by Mantel (1963) 
and Tarone (1975).  Using these methods in conjunction with 
stratification of the data as suggested above, it is possible to 
examine the relationships while adjusting for interval since 
exposure, age, and other variables. 

    Other techniques have been described for relating risk factors 
to disease or death in prospective and follow-up studies, based on 
the multiple logistic model used by Truett et al. (1967); see 
reports by Brown (1975); Byar & Mantel (1975); Dyer (1975a,b) and 
Farewell (1977).  Kullback & Cornfield (1976) described the use of 
log-linear models for analysing multiple cross-classified data and 
gave an illustration dealing with the effect of a number of smoking 
variables on mortality from coronary heart disease. 

    The role that epidemiological methods play in the process of 
assessing human risk from cancer, including sample size requirements 
for cohort and case-control studies were discussed by Mack et al. 
(1977).  Applications of multistage models of disease induction to 
problems of design, analysis, and interpretation of epidemiological 
studies were considered by Berry (1977), Berry et al. (1979), 
Whittemore (1977), and Peto (1978a,b). 

    Many of the most recent developments in statistical methods for 
analysing data from prospective and follow-up studies are based on 
applications and extensions of the important general model for 
survival data proposed by Cox (1972).  The flexibility of this 
approach to the study of time-related and other explanatory 
variables in an epidemiological setting is explored below, in the 
context of case-control studies (section et seq.). 

6.4.5.  Analysis of data from case-control studiesa  Relative and absolute risks

    The difficulties in reaching valid conclusions from the case-
control approach are discussed in a paper on how to analyse data 
from retrospective studies by Mantel & Haenszel (1959).  The authors 
emphasize that the investigators must satisfy themselves about "the 
fundamental assumption underlying the analysis of retrospective 
data":  that the assembled cases and controls are representative of 
the statistical population defined for the investigation. 

    The contingency table methods described by Mantel & Haenszel 
concentrate on how to summarize the overall  relative risk from 
substrata among those being studied, where the strata refer to 
groups similar in age or some other factor expressed as a discrete 

    Note that the results are expressed as estimates of relative 
risk; on their own, case-control studies do not provide measures of 
absolute risks.  Relation between prospective and case-control studies

    Mantel & Haenszel (1959) noted that, in a case-control study, 
 "a primary goal is to reach the same conclusions ... as would have 
 been obtained from a forward study, if one had been done."  This 
viewpoint is fundamental to a relatively new methodological 
approach described here.  The analytical methods for prospective 
and follow-up studies outlined in section 6.4.4 may relate the 
probability of disease occurrence, or more precisely the rate of 
occurrence, to each individual's exposure history.  The following 
discussion shows that the same methods of analysis can be adapted 
for use with the case-control design. 

    Consider first the prospective and follow-up study in which a  
population of disease-free persons is followed over a study period 
of defined length.  Subjects are classified at the beginning of the 
period, according to their exposure history, and at the end, 
according to whether or not they have developed the disease. (Times 
of disease occurrence during the study period are ignored for the 
moment).  A common method of analysis for this situation is to 
apply the linear logistic model (Cox, 1970), which relates a binary 
response variable y to a vector of K independent explanatory 
variables X = (X1, ..., XK) via the conditional probability 

a Note the different meaning attached to the word "control" here, 
  compared with section  et seq.  There the "control group" 
  was required for a comparison of incidence of disease, not for a 
  comparison of previous history.  See also the remarks about 
  nomenclature for variables in section  

    pr (y=1|X) = {1 + exp (-alpha - sigma betakxk)}-1  (1)

Here y denotes whether (y = 1) or not (y = 0) the disease occurs, 
while X represents the exposure history and other relevant risk 
variables.  A large number of possible relationships can be 
represented in this form by including among the Xs both discrete 
and continuous variables, transformations of these variables, and 
interaction (cross product) terms.  The classic illustration of this 
methodology is that of Truett et al. (1967), who used it to study 
the simultaneous effects of age, systolic blood pressure, serum 
cholesterol, and other variables on the risk of coronary heart 

    Provided that the disease is sufficiently rare, or the study 
period sufficiently short, the ratio of risks (RR) for individuals 
with two separate sets of values X* and X is well approximated 
by the corresponding odds ratio.   

   RR  =  -------------------               (2)

Under the logistic model (1) this takes the particularly simple 

    exp {sigma beta (X * - X )}              (3)
           1       k  k     k

Thus the effect of a unit increase in the value of the kth risk 
variable Xk is to multiply the risk of disease by the factor 
exp (betak).  In other words, betak represents the natural 
logarithm of the relative risk accompanying a unit change in Xk. 

    Case-control studies should involve random sampling from the 
population at risk.  Typically, all or nearly all available cases 
are used, while the controls represent only a small fraction of 
persons who remain disease-free throughout the study period.  The 
essential requirement is that the  sampling fractions for cases and 
 controls must not depend on any of the risk variables under study. 
Under these circumstances, the relative risk corresponding to 
different histories can be estimated by fitting to the sampled case-
control data exactly the same logistic model (1) as would have been 
fitted to the data on the entire cohort, had they been available.  
The regression coefficient beta has exactly the same interpretation 
as for the cohort study; only the constant term alpha is modified, 
depending on the ratio of the sampling fractions for cases and 
controls.  See Anderson (1972), Mantel (1973), and Prentice & 
Breslow (1978) for technical details that justify this approach. 

    One of the major difficulties of hospital-based studies, in 
which persons with diseases other than that under study are chosen 
as controls, is that they often violate the requirement that the 
control sampling fractions should not depend on the risk variables.  

Since the same exposure may be related to several diseases, exposed 
persons can easily be over-represented in the control sample, which 
leads to the relative risk associated with such exposures being 
underestimated.  Analysis of stratified samples

    In practice, it would be rare to draw an unrestricted sample of 
cases and controls from the population at risk.  The age and sex 
distribution of patients with particular diseases usually departs 
markedly from that of the general population.  Since these 
variables are often related also to exposure, they may confounda 
the relationship between risk factor and disease (Miettinen, 1974).  
This suggests that a  stratified control sample should be drawn with 
roughly the same age and sex distribution as the cases.  Additional 
nuisance factors may be used for strata formation, when their 
influence on the risk of disease occurrence is not itself of 
intrinsic interest but serves mainly to confound the issue.  Strata 
may be formed also on the basis of variables which could interact 
with the exposures to modify their effects.  Since variables 
selected for stratification at the design stage cannot be evaluated 
as potential risk factors, an appropriate choice requires 
considerable care as well as substantial knowledge of the disease 
process.  Factors considered to be a nuisance on one  occasion may 
well be the risk variables of prime importance in another study.         

    Even if a stratified sample has not been drawn at the design 
stage, it is possible to form the strata at the time of analysis. 
The same criteria are applied, namely the variables selected for 
strata formation are those that may confound or modify the       
disease/risk factor association.  Of course, with  post-hoc 
stratification there is a risk of having serious imbalances between 
the  number of cases and controls in some strata.                       
    Thus, the strategy of statistical analysis suggested here is to 
control the effects of the nuisance factors by stratification, 
while modelling the exposure main effects and interactions via the 
linear logistic model.  Equation (1) is generalized to: 
pri(y=1|X) = {1 + exp (-alphai-sigma beta X )}-1  (4)
                                     1  k k

a The word "confounded" is used in the statistical theory of            
  experimental design to describe an arrangement whereby 
  information about the independent effects of some factors is 
  sacrificed deliberately for the sake of economy of effort.  The 
  more general use of the term in epidemiology refers to 
  inadvertent associations between the main factor under 
  investigation and other factors which may also affect the 
  response variable.  Special care is then necessary to try to 
  distinguish between the effects of the confounded variables.  See 
  also the remarks in sections  2.5 and 6.4.6. 

where i = 1, .....,  I indexes the stratum.  If none of the 
regression variables X are interaction terms between exposure and 
stratification factors, a consequence of this formulation is that 
the RR associated with particular exposures remains constant over 
strata.  However, by including such interaction terms, changes in 
the RR can be modelled that accompany changes in age or other 
stratification factors. 

    If the study period is very long, say more than a year or two, 
dividing it into intervals and using calendar time as one of the 
stratification variables in addition to age should be considered.  
Cases developing the disease during a particular time interval are 
matched with controls who remain disease-free at that time.  Thus, 
the probability (Equation 4) refers more specifically to the 
conditional probability of developing the disease during the 
associated interval, given that the subject was disease-free at the 
appropriate instant. An advantage of this formulation is that, 
provided the intervals are made sufficiently short, the probability 
of developing the disease during any one of them is so small that 
there is no doubt about the odds ratio approximation (Equation 3). 

    A convenient computer package for fitting the model (Equation 
4), among several that are available, is the General Linear 
Interactive Modelling Program (GLIM) developed by Nelder (1975).  
This provides maximum likelihood estimates of the relative risk 
parameters beta and large sample estimates of their standard   
errors.  Evaluation of the statistical significance of terms in 
the regression equation, whether singly or in groups, is made via 
the likelihood ratio test. Breslow & Day (1980) present several 
worked examples that illustrate this approach.  Analysis of matched samples

    If closer control over the nuisance factors is desired than 
that provided by stratification into broad categories, each case 
may be matched individually to one or more controls.  Studies 
carried out entirely in hospital often involve sets of cases and 
controls matched on the basis of age, sex, race, and date of 
admission.  For occupational studies, the control may be chosen as 
having the same dates of birth and employment and being disease-
free at the time that the case is diagnosed. 

    One view of matching is as a limiting form of stratification in 
which the intervals of age and time used to form the strata become 
infinitesimally small.  This results in a continuous time analogue 
of Equation (4) wherein the ratio of age- and time-specific 
incidence rates for persons having exposure variables X* and X is 
given exactly by Equation (3).  It  will be recognized as the 
proportional hazards model of Cox (1972), which was mentioned 
earlier as a major tool for the analysis of cohort studies.  In the 
analysis stemming from this model, each case of disease occurring 
in the cohort is compared to the risk set of all persons remaining 
disease-free at the time the case was diagnosed.  With the case-
control approach, the case is compared only with those matched 

controls actually sampled from the risk set at that time.  Prentice 
& Breslow (1978) give a more detailed account of the proportional 
hazards model as it relates to case-control studies. 

    Alternatively, each matched set may simply be regarded as a 
stratum for which the logistic model (Equation 4) continues to 
hold.  However, when the number of alphai parameters is of the 
same order of magnitude as the number of subjects, then it is no 
longer feasible nor advisable to estimate them all.  Instead, they 
may be eliminated from the model by conditioning on the collection 
of risk vectors X actually observed for each matched set (Breslow 
et al., 1978).  Xijk denotes the value of the kth variable for 
the case (j = o) or jth control in the ith matched set.  Then 
the conditional probability that the vector of variables Xio = 
(Xio1, ..., Xiok) pertains to the case, as observed, and the 
remainder to the controls is 

    1 + sigma exp{sigma beta (X   - X   )}      (5)
          j            k    k  ijk   iok

    A likelihood function for beta is obtained by taking the 
product of the conditional probabilities (Equation 5) over all the 
matched sets.  This may be used to generate maximum likelihood 
estimates of the regression coefficients, standard errors, and 
likelihood ratio tests as in the stratified analysis based on the 
model (Equation 4).  Computer programs for implementing the matched 
analyses are available (Smith et al., 1981; Gail et al., 1981), and 
worked examples are presented in an IARC/WHO publication on the 
analysis of case-control studies (Breslow & Day, 1980).  Effect of ignoring the matching

    Prior to the advent of methods for the multivariate analysis of 
matched case-control studies, it was common practice to ignore the 
matching in the analysis.  This was thought to do little harm since, 
at least in the case of a single binary risk factor, it was known to 
yield conservative results (Seigel & Greenhouse, 1973; Armitage, 
1975).  Moreover in many problems, accounting for the matching in 
the analysis did not make any perceptible changes in the estimates 
of relative risk. 

    When matching variables are strongly correlated with both 
disease and exposure, however, it must be anticipated that an 
unmatched analysis may lead to serious underestimation of the 
relative risk.  Breslow & Day (1980) present an example which 
contrasts the correct analysis based on the conditional likelihood 
derived from (Equation 5) with several unconditional analyses that 
take increasing account of the matching variables through inclusion 
of additional A1 stratum parameters in the logistic equation (4).  
Estimates of the log relative risk (beta) parameters in the 
unmatched analysis are approximately half, in absolute value, of 
those obtained with the fully-matched conditional analysis.  
Estimates based on unconditional models incorporating some of the 
matching variables occupy an intermediate position.  This suggests 

that the bias in an unmatched analysis may be avoided, at least in 
part, by modelling the effects of the matching variables in the 
regression equation.  Of course, this is only feasible when the 
matching factors are quantifiable, and would be of little help if 
cases and controls were sibs, for example. 

    Situations will occur when variables that were thought to have 
confounding effects during the planning process and were therefore 
used for matching, turn out later not to have such properties.  The 
question then arises concerning the loss in efficiency that may 
accompany a fully matched analysis when it is not needed.  If, in a 
matched-pairs design, the probabilities of exposure to a binary risk 
factor are constant over the matched sets at p1 and po for cases and 
controls, the matched analysis is unnecessary to avoid bias and has 
an efficiency of: 

        p1 q1  + po qo
        p1 qo  + po q1

relative to the analysis that ignores the matching.  The efficiency 
is 100%, when the relative risk is 1 (p1 - p0), and generally falls 
below 70% only if relative risks larger than 4 or smaller than 0.25 
are being estimated (Breslow & Day, 1980). 

    A related question is whether or not matching at the design 
stage ought to have been used at all as a method of controlling the 
confounding effects of quantitative variables that could be 
controlled in analysis.  Several papers (Kupper et al., 1981; 
Samuels, 1981; Smith & Day, 1981) suggest that such matching is 
helpful, only if the confounding variables are strongly related to 
the disease.  Otherwise, the gains in efficiency from having case-
control samples that are balanced  vis-ą-vis the confounding 
variables are not important and, in some situations with large 
relative risks, there may even be a loss.  Alternative methods of analysis

    The methods described above require the solution of nonlinear 
equations by iterative numerical methods and are thus practical 
only if the research worker has access to the appropriate computer 
hardware and software.  When interest is focused on a single risk 
variable, and particularly when that variable takes on only two 
values, such machinery is not needed.  The techniques for 
stratified analyses proposed by Mantel & Haenszel (1959) have 
served many epidemiologists well for nearly two decades.  The same 
non-iterative techniques can also be used with matched designs and, 
if the number of controls per case is constant, they lead to quite 
simple expressions for estimates and test statistics (Miettinen, 
1970).  The required calculations are easily programmed for a 
pocket calculator, or may be performed by hand, if the data are not 
too extensive.  Fleiss (1981) gives many examples. 

    If high-speed computing machinery is available, however, there 
is considerable advantage in using the logistic modelling approach.  
Since it is couched in a general regression framework, it allows 
the user a great deal of flexibility with regard to the treatment 
of the various risk variables in the analysis.  Continuous 
variables such as age, weight, or blood pressure are probably best 
categorized into several levels, a separate estimate of relative 
risk being made for each, relative to a designated baseline level. 
However, they may be analysed as continuous variables, using 
transformations and polynomial expressions, where necessary, to 
achieve an adequate fit to the data.  Interactions among risk 
variables, or between risk and stratification variables, are easily 
explored.  Powerful tests are available for assessing the 
statistical significance of such interactions, which would not be 
feasible with the simpler stratified analyses.  Using general 
purpose computer software such as the General Linear Interactive 
Modelling (GLIM), all this may be accomplished with relative ease 
using standard statistical nomenclature.  In short, the logistic 
model and its analogues provide a link between the fundamental 
epidemiological measure of relative risk and the mainstream of 
statistical concepts and methods. 

6.4.6.  Drawing conclusions from analyses

    Most of the statistical methods referred to above are based 
implicitly on an assumption that is very rarely justified for 
epidemiological data.  The assumption is that the observed response 
measurements are random samplesa from hypothetical statistical 
populations of similar responses. 

    Consider a well-designed clinical drug trial:  it is possible, 
at least in principle, to allocate one or other treatment to persons 
selected randomly from a defined population of patients.  The 
theoretical requirement of random sampling may be met, or very 
nearly met.  But a group of individuals who happen to be exposed to 
a pollutant because they are resident in a particular part of a 
district, are certainly not a random sample of all people in that 
district.  Moreover, their residence there is likely to be 
associated with a number of social factors, some of which may be 
associated in turn with the same measures of response that are 
thought to be affected by the pollutant; the effects of the social 
factors are then partly confounded with the effect of exposure to 
the pollutant (footnote to section 

    The problem of obtaining random samples is common to all 
observational studies.  Statistical implications have been reviewed 
in detail by McKinlay (1975) who refers to many earlier discussions 

a A random sample consists of items which have been selected from a 
  population in a way which ensures that all items in that 
  population have an exactly equal chance of being in the sample.  
  See section for an explanation of the statistical concept 
  of "population".                                   

in the literature (see also the comments by Fienberg (1975) on     
McKinlay's paper and remarks by Jacobsen (1972) about 
epidemiologically determined exposure/effect relationships). 

    The fact that it is not usually possible to achieve truly 
random sampling in epidemiological studies determines that reliance 
is placed on the hope that the data are quasi-random.  Therefore, 
every effort has to be made to avoid biases in the way that the 
material is selected for study (a question of design, organization, 
and conduct of the work) and attempts must be made to detect biases 
that are unavoidable.  Data processing and descriptions should have 
been pursued with an alert eye open for artefacts and impossible 
data.  Model building should reflect the realities of the 
situation, as determined from data descriptions, so that the 
analyses can search for intercorrelations between hypothesized 
explanatory variables and other interfering factors that may have 
distorted the structure of the data.  Estimates of effects may be 
adjusted accordingly, whenever possible, using assumptions that can 
be tested from the data themselves.  In some cases, quantitative 
statements can be made about the degree to which biases may have 
affected results.  Conclusions based on the application of 
statistical methods can then be qualified by warnings about the 
restricted range of their validity. 

    These caveats underline the importance of the suggestion in 
the introduction to this chapter:  the interpretation of 
epidemiological data should be an activity involving all members of 
the study team, not just statisticians.  Results from statistical 
analyses must be considered in the wider, extrastatistical context 
of the biological and environmental phenomena that are being 
investigated (Merkov, 1979). 

    The need for an interdisciplinary approach becomes particularly 
urgent when attempting to draw conclusions from statistically 
significant associations between an environmental factor and some 
index of disease.  Can that association be interpreted legitimately 
as evidence that the environmental factor caused the disease?  The 
question touches on fundamental and controversial philosophical 
issues that have been debated for centuries; no universally accepted 
formula has yet been devised that resolves the difficulty. 
Nevertheless, several authorities have made suggestions on how to 
assess the plausibility of a causal interpretation of an 
epidemiologically-determined association (e.g., US Public Health 
Service, 1964; Hill, 1965; MacMahon & Pugh, 1970; Merkov, 1979).  
Examples of efforts to apply these ideas in practice include studies 
of smoking and lung cancer (US Public Health Service, 1964), air 
pollution and health effects (Lave & Seskin, 1977), and cadmium 
poisoning and Itai-Itai disease (Shigematsu, 1978).  Some of the 
principles involved are discussed below. 

    In the first place, it is important to rule out the most 
obvious statistical artefacts:  spurious correlations of the kind 
discussed in section; lack of attention to confounding 
variables; population selection effects; and other biases common in 
observational studies.  A causal interpretation is generally more 

plausible, if it can be demonstrated that such possible distortions 
are unlikely to have affected results, and, if the observed effect 
on health tends to occur after exposure to the suspected agent, 
rather than before such exposure.  The latter  desideratum may be 
difficult to achieve from results based on retrospective (or even 
from cross-sectional) study designs, if the effect on health is a 
chronic condition with a well-recognized latency period (e.g., 

    Second, the so-called strength of the association should be 
considered.  In an epidemiological context, this refers to how 
frequently the health effect is observed contiguously with the 
hypothesized environmental factor.  However impressive the 
statistical significance of an association, that is, however low 
the probability that it is due simply to chance, an interpretation 
that the environmental agent caused the effect on health will be 
less attractive if the same effect also occurs frequently in the 
absence of the suspected causal factor. 

    The strength of an association will be reflected in the 
magnitude of the estimated relative risk or some other statistical 
index of association that may be appropriate.  But of course, that 
index is calculated generally from a particular sample of data.  
If other well-designed studies fail to demonstrate similar 
associations, then this weakens the suggestion that the association 
found can be regarded as a demonstration of cause and effect.  To 
strengthen the case for causality, consistency in the association 
in different studies, circumstances, and population groups should 
be sought. 

    Such consistency is analogous to the concept of stability of 
regression relationships, which was discussed in section  
If the suspected environmental factor can be expressed in the form 
of a continuous exposure variable, and, if the corresponding fitted 
regression model is reasonably convincing, then this demonstration 
of an exposure/effect relationship is powerful evidence supporting 
a causal interpretation. 

    Confidence that the apparent stability of an exposure/effect 
relationship is real and not artefactual will be strengthened 
further, if it can be shown, after the survey has been completed, 
that a reduction in the level of the pollutant concerned does 
indeed lead to the reduced effect predicted by the fitted model.  
Intensified preventive measures, which often follow publication of 
results from epidemiological studies, may thus provide further very 
strong quasi-experimental evidence supporting the original 
hypothesis of cause and effect. 

    A cause-effect explanation for a statistically demonstrated 
association will not be convincing, if the explanation appears to 
conflict seriously with other well established knowledge concerning 
the biological processes involved or the environmental conditions 
posited as having led to the health effects.  Coherence of an 
hypothesized causal model with previously established scientific 
facts will tend to strengthen the argument. 

    All the above presupposes that the statistical evidence 
demonstrating an association is convincing, i.e., that the 
probability of having made the Type I error is acceptably low.  But 
the existence of the Type II error should not be forgotten.  A 
strong association (say an apparently high relative risk) may 
justify concern, even if it fails to reach statistical significance 
at some arbitrarily chosen level.  Results of this kind should 
prompt at least two questions before conclusions are formulated.  
First, what precisely was the level of significance found?  
Obviously  P ca. 0.45 is less impressive than  P ca. 0.06, although 
both may be reported as not significant at the 5% level.  It is a 
mistake to treat such very different situations as if they were 
equivalent.  In research-oriented studies, signficance testing 
should be regarded as a tool to help quantify evidence, not as a 
procedural panacea for solving scientific problems.  The second 
question should be "what, approximately, is the probability of 
having made the Type II error?".  The absence of statistical 
significance when beta is high (perhaps because of constraints on 
the numbers of observations that were possible in the survey 
situation) should be interpreted more cautiously than would the 
same results in circumstances where beta is low. 

    Conclusions derived from a critical review of results along 
these lines will remain sterile unless they are communicated 
clearly and convincingly.  The planning of the study should have 
allowed adequate time and resources for reporting. 

6.5.  Reporting

6.5.1.  The variety of epidemiological reports

    The potential audience for reports on results from an 
epidemiological study of environmental factors is likely to be 
wider than is usual in many other areas of scientific activity.  
Individuals and groups directly involved, the sponsors of the work, 
public authorities, trade unionists, politicians, the press, and of 
course scientific colleagues - all may be anxious to hear about the 
findings in more or less detail and to learn of the conclusions 

    Pressures for statements and summaries before the data have        
been properly processed and analysed may have to be resisted           
(section  Premature release of material before it has        
been verified may mislead and confuse rather than assist.  Priority    
should therefore have been given to sound data description (section    
6.3) as distinct from formal analysis.  The format of such 
descriptions should have been planned in advance, and the material 
should be made available to members of the study team, for comment, 
before modelling begins, even if some tables or graphs have to be 

a The epidemiological or environmental hygiene implications of the 
  results should be distinguished clearly from any clinical 
  observations on individuals, which will have been communicated in 
  confidence to the persons concerned, soon after examination.

endorsed with warnings that they are subject to correction on 
details.  Clear captions and legends and perhaps short explanatory 
notes may be all that is required for the first informal reports, 
to stimulate comments and suggestions that will be of help in the 
subsequent more detailed statistical work.   

    Continuation of such useful interaction between various parts 
of the study team may require additional papers and notes, of           
increasing complexity, during the course of the analysis.  Placing   
results and ideas on paper in a form that can be understood by       
others is a useful aid to clear thinking, and the accumulated 
papers will be helpful in the preparation of the main report.               

6.5.2.  Main scientific report

    The funding agency or other organization that has requested or 
supported the work may have specified the format and amount of     
detail required for the final, as distinct from the interim        
progress, reports.  Some useful guidelines for the documentation of
epidemiological studies have been suggested by the "Epidemiology   
Work Group of the Interagency Regulatory Liaison Group" (1981).  In
any case, it is sensible to assume that the sponsors will wish to  
receive a workmanlike scientific document that states clearly how  
the resources were used.  The report should include:               
-   an explanation of the objectives of the study and comments     
    on why they were chosen;                                       

-   a description of how the work was carried out;
-   presentation of the results; and              
-   discussion on their implications.             

    These elements and their sequence correspond to the four        
familiar main headings in scientific papers:  Introduction; Methods;
Results; Discussion.                                                

    The introductory explanation as to why the study was undertaken 
may refer to previously published material on the same subject, to 
justify the new effort that is being reported.  Authors will wish 
to highlight gaps in existing knowledge or new hypotheses that have 
been stimulated by earlier work.  However, there is no merit in 
prefacing the report with a formal bibliography or mini-review 
article, unless this has been specifically requested; it will tend 
to obscure the main purpose of the section, i.e., a clear statement 
of objectives.  Methods

    The description of methods used should obey the norms of good 
scientific reporting - clarity and precision of expression and 
economy of words. 

    In an epidemiological report, this section should include 
unambiguous statements:  describing the study design that was used; 
defining the group(s) being studied with respect to location and 
time; recording how the defined individuals were identified; and 
explaining what steps were taken to establish that identification 
of the defined groups was complete and accurate.  The methods for 
defining and ascertaining the health outcomes (end points, indices) 
must be clearly reported, as well as the methods for determining 
environmental exposures. 

    There should be references to the equipment, instruments, and 
questionnaires that were used, how they were used, by whom, and 
after what training.  Precautions taken to avoid errors should also 
be mentioned.  Non-standard methods should be described in 
sufficient detail to permit reproduction by others.  Lengthy 
descriptions (for instance, a newly developed questionnaire) may be 
conveniently placed in an appendix. 

    The mechanics of data processing and verification procedures 
should be summarized, from the point that data were collected or 
generated to the point where analyses began. 

    The Methods section may include a brief account of findings 
from any pilot trials that were conducted, particularly with regard 
to estimates of intra- and interinstrument and observer 
variability.  If this material is extensive and interesting in its 
own right, then it is better placed in an appendix.  Results

    Results usually start with a statement of how many in the 
defined group(s) were in fact identified and surveyed (response 
rates and follow-up rates).  These facts should be presented in a 
form that enables the reader to assess the degree to which lack of 
response or other gaps in the data might have biased or distorted 
the results.  This means that the distribution of individuals or 
items of data, which were not included, should be reported in 
relation to the main variables of interest, as far as possible, 
(e.g., age, or index of disease at an earlier survey).  Such tables 
may be the basis for estimating the likely or the maximum possible 
bias in the results caused by the omissions.  But note that tests 
of significance on apparent differences between those surveyed and 
those not surveyed with respect to explanatory variables are 
usually irrelevant.a 

a Whether or not the gaps in the data arose by chance, by 
  negligence, or by force of circumstances might be of interest to
  those responsible for auditing the survey procedures, with a view 
  perhaps to improving them in the future.  The absence of a 
  significant differ difference in these circumstances is certainly 
  no assurance that there is negligible bias in the results.

    Data description follows, based on careful selection of 
available tables, graphs, and summary statistics, in order to 
convey the overall trends and the complexities that are relevant to 
the study objectives, including negative results.  Very detailed 
tables containing many rows and columns may be included as 
appendices, with appropriate summary tables in the text, if the 
data they contain are central to the main research objective.  The 
principle is to ensure that all the essential information is 
documented and available for detailed study by the interested 
specialist reader without disturbing the flow of the text 
describing the results.  Summary statistics should be referred to 
with standard errors or other measure of dispersion, and with 
results from significance tests where appropriate.b 

    Narrative, tabular and graphical descriptions of results from 
more complex modelling of the data are often easier to follow if the 
model used is stated explicitly rather than just in general terms, 
perhaps as a footnote.  Graphs of fitted curves are generally 
unhelpful, unless they are accompanied by descriptions of the 
dispersion of results around the estimated lines, either in the form 
of confidence limits, or standard errors, or scattergrams of the 

    Tabular presentation of estimates of parameters, such as means 
or regression coefficients, are best accompanied by standard errors 
rather than " P-values".  The former allow the latter to be 
derived by the reader; the converse is not true. 

    The test accompanying the tables and graphs might explain why   
particular models were considered to be appropriate and why others 
were not investigated at all, but an effort should be made to       
separate the statement of results (with comments to clarify the   
meaning of the analysis) from the discussion of their implications.  Discussion

    This should include a critical review of imperfections in the 
design as originally conceived and as realized in practice.  It 
should draw attention to weaknesses in the data revealed by the 
description and analysis, with particular attention to possible 
population selection effects or other biases.  The discussion may 
include comparisons of results with those in earlier reports and 
any apparent contradictions should be commented on.  The discussion 
may include hypotheses generated from the work described and future 
steps to be taken. 
b For instance, whether or not a difference in age between a group 
  being studied and a control group is due to chance is immaterial.  
  The important point is to adjust estimates of response to allow 
  for this factor.  See also the last paragraph of section  
  A useful guide to deciding whether or not a significance test is 
  appropriate is to ask: "What is the null hypothesis?  What is the 
  alternative hypothesis?  Are they relevant to the research 

    The authors' main conclusions may be recapitulated in the last 
paragraph of the discussion, or perhaps in a short additional 
section.  Abstract

    A report of this kind should always include an  Abstract  
rather than a  Summary.  An abstract means a concise statement, 
usually not exceeding one or two pages, which refers to the key 
points in each of the four main sections of the report; i.e., why 
the work was done, how it was done, the findings, and the 
conclusions.  The word "Summary" is used more loosely to include, at 
the one extreme, the kind of abstract described above, and at the 
other extreme, a single sentence paraphrasing the main conclusion or 
even just elaborating on the title of the report.  Neither of these 
options are recommended for the kind of report considered here, but 
shorter papers that are intended for submission to scientific 
journals should comply with the convention of the journal concerned. 

6.5.3.  Non-technical reports

    A well-prepared scientific report should be clear and pleasant 
to read, however long or complex the argument.  Unnecessary jargon 
will have been excised from early drafts and obscurities will have 
been clarified as a result of discussion and criticism from 
colleagues.  But the detailed documentation that is proper for such 
a report will often be of little interest or will be unintelligible 
(because of its technical complexity) to many non-specialist who 
are very anxious to understand what was found.  However, the 
abstract may not provide sufficient information to satisfy them or 
it may not convey the essential message adequately to those 
unaccustomed to reading scientific reports.  It will often be 
useful therefore, and sometimes it will be essential, to prepare an 
additional short paper that recapitulates and supplements the 
abstract without necessitating familiarity with the various 
technical disciplines typically involved in epidemiological studies 
of environmental problems. 

    The difficulty of preparing such a short document should not be 
underestimated.  A balance has to be struck between providing all 
the information strictly necessary to sustain the argument 
rigorously on the one hand, and oversimplifying the issues on the 
other.  The busy people who will want to rely primarily on this 
paper will not wish to be blinded by science; but neither will they 
wish to be patronized.  It will be helpful to seek criticisms of 
drafts not only from professional colleagues, but also from others 
unconnected with epidemiology. 

    The effort required is usually justifiable, since it is better 
that an attempt to summarize the findings in non-technical language 
should be made by someone thoroughly familiar with the complexities 
and nuances of the particular study, rather than by someone 
unconnected with the project, however competent. 

    A report of this kind will often be the key document informing 
public debate and for briefing policy makers.  These matters are 
discussed in Chapter 7. 


ANDERSON, J.A.  (1972)  Separate sample logistic discrimina-
tion.  Biometrika,  59: 19-35.

ARMITAGE, P.  (1975)  The use of the cross-ratio in aetio-
logical surveys. In Gani, J., ed.  Perspectives in probability
 and statistics.  London, Academic Press, pp. 349-355.

BERRY, G.  (1974)  Longitudinal observations, their usefulness
and limitations, with special reference to the forced
expiratory volume.  Bull. Physiol. Pathol. Resp.,  10: 643-655.

BERRY, G.  (l977)  Discussion of the paper by Professors
Liddell and McDonald and Dr Thomas.  J. R. Stat. Soc., A,  140:

S.A.  (1979)  Asbestosis: a study of dose-response relation-
ships in an asbestos textile factory.  Br. J. ind. Med.,  36:

BRESLOW, N.  (1977)  Some statistical models useful in the
study of occupational mortality. In: Whittemore, A., ed.
 Environmental health: Quantitative methods - Proceedings of a
 Conference on Environmental Health, Alta, Utah, 5-9 July 1976,
Philadelphia, Society of Industrial and Applied Mathematics,
pp. 88-103.

BRESLOW, N. & DAY, N.E.  (1980)   Statistical methods in cancer
 research, Vol. I: The analysis of case-control studies.  Lyons:
International Agency for Research on Cancer. 338 pp (IARC
Scientific Publication No. 32).

SABAI, C.  (1978)  Estimation of multiple relative risk
functions in matched case-control studies.  Am. J. Epidemiol.,
108: 199-307.

BROWN, C.C.  (1975)  On the use of indicator variables for
studying the time-dependence of parameters in a response-time
model.  Biometrics,  31: 863-872.

BYAR, D.P. & MANTEL, N.  (1975)  Some interrelationships among
the regression coefficient estimates arising in a class of
models appropriate to response-time data.  Biometrics,  31:

CAREY, G.C.R., DAWSON, T.A.J., & MERRETT, J.D.  (1967)  Daily
changes in ventilatory capacity in smokers and in non-smokers.
 Br. J. prev. soc. Med.,  21: 86-89.

CAREY, G.C.R., DAWSON, T.A.J., & MERRETT, J.D.  (1968)  Adden-
dum to daily changes in ventilatory capacity in smokers and
non-smokers.  Br. J. prev. soc. Med.,  22: 59.

THOMPSON, D.J., WOLTER, D.W., & McCARROL, J.R.  (1969)  Air
pollution, weather and illness in a New York population.  Arch.
 environ. Health,  18: 523-530.

COHEN, J.  (1968)  Multiple regression as a general data-
analytic system.  Psychol. Bull.,  70: 426-443.

COLE, T.J.  (1975)  Linear and proportional regression models
in the prediction of ventilatory function.  J. R. Stat. Soc.,
 A,  138: 297-338.

COLE, T.J.  (1977)  Height standardization of ventilatory
function.  Proc. R. Soc. Med.,  70: 165-166.

COX, D.R.  (1968)  Notes on some aspects of regression
analysis.  J. R. Stat. Soc., A,  131: 265-279.

COX, D.R.  (1970)   The analysis of binary data,  London,
Methuen, 142 pp.

COX, D.R.  (1972)  Regression models and life tables (with
discussion).  J. R. Stat. Soc., B,  34: l87-220

COX, D.R. & SNELL, E.J.  (1974)  The choice of variables in
observational studies.  Appl. Stat.,  23: 51-59.

DENISTON, D.L. & ROSENSTOCK, I.M.  (1973)  The validity of
non-experimental designs for evaluating health services.
 Health Serv. Rep.,  88: 153-164.

DIXON, W.J. & MASSEY, F.J., Jr  (1957)   Introduction to
 statistical analysis,  Second Edition, New York, McGraw-Hill,
488 pp.

DRAPER, N.R. & SMITH, H.  (1966)   Applied regression analysis,
New York, Wiley, 407 pp.

DUNCAN, A.J.  (1959)   Quality control and industrial
 statistics,  revised ed., Homewood, Illinois, Irwin, Inc., 946

DYER, A.R.  (1975a)  An analysis of the relationship of
systolic blood pressure, serum cholersterol, and smoking to
14-year mortality in the Chicago Peoples Gas Company study -
I. Total mortality in exponential-Weibull model.  J. chron.
 Dis.,  28: 565-570.

DYER, A.R.  (1975b)  An analysis of the relationship of
systolic blood pressure, serum cholesterol, and smoking to
14-year mortality in the Chicago Peoples Gas Company study -
II. Coronary and cardiovascular-renal mortality in two
competing risk models.  J. chron. Dis.,  28: 571-578.

ENTERLINE, P.E.  (1976)  Pitfalls in epidemiological re-
search.  J. occup. Med.,  18: 150-156.

GROUP  (1981)  Guidelines for documentation of epidemiologic
studies (with comments by the Joint Committee on Governmental
Affairs of the Epidemiology Section of the American Public
Health Association, and the Society for Epidemiologic
Research).  Am. J. Epidemiol.,  114: 609-618.

FAREWELL, V.T.  (1977)  The combined effect of breast cancer
risk factors.  Cancer,  40: 931-936.

FIENBERG, S.E.  (1975)  Comment on the paper by McKinlay.  J.
 Am. Stat. Assoc.,  70: 521-523.

FINNEY, D.J.  (1975)  Numbers and data.  Biometrics,  31:

FLEISS, J.L.  (1981)   Statistical methods for rates and
 proportions,  2nd ed., New York, Wiley.

FLEISS, J.L. & STROUT, P.E.  (1977)  The effects of measure-
ment errors on some multivariate procedures.  Am. J. Pub.
 Health,  67: 1188-1191.

 The natural history of chronic bronchitis and emphysema.
Oxford, Oxford University Press, 272 pp.

GAIL, M.H., LUBIN, J.H., & RUBINSTEIN, L.V.  (1981)  Likeli-
hood calculations for matched case-control studies and surv-
ival studies with tied death times.  Biometrika,  68: 703-707.

GARDNER, J.J. & HEADY, J.A.  (1973)  Some effects of
within-person variability in epidemiological studies.   J.
 chron. Dis.,  26:  781-795.

HANKEY, B.F. & MYERS, M.H.  (1971) Evaluating differences in
survival between two groups of patients.   J. chron. Dis.,  24:

(1974)   Comparability of nine methods for monitoring NO2 in
 ambient air,  Washington DC, US Environmental Protection
Agency, pp. 1-107 (Publication No. EPA-650/4-74-012).

HILL, A.B. (1965)  The environment and diseases: associations
and causation.  Proc. R. Soc. Med., Sect. occup. Med.,  58:

HOCKING, R.R.  (1976)  The analysis and selection of variables
in linear regression.  Biometrics,  32: 1-49.

M.  (1982)  Coalworkers' simple pneumoconiosis and exposure to
dust at 10 British coalmines.  Br. J. ind. Med.,  39: l20-127.

JACOBSEN, M.  (1972)  Evidence of dose-response relation in
pneumoconiosis (2).  Trans. Soc. Occup. Med.,  22: 88-94.

JACOBSEN, M.  (1975a)  Quantifying radiological changes in
simple pneumoconiosis.  J. R. Stat. Soc. C,  24: 229-249.

JACOBSEN, M.  (1975b)  Discussion of Dr Cole's Paper.  J. R.
 Stat. Soc., A,  138: 331-322.

JOHNSTON, J.  (1963)   Econometric methods,  New York,
McGraw-Hill, 300 pp.

KASAP, H.S. & CORKHILL, R.T.  (1973)  A review of multivariate
statistical techniques and the treatment of missing values in
epidemiological research. In:  Uses of epidemiology in planning
 health services. Proceedings of the sixth International
 Scientific Meeting, International Epidemiological Association,
 Primoste, Yugoslavia, 29 August-3 September, 1971,  Belgrade,
Savremena Administracija, pp. 225-234.

KING, L.J.  (1969)   Statistical analysis in geography,
Englewood Cliffs, NJ, Prentice-Hall, 288 pp.

KULLBACK, S. & CORNFIELD, J.  (1976)  An information theoretic
contingency table analysis of the Dorn study of smoking and
mortality.  Comput. biomed. Res.,  9: 409-437.

(1978)  On the utility of proportional mortality analysis.  J.
 chron. Dis.,  31: 15-22.

LEWIS, D.K.  (1981)  Matching in epidemiologic studies:
validity and efficiency considerations.  Biometrics,  37:

(1972)  Changes in ventilatory function in coal miners after a
work shift.  Arch. environ. Health,  24: 204-208.

LAVE, L.B. & SESKIN, E.P.  (1977)   Air pollution and human
 health,  Baltimore, Johns Hopkins University Press, 368 pp.

(1982)  Significance of intra-individual changes in maximum
expiratory flow volume and peak expiratory flow measurements.
 Chest,  81: 566-570.

LEUNG, H.M. & KUPPER, L.L. (1981) Comparisons of confidence
intervals for attributable risk.  Biometrics,  37: 292-302.

LIDDELL, F.D.K.  (1972)  Validation of classifications of
pneumoconiosis.  Ann. N.Y. Acad. Sci.,  200: 527-551.

LOVE, R.G. & MILLER, B.G.  (1982)  Longitudinal study of lung
function in coalminers.  Thorax,  37: 193-197.

MACK, T.M., PIKE, M.C., & CASAGRANDE, J.T.  (1977)  Epidemio-
logic methods for human risk assessment. In: Hiatt, H.H.,
Watson, J.D., & Winsten, J.A., ed.  Origins of human cancer.
Cold Spring Harbor, NY, Cold Spring Harbor Laboratory, Vol. 4,
pp. 1749-1763.

MCKERROW, C.B. & ROSSITER, C.E.  (1968)  An annual cycle in
the ventilatory capacity of men with pneumoconiosis and of
normal subjects.  Thorax,  23: 340-349.

MCKINLAY, S.M.  (1975)  The design and analysis of the
observational study - a review.  J. Am. Stat. Assoc.,  70:

MCLINTOCK, J.S., RAE, S., & JACOBSEN, M.  (1971)  The attack
rate of progressive massive fibrosis. In: Walton, W.H., ed.
 Inhaled particles III,  Surrey, Unwin, Vol. 2, pp. 933-952.

MACMAHON, B. & PUGH, T.F. (1970)   Epidemiology, principles
 and methods,  Boston, Little, Brown. 376 pp.

MANTEL, N.  (1963)  Chi-square tests with one degree of
freedom: extensions of the Mantel-Haenszel procedure.  J. Am.
 Stat. Assoc.,  58: 690-700.

MANTEL, N.  (1966)  Evaluation of survival data and two new
rank order statistics arising in its consideration.  Cancer
 Chemother. Rep.,  50: 163-170.

MANTEL, N.  (l973)  Synthetic retrospective studies and
related topics.  Biometrics,  29: 479-486.

MANTEL, N. & BYAR, D.P.  (1974)  Evaluation of response-time
data involving transient states: an illustration using heart
transplant data.  J. Am. Stat. Assoc.,  69: 81-86.

MANTEL, N. & HAENSZEL, W.  (1959)  Statistical aspects of the
analysis of data from retrospective studies of disease.  J.
 Natl Cancer Inst.,  22: 719-748.

MAXWELL, A.E.  (1961)   Analysing qualitative data,  London,
Methuen, 163 pp.

MAXWELL, A.E.  (1970)  Multivariate analysis. In: Holland,
W.W., ed.  Data handling in epidemiology,  London, Oxford
University Press, pp.149-168.

MERKOV, A.M.  (1979)   The health of population and methods for
 its study,  Moscow, Statistika, 334 pp (in Russian).

MIETTINEN, O.S.  (1970)  Estimation of relative risk from
individually matched series.  Biometrics,  26: 75-86.

MIETTINEN, O.S.  (1972)  Standardization of risk ratios.  Am.
 J. Epidemiol.,  96: 383-388.

MIETTINEN, O.S.  (1974)  Confounding and effect modification.
 Am. J. Epidemiol.,  100: 350-353.

MIETTINEN, O.S.  (1976)  Stratification by a multivariate
confounder score.  Am. J. Epidemiol.,  104: 609-620.

MURPHY, E.A. & ABBEY, H. (1967)  The normal range - a common
misuse.  J. chron. Dis.,  20: 79-88.

NELDER, J.A.  (1975)   General linear interactive modelling,
Oxford, Royal Statistical Society, 58 pp.

OLDHAM, P.D.  (1962)  A note on the analysis of repeated
measurements of the same subjects.  J. chron. Dis.,  15: 969-977.

OLDHAM, P.D.  (1970)  The uselessness of normal values. In:
Arcangeli, P., ed.,  Introduction to the definition of normal
 values for respiratory function in man,  Turin, Panminerva
Medica, pp. 49-56.

OLDHAM, P.D. & ROSSITER, C.E.  (1965)  Mortality in coal-
workers' pneumoconiosis related to lung function: a prospec-
tive study.  Br. J. ind. Med.,  22: 93-100.

PASTERNACK, B.S. & SHORE, R.E.  (1977)  Statistical methods
for assessing risk following exposure to environmental car-
cinogens. In: Whittemore, A.  ed.,  Environmental health:
 quantitative methods - Proceedings of a Conference on Environ-
 mental Health, Alta, Utah, 5-9, July, 1976,  Phildelphia,
Society of Industrial and Applied Mathematics, pp.49-71.

PASTERNACK, B.S., SHORE, R.E. & ALBERT, R.E.  (1977)  Occupa-
tional exposure to chloromethyl ethers.  J. occup. Med.,  19:

PETO, R.  (1978a)  Control of industrially induced cancers.
 Environ. Health Perspect.,  22: 153-154.

PETO, R. (1978b)  Carcinogenic effects of chronic exposure to
very low levels of toxic substances.  Environ. Health
 Perspect.,  22: 155-159.

PRENTICE, R.L. & BRESLOW, N.E.  (1978)  Retrospective studies
and failure time models.  Biometrika,  65: 153-158.

PRENTICE, R.L. & PYKE, R.  (1979)  Logistic disease incidence
models and case-control studies.  Biometrika,  66: 403-411.

REGISTRAR GENERAL (1978)   Occupational mortality. The
 Registrar General's decennial supplement for England and
 Wales, 1970-1972.  London, HMSO (Series DS No. 1).

ROBINSON, C.V. & UPTON, A.C.  (1978)  Competing-risk analysis
of leukemia and non-leukemia mortality in X-irradiated, male
RF mice.  J. Natl Cancer Inst.,  6O: 995-1007.

D.D., & WALTON, W.H.  (1973)  Role of dust in the working
environment in development of chronic bronchitis in British
coal miners.  Br. J. ind. Med.,  30: 217-226.

SAMUELS, M.L.  (1981)  Matching and design efficiency in
epidemiological studies.  Biometrika, 68: 577-588.

SCOTT, A.J.  (1964)   Optimizing statistical analysis:  Data
 screening and preconditioning.  Evanston, Northwestern Univer-
sity (USGRR Document Number AD-433-551).

SANDALA, M.G. & ZVINJACKOVSKIJ, Ja.I.  (1981)  [Identification
of the role of separate factors in the complex exposure of
human populations to environmental factors.]  Gig. i Sanit.,  9:
4-6 (in Russian).

SHIGEMATSU, I., ed.  (1978)  [Epidemiology - methodology for
clinicians.] Tokyo, Kodansha, pp. 198-204 (in Japanese).

SIEGEL, D.G. & GREENHOUSE, S.W.  (l973)  Validity in estimat-
ing relative risk in case control studies.  J. chron. Dis.,  26:

N.E.  (1981)  Multivariate conditional logistic analysis of
stratum matched case-control studies.  Appl. Stat.,  30: 190-197.

SMITH, P.G. & DAY, N.E.  (1981)  Matching and confounding in
the design and analysis of epidemiological case-control
studies. In: Bithell, J.F., Coppi, R., ed.  Perspectives in
 medical statistics: Proceedings of the European Symposium on
 Medical Statistics, Rome 1980,  London, Academic Press, pp.

STEBBINGS, J.H., Jr  (1971a)  Chronic respiratory disease
among non-smokers in Hagerstown, Maryland. II. Problems in the
estimation of pulmonary function values in epidemiological
surveys.  Environ. Res.,  4: 163-192.

STEBBINGS, J.H., Jr  (1971b)  Chronic respiratory disease
among non-smokers in Hagerstown, Maryland. III. Social class
and chronic respiratory disease.  Environ. Res.,  4: 213-232.

STEBBINGS, J.H., Jr  (1972)  A survey of respiratory disease
among New York City postal and transit workers. III. Anthropo-
metric, smoking, occupational, and ethnic variables affecting
the FEV1 among white males.  Environ. Res.,  5: 451-466.

STEBBINGS, J.H., Jr  (1973)  A survey of respiratory disease
among New York City postal and transit workers. IV. Racial
differences in the FEV1.  Environ. Res.,  6: 147-158.

STEBBINGS, J.H., Jr  (1974)  Fractional carbon monoxide uptake
in an employed population.  Thorax,  29: 505-510.

SUITS, D.B.  (1957)  Use of dummy variables in regression
equations.  J. Am. Stat. Assoc.,  52: 548-551.

TARONE, R.E.  (1975)  Tests for trend in life table analysis.
 Biometrika,  62: 679-682.

THOMAS, D.G.  (1975)  Exact and asymptotic methods for the
combination of 2 x 2 tables.  Comput. biomed. Res.,  8: 423-446.

TIETJEN, G.L. & MOORE, R.H.  (1972)  Some Grubbs-type
statistics for the detection of several outliers.
 Technometrics,  14: 583-597.

TRUETT, J., CORNFIELD, J., & KARREL, W.  (1967)  A multi-
variate analysis of the risk of coronary heart disease in
Framingham.  J. chron. Dis.,  20: 511-524.

US PUBLIC HEALTH SERVICE  (1964)   Smoking and health.  New
York, Van Nostrand, 387 pp (Report of the Advisory Committee
to the Surgeon-General of the Public Health Service. PHS Pub.
No. 1103).

WALTER, S.D.  (1976)  The estimation of attributable risk in
health research.   Biometrics, 32: 829-849.

WESOLOWSKY, G.D.  (1976)   Multiple regression and analysis of
 variance,  New York, Wiley, 291 pp.

WHITTEMORE; A.A.  (1977)  The age distribution of human cancer
for carcinogenic exposures of varying intensity.  Am. J.
 Epidemiol.,  106: 418-431.

WILLIAMS, E.J.  (1959)   Regression Analysis,  New York, John
Wiley & Sons, 214 pp.

ZVINJACKOVSKIJ, Ja.I.  (1979)  [The effects of the complex of
environmental factors on human morbidity.]  Gig. i Sanit.,
No. 4, 7-11 (in Russian).


7.1.  Introduction

    Epidemiological research provides a variety of information, the 
main aim of which is to answer the hypothesis as to whether there 
are associations between suspected environmental agents and the 
health of those exposed to them. 

    The objectives of epidemiological studies of the effects of 
environmental agents on health may be summarized as follows: 

(a) to provide decision makers and health workers with
    information needed for the establishment of health
    criteria and programmes for the control of pollution and
    other environmental hazards;

(b) to assist in evaluating the efficacy of preventive and
    control measures in protecting human health from
    environmental hazards and to improve the quality of life;

(c) To improve scientific knowledge of the effects on health
    of environmental conditions.

    Thus, epidemiology is expected to provide the bulk of the 
answers that the scientist, workman, employer, citizen, or 
government needs about the relationship between various aspects of 
environment and human health. 

    The objective of this chapter is to present guidance on the 
practical application of epidemiological information in the 
identification, management, and solution of some of the principal 
health problems related to environmental pollution by chemical and 
physical agents.  The use of such information could be different 
from one situation to another, in particular, in relation to the 
social, economical, and cultural differences of communities, but 
some general principles may be advanced. 

7.2.  Communication with the Public

    As discussed in Chapter 6, a report of the results of a study 
has to be prepared in precise and accurate scientific language on 
the one hand, but on the other hand, there will frequently be a 
need for a simplified presentation of results that can be addressed 
to policy makers, the public and, in some cases, the mass-media.  
The second type of presentation cannot tell the story with all the  
technical details, and this frequently brings in a risk of 
misinterpretation.  It is therefore important that any material 
written for the public in this way should be cleared by the 
scientists and epidemiologists involved in the basic study.  They 
must carefully check the degree of simplification tolerable so that 
the communication remains comprehensible to its intended audience.  
In such a presentation, limitations of the epidemiological approach, 
and the need for support from other studies, before firm conclusions 
can be reached, may have to be spelt out, as discussed below. 

7.3.  Important Features and Limitations of Epidemiological

    One of the most important features of epidemiological research 
is that it often leads to valuable findings, without elucidating 
the detailed mechanisms involved.  As an example, one can have 
confidence in the epidemiological relationship between lung cancer 
and cigarette smoking, although it has not yet been proved what the 
carcinogenic factors are.  The Minamata disease incidence (section 
7.4.2) is another good example of this kind, where the disease had 
been related to the consumption of contaminated fish and shellfish 
and preventive action had been taken, though the causative factor, 
methylmercury, became known much later. 

    However, this in turn means that an epidemiological association 
does not necessarily provide firm evidence of a cause/effect 
relationship.  Quantitative exposure information necessary for 
establishing exposure/effect relationships is always difficult to 
obtain.  Public health administrators and decision makers have to 
be made aware of these problems. 

    For a number of reasons, epidemiological studies of the working 
population are of importance and of relevance to the general 
population.  Exposures are often much higher in the workplace than 
those in the outside environment and, therefore, the health risk 
would be higher.  For this reason, the first group from which to 
seek information on an environmental impact will often be the 
workers.  However, it has to be borne in mind that a working 
population is partly selected, since it excludes children, elderly 
people, those whose health is already impaired or who may be 
hypersensitive to certain agents, and a proportion of women.  
Furthermore, exposure to a contaminant at work is limited in most 
cases to eight hours a day.  Therefore, extrapolation from work-
place results to the broader use of the data has to be done with 
great caution. 

    It should also be noted that existing or routinely-collected 
indices of morbidity or mortality are not, in general, of as much 
value for standard setting as specific studies directed to the 
effects under consideration.  A further complication is that it is 
unusual for exposure to be to one agent only; not only are the 
effects due to various pollutants or contaminants liable to be 
similar, but also there is increasing evidence of interactions 
between different agents.  Even the most sophisticated 
epidemiological techniques cannot always provide answers to these 

    In the past, epidemiologists have learned a lot from various 
emergencies caused by the accidental release of toxic chemicals 
(sometimes called a "natural experiment", see section 2.10) by 
following-up those who might have been exposed.  However, though 
major incidents attract distinct alarm and attention on many 
occasions, sometimes valuable data have not been secured, because 
of the inadequate organization for data collection in the early 
stages of the emergency. 

    This type of problem is well illustrated by an example where 
great difficultires were encountered in the epidemiological survey 
of the population exposed to 2,3,7,8-tetrachlorodibenzo- p-dioxin 
(TCDD) after an industrial accident in Seveso, Italy.  Many social, 
political, and ideological debates took place, particularly during 
the months immediately after the accident.  In addition, a great 
number of "scientific suggestions" for the protection of the 
population and land reclamation, often contradictory to each other, 
were sent to the regional health authorities and to the inhabitants 
of the polluted area, from several Italian and foreign research 
workers and institutions.  The decision of the Regional Council to 
improve local services met with many difficulties in finding 
technical and administrative manpower for efficient management of 
this critical situation.  All this and the uncertainties due to 
scant public health services dismayed the population and deterred 
them from participating as "guinea-pigs" in a big international 
laboratory.  They felt that they had the right in the investigation 
of this accident not to allow clinicians and epidemiologists to 
perform  any test on them they wanted, and not to allow politicians 
and social groups to use them to further their own interests 
(Homberger et al., 1979; Bisanti et al., 1980; Del Corno et al., 
1980; Favaretti et al., 1980). 

7.4.  Standard setting

    One of the most important fields in which epidemiological 
information is required is that of standard setting.  It is, 
however, important to emphasize that epidemiological data are only 
one of many factors that have to be taken into account when 
developing standards.  Even when an appropriate standard for any 
one country has to be considered, it is likely that the scientific 
information available from all sources, including toxicological 
research, clinical studies, epidemiological surveillance, and 
environmental monitoring, may still fall short of that which is 
essential for deriving an exposure/effect relationship.  Even if in 
an ideal situation such a relationship could be constructed, there 
must still be another layer of activity before a standard can be 
postulated.  That is, a standard has to take account, not only of 
the scientific data from which the exposure/effect relationship is 
derived, but also of the national resources to ensure compliance 
with the standard. 

    No human activity is devoid of all risk, and, in many cases, 
it is implicit that a threshold cannot be proved. Therefore, any 
standard involves some degree of risk, either to susceptible 
individuals or to a vulnerable proportion of the population.  Hence, 
it is essential that standard setting should be seen not only as a 
scientific exercise, but as something requiring the cooperation of 
those likely to be exposed and of the government agencies and 
managements responsible. 

    The value judgements to be made on the information available 
would be the responsibility of policy decision makers, and not of 
epidemiologists or doctors in their professional roles.  The role of 

an epidemiologist is to provide the best data and exposure/effect 
relationships possible and, in their interpretation, to underline 
the limits of confidence to be placed in them. 

    Where attempts are made to set standards in the international 
field, the difficulties are greater, because of cultural, political, 
geographical, and other differences.  Thus, there is always a danger 
of an agreement or apparent agreement being reached that cannot be 
applied effectively, in practice. 

7.4.1.  Factors in standard setting

    One of the basic questions on the assessment of health risks 
from chemicals and other environmental hazards is whether 
extrapolation from experimental animal studies is appropriate.  
Experience shows that extrapolation from animal to animal, even 
with the same species, is often difficult because of variations in 
factors such as nutrition, metabolism, or habits.  Extrapolation 
from animal to man is, in consequence, generally much less reliable.  
Consideration has to be given also to the adequacy of any "safety 
factor" that is introduced as a result of the incompleteness of the 
available information.  The development of laboratory tests for 
mutagenesis, with all the technical problems involved, leads to 
further questions about the continuity of the significance of 
findings from cell systems through various animal species to man.  
In the long run, however, only human studies will support or 
challenge the control limits that are adopted. 

    The importance of social and economic factors in the required 
decisions about acceptable standards is well illustrated by an 
example on the problem of exposure to arsenic through drinking-
water in Mexico, in an area where the water supply is limited.  In 
this case, if assurance could be given that there would be no 
significant health risk to the population through the food chain, 
water with a content of arsenic unacceptable for human beings could 
be used for agricultural purposes and possibly for cattle.  The 
economic significance of such action would be great, since it might 
avoid the necessity of bringing water with a lower arsenic content 
from distant hydrologic basins.  It would be an oversimplification 
to consider that, under all circumstances, lower levels of a 
standard are better for man (Castellano Alvarado et al., 1964; 
Sanchez de la Fuente, 1976). 

    Fluoridation of drinking-water is another example that has long 
been controversial.  This health measure was introduced in Canada 
in 1945, and today some 45% of those on public water supplies 
receive fluoridated water (Canada, Health and Welfare, 1978).  
Allegations that fluoridation increased cancer rates prompted an 
epidemiological study of the cancer mortality data from some 79 
groups of municipalities throughout Canada (Canada, Health and 
Welfare, 1977).  Comparisons of the death rates from all types of 
cancer, for the period 1954-73, in some 58% of the Canadian 
population did not show any appreciable differences between 
municipalities with fluoridated and non-fluoridated water supplies.  
Nor were any significant differences apparent between death rates 

from all types of cancer when compared within the same group of 
municipalities prior to and after fluoridation.  This is an 
instance where an environmental policy decision to fluoridate 
public water supplies, the benefit of which had been demonstrated 
by an epidemiological study, was further supported and reinforced 
by a population study based on disease incidence. 

    There are obviously great difficulties in establishing that no 
effects exist.  However, it is of great importance that negative 
evidence should be made available, with indications of the degree of 
confidence that can be placed on the results, since a summation of 
marginally positive results might create mistaken impressions and 
mislead those concerned with decision making. 

7.4.2.  Interim standards

    In one sense, all standards are interim, since they have to be 
reviewed at intervals, but there are also occasions when action 
levels have to be set to limit exposures while further studies are 
pursued.  Epidemiological studies are often unable to provide the 
unequivocal evidence required by decision makers.  Similarly, since 
adverse effects may in some cases only become manifest after a long 
latent period, interim standards for new or newly-introduced 
substances might have to be maintained for extended periods, before 
appropriate epidemiological data become available. 

    In applying epidemiological studies in the establishment of 
working assumptions about a disease of unknown cause, the 
experience gained in the initial stages of Minamata disease taught 
valuable lessons about possible approaches.  A patient with an 
undiagnosed cerebral disease was seen in the paediatric clinic of a 
hospital at Minamata City, Japan, in early May 1956.  A doctor in 
the clinic remembered four similar patients in the recent past.  He 
thought that this was a sign of an epidemic outbreak of an unknown, 
unusual cerebral disease and notified the Minamata Health Centre.  
An epidemiological study was initiated by the Health Centre with 
the cooperation of the local medical society and the city health 
department.  This was reported to the Department of Health of 
Kumamoto Prefecture, and then to the Ministry of Health and 
Welfare.  The Medical School of Kumamoto University organized a 
study group on Minamata disease in August 1956 (Study Group of 
Minamata Disease, 1968).  In November 1956, an initial conclusion 
of the epidemiological studies was presented suggesting that this 
disease was caused by long-term exposure to a common causative 
actor, which was assumed to be polluted fish and shellfish in 
Minamata Bay.  Another important and interesting finding was an 
abnormally high mortality rate in the cat population, with 
associated cerebral disorders similar to those in man. 

    Although the findings of the epidemiological studies were still 
preliminary, and although no exact cause had been identified at that 
time, considering the grave health damage caused in the community, 
the Prefectural Governor issued a probibition order on sales of fish 
and shellfish harvested in Minamata Bay.  This is a typical example 
of the application of epidemiological findings to the decision of 
public health administrators in a health crisis in a community. 

    The second outbreak of Minamata disease was reported along the 
Agano River in Niigata Prefecture, Japan, in May 1965 (Special 
Research Team, 1967).  In the light of widespread concern about the 
mercury pollution, nation-wide studies on the mercury and methyl-
mercury contents of fish and of the hair of the general human 
population and workers in mercury-handling industrial plants were 
conducted by the Ministry of Health and Welfare.  Although 
scientific evidence sufficient to establish tolerable limits of 
mercury and alkylmercury had not yet been obtained from 
epidemiological and other studies, the Ministry of Health and 
Welfare was urged to take the necessary action to set up 
provisional guidelines for monitoring, surveillance, and control of 
these compounds.  These guidelines were developed, on the basis of 
the results of studies in Minamata, Agano, and from the nationwide 
survey (Ministry of Health and Welfare, Japan, 1968).  It was only 
in 1972 that the Joint FAO/WHO Expert Committee issued provisional 
tolerable weekly intakes of total mercury and methylmercury (WHO, 

    These examples illustrate the need for preventive action before 
all the scientific facts, including epidemiological evidence, become 
available and the subsequent need for more detailed studies. 

7.5.  Assessment of Effectiveness of Control Measures Taken

    Once the environmental risk, to which a population group is 
exposed, is determined and certain corrective measures have been 
taken, it is useful to conduct epidemiological studies to see 
whether the corrective measures taken have proved to be effective, 
and whether the effects on health or the risk of exposure have been 
reduced in the expected manner.  The following is an examplea to 
illustrate the use of an epidemiological study in this respect. 

    In a region in Mexico, chromium salts were entering the 
environment through an inappropriate dumping arrangement for 
hundreds of tonnes of solid wastes, resulting in the salts leaching 
into an underground water supply.  The corrective measures taken 
fundamentally consisted in prohibiting the dumping of wastes in 
order to stop them entering the water supply and in continuously 
recycling within the industrial process.  Also, at solid waste 
disposal sites, necessary preventive measures were taken.  Drinking-
water, which was not unduly contaminated, was brought in from 
elsewhere.  A second epidemiological study was organized, a few 
years later, to evaluate the benefits that these actions had had on 
the population.  The study included medical examinations of a 
representative sample of the affected population.  The levels of 
chromium compounds in urine were also determined.  At the same time, 
environmental measurements were conducted to measure chromium 
concentrations in soil, in water samples from deep wells, and in the 
effluent to the air from the factory. From all this information, it 
was found that the corrective measures implemented were effective. 
a   Based on the contribution from Dr B. R. Ordonez,
    Autonomous Metropolitan University, Mexico City, Mexico.

7.6.  Policy of Openness

    Some of the main uses of epidemiological information have been 
described and illustrated.  An attempt has been made also to show 
the limitations implicit in the epidemiological method.  Nowhere 
are these difficulties more obvious than in the consideration of 
multifactorial disease and the setting of appropriate standards for 
environmental factors to which individuals may be particularly 
sensitive.  Decisions have to be taken about the appropriateness of 
safety factors in the standard-setting process and such decisions 
are not purely scientific by any means.  In many cases, the 
crudeness of the assessment of exposure is also a serious matter, 
and it is important to avoid creating a false impression of 
precision. This means that, where no threshold levels are known, or 
where measurement is unreliable, no statement of absolute safety 
can be issued in relation to any effects.  If there is no threshold 
for carcinogenic effect, the only way of ensuring that additional 
cases would not occur under any circumstances would be to ban the 
material concerned.  There are usually many side-issues and other 
consequences involved in banning a substance, or replacing it with 
something else, which may again be adverse to the health of the 

    In recent years, philosophical considerations arising from such 
questions have led to an extensive literature on the subject of the 
relative risks of various aspects of human activity (Knox, 1975; 
Reissland & Harries, 1979).  The expression "acceptable risk" has 
also been used, though it appears to evade the question of who 
decides to accept it and on whose behalf.  While it is unlikely that 
a true balancing of relative risks or a true understanding of what 
the term "acceptable" risk means will ever be achieved, there are 
certain principles which, to the epidemiologist, the standard 
setter, the economist, and the administrator, must be made very 

    It is the responsibility of the leader of a study team to make 
results with their proper interpretation available to the study 
participants, the public, policy and decision makers, and to the 
scientific community.  Sometimes, hasty conclusions may have been 
drawn from imperfect studies or from misinterpretation of existing 
results, which may confuse the public and the decision makers.  The 
scientist must clearly indicate the unsatisfactory nature of such 
conclusions in dialogues with the public and decision makers.  It 
may be stated that, in certain situations, conflicting interests 
make it difficult to transmit results to both the public and 
administrators.  It follows that, as far as possible, all those who 
are involved contribute to the dialogue leading to prevention. 

    There are differences of opinion about the role of an 
epidemiologist in relation to political, economic, and other 
spheres of activity.  It would appear that two positions are 
possible, and these are not, in fact, incompatible.  In the first 
place, the epidemiologist is responsible for analysing and studying 
the available scientific evidence and arriving at as many definite 
conclusions as possible.   Any reservations that may be held about 

the firmness of the inferences should be included in the statements.  
If the scientific position is incorrect, everything else that 
follows will be incorrect also. 

    Once a statement has been prepared about the relation of a 
measurement of biological effect to a quantum of exposure, the 
broader dialogue must proceed.  In this dialogue, the epidemiologist 
has a role as an expositor, recognizing, however, that the evidence 
being presented is only one factor.  When there are discussions 
outside the realm of epidemiology, such as those on economic or 
social factors, the epidemiologist can speak only as a citizen with 
no more authority than any other citizen.  A failure to recognize 
this difference has led undoubtedly to friction in the past. 

    Unfortunately, non-scientific political factors can lead to a 
blurring of scientific evidence, and unreal and unjustified alarm 
can possibly arise.  It would appear that the only way of dealing 
with these problems, systematically and correctly, is by a declared 
policy of openness on the part of the scientific community and by as 
much exchange of scientific information as can be arranged 
effectively.  It must be hoped that there is a scientific integrity 
among those responsible for measurements, assessments, and 


ZAMBRELLI, E.  (1980)  Experience from the accident of Seveso.
 Acta. Morpho. Acad. Sci. Hung.,  28: 131-157.

CANADA, HEALTH AND WELFARE  (1977)   Fluoridation and Cancer,
Ottawa, Department of National Health and Welfare, 58 pp

CANADA, HEALTH AND WELFARE  (1978)   Fluoridation in Canada as
 of December 31, 1976,  Ottawa, Department of National Health
and Welfare, 26 pp (78-EHD-19).

ALVAREZ ACEVEDO, J.  (1964)  [Epidemiological Study of Arsenic
in the Miguel Aleman and Eduardo Guerra Colonies in Torréon,
Coahuila.]  Salud Publ. Mex.,  6: 375-385 (in Spanish).

[Distribution of chloracne cases in the area of Seveso,
polutted by TCDD.]  Quad. Aggiorn. Reg. Lombardia,  6: 195-224
(in Italian).

[Chloracne and clinical impairement in children of 0 to 14
years of age exposed to TCDD in the area of Seveso.]  Quad.
 Aggiorn. Reg. Lombardia,  6: 225-243 (in Italian).

The Seveso accident: its nature, extent and consequences.  Ann.
 Occup. Hyg.,   22 (4): 327-367.

 control measures for mercury poisoning;  August 1968, Tokyo,
Ministry of Health & Welfare.

KNOX, E.G.  (1975)  Negligible risks to health.  Commun.
 Health,  6: 244

REISSLAND, B. & HARRIES, V.  (1979)  A scale for measuring
risks.  New Scientist,  83: 809-11.

SANCHEZ DE LA FUENTE, E.  (1976)   Collective intoxication of
 cattle in the Comarca Lagunera  (Technical Information from the
Preventive Programs Service of Torréon's Health Center,
Torréon, Coahuila, Mexico).

SPECIAL RESEARCH TEAM  (1967)   Report of investigation of
 Niigata mercury poisoning incidence.  Tokyo, Ministry of Health
& Welfare.

STUDY GROUP OF MINAMATA DISEASE  (1968)   Minamata Disease,
Study Group of Minamata Disease, Kumamoto, Japan, Kumamoto
University, 338 pp.

WHO  (1972)   Evaluation of certain food additives and the
 contaminants: mercury, lead, and cadmium.  Geneva, World Health
Organization (WHO Technical Report Series No.505).

    See Also:
       Toxicological Abbreviations