IPCS INCHEM Home


    UNITED NATIONS ENVIRONMENT PROGRAMME
    INTERNATIONAL LABOUR ORGANISATION
    WORLD HEALTH ORGANIZATION





    INTERNATIONAL PROGRAMME ON CHEMICAL SAFETY



    Environmental Health Criteria 214




    HUMAN EXPOSURE ASSESSMENT


    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 Organization, or the World Health Organization.


    First draft prepared by Dr D. L. MacIntosh, University of Georgia,
    Athens, GA, USA and Professor J. D. Spengler, Harvard University,
    Boston, MA, USA



    Published under the joint sponsorship of the United Nations
    Environment Programme, the International Labour Organization, and the
    World Health Organization, and produced within the framework of the
    Inter-Organization Programme for the Sound Management of Chemicals.





    World Health Organization
    Geneva, 2000

         The International Programme on Chemical Safety (IPCS),
    established in 1980, is a joint venture of the United Nations
    Environment Programme (UNEP), the International Labour Organization
    (ILO), and the World Health Organization (WHO).  The overall
    objectives of the IPCS are to establish the scientific basis for
    assessment of the risk to human health and the environment from
    exposure to chemicals, through international peer review processes, as
    a prerequisite for the promotion of chemical safety, and to provide
    technical assistance in strengthening national capacities for the
    sound management of chemicals.

         The Inter-Organization Programme for the Sound Management of
    Chemicals (IOMC) was established in 1995 by UNEP, ILO, the Food and
    Agriculture Organization of the United Nations, WHO, the United
    Nations Industrial Development Organization, the United Nations
    Institute for Training and Research, and the Organisation for Economic
    Co-operation and Development (Participating Organizations), following
    recommendations made by the 1992 UN Conference on Environment and
    Development to strengthen cooperation and increase coordination in the
    field of chemical safety.  The purpose of the IOMC is to promote
    coordination of the policies and activities pursued by the
    Participating Organizations, jointly or separately, to achieve the
    sound management of chemicals in relation to human health and the
    environment.

    WHO Library Cataloguing-in-Publication Data

    Human exposure assessment.

    (Environmental health criteria ; 214)

         1.Environmental monitoring - methods   2.Environmental exposure
         3.Models, theoretical   4.Data collection - methods    
         5.Toxicity tests
         I.International Programme on Chemical Safety II.Series

         ISBN 92 4 157214 0                  (NLM Classification: QT 162)
         ISSN 0250-863X

         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 2000

         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 letters.

    CONTENTS

    ENVIRONMENTAL HEALTH CRITERIA FOR HUMAN EXPOSURE ASSESSMENT

    PREAMBLE

    ABBREVIATIONS

    FOREWORD

    1. DEFINING EXPOSURE

         1.1. Introduction
         1.2. Defining exposure
              1.2.1. Exposure and exposure concentration
              1.2.2. Exposure estimation by integration and averaging
              1.2.3. Exposure measurements and models
              1.2.4. Exposure in the context of an environmental health
                     paradigm
         1.3. Elements of exposure assessment
         1.4. Approaches to quantitative exposure assessment
         1.5. Linking exposure events and dose events
         1.6. Summary

    2. USES OF HUMAN EXPOSURE INFORMATION

         2.1. Introduction
         2.2. Human exposure information in environmental epidemiology
         2.3. Human exposure information in risk assessment
              2.3.1. Risk allocation for population subgroups or
                     activities
              2.3.2. Population at higher or highest risk
         2.4. Human exposure information in risk management
         2.5. Human exposure information in status and trend analysis
         2.6. Summary

    3. STRATEGIES AND DESIGN FOR EXPOSURE STUDIES

         3.1. Introduction
         3.2. Study design
         3.3. Sampling and generalization
         3.4. Types of study design
              3.4.1. Comprehensive samples
              3.4.2. Probability samples
              3.4.3. Other sample types
         3.5. Exposure assessment approaches
              3.5.1. Direct approaches to exposure assessment
                      3.5.1.1  Personal monitoring of inhalation exposures
                      3.5.1.2  Personal monitoring of dietary exposures
                      3.5.1.3  Personal monitoring of dermal absorption
                               exposures

              3.5.2. Indirect approaches to exposure assessment
                      3.5.2.1  Environmental monitoring
                      3.5.2.2  Models as an indirect approach to assessing
                               exposure
                      3.5.2.3  Questionnaires as an indirect approach to
                               assessing exposure
         3.6. Summary

    4. STATISTICAL METHODS IN EXPOSURE ASSESSMENT

         4.1. Introduction
         4.2. Descriptive statistics
              4.2.1. Numerical summaries
              4.2.2. Graphical summaries
                      4.2.2.1  Histograms
                      4.2.2.2  Cumulative frequency diagrams
                      4.2.2.3  Box plots
                      4.2.2.4  Quantile-quantile plots
                      4.2.2.5  Scatter plots
         4.3. Probability distributions
              4.3.1. Normal distribution
              4.3.2. Lognormal distribution
              4.3.3. Binomial distribution
              4.3.4. Poisson distribution
         4.4. Parametric inferential statistics
              4.4.1. Estimation
              4.4.2. Measurement error and reliability
              4.4.3. Hypothesis testing and two-sample problems
              4.4.4. Statistical models
                      4.4.4.1  Analysis of variance and linear regression
                      4.4.4.2  Logistic regression
              4.4.5. Sample size determination
         4.5. Non-parametric inferential statistics
         4.6. Other topics
         4.7. Summary

    5. HUMAN TIME-USE PATTERNS AND EXPOSURE ASSESSMENT

         5.1. Introduction
         5.2. Methods
              5.2.1. Activity pattern concepts
                      5.2.1.1  Time allocation parameters
                      5.2.1.2  Microenvironment parameters
                      5.2.1.3  Intensity of contact
              5.2.2. Surrogates of time-activity patterns
              5.2.3. Data collection methods
         5.3. Potential limitations
              5.3.1. Activity representativeness
              5.3.2. Validity and reliability
              5.3.3. Inter- and intra-person variability
         5.4. Summary

    6. HUMAN EXPOSURE AND DOSE MODELLING

         6.1. Introduction
         6.2. General types of exposure model
         6.3. Environmental media and exposure media
         6.4. Single-medium models
              6.4.1. Outdoor and indoor air
              6.4.2. Potable water
              6.4.3. Surface waters
              6.4.4. Groundwater
              6.4.5. Soil
         6.5. Multiple-media modelling
              6.5.1. Inter-media transfer factors
                      6.5.1.1  Diffusive partition coefficients
                      6.5.1.2  Advective partition coefficients
              6.5.2. Exposure factors
              6.5.3. Multiple-media/multiple-pathway models
         6.6. Probabilistic exposure models
              6.6.1. Variability
              6.6.2. Uncertainty
              6.6.3. Implementing probabilistic exposure models
         6.7. A generalized dose model
         6.8. Physiologically based pharmacokinetic models
         6.9. Validation and generalization
         6.10. Summary

    7. MEASURING HUMAN EXPOSURES TO CHEMICALS IN AIR, WATER AND FOOD

         7.1. Introduction
         7.2. Air monitoring
              7.2.1. Gases and vapours
                      7.2.1.1  Passive samplers
                      7.2.1.2  Active samplers
                      7.2.1.3  Direct-reading instruments
              7.2.2. Aerosols
              7.2.3. Semivolatile compounds
              7.2.4. Reactive gas monitoring
         7.3. Water
              7.3.1. Factors influencing water quality
              7.3.2. Water quality monitoring strategies
              7.3.3. Sample collection
         7.4. Assessing exposures through food
              7.4.1. Duplicate diet surveys
              7.4.2. Market basket or total diet surveys
              7.4.3. Food consumption
                      7.4.3.1  Food diaries
                      7.4.3.2  24-h recall
                      7.4.3.3  Food frequency questionnaires
                      7.4.3.4  Meal-based diet history
                      7.4.3.5  Food habit questionnaires
              7.4.4. Contaminants in food
         7.5. Summary

    8. MEASURING HUMAN EXPOSURE TO CHEMICAL CONTAMINANTS IN SOIL AND
         SETTLED DUST

         8.1. Introduction
         8.2. Selected sampling methods
              8.2.1. Soil
                      8.2.1.1  Surface soil collection
                      8.2.1.2  Soil contact and intake measurements
              8.2.2. Settled dust
                      8.2.2.1  Wipe sampling methods
                      8.2.2.2  Vacuum methods
                      8.2.2.3  Sedimentation methods
         8.3. Sampling design considerations
              8.3.1. Concentration and loading
              8.3.2. Collection efficiency
         8.4. Sampling strategies
         8.5. Summary

    9. MEASURING BIOLOGICAL HUMAN EXPOSURE AGENTS IN AIR AND DUST

         9.1. Introduction
         9.2. House dust mites
              9.2.1. Air sampling for house dust mites
              9.2.2. Dust sampling for house dust mites
              9.2.3. Available methods of analysis for house dust mites
                      9.2.3.1  Mite counts
                      9.2.3.2  Immunochemical assays of dust mite
                               allergens
                      9.2.3.3  Guanine determination
              9.2.4. Mite allergens
         9.3. Allergens from pets and cockroaches
              9.3.1. Air sampling for allergens from pets and cockroaches
              9.3.2. Dust sampling for allergens from pets and
                      cockroaches
              9.3.3. Available methods of analysis
              9.3.4. Typical allergen concentrations
         9.4. Fungi
              9.4.1. Air sampling for fungi
              9.4.2. Settled dust for fungi
              9.4.3. Available methods of analysis for fungi in air
                      9.4.3.1  Total counts of viable and non-viable
                               fungal particles
              9.4.4. General considerations for fungi
         9.5. Bacteria (including actinomycetes)
              9.5.1. Air sampling for bacteria
              9.5.2. Dust sampling for bacteria
              9.5.3. Available methods of analysis for bacteria
                      9.5.3.1  Total count of viable and non-viable
                               bacteria
                      9.5.3.2  Viable bacteria
                      9.5.3.3  Endotoxins

         9.6. Pollen
              9.6.1. Air sampling for pollen
              9.6.2. Dust sampling for pollen
              9.6.3. Available methods of analysis for pollen in air
              9.6.4. General considerations for pollen sampling
         9.7. Summary

    10. ASSESSING EXPOSURES WITH BIOLOGICAL MARKERS

         10.1. Introduction
         10.2. General characteristics
         10.3. Considerations for use in environmental exposure assessment
              10.3.1. Toxicokinetics and toxicodynamics
              10.3.2. Biological variability
              10.3.3. Validation of biological markers
              10.3.4. Normative data
         10.4. Advantages of biological markers for exposure assessment
              10.4.1. Characterizing inter-individual variability
              10.4.2. Efficacy of use
              10.4.3. Internal exposure sources
         10.5. Limitations of biological markers for exposure assessment
              10.5.1. Source identification
              10.5.2. Biological variability and altered exposure response
              10.5.3. Participant burden
              10.5.4. Biosafety
         10.6. Media available for use
              10.6.1. Blood
              10.6.2. Urine
              10.6.3. Exhaled breath
              10.6.4. Saliva
              10.6.5. Keratinized tissue (hair and nails)
              10.6.6. Ossified tissue
                      10.6.6.1 Teeth
                      10.6.6.2 Bone
              10.6.7. Breast milk
              10.6.8. Adipose tissue
              10.6.9. Faeces
              10.6.10. Other media
         10.7. Summary

    11. QUALITY ASSURANCE IN EXPOSURE STUDIES

         11.1. Introduction
         11.2. Quality assurance and quality control
         11.3. Elements of a quality assurance programme
         11.4. Quality assurance programme
              11.4.1. Organization and personnel
              11.4.2. Record-keeping and data recording
              11.4.3. Study plan and standard operating procedures
              11.4.4. Collection of samples
              11.4.5. Equipment maintenance and calibration
              11.4.6. Internal audit and corrective action

         11.5. Quality control/quality assurance for sample measurement
              11.5.1. Method selection and validation
                      11.5.1.1 Accuracy
                      11.5.1.2 Precision
                      11.5.1.3 Sensitivity
                      11.5.1.4 Detection limits
              11.5.2. Internal quality control
                      11.5.2.1 Control charts
              11.5.3. External quality control
              11.5.4. Reference materials
         11.6. Quality assurance and control issues in population-based
              studies
         11.7. Summary

    12. EXAMPLES AND CASE STUDIES OF EXPOSURE STUDIES

         12.1. Introduction
         12.2. Exposure studies
         12.3. Air pollution exposure studies
              12.3.1. Particle studies
              12.3.2. Carbon monoxide
              12.3.3. Nitrogen dioxide
              12.3.4. Ozone
              12.3.5. Combined exposure studies
              12.3.6. Assessing ambient pollution impacts indoors
              12.3.7. Volatile organic compounds
              12.3.8. Commuter exposures
         12.4. Exposures and biomarkers
              12.4.1. Exposure to lead and cadmium
              12.4.2. Exposure to furans, dioxins and polychlorinated
                      biphenyls
              12.4.3. Exposure to volatile organic compounds and urinary
                      metabolites
         12.5. Exposure to contaminants in drinking-water
         12.6. Exposure to microbes
         12.7. Exposure studies and risk assessment
              12.7.1. The German Environmental Survey
              12.7.2. The National Human Exposure Assessment Survey
              12.7.3. Windsor, Canada exposure and risk study
              12.7.4. Pesticide exposure study
              12.7.5. Czech study of air pollution impact on human health

    REFERENCES

    RÉSUMÉ

    RESUMEN
    

    NOTE TO READERS OF THE CRITERIA MONOGRAPHS

         Every effort has been made to present information in the criteria
    monographs as accurately as possible without unduly delaying their
    publication.  In the interest of all users of the Environmental Health
    Criteria monographs, readers are requested to communicate any errors
    that may have occurred to the Director of the International Programme
    on Chemical Safety, World Health Organization, Geneva, Switzerland, in
    order that they may be included in corrigenda.



                               *     *     *



         A detailed data profile and a legal file can be obtained from the
    International Register of Potentially Toxic Chemicals, Case postale
    356, 1219 Châtelaine, Geneva, Switzerland (telephone no.
    + 41 22 - 9799111, fax no. + 41 22 - 7973460, E-mail irptc@unep.ch).



                               *     *     *



         This publication was made possible by grant number
    5 U01 ES02617-15 from the National Institute of Environmental Health
    Sciences, National Institutes of Health, USA, and by financial support
    from the European Commission.



    Environmental Health Criteria

    PREAMBLE

    Objectives

         In 1973 the WHO Environmental Health Criteria Programme was
    initiated with the following objectives:

    (i)   to assess information on the relationship between exposure to
          environmental pollutants and human health, and to provide
          guidelines for setting exposure limits;

    (ii)  to identify new or potential pollutants;

    (iii) to identify gaps in knowledge concerning the health effects of
          pollutants;

    (iv)  to promote the harmonization of toxicological and
          epidemiological methods in order to have internationally
          comparable results.

         The first Environmental Health Criteria (EHC) monograph, on
    mercury, was published in 1976 and since that time an ever-increasing
    number of assessments of chemicals and of physical effects have been
    produced.  In addition, many EHC monographs have been devoted to
    evaluating toxicological methodology, e.g., for genetic, neurotoxic,
    teratogenic and nephrotoxic effects.  Other publications have been
    concerned with epidemiological guidelines, evaluation of short-term
    tests for carcinogens, biomarkers, effects on the elderly and so
    forth.

         Since its inauguration the EHC Programme has widened its scope,
    and the importance of environmental effects, in addition to health
    effects, has been increasingly emphasized in the total evaluation of
    chemicals.

         The original impetus for the Programme came from World Health
    Assembly resolutions and the recommendations of the 1972 UN Conference
    on the Human Environment.  Subsequently the work became an integral
    part of the International Programme on Chemical Safety (IPCS), a
    cooperative programme of UNEP, ILO and WHO.  In this manner, with the
    strong support of the new partners, the importance of occupational
    health and environmental effects was fully recognized. The EHC
    monographs have become widely established, used and recognized
    throughout the world.

         The recommendations of the 1992 UN Conference on Environment and
    Development and the subsequent establishment of the Intergovernmental
    Forum on Chemical Safety with the priorities for action in the six
    programme areas of Chapter 19, Agenda 21, all lend further weight to
    the need for EHC assessments of the risks of chemicals.

    Scope

         The criteria monographs are intended to provide critical reviews
    on the effect on human health and the environment of chemicals and of
    combinations of chemicals and physical and biological agents.  As
    such, they include and review studies that are of direct relevance for
    the evaluation.  However, they do not describe  every study carried
    out.  Worldwide data are used and are quoted from original studies,
    not from abstracts or reviews.  Both published and unpublished reports
    are considered and it is incumbent on the authors to assess all the
    articles cited in the references.  Preference is always given to
    published data.  Unpublished data are only used when relevant
    published data are absent or when they are pivotal to the risk
    assessment.  A detailed policy statement is available that describes
    the procedures used for unpublished proprietary data so that this
    information can be used in the evaluation without  compromising its
    confidential nature (WHO (1990) Revised Guidelines for the Preparation
    of Environmental Health Criteria Monographs. PCS/90.69, Geneva, World
    Health Organization).

         In the evaluation of human health risks, sound human data,
    whenever available, are preferred to animal data.  Animal and  in 
     vitro studies provide support and are used mainly to supply evidence
    missing from human studies.  It is mandatory that research on human
    subjects is conducted in full accord with ethical principles,
    including the provisions of the Helsinki Declaration.

         The EHC monographs are intended to assist national and
    international authorities in making risk assessments and subsequent
    risk management decisions.  They represent a thorough evaluation of
    risks and are not, in any sense, recommendations for regulation or
    standard setting.  These latter are the exclusive purview of national
    and regional governments.

    Content

         The layout of EHC monographs for chemicals is outlined below.

    *  Summary -- a review of the salient facts and the risk evaluation of
       the chemical

    *  Identity -- physical and chemical properties, analytical methods

    *  Sources of exposure

    *  Environmental transport, distribution and transformation

    *  Environmental levels and human exposure

    *  Kinetics and metabolism in laboratory animals and humans

    *  Effects on laboratory mammals and  in vitro test systems

    *  Effects on humans

    *  Effects on other organisms in the laboratory and field

    *  Evaluation of human health risks and effects on the environment

    *  Conclusions and recommendations for protection of human health and
       the environment

    *  Further research

    *  Previous evaluations by international bodies, e.g., IARC, JECFA,
       JMPR

    Selection of chemicals

         Since the inception of the EHC Programme, the IPCS has organized
    meetings of scientists to establish lists of priority chemicals for
    subsequent evaluation.  Such meetings have been held in: Ispra, Italy,
    1980; Oxford, United Kingdom, 1984; Berlin, Germany, 1987; and North
    Carolina, USA, 1995. The selection of chemicals has been based on the
    following criteria: the existence of scientific evidence that the
    substance presents a hazard to human health and/or the environment;
    the possible use, persistence, accumulation or degradation of the
    substance shows that there may be significant human or environmental
    exposure; the size and nature of populations at risk (both human and
    other species) and risks for environment; international concern, i.e.
    the substance is of major interest to several countries; adequate data
    on the hazards are available.

         If an EHC monograph is proposed for a chemical not on the
    priority list, the IPCS Secretariat consults with the Cooperating
    Organizations and all the Participating Institutions before embarking
    on the preparation of the monograph.

    Procedures

         The order of procedures that result in the publication of an EHC
    monograph is shown in the flow chart.  A designated staff member of
    IPCS, responsible for the scientific quality of the document, serves
    as Responsible Officer (RO).  The IPCS Editor is responsible for
    layout and language.  The first draft, prepared by consultants or,
    more usually, staff from an IPCS Participating Institution, is based
    initially on data provided from the International Register of
    Potentially Toxic Chemicals, and reference data bases such as Medline
    and Toxline.

         The draft document, when received by the RO, may require an
    initial review by a small panel of experts to determine its scientific
    quality and objectivity.  Once the RO finds the document acceptable as
    a first draft, it is distributed, in its unedited form, to well over
    150 EHC contact points throughout the world who are asked to comment
    on its completeness and accuracy and, where necessary, provide

    additional material.  The contact points, usually designated by
    governments, may be Participating Institutions, IPCS Focal Points, or
    individual scientists known for their particular expertise.  Generally
    some four months are allowed before the comments are considered by the
    RO and author(s).  A second draft incorporating comments received and
    approved by the  Director,  IPCS, is then  distributed to Task Group
    members, who carry out the peer review, at least six weeks before
    their meeting.

         The Task Group members serve as individual scientists, not as
    representatives of any organization, government or industry.  Their
    function is to evaluate the accuracy, significance and relevance of
    the information in the document and to assess the health and
    environmental risks from exposure to the chemical.  A summary and
    recommendations for further research and improved safety aspects are
    also required.  The composition of the Task Group is dictated by the
    range of expertise required for the subject of the meeting and by the
    need for a balanced geographical distribution.

         The three cooperating organizations of the IPCS recognize the
    important role played by nongovernmental organizations.
    Representatives from relevant national and international associations
    may be invited to join the Task Group as observers.  While observers
    may provide a valuable contribution to the process, they can only
    speak at the invitation of the Chairperson. Observers do not
    participate in the final evaluation of the chemical; this is the sole
    responsibility of the Task Group members.  When the Task Group
    considers it to be appropriate, it may meet  in camera. 

         All individuals who as authors, consultants or advisers
    participate in the preparation of the EHC monograph must, in addition
    to serving in their personal capacity as scientists, inform the RO if
    at any time a conflict of interest, whether actual or potential, could
    be perceived in their work.  They are required to sign a conflict of
    interest statement. Such a procedure ensures the transparency and
    probity of the process.

         When the Task Group has completed its review and the RO is
    satisfied as to the scientific correctness and completeness of the
    document, it then goes for language editing, reference checking, and
    preparation of camera-ready copy.  After approval by the Director,
    IPCS, the monograph is submitted to the WHO Office of Publications for
    printing.  At this time a copy of the final draft is sent to the
    Chairperson and Rapporteur of the Task Group to check for any errors.

         It is accepted that the following criteria should initiate the
    updating of an EHC monograph: new data are available that would
    substantially change the evaluation; there is public concern for
    health or environmental effects of the agent because of greater
    exposure; an appreciable time period has elapsed since the last
    evaluation.

         All Participating Institutions are informed, through the EHC
    progress report, of the authors and institutions proposed for the
    drafting of the documents.  A comprehensive file of all comments
    received on drafts of each EHC monograph is maintained and is
    available on request.  The Chairpersons of Task Groups are briefed
    before each meeting on their role and responsibility in ensuring that
    these rules are followed.


    FIGURE

    WHO TASK GROUP ON HUMAN EXPOSURE ASSESSMENT

     Members 

    Dr J. Alexander, Department of Environmental Medicine, National
         Institute of Public Health, Folkehelsa, Torshov, Oslo, Norway

    Dr M. Berglund, Institute of Environmental Medicine, Division of
         Metals and Health, Karolinska Institute, Stockholm, Sweden

    Dr M. Dellarco, US Environmental Protection Agency,
         Washington, DC, USA

    Mrs B. Genthe, Environmentek, CSIR, Stellenbosch, South Africa

    Dr L. Gil, Department of Biochemistry, University of Chile -
         Faculty of Medicine, Casilla, Santiago, Chile

    Dr S. Goto, Department of Community Environmental Sciences,
         Institute of Public Health, Minato-ku, Tokyo, Japan

    Professor M. Jantunen, Department of Environmental Hygiene and
         Toxicology, National Public Health Institute, Kuopio, Finland

    Dr N. Künzli, Department of Environment and Health, Institute of
         Social and Preventive Medicine, University of Basel, Basel,
         Switzerland

    Dr D. MacIntosh, Environmental Health Science, University of
         Georgia, Athens, GA, USA

    Dr M. Morandi, Environmental Sciences, Houston School of Public
         Health, Houston Health Science Center, University of Texas,
         Houston, TX, USA

    Dr S. Pavittranon, National Institute of Health, Department of
         Medical Sciences, Bamrasnaradura Hospital, Nonthari, Thailand

    Dr N. Rees, Risk Assessment, Management and International
         Coordination Branch, Ministry of Agriculture, Fisheries and Food,
         London, United Kingdom

    Dr B. Schoket, Department of Biochemistry, National Institute of
         Environmental Health, "Fodor József" National Public Health
         Centre, Budapest, Hungary

    Dr L. Sheldon, US Environmental Protection Agency, National
         Research Laboratory, Research Triangle Park, NC, USA

    Professor J. D. Spengler, School of Public Health, Harvard
         University, Boston, MA, USA

    Dr P. Straehl, Swiss Federal Agency for Environment, Forestry and
         Landscape, Swiss Department of the Interior, Bern, Switzerland

     Observers 

    Mrs S. Munn, European Commission, European Chemicals Bureau,
         Environment Institute, Ispra (VA), Italy


     Secretariat 

    Mr C. Corvalan, Office of Global and Integrated Environmental
         Health, World Health Organization, Geneva, Switzerland

    Dr K. Gutschmidt, International Programme on Chemical Safety,
         World Health Organization, Geneva, Switzerland

    Dr M. Krzyzanowski, European Centre for Environment and
         Health, World Health Organization, Regional Office for Europe,
         Bilthoven Division, De Bilt, Netherlands

    Dr G. Moy, Food Safety, World Health Organization, Geneva,
         Switzerland

    Dr H. Tamashiro, Office of Global and Integrated Environmental
         Health, World Health Organization, Geneva, Switzerland

    Dr M. Younes, International Programme on Chemical Safety,
         World Health Organization, Geneva, Switzerland


    ENVIRONMENTAL HEALTH CRITERIA FOR HUMAN EXPOSURE ASSESSMENT

         A Task Group on the Environmental Health Criteria for Human
    Exposure Assessment met in Glion-sur-Montreux, Switzerland, from 16 to
    20 February 1998. Dr M. Younes, IPCS, welcomed the participants on
    behalf of the Manager, IPCS, and the three IPCS cooperating
    organizations (UNEP/ILO/WHO). The Task Group reviewed and revised the
    final draft of the monograph. In preparation for the final draft a
    review meeting was held at the National Institute of Health Sciences
    (NIHS), Tokyo, from 17 to 19 July 1996.

         The first draft was prepared by Dr D. L. MacIntosh, University of
    Georgia, USA and Professor J. D. Spengler, Harvard University, USA.

         Dr K. Gutschmidt was responsible officer in IPCS for the overall
    scientific content of the monograph and the organization for the
    meetings, and Ms K. Lyle (Sheffield, United Kingdom) was responsible
    for the technical editing of the monograph.

         The efforts of all who helped in the preparation and finalization
    of the monograph are gratefully acknowledged.

    ABBREVIATIONS

    ACGIH     American Conference of Governmental Industrial Hygienists
    ADD       average daily dose
    AI        acceptance intervals
    ALAD      Delta-aminolaevulinic acid dehydratase
    AMIS      Air Monitoring Information System
    ANOVA     analysis of variance
    AOAC      Association of Official Analytical Chemists
    ASTM      American Society for Testing of Materials
    CDF       chlorinated dibenzofurans; cumulative distribution function
    CFU       colony-forming units
    CI        confidence interval
    DG18      dichloran 18% diglycerol agar
    DVM       dust vacuum method
    EDTA      ethylenediamine tetra-acetic acid
    ELISA     enzyme-linked immunosorbent assays
    EPS       extracellular polysaccharides
    ETS       environmental tobacco smoke (exposure)
    EU        endotoxin unit
    FDA       US Food and Drug Administration
    FFQ       food frequency questionnaire
    GEMS      Global Environment Monitoring System
    GerES     German Environmental Survey
    GM        geometric mean
    GSD       geometric standard deviation
    HEAL      Human Exposure Assessment Location
    HPLC      high-pressure liquid chromatography
    HUD       US Department of Housing and Urban Development
    IAEA      International Atomic Energy Agency
    IAQ       internal air quality
    ISEA      International Society of Exposure Analysis
    ISO       International Organization for Standardization
    LADD      lifetime average daily dose
    LAL        Limulus amoebocyte lysate
    LOD       limit of detection
    LOQ       limit of quantification
    LWW       Lioy-Weisel-Wainman
    MAD       maximum allowable deviations
    MCS       multiple chemical sensitivity
    MDL       method detection limit
    MEA       malt extract agar
    NAAQS     National Ambient Air Quality Standard
    NHEXAS    National Human Exposure Assessment Survey
    NIOSH     National Institute for Occupational Safety and Health
    NTA       nitriloacetic acid
    OR        odds ratio
    PAH       polycyclic aromatic hydrocarbons
    PBPK      physiologically based pharmacokinetic (method)
    PCB       polychlorinated biphenyls
    PCDD      polychlorinated dibenzo- p-dioxin
    PCP       pentachlorophenol
    PDF       probability distribution function

    PEM       personal exposure monitor
    PMn       particulate matter with aerodynamic diameter <  n µm
    PTEAM     particle total exposure assessment methodology
    QA        quality assurance
    QC        quality control
    RAST      radioallergosorbent tests
    RIA       radioimmunoassay
    RSP       respirable particulate matter
    SAM       stationary outdoor monitor
    SBS       sick building syndrome
    SD        standard deviation
    SEM       scanning electron microscope
    SIM       stationary indoor monitor
    SOP       standard operating procedure
    SVOC      semivolatile organic compound
    TCCD      2,3,7,8-tetrachloro dibenzo- p-dioxin
    TDS       US FDA Total Diet Study
    TEQ       TCCD toxic equivalents
    TSP       total suspended particulates
    TWI       tolerable weekly intake
    UNEP      United Nations Environment Programme
    VOC       volatile organic compound
    XRF       X-ray fluorescence

    FOREWORD

         The International Programme on Chemical Safety (IPCS), launched
    in 1980, is a joint collaborative programme of the International Labor
    Organization (ILO), the United Nations Environment Programme (UNEP),
    and the World Health Organization (WHO); WHO is the Administrating
    Organization of the Programme. The two main roles of the IPCS are to
    establish the scientific health and environmental risk assessment
    basis for safe use of chemicals  (normative function) and to
    strengthen national capabilities for chemical safety  (technical 
     cooperation). In the field of methodology, the work of the IPCS aims
    at promoting the development, improvement, validation, harmonization
    and use of generally acceptable, scientifically sound methodologies
    for the evaluation of risks to human health and the environment from
    exposure to chemicals. The work encompasses the development of
    Environmental Health Criteria monographs on general principles of
    various areas of risk assessment covering various aspects related to
    risk assessment such as, in this publication, on exposure assessment.

         The WHO and the World Meteorological Organization coordinate the
    assessment of climate, urban air and water pollution, and health
    status of populations. These measures provide the indicator of trends
    and status.

         Until 1995, the basic source for internationally comparable urban
    air pollution data was the Global Environment Monitoring System
    (GEMS/Air) of UNEP and WHO. Started in 1974, shortly after the
    Stockholm Environment Conference, GEMS had built up a system that
    collected comparable ambient air pollution data in about 50 cities of
    35 countries, varied in geography and income (UNEP/WHO, 1988, 1992).
    Typically, sulfur dioxide and total suspended particulates (TSP) had
    been monitored in three stations of each city, one each in industrial,
    commercial, and residential zones. Later, GEMS also collected
    monitoring data for carbon monoxide, nitrogen dioxide, and lead, and
    made emissions estimates for all five pollutants. The results were
    published periodically by GEMS, and also often appeared in other
    periodic international data sets, such as those of the World Bank
    (World Bank, 1992), the World Resources Institute (World Resources
    Institute, 1992), the United Nations (UN ESCAP, 1990) and UNEP itself
    (UNEP, 1991).

         More recently, WHO created with the Air Management Information
    System (AMIS) the successor of GEMS/Air. Like GEMS/Air, AMIS provides
    air quality data for major and megacities. Data on sulfur dioxide,
    nitrogen dioxide, carbon monoxide, ozone, black smoke, suspended
    particulate matter, PM10, lead and others are available. AMIS also
    includes information on air quality management (WHO, 1997).

         Much of what is known about contaminants in food, soils, water
    and air has become available through WHO and UNEP publications. For
    more than 20 years WHO/UNEP has been promoting an appreciation for
    improved assessments of human exposures through training sessions,

    workshops, demonstration projects, and published methodologies and
    reports. Through a series of WHO-sponsored studies in every populated
    continent, the principles of human exposure assessment have been
    illustrated for indoor and outdoor air pollutants, food contamination
    and water. In 1984, after some background reports (e.g., UNEP/WHO,
    1982), WHO and UNEP conducted the Human Exposure Assessment Location
    (HEAL) Project, which facilitates research and information sharing
    among 10-15 institutions worldwide concerned with exposure assessment
    for a limited number of pollutants (Ozolins, 1989). Unfortunately,
    although providing important functions, the HEAL project has not had
    the mandate or anything approaching the resources required to actually
    make comparable international estimates of population exposures. HEAL
    projects, for the most part, have investigated exposures to
    conventional inorganic air pollutants such as carbon monoxide,
    nitrogen dioxide and general undifferentiated particle mass where
    inhalation is the primary route of exposures. However, the HEAL
    programme does offer examples of lead, cadmium and pesticide studies
    which illustrate multiple exposure pathways and demonstrate the
    necessity of extensive analytical training and quality programmes. An
    analytical quality control programme which involved all participating
    laboratories enabled reliable international comparisons of exposure
    despite differences in methodologies applied by the different
    laboratories.

         Preceding this criteria document the UNEP, FAO and WHO have been
    actively advancing the concepts and methodologies for human exposures.
    GEMS/Air, GEMS/Water and GEMS/Food are establishing the uniformity
    among data collected worldwide to establish national and international
    status and trends. These efforts, together with others, such as the
    Codex Committee on Pesticide Residues, the several Joint FAO/WHO
    Consultations on food consumption, pesticide residues, veterinary
    drugs, additives and chemical contaminants, have been developing the
    basis of quantitative assessment of human exposures and risk. Table 38
    (pg. 279) provides a listing of pertinent publications related to
    assessment of air, water and food contamination.

     Scope 

         This current criteria document on human exposure assessment
    presents in one publication the concepts, rationale, and statistical
    and procedural methodologies for human exposure assessment. The
    underpinnings of exposure assessment are the basic environmental and
    biological measurements found in the more familiar specialties of air
    and water pollution and food and soil sciences. Therefore, throughout
    this document readers are referred to other publications for technical
    details on instrumental and laboratory methods. This criteria document
    is intended for the community of scientific investigators inquiring
    about the human health consequences of contaminants in our
    environment. As such, this text will be of interest to physical
    scientists, engineers and epidemiologists. It is intended also for
    those professions involved in devising, evaluating and implementing
    policy with respect to managing the quality of environmental health,
    inclusive of air, water, food and soil. By necessity environment is

    defined broadly to include place, media, and activities where we
    humans encounter contaminants.

         Of primary concern in this document are those environmental
    contaminants that exist in various media as a consequence of direct or
    indirect human intention. We have included some biological agents that
    are "natural" but, through actions of irritation and allergy, can
    contribute to or cause morbidity and mortality as a result of
    inadequate building design and maintenance. We recognize that viral,
    bacterial and other biological agents in air, food, soil and water
    contribute significantly to the burden of disease worldwide. However,
    in the context of environmental exposure assessment the focus is on
    chemical contaminants and a few specific allergens that might
    contribute directly to disease or, in combination with biopathogens,
    alter susceptibility and expression of disease.

         To say that exposure assessment of environmental contaminants is
    exclusive of any population or location is, in principle, a
    contradiction. There are practical considerations, however, for
    identifying the industrial workplace as a separate domain.
    Administratively, many nations handle occupational health and safety
    concerns separately from the environment. The management of workplace
    hazards through well-established industrial hygiene practices of
    source control, ventilation and worker protection are widely
    recognized. This separation of workplace exposures from the general
    environmental exposure focus in this document is not hard and fast.
    Occupationally acquired contaminants can expose family members not
    working in the specific industry. Industrial control strategies that
    increase ventilation can adversely affect the neighbouring community.
    In many societies, commercial and residential use of property are
    integrated. Family operated business along congested streets means
    that contaminants generated in outdoors, indoors and workplaces are
    intermingled. Even where commercial and residential property are
    distinct, chemical and biological contaminants can lead to non-worker
    exposures.

         Information on human exposures has a well-recognized role as a
    corollary to epidemiology. But it is more than this, because
    understanding human exposures to environmental contaminants is
    fundamental to public policy. The adequacy of environmental mitigation
    strategies is predicated on improving or safeguarding human and
    ecological health. The public mandate for and acceptance of controls
    on emissions is first based on sensory awareness of pollution.
    Irritated airways, foul-smelling exhaust, obscuring plumes, oil slicks
    on water, dirty and foul-tasting water, and medical waste and debris
    on beaches are readily interpreted as transgressions against us and
    threaten commonly shared natural resources. As we enter the
    twenty-first century, we recognize that we, humans have had profound
    but often subtle impacts on the chemistry of the biosphere and
    lithosphere. Metals, organic compounds, particulate matter, and
    photochemically produced gases are widely dispersed, recognizing no
    geographic or political boundaries. Global markets, urbanization, and
    increased mobility have environmental contamination as a consequence.

    Assessing the quantities and distribution of potentially harmful
    contaminant exposures to human populations is a critical component of
    risk management. As long as disease prevention and health promotion
    are the principal tenets of public health, then assessing the levels
    of contaminant exposures in environmental and biological samples will
    be necessary.

         This book presents the methodologies for surveying exposures,
    analysing data and integrating findings with the ongoing national and
    global debate defining natural limits to human behaviour. It serves
    the cross-disciplinary needs of environmental managers, risk assessors
    and epidemiologists to learn something about the design, conduct,
    interpretation and value of human exposure studies of multimedia
    environmental contaminants. For investigators considering exposure
    studies, this book guides them to contemporary information on
    measurement of analysis methods and strategies.

         In Chapter 1 of the document the basic terms and concepts used in
    exposure assessment are defined. Similar understanding of terms used
    commonly among health assessors working in the different fields of
    air, water, soil and food sciences is a critical starting point in
    defining the emerging specialist area of exposure assessment.
    Application of exposure research and routine assessments to the
    information needs of risk managers, policy-makers and epidemiologists
    is established in Chapter 2. Discussion of these information needs is
    developed in Chapter 3, which presents the objectives for various
    study designs.

         Chapter 4 covers basic statistical concepts used in exposure
    assessment. The intent is to inform the reader of how statistical
    analysis is vital to all components of an exposure assessment. By
    examples and references the reader is directed to more substantial
    texts on study design, data analysis, modelling and quality control.

         Chapter 5 is devoted to a component of exposure assessment
    related to the collection and interpretation of human activity
    patterns. Information on how, where and when people contact
    potentially contaminant media is useful for data interpretation,
    establishing risk scenarios and identifying activities, locations and
    populations at differential risk. The emphasis here is primarily
    related to air pollution exposure studies. In the conduct of total
    multimedia exposure investigations or modelling analogous information
    is needed for the ingestion of water and food, as well as for dermal
    contact.

         Chapter 6 extends the concepts of the preceding chapters in
    discussing models for human exposure assessment. The data requirements
    for various pathways and various modelling approaches are presented.

         Chapter 7 separates the conceptual first half of the text from
    the pragmatic guidelines offered in the rest of the document. The
    chapter contains a discussion of air monitoring, water monitoring and
    food sampling. These particular fields are rather well developed

    individually, if not well integrated into multimedia studies. The
    reader is referred to many other resources that can guide the
    investigator to details on instruments, sampling methods and
    laboratory analysis.

         In Chapter 8, proportionally more emphasis is placed on soil and
    settled dust sampling. Again, the laboratory methods for metals,
    organics and various chemical compounds are readily available in the
    published literature. This chapter, then, focuses on relatively new
    sampling techniques to quantify in a standardized way the contaminant
    levels in soil and settled dust.

         In Chapter 9, on microbiological agents, assessment techniques
    for commonly encountered allergens, mycotoxins, fungal and pollen
    spores, microbiological bacteria and endotoxins are presented. These
    agents have been included because of their imputed contribution to
    respiratory disease and potential interactions with chemical
    pollutants. There is growing recognition that exposure to these agents
    in schools, homes, hospitals and office buildings constitutes a
    specific risk to atopic, asthmatic and compromised individuals.

         The use of biomarkers for exposure assessments is presented in
    Chapter 10. Biological samples derived from human tissue or fluids
    have been used as markers of both effects as well as exposure (dose)
    to a variety of occupational and environmental contaminants. The
    chapter describes the applications of biomarkers in exposure studies.

         The quality assurance (QA) activities that should be considered
    in conducting and evaluating exposure studies are addressed in Chapter
    11. Contributors to this document intended to impart their experiences
    to improve future exposure study. It is emphasized that QA aspects
    must be considered in all components of exposure studies, to enhance
    comparability and interpretation.

         Chapter 12 presents brief synopses of exposure studies.
    Selections illustrate a variety of study designs with different
    objectives and target pollutants and populations. Relatively more
    emphasis has been given to particles and passive exposure to cigarette
    smoke. The evidence is that cigarette consumption has increased almost
    worldwide, suggesting that greater attention be given to
    characterizing and reducing exposures to non-smokers, in particular,
    infants and young children. Epidemiological studies conducted over the
    last 15 years indicate that ambient particulate matter is adversely
    affecting human health at levels well below many of the established
    standards. Exposure assessment along with toxicology and epidemiology
    will be needed to answer many of the remaining unresolved issues about
    ambient and indoor suspended particles.

         Other studies summarized show how exposure assessment is
    supportive of epidemiology and risk management. The reader should
    recognize that Chapter 12 is not comprehensive but is intended to help
    educate the research community and others about the application, use
    and limitations of exposure assessment methodologies.

    1.  DEFINING EXPOSURE

    1.1  Introduction

         People are exposed to a variety of potentially harmful agents in
    the air they breathe, the liquids they drink, the food they eat, the
    surfaces they touch and the products they use. An important aspect of
    public health protection is the prevention or reduction of exposures
    to environmental agents that contribute, either directly or
    indirectly, to increased rates of premature death, disease, discomfort
    or disability. It is usually not possible, however, to measure the
    effectiveness of mitigation strategies directly in terms of prevented
    disease, reduced premature death, or avoided dysfunction. Instead,
    measurement or estimation of actual human exposure, coupled with
    appropriate assumptions about associated health effects or safety
    limits (e.g., acceptable daily intake, tolerable daily intake), is the
    standard method used for determining whether intervention is necessary
    to protect and promote public health, which forms of intervention will
    be most effective in meeting public health goals, and whether past
    intervention efforts have been successful (Ott & Roberts, 1998).

         The purpose of this chapter is to define the concept of exposure,
    and the direct and indirect method of exposure assessment. A brief
    discussion of exposure in the environmental health paradigm and its
    relationship to dose is presented.

    1.2  Defining exposure

         Exposure is defined as contact over time and space between a
    person and one or more biological, chemical or physical agents (US
    NRC, 1991a). Exposure assessment is to identify and define the
    exposures that occur, or are anticipated to occur, in human
    populations (IPCS, 1993). This can be a complex endeavour requiring
    analysis of many different aspects of the contact between people and
    hazardous substances (see Table 1). Although exposure is a
    well-established concept familiar to all environmental health
    scientists, its meaning often varies depending on the context of the
    discussion. It is important however, that exposure and related terms
    be defined precisely. In the following sections, we describe and
    define important exposure-related terms used in this document. The
    definitions are consistent with the US EPA's Exposure Assessment
    Guidelines and related WHO publications (WHO, 1987, 1996a; US EPA,
    1992a; IPCS, 1994). It is important to recognize, however, that
    terminology and definitions vary among organizations and nations.
    Thus, the reader is advised to concentrate on the concepts, rather
    than the specific terms, as they represent the crux of exposure
    assessment.

    Table 1.  Different aspects of the contact between people and pollution
              that are potentially important in exposure analysis
              (Sexton et al., 1995b)

                                                                          
    Agent(s)                    biological, chemical, physical, single
                                agent, multiple agents, mixtures

    Source(s)                   anthropogenic/non-anthropogenic, area/point,
                                stationary/mobile, indoor/outdoor

    Transport/carrier medium    air, water, soil, dust, food, product/item

    Exposure pathways(s)        eating contaminated food,
                                breathing contaminated workplace air
                                touching residential surface

    Exposure concentration      mg/kg (food), mg/litre (water), µg/m3 (air),
                                µg/cm2 contaminated surface), % by weight,
                                fibres/m3 (air)

    Exposure route(s)           inhalation, dermal contact, ingestion,
                                multiple routes

    Exposure duration           seconds, minutes, hours, days, weeks,
                                months, years, lifetime

    Exposure frequency          continuous, intermittent, cyclic, random,
                                rare

    Exposure setting(s)         occupational/non-occupational,
                                residential/non-residential, indoors/outdoors

    Exposed population          general population, population subgroups,
                                individuals

    Geographic scope            site/source specific, local, regional,
                                national, international, global

    Time frame                  past, present, future, trends
                                                                          


    1.2.1  Exposure and exposure concentration

         Exposure, as defined earlier, is the contact of a biological,
    chemical, or physical agent with the outer part of the human body,
    such as the skin, mouth or nostrils. Although there are many instances
    where contact occurs with an undiluted chemical (e.g., use of
    degreasing chemicals for cleaning hands), contact more often occurs
    with a carrier medium (air, water, food, dust or soil) that contains
    dilute amounts of the agent. "Exposure concentration" (e.g., mg/litre,
    mg/kg, µg/m3) is defined as the concentration of an environmental
    agent in the carrier medium at the point of contact with the body.

    1.2.2  Exposure estimation by integration and averaging

         A minimal description of exposure for a particular route must
    include exposure concentration and the duration of contact. If the
    exposure concentration is integrated over the duration of contact
    (Table 2), the area under the resulting curve is the magnitude of the
    exposure in units of concentration multiplied by time (e.g.,
    mg/litreÊday, mg/kgÊday, µg/m3Êh). This is the method of choice to
    describe and estimate short-term doses, where integration times are of
    the order of minutes, hours or days.

         Over periods of months, years or decades, exposures to most
    environmental agents occur intermittently rather than continuously.
    Yet long-term health effects, such as cancer, are customarily
    evaluated based on an average dose over the period of interest
    (typically years), rather than as a series of intermittent exposures.
    Consequently, long-term doses are usually estimated by summing doses
    across discrete exposure episodes and then calculating an average dose
    for the period of interest (e.g., year, lifetime). Although the
    integration approach can also be used to estimate long-term exposures
    or doses, its application to time periods longer than about a week is
    usually difficult and inconvenient.

    1.2.3  Exposure measurements and models

         Direct measurements are the only way to establish unequivocally
    whether and to what extent individuals are exposed to specific
    environmental agents. But it is neither affordable nor technically
    feasible to measure exposures for everyone in all populations of
    interest. Models, which are mathematical abstractions of physical
    reality, may obviate the need for such extensive monitoring programmes
    by providing estimates of population exposures (and doses) that are
    based on a smaller number of representative measurements (Fig. 1). The
    challenge is to develop appropriate and robust models that allow for
    extrapolation from relatively few measurements to estimates of
    exposures and doses for a much larger population (US NRC, 1991b).

         For relatively small groups, measurements or estimates can be
    made for some or all of the individuals separately, and then combined
    as necessary to estimate the exposure (or dose) distribution. For
    larger groups, exposure models and statistics can sometimes be used to
    derive an estimate of the distribution of population exposures,
    depending on the quantity and quality of existing data. Monte Carlo
    and other statistical techniques are increasingly being used to
    generate and analyse exposure distributions for large groups (US EPA,
    1992a).

    1.2.4  Exposure in the context of an environmental health paradigm

         The presence of hazardous substances in our environment does not
    necessarily imply a risk to human health or to the ecosystem. Exposure
    is an integral and necessary component in a sequence of events having
    potential health consequences. An expanded and more detailed version

    TABLE 2

    of the environmental health paradigm also showing the role of exposure
    is depicted in Fig. 2. The role of exposure assessment in the risk
    assessment framework applied by EU and US EPA is shown in Fig. 3.

         The release of an agent into the environment, its ensuing
    transport, transformation and fate in various environmental media, and
    its ultimate contact with people are critical events in understanding
    how and why exposures occur. Definitions for key events in the
    continuum are summarized below. They were compiled from three sources:
    Ott (1990); US EPA (1992a); Sexton et al. (1995a).

    *   Sources. The point or area of origin for an environmental agent
       is known as a source. Agents are released into the environment from
       a wide variety of sources, which are often categorized as
        primary sources including point sources (e.g., incinerator)
       versus area sources (e.g., urban runoff), stationary sources (e.g.,
       refinery) versus mobile sources (e.g., automobile) and
       anthropogenic sources (e.g., landfill) versus non-anthropogenic
       sources (e.g., natural vegetation) and  secondary sources 
       including condensation of vapours into particles and chemical
       reactions of precursors producing new pollutants.

    *   Exposure pathway. An exposure pathway is the physical course
       taken by an agent as it moves from a source to a point of contact
       with a person. The substance present in the media is quantified as
       its concentration.

    FIGURE 1

    FIGURE 2

    FIGURE 3

    *   Exposure concentration. As discussed in 1.2.1, exposure is the
       concentration of an agent in a carrier medium at the point of
       contact with the outer boundary of the human body. The
       concentration is the amount (mass) of a substance or contaminant
       that is present in a medium such as air, water, food or soil
       expressed per volume or mass. Assessments are often not at exposure
       or exposure concentration, since that information alone is not very
       useful unless it is converted to dose or risk. Assessments
       therefore usually estimate how much of an agent is expected to
       enter the body. This transfer of an environmental agent from the
       exterior to the interior of the body can occur by either or both of
       two basic processes: intake and uptake.

    *   Exposure route. Exposure route denotes the different ways the
       substance may enter the body. The route may be dermal, ingestion or
       inhalation.

    *   Intake. Intake is associated with ingestion and inhalation routes
       of exposure. The agent, which is likely to be part of a carrier
       medium (air, water, soil, dust, food), enters the body by bulk
       transport, usually through the nose or mouth. The amount of the
       agent that crosses the boundary per unit time can be referred to as
       the "intake rate", which is the product of the exposure
       concentration times the rate of either ingestion or inhalation. For
       inhalation, intake may be calculated for any time period. For
       ingestion, intake is usually expressed as the amount of food or
       water consumed times the pollutant concentration in that medium
       during a certain time period.

    *   Uptake. Uptake is associated with the dermal route of exposure,
       as well as with ingestion and inhalation after intake has occurred.
       The agent, as with intake, is likely to be part of a carrier medium
       (e.g., water, soil, consumer product), but enters the body by
       crossing an absorption barrier, such as the skin, respiratory tract
       or gastrointestinal tract. The rates of bulk transport across the
       absorption barriers are generally not the same for the agent and
       the carrier medium. The amount of the agent that crosses the
       barrier per unit time can be referred to as the  uptake rate. This
       rate is a function of the exposure concentration, as well as of the
       permeability and surface area of the exposed barrier. The uptake
       rate is also called a  flux. 

    *   Dose. Once the agent enters the body by either intake or uptake,
       it is described as a dose. Several different types of dose are
       relevant to exposure estimation. All these different dose measures
       are approximations of the target or biological effective dose.

       -   Potential (administered) dose. Potential or administered dose
          is the amount of the agent that is actually ingested, inhaled or
          applied to the skin. The concept of potential dose is
          straightforward for inhalation and ingestion, where it is
          analogous to the dose administered in a dose-response
          experiment. For the dermal route, however, it is important to

          keep in mind that potential (or administered) dose refers to the
          amount of the agent, whether in pure form or as part of a
          carrier medium, that is applied to the surface of the skin. In
          cases where the agent is in diluted form as part of a carrier
          medium, not all of the potential dose will actually be touching
          the skin.

       -   Applied dose. Applied dose is the amount of the agent directly
          in contact with the body's absorption barriers, such as the
          skin, respiratory tract and gastrointestinal tract, and
          therefore available for absorption. Information is rarely
          available on applied dose, so it is calculated from potential
          dose based on factors such as bioavailability (Fig. 2).

       -   Internal (absorbed) dose. The amount of the agent absorbed,
          and therefore available to undergo metabolism, transport,
          storage or elimination, is referred to as the  internal or
           absorbed dose (Fig. 2). Bioavailability has been used to
          describe absorbed dose.

       -   Delivered dose. The delivered dose is the portion of the
          internal (absorbed) dose that reaches a tissue of interest.

       -   Biologically effective (target) dose. The biologically
          effective dose is the portion of the delivered dose that reaches
          the site or sites of toxic action.

         The link, if any, between biologically effective (target) dose
    and subsequent disease or illness depends on the relationship between
    dose and response (e.g., shape of the dose-response curve), underlying
    pharmacodynamic mechanisms (e.g., compensation, damage, repair), and
    important susceptibility factors (e.g., health status, nutrition,
    stress, genetic predisposition).

    *   Biological effect. A measurable response to dose in a molecule,
       cell or tissue is termed a biological effect. The significance of a
       biological effect, whether it is an indicator or a precursor for
       subsequent adverse health effects, may not be known.

    *   Adverse effect. A biological effect that causes change in
       morphology, physiology, growth, development or life span which
       results in impairment of functional capacity to compensate for
       additional stress or increase in susceptibility to the harmful
       effects of other environmental influences (IPCS, 1994).

    1.3  Elements of exposure assessment

         Assessing human exposure to an environmental agent involves the
    qualitative description and the quantitative estimation of the agent's
    contact with (exposure) and entry into (dose) the body. Although no
    two exposure assessments are exactly the same, all have several common
    elements: the number of people exposed at specific concentrations for
    the time period of interest; the resulting dose; and the contribution

    of important sources, pathways and behavioural factors to exposure or
    dose. A list of the types of estimates that might comprise a
    comprehensive exposure assessment could include the following (as
    described in part by Brown (1987) and Sexton et al. (1995a)):

    *   Exposure 
       -  routes, pathways and frequencies
       -  duration of interest (short-term, long-term, intermittent or
          peak exposures)
       -  distribution (e.g., mean, variance, 90th percentile) --
          population, important subpopulations (e.g., more exposed, more
          susceptible)
       -  individuals -- average, upper tail of distribution, most exposed
          in population.

    *   Dose 
       -  link with exposures
       -  distribution (e.g., mean, variance, 90th percentile) --
          population important subpopulations (e.g., higher doses, more
          susceptible)
       -  individuals -- average, upper tail of distribution, highest dose
          in population.

    *   Causes 
       -  relative contribution of important sources
       -  relative contribution of important environmental media
       -  contribution of important exposure pathways
       -  relative contribution of important routes of exposure.

    *   Variability 
       -  within individuals (e.g., changes in exposure from day to day
          for the same person)
       -  between individuals (e.g., differences in exposure on the same
          day for two different people)
       -  between groups (e.g., different socio-economic classes or
          residential locations)
       -  over time (e.g., changes in exposure from one season to another)
       -  across space (e.g., changes in exposure/dose from one region of
          a city, country to another).

    *   Uncertainty 
       -  lack of data (e.g., statistical error in measurements, model
          parameters, etc.; misidentification of hazards and causal
          pathways)
       -  lack of understanding (e.g., mistakes in functional form of
          models, misuses of proxy data from analogous contexts).

         Although comprehensive exposure assessments could be considered
    the ideal, they are very costly; decisions therefore need to be made
    on the most important elements for inclusion. For any study, the
    purpose must first be defined. Possible purposes include environmental
    epidemiology, risk assessment, risk management or status and trend
    analysis (see Chapter 2). The data elements and measuring approaches

    that are needed for this purpose are then determined. Table 3
    summarizes the basic information that is required for each study. It
    should be mentioned that different elements of the exposure assessment
    framework might be selected to meet different study requirements.


    Table 3. Basic information needed for exposure assessments in 
             different contexts
                                                                        

                              Information required
                                                                        

    Risk assessment           Point estimates or distributions of 
                              exposure and dose
                              Duration of exposure and dose

    Risk management           Pollutant source contributing to 
    (conducted once hazard    exposure and dose
    is identified)            Personal activities contributing 
                              to exposure and dose
                              Effectiveness of intervention measures

    Status and trend          Change of exposure and dose of 
                              populations over time

    Epidemiology              Individual and population exposures and 
                              doses, exposure dose categories
                                                                        


    1.4  Approaches to quantitative exposure assessment

         Quantitative estimation of exposure is often the central feature
    of assessment activities. The quantitative estimation of exposure can
    be approached in two general ways:  direct assessment, including
    point-of-contact measurements and biological indicators of exposure;
    and  indirect assessment, including environmental monitoring,
    modelling, questionnaires (US NRC, 1991b) (see Chapter 3.5). These two
    generic approaches to quantitative estimation of exposure are
    independent and complementary. Each relies on different kinds of data
    and has different strengths and weaknesses. It is potentially useful,
    therefore, to employ multiple approaches as a way of checking the
    robustness of results. Among other factors, the choice of which method
    to use will depend on the purpose of the assessment and the
    availability of suitable methods, measurements and models.

         Direct approaches for air, water and food include personal air
    monitors, measurements of water at the point of use and measurement of
    the food being consumed. Indirect approaches include
    microenvironmental air monitoring and measurements of the water supply
    and food supply (contents of a typical food basket, for instance).

         Exposure models are constructed to assess or predict personal
    exposures or population exposure distributions from indirect
    measurements and other relevant information.

         Measures of contaminants in biological material (biomarkers)
    afford a direct measure of exposure modified by and integrated over
    some time in the past which depends on physiological factors that
    control metabolism and excretion. Such measures give no direct
    information about the exposure pathways. Examples of the type of
    biomarkers measured in human material that can be used for
    reconstructing internal dose and their relevance to exposure
    assessment are discussed in Chapter 10.

    1.5  Linking exposure events and dose events

         The schematic framework in Fig. 2 shows how the
    interrelationships among significant exposure- and dose-related events
    in the paradigm can be conceived.

         It is important to keep in mind that, although events along the
    continuum are correlated, the relative position of a particular
    individual within a distribution may change dramatically from one
    event to the next as the agent or its metabolite/derivative moves
    through the various stages from exposure concentration to biologically
    effective dose.

         To make realistic estimates for a specific event (e.g., an
    internal dose), it is necessary to have at least one of two types of
    information: measurements of the event itself (e.g., internal dose),
    or measurements of an earlier (e.g., potential dose) or later (e.g.,
    delivered dose) event in the continuum. It is also necessary to
    understand the critical intervening mechanisms and processes (e.g.,
    pharmacokinetics) that govern the relationship between the event
    measured and the event of interest (e.g., internal dose). Unless such
    data are on hand, extrapolating from one event to another, moving
    either from exposure to dose (downwards in Fig. 2) or from dose to
    exposure (upwards in Fig. 2) is problematic.

         Suitable data and adequate understanding are seldom, if ever,
    available to describe and estimate all of the significant events for
    the groups and individuals of interest. Generally speaking,
    measurement of exposure concentration and delivered dose  (body 
     burden) is in many cases relatively straightforward, whereas
    measurement of potential (administered) dose and internal (absorbed)
    dose is usually possible only with substantially greater effort.
    Measurement of biologically effective (target) dose may also be
    possible in some cases, although it is usually impossible to measure
    the applied dose.

         This situation presents us with a conundrum. We would like to
    have realistic estimates of exposure concentrations of an agent for
    all important pathways, and the resulting biologically effective dose.
    Typically, however, if relevant data are available at all, they are

    related to exposure concentrations for one pathway or route of
    exposure. In the few cases where data on dose are also available,
    these data usually reflect delivered dose (body burden) rather than
    biologically effective dose. Even if suitable measurements of both
    exposure concentration and delivered or target dose are on hand, the
    absence of pharmacokinetic understanding to relate these measurements
    to each other, as well as to other significant events along the
    continuum, seriously impairs efforts to establish the link between
    exposure and dose.

         We are thus left with a situation in which we can measure
    specific events on either side of the body's absorption boundaries,
    but we can relate them to each other only by using a series of
    unsubstantiated assumptions. Yet it is this relationship between
    exposure and dose that is critical to, for example, establishing cause
    and effect relationships between exposure and diseases.

    1.6  Summary

         Exposure requires the occurrence of the presence of an
    environmental toxicant at a particular point in space and time; and
    the presence of a person or persons at the same location and time. In
    addition, the amount which comes in contact with the outer boundary of
    the human body is required.

         As the intrinsic value of exposure-related information has become
    recognized, "exposure analysis" has emerged as an important field of
    scientific investigation, complementing such traditional public health
    disciplines as epidemiology and toxicology, and is an essential
    component in informed environmental health decision-making (Goldman et
    al., 1992; Sexton et al., 1992, 1994; Wagener et al., 1995).

    2.  USES OF HUMAN EXPOSURE INFORMATION

    2.1  Introduction

         Exposure assessments collect data on the route magnitude,
    duration, frequency and distributions of exposures to hazardous agents
    for individuals and populations. Human exposure data have been used
    for the evaluation and protection of environmental health in four
    interrelated disciplines: epidemiology, risk assessment, risk
    management, and status and trends analysis. The fundamental goal of
    exposure assessment studies is to reduce the uncertainty of the
    exposure estimates that are used within each discipline to make public
    policy decisions or reach research conclusions.

          Epidemiology is the examination of the link between human
    exposures and health outcomes (Sexton et al., 1992).  Risk 
     assessment is the estimation of the likelihood, magnitude and
    uncertainty of population health risks associated with exposures. In
    contrast,  risk management is the determination of the source and
    level of health risks and which health risks are acceptable and what
    to do about them. Status and trends analysis comprises the evaluation
    of historical patterns, current status and possible future changes in
    human exposures.

         The purpose of this chapter is to describe the disciplines from
    environmental epidemiology through risk assessment. It also describes
    how human exposure assessment data are used in each of these
    disciplines

    2.2  Human exposure information in environmental epidemiology

         Epidemiology is the study of the determinants and distribution of
    health status (or health-related events) in human populations.
    Environmental epidemiology searches for statistical associations
    between environmental exposures and adverse health effects (presumed)
    to be caused by such exposures. It is a scientific tool that can
    sometimes detect environmentally induced health effects in
    populations, and it may offer opportunities to link actual exposures
    with adverse health outcomes (US NRC, 1991c, 1994; Matanoski et al.,
    1992; Beaglehole et al., 1993).

         Exposure assessment methods can be used for identifying and
    defining the low or high exposure groups. They can also be used for
    devising more accurate exposure data from measured environmental
    contaminant levels and personal questionnaire or time-activity diary
    data, or estimating population exposure differences between days of
    high and low pollution, or between high and low pollution in
    communities using measured environmental and population behavioural
    data (see also Chapters 3 and 5).

         In particular, to establish long-term health effects of "low
    dose" environmental exposures, epidemiological methods are the
    predominant, if not only, tools at hand for health-effect assessment.
    However, the excess risk of most environmentally related health
    effects is small, with relative risks and odds ratios usually being
    less than 2 across the observed range of exposure experienced by
    populations. Furthermore, there are usually no "non-exposed"
    comparison groups, and the factors contributing to the development of
    diseases are numerous. As a consequence, environmental epidemiology
    faces considerable methodological challenges. Adequate exposure
    assessment is one key issue, as well as the need for studies conducted
    with large populations.

    2.3  Human exposure information in risk assessment

         Risk assessment is a formalized process for estimating the
    magnitude, likelihood and uncertainty of environmentally induced
    health effects in populations. Exposure assessment (e.g., exposure
    concentrations and related dose for specific pathways) and effects
    assessment (i.e., hazard identification, dose-response evaluation) are
    integral parts of the risk assessment process. The goal is to use the
    best available information and knowledge to estimate health risks for
    the subject population, important subgroups within the population
    (e.g., children, pregnant women and the elderly), and individuals in
    the middle and at the "high end" of the exposure distribution (US NRC,
    1983; Graham et al., 1992; Sexton et al., 1992).

         Environmental health policy decisions should be based on
    established links among emission sources, human exposures and adverse
    health effects. The chain of events depicted in Fig. 4 is an
    "environmental health paradigm": a simplified representation of the
    key steps between emission of toxic agents into the environment and
    the final outcome as potential disease or dysfunction in humans. This
    sequential series of events serves as a useful framework for
    understanding and evaluating environmental health risks (Sexton, 1992;
    Sexton et al., 1992, 1993). It is directly related to the risk
    assessment process.

    *   Exposure assessment in the risk assessment framework focuses on
       the initial portion of the environmental health paradigm: from
       sources, to environmental concentrations, to exposure, to dose. The
       major goal of exposure assessment is to develop a qualitative and
       quantitative description of the environmental agent's contact with
       (exposure) and entry into (dose) the human body. Emphasis is placed
       on estimating the magnitude, duration and frequency of exposures,
       as well as estimating the number of people exposed to various
       concentrations of the agent in question (US NRC, 1983, 1991a;
       Callahan & Bryan, 1994).

    FIGURE 4

    *   Effects assessment examines the latter portion of the events
       continuum: from dose to adverse health effects (Fig. 4). The goals
       are to determine the intrinsic hazards associated with the agent
       (hazard identification) and to quantify the relationship between
       dose to the target tissue and related harmful outcomes
       (dose-response/effect assessment). The overlap between exposure
       assessment and effects assessment reflects the importance of the
       exposure-dose relationship to both activities (Sexton et al.,
       1992).

    *   Risk characterization is the last phase of the risk assessment
       process. The results of the actual exposure assessment and the
       effects assessments are combined to estimate the human health risks
       from the exposures.

         Systemic (non-cancer) toxicants are usually assumed to have
    thresholds below which no effects occur. For these toxicants, safety
    assessments are performed with establishment of  tolerable intakes 
    (IPCS, 1993) or  reference concentrations/doses (USEPA). From these,
    guidelines are derived and standards designed to protect public
    health. Ambient concentration standards, and workplace personal
    exposure limits, are often established at or below threshold levels
    determined as part of the risk assessment process. Although these
    standards are set with safety margins, exposures that exceed these
    reference levels raise concerns about potentially elevated health
    risks for the exposed population (Fig. 5a).

         Quantitative risk assessment for carcinogens is a well
    established, albeit controversial, procedure. As part of the
    guidelines developed by the WHO, it is common practice to extrapolate
    from high to low dose by assuming a linear, non-threshold model for
    carcinogenicity. Under this assumption, cancer risk for individuals
    can be estimated directly from the exposure or dose distribution, and
    the number of excess cancer cases (i.e., the increase above background
    rates) in the exposed population can usually be estimated by
    multiplying the average dose by both the total number of people
    exposed and the dose-response slope factor (Fig. 5b). Although
    individual risk is assumed to increase with increasing exposure and
    dose all along the distribution, exposures of concern are typically
    defined to be those above some minimal level of risk (e.g., WHO
    considers this to be a 1 in 105 or 106 excess lifetime risk of
    developing cancer). Unit cancer risk numbers are given in inverse
    concentration units for food, water and air as (ppm)-1, (ppb)-1 or
    mg-1m-3). Expressed in inverse dose units (mg kg-1day-1), the cancer
    slope risk factor is multiplied by ingestion or inhalation rates and
    adjusted for body weight. Individual cancer risk is calculated by
    assuming a lifetime of exposure at a given level of contamination.
    When exposure data are available, it is then possible to approximate
    the cancer risk of the typical or average person in the population or
    one who might be at maximum risk due to a greater level of exposure.

    FIGURE 5

         In regulatory applications of risk assessments, exposure
    estimates are often constructed using existing data or single point
    measurements to estimate the risk of a facility, hazardous waste site
    or chemical waste site, or even the use of a chemical product. This
    approach can result in large errors in the exposure assessment and
    hence the risk assessment. Exposure assessment studies are used to
    obtain a more accurate determination of the exposure associated with a
    health impact outcome of concern. Population-based risk assessments
    benefit from the use of population-based measurements derived from
    surveys or models (see Chapter 3) to estimate the distribution of
    health effect outcomes in the total exposed population over a
    specified time period.

    2.3.1  Risk allocation for population subgroups or activities

         Exposure studies may also be conducted to provide more realistic
    and location-specific information for use in human health risk
    assessments. Measurement data on pollutant concentrations and exposure
    factors, such as contact rates, can be used instead of relying on
    assumed "default" values for an "averaged" or representative
    individual. An example of an exposure study designed to collect data
    for the purpose of allocating risk to locations, sources and
    activities is the Windsor Air Quality Study conducted in Windsor,
    Ontario, Canada (Bell et al., 1994).

         The Windsor Air Quality Study was designed to investigate the
    Windsor airshed characteristics with respect to airborne toxic
    compounds and to determine personal inhalation exposures to these
    compounds. Data were then used as inputs for a multimedia assessment
    of risk due to total pollutant exposure. The air quality study
    examined just one aspect, the inhalation route. It was designed to
    separately attribute risk to several airborne contaminants by indoor
    and outdoor locations. Statistical analysis and inference were used to
    impute source contributions to population risk (i.e., the waste
    incinerator across the river in Detroit, Michigan, USA) for selected
    volatile organic compounds (VOCs), carbonyls and trace metals (see
    Table 4) based on microenvironmental and personal measurements and
    time activity patterns. In general, air quality was determined to be
    relatively poor in recreation halls, new office buildings, cars and
    garages when compared to outdoor air quality standards and criteria.
    Although high contaminant concentrations were detected in various
    microenvironments, population exposures (defined as the product of
    concentration and time) were relatively low because the study subjects
    did not spend any appreciable time in those microenvironments. This
    point is illustrated in Fig. 6. For all of the VOCs, the highest
    concentrations were measured during the commuting periods, with
    comparable concentrations being measured indoors at the office and
    home and the lowest outdoors (Table 3). When time in each
    microenvironment is considered, exposure in the home accounted for
    over 70% of the total exposure profile for that individual.

        Table 4.  Target analytes in the Windsor air quality study

                                                                                              
    Volatile organic compounds

    Propane, chloromethane, 2-methylpropane, chloroethene, 1,3-butadiene, butane, 
    2-methylbutane, pentane, isoprene, 1,1-dichloroethene, dichloromethane, allyl chloride, 
    hexane trichloromethane, 1,2-dichloroethane, 1,1,1-trichloroethane, benzene, 
    tetrachloromethane, xylenes, styrene, o-xylene, 1,1,2,2-tetrachloroethane, nonane, 
    1,3,5-trimethylbenzene, 1,2,4-trimethylbenzene, 1,4-dichlorobenzene; decane, 
    1,2-dichlorobenzene, undecane, 1,2,4-trichlorobenzene, dodecane, tridecane

    Carbonyls

    Formaldehyde, acetaldehyde, acrolein, acetone, propianaldehyde, crotonaldehyde, methyl 
    ethyl ketone, benzaldehyde, isovaleraldehyde, 2-pentanone, valeraldehyde,  o-tolualdehyde, 
     m-tolualdehyde,  p-tolualdehyde, methyl isobutyl ketone, hexanal, 2,5-dimethylbenzaldehyde

    Trace metals

    Beryllium, chromium, manganese, nickel, arsenic, selenium, cadmium, lead
                                                                                              
    

         Results of the study emphasize the importance of exposure
    assessments for policy decisions. For this community, changes in
    lifestyle, consumer product formulations, cleaning of indoor air and
    increased ventilation would probably have more impact on reducing
    health risks from exposures to VOCs than reliance on
    government-mandated abatement strategies for ambient sources.

    2.3.2  Population at higher or highest risk

         Risk assessment may be used to identify and evaluate those
    populations, subpopulations and individuals at potentially greater
    risk so that, if warranted, appropriate mitigation actions can be
    implemented. Individuals and groups are deemed to be at potentially
    higher risk because they are exposed to high concentrations of
    hazardous pollutants (Sexton et al., 1993). Individuals and groups can
    also be at increased risk because they are more susceptible to the
    adverse effects of a given exposure. Among the potential causes of
    enhanced susceptibility are inherent genetic variability, age, gender,
    pre-existing disease (e.g., diabetes, asthma), inadequate diet,
    environmental or lifestyle factors (e.g., smoking), stress and
    inadequate access to health care. As far as possible, it is important
    to identify these susceptible individuals and groups so that we can
    understand their exposures and take account of this information in
    assessing and managing risks. Exposure and risk information for
    susceptible populations is critical since health standards and
    regulations are often developed with the intent of protecting these
    individuals.

         Exposure studies provide valuable information for the risk
    assessment by quantifying the distribution of exposures in a
    population and identifying those subpopulations or individuals who
    have the highest exposures. Information is also gathered on
    characteristics of the populations and factors that could contribute
    to elevated exposures. In these studies, measures of central tendency,
    such as the median and average, along with expressions of variability,
    such as the standard deviation, are commonly used to describe the
    distribution of exposures for a population (Fig. 7). Often, the
    relative position of an individual or group in the exposure
    distribution is of primary interest to the exposure assessor. Among
    the most frequently used descriptors for individual and subgroup
    exposures are values near the middle of the distribution, values above
    the 90th percentile and values at the extreme upper end, such as for
    the most exposed person in the population. Exposure studies that are
    targeted on susceptible populations are used with the same type of
    inputs in risk assessment for these groups.

    2.4  Human exposure information in risk management

         Risk management decisions carried out by policy-makers are of
    four basic types: priority setting, selection of the most
    cost-effective method to prevent or reduce unacceptable risks, setting
    and evaluating compliance with standards or guidelines, and the
    evaluation of the success of risk mitigation efforts. Exposure
    information is crucial to these decisions. In addition to data on
    exposures and related health effects, decision-makers also must
    account for the economic, engineering, legal, social and political
    aspects of the problem (Burke et al., 1992; Sexton et al., 1992).

         Conceptually, as shown in Fig. 8, estimating and prioritizing
    health risks are seemingly straightforward. Risk is a combination of
    effects estimates, where "highest" priorities can be thought of as
    those that entail both "high" toxicity for the agent of interest
    (adverse effects are likely to occur in humans at relatively low
    exposures or doses), and "high" exposures for the population,
    subpopulation or individuals of interest (exposures or doses are above
    a health-based standard). Conversely, "lowest" priority risks involve
    "low" toxicity and "low" exposures. "Medium" priority risks are those
    for which either toxicity or exposure is "low" while the other is
    "high" (Sexton, 1993). The Windsor Air Quality Study, for example,
    showed that incinerator emissions contributed little to total human
    exposure for VOCs. Despite the fact that the pollutants were of high
    toxicity, incinerator emissions were considered to be of relatively
    low risk to the population. In contrast, studies show that second-hand
    smoke has both high toxicity and high human exposures, and should
    therefore be identified as a high priority risk.

    FIGURE 6

    FIGURE 7


    FIGURE 8

         Risk mitigation proceeds from first determining that an exposure
    is a hazard (risk assessment) to identifying and quantifying the route
    and the environmental pathways for a contaminant. Where a contaminant
    has multiple sources or routes of exposure, relative contributions to
    individual and population risk must be determined. Exposure
    assessments are crucial for developing this information, and may rely
    on both measurements and modelling. Once this information is obtained,
    then effort can be directed toward the most effective mitigation
    strategies.

         In fact, intervention studies are implicitly or explicitly
    predicated on the sequence of risk assessment and mitigation.
    Intervention at the source, transmission or receptor (receiving
    person) is intended to reduce the effect or risk of an effect.
    Prohibiting smoking in public buildings or sections of restaurants is
    designed to separate sources from receptors. Specific ventilation
    requirements for operating theatres or isolation rooms of infectious
    patients are designed to dilute potential contaminants and pathogens.
    On a larger scale, substitution of cleaner fuels (e.g., reformulated
    or unleaded gasoline, cleaner coal, low-sulfur oil, natural gas)
    radiation of food or ozonation of drinking-water are examples of risk
    mitigation interventions based on the assumption that contaminant
    reductions experienced in the environmental media will result in a
    corresponding reduction in actual exposures and hence risk. It is
    essential, then, to understand the efficacy of mitigation strategies
    with respect to their effect on human exposures.

         The combined use of total exposure assessment for air,
    receptor-source modelling and economic principles can assist
    environmental policy and regulation in developing risk mitigation
    strategies. The hybridization of these well-developed models can be
    used to assist in the identification of priority sources to target
    regulatory programmes, and in the development of cost-effective
    strategies for air pollution control to bring about the greatest and
    earliest reduction in pollutant exposures.

         Epidemiological information about the health effects of
    relatively low levels of air pollutants now raises controversial
    policy issues for risk management. On the one hand, the economic
    consequences of these health effects may be substantial; on the other
    hand, for some pollutants, control measures may become very expensive.
    For pollutants such as VOCs, for example, exposure monitoring rather
    than ambient air monitoring may lead to more rapid and cost-effective
    risk reduction policies.

         Developed countries have experimented with regulatory reforms
    that include emission trading. Basically, the concept calls for
    emission reduction at one source to be credited to the emission levels
    at another source. These trading schemes are based on the assumption
    that equal mass emission reduction of a pollutant would result in
    equal health or ecological benefits. Thinking in terms of total 
    exposure assessment reorients the relative importance of sources and

    their impacts on different populations. Accordingly, control options
    for reducing exposures can be broadened (Smith, 1995).

    2.5  Human exposure information in status and trend analysis

         Evaluating the current status of exposures and doses in the
    context of historical trends is an important tool for both risk
    assessment and risk management. In many cases it requires collecting
    exposure data over a relatively long period of time (e.g., 5-20
    years). This can only be done through an exposure assessment study and
    often when the contaminant has a long residence time in the
    environment or biological tissue. If concentrations of a contaminant
    exhibit high variability in environmental media, the study may require
    relatively large sample sizes, the use of probability samples and/or
    extensive follow-up to observe trends. Data on status and trends can
    be invaluable for identifying new or emerging problems, recognizing
    the relative importance of emission sources and exposure pathways,
    assessing the effectiveness of pollution controls, distinguishing
    opportunities for epidemiological research and predicting future
    changes in exposures and effects (Goldman et al., 1992; Sexton et al.,
    1992).

         Exposure studies may be conducted to document the status and
    trends of human exposure (e.g., Kemper, 1993; Noren, 1993). A good
    example of a study designed for this purpose is the German
    Environmental Survey (GerES). The nationwide representative survey was
    conducted for the first time in 1985-1986, on behalf of the Federal
    Ministry for the Environment, Nature Conservation and Reactor Safety.
    In 1990-1991 the survey was repeated in West Germany (the FRG before
    reunification) and in 1991-1992 it was extended to East Germany
    (former GDR) (Krause et al., 1992; Schulz et al., 1995).

         The purpose of the survey was to establish a representative
    database on the body burden of the general population. Biological
    monitoring was used to characterize exposure to pollutants
    (predominantly heavy metals). In addition, the occurrence of a number
    of pollutants in the domestic area likely to contribute to total
    exposure (house dust and drinking-water) was studied. The design of
    the study is summarized as follows:

    *   Population samples. Cross-sectional samples using a stratified
       two-step random sampling procedure according to the size of the
       community, gender and age. The final set included 2731 West Germans
       in 1985-1986 and 4287 adults from East and West Germany in
       1990-1992 (aged 25-79 years). In addition about 700 children (aged
       6-14 years) living in the same households were included in
       1990-1992.

    *   Human biomonitoring. Analysis of blood (lead, cadmium, copper,
       mercury), spot urine (arsenic, cadmium, copper, chromium, mercury)
       and scalp hair (aluminium, barium, cadmium, chromium, copper,
       magnesium, phosphorus, lead, strontium and zinc).

    *   Questionnaires. Questions about social factors, smoking habits,
       potential sources of exposure in the domestic, working, and general
       environment, and nutritional behaviour.

    *   Domestic environment. Concentration of trace elements in dust
       deposit indoors, in vacuum cleaner bags (pentachlorophenol [PCP],
       lindane and pyrethroids) and in household tap water; determination
       of VOCs in homes of a subsample of 479 participants (passive
       sampling) in 1985-1986.

    *   Personal sampling. Determination of VOCs by personal sampling
       using a subsample of 113 people in 1991.

    *   Dietary intake. A 24-h duplicate study in 1990-1992 with a
       subsample of 318 people.

         Characteristics of the frequency distributions (percentiles) and
    other statistical parameters of the concentration of elements and
    pollutants in the different media were calculated. As an example, the
    concentrations of elements and compounds in blood and urine of the
    German adult population analysed in 1990-1992 are shown in Table 5.
    The 1990-1991 and 1991-1992 surveys showed differences between East
    and West Germany. The mercury concentrations in blood and urine as
    well as the cadmium, chromium and copper concentrations in urine were
    significantly higher ( p < 0.001) in East Germany than in West
    Germany. The blood lead level was identical in both study populations
    (geometric mean 45 µg/litre).

         The comparison of the results for the biological, personal and
    microenvironmental exposure measurements taken in East Germany in
    1985-1986 and in 1990-1992 permits an analysis of trends over time.
    The success of abatement measures could be shown in a number of cases:
    the reduction of lead concentrations in petrol and of industrial
    cadmium emissions resulted in decreased lead and cadmium
    concentrations in the blood of the general population. The ban on PCP
    led to a decrease of PCP in house dust. The results of the GerES have
    provided a useful set of reference data to characterize and to assess
    exposures of the general population. They have also been useful for a
    number of risk assessments, for example the role of copper in
    drinking-water and liver cirrhosis in early childhood, and presence of
    mercury in amalgam fillings.

    2.6  Summary

         The specifics of any particular exposure analysis hinge on its
    intended use or uses. For example, the pertinent aspects of exposure
    to be considered, the nature of the information required and the
    necessary quantity and quality of the data will depend on whether the
    exposure assessment is being conducted in the context of an
    epidemiological investigation (Matanoski et al., 1992), risk
    assessment (Graham et al., 1992), risk management (Burke et al., 1992)
    or status and trend analysis (Goldman et al., 1992) (see also Chapter
    1, Table 1).


        Table 5.  Elements and compounds in blood and urine of the German population (aged 25-69 years, 1990-1992)
    (Krause et al., 1992)

                                                                                                                                  
                                 QL      N       <QL    10      50      90      95      98      MAX     AM      GM      CI GM
                                                                                                                                  

    Blood
    Lead (µg/litre)              15      3966    61     24.0    45.3    86.8    105.6   134.2   708.0   52.4    45.3    44.5-46.0
    Cadmium (µg/litre)           0.1     3965    231    0.1     0.3     1.9     2.6     3.6     11.3    0.7     0.4     0.4-0.4
    Copper (mg/litre)            0.1     3968    0      0.8     0.9     1.2     1.3     1.5     2.5     1.0     0.9     0.9-1.0
    Mercury (µg/litre)           0.2     3958    632    <0.2    0.6     1.6     2.1     3.0     12.2    0.8     0.5     0.5-0.5

    Urine
    Arsenic (µg/litre)           0.6     4001    210    1.8     7.1     19.8    29.9    56.7    205.5   10.5    6.3     6.1-6.5
    Arsenic (µg/g creatinine)            4001           1.4     4.9     15.3    24.1    40.0    147.6   7.6     4.6     4.5-4.8
    Cadmium (µg/litre)           0.1     4002    150    0.1     0.3     0.9     1.3     1.7     6.9     0.4     0.3     0.3-0.3
    Cadmium (µg/g creatinine)            4002           0.1     0.2     0.7     0.9     1.3     6.1     0.3     0.2     0.2-0.2
    Chromium (µg/litre)          0.2     4002    1716   0.15    0.2     0.4     0.6     1.0     21.2    0.3     0.2     0.2-0.2
    Chromium (µg/g creatinine)           4002           0.0     0.1     0.3     0.5     0.9     10.6    0.2     0.1     0.1-0.1
    Copper (µg/litre)            1.1     4002    20     4.5     9.7     18.7    22.9    28.7    444.2   11.6    9.5     9.3-9.7
    Copper (µg/g creatinine)             4002           3.5     6.7     13.1    17.7    28.5    420.7   8.9     6.9     6.8-7.1
    Mercury (µg/litre)           0.2     4002    785    <0.2    0.5     2.6     3.9     6.0     53.9    1.1     0.5     0.5-0.6
    Mercury (µg/g creatinine)            4002           0.1     0.4     1.6     2.2     3.2     73.5    0.7     0.4     0.4-0.4
    Nicotine (µg/litre)          5       3750    1566   <5      9.3     1438    2431    3567    10 984  422     24.9    23.0-27.1
    Nicotine (µg/g creatinine)           3748           1.3     7.0     1003    1636    2431    10 478  292     18.4    17.0-20.0
    Cotinine (µg/litre)          5       3800    1813   <5      5.6     2037    2681    3483    6573    537     26.6    24.3-29.1
    Cotinine (µg/g creatinine)           3798           1.3     4.9     1396    1940    2788    8111    388     19.6    17.9-21.4
    Creatinine (mg/100 ml)       0       4002           0.7     1.5     2.5     2.9     3.2     5.7     1.5     1.4     1.3-1.4
                                                                                                                                  

    Annotations: QL = quantification limit, N = sample size, n < QL = number of values below QL, 10, 50, 90, 95, 98 = percentiles, 
    MAX = maximum value, AM = arithmetic mean, GM = geometric mean.

    Source: UBA, WaBoLu, Environmental Survey 1990-1992, Federal Republic of Germany.
    


         Knowledge of human exposures to environmental contaminants is an
    important component of environmental epidemiology, risk assessment,
    risk management and status and trends analysis. Exposure information
    provides the critical link between sources of contaminants, their
    presence in the environment and potential human health effects. This
    information, if used in the context of environmental management
    predicated on human risk reduction, will facilitate selection and
    analysis of strategies other than the traditional "command and
    control" approach. Most of the environmental management structures
    around the world rely directly on the measured contaminants in various
    media to judge quality, infer risk and interpret compliance. Even in
    these cases, exposure information can evaluate the effectiveness of
    protecting segments of population more susceptible or at higher risk.

         It is this direct connection that makes exposure measures
    invaluable for evaluation of environmental health impacts on a local,
    regional and global scale.

    3.  STRATEGIES AND DESIGN FOR EXPOSURE STUDIES

    3.1  Introduction

         Accurate estimates of human exposure to environmental
    contaminants are necessary for a realistic appraisal of the risks
    these pollutants pose and for the design and implementation of
    strategies to control and limit those risks. Three aspects of exposure
    are important for determining related health consequences:

    *   Magnitude: What is the pollutant concentration?

    *   Duration: How long does the exposure last?

    *   Frequency: How often do exposures occur?

    The design of an exposure study specifies the procedures that will be
    used to answer these three questions.

         In this chapter, strategies and designs for exposure studies are
    discussed with emphasis on their relative advantages and
    disadvantages. The brief discussion of study design presented in
    Chapter 1 is expanded upon here in terms of fundamental types of
    generic study designs and approaches to assessing human exposure to
    chemicals in the environment. Statistical considerations for study
    design are presented in Chapter 4. The reader is referred to
    subsequent chapters for details on implementing exposure study designs
    through modelling (Chapter 6), monitoring of environmental media
    (Chapters 7, 8 and 9) and monitoring of biological tissue (Chapter
    10).

    3.2  Study design

         A good study design is the most important element of any exposure
    study. A flow chart that includes critical elements is shown in Fig.
    9. First the purpose of the study is defined: epidemiology, risk
    assessment, risk management or analyses of status and trends (see also
    Chapter 2). Within this context, specific study objectives are
    formulated. Often studies have several objectives, which must be
    prioritized to ensure that the primary objective is fulfilled. Study
    parameters must be selected that are consistent with the objective. A
    study design is formulated which links objectives to measurement
    parameters in a cost-effective manner. Two critical and often
    overlooked elements of the study design are development of a
    statistical analysis plan and quality assurance (QA) objectives. For
    general population studies, methods for measurement and analysis of
    contaminants in collected environmental or biological samples must be
    sufficiently sensitive to determine their concentration at typical
    ambient levels. For multimedia studies, method detection limits must
    be consistent across media. The study design is not complete until a
    pilot study has been conducted to evaluate sample and field study
    procedures.

    FIGURE 9

    3.3  Sampling and generalization

         Decisions on population sampling strategies involve consideration
    both of the populations that are available and of the types of
    measurements needed. Of prime consideration are the people, place and
    time (i.e., individuals, locations, sampling period or conditions)
    from which exposure samples are to be collected. Also, it is important
    to determine if the estimates to be derived from the proposed sample
    could be generalized to a wider population of interest. For example,
    consider an exposure assessment study from a sample population of a
    small town in southwestern Australia. The many potential populations
    of interest which this sample might generalize include: all people
    living in that town; people living in a small southwestern Australia
    town; people living in southwestern Australia; people living in
    Australia; people living in any small town; people in general. In this
    case, the sample population is not likely to provide a representative
    sample of the latter two populations.

         The appropriateness of the generalization is determined by
    considering if the sample is randomly selected in such a way as to be
    representative of the larger population of interest (Whitmore, 1988).
    This randomization is in terms of the distribution of the collected
    data. For continuous outcomes, the percentages of key attributes, such
    as demographic factors, should be similar between the sample and the
    population. However, when this is not possible, owing to limited
    funding for example, a descriptive study (described below) can provide
    credible data, although the extent to which these can be generalized
    is limited.

    3.4  Types of study design

         Once the population is defined, then the attention shifts to
    sampling strategies; in particular, comprehensive samples, probability
    samples, and other types of samples. A  comprehensive sample includes
    all members of the selected population. In a  probability sample each
    member has a known likelihood of being selected.  Simple random 
     sampling is a special case where each member of the population has
    an equal probability of being selected. Other types of study groups
    are selected on the basis of other characteristics, such as
    availability or convenience.

    3.4.1  Comprehensive samples

         Complete populations can be used to collect a full picture of the
    process being studied, especially when the total population is
    relatively small such as families in a neighbourhood. In these cases,
    an exhaustive collection of measurements is taken from every potential
    subject, and the completed data describe the situation exactly. There
    is no sample variability except through the methods and procedures
    used for measurement and monitoring. The main reasons for studies of
    this nature are either a small population size, a need for a complete
    evaluation of the problem, high potential risk, high variability among
    units or legal requirements. The advantages of this type of study are

    that a complete description of the exposure is given, and there is no
    need for generalization because all potential subjects are covered.
    The disadvantage of this approach, if the population is large, lies in
    the expense: all individuals in all locations must be monitored at all
    times.

    3.4.2  Probability samples

         Surveys consist of a random sampling of subjects from the
    population of interest. This approach aims to remove selection bias
    and is useful for generalizing results beyond the study sample. It is
    important to distinguish that "random" does not translate to
    "haphazard". A truly random sample is independent of human judgement.
    Every unit in the total population has a known above-zero likelihood
    of being included in the sample. Effective study design allows
    researchers to draw statistically valid inferences about the general
    population that the sample is designed to represent (Kish, 1965). For
    these studies, one needs to (Sexton & Ryan, 1988):

    *  choose a population for investigation

    *  choose an appropriate unit for sampling and analysis (e.g., person,
       household, neighbourhood, city, etc.)

    *  stratify as appropriate

    *  choose a sampling strategy (e.g., simple random sampling,
       multistage sampling).

         The results of a probability survey can be used to make general
    statements about the population under investigation. The advantages
    include having results that represent the population, taking into
    account the possible error due to sampling. The disadvantages of this
    scheme lie in the complicated sample selection, difficulty in
    maintaining compliance from participants and the potentially complex
    statistical analysis. In addition, randomized surveys of insufficient
    sample size may miss rare hazardous events or small populations with
    high exposure or risk.

         Sampling strategies for survey studies include randomization
    methods for choosing subjects to enroll in the study. Simple random
    sampling is a scheme in which all sampling units of the same size have
    equal probability of being selected. It can be difficult to implement
    but relatively easy to generalize. Simple random sampling presents
    logistic and fiscal constraints when considered for exposure surveys
    that are large in geographic scope. For example, a national survey of
    5000 personal exposures to respirable particulate matter that utilizes
    simple random sampling may result in individuals selected from 1000
    cities and towns. The travel and site preparation costs of such a
    design may not be feasible in many situations.

         A variety of alternatives to simple random sampling exist that
    may be used to provide practical and efficient samples of large
    populations (Callahan et al., 1995).  Stratified sampling may be used
    to obtain more precise survey results if exposures are more
    homogeneous within strata than between them. Possible strata include
    urban, suburban and rural populations, or occupationally exposed and
    non-occupationally exposed individuals.

          Oversampling of target populations or contaminants also may
    yield substantial increases in the precision of results. Because the
    individuals anticipated to have the highest exposures to a particular
    pollutant may be rare in the population being studied, oversampling
    can be considered to obtain more precise estimates of exposure. Before
    committing substantial resources to oversampling, special care must be
    taken to ensure that assumptions or data used to support a rationale
    for selecting the oversampled population are accurate; otherwise
    erroneous oversampling may decrease the precision of the study results
    (Callahan et al., 1995).

          Multistage sampling designs utilize clusters of sampling units
    thereby limiting sampling locations to manageable areas. Depending on
    the scope of the study, the stages of probability sampling necessary
    may include:

    *  selection of primary sampling unit (e.g., a city)

    *  selection of sample area segments (e.g., blocks within the city)

    *  selection of sample housing units within sample segments (e.g.,
       residences within the blocks)

    *  selection of sample individuals within sample housing units

    *  selection of sample time points within the monitoring period
       (Callahan et al., 1995).

    The optimal degree of clustering depends on the variability of the
    survey variables between and within the clusters and the costs of
    fieldwork relative to sample collection and analysis costs. Although
    details of this approach are beyond the scope of this chapter, it
    should be recognized that cluster sampling introduces correlation
    among the sample individuals that affects the validity of the survey
    estimates. Thus, tradeoffs between increased sample size achieved
    through clustering and loss of validity must be considered carefully.
    Details of multistage and cluster sampling may be found in Hansen et
    al. (1953), Kish (1965), Cochran (1977), Kalton (1983), Kollander
    (1993) and Callahan et al. (1995).

         One concern with survey studies is maintaining participation of
    subjects who did not initially volunteer. Another issue, which is more
    conceptual, is subject selection for the more complex sampling
    strategies. In particular, stratification factors need to be carefully
    chosen so that potential confounders can be determined and the

    adjustments can be made from the resultant effects. Important
    considerations for the design of population-based (e.g., national or
    regional) exposure surveys, including response rates and confounders,
    are discussed by Whitmore (1988) and Callahan et al. (1995).

    3.4.3  Other sample types

         Non-probability sample studies ("anecdotal studies") may consist
    of selecting a sample based on the self-reporting of conditions, such
    as complaint cases for "sick building" syndrome. Data collected in
    this manner are potentially subject to biased reporting. It is
    difficult to generalize results unless causal relationships are very
    strong or unless there is little reason to believe that a confounder
    or an unmeasured significant factor is relevant. In general, such
    studies are used for description or exploration of a given situation.
    In particular, they can be used to evaluate the variability of
    outcomes and explore unknown situations for further explanatory study.
    When choosing subjects, it is useful to focus on variability in the
    expected outcome and also on the likelihood of completing the study.
    It is also helpful to focus on a simple, preferably dichotomous,
    hypothesis. Extensive validation will be necessary before accepting or
    rejecting the hypothesis since the generalization of the results is
    uncertain.

         The advantages of targeted anecdotal studies are the inexpensive
    and quick ways in which they aid in the design of future studies. For
    example, when exploring protocols, determining stratification
    variables, potential biases and confounders, and identifying the units
    of analysis, the use of cooperative volunteers can simplify field
    operations. The uncertainty of the results of these studies is due to
    potential biases from the non-random and possibly non-representative
    sample (i.e., responder bias). Since the population in such
    non-probability sample studies is often made up of volunteers, there
    is usually some factor present which distinguishes them from those who
    do not choose to participate. This factor could influence the results;
    in particular, those who participate may tend to consider themselves
    strongly affected or not affected by the pollutant being studied and
    may alter their responses or behaviours as a result. This phenomenon
    is a special case of responder bias, often termed  self-selection 
     bias. Also, a poorly designed study can fail to control for temporal
    and spatial variability, as well as meteorological, site and source
    bias. This bias is a result of a single, "random-day", or grab
    sampling and single-location sampling, which decreases the potential
    for generalization.

         Controlled experiments are useful to examine a few factors and to
    study their influence on the resulting exposure. The use of
    randomization and control ensures that the effects are real and not
    the result of confounding causes, incorrectly measured variables or
    missing variables. Examples include chamber studies and other
    situations where the investigator has control over most of the
    environmental factors.

    3.5  Exposure assessment approaches

         As discussed in Chapter 1, strategies for assessing environmental
    exposure can be categorized as one of two general approaches; direct
    or indirect.  Direct approaches include personal exposure monitoring
    and biological markers of exposure.  Indirect approaches include
    environmental sampling, combined with exposure factor information,
    modelling and questionnaires.

    3.5.1  Direct approaches to exposure assessment

         Direct measures of exposure include samples collected at the
    interface between an exposure medium and the human body, e.g., at the
    breathing zone in the case of air pollutant exposure, or samples of
    biological tissue in which concentrations of target pollutants can be
    quantitated. Measurements in food or drinking-water (duplicate
    portions) which are ingested could also be viewed as a direct way of
    assessing exposure through these media. Thus, direct approaches to
    exposure assessment include personal exposure monitoring and
    biological markers of exposure. Personal monitoring methods are
    discussed below, and the subject of biomarkers of exposure is
    presented in detail in Chapter 10.

         Personal monitoring of exposure to environmental contaminants
    refers to collection of samples at the interface between the exposure
    medium and the human receptor (e.g., the breathing zone). Personal
    monitoring approaches are summarized in Table 6. Personal monitors
    make it possible to measure exposures for an identified subset of the
    general population. Moreover, if study participants maintain records
    of their activities, then locations where highest exposure
    concentrations occur as well as the nature of emission sources can
    often be inferred. Personal monitoring can be done for all potential
    exposure media (e.g., air, water, soil, food) and pollutants of
    interest. Although available, personal monitoring methods may not be
    employed in a particular investigation due to study design, time or
    expense considerations. The principal limitation on the use of
    personal monitoring for exposure assessment is the availability of
    sample collection methods that are sensitive, easy to operate, able to
    provide sufficient time resolution, free from interferences and
    cost-effective. Consideration should be given to the likelihood that
    the inconvenience of complying with personal monitoring protocols may
    alter the normal behaviour of the study participants. For example,
    participants tend to wear personal air monitors on days that they do
    not go to work. In duplicate portion studies, participants may not
    provide equal portions of expensive or well-liked foods, leading to
    underestimation of intake. Approaches to personal monitoring of
    inhalation, dietary and dermal exposures are discussed below.

    Table 6.  Summary of personal monitoring approaches

                                                                          
    Exposure route   Media       Environmental sample   Biological sample
                                                                          

    Inhalation       air         personal monitor       breath
                                                        urine

    Ingestion        water       tap water              blood

    Ingestion        food        duplicate portion      faeces
                                                        breast milk

    Dermal           soil/dust   dermal patch           others
                                                                          


    3.5.1.1  Personal monitoring of inhalation exposures

         Personal monitoring of human exposure to air pollutants requires
    that study participants transport their sample collection device with
    them at all times during the assessment period. Examples include a
    diffusion tube used for passive sampling of gases, such as VOCs, or a
    filter with a battery-operated pump for active sampling of aerosols
    and their components (ACGIH, 1995).

         Personal air monitors can be grouped into two general categories:
     integrated samplers that collect the pollutant over a specified time
    period and then are returned to the laboratory for analysis, and
     continuous samplers that use a self-contained analytical system to
    measure and record the pollutant concentration on the spot.
    Instruments in both categories can be either active or passive.
     Active monitors use a pump and a power source to move air past a
    collector or sensor.  Passive monitors depend on diffusion to bring
    the pollutants into contact with the collector or sensor. Additional
    information may be found in Chapter 7 and ACGIH (1995).

         As Wallace & Ott (1982) pointed out, the direct measurement of
    exposures using personal monitors raises several methodological
    issues. Personal monitoring studies are complex, expensive, time
    consuming and labour intensive. Other challenges include selection and
    recruitment of representative subjects; distribution, maintenance and
    retrieval of many monitors; laboratory analysis of many air samples
    returned from monitors in the field or calibration and validation of
    many real-time monitors; and the transcription and statistical
    analysis of data on pollutant concentrations and time-activity
    patterns.

    3.5.1.2  Personal monitoring of dietary exposures

         Exposures to contaminants in food may be directly measured by
    collecting meals as prepared for consumption by members of the study
    population; such samples are often termed duplicate portion samples.
    Duplicate portion study designs provide food samples as actually
    consumed, rather than samples of unprepared, individual food items
    that are typical of surveillance approaches to characterizing dietary
    exposures (US NRC, 1993). This distinction is important because the
    method by which food is prepared for consumption (e.g., washed, washed
    and cooked, or commercially processed) can influence contaminant
    residues. In addition, some pollutants can be generated during
    cooking, for example, benzo [a]pyrene (Waldman et al., 1991a) and
    heterocyclic amines (Skog et al., 1998). Thus, residue levels measured
    in duplicate portion samples are likely to more accurately reflect
    personal dietary ingestion exposures than raw agricultural commodities
    and other foods collected at the producer, processor or distributor
    level. Depending on the objectives of the study, water may also be
    included as part of the duplicate portion sample.

         Duplicate portion study designs use either collection of
    individual servings or meals or composite samples. In studies of this
    type, participants are often monitored over one or more days, and the
    duplicate portion samples are collected daily over the monitoring
    period. The former affords a detailed examination of contaminant
    levels in specific commodities or foods comprised of several
    commodities; however, the analytical chemistry costs associated with
    this degree of temporal resolution may be prohibitive. Composite
    samples provide an integrated measure of dietary exposure and provide
    an efficient means for characterizing total dietary exposures. Both
    collection schemes require a high level of effort from study
    participants, and the complex food matrices may present analytical
    chemistry challenges.

         Duplicate portion studies require a high degree of participation
    by the study respondents, because they are primarily responsible for
    preparation and storage of an additional serving of every food or meal
    consumed over the monitoring period. This burden makes it difficult to
    collect representative samples of all foods consumed by the
    respondent, especially when food is relatively expensive or scarce or
    is consumed outside the home. Respondent burden also makes it
    difficult to conduct studies of chronic dietary exposures using the
    duplicate portion approach. Additional information on assessment of
    dietary exposure, including both direct and indirect approaches, may
    be found in Chapter 7.4 as well as WHO (1985a, 1995c); EC (1997a).

    3.5.1.3  Personal monitoring of dermal absorption exposures

         Personal monitoring of dermal exposure is used for those
    situations where a pollutant comes in contact with the skin and intake
    occurs via absorption through the skin. Dermal patches and skin wipe
    samples are used to evaluate exposures for residues adhering to the
    surface of the skin (US EPA 1992b; Fenske, 1993; Geno et al., 1996;

    Shealy et al., 1997). These methods have typically been used for
    industrial hygiene assessments where very high exposures are expected.
    Dermal patches and skin wipe samples have been used to characterize
    transfer of pesticide residues from soil and grass to skin as well as
    spot concentrations of residues on skin (Fenske et al., 1991). Dermal
    absorption can also occur during bathing, showering or swimming. In
    this case, the contaminant is in the water and exposure occurs when
    the water contacts the skin. Dermal exposure in this situation is
    defined as the concentration of the contaminant in the water and the
    duration of contact.

    3.5.2  Indirect approaches to exposure assessment

         Indirect measures of exposure include estimates derived from
    environmental monitoring (i.e., measurements made in locations
    frequented by the study participants), models and questionnaires.

    3.5.2.1  Environmental monitoring

         Indirect estimates of exposure may be made by combining
    measurements of pollutant concentrations at fixed sites with
    information on rates of contact with these media recorded in data logs
    and diaries or time-activity surveys. Examples include air pollutant
    concentrations in specific areas combined with time budget records
    (see Chapter 5), food contaminant data combined with information on
    dietary patterns (see Chapter 7.4 for details), and pollutant
    concentrations on skin combined with data on frequency and duration of
    hand-to-mouth contact. Although collection of environmental,
    time-activity and questionnaire data needed for this exposure
    assessment approach is simpler than for personal monitoring, it is
    still invasive and laborious, and may lead to selection bias.

         Microenvironmental monitoring is a special case of environmental
    monitoring in which the location where measurements are made is
    considered to be homogeneous with respect to concentrations of the
    target pollutants over the averaging time of interest. The concept of
    a microenvironment has been widely applied in air pollution exposure
    assessments. Examples of potentially important micro-environments used
    for air pollution exposure assessment are listed in Table 7. The
    general form of the equation used to calculate time-weighted
    integrated exposure from micro environmental monitoring data is

    FIGURE
                                                                (3.1)

    where  E is the time-weighted integrated exposure (e.g., mg/m3),
     C is the concentration (e.g., mg/m3),  t is the unit time (e.g.,
    minute),  T is the total elapsed time (e.g., minutes). The subscripts
     i, j and  k denote the medium, the pathway and the microenvironment
    respectively (Duan, 1982). The most important assumptions inherent in
    this model are:

    *  The concentration  Cj in microenvironment  j is assumed to be
       constant during the time that person  i is there.

    *  The concentration  Cj within microenvironment  j and the time
       that person  i spends there are assumed to be independent events.

    *  The number of microenvironments necessary to characterize personal
       exposure adequately is assumed to be small.

         The concept of a time-weighted integrated exposure is illustrated
    in Fig. 10. A unit width is indicated on the  j axis for each of five
    microenvironments: indoors at home, indoors at work, indoors in other
    locations, in transit, and outdoors. The concentration of respirable
    particles (RSP) is displayed on the  y axis, and the fraction of time
    that person  i spends in each microenvironment over the 24-h period
    is plotted on the  t axis. Even though the RSP concentration was low
    inside the home, it contributed significantly to the time-weighted
    exposure because this person spent 18 out of 24 h there. Conversely,
    the RSP concentration outdoors made only a minor contribution because
    this person was outdoors less than half an hour during the 24-h
    period.

         Indirect monitoring of ingestion exposures via hand-to-mouth
    contact may be obtained by collection of dermal wipe samples. However
    as indicated above, the use of this method has been limited to date. A
    drawback of the dermal wipe approach is that the integration time may
    be highly variable among subjects owing to variations in frequency of
    hand and body-washing, making interpretation of the results difficult
    (Fenske, 1993). Information on rates of contact between the
    contaminated skin and mouth is also required to complete the exposure
    assessment. A discussion of these types of data may be found in
    Chapter 5.

         Given the diversity of microenvironments that people move through
    each day (see Table 7), application of the indirect approach to
    exposure assessment is not straightforward. Its utility depends on
    identification of and sampling in the microenvironments with the
    greatest potential to influence human exposure. The costs and
    practical difficulties of monitoring in all, or even most, of the
    locations where people are likely to spend their time limits the scope
    of indirect measurements.


        Table 7.  Potentially important microenvironments for air pollution exposure assessment

                                                                                                                              
    Microenvironments      Comments
                                                                                                                              

    Outdoors
    Urban                  metropolitan areas where air pollution levels are high as a result of high density of mobile and 
                           stationary sources

    Suburban               small- to medium-sized cities where air pollution levels tend to be lower than in metropolitan 
                           areas, although transport of urban pollution can affect local air quality under certain conditions

    Rural                  agricultural communities and small towns with few major anthropogenic sources of air pollution. 
                           Air pollution levels tend to be low, although transport of urban and suburban pollution can affect 
                           local air quality under certain conditions

    Indoors-occupational
    Industrial             manufacturing and production processes, such as those in petrochemical plants, pulp mills, power 
                           plants, and smelters

    Non-industrial         primarily service industries where workers are not involved in manufacturing and production 
                           processes, such as insurance companies, law offices, and retail sales outlets

    Indoors-Non-occupational
    Residential            single-family houses, apartments, mobile homes, condominiums

    Commercial             restaurants, retail stores, banks, supermarkets

    Public                 post offices, courthouses, sports arenas, convention halls

    Institutional          schools, hospitals, convalescent homes

    Indoors-Transportation
    Private                automobiles, private aeroplanes

    Public                 buses, subways, trains, commercial aeroplanes
                                                                                                                              
    


    FIGURE 10

    3.5.2.2  Models as an indirect approach to assessing exposure

         The microenvironmental exposure equation describes a model
    commonly used for assessment of air pollutant exposure. More
    generally, models are useful tools for quantifying the relationship
    between pollutant exposure and important explanatory variables, as
    well as for expanding existing exposure information to estimation of
    exposures of new populations and subgroups, and future exposure
    scenarios. Validated exposure models reduce the need for expensive
    measurement programmes. The challenge is to develop exposure databases
    and models that allow maximum extrapolation from minimum measurements
    or costs. Such models need to reflect the structures of the physical
    environments and human activities of interest in exposure assessment.

         In addition to the essentially physical (deterministic) exposure
    models, physical-stochastic (probabilistic) and statistical
    (regression) models are used. The former type is particularly useful
    for population exposure distribution assessments, the latter requires
    less supporting information but cannot be used for extrapolation
    outside of the study population. Exposure models are discussed in
    detail in Chapter 6.

    3.5.2.3  Questionnaires as an indirect approach to assessing exposure

         Questionnaires typically provide qualitative, often
    retrospective, information. They may be used to categorize respondents
    into two or more groups with respect to potential exposure (e.g.,
    exposed or unexposed, high exposure or low exposure) and are commonly
    used for this purpose in epidemiological studies. As noted earlier,
    questionnaires may also be used to aid in interpretation of personal
    and environmental monitoring results.  A priori knowledge of the
    determinants of the exposure of interest is required to develop
    effective questionnaires relevant to exposure assessment (e.g., high
    formaldehyde exposure for workers in a certain industry, or high
    carbon monoxide and lead exposure for traffic policemen, bus drivers
    and road toll collectors). Most often the information necessary to
    develop questionnaires is obtained from previous studies that utilized
    environmental measurements, models or biological monitoring to measure
    exposure. In many cases, basic socio-demographic questionnaire data
    may provide extremely valuable information as they might be strong
    surrogates of exposure. It has long been known that rates of disease
    differ in social strata. In addition, it is readily apparent in many
    countries that the physical characteristics of one's residential
    environment are linked to income level. For lead exposure, differences
    in exposure among groups defined by income and social status have been
    demonstrated. Phoon et al. (1990) have shown that diet and job
    category were the most important predictors of blood lead levels among
    men in Singapore. In the USA, elevated blood lead levels have been
    linked to children who live in older, inner-city housing, particularly
    properties in poor repair (MMWR, 1997). Homes in these areas are more
    likely to have been painted with leaded paints (pre-1950) and have
    higher concentrations of lead in soil owing to deposition of emissions
    from leaded gasoline prior to the 1970s. Haan et al. (1987) found an

    increased risk of death among people living in a poverty area in the
    USA as compared to an adjacent non-poverty area, even after adjusting
    for differences in smoking, race, baseline health status, access to
    medical care, employment status, marital status, depression, sleep
    patterns and body mass index. These results suggest that sociophysical
    aspects of the environment, such as increased exposure to contaminants
    from poorer housing, may be important contributors to the association
    between socio-economic status and excess death rates.

    3.6  Summary

         A good study design is the most important element of any exposure
    assessment. It includes the purpose and objectives of the
    investigation as well as relevant methods for sampling, measurements,
    statistical analyses, and quality assurance. Methods for
    characterizing the magnitude, duration and time patterns of human
    contact with environmental contaminants may follow a direct approach
    or an indirect approach. Direct approaches to exposure assessment
    include point-of-contact measurements and measures of biological
    markers of exposure. Indirect approaches include environmental
    monitoring, modelling and questionnaires. These approaches may be
    employed in various types of exposure studies that are typified by the
    manner in which the study population is selected; for example,
    comprehensive studies that include all members of the study
    population, descriptive studies consisting of a non-probability
    sample, or surveys based on a randomly selected, representative sample
    of individuals.

    4.  STATISTICAL METHODS IN EXPOSURE ASSESSMENT

    4.1  Introduction

         Statistics is a necessary and critical tool in exposure
    assessment studies. Statistics can be employed at each stage of the
    exposure assessment study. At the planning stage, statistics aids in
    selection of study design and determination of the amount and form of
    data to collect. After the data are collected, statistical description
    of the results helps understanding of the basic characteristics of
    exposure and its determinants. Statistics is also essential during
    final analysis of the data for hypothesis testing, characterizing
    exposure through various routes and media, and exploring relationships
    between ideal measurements (e.g., exact lung uptake) and feasible
    measurements (e.g., ambient, indoor, or personal measures).
    Furthermore, statistical inference allows one to generalize the
    observations derived from a sample to a wider population from which
    the sample was drawn. Finally, as noted in Chapter 11, statistics play
    an important role in quality assurance (QA) programmes.

         Selected applications of descriptive and inferential statistics
    in exposure assessment studies are discussed in the following
    sections. This chapter is not a substitute for a course in statistical
    methods, but is intended to provide a brief review and useful
    references. Widely available statistical software for personal
    computers can be used to perform data processing and necessary
    calculations. One example of such packages is the statistical
    programme Epi Info developed for and distributed by WHO (Dean et al.,
    1995).

         Throughout the chapter, data collected as part of a lead exposure
    study performed in Malta and Mexico (WHO, 1985b) (Table 8) will be
    used to illustrate some key statistical concepts and methods. The
    purpose of this study was to investigate the relative importance of
    lead exposure via different routes of exposure. Blood lead
    concentrations were considered to be an indicator of lead uptake from
    all exposure routes, whereas faeces lead concentrations were
    considered to represent exposure via ingestion. In the course of this
    study, blood lead and faeces lead measurements were obtained from 36
    and 19 individuals in Malta and Mexico, respectively.

    4.2  Descriptive statistics

         Descriptive statistics summarize data in a simple manner to
    discern key points about the collected information. We typically
    assume that the collected data are a sample from a larger population
    of possible measurements and that the sample is representative of the
    population. The sample consists of the individual observations from
    the study population, with multiple variables or covariates recorded
    for each observation.  Univariate methods examine the distribution of
    a single variable;  multivariate methods describe relationships among
    two or more variables. That is, if we consider a single observation
    and know the value of one variable, multivariate methods indicate what

    Table 8.  Blood lead (PbB) and faeces lead (PbF) data from sample 
              populations in Malta and Mexico. Source: WHO, 1985b

                                                                          
                                 Malta                       Mexico
    Number                                                                
                           PbB          PbF            PbF          PbB
                           (µg/litre)   (µg/g)         (µg/litre)   (µg/g)
                                                                          

    1                      171           2.9           239          6.3
    2                      270          30.5           263          4.2
    3                      198           5.6           198          5.7
    4                      122           3.8           163          5.3
    5                       96          16.6           217          4.3
    6                      385          35.5           188          4.7
    7                      359          49.6           190          3.3
    8                      267           6.8           248          5.2
    9                      261           8.1           225          4.5
    10                     301          25.6           152          3.4
    11                     202           7.7           177          5.9
    12                     222          32.3           157          3.8
    13                     339          10.9           297          5.3
    14                     156           5.7           144          3.6
    15                     262          18.7           257          9.8
    16                     290          16.5           131          4.8
    17                     158           4.9           187          5.1
    18                     343          37.8           168          3.2
    19                     228           9.1           112          2.8
    20                     256          14.1
    21                     270           9.9
    22                     245           4.9
    23                     337          14.3
    24                     362          19.2
    25                     155           4.9
    26                     194           9
    27                     206           6.7
    28                     276          12.4
    29                     222          11.2
    30                     214          21.3
    31                     248           7.8
    32                     283          17.8
    33                     215          10.9
    34                     279          14.9
    35                     229           8.6
    36                     127          17.3
                                                                          

    Table 8.  (continued)

                                                                          
                                 Malta                       Mexico
    Number                                                                
                           PbB          PbF            PbF          PbB
                           (µg/litre)   (µg/g)         (µg/litre)   (µg/g)
                                                                          

    Median               246.5          11.1           188          4.7
    Mean                   243          14.8         195.4          4.8
    Standard deviation    70.9          10.8          49.5          1.6
    Standard error        11.8           1.8          11.4          0.4
    Minimum                 96           2.9           112          2.8
    Maximum                385          49.6           297          9.8
    Range                  289          46.7           185          7
                                                                          


    we can infer about the other variables. Both numerical and graphical
    techniques may be used to characterize the sample and any relevant
    subsets, and to obtain preliminary results from the study.

    4.2.1  Numerical summaries

         Numerical approaches include calculating descriptive statistics
    that describe the distribution of a variable (e.g., blood lead
    concentrations) in terms of central tendency and dispersion as well as
    descriptions of associations between pairs of variables. Other
    numerical descriptive measures can be used to describe points in the
    distribution (e.g., percentiles). Each of these descriptive statistics
    is described below and where appropriate the formulas used to
    calculate them are provided in Table 9.

         Standard measures of central tendency include the  sample 
     median (i.e., midpoint observation) and  sample mean (i.e.,
    average). Referring to the lower half of Table 8, note that the median
    blood lead concentration for the Maltese study population was 246.5
    µg/litre, intermediate between the eighteenth and nineteenth
    observations. Thus, 50% of the individuals in this sample had a blood
    level less than 246.5 µg/litre and 50% had a greater blood lead
    concentration. The sample mean blood lead concentration in the Maltese
    population was 243 µg/litre compared to 195.4 µg/litre in Mexico,
    indicating that blood lead levels were higher in the Maltese
    population. Methods for assigning confidence levels to statements such
    as this are described in Section 4.4. The sample mean is more precise
    for estimating the average of the distribution, but it is sensitive to
    measurement imprecision, errors and extreme values. Although the
    sample median is less precise for estimating the average, it is more
    robust with respect to errors in the data. Therefore, when outliers or
    extreme values are present, or when possible errors and contamination
    in the observed data are suspected, the median is likely to be a
    better descriptor of central tendency than the mean.

    TABLE 9

         Standard measures of dispersion include the sample variance, the
    sample standard deviation and the sample range (formulas in Table 9).
    These measures describe the spread of the observations. Examination of
    Table 8 reveals that blood lead concentrations are more variable in
    the Maltese sample population (standard deviation = 70.9 µg/litre,
    range = 289 µg/litre) than that in Mexico (standard deviation = 49.5
    µg/litre, range = 185 µg/litre). Measures of dispersion are useful for
    characterizing the degree of variability of a given measure among the
    members of a study population. As we will see later in this chapter,
    dispersion is also a key component of some study design issues.

         The concept of sample percentile is an important aspect of
    exposure assessment. A sample percentile for a variable in a data set
    is the value of the data such that at least  p % are at or below this
    value, and (1 -  p)% are at or above this value. A percentile is
    determined by first ordering the sample (i.e., rank from lowest to

    highest) and then identifying the observation that corresponds with
    the desired fraction of the data set. In the case of blood lead
    concentrations measured in the Maltese sample population, 283 µg/litre
    is the 75th percentile since it is the 27th of 36th rank-ordered
    values in the data set. Graphical representation of percentiles is
    discussed in the next section.

         Multivariate summary statistics allow one to evaluate
    relationships between or among different variables. Most of these
    examine correlation (the strength of the linear relationship) between
    variables, where the direction and magnitude of the relationship, or
    association, is described by the correlation coefficient  (p). The
    correlation coefficient ranges from -1 to +1, where negative values
    indicate an inverse relationship between two variables, positive
    values indicate a direct relationship, and values near zero, whether
    negative or positive, indicate a weak relationship. In the example
    case, the correlation between blood lead and faeces lead in the
    Maltese study population is 0.57, indicating these biomarkers of lead
    exposure have a moderate to strong positive association.

    4.2.2  Graphical summaries

         Graphical summaries of data provide illustrative information
    about the distribution of the observed values and associations between
    variables. Graphical presentations of data can suggest the shape of
    the distribution and aid in exploring hypothesized relationships
    between factors included in the study. In many situations and for many
    exposure analysts, graphical summaries of data convey information more
    readily than numerical summaries. Fundamental graphical presentation
    methods are described here. A description of advanced visualization
    methods may be found in Cleveland (1993) and Tufte (1983, 1997).

    4.2.2.1  Histograms

         Histograms are bar charts used to illustrate the relative
    frequency of values or ranges of values within an exposure metric.
    Observations are assigned to ranges of the data, and the height of the
    bar represents the frequency of observations in that range. It is
    important to note that the choice of ranges can be arbitrary,
    resulting in many possible different pictures of the results. A
    histogram of the Maltese blood lead data is shown in Fig. 11. Here,
    the data were grouped into bins with interval ranges of 25 µg/litre.
    Blood lead concentrations between 200-225 µg/litre and 275-300
    µg/litre were observed the most often. Histograms can be used to
    illustrate absolute or relative frequency.

    4.2.2.2  Cumulative frequency diagrams

         Cumulative frequency or probability diagrams can be used to
    graphically express percentiles of a distribution. A cumulative
    probability chart for the Maltese blood lead data is shown in Fig. 12.
    The value associated with a given percentile, or vice versa, can
    easily be determined from such a figure.

    FIGURE 11

    FIGURE 12

    4.2.2.3  Box plots

         A box plot is another approach for graphically describing the
    distribution of a measurements in an exposure study. Some details of
    box plots vary among users; however, all of them display the sample
    median, mean, 25th percentile and 75th percentile. Selected other
    values, such as 10th and 90th percentiles or 5th and 95th percentiles
    as well as the extremes (i.e., the minimum and maximum) of the
    distribution are displayed, too. Fig. 13 shows box plots of the blood
    lead measurements from the Maltese and Mexican sample populations. The
    bottom and top horizontal lines of each box denote the interquartile
    range (i.e., the 25th and 75th percentiles) and the solid horizontal
    line across the centre indicates the sample median. The dotted line
    across the box indicates the mean of the distribution. The whiskers on
    the boxes in Fig. 12 extend to the 10th and 90th percentiles of the
    distributions, and the open circles denote all observations beyond
    those percentiles. As illustrated here, box plots are a convenient
    method for displaying information on the central tendency, dispersion,
    symmetry and tails of an exposure measure.

    FIGURE 13


    4.2.2.4  Quantile-quantile plots

         Quantile-quantile plots can be used to compare the distribution
    of a variable with a different sample or a known distribution.
    Exposure measures are commonly compared to the normal or lognormal
    distribution (see section 4.3) for purposes of evaluating whether the
    normality assumptions inherent in numerous statistical inference
    methods are met. While a discussion of probability distributions and
    statistical inference methods is reserved for later in the chapter, a
    quantile-quantile plot is shown in Fig. 14. Here, the Maltese blood
    lead data are plotted against the standard normal distribution (see
    section 4.3). This special form of quantile-quantile plot is known as
    a  normal probability chart. Data that form an approximately straight
    line on such a chart are approximately normally distributed. Data that
    do not form a straight line follow a non-normal probability
    distribution.

    4.2.2.5  Scatter plots

         Scatter plots display the relationship between two exposure
    variables measured from the same unit of observation (e.g., a person
    or location). Scatter plots are useful for graphically illustrating
    associations that are summarized numerically by correlation
    coefficients. Possible results include noticeable linear or non-linear
    trends, the absence of trend (a big "cloud") or a general trend with
    some observations as outliers.  Outliers are observations that do not
    follow the trends of the rest of the data and can strongly affect
    estimates of associations by masking real effects. Outliers can be the
    result of measurement error, human error or a correct but abnormal
    observation. Regardless, all potential outliers should be checked for
    accuracy and corrected or removed if this is justifiable. Fig. 15
    contains a scatter plot of blood lead and faeces lead measurements
    made concurrently on the Maltese sample population. Note that the plot
    indicates a positive association between the two measurements, but
    that the relationship is not 1 : 1, i.e., a unit change in blood lead
    levels is not accompanied by a constant change in blood faeces
    concentrations. This observation is consistent with the correlation
    coefficient between these measures of 0.57 that was noted in the
    previous section.

    4.3  Probability distributions

         Most exposure measurements can be considered random variables;
    that is, the different values obtained for a measurement of a given
    type are a function of a set of causative variables that may or may
    not be known to the analyst (Ott, 1995). Statistics allows for
    analysis of random variables by incorporation of variation through
    probability. This addition of variation allows for the generalization
    of results to populations larger or different than the sample under
    consideration. 

    FIGURE 14

    FIGURE 15

         Continuous probability distributions are described by their
    probability density function (PDF), which provides the probability of
    an outcome taking values in a small interval, and by their cumulative
    distribution function (CDF), which describes the probability of an
    outcome being less than a particular value. The PDF and CDF are
    directly analogous to the concepts of a histogram and cumulative
    probability distribution discussed in Section 4.2.

         Probability models are used to make statements such as, "The
    probability that the daily maximum ozone concentration will be greater
    than 120 ng/litre today is 0.08." Such estimates can be based upon
    empirical evidence (i.e., by looking at the number of observed
    concentrations greater than 120 ng/litre in comparison with the total
    number of observed concentrations) or by choosing a distribution and
    parameters that describe the observed data. An example of the latter
    would be to model the distribution of blood lead levels in Maltese
    subjects as normally distributed with a mean of 243 µg/litre and
    standard deviation of 70.9 µg/litre and to use the properties of the
    distribution to estimate the probability. The amount of confidence in
    the accuracy of the estimates is related to the amount of data
    available and the sampling scheme used to collect the data, as well as
    the degree to which the mathematical distribution fits the observed
    data.

         Two standard distributions commonly used in exposure assessment
    for modelling continuous outcomes are the  normal and the
     lognormal distributions. The  binomial and  Poisson distributions
    are often used in exposure studies as well. Many other probability
    distributions are available which have more flexibility (Johnson &
    Kotz, 1970a,b), but these four are frequently used and thus warrant
    attention here.

    4.3.1  Normal distribution

         The normal distribution, also known as the  Gaussian 
     distribution, is one of the most important statistical
    distributions. It is characterized by a symmetric, bell-shaped
    frequency distribution and is commonly used as a basis for analysis of
    environmental exposure data. Usually, a random variable  (X) that
    follows a normal distribution with mean µ and variance rho2 is
    denoted by  X ~ N(µ, rho2). The probability density function of the
    normal distribution with parameters µ and rho2 is given in Table 10.

         Since the cumulative distribution function cannot be integrated
    in a closed form, the best we can do is to numerically compute the
    integral. The values µ = 0 and rho = 1 specify the  standard normal 
     distribution. The values of the CDF for the standard normal
    distribution have been tabulated and are available from most
    statistical textbooks and computer packages. The capital letter  Z is

    usually reserved to denote a standard normal random variable, i.e.,
     Z ~ N(0,1). The normal distribution ranges from positive infinity to
    negative infinity and is symmetric. Equation 4.7 can be used to
    transform any normal random variable  X to a standard normal random
    variable (Table 10). Standardized normal random variables are useful
    for computing the probability of an event occurring, e.g., the
    likelihood that someone in Malta has a blood lead concentration
    greater than 384 µg/litre. Assuming the Maltese blood lead data
    presented earlier are representative of the general population and the
    blood lead concentrations are approximately normally distributed, the
    standard normal distribution can be used to calculate the desired
    probability.

    4.3.2  Lognormal distribution

         Many exposure measurements are strictly positive and right skewed
    (i.e., asymmetric). Examples include the size distribution of
    suspended particulate matter, personal exposures to various air
    pollutants and human time-activity patterns. The lognormal
    distribution is one possible model for describing data with these
    characteristics. The natural log (ln) transform of a lognormally
    distributed random variable has the properties of a normally
    distributed random variable. In other words, the distribution defined
    by the mean (µln x) and standard deviation (rholn x) of the
    ln-transformed values is bell-shaped and symmetric and can be
    standardized according to the procedure outlined in the previous
    section. Exponentiation of µln x and rholn x gives values termed the
    geometric mean (GM) and geometric standard deviation (GSD),
    respectively. The GM and GSD can also be used to define a lognormally
    distributed exposure measure.

         A histogram of the blood faeces data from the Maltese sample
    population is presented in Fig. 16a. The data depart from normality as
    they are clearly right skewed. The histogram in Fig. 16b shows that
    the ln-transformed values are approximately symmetric and indicates
    that the data approximate a lognormal distribution rather than a
    normal distribution. In this data set, µln x = 2.5 and rholn x = 0.7
    with corresponding GM = e2.5 = 11.8 µg/litre and GSD = e0.7 = 2.0. The
    degree to which the lognormal distribution accurately describes the
    data can be evaluated by plotting the raw data on lognormal
    probability paper. This procedure is identical to that described in
    relation in to Fig. 13, except that the  y axis is expressed on a
    logarithmic scale. The Maltese faecal lead data are plotted on
    lognormal probability paper in Fig. 17. The nearly straight line
    formed by the faecal lead measurements displayed on a logarithmic
    scale versus  Z indicates that the data are approximately lognormally
    distributed.

    4.3.3  Binomial distribution

         In some situation, the analyst may be interested in
    characterizing the frequency of a binary exposure outcome (e.g.,
    yes/no; true/false). The binomial distribution is useful for modelling

    TABLE 10
    FIGURE 16a;V214EH21.BMP

    FIGURE 16b;V214EH22.BMP

    FIGURE 17

    binary responses. The possible responses can be generally labelled as
    success or failure. Often we are not interested in a single outcome,
    but rather in the number of successes  (k) and failures  (n - k) for
    a specific number  (n) of repeated independent trials for the
    outcome. The probability of exactly  k successes in n independent
    trials, given a probability of success  (p) in a single trial, is
    given by the binomial probability distribution ( Pk) in Table 10.

         For example, assume daily exceedances of an ozone air quality
    standard are independent events in a study of 1-year and 3-year time
    periods. Let  k be a random variable describing the total number of
    exceedances encountered in a 1-year period ( n = 365 days). Further
    assume from historical data that the expected number of exceedance
    days each year is 1, thus  p = 1/365 = 0.00274. The calculated
    probabilities of  k days of exceedance per year are shown in Table
    12. Examination of the resulting probabilities in this example reveals
    a right-skewed distribution with the greatest probability occurring
    between  k = 0 and  k = 4 days.

    4.3.4  Poisson distribution

         Some exposure-related measurements are expressed as a rate of
    discrete events, i.e., the number of times an event occurs per unit
    time, such as the frequency (times per week) that a person consumes an
    ocean fish containing a methylmercury concentration greater than
    5.0 ppm. The Poisson distribution is used for describing potentially
    unlimited counts or events that take place during a fixed period of
    time (i.e., a rate), where the individual events are independent of

    one and other. The Greek letter lambda is typically used to denote the
    average or expected number of counts per unit time. In a Poisson
    distribution, the parameter lambda also describes the variance of the
    random variable. We can think about this intuitively by noting that as
    the expected number of counts or events increases (i.e., the rate of
    events increases), the amount of variability should increase as well.
    For example, if we expect a count of 1 then it is not too difficult to
    imagine observing 0 or 2. Likewise, if we expect a count of 20 000
    then it is not difficult to imagine 20 100 or 19 900 as reasonable
    observations. However, the variance is definitely larger in the second
    case. The formula used to compute the probability of a specific number
    of counts being observed over a fixed time interval is listed in Eq.
    4.11 of Table 10.

         For example, the Poisson distribution can be used in an exposure
    model to characterize the frequency with which a person comes in
    contact with a contaminant; say, the number of times per day a person
    encounters benzo [a]pyrene associated with environmental tobacco
    smoke. Assume that based on existing data, the expected number of
    encounters is anticipated to be 2 per day. Using Eq. 4.11, with lambda
    = 2, there is a 9% chance that an individual will have 4 (i.e.,
     n =4) encounters with benzo [a]pyrene on a given day. Subject to
    limitations associated with the independence assumption noted above,
    the Poisson distribution can be a useful exposure modelling tool.

    4.4  Parametric inferential statistics

         Inferential statistics is the process of using the observed data
    and assumptions about the distribution and variation of the data to
    draw conclusions. The two complementary components of inference are
     parameter estimation (either point or interval estimation) and
     hypothesis testing. Only frequentist, or classical, inference will
    be discussed here. However, Bayesian statistical inference, as well as
    decision theory, can be valuable for incorporating other aspects such
    as prior belief and loss into a statistical analysis, and they are
    worth consideration. Further information on Bayesian statistics may be
    found in Carlin & Louis (1996).

    4.4.1  Estimation

         Exposure measurement data can be used to estimate the parameters
    of a model (e.g., a probability distribution), especially those that
    describe the mean and variance of the variable. The two types of
    common reported estimates are  point estimates and  interval 
     estimates.

         Point estimation for quantities is commonly performed using
    maximum likelihood, ordinary least squares or weighted least squares
    methods. All estimates are chosen because they optimize (i.e., find
    the maximum or minimum of) some objective function such as the
    likelihood function or squared error function. One example is the
    sample mean for the population mean when the data are normal, using
    maximum likelihood, or for any data, using least squares.

         Two different forms of interval estimation are used to
    characterize variability in point estimates. The first is based on
    error propagation and is the result of simulating data to see what
    distribution of results might be expected under the model; the second
    is the usual statistical notion of confidence intervals. This approach
    focuses more on the variability of a modelled outcome due to
    variability of the input, and is useful in designing studies and
    determining which factors will have the greatest effect on the
    variability of the exposures. These procedures are described more
    fully in Chapter 6.

         The second form of interval estimate, the statistical (1-alpha)%
    confidence interval, gives a range of estimates, for a parameter,
    which is generated in a manner such that it contains the true
    parameter value (1-alpha)% of the time. For a normally distributed
    random variable, a one-sided confidence interval for the estimate of
    the mean is derived from the standard error and  Z1-alpha, while
     Z1-alpha/2 is used for a two-sided confidence interval. The standard
    error (rho×) is an expression of uncertainty about the mean and is
    calculated as the standard deviation divided by the square root of the
    number of observations  (n) (Table 9). Continuing with the example
    from the Maltese study, the standard error of the blood lead sample
    data is 11.8 µg/litre (Table 8). For alpha = 0.05, the two-sided 95%
    confidence interval about the estimated mean is computed as 243
    µg/litre ± 23.1 µg/litre, where the latter is equal to  Z1-(alpha/2) ×
    rho× or 1.96 × 11.8 (Table 11). Details of this procedure and
    related considerations may be found in most introductory statistics
    textbooks, for example Kleinbaum et al. (1988).

    4.4.2  Measurement error and reliability

         The term measurement error refers to the accuracy and precision
    of a given sample collection and analysis methodology.  Accuracy 
    describes the degree to which a measurement is free of bias.  Bias is
    systematic deviation in a measurement from the true value of the
    process being measured.  Precision refers to the reproducibility of a
    particular measurement system. Measurement reliability is a closely
    related concept in that a measurement with a high degree of accuracy
    and precision can be considered to be more reliable than one with a
    low degree of accuracy and precision. Additional information on
    measurement error and reliability is contained in Chapter 11, where
    the topic is discussed in the context of QA in exposure studies.
    Methods for assessing the accuracy of an exposure measure are also
    discussed in Chapter 11. Here, an approach for quantitatively
    estimating the precision of an exposure measurement system is
    presented.

         Statistical analysis of environmental samples collected
    simultaneously in space and time can be used to estimate the precision
    of a measurement method. Such samples are often referred to as
     duplicates and are often collected in pairs. The difference in the
    measurement parameter (e.g., concentration) between a duplicate pair

    Table 11.  Standard normal cumulative probabilities

                                                                      
    z              p(Z < z)                z              p(Z < z)
                                                                      

    -4.265         0.00001                 0              0.50
    -3.891         0.00005                 0.126          0.55
    -3.719         0.0001                  0.253          0.60
    -3.291         0.0005
    -3.090         0.001                   0.385          0.65
    -2.576         0.005                   0.524          0.70
    -2.326         0.01                    0.674          0.75
                                           0.842          0.80
    -2.054         0.02                    1.036          0.85
    -1.960         0.025
    -1.881         0.03                    1.282          0.90
    -1.751         0.04                    1.341          0.91
    -1.645         0.05                    1.405          0.92
                                           1.476          0.93
    -1.555         0.06                    1.555          0.94
    -1.476         0.07
    -1.405         0.08                    1.645          0.95
    -1.341         0.09                    1.751          0.96
    -1.282         0.10                    1.881          0.97
                                           1.960          0.975
    -1.036         0.15                    2.054          0.98
    -0.842         0.20
    -0.674         0.25                    2.326          0.99
    -0.524         0.30                    2.576          0.995
    -0.385         0.35                    3.090          0.999
                                           3.291          0.9995
    -0.253         0.40                    3.719          0.9999
    -0.126         0.45                    3.891          0.99995
     0             0.50                    4.265          0.99999
                                                                      


    is indicative of the precision of the collection and analysis
    methodology. Descriptive statistics generated from a set of
    differences between duplicate samples can be used to characterize the
    average degree of precision as well as variability in precision.

         Consider a hypothetical study of respirable particulate matter
    (RSP) in outdoor air where 20 duplicate pairs of 24-h average
    measurements were made. Assume the average 24-h average concentration
    among the 40 measurements was 50 µg/m3. Further assume that the
    distribution of differences between the 20 pairs of duplicate samples
    was normally distributed with a mean and standard deviation of 5 and 1
    µg/m3, respectively. On average, then, a single measurement can be
    expected to be within 5 µg/m3 of the actual concentration. Utilizing

        Table 12.  Probability distribution for the number of exceedances, using the binomial model 
               with expected number of exceedances of 1.0

                                                                                               
                                 1-Year period                           3-Year period
                                                                                               
    Number of            Probability      Cumulative            Probability       Cumulative
    exceedances                           probability                             probability
    k                    Pk {k}           Fk{k}                 PM{k}             FM {k}
                                                                                               

    0                    0.36737          0.36737               0.04958           0.04958

    1                    0.36838          0.73576               0.14916           0.19874

    2                    0.18419          0.91995               0.22414           0.42288

    3                    0.06123          0.98118               0.22435           0.64723

    4                    0.01522          0.99640               0.16826           0.81549

    5                    0.00302          0.99942               0.10086           0.91635

    6                    0.000498         0.999920              0.05034           0.96670

    7                    0.000070         0.9999904             0.02152           0.98821

    8                    0.0000086        0.9999989             0.00804           0.99625

    9                    0.0000009        0.9999999             0.00267           0.99892

    10                   0.00000009       0.9999999             0.00080           0.99972

    11                   0.000000008      0.9999999             0.00022           0.99993

    12                   0.000000001      0.9999999             0.00005           0.99998
                                                                                               
    

    concepts presented in Chapter 4.4.1, a single measurement can be
    expected to be within approximately 3-7 µg/m3 of the true
    concentration 95% of the time, i.e., within ±2 standard deviations of
    the average difference.

         For a probability distribution, the coefficient of variation is
    defined as the ratio of the standard deviation to the point estimate
    of the mean. In this way, the coefficient of variation error describes
    the degree of dispersion of a data set relative to a measure of its
    central tendency. The coefficient of variation provides a quantitative
    estimate of the relative degree of variability among the observations

    in a data set. Using data from the hypothetical example described
    above, the coefficient of variation among the pairs of duplicate
    samples is 0.2. Thus, on average, a single measurement can be expected
    to be within 20% of the actual concentration.

    4.4.3  Hypothesis testing and two-sample problems

         Exposure assessments are often performed to determine whether the
    level of exposure to a pollutant is different between two or more
    groups of people or locations or periods of time. Additional
    attributes typically considered to be determinants of exposure include
    any number of demographic factors (e.g., age, gender, ethnicity) and
    behaviour patterns. This section describes the statistical procedure
    used to address this type of study objective.

         Statistical hypothesis testing is a procedure where a choice is
    made between two hypotheses that are not weighed equally; the null and
    the alternative. The  null hypothesis typically reflects what can be
    stated with confidence about a particular phenomenon on the basis of
    existing information. In practice, concluding that the null hypothesis
    is false indicates that the data provide strong evidence for a
    departure from conventional wisdom or practice. Thus, hypothesis tests
    are generally constructed such that the conclusion will lie with the
    null unless the evidence strongly suggests otherwise.

         Two types of errors can arise from hypothesis testing:

    *  concluding that the alternative hypothesis is true when it is in
       fact false (false negative)

    *  concluding that the null hypothesis is true when in fact it is
       false (false positive).

    The first type of error is known as a  type I error and the second
    one is a  type II error. The probability of a type I error is denoted
    by alpha and the probability of a type II error by ß. Only alpha is
    considered in the construction of the hypothesis test. However, as
    described later, both type I and type II errors are considered in
    sample size determinations.

         The general procedure for common tests that try to determine if
    some factor has an effect on the exposure outcome is as follows: a
    test statistic is constructed whose value is known if the null
    hypothesis is true. For example, if the null hypothesis is that the
    population mean is 1 (H0: µ=1), then under the null hypothesis, × =
    0, where × is the sample mean. Next, adjustments are made so that
    the distribution of this test statistic is known. For example, with
     s denoting the sample standard deviation and  n the sample size,
    the test statistic  T defined by Eq. 4.12 in Table 13, where  T has
    a distribution which follows a  t-distribution with  n-1 degrees of
    freedom. Now, using the known distribution of the test statistic, we
    construct ranges of values for which we reject (rejection region) and

    fail to reject (acceptance region) the null hypothesis. The rejection
    region is any area which has probability alpha, usually chosen to
    correspond to likelihoods between 0.025 and 0.05.

    TABLE 13


         A large number of problems in exposure assessment involve the
    comparison of two groups, for example, control and treatment; old
    method and new method; or normal and abnormal. If we focus on the
    location problem, where the means or the medians are compared, then
    depending on the assumptions we make with regard to the data,
    different tests can be performed. Assumptions typically made include:

    *  The data consist of a random sample from population 1 ( X1,i,
       i = 1, ...,  n), and a random sample from population 2 ( X2,i =
       1, ...,  n2)

    *  The two samples are independent of each other.

    *  Observed variables are on a continuous scale.

    *  Measurement scale is at least ordinal.

    *  Population 1 ( X1) has approximately the same distribution as
        X2.

         If we assume that the data follows a normal distribution and that
    the data are independent, with the first group distributed  N(µ,
    rho21) and the second group distributed  N(µ, rho22) so that the
    variances are possibly different, a test can be constructed to see if
    the difference (Delta) between the means for the groups is equal to a
    hypothesized value (Delta0), typically set to zero. This scenario
    would result in a two-sample  t-test, and the test statistic is
    presented in Eq. 4.13 in Table 13, where  t is compared with a

     t-distribution with  df = min( n1-1,  n2-1) degrees of freedom,
    and  si2 is computed as described in section 4.2.1. The possible
    alternatives are that Delta > Delta0, Delta < Delta0, or the
    general alternative that Delta not equal Delta0. If we are looking
    for differences, we reject the null hypothesis that the groups are the
    same for the respective alternative if  t >  Tdf,alpha,  t < -
     Tdf,alpha, or | t| >  Tdf, alpha/2, where alpha is the prespecified type I
    error for the decision to be made.

         Referring once again to the blood lead example presented earlier,
    the following null hypothesis may be tested: mean blood lead
    concentrations in the Maltese sample population are equal to those in
    the Mexican sample population. The corresponding alternative
    hypothesis is: mean blood lead concentrations are not equal in the two
    sample populations. As indicated in Fig. 13, the point estimates of
    the respective sample means are different. Completion of the
    two-sample  t-test will allow for determination of whether the
    difference is statistically significant with 1-alpha% confidence.
    Using Eq. 4.13, the  t-statistic is computed to be 3.30. Setting
    alpha = 0.05, the critical  t-value is 1.96. Thus, the Maltese and
    Mexican sample mean blood concentrations are significantly different
    at the 0.05 level.

    4.4.4  Statistical models

         Statistical models make explicit the potential sources of
    variability to be measured. The response, exposure, is dependent upon
    a combination of measured factors and background variation from
    unmeasured influences. For example, in examining pesticide exposures,
    one might consider methods and amounts of applications, climate
    conditions and duration of potential exposure. Unmeasured factors
    might include exact knowledge of individual behaviours and locations,
    which may cause different levels of exposure between two individuals
    who are equal with respect to other exposure characteristics. One must
    consider as many of the potential relationships between the responses
    as possible, as well as how the possible factors will affect each
    other, before finalizing a study design.

         Since no simple model will perfectly describe all relationships,
    the goal is to construct a parsimonious model that describes the major
    factors of the process resulting in exposure. For example, in studying
    the exposure of children to lead, the presence of lead in paint, in
    house dust or in water could be important factors, whereas gender and
    age might have an indirect effect on exposure by influencing the
    location and patterns of play. However, both types of data will be
    important in determining response, even though one is only an indirect
    cause. The average outcome described above could be the annual average
    exposure to lead or perhaps the maximum daily exposure, depending upon
    whether a cumulative or a threshold effect is the focus. 

         As noted in Chapter 3, by considering the statistical model
    before finalizing the study design one can help ensure that most
    influential factors would be accounted for, and more importantly, that
    the true effects of factors can be estimated from the study data. It
    is possible to design a study where some influential factors were not
    accounted for. Suppose there is interest in the effects of location
    and time of day on outdoor ultraviolet radiation exposures. If
    measures are only taken at one site at one time of day, and then at
    another site at a different time of day, then the effects of location
    and time of day are not distinguishable from the collected data.

         The mean, or average, outcome is the most common summary used for
    modelling and testing of situations of different conditions, but other
    parameters, such as the variance, the percentiles or the median, can
    be used for estimation and testing. Common models and statistical
    analyses, such as the multiple linear regression model, the  t-test
    and analysis of variance (ANOVA) use the mean for modelling and
    testing. The models can be as simple as taking the sample, dividing it
    into groups and comparing the means in the different groups. The
    models can also be as complex as trying to construct a physical model
    for the means with the addition of terms which incorporate randomness
    due to unmeasured factors or other sources of variation.

    4.4.4.1  Analysis of variance and linear regression

         ANOVA is a technique for assessing how several nominal
    independent variables affect a continuous dependent variable, and is
    usually concerned with comparisons involving several group means.
    Regression and ANOVA models are closely related and can be analysed
    within the same framework. The major difference is that for ANOVA, all
    the independent variables are treated as being nominal; whereas for
    regression analysis, any mixture of measurement scales (nominal,
    ordinal, or interval) is allowable for the independent variables.
    Examples of ANOVA used in exposure assessment can be found in Liu et
    al. (1994a), who used ANOVA models to examine the effects of wind
    speed, ozone concentrations, human subject and interaction between
    wind speed and concentration on the performance of an ozone passive
    personal sampler.

         Estimation for both ANOVA and linear regression models consists
    of obtaining point estimates for the parameters that describe the mean
    exposure under a certain set of conditions. Part of the estimation
    procedure is to determine how well the model fits. The first
    diagnostic is to examine the residual error (residual). A residual is
    simply the difference between the exposure estimated by the model and
    the actual exposure. By examining the residuals, one can determine for
    what ranges of actual exposures or conditions the model does not fit
    well, and use this to decide how to adjust the model.

         The simplest design (and corresponding model) occurs when
    measurements are taken while varying only one possible factor over a
    finite,  k, number of levels. Consider PM2.5 exposure; let the factor
    be the time of day when the levels are measured. For simplicity,
    divide time into three categories -- morning, afternoon, or evening --
    so  k = 3. If there is no known or hypothesized functional form for
    the relationship, the resulting abstract model for exposure,  Y, 
    during a particular time period,  i, should be the sum of the mean
    (average) during the time period  i, denoted by gammai, and an
    error, epsilon, which will represent the natural variation of the
    measurement. It is common to assume that the variation of the outcome
    is the same among all levels of the factor; this assumption is known
    as  homoscedasticity. 

         This model is referred to as the one-way ANOVA. The resulting
    model for the observed data, with  Yij denoting the  jth PM2.5
    measurement collected during the  ith time period with  i ranging
    from 1 to 3, is  Yij = gammai+ epsilonij where  gammai represents
    the average outcome due to the  ith factor level (in this example
     i ranges from 1 to 3), and  epsilonij (the error term) represents
    independent random variation. One common assumption is that the error
    terms follow a normal distribution with variance rho2. The parameters
    which need to be estimated in this model from the data are the means
    of the subsamples, gammai, and the variance of the outcome, rho2.
    This type of model, which compares the means of distinct groups, is
    the basis for ANOVA.

         Increasing the level of complexity leads to multiway or
    multifactor ANOVA as well as the multiple linear regression model,
    which is a more specific model for the effects of independent
    variables on the dependent variable. Let  Y denote the exposure level
    for a particular person or location; this is the dependent variable.
    Let  X, ..., Xn denote  n independent variables (known as
    covariates) which potentially influence the exposure level  Y. If the
    assumption of the existence of a linear relationship between the
    independent and dependent variables is reasonable, then a model for
    the outcome,  Y, based on the covariates  Xi, can be written as

    FIGURE

    where the information not conveyed by the covariates results in the
    error (epsilon), which is assumed to be normally distributed. theta0
    denotes the average exposure when all the  X values are zero, and
    thetai denotes the change in exposure for a unit change in the  ith
    variable. An example would be 24-h personal exposures to nitrogen
    dioxide. In this case, the factors may be distinct times and locations
    (or microenvironments) where nitrogen dioxide exposure may occur; for
    example, outdoors, indoors while home cooking on a gas range, and in
    an automobile.

         A regression model is used to evaluate the relationship of one or
    more independent variables  X1,  X2, ...,  Xk to a single,
    continuous dependent variable  Y. It is often used in exposure
    assessment to characterize the relationship between the dependent and
    independent variables (continuous and discrete) by determining the
    extent, direction and strength of the association. For example, in the
    particle total exposure assessment methodology (PTEAM) study, indoor
    PM2.5 concentration  (Y) was regressed against outdoor air
    concentrations ( X1), smoking rates ( X2), cooking durations
    ( X3), air exchange rates ( X4) and house volumes ( X5) to
    determine the major factors affecting indoor PM2.5 concentrations
    (Ozkaynak et al., 1996).

         Further information on ANOVA and linear regression may be found
    in Ott (1995), Kleinbaum et al. (1988) and most introductory
    statistics textbooks.

    4.4.4.2  Logistic regression

         An approach which is different from the linear or additive
    relationship described above is to consider a categorical outcome for
    exposures, e.g., exposure measurements grouped into ordinal levels
    such as low, medium and high. When the response is binary, that is, if
    an exposure is either present or absent (i.e.,  a threshold effect), 
    then a linear relationship is not appropriate. In this case, we must
    use an alternative model, for example:

    logit  P(Y = 1) = alpha + ß1 X1 + ß2 X2 + ... + ßk Xk

    where the logit function is logit  (x) = ln ( x/(1- x)), the
    function  P(Y = 1) denotes the probability that the response variable
     Y will take on the value 1 (denoting "success"), and the role of
    epsilon from the previous model is taken by modelling the parameter
    representing the probability that  Y = 1, as opposed to  Y itself.
    This model is known as logistic regression. In this model, alpha
    denotes the baseline odds for exposure given that the associated
    factors,  X ..., Xk , are zero, and ßi denotes the change in the
    log-odds that the response is  Y = 1 given a 1-unit increase in
     Xi. This approach can be adjusted to allow for the analysis of
    other types of categorical outcomes (McCullagh & Nelder, 1990). One
    common parameter which describes logistic regression results is the
     odds ratio. For a particular set of covariates,  Xi, the odds of
    the event occurring ( Y = 1), is exp(ß Xi). To compare the odds
    ratio for two situations, compute the first set of odds and take its
    ratio over the second odds. Usually, the situations will be identical,
    except that the covariate of interest will be zero for one of the
    sets. For example, if the model for the linear combination of
    covariates is 1.3 +2.5 X, then the odds ratio for  Y = 1, comparing
     X = 1 versus  X = 0, is e(1.3+2.5)/e1.3. A similar computation can be
    done when  X is a continuous random variable, for two different
    values of  X. In exposure assessment, a logistic regression model

    could be used to evaluate the importance of demographic or temporal
    factors on the likelihood that an individual will engage in an
    activity such as applying pesticides.

    4.4.5  Sample size determination

         Hypothesis testing attempts to determine if the data reject or
    fail to reject a particular (null) hypothesis. The test is based upon
    statistical considerations, and hence just reports how likely or
    unlikely the null hypothesis is. If there is minimal information, it
    will be difficult to statistically reject the null hypothesis; hence,
    sample size calculations are done in order to ensure that there is
    sufficient information from which to make a decision. The decision is
    between two unequally weighted hypotheses; the first is the null
    hypothesis, H0, which is the safe hypothesis, and the second is H1,
    the alternative or sceptical hypothesis, which requires sufficiently
    large evidence to believe in. The specifications of the test, based
    upon sceptical scientific belief or common usage, are the type I error
    and the type II error (defined in section 4.4.3). Introductory
    information on sample size determination is presented here; the reader
    is referred to Lemeshow et al. (1990) for details.

         To determine the minimum sample size required to observe the
    desired outcome, one needs to determine the smallest effect that is
    scientifically worth detecting (i.e., based on measurement limit or
    scientific principles), and use that to collect a sample with enough
    information to detect such a difference. The effect is the minimum
    significant difference in exposure between two groups. The smaller the
    effect, the more information is required to distinguish it. This
    effect is related to the type I error of a hypothesis test. There are
    two components, the type II error and the sample size, which are
    unspecified. When the type II error is specified, the resulting sample
    size can be determined. Once the sample size is determined, the power
    (1-type II error) of a study can be computed. The smaller the
    difference to be detected, the larger the sample size needed for fixed
    type I and type II error probabilities.

         The following describes the formula for a two-group comparison.
    To compare the means of two groups with equal sample size, let Delta
    represent a scientifically significant difference that we would like
    to detect, if it exists. Suppose that the first group can be modelled
    by  Y1 = µ1+epsilon1, where µ1 is the mean and epsilon1 ~
     N(0,rho2) is the error term, and the second group can be modelled
    by  Y22 + epsilon2 with the error term epsilon2 ~  N(0,
    rho2). The only difference between the two groups is the mean
    response. The groups will be considered statistically different only
    if |µ21| > Delta. The minimum number of observations in each
    group is given by Eq. 4.14 in Table 14 where  zgamma is defined as
    the value satisfying  P(Z >  z) = gamma, where  Z follows the
    distribution of a standard normal random variable, alpha is the type I
    error, and ß is the type II error. This formula can also be used to
    approximate the sample size needed for a difference of proportions

    (e.g., for dose-response models comparing two groups), by letting
    Delta represent the difference in proportions (instead of a difference
    in means).

    TABLE 14


    4.5  Non-parametric inferential statistics

         Each of the statistical analysis methods described previously
    assumes that the data can be adequately described by a probability
    distribution with known parameters, and that distribution can be
    transformed, if necessary, to meet the assumptions of the statistical
    model (e.g., normally distributed, independence, etc.). Many
    exposure-related data sets do not fit this description, however. One
    reason for this is that the data may not be normally distributed or
    cannot be transformed so that they are approximately normal. A more
    common reason is that although the underlying distribution of the
    population from which the sample is drawn may be reasonably assumed to
    be approximately normal or lognormal, there are too few samples to
    allow the nature of the underlying distribution to become apparent. In
    exposure studies sample sizes are often small (e.g., 10 or less)
    because of logistic difficulties in collecting samples and the expense
    of collecting and analysing the samples. In this case, the point
    estimates of the standard deviation and standard error are considered
    to be highly unstable. Consequently, confidence intervals generated
    using the estimation methods described above are considered to be
    unreliable. Non-parametric statistical analysis methods can be used to
    analyse data with these characteristics.

         Non-parametric statistical methods rely on rank statistics, i.e.,
    the order of observations in a data set. Glantz (1987) provides a
    concise introduction to non-parametric statistical methods with regard
    to health statistics. The sign test and Mann-Whitney rank sum test are
    two non-parametric methods for evaluating the equivalence of the
    median from two sample populations. These methods are analogous to the
    two-sample  t-test described in Section 4.4.3. The Kruskal-Wallis
    test is analogous to the  k-sample ANOVA method described in Section
    4.4.4 and is used to test whether the medians of more than two sample
    populations are equal. For further information on this topic, the
    reader is referred to Mosteller & Rourke (1973), a classic text on
    non-parametric methods, and also to Gilbert (1987) and Ott (1995) for
    a discussion of statistics based on rank order in an environmental
    context.

    4.6  Other topics

         Many new developments in statistical theory can be applied to the
    analysis of exposure assessment data. These include the topics of
    measurement error, missing data, spatial statistics, non-linear
    models, mixed effects, generalized mixed effects models, simulation
    models (e.g., Monte Carlo analysis), as well as others. Modern
    computing methods such as re-sampling and the bootstrap have made
    possible estimation, evaluation, and testing of complex models. In
    addition, other inferential philosophies, such as Bayesian and
    decision-theoretic approaches, can be useful. Recommended references
    for further reading on these and related subjects are Sachs (1986),
    WHO (1986), Gilbert (1987), Glantz (1987), Kleinbaum et al. (1988) and
    Ott (1995).

    4.7  Summary

         Statistical methods are a critical tool in applied and
    research-oriented exposure assessment studies. It is recommended that
    a statistician be involved in all aspects of an exposure
    investigation, especially during the design and data analysis stages.
    Sample size determination is an important use of statistics during the
    planning of an exposure assessment study. Numerical and graphical
    descriptive statistics can be used to summarize exposure data and
    perform preliminary analyses of relationships between and among
    exposure variables. In many cases, exposure data are approximately
    normally or lognormally distributed and can thus be readily
    incorporated into standard parametric statistical inference methods
    such as estimation and hypothesis testing. In addition, other
    parametric statistical models such as ANOVA, linear regression and
    logistic regression can be used to quantify associations among
    exposure measures. In situations where the number of observations is
    small or the data cannot be transformed to an approximately normal
    distribution, non-parametric methods such as the sign, Mann-Whitney
    and Kruskal-Wallis tests can be used to test hypotheses.

    5.  HUMAN TIME-USE PATTERNS AND EXPOSURE ASSESSMENT

    5.1  Introduction

         Methods for the collection and application of time-use data in
    exposure studies are critically reviewed in this chapter. All methods
    have their limitations. With appropriate quality assurance, however,
    information on time use and activity patterns collected by
    questionnaire, diary, interview, observation or technical means can be
    very valuable for interpreting and modelling exposures. Although the
    methodologies of time-activity data collection are universal, they
    need to be applied and their vocabularies selected keeping in mind the
    population and culture of concern and objectives of the study.
    Accurately and reliably documenting the time-activity patterns of the
    general and target populations are important components of
    understanding and mitigating human exposure (see Table 15).

         People's activity patterns, eating and drinking habits, and
    lifestyle characteristics must be superimposed over concentrations in
    environmental media before it is possible to derive realistic
    estimates of actual human exposure. Too often in the past, pollutant
    concentrations in a particular medium have been assumed to represent
    exposure, only for it to be found later that they did not provide an
    accurate picture owing to modifying factors such as the time people
    spend indoors rather than outdoors, food preparation and cooking, and
    use of bottled water instead of tap water. Experience has shown that
    exclusive reliance on central monitoring sites (e.g., urban air
    pollution monitoring sites, samples from drinking-water reservoirs)
    and bulk sampling procedures (e.g., spot checks for pesticides in
    food) for determining human exposures may be insufficient in many
    cases.

         In an exposure context, data about human time use and activity
    patterns (often referred to as time-activity data) have four related
    purposes.

    1.   Knowledge of the activities performed while a study participant
    carried a personal monitor can aid in identifying the determinants of
    exposure, i.e., "What did this person do that led her/him to have such
    a high exposure?" and "To what extent can exposure be explained the
    amount of time spent in specific activities or locations?" For
    instance, several studies in which activity pattern data were
    collected in conjunction with monitoring data have shown that
    indicators such as commuter status, work status, cooking fuel type,
    season, residential location and day of week are important in
    differentiating exposure to carbon monoxide and nitrogen dioxide
    (Akland et al., 1985; Ryan et al., 1990; Schwab et al., 1990; Berglund
    et al., 1994a). Investigations of VOC exposure have found that people
    who reported engaging in auto-related activities (e.g., exposure to
    vehicle exhausts, pumping gasoline and visiting a service station)
    were associated with statistically significant increases in breath and
    personal exposure levels of several aromatic and aliphatic compounds;


        Table 15.  Features of time-activity studies aimed at exposure assessment

                                                                                                                                              
    Location        Pollutant      Participant characteristics   Survey characteristics        Spatial and source           Reference
                                                                                               characteristics
                                                                                                                                              

    Cincinnati,     No pollutant   487 people under age 70;      March and August 1985;        28 microenvironments;        Johnson, 1989
    Ohio, USA                      representative; includes      diary; minute resolution;     location data; breathing 
                                   children; oversample          3-day sample; time of         rate; smoking status; 
                                   asthmatics; data on age,      year; day of week; time       pollutant-related activity 
                                   gender, race, income,         of day                        questionnaire
                                   work status, health status

    California,     No pollutant   1780 people over age 11;      October 1987-July 1988; 24-h  50 microenvironments;        Wiley et al., 
    USA                            representative of             recall and questionnaire;     stressed activities with     1991; Jenkins et 
                                   English-speaking households;  time of year; day of week;    respect to toxics exposure   al., 1992
                                   stratified by region; data    time of day                   and high breathing rates; 
                                   on demographics and                                         location/ region; housing 
                                   socio-economic status                                       unit characteristics

    California,     No pollutant   1200 children under age       April 1989-March 1990; 24-h   113 activities; 63           Wiley et al., 
    USA                            12; representative of         recall and questionnaire;     locations; proximity to      1991
                                   English-speaking households;  time of year; day of week;    sources; location/region; 
                                   stratified by region; data    time of day                   housing unit characteristics
                                   on demographic and 
                                   socio-economic status

    Kanawha         No pollutant   90 children aged 9-11;        July and September 1989;      Home/near home/far;          Schwab et al., 
    Valley, West                   longitudinal (4 weeks);       diaries; 30-min resolution    school; indoor vs. outdoor;  1991, 1992
    Virginia, USA                  stratified by gender,         except travel (15 min);       exertion level; housing unit
                                   respiratory health; data on   time of year; day of week;    characteristics
                                   demographic, socio-economic   time of day
                                   status and health variables
                                                                                                                                              

    Table 15.  (continued)

                                                                                                                                              
    Location        Pollutant      Participant characteristics   Survey characteristics        Spatial and source           Reference
                                                                                               characteristics
                                                                                                                                              

    New York,       No pollutant   1000 children aged 5-12;      Mid-1990 to mid-1991;         Usual commuting; frequency   Silvers et al., 
    New Jersey;                    stratified by state,          24-h recall of child's        of bathing, hand washing;    1994
    Pennsylvania,                  weekday/weekend and season;   activities by adult           weather conditions; clothing 
    Oregon;                        data on demographic,          caregiver; 30-min resolution; type; play surface; dwelling 
    Washington;                    socio-economic status and     questionnaire; time of        type
    California,                    community type                year; day of week; time 
    USA                                                          of day

    Berkeley,       Ozone          168 college freshmen (aged    Test-retest reliability       Time spent outdoors, time    Künzli et al., 
    California,                    17-21) raised in California;  study to recall lifetime      spent in physical activity   1997a,b
    USA                            convenience sample;           residential history           (outdoors)
                                   long-term ozone exposure

    Athens;         PM25, CO,      450 adults (aged 25-55)       Mostly 1997; 48-h personal,   Time spent in                Jantunen et al., 
    Basel;          VOC, NO2       personal air sampling;        indoor, outdoor and at work   microenvironments (e.g.,     1998
    Grenoble;                      approximately 1200 adults     monitoring; time-activity     Indoors, outdoors, at 
    Helsinki;                      with time-activity diary;     diary (Fig. 19); specific     work); traffic categories
    Milano;                        random population sample;     tasks
    Praha                          demographic and 
                                   socio-economic status

    Washington,     CO             700 adults aged 18-65;        1982-1983 (winter); diary     8 locations; transport       Hartwell et al., 
    USA                            representative; oversample    and questionnaire; minute     mode use; activity index;    1984; Akland et 
                                   long commutes and gas         resolution; 1-day sample;     smokers present; range use;  al., 1985
                                   ranges; excluded smokers;     time of year; day of week     in gavage; census tracts 
                                   data on age, gender, work     and day                       for work, home, other; 
                                   status                                                      housing unit characteristics

                                                                                                                                              

    Table 15.  (continued)

                                                                                                                                              
    Location        Pollutant      Participant characteristics   Survey characteristics        Spatial and source           Reference
                                                                                               characteristics
                                                                                                                                              

    Denver,         CO             452 adults aged 18-65;        1982-1983 (winter); diary;    8 locations; transport       Johnson, 1984
    Colorado,                      representative; oversample    questionnaire; minute         mode; activity index; 
    USA                            gas ranges and long commutes; resolution; 2-day sample;     smokers present; range use; 
                                   excluded smokers; data on     time of year; day of week;    in gavage; census tracts 
                                   age, gender, work status      time of day                   for work, home, other; 
                                                                                               housing unit characteristics

    Elizabeth/      VOCs           355 people; representative;   Fall 1981; follow-up: 157     Activities > 1 h; high       Wallace et al., 
    Bayonne,                       oversampled high-exposure     in summer 1982; follow-up:    exposure activities (e.g.,   1985, 1986
    New Jersey,                    occupations; data on age,     49 in early 1983; 24-h        smokers, occupations, 
    USA                            gender, race, socio-economic  recall diary; activity        travel); proximity to 
                                   status, and proximity to      questionnaire                 industry; housing unit 
                                   VOC sources                                                 characteristics

    Portage,        NO2, RSP       120 children; selected from   1987; retrospective, actual,  11 microenvironments; home   Adair & Spengler, 
    Wisconsin                      larger (600) cohort;          and prospective diary;        zip codes; school location;  1989a,b
                                   stratified by cooking fuel;   10-15-min resolution; time    housing unit characteristics
                                   data on gender, age,          of year; day of week; time 
                                   parental education            of day

    Steubenville,   NO2, RSP       150 winter, 250 summer;       1987; retrospective actual    11 microenvironments; home   Adair & Spengler 
    Ohio                           selected from cohort of 600   and prospective diary;        zip codes; school location;  1989a,b
                                   children; stratified by       10-15-min resolution; time    housing unit characteristics
                                   cooking fuel; data on gender, of year; day of week; 
                                   age, parental education       time of day.

    Topeka,         NO2, RSP       300 winter, 300 summer;       1988; retrospective, actual   11 microenvironment; home    Adair & Spengler, 
    Kansas,                        selected from cohort of       and prospective               zip codes; school location;  1989a,b
    USA                            600 children; stratified by   questionnaires; 10-15-min     housing unit characteristics
                                   cooking fuel; data on gender, resolution; time of year; 
                                   age, parental education       day of week; time of day

                                                                                                                                              

    Table 15.  (continued)

                                                                                                                                              
    Location        Pollutant      Participant characteristics   Survey characteristics        Spatial and source           Reference
                                                                                               characteristics
                                                                                                                                              

    California,     VOCs           188 people; representative;   February-March 1984;          Activities >1 h;             Wallace et. al., 
    USA                            oversampled high-exposure     follow-up: 52 in May-June     high-exposure activities     1988, 1991a,b
                                   occupation; data on age,      1984 and 51 in Feb and        (e.g., smokers, occupations, 
                                   gender, race, socio-economic  March 1987; 24-h recall       travel); proximity to 
                                   status and proximity to VOC   diary; activity               industry; housing unit 
                                   sources                       questionnaire                 characteristics

    Boston,         NO2            325 (winter), 298 (summer)    1986; diary and               6 microenvironments; range   Ryan et al., 
    Massachusetts,                 ages 8 and above;             questionnaire;                on; near roads; combustion;  1990
    USA                            representative; stratified    15-30-min resolution; 2-day   outside home; home location; 
                                   by range type; no personal    sample; time of year; day     housing unit characteristics
                                   data                          of week; time of day

    Los Angeles     NO2            620 people ages 8 and above   May 1987-May 1988; diary,     17 microenvironments:        Spengler et al., 
    and Orange                     sampled two 24-h periods; 65  questionnaire; 15-min         including near roads; home   1994; Schwab et 
    Counties;                      sampled eight cycles;         resolution; two-day sample;   zip codes; work zip codes;   al., 1990
                                   representative; data on age,  time of year; day of week;    climate region; housing 
                                   gender, work status           time of day                   unit characteristics

    Albuquerque,    NO2            1000+ infants; stratified     January 1988-December 1991;   Room in house; outside       Samet et al., 
    New Mexico,                    by range type; data on        every 2 months for the        of house (including travel); 1992
    USA                            child's health and parents'   first 18 months of life;      range use; housing unit 
                                   socio-economic and            60 min; time of year;         characteristics
                                   demographic characteristics   day of week; time of day
                                                                                                                                              
    

    reporting a smoker present in the home was associated with increased
    indoor concentrations and personal exposures of aromatic compounds;
    visiting dry cleaners, self-reports of proximity to smokers, pesticide
    use, exposure to solvent, degreasing compounds, and odorous chemicals,
    and employment status in certain occupations (e.g., paint, chemical or
    plastics plants) were associated with increased personal exposure to
    several VOCs (Wallace et al., 1985, 1986, 1988). Occupational exposure
    may be an important component of total exposure for some individuals
    or sub-populations.

    2.   Time-activity data allow modelling of human exposure to
    pollutants for which personal monitors are not yet available or are
    very expensive, or for which exposure is a function of multiple
    pathways. Total exposure can be simulated from information on the time
    spent doing various activities and/or in specific locations, coupled
    with knowledge about the likely range of pollutant concentrations in
    each situation. For example, the models SHAPE (Ott et al., 1988), NEM
    (Johnson et al., 1990), SIMSYS (Sexton & Ryan, 1988), and REHEX (Hall
    et al., 1992) are currently being used to estimate exposure to carbon
    monoxide, ozone, particulates, sulfur dioxide and nitrogen dioxide.
    Techniques are also being developed to allow prediction of dermal and
    ingestion exposures based on assumptions about human activity patterns
    (e.g., Fenske, 1993). The usefulness of all of these models is
    dependent upon the accurate characterization of pollutant-relevant
    time-activity patterns.

    3.   From an epidemiological perspective, activity patterns can be
    used to assess the relationship between exposure and health status
    (e.g., Armstrong, 1985). For instance, "Do those who engage in
    potentially high-exposure activities experience more frequent or
    severe illnesses?" or "Do sensitive individuals avoid potentially
    high-exposure activities or limit them to certain times of day or
    locations?" In epidemiology, time-activity data may serve four
    purposes:

    *  They may be a surrogate of the exposure of interest. For example,
       people may be asked about the hours they spend indoors with smokers
       to assess health effects of environmental tobacco smoke
       (Leuenberger et al., 1994).

    *  They may be used to improve another imperfect measure of exposure.
       For example, estimates of long-term exposure to ozone may be
       derived from fixed site monitor data, weighted, however, for
       duration of time spent in outdoor activities (Künzli et al.,
       1997a,b).

    *  They may be used as a surrogate for a cofactor which might confound
       the association between health and some other exposure. For
       example, the effect of ambient air pollution on lung function may
       be thought to be confounded by environmental tobacco smoke exposure
       (ETS). Time spent with smokers could thus be used to control this
       potential confounding.

    *  The association of an exposure with some health outcome might not
       be the same in subgroups of different time-activity patterns
       (modified effect). In this case, time-activity data will allow the
       investigator to address such interactions.

    4.   Another purpose of time-activity data is to describe patterns of
    population behaviour. The proportion of time spent by the population
    in various microenvironments or frequency of use of various facilities
    (e.g., swimming pools) may provide an indication for the contribution
    of each of the microenvironments or activities on total population
    exposure. In such studies, the emphasis is on characteristics of
    groups, and not on individual data. Therefore the precision of the
    estimates may be improved by the increased sample size although the
    survey tools may remain relatively simple and inexpensive.

         An understanding of the frequency and duration of the activities
    in which the target population engages can be used to set priorities
    for public health strategies designed to reduce exposure by limiting
    contact with contaminated media. Comprehensive exposure factor data
    for the US population may be found in AIHC (1994) and US EPA (1996a).
    Although this information is focused on the USA it may serve as a
    useful model to follow in other countries.

    5.2  Methods

    5.2.1  Activity pattern concepts

         Activity pattern data that may be useful in assessing exposure
    can be divided into three categories:

    *  the distribution of time among activities, referred to in this
       document as  time allocation parameters 

    *  the factors that influence the degree of media contamination in the
       activities or locations of interest, referred to in this document
       as  microenvironmental parameters 

    *  the  intensity of contact while engaging in each activity.

    5.2.1.1  Time allocation parameters

         Time allocation parameters include the amount of time spent in a
    given activity, the time of day, week and year of contact, and the
    expected frequency with which the person or population engages in the
    activity. The relevant spatial resolution for describing time-use
    patterns, thus grouping activities for exposure assessment, depends
    upon the characteristics of the pollutant, the media, the location and
    the emission source(s).

         The concept of microenvironment has been used to define an area
    across which the concentration of an air pollutant is assumed to be
    homogeneous (Duan, 1982). The most basic division of microenvironments
    is whether a person is indoors or outdoors, although more refinement
    is necessary for many exposure assessments. Time spent indoors is
    especially important with regard to pollutants which depend on indoor
    sources. Other typical microenvironments of interest in studying air
    pollution are home, work or school, and modes of transportation.

         Depending on the characteristics of the media and the pollutant,
    a description of the actual activity may also be required to
    understand exposure. General activity categories such as "socializing"
    and "recreation" are less important than knowing whether the
    participant is involved in specific activities that lead to contact
    with environmental media in addition to or other than air. For
    instance, swimming leads to water contact, and farming and gardening
    lead to soil contact.

    5.2.1.2  Microenvironment parameters

         The distinction between people's activities and the pollutant
    concentration in a microenvironment is not always clear. The use of
    household appliances and consumer products that emit environmental
    contaminants and/or influence pollutant fate and transport affect
    microenvironmental concentrations. Thus, information on the
    microenvironmental parameters, i.e., the factors affecting the
    concentration in a given location, have also been included under the
    rubric of time-activity data. Important microenvironmental parameters
    for air pollution exposure assessment include building structure and
    household characteristics (e.g., the type of heating and cooking fuel
    used, the presence of parking garages and air conditioning units),
    information on proximity to specific sources (e.g., heavy traffic,
    cigarette smoking, cooking, solvent, pesticides), timing of emissions
    for each source, indoor/outdoor air exchange rates and meteorological
    and topographic factors.

    5.2.1.3  Intensity of contact

         In addition to time allocation measures and microenvironmental
    parameters, information on the intensity of contact is needed to
    assess exposure. Here the focus is on micro-level activities that
    affect the rate of contact with the contaminated media while the
    person is in a certain microenvironment (e.g., outdoors at home) and
    performing a specific activity (e.g., cleaning). The potential for
    dermal contact depends upon the surface area of exposed skin, thus
    clothing type and fabric consistency as well as the size of the
    person, whether the individual is sitting, crawling, kneeling or using
    their hands on the contaminated surface, or otherwise handling the
    contaminant. In addition, exposure for the given event depends upon
    the duration and frequency of each contact between the exposed skin
    and the contaminated media; e.g., 50 1-min contacts between the
    person's hand and the floor while cleaning. As described in Chapter 7,
    dietary factors, including the type of foods that are consumed and the

    amount consumed per time period of interest, are the most obvious.
    Concern also has been raised about the potential for contamination of
    foods from contact with surfaces during storage, preparation and
    consumption (Berry, 1992, Freeman et al., 1997). Hand-mouth and
    object-mouth contact, although difficult to measure, may be one of the
    most important routes of exposure to contaminants such as pesticides
    and lead that reside in house dust, especially in children (Charney et
    al., 1980; Rabinowitz & Bellinger, 1988; Davies et al., 1990). For
    pollutants for which inhalation is the primary route of exposure, the
    intensity of contact is influenced by one's level of exertion (often
    referred to as "activity level"). Breathing rate or heart rate is
    needed to predict dose (the amount of contaminant that enters the
    body), thereby producing a more accurate estimate of the resulting
    health effects.

         Finally, depending on the purpose of the exposure assessment, the
    required temporal resolution of the time-activity data may vary
    substantially. Whereas short-term time-activity patterns may be
    important for acute exposures, long-term average time-activity
    patterns may be more relevant in other circumstances. If long-term
    exposure is of major interest, e.g., over years or lifetime,
    residential history is an important information to assign respective
    ambient monitor data for the entire period of interest (Künzli et al.,
    1996).

    5.2.2  Surrogates of time-activity patterns

         For many exposures surrogates of time-activity patterns may be
    developed on the basis of generalizations about the activities of
    people at a particular time, who live in a specific geographic
    location or who share a specific set of living conditions. Usually the
    most important time-activity surrogate is age group. Some activities
    that are useful for predicting exposure to air pollutants, such as
    distance and timing of travel or duration of work and its locations,
    also show systematic differences in their frequency and duration by
    demographic characteristics. For instance, Schwab et al. (1990)
    documents how time in the kitchen, which influences exposure to
    combustion products, is greater among women in the USA than among men,
    even after adjusting for whether the woman works outside the home;
    likewise, men spend more time in transit, regardless of their age or
    employment status. It is likely that the frequency of contact with a
    wide variety of toxins differs across groups defined by gender and
    age, owing to traditional divisions of labour in many cultures.

         Similarly, information about an individual's health condition may
    be important in characterizing their time-activity pattern. For
    instance, the limited data available on asthmatics suggests they may
    spend more time indoors than the general population (Goldstein et al.,
    1986, 1988; Lichtenstein et al., 1989; Schwab et al., 1991). As
    asthmatics are particularly sensitive to air pollutants, this activity
    information is important.

         Socioeconomic status may influence time-activity patterns related
    to, for instance, time spent travelling to work or outdoors.
    Currently, however, the gap existing in time-activity databases with
    respect to the activity patterns of sensitive (e.g., elderly) and
    potentially high-risk (e.g., low socioeconomic status) subgroups, is a
    limitation for extension of exposure models to these groups. Further
    study is needed to determine the extent to which income, education and
    occupation are reliable surrogates for exposure-related factors (e.g.,
    housing unit size and condition).

    5.2.3  Data collection methods

         Sociologists pioneered studies of activity patterns (Szalai,
    1972; Chapin, 1974; Robinson, 1977). These "time budget"
    investigations, which have been conducted in several nations,
    emphasize the purpose of activities (cooking, eating, TV watching).
    Ott (1989) summarizes such studies in relation to their usefulness to
    exposure assessment; a basic drawback for exposure assessment
    applications is the lack of information on location, particularly
    distinguishing whether the participant was indoors or outdoors. In the
    1960s and 1970s, a series of time-activity studies was conducted by
    geographers interested in the influence of the economic and physical
    structure of cities on travel patterns, e.g., journey to work (Hanson
    & Hanson, 1981), access to facilities (Fox, 1983) or shopping
    behaviour (Douglas, 1973). As such, the emphasis was on collecting
    information on the geographic location of trip origins and
    destinations as well as timing and mode use. Finally, the US
    Department of Transportation, in conjunction with the Census Bureau,
    has been collecting information on the travel activities (durations
    and mode use) of a representative national sample approximately every
    7 years since 1969 (US Federal Highway Administration, 1986, 1992).

         A variety of methods are available for collecting data about
    time-activity patterns, including interviewer-administered recall
    questionnaires, self-administered real-time diaries, direct
    observation and video recording. The diary techniques used in the
    social sciences for eliciting time-activity data have been applied to
    studies of total human exposure to air pollutants (see methodological
    reviews by Robinson (1988), Ott (1989), Quackenboss & Lebowitz
    (1989)). Specifically, participants are asked to complete a diary or
    questionnaire regarding their activities during the designated period
    (usually 12-48 h). The survey instruments used in these exposure
    studies, however, depart from any single type used previously. Rather
    than focusing on activity purposes or transportation exclusively, the
    instruments used in exposure studies probe for changes in location or
    activity that might lead to changes in the level of pollution to which
    the person came into contact.

         Time allocation measures for assessing exposure to air pollutants
    frequently have been collected using self-completed real-time diaries.
    Because this approach requests that participants record all activities
    over one or more 12-h or 24-h periods, it has the potential to provide
    the most comprehensive information on time allocation, sequencing, and

    frequency. Real-time diaries are particularly useful when it is
    important to know the time of day during which each activity was
    performed (e.g., the amount and location of exercise in the morning
    versus the afternoon when ozone levels tend to be higher).

         Two diary formats are common for collecting time-activity data:
    the open-ended style requires participants to describe their exact
    activity (see, for example, the instruments described in Akland et al.
    (1985), Johnson (1989), and Jenkins et al. (1992)), whereas the
    close-ended format (Fig. 18) involves simply checking the appropriate
    microenvironment for the given time of day (see, for example, the
    instruments used in EXPOLIS (Fig. 18) (Jantunen et al., 1998) or those
    described by Schwab et al. (1990) and Samet et al. (1992). Several
    researchers are developing electronic monitors to supplement diaries
    by detecting whether a participant is indoors or outdoors, a key
    parameter for assessing exposure to air pollutants (e.g., Hinton,
    1990; Moschandreas & Relwani, 1991; Waldman et al., 1991b).

    FIGURE 18

         Interviewer-administered questionnaires that ask participants to
    recall frequency and duration of time spent in specific activities
    during either the previous or typical day, month, year or age-period
    (i.e., usual activity patterns) also have been used to collect time
    allocation measures, microenvironmental parameters and exposure
    surrogates. Juster et al. (1985a) points out that data collected in
    this fashion are most accurate when the survey focuses on activities
    that are done frequently or on a routine basis (e.g., the daily
    commute to work). Questionnaires that take the form of checklists are
    also particularly useful when the researcher is only interested in
    certain well-defined activities. Questions to recall activity patterns
    over a long period may refer to defined age groups and/or to each
    residential location lived in (see Fig. 19). In environmental exposure
    studies, information on the proximity of the study participant to
    local contaminant sources is typically collected via questionnaires
    that ask whether or not the participant engaged in a certain activity.
    For instance, studies of VOC exposure have asked about use of
    household cleaners, visits to petrol stations and storage of gasoline
    products indoors (Wallace et al., 1987a,b). Questionnaires are also
    used to solicit information on housing unit characteristics (e.g.,
    type of cooking equipment or house volume) that influence
    concentrations indoors (Lebowitz et al., 1989). Surveys may request
    information on a variety of parameters that affect the concentration
    of combustion products to which an individual is exposed during
    travel, including traffic speed, time of day, mode of transportation,
    age of vehicle, trip timing and roadway used.

         Researchers have experimented with a variety of methods for
    collecting information on the intensity of contact. As described in
    Chapter 7, a number of approaches are used for quantifying food
    consumption rates. An inexpensive technique that has been used to link
    breathing and activity patterns is to have participants record the
    level of activity (e.g., high, medium, low) associated with each
    activity entry in the diary. This method has been used in several
    population-based studies (Johnson, 1989; Lichtenstein et al., 1989;
    Schwab et al., 1990; Wiley et al., 1991). Others have used
    questionnaires that request information about specific high-exertion
    activities such as exercising and working outdoors (Goldstein et al.,
    1986; Lebowitz et al., 1989). Categorical exertion-level data is not
    useful for calibrating activity pattern data, however, without an
    increased understanding of (1) the range of reported activities
    associated with each exertion level and (2) the range of breathing
    rates associated with each exertion level. A compendium on energy
    expenditure, which closely relates to ventilation rate, has been
    published for a variety of physical activities (Ainsworth et al.,
    1993). These data may be used to categorize activity data depending on
    levels of exertion (Künzli et al., 1997a,b). Data are becoming
    available through the application of electronic methods of tracking
    exertion levels; heart-rate and breathing-rate monitors have been used
    in the field studies by Raizenne & Spengler (1989), Shamoo et al.
    (1991) and Terblanche et al. (1991).

    FIGURE 19

         Standardized methods are not available for collecting information
    on hand-mouth contact. Several researchers (Charney et al., 1980;
    Brunekreef et al., 1983; Bellinger et al., 1986) have administered
    questionnaires to parents of toddler-age participants in order to
    qualitatively characterize the frequency with which children suck
    their fingers (i.e., usually, sometimes, never). Direct observation
    may be better suited to capturing micro-level activities, but such
    approaches have rarely been used in large-scale field studies owing to
    the expense of following more than a few participants and because of
    concerns that the observation process will lead to bias or alterations
    in typical patterns. Video techniques have now made it possible to
    record participant activities with less interference. Davies et al.
    (1990), for instance, used video methods to obtain data on the number
    of times 2 year olds put their hands and objects in their mouth while
    in standardized play situations. Zartarian et al. (1995) used
    videotape data to collect micro-level data on four young farm children
    at play inside their homes to quantify dermal and ingestion exposure
    to pesticides. As Zartarian et al. point out, however, researcher
    presence may still have influenced the participants' behaviour.
    Observation of children's hand-mouth contact also has been performed
    in clinical settings (e.g., Madden et al., 1980). All of these
    methods, however, share the limitation that they cannot quantify the
    full variability in factors that influence hand-mouth contact. Indeed,
    capturing this variability may not even be possible, as is discussed
    in a subsequent section of this chapter. In the absence of information
    on hand-mouth contact, several researchers have measured mineral
    levels in children's faeces to estimate typical soil ingestion rates
    (Binder et al., 1986; Calabrese et al., 1989, 1990). Such estimated
    ingestion rates can then be used to model exposure in areas with
    measured soil contamination levels.

         For dermal exposure, questionnaires are most appropriate for
    collecting categorical-type information, such as whether a person
    performed a certain activity during a designated activity. The US
    Environmental Protection Agency 1992 report entitled  Dermal 
     Exposure: Applications and Principles, reviews the literature
    regarding methods for estimating soil and water contact (US EPA,
    1992b). Hawley (1985) has used data from previous studies and
    professional judgement to develop assumptions for use in estimating
    outdoor soil contact time, but these estimates do not account for
    indoor exposure such as soil tracked into the house or for exposure to
    contaminants that reside primarily in indoor dust (e.g., pesticides)
    (US EPA, 1992b). The US EPA report cites Tarshis (1981) and James &
    Knuiman (1987) as sources of data on the frequency with which people
    shower and bathe. Few data are available on swimming (US EPA, 1992b)
    which could be important because of elevated chloroform concentrations
    found within air just above the pool-water surface, or other
    contaminants which can be swallowed or dermally absorbed from lakes or
    river waters.

         Linking activities with measurements of dermal exposure,
    researchers are testing innovative approaches to assessing skin
    contact with contaminated surfaces. For instance, Fenske et al.
    (1986a,b) applied non-toxic fluorescent tracers to lawns in lieu of
    insecticides; after participants engaged in a standard set of play
    activities, video imaging was used to ascertain the amount of tracer
    on the exposed skin. The degree of soil adherence to skin is a closely
    related issue and has been examined by several researchers (Driver et
    al., 1989; Finley et al., 1994a; Kissel et al., 1996).

    5.3  Potential limitations

         Time-activity data can enhance an understanding of sources and
    behaviours important in assessing exposures. Inferences can be drawn
    from simulations, case studies or even studies using large randomized
    designs. However, all users of time-activity data should be aware of
    its limitations for assessing human exposure to environmental
    contaminants.

         The feasibility of collecting time-activity data is often limited
    by the burden which such studies impose on participating individuals.
    The data collection requires constant, or regular, attention to the
    fact that the subjects are participating in the study, that they have
    to remember about all activities and to fill in the diaries. This is
    often inconvenient and takes respondent's time. Collection of the data
    by an observer, which often is a method of choice in studies involving
    children, may be of limited feasibility owing to the restricted access
    of the observer to the subject under study and because typical
    activities may possibly be modified by the fact of being under
    observation.

    5.3.1  Activity representativeness

         One of the uses of time-activity data is to allow
    characterization of the distribution of exposure for a given
    geographic, demographic or socioeconomic segment of the population.
    However, the study protocol may call for certain types of days or
    individuals to be excluded (e.g., travel that takes the participant
    away from the home for more than the 24-h or 48-h sampling period may
    lead to disqualification). Although standard techniques such as
    weighting and imputation can be used to treat non-response, these
    methods assume that refusal to participate is random and there is
    information about the non-respondents (Kalton & Kasprzyk, 1986). In
    the case of time-activity studies, however, once contacted, people may
    participate or not because of the variables that the study is designed
    to predict. As shown in the European multi-city study EXPOLIS, the
    subjects in Basel ready to participate had lower traffic density
    around their homes than non-participants (Oglesby, 1998). The
    potential for misrepresenting the exposure distribution must,
    therefore, be considered because there is no method for quantifying
    the direction and/or extent of the bias with respect to high-exposure
    behaviours.

         The representativeness of the activity data collected may also be
    influenced by the increased burden imposed upon participants by
    exposure assessment studies. Epidemiologists and social scientists
    have found that participation rates and compliance with instructions
    may decline with increasing study periods, longer questionnaires, more
    complicated questions and more complex tasks. Whitmore (1988)
    speculated that the higher than average refusal rates experienced in
    air pollution exposure studies are related to the burden associated
    with carrying monitors and completing activity diaries. This has been
    shown in the European multi-city EXPOLIS study in Grenoble where
    participants had different time activity patterns in days with
    personal exposure monitors compared to days when only time-activity
    data was collected (Boudet et al., 1997).

    5.3.2  Validity and reliability

         Survey researchers in a number of fields have raised questions
    about the validity of data collected via self-administered surveys:
    i.e., is the instrument measuring what is intended (Laporte et al.,
    1985). Data validity is of particular importance when trying to link
    measured exposure with a given day's activity diary. The error
    introduced by an inaccurate diary affects both efforts to explain the
    contribution of certain activities to personal exposure and efforts to
    estimate the distribution of personal exposure from time-weighted
    microenvironmental measurements. The relationship between the degree
    of error in the diary and the degree of error in the predictive model
    depends upon the concentration in the microenvironment and the total
    time spent there. Neglecting to report even short-duration activities
    in high-concentration microenvironments will have more effect than
    underestimating a similar amount of time in a low-concentration
    microenvironment in which a large portion of the day is spent.

         Scientists who use activity pattern data have raised a variety of
    concerns about the effects of inadvertent and/or deliberate errors in
    reporting. For instance, activity diary data may be compromised by
    participants' misunderstanding of the definitions of various locations
    (microenvironments). Discussions with participants have revealed the
    potential for confusion about: How far is "far from home?" Is a
    "parking garage" inside or outside? Is "walking" a light- or
    medium-exertion activity? (Schwab et al., 1991, 1992).

         To a certain extent, the quality of the data can be controlled
    during data collection. Detailed instructions can improve participant
    compliance. Field and laboratory pretesting of the survey instrument
    and instructions, important components of the survey design process,
    can yield improvements in protocol and clearer definitions of survey
    terminology such as distinctions between microenvironmental categories
    (Bercini, 1992). Extensive training of participants in keeping the
    diary can be expensive, but detailed reference sheets and one-on-one
    sessions can greatly improve data quality. One of the more
    time-consuming but necessary steps is reviewing the returned diaries
    for temporal completeness and clarity of responses. Ideally, this
    should be done in the presence of the participant, and within 24 h of

    completion of the monitoring period. Another quality assurance step
    involves the use of a uniform system to code information on individual
    activities into microenvironmental categories.

         The validity and reliability of the diary data may be increased
    by the use of study forms that are simple and easy to understand. The
    language of the questions and instructions must be simple and the
    method of selection of answers, or of filling in data, obvious to
    minimize coding errors. The number of items on the questionnaire
    should be kept to a necessary minimum. Only the information for which
    there is clear use in analysis and data interpretation and which
    serves directly the study objectives should be included in the diary
    form.

         Verifying the validity of time-activity data is extremely
    difficult, if not impossible, because an absolute standard does not
    exist. Several researchers have sought to assess the reliability of
    self-reported data through test-retest procedures and by comparing
    different methods of collecting the same type of information (Laporte
    et al., 1985). The University of California at Berkeley ozone study
    required college students to recall time spent in physical activities
    outdoors, over years. The information was used as a surrogate to
    improve long-term ozone exposure assignment in an epidemiological
    study (Künzli et al., 1997b). The test-retest study revealed rather
    high correlations for time spent in heavy ( r = 0.81) or moderate
    ( r = 0.61) activity (Künzli et al., 1997b). This level of
    concordance is similar to that observed in dietary intake validation
    studies where food-frequency questionnaires and diet records
    correlated in the order of  r = 0.6 for the intake of a variety of
    nutrients (Rimm et al., 1992). Robinson (1985) found that a variety of
    methods for collecting time-activity data, including 24-h recall
    surveys, same-day diaries, records of the activities during 40
    randomly selected moments throughout the day (signalled using a
    beeper), and recall of the activities during a randomly chosen hour
    yielded essentially similar sample distributions of time the sample
    spent in a variety of activities. Quackenboss et al. (1986) also found
    consistency between diaries and the responses to self-administered
    recall questionnaires. Juster (1985b) found reasonably strong
    agreement in the reports of spouses regarding whether their partner
    was present at any given time throughout the day. Other comparisons of
    methods show that when asked about the usual time spent in selected
    activities, respondents tend to over-report time in unscheduled
    activities (relative to that recorded on their diaries), but reports
    are consistent for habitual activities such as commuting to work
    (Robinson, 1985). Waldman et al. (1991b) showed similar results when
    comparing activities recorded in electronic diaries with next-day
    recall; concordance between the methods was highest for routine,
    long-duration activities. Additional research, however, is necessary
    to determine the extent and direction of bias for the activities and
    the time frames of most concern in an exposure context (e.g., the
    frequency with which a person uses household cleaning products rather
    than the total time spent cleaning).

    5.3.3  Inter- and intra-person variability

         To be of use in exposure assessment, time-activity data must
    describe the aspects of human behaviour that influence the variability
    in pollutant concentrations contacted. There is both between- and
    within-individual variability in people's activities, which has
    implications for the use of time-activity data in exposure assessment.

         At one end of the spectrum are aspects of human activity patterns
    that tend to be highly regular. For instance, many people tend to
    follow daily routines with respect to how long they sleep and the time
    they depart for work. In addition, because basic routines are fairly
    uniform across individuals, diary data from several studies has shown
    that the distribution of time reported in the microenvironments that
    comprise the majority of the day (i.e., inside at home and inside at
    work/school) exhibit relatively little variation from year to year
    within a given study population or from place to place within the USA
    (Robinson, 1985; Schwab et al., 1990).

         The only large time-activity study done in conjunction with a
    continuous monitoring device was the Denver/Washington, DC study of CO
    exposures (Akland et al., 1985); this study yielded time-weighted
    concentrations in specified microenvironments. Analyses of these
    results suggest that variations in activities or locational attributes
    (e.g., variations in source strength) that are finer than those
    captured by these simple microenvironments explain much of the
    variability in exposure. Although less variability in the
    concentrations of some other air pollutants may be expected, these
    results confirm the concerns raised above regarding the ability to
    predict variations in exposure from the time allocation measures
    typically collected in diary-type studies.

         At the other end of the spectrum with respect to consistency in
    activity patterns are aspects of human behaviour that influence the
    intensity of contact with contaminated media. By their nature, these
    activities are highly variable both across individuals and across time
    for a given person. First, physical and demographic characteristics
    influence the frequency and duration of activities. For instance, in
    the case of dermal exposure it may be hypothesized that contamination
    from lying on a surface (e.g., a lawn to which a weedkiller has
    recently been applied) will be greater for a heavy person than a
    lighter person. Similarly, a child's standing and sitting height, in
    addition to crawling activities, mean that its breathing zone is much
    closer to the floor than that of an adult, raising the possibility of
    dust inhalation. Children also choose play locations that typical
    monitoring studies might ignore, such as stairwells and corners.

    5.4  Summary

         Information on people's activity patterns can be used to identify
    the determinants of measured exposures, predict unmeasured or
    unmeasurable exposures, assess relationships between exposure and
    health status, and identify high risk exposure situations that may be

    amenable to public health actions. Some of the main activity patterns
    important for assessing exposures by various media that were discussed
    in this chapter are summarized in Table 16.

         The relative cost of field sampling and laboratory analysis for
    environmental and biological measurements highlights the potential
    value of time-activity data. Assessments of long-term activity
    patterns (e.g., lifetime) may only be feasible using time-activity
    questionnaires. Various methods are used to collect information about
    human activities, including diaries and questionnaires, mechanical
    devices, and observation. Methods have only recently begun to be
    developed for assessing the role of time-activity patterns on dietary
    and non-dietary ingestion and dermal exposure pathways. Concerns about
    the ability of data collection methods to ensure activity
    representativeness and data validity and about the implications of
    inter- and intra-person variability in behaviour place limits on the
    application of time-activity data for human exposure assessment.
    However, with appropriate quality assurance programmes, information on
    time use and activity patterns can be very valuable for interpreting
    and modelling exposures.


    Table 16.  Type of information obtained from time-activity data 
               relevant to specific exposure pathways

                                                                          

    Personal air
                        time and location spent outdoors
                        type of indoor location
                        use of sources
                     In the presence of sources:
                        ventilation and filtration of indoor location
    Water
                        quantity of water consumed direct and indirectly
                        accidental ingestion from swimming (pools, rivers, 
                        etc.)
                        dermal contact, time in showering/bathing
                        hand/body washing frequency

    Food
                        amount and type of food products consumed
                        preparation methods including cleaning
                        preparation location (e.g., street vendors)
                        storage practices

    Soil
                        amount of contact time and type of soil 
                        (e.g., farm, garden/possible pesticide 
                        application)
                        skin surface contact
                        frequency and duration of washing since contact
                                                                          

    6.  HUMAN EXPOSURE AND DOSE MODELLING

    6.1  Introduction

         An exposure model is a logical or empirical construct which
    allows estimation of individual or population exposure parameters from
    available input data. Such data may be measured or collected for this
    purpose, or obtained from other sources. Technological, logistic and
    financial constraints can make it difficult to monitor the exposure of
    humans to the various environmental agents. It is, therefore, prudent
    in many situations to use models to assess contaminant exposures.
    Models provide an analytic structure for combining data of different
    types collected from disparate studies in a manner that may make more
    complete use of the existing information on a particular contaminant
    than is possible from direct study methods (EC, 1997b). Exposure
    models, if supported by adequate observations, can be used to estimate
    group exposures (e.g., a population mean) or individual exposures
    (e.g., the distribution of exposures among members of a population).
    Model results also can be used to evaluate exposures at various points
    of population distributions which cannot be measured directly because
    of limitations of methods or resources (e.g., the upper 5% of
    exposures for a population). This chapter introduces the principal
    aspects of exposure modelling, including those for single and multiple
    environmental media. In addition, the concepts of variability,
    uncertainty and model validation are discussed.

    6.2  General types of exposure model

         Exposure models can be divided into three broad categories;
    statistical, deterministic and practical or combinations of
    statistical and deterministic models (Fig. 20). Statistical (often
    regression) models are in their simplest form numerical best fits
    between collected exposure measurements and potentially related
    factors (e.g., demographics). In statistical models, the magnitude and
    direction of association between the variables are inferred from the
    observations themselves. Such models cannot be considered reliable for
    predicting exposures outside the original study population and
    environmental setting without first validating them for that specific
    purpose. Deterministic (or physical) models are based on a logical
    expression of the physical environment and human behaviour in it. Such
    models need to be validated by actual exposure data, and can in
    principle be used for exposure prediction of new populations and
    settings. Although deterministic models can be useful for estimating
    mean population exposure, input data to estimate the distribution of
    exposure within a population are often not available. Probabilistic
    exposure models (section 6.6.3) are normally based on deterministic
    models, but because they incorporate the measured or estimated
    distributions of the input variables, they produce more realistic
    population exposure distributions than deterministic models. Practical
    models can combine features from these different types, e.g., a
    statistical model may include parts of a logical construct. Several
    important types of statistical models are discussed in Chapter 4, and
    deterministic and practical models are discussed here.

    FIGURE 20

         Using a deterministic model for a given contaminant, exposure
    concentration is estimated as a concentration averaged over a given
    period of time (see Eq. 3.1, p. 46).

         For the inhalation and dermal exposure routes, concentrations in
    the different microenvironments occupied by a person are integrated
    over time. The integrated time period is usually 24 h, 1 year or a
    lifetime of 70 years, although any time period may be used. The
    concept of microenvironment is often unnecessary for the ingestion
    route. In this case, the concentration of contaminants in the food
    consumed and the amount of food and beverages consumed during a given
    period of time are sufficient to determine exposure.

    6.3  Environmental media and exposure media

         In exposure analysis, we use human exposure assessments to
    translate contaminant levels in environmental media into quantitative
    estimates of the amount of contaminant that comes in contact with the
    human-environment boundaries, that is, the lungs, the gastrointestinal
    tract and the skin surface of individuals within a specified
    population. Environmental media of principal relevance to human
    exposure include air, ground-surface soil, root-zone soil, plants,
    groundwater and surface water in the contaminated landscape. As
    described in Chapter 2, exposure pathways define a link between an
    environmental medium and an exposure medium. Important exposure media
    include outdoor air, indoor air, food (commercial and homegrown),
    exterior soil, interior soil or household dust, and drinking and
    cooking water. Exposure then occurs by contact with contaminants in
    these exposure media via inhalation, ingestion and dermal uptake. Fig.
    21 illustrates the types of exposure pathways we use to carry out a
    multiple-media, multiple-route, multiple-pathway exposure assessment.

         Exposure assessments often rely implicitly on the assumption that
    exposure can be linked by simple parameters to ambient concentrations
    in air, water and soil. However, total exposure assessments that
    include time-activity patterns and microenvironmental data reveal that
    an exposure assessment is most valuable when it provides a
    comprehensive view of exposure routes and pathways and identifies
    major sources of uncertainty. Listed in Table 17 are potential
    interactions among environmental media, exposure media and exposure
    pathways that are addressed in this chapter.

         An assessment of intake requires that we determine how much
    crosses these boundaries. Thus, we see the need to address many types
    of "multiples" in the quantification of human exposure, such as
    multiple media (air, water, soil); multiple exposure pathways (or
    scenarios); multiple routes (inhalation, ingestion, dermal); multiple
    chemicals; multiple population subgroups; and multiple health
    end-points. The matter is further complicated by the fact that
    pollutants may have both systemic and route specific health effects.
    For the compounds that have mainly systemic effects the total exposure
    -- sum of all routes -- is most relevant; for other agents such as
    pneumococci aerosols in the lung, dermal vs. ingestion absorption of

    FIGURE 21


        Table 17.  Interactions among environmental media, exposure media and exposure pathways

                                                                                                                                    
    Exposure routes                                              Media
                                                                                                                                    

                       Air                                Soil                                  Water
                       (gases and particles)              (ground-surface soil,                 (surface water and groundwater)
                                                          root-zone soil)
                                                                                                                                    

    Inhalation         gases and particles in             soil vapours that migrate to          contaminants transferred from 
                       outdoor air                        indoor air                            tap water

                       gases and particles                soil particles transferred to 
                       transferred from outdoor air       indoor air
                       to indoor air

    Ingestion          fruits, vegetables, and grains     soil                                  tap water
                       contaminated by transfer of 
                       atmospheric chemicals to plant     fruits, vegetables, and grains        irrigated fruits, vegetables, and 
                       tissues                            contaminated by transfer from soil    grains

                       meat, milk, and eggs               meat, milk, and eggs contaminated     meat, milk, and eggs from animals 
                       contaminated by transfer of        by transfer from soil to plants       consuming contaminated water
                       contaminants from air to plants    to animals
                       to animals

                       meat, milk and eggs contaminated   meat, milk, and eggs contaminated     fish and sea food
                       through inhalation by animals      through soil ingestion by animals

                       mother's milk                      mother's milk                         mother's milk

    Dermal contact     (not included)                     soil                                  baths and showers
                                                                                                swimming, etc.
                                                                                                                                    
    

    solvents which are rapidly metabolized in the liver, or fine
    particulate matter in the ambient air, the route of exposure is
    crucial, and total exposure as a sum of all exposure routes may be
    meaningless. Multiple media exposure models are discussed in section
    6.5.

    6.4  Single-medium models

         Most of the transport models that have been developed for
    describing the behaviour of contaminants in the environment have dealt
    with specific environmental media, such as indoor and outdoor air,
    surface water and sediments, groundwater and soils. These
    single-medium models operate at various levels of spatial and temporal
    detail, depending on the particular conditions being assessed. The
    following discussion will highlight some of the more commonly used
    methods for characterizing contaminant transport in environmental
    media. Additional information on transport modelling for use in
    exposure assessments can be found in Masters (1991).

    6.4.1  Outdoor and indoor air

         Substances in outdoor air are transported from sources to
    receptors by atmospheric advection and dispersion. In general,
    pollutant concentrations in outdoor air are directly proportional to
    emission strength and inversely proportional to dispersion. The
    physical relationship, e.g., lateral and vertical distance, between
    sources and receptors is also an important factor. Meteorological
    parameters have an overwhelming influence on the dispersion of
    contaminants in the lower atmosphere. Among them, wind parameters
    (direction, velocity, and turbulence) and thermal properties
    (stability) are the most important. A number of models are available
    for estimation of ambient concentrations of pollutants. Most of them
    are founded on the Gaussian air dispersion model, an introduction to
    which may be found in Wilson & Spengler (1996). Two of the seminal
    works in this field are Pasquill (1961) and Gifford & Hanna (1973).

         Another area of air quality models focuses on determining the
    sources of pollutants in outdoor air. As discussed in Chapter 2,
    information on sources of exposure is important for evaluating
    alternative strategies for managing risk. These models are commonly
    used for apportioning concentrations of airborne particulate matter
    among its various sources (e.g., coal-fired power plants,
    gasoline-powered vehicles and diesel-powered vehicles). In such source
    apportionment models, profiles of element concentrations in
    particulate matter emitted from different sources are combined with
    sophisticated statistical methods (e.g., principal component or factor
    analysis) to estimate the relative abundance of particles from each
    source type. Glover et al. (1991) and Daisey et al. (1986) provide a
    good introduction to source apportionment models for particulate
    matter, while Edgerton & Shah (1991) describe a source apportionment
    model for VOCs.

         Several approaches have been used to estimate expected indoor air
    pollution concentrations (for reviews see Cooke, 1991; WHO, 1997b).
    These include deterministic models based on a pollutant mass balance
    around a particular indoor air volume; a variety of empirical
    approaches based on statistical evaluation of test data and (usually)
    a least squares regression analysis; or a combination of both
    approaches, empirically fitting the parameters of a deterministic
    model with values statistically derived from experimental measurements
    (see Chapter 4). All three approaches have advantages and weaknesses.
    The deterministic model provides more generality in its application,
    but the results lack accuracy and precision. Deterministic models
    include single- and multiple-compartment models. The empirical models,
    when applied within the range of measured conditions for which they
    were fitted, provide more accurate information. An example of an
    empirical model for indoor concentrations of respirable particulate
    matter may be found in Chapter 12. Often the compartment of the indoor
    air mass balance models that is most difficult to represent is the
    role of indoor surfaces as sources or sinks for contaminants. This is
    an important field of inquiry with respect to inhalation exposures to
    ozone and VOCs (Reiss et al., 1995).

    6.4.2  Potable water

         Exposure to contaminants in water may occur via the ingestion,
    dermal absorption and inhalation routes. Ingestion of water primarily
    occurs via two pathways: direct ingestion via drinking or cooking and
    intrinsic water intake (i.e., the water intrinsic in foods prior to
    preparation). It is important to consider both routes. Drinking-water
    ingestion rates have also been shown to vary according to cultural
    differences and can be an important source of uncertainty about
    chemical exposure when extrapolating results of epidemiological
    studies from one culture to another (e.g., Mushak & Crocetti, 1995).
    Lognormal distributions of drinking-water ingestion rates for
    individuals comprising various age groups in the USA (Table 18) are
    available in the literature (Roseberry & Burmaster, 1992). Additional
    information on drinking and cooking water as exposure media may be
    found in Chapter 7.

         Dermal absorption of contaminants in residential water sources
    may occur during bathing and other forms of washing or cleaning. There
    are three principal mechanisms by which molecules can transverse the
    skin and enter the body: passive transfer or diffusion, facilitated
    diffusion and active transport. Passive diffusion is the mechanism
    most commonly expressed in dermal exposure models. The rate of passive
    diffusion is a function of the concentration gradient of the
    contaminant on the surface of the skin and in the tissue immediately
    below the skin and the ease with which a molecule of the contaminant
    can move through the lipophilic interior of the skin membrane. Ease of
    passage is a function of the partition coefficient of the contaminant
    (e.g., the octanol-water partition coefficient,  Kow), molecular
    size, the degree of ionization and the porosity of the skin. Porosity
    of the skin to VOCs present in drinking-water treated with chlorine
    has been shown to be temperature dependent (Gordon et al., 1998).

    Table 18.  Lognormal distributions of water intake by age group in 
               the USA. Source: Roseberry & Burmaster (1992)

                                                                          
    Age group      Geometric mean (ml/day)    Geometric standard deviation
                                                                          
    Drinking and cooking water intake
    < 1 year                267                         1.85
    2-11                    620                         1.65
    12-20                   786                         1.72
    21-65                  1122                         1.63
    > 65                   1198                         1.62

    Total water intake (direct + intrinsic)
    < 1 year               1074                         1.34
    2-11                   1316                         1.40
    12-20                  1790                         1.41
    21-65                  1926                         1.49
    > 65                   1965                         1.50
                                                                          


         Inhalation exposures to VOCs transferred from water to air could
    be as great as, or even greater than, exposures from ingestion.
    Inhalation pathways include contaminants transferred to the air from
    showers, baths, toilets, dishwashers, washing machines and cooking.
    Several models have been proposed to explain the mass-transfer
    process; in particular, a time-dependent, three-compartment model for
    residential exposure (McKone et al., 1987). The three compartments
    used in such a model are the shower/bath stall, the bathroom and the
    remaining residential volume. Factors that affect the projected
    exposure are chemical mass-transfer rates from water to air,
    compartment volumes, air-exchange rates and human occupancy factors.

    6.4.3  Surface waters

         The transport of contaminants in surface waters is determined by
    two factors: the rate of physical transport in the water system and
    the chemical reactivity. Physical transport processes are dependent to
    a large extent on the type of water body under consideration (e.g.,
    oceans, seas, estuaries, lakes, rivers or wetlands). Schnoor (1981)
    and Schnoor & MacAvoy (1981) have summarized important issues related
    to surface water transport. At low concentrations, contaminants in
    natural waters exist in both dissolved and sorbed phases. In rapidly
    moving water systems, advection controls mass transport and dissolved
    substances move at essentially the same velocity as the bulk of the
    water in the water system. Contaminants that are sorbed to colloidal
    materials and fine suspended solids can also be entrained in the
    current, but they may undergo additional transport processes that
    increase their effective residence time in surface waters. Such
    processes include sedimentation, deposition, scour and resuspension.
    Thus, determining the transport of contaminants in surface waters
    requires that we follow both water movement and sediment movement.

         A water balance is the first step in assessing surface water
    transport. A water balance is established by equating gains and losses
    in a water system with storage. Water can be stored within estuaries,
    lakes, rivers and wetlands by change in elevation or stage. Water
    gains include inflows (both runoff and stream input) and direct
    precipitation. Water losses include outflows and evaporation.

    6.4.4  Groundwater

         In groundwater, the dilution of contaminants occurs much more
    slowly than it does in surface water. After precipitation, water
    infiltrates the ground surface where it travels vertically down
    through the unsaturated zone until it contacts the water table, and
    then flows approximately horizontally. This horizontal movement is
    driven by the hydraulic gradient, which is the difference in hydraulic
    head at two points divided by the distance (along the flow path)
    between the points. Bear & Verruijt (1987) and Freeze & Cherry (1979)
    have compiled extensive reviews on the theory and modelling of
    groundwater flow and on transport of contaminants in groundwater. The
    movement of contaminants in groundwater is described by two principal
    mechanisms: gross fluid movement (advective flow), and dispersion.
    Dispersion depends on both fluid mixing and molecular diffusion. The
    transport of many chemical species in groundwater is often slowed or
    "retarded" relative to the flow of the bulk fluid by sorption of the
    contaminant material to soil particles or rock. As is pointed out by
    Bear & Verruijt (1987), many groundwater models are available for
    assessing the transport of contaminants in the subsurface environment,
    ranging from simple one-dimensional hand calculations to large
    three-dimensional computer programmes. The choice of an appropriate
    model for any situation depends to a large extent on the information
    available, the type of information needed to carry out an exposure
    assessment and the tolerance of the analyst for large, complex
    computer programmes.

    6.4.5  Soil

         Soil, the thin outer zone of the earth's crust that supports
    rooted plants, is the product of climate and living organisms acting
    on rock. A true soil is a mixture of air, water, mineral and organic
    components (Horne, 1978). The relative mix of these components
    determines to a large extent how a chemical will be transported and/or
    transformed within the soil. The movement of water and contaminants in
    soil is typically vertical as compared to horizontal transport in the
    groundwater (i.e., saturated) zone. A chemical contaminant in soil is
    partitioned between soil water, soil solids, and soil air. For
    example, the rate of volatilization of an organic compound from the
    soil solids or from soil water depends on the partitioning of the
    compound into the soil air and on the porosity and permeability of the
    soil.

         Models developed for assessing the behaviour of contaminants in
    soil can be categorized in terms of the transport/transformation
    processes being modelled. Partition models such as the fugacity models
    of Mackay (1979) and Mackay & Paterson (1981, 1982) describe the
    distribution of a contaminant among the liquid, solid and water phases
    of soils. Jury et al. (1983) have developed an analytical screening
    model that can be used to calculate the extent to which contaminants
    buried in soil evaporate to the atmosphere. The multiple-media model
    GEOTOX (McKone & Layton, 1986) has been used to determine the
    inventory of chemical elements and organic compounds in soil layers
    following various contamination events. This model addresses
    volatilization to atmosphere, runoff to surface water, and leaching to
    groundwater and first-order chemical transformation processes.

    6.5  Multiple-media modelling

         Human beings come directly into contact with certain media via
    certain routes and are exposed to the chemicals therein as depicted in
    Table 19. Efforts to assess human exposure from multiple media date
    back to the 1950s when the need to assess human exposure to global
    fallout led rapidly to a framework that included transport both
    through and among air, soil, surface, water, vegetation and food
    chains (Whicker & Kirchner, 1987). Efforts to apply such a framework
    to non-radioactive organic and inorganic toxic chemicals have been
    more recent and have not as yet achieved such a high level of
    sophistication. In response to the need for multiple-media models in
    exposure assessment, a number of transport and transformation models
    have recently appeared. In an early book on multiple-media transport,
    Thibodeaux (1996) proposed the term "chemodynamics" to describe a set
    of integrated methods for assessing the cross-media transfers of
    organic chemicals. The first widely used multiple-media compartment
    models for organic chemicals were the fugacity models proposed by
    Mackay (1979, 1991) and Mackay & Paterson (1981, 1982). Cohen and his
    co-workers introduced the concept of the multiple-media compartment
    model and more recently the spatial multiple-media compartment model,
    which allows for non-uniformity in some compartments (Cohen & Ryan,
    1985, Cohen et al., 1990). Another multiple-media screening model,
    called GEOTOX (McKone & Layton, 1986; McKone et al., 1987), was one of
    the earliest to explicitly address human exposure.

         The preceding models deal with inter-media transfer of
    contaminants on a relatively large scale, but other models are scaled
    to the residence and exposures that may occur therein. Exposure to
    chemicals in consumer products such as cleaning agents and paint are
    the focus of a model called CONSEXPO (van Veen, 1996).

         All multiple-media exposure models have at least two features in
    common, regardless of the objective for which they were designed.
    First, movement of contaminants from one medium to another is
    characterized. Second, the rate and/or frequency of human contact with
    environmental media is modelled. The former may be referred to as
     inter-media transfer factors and the latter as  exposure factors. 

    Table 19.  Potential human exposure media and routes

                                                          
    Environmental medium      Exposure routes
                                                          

    Air                       dermal contact inhalation

    Tap water                 dermal contact ingestion

    Food and beverages        ingestion

    Surface soil              dermal contact ingestion

    Surface water             dermal contact ingestion
                                                          


    6.5.1  Inter-media transfer factors

         Transfer of contaminants between media is commonly modelled as
    partitioning of a chemical between two or more media. Thus,
    multiple-pathway models require the measurement or estimation of
    partition coefficients of contaminants between several pairs of
    environmental media. There are two general classes of partitioning
    coefficients. The first class relies on basic physicochemical
    properties of the compounds of interest such as aqueous solubility,
    vapour pressure and dipole moment; they describe partitioning due to
    diffusive processes. Coefficients in the second class describe
    partitioning resulting from what may be considered advective
    processes, but also implicitly include diffusive partitioning.

    6.5.1.1  Diffusive partition coefficients

         The class of diffusive partition coefficients includes those
    between soil and water in soil (e.g., groundwater), air and plants,
    soil and plants, animal intake and food, surface water and fish,
    mother's uptake and breast milk, residential water and indoor air,
    soil-gas and indoor air, human skin and soil, and human skin and
    water. In many cases, partition coefficients developed from
    laboratory-scale experiments are the basis for modelling partitioning
    of a compound between environmental media (Lyman et al., 1990). For
    example, the octanol-water partition coefficient is often used as a
    proxy for partitioning non-polar organic compounds (e.g.,
    organochlorine substances) between water and fish lipids. In this
    case,  n-octanol is considered a good model for fish lipids.
    Similarly, the organic carbon-water partition coefficient is used to
    characterize partitioning of non-polar substances between organic
    matter in soil and water. Finally, Henry's constant describes
    partitioning of volatile and non-volatile compounds between air and
    water. Connell et al. (1997) provide a comprehensive introduction to
    the use of this type of partition coefficient in environmental science
    and exposure assessment.

    6.5.1.2  Advective partition coefficients

         The second class of partitioning coefficients jointly describe
    bulk transfer of compounds from one medium to another and diffusive
    partitioning. They are often used to model active uptake of
    contaminants by animals, principally livestock and game such as fish
    or fowl. Factors of this type are used to model transfer of
    semi-volatile compounds (SVOCs) such as dioxins from air to soil, soil
    to beef and soil to cow's milk (e.g., Nessel et al., 1991; Fries,
    1995). Bioaccumulation of lipophilic compounds and some forms of heavy
    metals (e.g., methylmercury) in fish from ingestion of contaminated
    prey and diffusive uptake through respiration is also modelled using
    partition coefficients such as these (e.g., MacIntosh et al., 1994).

    6.5.2  Exposure factors

         In constructing exposure models one needs to define the
    characteristics of individuals in various age and sex categories and
    the characteristics of the microenvironments in which they live or
    from which they obtain water and food. The types of data needed to
    carry out the exposure assessment include exposure duration and
    averaging time, time-activity patterns of individuals, food
    consumption patterns, household parameters, human factors such as body
    weight, surface area, soil ingestion and breast milk intake, and
    parameters associated with food crops and food-producing animals.

         Time-activity patterns provide information on how individuals
    distribute their time among a number of potential exposure media.
    Time-activity pattern data describe such things as the average number
    of hours spent indoors at home and in what rooms and the nature of
    activity. Time-activity data also includes information on time spent
    outdoors at home or spent in microenvironments, such as bathrooms
    (including shower and bathing time). Exposure times are activity data
    that involve the number of days per year and hours per day spent in
    contact with soil during recreation and home gardening and in contact
    with surface water during swimming or other water recreation.
    Household factors relate to drinking-water supply and use,
    room-ventilation rates, and soil and dust concentrations within homes.
    Soil ingestion rates and soil contact on skin are also needed. Methods
    for measuring time-activity patterns and related considerations are
    discussed in detail in Chapter 5.

         Input data of these types may be measured in the population under
    investigation, i.e., site specific, or may be drawn from standard
    references such as AIHC (1994), Finley et al. (1994b) and US EPA
    (1996a). Site-specific data are preferred, in case the population of
    interest may exhibit unique characteristics expected to influence
    exposure. If site-specific data are not available, values observed in
    other populations or estimates may be applied. Some model applications
    may rely solely on estimated inputs. For example, screening models are
    often used to assess exposure and health risks associated with new
    products such as pesticides designed for agricultural and residential
    use. In this case, model inputs may be determined in a manner such

    that the model result is unlikely to underestimate the true level of
    exposure experienced by the population of interest. Models such as
    these are often referred to as "worst-case" models. An exposure
    modelling system recently developed by the European Union contains a
    suite of screening models (EC, 1996).

    6.5.3  Multiple-media/multiple-pathway models

         Multiple-media or so-called "total" exposure models provide
    methods for integrating multiple exposure pathways from multiple
    environmental media into a model system that relates concentrations of
    toxic chemicals to potential total human dose at toxic substances
    release sites. This type of simulation matrix is used to generate the
    hypothetical histogram shown in Fig. 22. The scenarios used to develop
    this particular histogram are for a representative VOC incorporated in
    the top several metres of soil. Here we can see that, based on a
    multiple-media and multiple-pathway assessment, we get indications of
    where it is most valuable to focus our resources to more fully
    characterize distributions of population exposure. In this way, we
    characterize total potential dose using comprehensive, simple and
    possibly stochastic models to focus efforts on those exposure
    pathways, media and scenarios that require more realistic assessment
    of the distribution of dose within the population. This matrix allows
    us to make both pathway-to-pathway and medium-specific comparisons of
    total potential doses from multiple environmental media.

    6.6  Probabilistic exposure models

          Variability and  uncertainty are two important and related
    concepts in exposure modelling, but it is important to distinguish
    between them.  Variability arises from true heterogeneity across
    people, places or time; uncertainty represents a lack of knowledge
    about factors affecting exposure (or risk). Thus, variability can
    affect the precision of model estimates and the degree to which they
    can be generalized, whereas uncertainty can lead to inaccurate or
    biased estimates (Hoffman & Hammonds, 1994). It should be noted that
    variability and uncertainty can complement or confound one another.
    They may also have fundamentally different manifestations. For
    example, uncertainty may force decision-makers to judge how
    practicable it is that exposures have been over- or underestimated for
    every member of the exposed population, whereas variability forces
    them to cope with the certainty that different individuals are subject
    to exposures both above and below any of the exposure levels chosen as
    reference points (US NRC, 1994).

         Failing to distinguish between variability and uncertainty makes
    it difficult to accurately characterize the population distribution of
    exposure and to make informed decisions about priorities for future
    research objectives. Exposure models can allow for consideration of
    both variability and uncertainty.

    FIGURE 22

    6.6.1  Variability

         Diverse sources of environmental contaminants lead to various
    contaminated media (e.g., soil, dust, water, air, food), which in turn
    result in a multitude of routes and pathways of human exposure. For a
    given contaminant, the magnitude and relative contribution of each
    exposure route and pathway may vary among geographic regions and over
    time. In addition, differences in activities among individuals lead to
    disparate rates of contact with contaminated media. In aggregate,
    these factors result in varying levels of personal exposure to a
    particular contaminant among the members of a population, i.e., a
     distribution of exposures.

         Exposure model inputs expressed as distributions can be used to
    model inter-individual variability of exposures. Examples of
    probabilistic human exposure models that explicitly consider
    variability of exposure among individuals may be found in Finley et
    al. (1994a) and MacIntosh et al. (1995, 1996). Variable parameters are
    those that are stochastic with respect to the reference unit of the
    assessment question (IAEA, 1989) and are described by probability
    distributions that reflect their intrinsic randomness. Exposure
    concentrations may vary between individuals owing to the influence of
    personal activities (e.g., cigarette smoking contributions to indoor
    respirable particulate levels). Such differences represent true
    variability of factors that affect exposure among individuals and can
    determine the relative position of an individual or type of individual
    within the distribution of exposures for the population.

    6.6.2  Uncertainty

         Several publications have stressed the importance of
    distinguishing among different types of uncertainty (IAEA, 1989; US
    EPA, 1992c). Explicit consideration of uncertainty in exposure and
    risk assessments is important for understanding the range and
    likelihood of potential outcomes and the relative influence of
    different assumptions, decisions, knowledge gaps and stochastic
    variability in inputs on these outcomes (Bogen & Spear, 1987; Iman &
    Helton, 1988; IAEA, 1989; Morgan & Henrion, 1990; US EPA, 1992c). This
    understanding can help the analyst advise the decision-maker on an
    appropriate course of remedial action, decide whether it is worthwhile
    to collect additional information regarding model parameters, choose
    the appropriate model to use and evaluate which of these actions could
    be most effective in reducing uncertainty about the outcomes (IAEA,
    1989; Morgan & Henrion, 1990).

         Three types of uncertainty are commonly considered:  scenario 
     uncertainty, arising from a lack of knowledge required to fully
    specify the problem;  model uncertainty, arising from a lack of
    knowledge required to formulate the appropriate conceptual or
    computational models; and  parameter uncertainty, arising from a lack
    of knowledge about the true value or distribution of a model parameter
    (US EPA, 1992c). In practice, scenario and model uncertainty are
    commonly considered to be negligible relative to parameter

    uncertainty, although in many cases they may be the largest sources of
    true uncertainty.

         Uncertain parameters are those for which the true value is not
    known or cannot be measured. For example, the true annual mean
    concentration of respirable particles in Mexico City during 1996 is
    uncertain because it can only be estimated from existing data which do
    not cover every day of the year nor every location of the city.
    Another example, is the mean and variance of soil ingestion by
    children aged 6-10 years in Taipei. Presumably, a single distribution
    can be used to describe this behaviour; however, its parameters can
    only be estimated.

         The uncertainty about various parameters of an assessment can be
    formally incorporated into exposure models to estimate uncertainty
    about the prediction end-point, identify the components that influence
    prediction uncertainty and prioritize future research needs (Bogen &
    Spear, 1987; IAEA, 1989). Uncertainty about the true population
    distributions is characterized by propagating the estimated
    uncertainty about model inputs through to the distributions of the
    prediction end-points.

    6.6.3  Implementing probabilistic exposure models

         Although probabilistic exposure models are computationally more
    challenging to implement than deterministic (i.e., point estimate)
    models, the advantages of being able consider population distributions
    and sources and magnitude of uncertainty are often worth the
    additional effort. Several tools are available for propagating input
    parameter variability and uncertainty through to the assessment
    end-point. Classical error propagation techniques may be convenient
    for models with relatively few inputs and small coefficients of
    variation (Bevington, 1969; Seiler, 1987). For more complex models,
    computer-based simulation techniques are likely to be the method of
    choice.

         Probabilistic exposure models may be run in one or two
    dimensions.  One-dimensional models estimate either variability among
    exposures to individuals or uncertainty about a single exposure
    metric; for example, the mean 8-h average carbon monoxide exposure for
    individuals in a specific area.  Two-dimensional simulation models 
    may be used to estimate both population distributions (i.e.,
    inter-individual variability) and uncertainty about the population
    distribution. The IAEA (1989) has suggested a Monte Carlo simulation
    approach for conducting two-dimensional simulations. In the first
    phase, a single realization is obtained from the distribution of each
    uncertain parameter. In the second phase, repeated realizations are
    obtained from the variable parameters. The entire process of a single
    sampling from the uncertain parameters, followed by repeated sampling
    from the variable parameters, is referred to as a  simulation. A
    single model run consists of generating  k simulations each composed
    of  i iterations, which produces a family of  k predicted
    distributions of population exposures. Prediction uncertainty is

    represented by the distribution of individual estimates for a specific
    percentile or summary statistic among the family of population
    distributions. In this way, the type of plot shown in Fig. 23 contains
    probabilistic information on estimates of both inter-individual
    variability in the prediction end-point, and uncertainty about any
    specific percentile of the population distribution.

    6.7  A generalized dose model

         The magnitude of exposure (dose) is the amount of agent available
    at human exchange boundaries (skin, lungs, gastrointestinal tract)
    where absorption takes place over a specified period of time.
    Depending upon boundary assumptions, a number of dose questions may be
    derived. The  average daily dose (ADD) is one of the most useful
    approaches, and is applied for exposure to non-carcinogenic compounds
    (for carcinogens,  lifetime average daily dose, LADD, is often
    employed). The ADD is calculated by averaging the potential dose
    ( Dpot) over body weight and the appropriate averaging exposure time:

         ADD = total potential dose/body weight × averaging time,

    where the potential dose is a product of contaminant concentration
     (C) in the exposure medium contacting the body, intake rate  (IR) 
    and exposure duration  (ED):

         total potential dose =  C ×  IR ×  ED.

    The intake rate refers the rates of inhalation, ingestion or dermal
    contact depending on the route of exposure.

         The concentrations in air, water and soil used for an exposure
    assessment are those measured or estimated to be available in these
    environmental media at the nearest receptor point to the source (e.g.,
    soil or groundwater at a hazardous waste site). When an environmental
    concentration is assumed constant over a long time period, the
    population-averaged potential dose (for ingestion or inhalation
    pathways) or absorbed dose (for dermal contact) is expressed as an
    average daily dose (ADD) in mg kg-1 day-1:

    FIGURE

    where [ Ci/ Ck] is the intermedia-transfer factor, which expresses
    the ratio of contaminant concentration in the exposure medium  i 
    (i.e., personal air, tap water, milk, soil, etc.) to the concentration
    in an environmental medium  k (ambient air gases or particles,
    surface soil, root-zone soil, surface water and groundwater);
    [ IUi / BW] is the intake or uptake factor per unit body weight
    associated with the exposure medium  i. For exposure through the

    FIGURE 23

    inhalation or ingestion pathway [ IUi / BW] is the intake rate per
    unit body weight of the exposure medium such as m3(air) kg-1 day-1,
    litres(milk) kg-1 day-1, or kg(soil) kg-1 day-1. For exposure
    through the dermal pathway, [ IUi / BW] is replaced by  UFi, the
    uptake factor per unit body weight as a fraction of the initial
    concentration in the applied medium with nominal units [litres(water)
    kg-1 day-1 or kg(soil) kg-1 day-1];  EF is the exposure frequency
    for the exposed population in days per year;  ED is the exposure
    duration for the exposed population in years;  AT is the averaging
    time for the exposed population in days; and  Ck is the contaminant
    concentration in environmental medium  k. 

         The potential dose factor, PDF( k-> i), is defined as the
    ratio of dose to concentration, as expressed in the following
    equation:

    FIGURE

         The ADD is used to make route and route-to-route comparisons and
    allows one to consider the relative significance of several exposure
    routes. With the ADD, we compare inhalation, ingestion or dermal
    exposures to the same medium such as tap water and compare exposures
    through indirect pathways (e.g., food-chain transfers) to those from
    direct pathways (e.g., inhalation or ingestion). As an example, the
    ADD for the ingestion route for chloroform for a 70-kg individual
    ingesting 2 litres/day of tap water containing 1 µg/litre chloroform,
    365 days/year for a lifetime is 2 µg/day divided by 70 kg or 0.029
    µg/kg-1 day-1. This ADD can be used as the basis for determining the
    relative significance of dermal, inhalation, and other ingestion
    exposures attributable to tap water.

    6.8  Physiologically based pharmacokinetic models

         Human exposure to contaminants results in dose to the critical
    organs. A mass balance on the contaminants that enter the body
    accounts for the distribution in the various organs, transformation
    into by-products, and excretion via specific mechanisms. The three
    major exposure routes by which contaminants enter the human body are
    inhalation, dermal absorption and ingestion. The vehicle that moves
    contaminants between organs is blood. Transformations include the
    metabolism of specific contaminants in specific organs. Mechanisms of
    excretion include exhaled air, sweat, urine and faeces.

         The above processes that occur in the human body can be modelled
    by using physiologically based pharmacokinetic (PBPK) principles
    (Masters, 1991). These principles can be applied at differing levels
    of complexity. Simple models assume steady states and total absorption

    and estimate dose to critical organs in a gross manner. They can be
    solved by using linear algebraic relationships. Complex models include
    time dependency, assume the human body to consist of multiple
    homogeneous boxes, each representing an organ or a portion thereof,
    and determine the distribution of contaminants in the different boxes
    as a function of time. The relationships usually end up as non-linear
    ordinary differential equations that are solved by using numerical
    integration techniques. Examples of PBPK models may be found in Cox
    (1996) for inhalation of benzene, Bookout et al. (1997) for dermal
    absorption of chemicals and Rao & Ginsberg (1997) for multiple-route
    exposure to methyl  tert-butyl ether. A wide array of PBPK models
    have been developed for other chemicals and chemical classes and may
    be found in the relevant literature.

         Whatever the complexity of the model representing the human body,
    the difficulty is interpreting the dose results to characterize risk.
    Usually, these human models are extrapolated to parallel animal models
    for which toxicological data are available.

    6.9  Validation and generalization

         The modelling approaches described above are mathematical
    abstractions of physical reality that may or may not provide adequate
    estimates of exposure. The preferred way to be sure that a model is
    capable of providing useful and accurate information is by validation,
    i.e., comparing model predictions with measurements independent of
    these used to develop the model. Models can be validated in terms of
    prediction accuracy and precision by comparing predicted values to
    those measured in the field. Although measurements are preferable as
    the "gold standard" in validation of models, comparison of results
    from different assessment methods or modelling approaches can also be
    used to evaluate validity, or at least agreement. This may be the only
    option when measurements are not feasible; for example, in
    retrospective assessment of exposure. Model validation is a necessary
    precondition for the generalization of model results to a different or
    larger population (Ott et al., 1988).

         In the statistical modelling approach, data collection is an
    integral part of model construction. If the data are known to be from
    a statistically representative sample of the population, then there is
    no need for further validation. However, validation is necessary if
    the results are to be extrapolated beyond the range for which the
    existing database provides a statistical description. The physical and
    physical-stochastic modelling approaches must be validated with actual
    data from separately conducted field studies. Care must be taken that
    the data used to validate a model are not biased with respect to
    crucial model parameters. The validation step is useful only to the
    degree that the sample population is representative of the group to
    which results will be extrapolated.

         Finally, when modelling environmental-response-health processes,
    and when validating such models, it is important to realize that in
    principle perfect modelling is possible only for closed systems, and

    the systems described in this report are very open-ended. The
    practical implication of this fact is that even the best models need
    to be validated for each new population and environmental setting
    before application.

    6.10  Summary

         An exposure model is a logical or empirical construct which
    allows estimation of individual or population exposure parameters from
    available input data. Exposure models, if supported by adequate
    observations, can be used to estimate group exposures (e.g., a
    population mean) or individual exposures (e.g., the distribution of
    exposures among members of a population). Models may be used to
    estimate exposure via single or multiple media. The latter is
    particularly useful for comparing the magnitude of exposures likely to
    occur from different media and thus for priority-setting. Exposure
    models may be statistical or deterministic in nature or a combination
    of both. Probabilistic methods may be applied to all three types as a
    means to estimate population distributions of exposure, i.e.,
    variability of exposure among individuals. In addition, probabilistic
    methods may be used to characterize uncertainty in model input
    parameters and propagate that uncertainty through to the prediction
    end-point. Evaluation of the accuracy of model results is critical
    before relying on model output for decision-making.

    7.  MEASURING HUMAN EXPOSURES TO CHEMICALS IN AIR, WATER AND FOOD

    7.1  Introduction

         This chapter describes sampling methods used in environmental
    exposure assessment to analyse chemical concentrations in air, water
    and food. The information presented provides a general description of
    available sampling methods and guidance for their selection. It is not
    intended to be comprehensive and the reader should refer to the
    research literature for specific details.

         Assessment of human exposures to contaminants in environmental
    media requires establishing measurement strategies and selecting
    appropriate sampling instruments and analytical methods. Taken
    together, these three elements define a monitoring programme.
    Monitoring methods can be used to determine the magnitude, duration
    and frequency of exposure to an environmental contaminant. Magnitude
    of exposure is defined as the concentration of a specific pollutant
    averaged over a predetermined time interval, such as 1 h, 24 h or a
    lifetime. Different measurement methods have specific characteristics
    that determine the locations in which they are feasible for use. In
    the case of air, the method's sensitivity to pollutants determines the
    averaging times over which it will provide reliable responses.
    Therefore, a clear understanding of the concentration range
    anticipated, averaging time of interest, and expected frequency of
    exposure events is needed to identify appropriate field and laboratory
    methods. In the absence of any prior information, pilot studies may be
    performed to obtain the information needed to finalize the design of
    the monitoring programme.

         Selection of instruments will depend on the target population
    (e.g., children or adults) and study objectives. In some situations,
    understanding the distribution or the average population exposure to a
    contaminant may be sufficient. In fact, most environmental monitoring
    of contaminants in outdoor air, water at the point of distribution and
    "market basket" surveys implicitly assumes that indicators of
    population exposure are more relevant than information at the
    individual level. Studies assessing individual exposures using such
    surrogate measures should select sampling instruments and analysis
    methods based on sensitivity, selectivity, response rate, portability,
    durability and cost, among other factors. Table 20 summarizes these
    concepts.

    7.2  Air monitoring

         Air sampling methodologies should conform to the exposure
    assessment approach selected, either direct or indirect, as described
    in Chapter 3.

         Direct monitoring methods for exposure measurements include the
    use of personal air monitors and/or analysis of human tissue and/or
    biological fluids. Aspects of biomonitoring are described in
    Chapter 10. Indirect air monitoring methods can include


        Table 20.  Selection factors for instruments and methods

                                                                                                                            
    Factor              Comment
                                                                                                                            

    Sensitivity         The magnitude and duration of contaminant exposure define the sensitivity required. As a general 
                        guide, one order of magnitude below and above the concentration of interest is desired. 
                        Reproducibility (precision) as measured by percentage relative error should be below 5%. Sensitivity 
                        is usually inversely proportional to integration time or amount of sample collected

    Selectivity         Response to a specific compound or analyte without interferences. In some cases, non-selective 
                        instruments may be appropriate if exposure situation (e.g., sources, emissions) are understood. 
                        Specific or selective response may require more expensive equipment or more time-consuming 
                        analytical procedures

    Response rate       There are two aspects of response rate: (i) time required for instrument to respond to 90% of a 
                        step change in concentration; (ii) time required between sampling and final processing of data. The 
                        appropriate instrument response rate depends, in part, on the relationship between the contaminant 
                        and the health effect of interest. Acute effects may require instrument methods that can resolve 
                        exposures over intervals of minutes. If health effects from chronic exposures are of primary concern 
                        or the metabolic half-life is long, then rapid response is not necessary

    Portability         Instruments and sampling procedures should not modify behaviour of subjects. Portability includes 
                        size, weight, noise, power, and safety considerations. Portability will influence study design and 
                        usually involves a tradeoff with sensitivity and response rate (e.g., integrated samples rather than 
                        continuous)

    Durability          Instruments used for air sampling are subjected to a broad range of conditions. Since temperature 
                        and humidity are potentially interferents and are not easily controlled, the performance of 
                        instruments/methods must be fully evaluated

    Cost                Instrument cost and analytical expenses will influence study design. It may be necessary to trade 
                        off sample cost for accuracy, precision, and response rate. Increasing the number of samples per 
                        subject and/or the number of subjects, or relaxing resolution requirements could compensate for the 
                        use of less expensive methods
                                                                                                                            
    

    microenvironmental sampling in combination with questionnaires and
    time-activity logs. Ambient air monitors can also be used to estimate
    exposures when combined with information such as building
    characteristics, indoor/outdoor contaminant ratios and time-activity
    patterns.

         The direct approach depends largely on the availability of
    sensitive, small, quiet, lightweight and portable personal monitors.
    Personal air monitors can be used for microenvironmental monitoring as
    well. In addition, microenvironmental monitors with larger sampling
    flows are used for indoor/outdoor sampling. Ambient monitors are
    generally high-volume samplers and are not suitable for indoor use.
    Suitable air monitors must also fulfil several requirements, such as
    detection limits, interferences, time resolution, easy operation and
    of course, cost. There are several good references on air monitoring
    and analysis. The reader is referred to  Air Sampling Instruments 
     for Evaluation of Atmospheric Contamination (ACGIH, 1995).
    Additional general publications include US EPA (1994, 1996b), and
    Lodge (1988). It is important, however, to refer to the published
    scientific literature for the most appropriate and recent air
    monitoring methods.

         The following sections describe methods available for air
    sampling of gases and vapours, airborne particulate matter, SVOCs and
    reactive gases. The methods are classified into active and passive or
    continuous monitors. A detailed list of sampling methods, air
    pollutants for which they are used, sources and other pertinent
    information is presented in Table 21-24. An indicator of their
    suitability for personal, indoor or ambient monitoring is also
    included.

    7.2.1  Gases and vapours

    7.2.1.1  Passive samplers

         Commercial passive samplers are available for a variety of air
    pollutants, including inorganic gases such as carbon monoxide,
    nitrogen dioxide, sulfur dioxide and ozone, and VOCs (e.g., benzene,
    toluene, xylene, etc.). Passive air samplers are probably the most
    convenient tool for conducting large-scale personal exposure
    assessments because they are small, inexpensive and easy to use.
    However, sampling rates are of the order of 10-50 ml/min and absorbing
    capacity is limited. Passive samplers operate on the principle of
    molecular diffusion. The rate of diffusion is related to the diffusion
    coefficient of the compound, the cross-sectional area of the absorbing
    surface and the length of diffusion path. Specific information on the
    calculation of sampling rates can be obtained from the manufacturers.
    The collection mechanism relies either on physico-chemical absorption
    or adsorption or chemical reactions. The samplers for inorganic gases
    rely on reaction of the contaminant with a chemical coating on the
    collection surface. The samplers for VOCs typically rely on absorption
    by a liquid or adsorption by a solid collection medium. Selection and
    use of passive samplers should take into consideration potential

    sources of error such as wind effects, temperature, humidity and
    interfering gases.

         In practical applications, personal monitoring is performed by
    mounting the passive sampler on a participant's collar to estimate air
    pollution concentrations in the breathing zone. After collection, the
    adsorbent is removed from the sampler and extracted with the
    recommended solvent. The extract is then analysed by a suitable method
    (e.g., spectrophotometry, gas chromatography with specific or
    unspecific detectors, HPLC, etc.). As with any monitoring procedure,
    measures should be taken to evaluate sample preservation and
    integrity. These procedures should be described as part of the quality
    assurance (QA) protocol and the standard operation procedures (SOPs)
    (see Chapter 11).

    7.2.1.2  Active samplers

         There are many commercially available liquid-media samplers for
    reactive and soluble gases, such as liquid-containing bottles, and
    solid-sorbent tubes for insoluble and non-reactive gases and vapours,
    such as activated charcoal, silica gel, porous polymers or other
    materials. Pollutants are transported with the carrier gas (air), and
    are captured by collecting media. The most frequently applied
    mechanisms in the collection of air pollutants in these media are
    chemical reactions (e.g., acid-base and colour-forming), and
    absorption/adsorption of the pollutant molecules on collecting media.
    Solid sorbent collection efficiency depends on contacting surface
    area, air flow rate, temperature, humidity and presence of interfering
    compounds.

         The sampling rate, breakthrough volume and method limit of
    detection are important parameters which need to be considered for an
    accurate exposure assessment by active samplers. The identification
    and quantification of collected air pollutants are usually performed
    by analytical instruments, such as spectrophotometry, gas
    chromatography with specific or non-specific detectors, HPLC, etc.
    Although not yet used extensively, small, evacuated canister samplers
    have been developed for personal monitoring (Pleil & Lindstrom, 1995).
    These have the advantage of not using sorbents. Analysis is typically
    done by gas chromatography following thermal desorption.

    7.2.1.3  Direct-reading instruments

         The concentration of gases and vapours (e.g., carbon monoxide,
    sulfur dioxide) in an individual's breathing zone can also be
    determined with the use of portable direct-reading instruments.
    Commercially available direct-reading instruments have data logging
    capabilities to store measurements at a rate of 1 s-1. Depending on
    the frequency of measurements, these instruments can operate up to
    2 weeks continuously. Instrument software allows for direct
    calculation of concentrations with different averaging times and
    statistical analysis of the data.


        Table 21.  Air sampling methods for inorganic gases

                                                                                                                            
    Carbon monoxide           Manufacturer            Comments                                     Application
                                                                                                                            

    Continuous
       Electrochemical        Energetic Sciences      0-50, 0-100 ppm; portable and personal;      environmental/personal
                                                      LOD ~ 2 ppm

                              Interscan               Various ranges; LOD ~ 1 ppm                  environmental

                              Bacharah                Based on the measurement of Hg vapour        environmental
                                                      from a pellet oxidized by CO. Range: 0-5, 
                                                      0-20 dl: 1 ppm Sample flowrate: 4.7 
                                                      litre/min

       Photometers            Beckman Instruments     Based on dual-isotope fluorescence,          environmental
                                                      LOD = 0.1 ppm

    Passive
       Diffusion detectors    Lab Safety Supply Co.   Changes colour; LOD ~ 50 ppm for 8 h         personal

                              Quantum Group Inc.      Simple colour change detector                personal

                              3M Corporation          Indicates presence of CO by colour           personal
                                                      change

                              Wilson Safety Products  Dosimeter badge. Colour change is            personal
                                                      proportional to CO concentration

    Active                    MSA                     Air is pumped through activated charcoal     personal
                                                      tubes that change colour when CO is present

                              Sensidyne
                                                                                                                            

    Table 21.  (continued)

                                                                                                                            
    Carbon monoxide           Manufacturer            Comments                                     Application
                                                                                                                            

    Continuous
       Infrared               GasTech                 300-5000 ppm                                 environmental

                              Rosemount Analytical    Measures CO, CO2, NO and hydrocarbons        environmental

                              SKC West                Sampling frequency: 8 s to 30 min            environmental

       Electrochemical        Devco Engineering       Based on conductivity in water due to        environmental
                                                      ionization of gas

    Nitrogen oxides
    Continuous
       Electrochemical        Trasducer Research      LOD > 2 ppb                                  environmental

                              Interscan               Various ranges; LOD > 20 ppm                 environmental

       Chemiluminescence      Beckman Instruments     Range: 0.1-1 ppm. Operates continuously      environmental
                                                      for 7 days. Analyses NO, NO2, NOx based 
                                                      on the excitation of molecules by light

                              Columbia Scientific     Uses the chemiluminescence reaction of O3    environmental
                                                      with NO. Sampling rate: 1.2 litre/min

                              Rosemount Analytical    Designed to monitor continuous emissions     environmental

       Colorimetric           Phillips Electronics    Set for a variety of chemicals, depending    environmental
                              Instruments             on the electrolyte. Measures concentration
                                                      based on a specific chemical reaction

                                                                                                                            

    Table 21.  (continued)

                                                                                                                            
    Carbon monoxide           Manufacturer            Comments                                     Application
                                                                                                                            

    Passive
       Diffusion              Env Sciences and        LOD ~ 500 ppb for a 1-h exposure             personal
       tubes/badges           Physiology

                              MDA Scientific          Palmes sampler is an acrylic tube with       personal
                                                      stainless steel grids coated with 
                                                      triethanolamine placed at the bottom

                              RS Landauer Jr. & Co    Pen-shaped badge for the collection of       personal
                                                      N2O on a molecular sieve. Analysis with 
                                                      IR

    Active
    Electrochemical           MDA Scientific          2-3 ppm; measurement on a 15-min basis       personal

    Ozone
    Continuous
       Chemiluminescence      Beckman Instruments     Operates continuously for 7 days Based on    environmental
                                                      the reaction of ozone with ethylene to 
                                                      produce light  Range: 0-0.0025 ppm, 
                                                      DL: 0.01 ppm

                              Philips Electronics     Operates continuously for 7 days. Based      environmental
                              Instruments             on the reaction of ozone with ethylene 
                                                      to produce light

                              Columbia Scientific     Based on the reaction of ozone with          environmental
                                                      ethylene. Ranges: 0-0.1, 0-0.2, 0-0.5, 
                                                      0-1.0 ppm

       UV Vis photometer      Dasibi Environmental    Concentration is determined by detecting     environmental
                                                      the absorption level of UV within a volume 
                                                      of air

                                                                                                                            

    Table 21.  (continued)

                                                                                                                            
    Carbon monoxide           Manufacturer            Comments                                     Application
                                                                                                                            

                              Mast Development        Portable. Sampling rate 2 litre/min,         environmental
                                                      measurement cycle 20 s

    Passive
       Diffusion monitors     Ogawa                   Uses 2 multitube diffusion barriers          environmental/
                                                      with collection on glass fibre filters       personal
                                                      coated with nitrite
                                                                                                                            

    LOD: Level of Detection

    Table 22. Air sampling methods for organic vapours

                                                                                                                                     
                                    Manufacturer                             Comments                                 Application
                                                                                                                                     

    Continuous
       Photo-ionization detector    Thermo Environmental Instruments         Based on UV light, photoionization       environmental
                                                                             detectors can detect a wide 
       Flame ionization detectors   Columbia Scientific                      variety of chemical compounds.

                                    Foxboro                                  Measures hydrocarbons as methane         environmental
                                                                             equivalents. Sample flowrate 
                                                                             20 ml/min

                                                                             Mainly used as a portable survey         environmental
                                                                             equipment. Based on hydrogen flame 
                                                                             ionization detection. Sample 
                                                                             flowrate 2 litre/min, LOD ~ 0.2 ppm

       Thermal ionization           Photovac International                   Semiquantitative response                environmental
       detector

       Infrared photometers         Foxboro                                  Miran portable air analyser. Owing       environmental
                                                                             to its tunable IR wavelength, can 
                                                                             detect several organic compounds. 
                                                                             Sampling rate 28 litre/min

                                    Infrared Industries                      2 models. LOD = 25 ppm                   environmental

       Portable gas                 Photovac International                   Portable. Can detect selected VOCs:      environmental
       chromatographs                                                        Benzene, C4-C8, halocarbons down 
                                                                             to ppb level

                                    H-Nu Systems                             Portable gas chromatographs with 5       environmental
                                                                             different detector options (FID, 
                                                                             PID, ECD, TCD, FPD)
                                                                                                                                     

    Table 22. (continued)

                                                                                                                                     
                                    Manufacturer                             Comments                                 Application
                                                                                                                                     

                                    Microsensor Systems                      Portable, isothermal gas                 environmental
                                                                             chromatograph. Samples are 
                                                                             concentrated in tubes, heated and 
                                                                             analysed. LOD = 2 ppb

    Passive
       Charcoal badges              3M                                       Single charcoal strips (300 mg).         personal/
                                                                             Sampling rate depends on the number      environmental
                                    SKC                                      of windows (1 or 2): 35-70 cm3/min. 
                                                                             Minimum collectable sample: 
                                    Gilian Instrument                        0.2 ppm/h

                                    Perkin Elmer                             Require laboratory analysis

                                    Pro-Tek

                                    3M                                       Two charcoal strips to avoid             personal/
                                                                             breakthrough and increase sample         environmental
                                    SKC                                      amount. Sampling rate depends on 
                                                                             the number of windows (1 or 2): 
                                    Gilian Instrument                        35-70 cm3/min. Minimum collectable 
                                                                             sample: 0.2 ppm/h.
                                    Perkin Elmer
                                                                             Desorption efficiency depends on 
                                    Pro-Tek                                  the amount and type of solvent used

                                                                             Require laboratory analysis

                                                                                                                                     

    Table 22. (continued)

                                                                                                                                     
                                    Manufacturer                             Comments                                 Application
                                                                                                                                     

    Active
       Charcoal tubes               Perkin Elmer                             The most commonly used adsorbent         personal/
                                                                             is activated charcoal.                   environmental
                                    National Draeger                         2 sizes of tubes are available : 
                                                                             100/50 mg or 200/100 mg.
                                    SKC

    Formaldehyde
    Passive                         GMD Systems                              LOD > 0.2 ppm for 15 min                 personal

                                    Interscan                                Various ranges; LOD > 20 ppm             personal/
                                                                                                                      environmental

                                    Air Quality Research                     LOD ~ 0.01 ppm for a 7-day exposure      personal

                                    DuPont                                   1.6-54 ppm up to 7 days                  environmental

                                    3M                                       LOD ~ 0.8 ppm for a 1-h exposure         personal
                                                                             Requires colorimetric analysis
                                    SKC                                                                               personal/
                                                                                                                      environmental

                                    AirScan Environmental Technologies       Based on crystal growth and              personal
                                                                             nucleation Length of stain is 
                                                                             proportional to concentration. 

                                    Environmental Science and Physiology     LOD ~ 500 ppb for a 1-h exposure         personal

                                    Envirometrics Products                   Based on electric reaction               personal
                                                                             with a lead-acid battery

                                                                                                                                     

    Table 22. (continued)

                                                                                                                                     
                                    Manufacturer                             Comments                                 Application
                                                                                                                                     

    Gases and Vapours
    Active
       Solid adsorbents             Barneby Cheney                           Large number of chemicals efficiently    personal/
                                                                             collected under a wide variety of        environmental
                                    Columbia Scientific Instruments          conditions.
                                                                             The choice of the adsorbent is 
                                    Draeger                                  designed to maximize collection 
                                                                             efficiency while retaining low 
                                    Fischer Scientific                       selectivity.
                                                                             Approximately 50 sorbent types 
                                    Perkin Elmer                             are available; some are chemically 

                                                                             treated to facilitate their 
                                    3M                                       collection properties.

                                    SKC                                      Most tubes contain a primary sorbent 
                                                                             section and a backup bed that is 
                                    Supelco                                  used to indicate breakthrough.

                                    Westvaco                                 Require laboratory analysis

       Polyurethane foam            Supelco                                  Collection of pesticides and PCBs        personal/
                                                                                                                      environmental

    Passive
       Diffusion monitors           3M                                       In general, the sorbent used is          personal/
                                                                             activated charcoal protected by a        environmental
                                    Gilian Instrument                        screen.

                                    SKC                                      Some monitors have a backup layer 
                                                                             used to indicate breakthrough.
                                    Supelco
                                                                             Each compound has a particular 
                                                                             diffusion rate
                                                                                                                                     

    Table 22. (continued)

                                                                                                                                     
                                    Manufacturer                             Comments                                 Application
                                                                                                                                     

                                                                             Require laboratory analysis

                                                                             Desorption efficiency will vary 
                                                                             with the amount of material on 
                                                                             the charcoal and with the amount 
                                                                             and type of desorber used.
                                                                                                                                     

    Table 23.  Air sampling methods for particulate matter/aerosols

                                                                                                                             
                                   Manufacturer              Comments                                         Application
                                                                                                                             

    Continuous
      Light-scattering             PPM                       LDL ~ 10 µm/m3 Handheld monitor                  environmental
      photometers
                                   Air Technique             Portable. Sample rate 28.3 litre/min             environmental
                                                             Suction: vacuum pump. Particles can be 
                                                             collected downstream of the filter 

                                   Virtis                    Near-forward sampling, Sample rate 28.3          environmental
                                                             litre/min. Suction: vacuum pump. 
                                                             Particles can be collected downstream 
                                                             of the filter 

                                   Hund                      Measures respirable aerosol mass                 personal or
                                                             concentration by IR scattering                   environmental
                                                             detection. Average of 8 h

                                                             Measures fine dust (0.2-10 µm) mass              environmental
                                                             concentration by IR scattering 
                                                             detection. Average of 8 h

                                   MIE                       Detects respirable dust. Portable.               fixed point/
                                                             Sample rate 2 litre/min                          environmental
                                                             Averages measurement over 8 h

                                                             Miniram dust monitor. Provides                   personal
                                                             instantaneous or 8 h average concentration

                                   Negretti                  Portable dust monitor, Range                     environmental
                                                             0.01-20/0.1-200 mg/m3. Particles can 
                                                             be collected on a filter

                                                                                                                             

    Table 23.  (continued)

                                                                                                                             
                                   Manufacturer              Comments                                         Application
                                                                                                                             

                                   Casella                   Handheld, Range 0.01-20/0.1-200 mg/m3),          environmental
                                                             size >0.1 µm. Particles can be collected 
                                                             on a filter

                                   TSI                       Integrating nephelometer averages over           mostly used 
                                                             30-s periods. Has different wavelengths          for visibility
                                                             depending on the aerosol characteristics

                                                             Laser photometer for particles >0.1              environmental
                                                             µm diameter. Measures aerosol 
                                                             concentration in mg/m3

                                   Belfort Instruments       Integrating nephelometer. Flowrate               environmental
                                                             10 litre/min

    Instantaneous                  MIE                       0.01-10 mg/m3 or 0.1-100 mg/m3                   personal
                                                                                                              aerosol
                                                                                                              monitor

      Condensation nucleus         Met One                   2 models. Sample flowrate 1.4 or 2.8             environmental
      counters                                               litre/min. Ultrafine particles are grown 
                                                             in alcohol vapour condensation

      Optical particle counters    Climet Instruments        6 models. Flowrate 0.3 -1.0 litre/min.           environmental
                                                             Size range: 0.3-20 µm, 5-16 size 
                                                             channels. Light source: white light or 
                                                             laser

                                   Hiac/Royco                6 models. Flowrate 0.01-1.0 litre/min            environmental
                                                             Range: 0.3-10 µm, 6 size channels
                                                             Light source: white light or laser

                                                                                                                             

    Table 23.  (continued)

                                                                                                                             
                                   Manufacturer              Comments                                         Application
                                                                                                                             

                                   Met One                   3 models. Flowrate 0.1-1.0 litre/min             environmental
                                                             Range: 0.1-1 µm, 6 size channels
                                                             Light source: white light or laser

                                   Particle Measuring        Flowrate 0.1-1.0 litre/min; range                environmental
                                   Systems                   0.05-5 µm, 4-16 size channels 
                                                             Illumination source: laser

                                   Faley International       4 models. Flowrate 0.017-1.0 litre/min           environmental
                                                             Range 0.3-5 µm, 2-5 size channels. 
                                                             Light source: white light

                                   TSI                       Flowrate: 0.1-1.0 litre/min; range               environmental
                                                             0.05-5 µm; 4-16 size channels

      Piezobalance                 TSI                       Less reliable for concentrations                 environmental
                                                             <10 µg/m3. Difficult to calibrate

      Beta gauge                   Wedding & Assoc           The particulate collected on the filter is       environmental
                                                             continuously measured by the attenuation of 
                                                             gamma-radiation

    Active (Total)
      Open cassette                SKC                       Air is pulled through a filter with no           personal
                                                             size selection device

      IOM inlet                    SKC                       Samples inhalable dust particles. Reusable       personal
                                                             filter cassettes. Sampling rate 2 litre/min. 
                                                             Cut point 100 µm

    Active (size selective)
      PM10 impactors               BGI                       Sampling rate 28.3 litre/min                     environmental
                                                             Suction: pump aerosol spectrometer

                                                                                                                             

    Table 23.  (continued)

                                                                                                                             
                                   Manufacturer              Comments                                         Application
                                                                                                                             

                                   MSP                       Sampling rate 4 or 10 litre/min;                 personal
                                                             Suction: pump

                                                             10-2.5 µm virtual impactor.                      environmental
                                                             Sampling rate 1130 litre/min

                                   TSI                       Sampling rate 1 litre/min;                       personal
                                                             Suction: pump

                                   MIE                       Sampling rate 2 litre/min;                       personal
                                                             Suction: pump

                                   Air Diagnostics           Sampling rate 4 litre/min;                       fixed 
                                                             Suction: pump                                    location/
                                                             environmental                                    environmental

                                   Graseby Andersen          Virtual dichotomous. Cutpoints 10                environmental
                                                             and 2.5 µm; sampling rate 16.7 litre/min

                                                             High-volume sampler. Sampling rate               environmental
                                                             1100 litre/min

                                   SKC                       Personal impactor. Single stage. Suction:        personal/
                                                             personal pump. Sampling rate 2, 4 or 10          environmental
                                                             litre/min

      PM2.5 impactors              URG                       Sampling rate 4 litre/min; part are              personal
                                                             collected in filters and organics in a 
                                                             polyurethane foam

                                   MSP                       Sampling rate 4 or 10 litre/min;                 personal
                                                             Suction: pump 

                                                                                                                             

    Table 23.  (continued)

                                                                                                                             
                                   Manufacturer              Comments                                         Application
                                                                                                                             

                                   SKC                       Personal impactor. Single stage.                 personal/
                                                             Suction: personal pump.                          environmental
                                                             Sampling rate 2, 4 or 10 litre/min

      Cascade impactors            BGI                       7 stages. Sampling rate 5 litre/min;             environmental
                                                             Cut points: 32, 16, 8, 4, 2, 1 µm

                                   Graseby Andersen          13 stages; sampling rate: 3 litre/min;           environmental
                                                             Cut points: 13-0.08 µm

                                                             9 stages; sampling rate: 7 litre/min;            environmental
                                                             Cut points: 18, 11, 4.4, 2.65, 1.7, 0.95, 
                                                             0.53, 0.32, 0.16 µm

                                                             8 stages; sampling rate 28 litre/min;            environmental
                                                             Cut points 10-0.4 µm

                                                             7 stages; sampling rate 28 litre/min;            environmental
                                                             Cut points 6, 4.6, 3.3, 2.2, 1.1, 0.7, 0.4 µm

                                                             6 stages; sampling rate 0.3-20 litre/min;        personal
                                                             Cut points 0.5-20 µm

                                                             5 stages; sampling rate 1132 litre/min;          environmental
                                                             Cut points 7.2, 3, 1.5, 0.95, 0.49 µm

                                                             4, 6 or 8 stages; flowrate 2 litre/min           personal
                                                             Cut points 20-0.6 with 8 stages, 10-0.6 
                                                             µm with 6, and 20-3.5 with 4

                                                             Radial slot impactor. 6, 8 or 10 stages with     environmental
                                                             an optional cyclone

                                                                                                                             

    Table 23.  (continued)

                                                                                                                             
                                   Manufacturer              Comments                                         Application
                                                                                                                             

                                   Hauke KG                  Sampling rate 30 litre/min                       environmental
                                                             Cut points below 0.1 µm

                                   in-Tox Products           4 models of 7 stages each.                       environmental
                                                             Cut points 3.1-0.33 µm @ 0.1 litre/min;
                                                             4.5-0.32 µm @ 1 litre/min; 5-0.25 µm @ 
                                                             2 litre/min; 5-0.0.5 µm @ 5 litre/min

                                   MSP                       8 stages, Sampling rate: 30 litre/min;           environmental
                                                             Cut points: 10, 5.62, 3.16, 1.78, 1, 
                                                             0.56, 0.316, 0.178, 0.1, 0.056 µm

      Virtual impactors            BGI                       3 virtual stages, flowrate: 30 litre/min;        environmental
                                                             Cut points 1.2, 4 and 14 µm

                                   Graseby Andersen          The dichotomous sampler fractions the            environmental
                                                             particles in 2 sizes: 10 and 2.5 µm. 
                                                             Sampling rate 17.6 litre/min

                                   MSP                       Sampling rate 30 litre/min;                      environmental
                                                             Cut points below 0.1 µm

                                                             High-volume operates at 1130 litre/min           environmental
                                                             Cut point 2.5 µm

      Cyclones                     Mine Safety Appliances    Measures respirable particles with a pre-cut     personal
                                                             diameter of 3.5 µm @ 2 litre/min

                                   Sensidyne                 Measures respirable particles in ambient         environmental
                                                             air @ 240 litre/min

                                                                                                                             

    Table 23.  (continued)

                                                                                                                             
                                   Manufacturer              Comments                                         Application
                                                                                                                             

                                                             Measures respirable particles in ambient         environmental
                                                             air @ 9 litre/min

                                                             Measures respirable particles @                  personal
                                                             1.7 litre/min

                                   SKC                       Measures respirable particles with a cut         personal
                                                             point of 3.5 µm @ 1.9 litre/min

      Elutriators                  Casella                   Horizontal elutriator that retains particles     environmental
                                                             with a cut point of 3.5 µm at a flowrate 
                                                             of 50 litre/min

                                                             Horizontal elutriator that retains particles     personal
                                                             with a cut point of 3.5 µm at a flowrate of 
                                                             2.5 litre/min
                                                                                                                             

    LOD: Level of detection

    Table 24.  Air sampling methods for reactive gases

                                                                                                                         
                          Manufacturer                   Comments                                       Application
                                                                                                                         

    Hydrogen sulfide/sulfur dioxide/ammonia/chloride

    Continuous
      Electrochemical     Devco Engineering

                          CEA Instruments
                                                         All based on conductivity change in            environmental
                          Sensidyne                      water due to ionization of gas

                          Teledyne

                          Bacharah

      Colorimetric        Phillips Electronics           Based on the reaction of the gas with          environmental
                          Instruments                    the reagent to produce a coloured 
                                                         product
                                                         Each compound has a specific reagent 
                          CEA Instruments                for its detection

      Potentiometric      AIM                            Conductivity of the reagent changes in         environmental
                                                         proportion to the concentration of the 
                          Calibrated Instruments         gas being sampled and is measured 
                                                         by an electrode
                          Eitel Manufacturing

      UV and visible      Barringer Research             Based on the correlation with the              environmental
      light photometers                                  absorption spectra of SO2 in the UV
                                                         Sensitivity 2 ppm

      UV and visible      Beckman Instruments            Based on the fluorescence of SO2               environmental
      light photometers                                  under UV light
                                                                                                                         

    Table 24.  (continued)

                                                                                                                         
                          Manufacturer                   Comments                                       Application
                                                                                                                         

                          Rosemount Analytical           For SO2 uses a non-dispersive UV               environmental
                                                         "transflectance" analysis

                          Columbia Scientific            Uses a continuous UV source of high            environmental
                                                         intensity to detect SO2

    Passive
      Solid adsorbents    Barneby Cheney                 Large number of chemicals are 
                                                         efficiently collected under a wide 
                                                         variety of conditions
                          Columbia Scientific            Choice of adsorbent is designed to 
                          Instruments                    maximize collection efficiency while 
                                                         retaining low selectivity
                                                         Approximately 50 sorbent types are 
                          Draeger                        available; some are chemically treated         personal
                                                         to facilitate their collection 
                          Fischer Scientific             properties
                                                         Most tubes contain a primary sorbent           personal
                                                         section and a backup bed that is 
                          Perkin Elmer                   used to indicate breakthrough

                          3M                             Require laboratory analysis                    

                          SKC

                          Supelco

                          Westvaco

                                                                                                                         

    Table 24.  (continued)

                                                                                                                         
                          Manufacturer                   Comments                                       Application
                                                                                                                         

    Active
      Annular denuders    URG                            Different models can be used to collect        personal/
                                                         only gases or gases and particles/aerosols.    environmental
                                                         Especially used for acid aerosols (SO2, 
                                                         H2SO4, HNO3, (NH4)2SO4, NH4HSO4, NH4NO3)
                                                                                                                         
    

    7.2.2  Aerosols

         At present, active sampling is the only feasible way to perform
    exposure assessments on particulates directly. Active particle
    samplers operate by drawing aerosols into a sensor or on to a
    collection surface (e.g., a filter) by means of a pump (Hinds, 1982;
    Lehtimäki & Willeke 1993). Large stationary samplers that operate with
    a standard flow rate of approximately 1000 litre/min are available
    commercially and are useful for collecting large sample volumes. Small
    stationary samplers that operate with flow rates in the range of 1-10
    litre/min are also commercially available. Both sizes are available in
    configurations that allow for sampling of total suspended particulate
    matter (i.e., not size separated) or specific size fractions (e.g.,
    PM2.5 or PM10). Personal aerosol samplers that allow collection of
    total inhalable particulate matter of specific size fractions are also
    available.

         The cyclone and, particularly, the impactor are the two most
    commonly used size preselectors. Cyclones can collect suspended
    particulate matter of various sizes depending on the geometry of the
    cyclone and the flow rate. It operates on the principle of centrifugal
    forces that drive particles in the direction of the outer wall of the
    cyclone (Hinds, 1982). Particles with aerodynamic diameter greater
    than the cut-point of the cyclone impact upon the wall and/or the
    bottom of the cyclone. Particles with aerodynamic diameter less than
    the cut-point remain in the air stream and are collected on a filter
    downstream.

         Impactors rely on inertial forces to separate particles based on
    aerodynamic diameter. Air is accelerated through a nozzle or jet and
    then forced to make a 90° turn around an impaction plate before
    passing through a filter and exiting the sampler. Depending on their
    size, particles suspended in the air stream pass through the
    acceleration nozzle and then either remain entrained in the flow or
    collide and are retained on the impaction plate. The cut-point of an
    impactor is determined by the flow rate, jet size and shape (e.g., the
    distance between the jet and the impaction surface) (Pastuska, 1988;
    Lehtimäki & Willeke, 1993). The air flow rate must be calibrated
    carefully because correct size selection depends largely on precise
    flow rates.

         Filters are made either from fibre mats of glass, cellulose or
    quartz or from synthetic membranes (e.g., Teflon). The selection of
    appropriate filters depends on the pump, filter static pressure,
    collection efficiency, extraction and analytical requirements, and the
    potential for sampling artefacts. Filter mass is determined by
    weighing the filter under controlled temperature and humidity
    conditions before and after use following a conditioning period of at
    least 24 h at those same conditions. The collected mass can be
    extracted and analysed for chemical composition. The extraction and
    analysis procedures used depend on the analytes of interest. A recent
    summary of methods for extraction and analysis of components of
    particulate matter may be found in Koutrakis & Sioutas (1996).

    7.2.3  Semivolatile compounds

         For airborne contaminants that are present in both the particle
    and the vapour phase at typical environmental conditions, it is
    necessary to use a combination of sampling methods. The most common
    approach consists of an aerosol sampling inlet (with or without size
    preselector) followed by a sorbent cartridge or tube. Examples of such
    contaminants include airborne PAHs, pesticides, polychlorinated
    biphenyls (PCBs), dioxins and furans. Semivolatile sampling systems
    are commercially available for personal air monitoring. Extraction and
    analysis of these samples are done separately for the particle and
    vapour phase and then the results are combined to provide a total
    concentration. An introduction to sampling and analysis methods for
    VOCs in air may be found in Binkova et al. (1995), Wallace & Hites
    (1996), Wallace et al. (1996) and Simonich & Hites (1997).

    7.2.4  Reactive gas monitoring

         Certain gases present in air may react with chemicals present in
    particles. For example, sulfuric acid particles collected on filters
    can be neutralized by the ammonia gas present in the sample or air
    stream. The preferred sampling approach to avoid this is to use a
    denuder to remove the reactive gas before it reaches the downstream
    filter. In the case of sulfuric acid monitoring, a citric-acid-coated
    denuder is used to remove the ammonia gas. Small denuder systems are
    commercially available for personal monitors. Denuder technologies are
    described in Lodge (1988) and Koutrakis & Sioutas (1996).

    7.3  Water

         The sampling and analysis of drinking-water characterizes the
    extent to which this carrier medium represents a source of specific
    chemical exposure. Contaminated drinking-water supplies contribute to
    the human intake of numerous chemical contaminants, including heavy
    metals, fertilizers, pesticides, aromatic hydrocarbons and
    organohalogens, among others. In some cases, drinking-water may be the
    primary source of human exposure. Chemical pollutants in water may
    originate from one or more of a myriad of sources, as summarized in
    Table 25. In the selection of measurement and sampling methods, it is
    important to consider raw water sources, water treatment processes,
    and distribution and service systems, all of which can either reduce
    or increase the contaminant concentrations in drinking-water.

         Samples collected at the end of the distribution system provide a
    better measure of potential exposure to individuals than samples
    collected at the source prior to any removal or treatment that might
    take place. Numerous texts detail sampling and analytical techniques
    specific to drinking-water, and these methods can be used to develop
    comprehensive exposure assessment protocols (UNEP/WHO, 1986; WHO,
    1992,1993).


        Table 25. Origins of chemicals commonly occurring in drinking-water (Hickman et al., 1982)

                                                                                                                            

    Substances affecting the source (raw water)
    "Naturally occurring"                            Leached from geological formation (e.g., calcium, heavy metals)
                                                     Derived from soil and sediments

    Pollutants derived from point sources            Domestic sewage treatment (e.g., nitriloacetic acid)
                                                     Industrial effluents (e.g., synthetic organics, metals, cyanide)
                                                     Landfill waste disposal (e.g., metals, synthetic organics)

    Pollutants derived from non-point sources        Agricultural run-off (e.g., fertilizers, pesticides)
                                                     Urban run-off (e.g., salt, PAHs)
                                                     Atmospheric fall-out (e.g., PAHs, chlorinated organics, heavy metals)

    Substances resulting from treatment
    Substances formed during disinfection            Trihalomethanes, chlorophenols

    Treatment chemicals                              Chloramines, fluorides

    Treatment chemical impurities                    Acrylamide monomer, carbon tetrachloride

    Substances arising from the distribution 
    and service systems
    Contaminants arising from contact with           Lead, vinyl chloride monomer and asbestos fibres from piping, 
    construction material and protective coatings    cadmium from fittings, PAHs from coal tar linings

    Substances arising from point-of-use devices     Sodium, silver
                                                                                                                            
    

         In developing countries it is quite common for individuals not to
    have access to treated water from distribution systems, so analysing
    water quality solely from distribution systems may not provide a true
    reflection of exposure. Even if drinking-water is obtained from piped
    supplies, it may not provide an adequate indication of exposure as
    many individuals are forced to store water after collection, when
    gross contamination may occur. In some areas of the world, run-off
    water is routinely collected from roofs for drinking and cooking
    needs. Dustfall attributable to traffic, industry, or construction may
    contribute to variable (potentially high) pollutant concentrations in
    this source.

         Exposure to contaminants in water is not limited to oral routes.
    For instance, disinfection by-products and radon gas dissolved in
    groundwater may be released into an indoor atmosphere providing an
    inhalation route. Heating water also releases dissolved VOCs. Exposure
    to contaminants may also occur through inhalation of aerosols from
    irrigation sprays. During other water-based activities (e.g.,
    swimming, showering and bathing), other contaminants may be absorbed
    via a dermal (percutaneous) route. Although the contribution of
    non-oral routes is usually much less than that of oral routes, these
    pathways should not be overlooked in the selection of measurement
    methods to assess exposure. Methods for modelling exposure through
    these pathways are discussed in Chapter 6.

    7.3.1  Factors influencing water quality

         In order to select appropriate measurement and monitoring
    methods, it is important to understand the following factors that
    influence the quality of the water being sampled, and the resultant
    exposure:

    *  treatment systems

    *  distribution networks

    *  storage practices

    *  spatial and temporal variations

    *  climatic and seasonal changes.

         Water treatment encompasses a variety of processes, ranging from
    simple screening and filtration to multi-step purification. The latter
    includes methods for coagulation, aeration, de-aeration, colour
    removal, softening, disinfection, fluoridation, stabilization and
    demineralization. Some of these steps constitute "removal", and others
    involve the "addition" of treatment chemicals to mitigate the hazards
    of contaminants in water. A list of chemical additives typically used
    in water treatment systems is shown in Table 26. The reaction of
    treatment chemicals with other substances present in raw (untreated)
    water often results in the generation of intermediate reaction
    products with adverse health significance. For instance, chlorine,

    accepted worldwide for disinfection and oxidation, results in the
    formation of disinfection by-products such as trihalomethanes (e.g.,
    chloroform).


    Table 26.  Water treatment chemicals

                                                                          
    Activated alumina             Sodium bicarbonate
    Aluminum sulfate              Sodium calcium magnesium polyphosphate
    Ammonia                            (glassy)
    Ammonium hydroxide            Sodium carbonate
    Bentonite clay                Sodium chlorite
    Calcium hydroxide             Sodium fluoride
    Calcium hypochlorite          Sodium hydroxide
    Calcium oxide                 Sodium metabisulfite
    Carbon (activated, granular,  Sodium polyphosphate (glassy)
      and powder)                 Sodium silicate
    Carbon dioxide                Sodium siliconfluoride
    Chlorine                      Sodium tripolyphosphate
    Ferric chloride               Sodium zinc polyphosphate (glassy)
    Ferric sulfate                Sodium zinc potassium polyphosphate
    Ferrous sulfate                 (glassy)
    Fluosilicic acid              Sulfur dioxide
    Potassium permanganate        Sulfuric acid
    Sodium aluminate              Tetrasodium pyrophosphate
                                                                          


         Distribution networks constitute another potential source of
    chemical contaminants in drinking-water. The materials used in
    distribution networks may serve as a pollutant source by leaching into
    the water over time. Some examples include lead from lead-containing
    solders and pipes, asbestos fibres from the surface of asbestos-cement
    pipes and cadmium from metallic fittings. Other contaminants include
    PAHs from coal-tar-based sealants, plasticizers, stabilizers and
    solvents used in the manufacture of plastic pipes.

         Water sources experience considerable variations in quality over
    time and geographic location. The quality of river water may change
    rapidly during heavy storms, melting snows and droughts. The quality
    of water in lakes may be affected by climate, season, location or some
    combination thereof. Groundwater historically has enjoyed the most
    consistent quality, with relatively constant composition. However, the
    vulnerability of groundwater to contamination is gaining widespread
    attention, with particular emphasis on synthetic organic substances,
    surface impoundments, landfills, agriculture, leaks and spills, land
    disposal of wastewater, septic tanks and the petroleum/mining
    production industries.

    7.3.2  Water quality monitoring strategies

         There are numerous considerations in the design of a monitoring
    and measurement strategy for water quality assessment. The
    International Organization for Standardization (ISO) has provided
    guidance on a number of issues related to sampling strategies for
    water quality assessment (Table 27). A sound monitoring methodology
    must be followed by the appropriate sample storage and transportation,
    to minimize changes in sample composition. Losses can occur due to
    several physical, chemical and biological changes, such as ion
    exchange, adsorption with the container material, oxidation to
    precipitated forms, loss of volatiles to the vapour space and
    biochemical conversions. For contaminants at low source
    concentrations, these changes can introduce significant errors in the
    analytical results.


    Table 27.  ISO standards of water quality giving guidance on sampling

                                                                             
    ISO standard     Title (water quality)
    number
                                                                             

    5667-1: 1980     Sampling - Part 1: Guidance on the design of sampling 
                     programmes
    5667-2: 1982     Sampling - Part 2: Guidance on sampling techniques
    5667-3: 1985     Sampling - Part 3: Guidance on the preservation and 
                     handling of samples
    5667-4: 1987     Sampling - Part 4: Guidance on sampling from lakes, 
                     natural and man-made
    5667-5: 1985     Sampling - Part 5: Guidance on sampling of 
                     drinking-water and water used for food and beverage 
                     processing
    5667-6: 1985     Sampling - Part 6: Guidance on sampling of rivers and 
                     streams
                                                                             


         The design of a water monitoring programme would be incomplete
    without consideration of the demographic and socioeconomic
    characteristics, and also an understanding of the historical
    development, of the potentially exposed community. The evolution of
    materials used in distribution systems changes the profiles of
    pollutants requiring measurement. Cultural and socio-economic factors
    affect usage patterns, which in turn influence the extent of exposure
    to contaminants in drinking-water.

         In order to ensure the representativeness and validity of water
    samples, sampling techniques must be carefully selected (WHO, 1992,
    1993). The first step in the design of a sampling programme is to
    develop concise objectives, accounting for

    *  the nature of the substance to be measured

    *  point of exposure

    *  the duration of time over which measurements will be taken.

         The type and magnitude of spatial and temporal variations in the
    concentration of water constituents will depend upon both their
    sources and their behaviour in the distribution and service systems.

    Substances can be classified into two main types: 

    *   Type 1. Substances whose concentration is unlikely to vary during
       distribution. The concentration of these substances in the
       distribution system is largely governed by the concentration in the
       water going into the supply, and the substances do not undergo any
       reaction in the distribution system. Examples of such substances
       are arsenic, chloride, fluoride, hardness, pesticides, sodium and
       total dissolved solids.

    *   Type 2. Substances whose concentrations may vary during
       distribution. These include

       -  substances whose concentration during distribution is dependent
          mainly on the concentration in the water going into the supply,
          but which may participate in reactions (which change the
          concentration) within the distribution system. Examples are
          aluminium, chloroform, iron, manganese and hydrogen ion (pH).

       -  substances for which the distribution system provides the main
          source, such as benzo [a]pyrene, copper, lead and zinc.

         This classification applies only to piped water supplies. In all
    other types of supply, water constituents should be regarded as type 1
    substances. The same substance may belong to different classes in
    different distribution systems.

    7.3.3  Sample collection

         The location, frequency and time of sampling is strongly
    dependent on the spatial and temporal variations for the particular
    pollutant of interest. There are many different methods to collect
    water samples and measure contaminant concentrations. The choice of a
    particular technique can have a profound effect on the analytical
    results. Some conventional measurement methods are briefly described
    below:

    *   Grab samples represent a "snapshot" of a situation at a
       particular time and place. Using samples taken at intervals and
       analysed individually, this method can characterize variations in
       source composition.

    *   First-draw (static) samples are collections immediately following
       a stagnation period (e.g., overnight). This reflects the influence
       of domestic plumbing on the inorganic content of water quality.

    *   Flushed samples are taken after taps have been run for a
       sufficient length of time to eliminate stagnant water.

    *   Composite samples involve regular sampling, usually over a 24-h
       period, followed by pooling of samples and analysis of the
       composite. This integrated method overcomes the disadvantages
       inherent in first-draw sampling. Time-composite samples approximate
       the potential exposure to drinking-water contaminants.

    7.4  Assessing exposures through food

         Exposure to chemical compounds in food can be measured directly
    by analysing duplicate diets or indirectly by analysing foods or total
    diets, matching food consumption data with information of chemical
    concentration in the foods or, for certain chemicals, estimating the
    total amount of the chemical available divided by the population of
    concern (called  per capita estimates). The consumption of water and
    the resulting exposure should also be determined if appropriate
    (FAO/WHO, 1997). The estimation of exposure to food chemicals is a
    complex activity and no single approach is suited to all
    circumstances. The method chosen depends on the information available,
    the population group of concern, whether acute or chronic effects of
    the chemical are being assessed, and the intended use of the result
    (Rees & Tennant, 1993). The Intake Assessment Group which has been
    added to the Joint FAO/WHO Expert Committee on Food Additives is also
    examining other means of evaluating dietary exposure assessments for
    food additives and contaminants.

         Direct approaches tend to consider samples of food as actually
    consumed because the method by which food is prepared for consumption
    (e.g., washing, peeling, cooking and commercial processing) can
    influence contaminant residue levels. For example, malathion
    concentrations were found to decrease by 99% when raw tomatoes were
    processed into canned tomatoes (Elkins, 1989). In contrast,
    concentrations of ethylenethiourea, a carcinogenic degradation product
    of maneb (manganese ethylene bisdithiocarbamate), rose 94% when turnip
    greens were washed, blanched, frozen and subsequently sautéed (Elkins,
    1989; Houeto et al., 1995). Although cooking may lead to a reduction
    in the lead content of vegetables, in areas where the lead
    concentrations in drinking-water are higher than average (e.g., due to
    lead pipes), cooking water can be a significant source of lead intake
    (UK MAFF, 1989).Therefore, preparation and processing can alter
    contaminant levels present in foods, or introduce new contaminants.
    For these reasons, the concentration of the target analyte in
     ready-to-eat foods is the most useful measure for purposes of
    dietary exposure assessment.

    7.4.1  Duplicate diet surveys

         Duplicate diet surveys are particularly useful because they
    reflect the range of preparation habits of the study population. These
    studies require that respondents save a serving of each meal or
    components of each meal and store them until collection by the
    research team. Following collection, the food is composited over
    predetermined time intervals (e.g., by meal or by day) and analysed
    for the target analytes. In duplicate diet studies, logistic and cost
    constraints typically require that foods be composited. The principal
    disadvantage of composite samples is that they do not allow for
    identification of the contribution of individual foods to total
    dietary exposure. A high degree of respondent burden is associated
    with duplicate diet studies, so they are not conducive to assessing
    chronic dietary exposures and may underestimate intake. Such
    approaches are only suitable for chemicals that can be analysed
    accurately, so direct diet methods are not traditionally used for
    assessing food additives exposure, for example. A summary of dietary
    exposure assessments for chemical contaminants in food using the
    duplicate diet performed worldwide may be found in Thomas et al.
    (1997).

         There are many indirect methods for estimating exposure to food
    chemicals because there are a variety of ways to collect consumption
    data, to express residue levels in the foods concerned (for example,
    legislative levels, manufacturer or industry use levels, predicted,
    proposed or analysed levels or any combination of these) and there are
    several approaches which can be used to combine the information to
    assess exposure (Rees & Tennant, 1994). Some methods are better than
    others, depending on the chemical; for example several countries have
    found it useful to assess food additive exposure by using  per 
     capita methods (Ito, 1993). More information on these indirect
    methods is given below, but the reader is strongly advised to refer to
    more comprehensive documents on dietary survey methodology and dietary
    exposure assessment approaches (WHO, 1985a, 1997c; FAO/WHO, 1995a,b,
    1996, 1997).

    7.4.2  Market basket or total diet surveys

         Market basket or total diet surveys utilize food chemical
    concentrations measured in ready-to-eat foods prepared in the
    laboratory linked to model diets derived from food consumption data
    and standard recipe preparation for large populations, households or
    individuals. Food products or food groups selected for sampling and
    analysis are generally intended to be representative of those most
    commonly consumed by the population of interest. Total diet studies
    have been carried out since the 1960s in many countries. Market basket
    surveys are often employed by regulatory agencies charged with
    ensuring and monitoring the safety of a national food supply (FAO/WHO,
    1995b). Initially this purpose was to estimate background exposures of
    the population to pesticides residues and radioactive contaminants.
    The emphasis has shifted from pesticides to toxic metals and more

    recently has included a variety of trace elements and organic
    contaminants.

         For example, the US FDA Total Diet Study (US TDS) is a market
    basket survey based on heavy metal and pesticide data measured in
    samples of 234 different ready-to-eat food products selected to be
    representative of over 4000 foods common in the diet of residents in
    the USA, and the results of national food consumption surveys
    (Pennington, 1992). However, more commonly total diet (market basket)
    studies consider smaller food groups rather than individual foods (UK
    MAFF, 1985). The main advantage of the total diet (market basket)
    approach for estimating exposure is the ability to monitor trends
    without burdening study participants. The total diet approach allows
    data from separate studies of food consumption and contaminant
    residues to be combined (e.g., Tomerlin et al., 1996). Moreover, this
    approach allows analytical chemistry resources to be directed to the
    foods that are most likely to yield the greatest exposure (e.g., the
    foods consumed in greatest amounts and foods that are likely to
    contain the highest residue concentrations). Such foods may be
    indicated by information available from existing data such as the
    GEMS/Food (WHO, 1978, 1997c) and the US TDS (Pennington & Gunderson,
    1987).

         However, this method cannot be used for all contaminants. This is
    because the analysis of food groups may be too expensive for some
    contaminants and may not be feasible for others. Analytical methods
    may not be sufficiently reliable, the limit of detection may be too
    high or the grouping of the foods (compositing) may decrease the
    likelihood of finding the source of the contaminant. Analysis of
    individual food products affords a detailed examination of contaminant
    levels in specific commodities -- either raw, processed or prepared.
    Sampling may be designed to characterize geographic and temporal
    variability of contaminant levels that may be a result of varying
    application rates of pesticides, natural levels of elements (e.g.,
    heavy metals), climate and other factors. In addition, samples can be
    collected at all steps in the process from field to consumer thereby
    providing insight into the sources and fate of contaminants in food.

         Further information on the strengthens and limitations of each of
    the approaches described above have been published in the
    comprehensive  Guidelines for the Study of Dietary Intake of 
     Chemical Contaminants (WHO, 1985a).

    7.4.3  Food consumption

         The FAO/WHO Consultation on Food Consumption and Exposure
    Assessment of Chemicals (called Exposure Consultation) reviewed
    current methodology for food additives, contaminants, pesticides,
    veterinary drugs and nutrients. The Exposure Consultation agreed to
    expand and revise the regional diets presently used by the GEMS/Food
    for pesticides and recommended that this consumption data can be used
    for estimate dietary exposure to certain other chemicals. The regional
    diets will be based on 1990-1994 FAO Food Balance Sheets which reflect

    a country's amount of raw commodities for consumption, and may not
    necessarily refer to foods in the forms people consume them. Waste at
    the household or individual level is not usually considered.

         Major methods for determining food consumption at the national
    levels were identified as population-based, household-based and
    individual-based. The report from a FAO/WHO consultation on the
    preparation and use of food-based guidelines (FAO/WHO, 1996) gives
    more information on food consumption study designs. The Exposure
    Consultation supported the concept that an improvement in dietary
    exposure assessments can be achieved by refining any combination of
    the contributing elements: food consumption data, food chemical
    concentration data or the method used to combine the two. This allows
    the risk assessor a greater flexibility in selecting cost-effective
    approaches to refine dietary exposure assessments using the resources
    available (WHO, 1997c).

         The five basic approaches discussed by the Exposure consultation
    for describing the diet of individual people are:

    *  food record/diary survey

    *  24-h recall

    *  food frequency questionnaire

    *  meal-based diet history

    *  food habit questionnaire (WHO, 1997c).

         The 24-h recall is a widely used dietary assessment method and is
    utilized in many exposure-related studies including the National
    Health and Nutrition Examination Survey conducted by the US Centers
    for Disease Control and Prevention (Witschi, 1990).

    7.4.3.1  Food diaries

         Food diaries are detailed descriptions of types and amounts of
    foods and beverages consumed, meal by meal, over a prescribed period,
    usually 3-7 days. Food diaries and recalls may be presented in
    numerous formats or combined with food models and weighing procedures
    to characterize serving size more accurately; however, regardless of
    the specific details, dietary recording places a substantial burden on
    the subject (Witschi, 1990).

    7.4.3.2  24-h recall

         The short-term nature of the 24-h recall and the facility to
    consider meal occasions or daily consumption from diary surveys make
    this method ideal for assessing exposure to substances that can give
    rise to acute health effects, such as the cholinesterase-inhibiting
    organophosphate and carbamate pesticides. Diary methods may be used
    for assessment of long-term exposure but the underlying assumption is

    that the food consumption is representative of usual habits.
    Probabilistic approaches can be useful to predict consumption and
    resulting exposure over longer periods of time.

    7.4.3.3  Food frequency questionnaires

         Food frequency questionnaires (FFQs) are a standard tool for
    characterizing food intake over extended periods of time. A food
    frequency questionnaire consists of two basic components: a list of
    foods and a frequency response section for respondents to indicate how
    often a specific serving size of each food is consumed (Table 28). The
    underlying principle of the food frequency approach is that average
    long-term diet, for example, intake over weeks, months or years, is
    important rather than intake on a few specific days. This may not be
    true for all contaminant-health effect combinations (e.g., acute and
    reversible effects such as cholinesterase inhibition); however, it is
    reasonable in the context of assessing health effects that may be
    caused by cumulative exposure, such as cancer, or reproductive and
    developmental effects that may follow a threshold dose-response curve.
    Some FFQs include questions on usual food preparation methods,
    trimming of meats, use of dietary supplements and identification of
    the most common type or brand consumed. FFQs can be used to rank
    individuals by exposure to selected chemicals. Although FFQs are not
    designed to measure absolute exposure, the method may be more accurate
    than other methods for estimating average exposure to chemicals having
    large day-to-day variability and for which there are relatively few
    food sources. FFQs have several disadvantages too: specifically, they
    are less reliable in estimating consumption of rarely consumed foods
    and the food lists are often designed to assess nutrients and may
    require substantial revision to assess chemical exposures.

    7.4.3.4  Meal-based diet history

         Meal-based diet history methods are designed to assess usual
    individual food consumption. It consists of a detailed listing of the
    types of foods and beverages commonly consumed at each meal over a
    defined time period which is often a "typical week".

    7.4.3.5  Food habit questionnaires

         Food habit questionnaires are designed to collect either general
    or specific types of information, such as food perception and belief,
    food likes and dislikes, methods of preparing foods, use of dietary
    supplements and social setting surrounding eating occasions. This type
    of information is frequently considered with other methods but may be
    used on its own.

         Although the last two methods are seldom used in dietary exposure
    assessments they can contribute very useful background information and
    may be the only information for specific population group issues
    (e.g., organic food consumption by vegetarians). They can be targeted
    to answer specific questions or prioritize issues of concern and
    provide a cost-effective tool for the risk assessor.


        Table 28.  An example of food listing and frequency response options of an FFQ

                                                                                                                               
    For each food listed, fill in the          Average use of the last 3 months
    circle indicating how often, on                                                                                            
    average, you have used the amount               Per month                Per week                    Per day
    specified, during the past 3 months                                                                                        
                                               Never or    1-3      1        2-4      5-6      1        2-3      4-5      6+
                                               less than 
                                               once
                                                                                                                               

    DAIRY FOODS    Skim or low-fat milk 
                   (8 oz glass)                0           0        0        0        0        0        0        0        0

                   Whole milk (8 oz glass)     0           0        0        0        0        0        0        0        0

                   Sherbet or ice milk 
                   (1/2 cup)                   0           0        0        0        0        0        0        0        0

                   Ice cream (1/2 cup)         0           0        0        0        0        0        0        0        0
                                                                                                                               
    


         Food consumption data is often collected for nutritional or
    economic purposes, and foods may not be described in the detail
    required for exposure assessment (e.g., fish consumption may be
    recorded but the contaminant of interest may be found primarily in
    fatty fish or fish caught in a particular location). There are number
    of difficulties using the different types of consumption data. A
    report from the European Commission provides a good summary of the
    practical problems in using consumption data to estimate dietary
    exposure (EC, 1997a).

    7.4.4  Contaminants in food

         The vast majority of food that is actually consumed has undergone
    some form of processing, ranging from simple washing to complete
    reconstitution, as it progresses from the producer to ultimately being
    ingested by a consumer (FAO/WHO, 1995b). Several factors can influence
    contaminant concentrations in foods that are ready to eat. These
    factors include those that may vary by season and/or geographic
    region, such as food source (e.g., homegrown, locally grown by a small
    producer, domestically grown by a mass producer and imported), and
    former or current application of pesticides (US NRC, 1993). The form
    in which food is consumed (e.g., raw apple, apple sauce, apple juice)
    can be very different in different subpopulations (e.g., adults,
    elderly or young children).

         Residue levels measured in raw agricultural commodities collected
    at the producer, processor or distribution level are unlikely to be an
    accurate reflection of contaminant concentration in food as actually
    consumed. With the exception of the GEMS/Food, which collects
    contaminant and pesticide residue data from member countries, there
    are no centrally coordinated reference databases for other food
    chemicals in foods. Potential data sources at the national level may
    include supervised trial data, government monitoring and surveillance
    data (Pennington & Gunderson, 1987), national food composition
    databases (nutrients) and industry funded surveys. A number of
    analytical methods for contaminants in food have been published by the
    US FDA, EOAC and US EPA (e.g., US FDA 1997a,b). Different approaches
    have been used for calculating exposure when the contaminant
    concentrations fall below the detection limit (e.g., assuming the
    concentration is zero or some percent of the detection limit).

    7.5  Summary

         This chapter has introduced available sampling methodology for
    chemicals in air, water, and food. Common to the selection of these
    methods are considerations of detection limits, interferences, ease of
    operation and cost. Personal, microenvironmental and ambient air
    sampling methods are available for monitoring gases and vapours, both
    passively and actively, aerosols, SVOCs and reactive gases.

         Sampling considerations for assessing water quality are numerous.
    An important consideration is that exposure to contaminants is not
    limited to oral routes and that not all individuals have access to
    treated water from distribution systems. Guidance for sampling and
    monitoring programmes is provided.

         There are a number of methods for measuring estimating food
    consumption and contamination. The method chosen will depend on the
    information available, the population group of concern, whether acute
    or chronic effects of the chemical are being expressed, the intended
    use of the results and available resources. The reader is strongly
    advised to refer to more comprehensive documents on dietary survey
    methodology and dietary exposure assessment approaches.

    8.  MEASURING HUMAN EXPOSURE TO CHEMICAL CONTAMINANTS IN SOIL AND 
        SETTLED DUST

    8.1  Introduction

          This chapter is intended to provide the reader with important
    concepts and a basic understanding of soil and settled dust sampling
    so that effective sampling strategies can be designed to meet specific
    research needs. Choices in sampling methods, sampling locations,
    sampling areas and the sampling time of the sample collection may be
    particularly important when the results are used for exposure
    assessment purposes. For these methods to be used successfully, it is
    important that the investigators understand the basic concepts behind
    collecting soil and settled dust and the limitations of different
    methods and strategies. Because this field of research is currently
    evolving rapidly, it is recommended that researchers consult the
    literature for new and complete information before designing a study
    to measure toxic metals, pesticides, PAHs, other products of
    incomplete combustion, fibres and biological matter. The most
    appropriate method for sampling soil and settled dust depends on the
    living conditions of the study population and the target contaminants.
    The information in this chapter is therefore intended to provide
    general guidance on approaches that might be taken.

          Soil is a mixture of air, water, mineral and organic components
    (Horne, 1978). The relative mix of these components determines to a
    large extent the capacity of a soil for containing chemical
    contaminants and the potential for it to be an important source of
    exposure. Settled dust, which may be found outdoors or indoors, is
    often a complex mixture of material from several sources. Outdoor
    settled dust is material deposited on roadways, streets and other
    paved surfaces. Indoor settled dust (house dust) is material deposited
    on indoor surfaces such as floors, carpets and furniture. Chemical
    contaminants present in indoor dust can originate from activities in
    the home or can be tracked into the home from road dust, soil or work
    sites (US EPA, 1991). Material present in soil, outdoor dust and
    indoor dust may include clay, sand, bacteria, viruses, allergens,
    products of incomplete combustion, environmental tobacco smoke, heavy
    metals, pesticides, asbestos fibres, paint fragments, solvents, flame
    retardants, cleaners, and residues from synthetic fibres, building
    products and many other materials and pollutants (Robert & Dickey,
    1995).

          Unintentional ingestion of house dust, particularly for children,
    may be a significant contributor to the total human exposure to many
    potentially toxic substances, depending on personal living conditions
    and frequency of contact with this media. Because children spend more
    time in contact with soil and indoor surfaces than adults and have a
    greater dose given the same exposure, these exposure pathways are
    particularly relevant to children. For example, it is likely that
    children's lead exposure from settled dust is an important
    contribution to total lead exposure because of the past and present
    use of gasoline, lead-based paint on housing and steel structures, and

    airborne emissions from industrial point sources that settle in
    residential environments. In the USA, house dust is considered a major
    source of lead to most children (CDC, 1991; Lanphear & Roghmann,
    1997). Older homes are especially susceptible to lead dust exposure if
    paint is peeling or renovations are being done (Roberts et al., 1992).

          Soil and settled dust can be a significant source of exposure to
    numerous other toxicants in addition to lead, including pesticides and
    PAHs. Pesticides, although designed to degrade to different extents
    through natural environmental processes such as sun, rain and soil
    microbial activity, may accumulate in soil and dust and persist for
    long periods of time. Because of the lack of these external
    degradation processes, pesticides may be particularly persistent in
    indoor settled dust (Simcox et al., 1995). Studies have shown that in
    the general population in the USA the highest concentrations and
    largest number of pesticides are found in house dust as compared to
    soil, air and food (Whitmore et al., 1993; Lewis et al., 1994).
    Although many pesticides in house dust come from outdoor sources, many
    households use pesticides indoors. Because little or no training is
    provided for users of household pesticides, unnecessary exposures
    often occur. Pesticides often found in house dust include those used
    for control of insects; e.g., chlordane and heptachlor in homes
    treated for termites, pentachlorophenol and lindane in homes where
    wood preservatives had been used, and other harmful pesticides
    contained in flea and garden treatment (Roberts et al., 1992).

          Hazardous substances that originate at the worksite may also find
    their way (e.g., via clothes) into the homes of workers. The US
    National Institute for Occupational Safety and Health compiled a
    bibliography of more than 350 published and unpublished accounts of
    take-home, or "para-occupational" contamination worldwide (NIOSH,
    1994). The reports identified by NIOSH document the spread from
    workplace to home of toxic metals (lead, beryllium, cadmium and
    mercury), asbestos, and various other potentially hazardous
    substances. Settled dust was a major source of familial exposure in
    most of these studies.

    8.2  Selected sampling methods

    8.2.1  Soil

          Soil constitutes a potential exposure pathway through direct
    contact and ingestion or inhalation of resuspended soil particles.
    Children's activities make them more likely to be affected by such
    exposures. In addition, contaminated soil can be tracked inside homes,
    or may infiltrate indoors when resuspended. In either case, soil may
    become a component of settled indoor dust. There are no standard
    collection methods for soil sampling, as discussed later for settled
    dust (section 8.2.2). This limitation affects the ability to make
    comparisons of results from soil sampling across studies. However,
    information on soil contamination can provide insights into the
    relative importance of multimedia contaminants as they may affect
    exposure.

    8.2.1.1  Surface soil collection

          The most commonly used approaches make use of an auger or similar
    sampler such that a sample is defined by cross-sectional area and
    predesigned depth of the auger. Alternatively, a predetermined amount
    of surface soil may be scooped with a small trowel, with less precise
    definition of sampler depth. In either case, the sample is stored in a
    clean, inert container and transferred to the laboratory for analysis.

    8.2.1.2  Soil contact and intake measurements

          Skin contact has been measured by methods similar to those used
    for settled dust (e.g., self-adhesive labels, hand wipes), and
    controlled application followed by recovery of the fraction of
    deposited soil on the skin (Lepow et al., 1975; Roels et al., 1980;
    Que Hee et al., 1985). The amount of soil that adheres to the skin
    depends on a number of variables including soil properties (e.g.,
    water content, particle size, carbon content), region of the body and
    activity (Kissel et al., 1996). A number of studies have attempted to
    estimate soil ingestion based on hand adherence estimates and
    scenarios of activities, as well as analyses of soil tracers (e.g.,
    concentrations of aluminium, silicon or titanium) (e.g., Calabrese et
    al., 1989, 1990).

    8.2.2  Settled dust

          Although indoor dust is becoming recognized as a reservoir for
    many toxic substances and a potentially significant source of human
    exposure, there is no uniform standard for sampling settled dust. More
    than 15 methods have been described in the literature to date.
    Scientists do not yet agree either on the definition of settled dust
    or on the methods to measure it. This issue is further complicated by
    the fact that results from one settled dust sampling method may not be
    directly comparable to results from others. Even with these
    limitations, settled dust sampling methods have been used effectively
    and provided valuable insights into the total human exposure paradigm.

          Selected sampling methods are described below to give the reader
    an indication of the diversity of techniques available. The list is by
    no means exhaustive. Several of the methods described are simple to
    use and readily available to researchers worldwide. Brief descriptions
    of how to use the simpler methods are provided. Other methods require
    specialized equipment that is relatively expensive and may be
    difficult to obtain in some regions of the world. The methods are
    distinct from one another, but most fall into three categories: wipe,
    vacuum sampling and sedimentation methods. These methods are widely
    used for sampling settled dust indoors; however, in principle they may
    be applicable for outdoor settled dust as well. Bulk sample collection
    methods, such as sweeping, are not covered here. Key features of the
    various methods for collecting settled dust samples described in this
    chapter are summarized in Table 29.


        Table 29.  Comparison of features of different methods for collected settled dust samples

                                                                                                                              
    Feature               Common    HUD     Preweighed   Commercial   DVM       Rotary   HVS3       Sirchee    Sedimentation
                          wipe      wipe    sample       vacuum       vacuum    vacuum   vacuum     -
                                                                                                    Spittler 
                                                                                                    vacuum
                                                                                                                              

    Widely available      Yes       Yes     Yes          Yes          No        No       No         No         Yes

    Cost                  Low       Low     Low          Medium       High      High     High       Medium     Medium

    Simple method         Yes       Yes     Yes          Yes          Yes       Yes      No         No         Yes

    Loading               Yes       Yes     Yes          No           Yes       Yes      Yes        Yes        Yes

    Concentration         No        No      Yes          Yes          Yes       Yes      Yes        Yes        Yes

    Sieving possible      No        No      No           Yes          No        No       Yes        Yes        Yes

    Portable              Yes       Yes     Yes          No           Yes       Yes      No         Yes        Yes

    Samples small areas   Yes       Yes     Yes          Yes          Yes       Yes      No         Yes        Yes

    AC powered            No        No      No           Yes          Yes       Yes      Yes        Yes        No

    Size selective        No        No      No           No           Yes       No       Yes        No         No
                                                                                                                              
    

    8.2.2.1  Wipe sampling methods

          A common wipe sampling method uses  premoistened towelettes to
    wipe a measured area defined inside a template placed on the sampling
    surface (Vostal et al., 1974; US HUD, 1995). Typical sampling areas
    are in the range of 0.1 m2 and masking tape is commonly used as a
    template. The actual surface area inside the template is not critical
    as long as it is measured and recorded. However, sampling areas
    greater than 0.2 m2 are not recommended because larger areas cannot
    be wiped effectively with one towelette. This method has been used
    extensively in the USA to measure lead amounts in settled dust, but
    has also been used to ascertain levels of cadmium, chromium and
    arsenic, as well as many other metals and organic compounds.

          With the  HUD method, the person collecting the sample should
    wear a clean disposable glove on the hand that will come in contact
    with the towelette. To collect a sample, the surface inside the
    template is wiped with a towelette back and forth in vertical
    S-strokes. The exposed side of the towelette is then folded inward,
    exposing a clean portion, and the same area is wiped with horizontal
    S-strokes. The towelette is folded once more, again exposing a clean
    portion, and the area is wiped a final time with additional vertical
    S-strokes. The towelette is then folded, exposed side in, placed into
    a clean sealable plastic bag or container, and sent to a laboratory
    for analysis.

          Several researchers have used  preweighed wipe material, such as
    cotton gauze or filter paper, in order to determine the quantity of
    settled dust collected (Lepow et al., 1974; Stark et al., 1982;
    Rabinowitz et al., 1985; Levallois et al., 1991). The sampling
    material is then reweighed in a laboratory after sample collection.
    Theoretically, the weight of total dust collected can be calculated by
    subtraction, and toxicant concentration could be determined after
    analysis on a mass basis.

          An important issue that needs to be addressed when using the
    preweighed wipe methods is the potential loss of sampling material or
    dust during handling in the field or laboratory. Furthermore,
    Chavalitnitikul & Levin (1984) noted that filter paper tends to fall
    apart when rough surfaces are wiped. Loss of sampling material in the
    field would underestimate the amount of total dust collected when
    final weights are obtained, which would in turn overestimate the
    calculated mass concentration results. Because of water loss or gain,
    changes in humidity may also significantly affect the before and after
    weights of the samples. These potential sources of error must be
    carefully controlled to make the results from preweighed wipe methods
    reliable.

          A specially designed preweighed wipe sampling method has been
    developed to minimize the potential sources of error mentioned above.
    This method, known as the  Lioy-Weisel-Wainman (LWW) method, was
    developed to quantitatively measure the toxicant concentration (mg/g)

    and surface loading (mg/m2) of dust on flat surfaces (Lioy et al.,
    1993). The sampling device is not made from common materials and is at
    this time only available from the research group that developed it.

    8.2.2.2  Vacuum methods

          Many researchers have collected samples from commercial household
    vacuum cleaners, which are often referred to in the refereed
    literature as research dust samplers. Some researchers state that they
    sampled only the fine dust that settled to the bottom of the bag.
    (Kaye et al., 1987; Moffat, 1989; Davies et al., 1990; Thornton et
    al., 1990; Jensen, 1992). Other researchers modified their vacuum
    cleaners to hold filters (Diemel et al., 1981; Watt et al., 1983).

          A settled  dust vacuum method, commonly called the DVM, is
    constructed from conventional industrial hygiene sampling materials
    that are likely to be available to researchers worldwide (Que Hee et
    al., 1985). The sampler consists of a common personal air-monitoring
    pump, usually operated at 2.5-3.0 litres/min. Sampling areas with this
    method are typically 25 cm × 25 cm, and often take more than 5 min to
    sample completely. A three-sided template is sometimes used on bare
    floors to vacuum dust that has migrated to the walls. Sampling areas
    are covered three times with overlapping passes in the horizontal and
    vertical directions. Que Hee et al. (1985) state that the sampler was
    designed to collect only small dust particles that would most likely
    stick to a child's hands, not total lead on a surface. Therefore, the
    amount of dust collected by this method from a given surface is
    usually less than collected by other methods. This sampler has been
    used in numerous studies in the USA and elsewhere, and its use has
    provided considerable information linking lead in settled dust to lead
    in children (e.g., Bornschein et al., 1985).

          Researchers have also used laboratory  rotary vane vacuum pumps 
    connected to the same three-piece filter cassettes as used with the
    DVM described above, but with a much higher flow rate. The filter
    cassette is often used openface or with a wide diameter nozzle so
    sampling areas can be covered in fewer passes than required for the
    DVM, thus reducing the time spent collecting samples (Solomon &
    Hartford, 1976).

          Prpic-Majic et al. (1992) described another vacuum pump sampling
    method that used a prescreen at its nozzle entrance to prevent coarse
    particles and small objects from being collected on the membrane
    filter that served as the sampling surface. Total dust measurement was
    obtained from the dust particles that reached the membrane filter.
    There was no mention of potential loss of fine dust trapped in the
    prescreen, especially after it was loaded with fibres and debris.

          A sophisticated vacuum sampling device, called the  HVS3, was
    designed to make dust collection efficiency from different surface
    types as consistent as possible (ASTM, 1993). The HVS3 is a
    high-powered vacuum cleaner equipped with a nozzle that can be
    adjusted to a specific static pressure and air flow rate to allow for
    consistent dust collection. The sampler uses a cyclone to separate
    particles greater than about 5 mm from the air stream and collects
    them in a 250 ml sample bottle screwed into the bottom of the cyclone.
    Smaller particles are not collected. The HVS3 can collect large,
    representative samples of settled dust from indoor surfaces, such as
    rugs and bare floors, and dust from outdoor surfaces, such as streets,
    sidewalks, lawns and bare, packed dirt. However, it cannot be used to
    sample from small or uneven areas because of the large size of the
    device. The HVS3 has been used in numerous exposure assessment studies
    to measure toxic metals and pesticides in settled dust. The sampler is
    not made from standard materials and is therefore relatively expensive
    to buy. Interested readers should consult the ASTM standard method
    (D5438-93) for more information on the specifications and availability
    of the HVS3 sampling device (ASTM, 1993).

          Farfel et al. (1994) modified the HVS3 by using the same cyclone
    as in the HVS3 but with a commercially available handheld vacuum to
    make the device smaller and more portable. These authors also used
    flexible tubing as the pickup nozzle to allow small surfaces, such as
    windowsills, to be sampled. This modification, called the  BRM 
     method, does not allow control of either the sampling flow rate or
    the static pressure at the pickup nozzle. The ASTM standard method for
    the HVS3 does not apply to the BRM, except for its description of the
    cyclone.

          Another settled dust vacuum sampling method that has been used in
    several research studies, the  Sirchee-Spittler method, is a
    hand-held, battery-powered vacuum unit (Rinehart & Yanagisawa, 1993;
    Weitzman et al., 1993; Aschengrau et al., 1994). The sampler is simple
    to use, highly portable and can cover large areas in a short period of
    time. Unfortunately, there are not many Sirchee-Spittler sampling
    devices in service and its availability to researchers worldwide is
    therefore limited.

    8.2.2.3  Sedimentation methods

          Sedimentation methods involve measuring the amount of dust which
    settles on a clean, preweighed surface over a given period of time.
    Such procedures can make use of a simple collecting cup (Aurand et
    al., 1983) or a flat plate (Pellizzari et al., 1995). After a
    specified period of time, the sample is collected and measured, and
    the dust is then analysed in a laboratory. Data from the German
    Environmental Survey (Schulz et al., 1995) on domestic dust
    precipitation is given in Table 30. Sedimentation methods are useful
    for collecting samples over a specific period of interest (e.g., a
    day, week or month). In contrast, the integration times of settled
    dust samples collected using the wipe or vacuum methods described
    above are not well characterized.


        Table 30.  Sedimentation of elements in indoor dust, Germany 1990-1992 (Schulz et al., 1995)

                                                                                                           
    Element        No. of     No. of              Percentiles                  Maximal        Confidence 
                   samples    values                                           value        interval 
                              >LOQ       10          50           95           (GM)           (GM)
                                                                                                           

    Dustfalla      3282       -          1.4         21.0         579          4.52           4.36-4.68

    Arsenicb       3279       965        < 4         33           1313         5.4            5.2-5.6

    Boron          2896       511        < 0.06      0.64         47.1         0.13           0.13-0.14

    Cadmiumb       3282       0          5           44           833          11.7           11.4-12.0

    Calcium        3277       25         17          273          2679         51.2           49.5-52.9

    Chromium       3282       14         0.02        0.28         3.92         0.07           0.06-0.07

    Copper         3277       1167       < 0.3       1.5          48.8         0.33           0.32-0.34

    Iron           3277       74         2           41           765          7.7            7.4-8.0

    Lead           3282       0          0.11        1.17         86.6         0.29           0.28-0.29

    Magnesium      3277       26         2           25           342          5.2            5.0-5.3

    Phosphorus     3277       1063       < 1.8       17           542          2.8            2.7-2.9

    Zinc           3277       15         0.9         8.6          108          2.2            2.1-2.3
                                                                                                           

    Units are µg m-2d-1 unless otherwise indicated.
    a  mg m-2d-1.
    b  ng m-2d-1.
    


    8.3  Sampling design considerations

          Section 8.2.2 describes numerous innovative methods that have
    been developed and used by researchers to collect settled dust from
    surfaces. Many more examples can be found in the literature. However,
    there has been little standardization among the methods. Differences
    in vacuum pump flow rates, nozzle shapes and sizes, and sampling
    technique will affect dust-pickup characteristics of vacuum sampling
    methods and will, therefore, affect the results. Differences in wipe
    sampling material and technique will also affect the results from wipe
    samples. Different recovery rates of dust from alternative
    sedimentation methods can also have a large effect on analytical
    results. These differences among methods, which are not well
    documented in the literature, can make interpretations and comparisons
    between studies difficult. It is important that sampling methods are
    well described when results from settled dust sampling are reported.

          Sampling design considerations for soil should follow the
    objectives of the study and consider the particular conditions of the
    site being monitored. For example, multiple soil samples can be
    obtained around the perimeter of a house at a sufficient distance so
    that the soil is representative of material that might be tracked into
    the home. In this case, the samples might be composited. Backyard soil
    might vary in the number and amounts of contaminants present, as well
    as usage and specific activities by residents. The number and location
    of samples to be obtained should be based on these considerations.

    8.3.1  Concentration and loading

          Almost all settled dust contains measurable levels of common
    environmental contaminants such as heavy metals and pesticides, and
    most residential surfaces, such as floors and windowsills, contain
    settled dust (CDC, 1991). The actual concentration of a target analyte
    in a sample of settled dust depends on the amount of dust collected
    that does not contain the analyte and the amount of dust collected
    that does contain the analyte.

          The analyte concentration, sometimes called a  mass 
     concentration, is usually expressed as micrograms of analyte per
    gram of dust (µg/g). The amount of dust on a surface can be expressed
    as grams of dust per unit area, such as per square metre, and is
    usually called  dust loading (g/m2). The analyte concentration,
    multiplied by the dust loading on a surface, gives a  analyte 
     loading value and is commonly expressed as micrograms of analyte per
    unit area (µg/m2). The dust loading and analyte loading measurements
    are both  area concentrations, that is, the concentration of dust or
    contaminant per unit area. In this report, "concentration" refers to
    mass concentration and "loading" refers to area concentration.

          The example of residential sampling for lead is used to simplify
    the discussion. Common wipe sampling methods, such as the HUD method,
    measure lead loading directly, without measuring lead concentration

    and dust loading. Fig. 24 illustrates what common wipe samples can
    measure, using realistic results collected from floors in a
    hypothetical residence. Assume that each diagonal line in the figure
    represents the lead loading results from one wipe sample. The diagonal
    lead loading lines show the infinite number of lead concentration
    ( y axis) and dust loading ( x axis) combinations that might result
    in the measured lead loading. As mentioned earlier, the product of the
    two parameters is the lead loading (µg/g × g/m2 = µg/m2). Using a
    log scale on the  x and  y axes ensures that the infinite number of
    combinations that result in the same lead loading value fall on a
    straight line. As noted in Chapter 4, the distribution of many
    measures of environmental exposure is skewed right and may often be
    approximated by a lognormal distribution. For lognormal distributions,
    geometric relationships (e.g., factorial) exist among quantiles of the
    distribution, in contrast to the linear relationships present in
    measures that follow a normal distribution. As described in Chapter 4,
    lognormal distributions can be "normalized" in a numerical sense by
    expressing the data as the log-transformed values or in a graphical
    sense by plotting data on log scales. This example assumes that lead
    concentration and dust loading are lognormally distributed and
    perfectly correlated with each other, i.e., lead loading in µg
    lead/m2 is assumed to be constant. A scatter plot of two perfectly
    correlated and lognormally distributed measures depicted on a normal
    scale would exhibit a curved relationship, but appears as a straight
    line when depicted on a log scale.

          Because common wipe sampling measures lead loading directly, but
    does not measure lead concentration and dust loading, the results from
    wipe sampling cannot be used to determine which combination of lead
    concentration and dust loading is present. Similarly, Davies et al.
    (1990) states that for a given contaminant loading value, the
    contaminant concentration can range from high where there is little
    dust to, conversely, low where there is a large volume of dust. The
    only way to measure both lead concentration and dust loading is to
    collect a house dust sample with one of the vacuum sampling methods,
    or with one of the preweighed wipe sampling methods. Common wipe
    sampling methods do not measure lead concentration.

          Although research studies have shown that estimates of both lead
    concentration and lead loading (area concentration) correlate
    significantly with children's blood lead levels, it is unclear which
    measure is better at predicting the true, long-term, lead dust
    exposures to children. Results from Davies et al. (1990) suggest that
    the average lead loading (lead area concentration) measured in a
    child's environment expressed more realistically the exposure of
    children to lead than did lead concentration (lead mass concentration)
    measurements. Results from the Lanphear et al. (1995) study also
    suggest that lead loading measurements correlate better with
    children's blood lead levels than does lead concentration. However,
    Bornschein et al. (1985) showed that, for their conditions, lead
    concentration and lead loading have very similar correlations with

    FIGURE 24

    children's blood lead levels. Laxen et al. (1987) found that blood
    lead levels did not correlate better with lead dust loading than with
    concentration.

    8.3.2  Collection efficiency

          Another important concept to understand is that the type of
    surface from which the dust is sampled directly affects the efficiency
    of dust collection from the surface. Furthermore, different sampling
    methods recover different amounts of total dust from the same sampled
    surface. These differences are due to different collection
    efficiencies of the methods. Differences in collection efficiency on
    different surface types and among sampling devices may influence
    measurements of toxicant levels in settled dust.

          Roberts et al. (1991) documented total dust recoveries that
    ranged from greater than 90% by weight on a smooth painted surface to
    about 30% on a carpet. Chavalitnitikul & Levin (1984) compared several
    types of wipe sampling methods. They conducted a laboratory wipe
    sampling experiment with wipe materials on a smooth surface (Formica)
    and a rough surface (plywood). The study examined different wipe
    materials, such as Whatman filters, paper towels and adhesives --
    paper labels, adhesive cloth and dermal adhesive. The researchers
    determined that, on smooth surfaces, all techniques were comparable,
    with about 85-90% recovery with carefully prescribed protocols. On
    plywood, however, recoveries dropped to less than 43%. They also noted
    that the Whatman filters fell apart on the rough surface. Other
    sampling method characterization studies document similar differences
    (US EPA, 1995a,b).

          Three commonly cited methods used to sample lead in settled dust
    (the DVM, BRM, and HUD methods) may collect very different amounts of
    total dust from the same surface (Lanphear et al., 1995). Assuming
    that a smooth hard surface is sampled, the difference in collection
    efficiency between the DVM and the other two methods may be greater
    than a factor of 10, with the DVM consistently collecting less dust
    than the BRM and HUD methods. The latter two methods would probably
    collect similar amounts of dust on a smooth hard surface.

          Since contaminant loading is directly related to total dust
    collected from the sampled surface, the DVM sampler will consistently
    measure lower contaminant loading values on hard surfaces than the BRM
    or HUD methods. This does not imply that a high collection efficiency
    is better than a low efficiency. An argument in favour of the DVM's
    low collection efficiency is that it measures the more biologically
    active fraction of leaded dust available to a child (Que Hee et al.,
    1985). However, results from the only study to use all three methods
    side by side in children's homes suggest that the BRM and HUD methods
    correlate slightly better with children's blood lead levels than the
    DVM method (Lanphear et al., 1995). The same study showed that the BRM
    collects much more dust from carpeted surfaces than the DVM or HUD
    methods. The point to note is that lead loading measurements on the
    same surface differ among sampling methods. Further research is needed

    to determine the importance of collection efficiency for exposure
    assessment studies.

          As with contaminant loading, differences in collection efficiency
    on different surface types and among sampling methods may affect
    measurements of contaminant concentration. Differences in the relative
    recovery of contaminant-containing dust and non-contaminant-containing
    dust can result in different contaminant concentration measurements.
    Theoretically, however, concentration measurements are likely to vary
    less among methods than are loading measurements. Results from the
    Lanphear study, which collected hundreds of side-by-side lead dust
    samples with the DVM and BRM methods, are consistent with this theory.
    Geometric mean lead levels and the corresponding standard deviations
    suggest that, on average, side-by-side lead loading measurements
    differ more between the two sampling methods than do the lead
    concentration measurements (Lanphear et al., 1995).

    8.4  Sampling strategies

          Choosing an appropriate sampling method is an important part of
    designing a study to measure toxicants in house dust. However, it is
    only part of designing a sampling strategy. The sampling method
    specifies how to collect settled dust, whereas the sampling strategy
    specifies the process of sampling. Several of the questions that need
    to be answered when developing a sampling strategy are:

    *   What age group is targeted by the study?

    *   Which surfaces and substrates should be sampled?

    *   When and how should sampling take place?

    *   Should a composite sample be created?

    *   How will the samples be analysed?

          As noted in the first section of this chapter, young children who
    play on floors are likely to have higher exposure to settled dust than
    adults. Children may be also routinely exposed to dust in areas of a
    residence that adults do not contact. Different sampling strategies
    may be appropriate for different age groups.

          The potential effect of the surface type and substrate on dust
    collection should be factored into the strategy because dust
    collection efficiencies from different surface types can vary greatly.
    For example, toxicant loading or concentration measurements may
    correlate relatively well with biological measurements when dust is
    collected on hard floors or on carpets. However, if the person's
    relative exposure to dust from floors versus carpets differs from the
    sampling method's relative collection efficiency on these surfaces,
    the relationship between biological and settled dust measurements will
    be different for each surface. Similar differences between a human's
    exposure and a sampling method's collection efficiency may be found

    between components within a room, such as between a windowsill and a
    floor.

          Another issue to note is that the sources of dust, its temporal
    and spatial variability, and accessibility to humans, especially to
    young children, may vary greatly from person to person, room to room
    and house to house. However, little research has been done to examine
    this variability across space and time. Interpretations of house dust
    sample results may, therefore, be affected by this variation in
    addition to the variation introduced by the choice of sampling method.
    Short-term changes in a person's environment before sampling, possibly
    influenced by sporadic house cleaning practices or by a person who has
    just returned home from vacation, may offset the dust/biological
    relationships owing to the timing of sample collection.

          The toxicant levels in settled dust to which a person is exposed
    may be thought of as a weighted average across the areas where the
    person has dust contact, with weights roughly proportional to the time
    a person spends in different areas. From a sampling perspective, the
    average toxicant level to which a person is potentially exposed may be
    estimated by collecting many individual samples of settled dust for
    separate analysis and combining the results by calculating a weighted
    average after analysis. Or, field composite samples can be collected
    before laboratory analysis by collecting and physically combining two
    or more settled dust samples from each of several areas in a dwelling.
    Researchers have used both strategies for collecting dust samples
    (Farfel & Rhode, 1995).

          A common criticism of composite sampling is that toxicant
    variation across a floor or throughout a residence cannot be
    determined; toxicant "hot spots" may be missed. It must be
    acknowledged, however, that any settled dust sampling strategy may
    miss hot spots. The important issue is how much these hot spots
    contribute to the total exposure of the average person. This question
    has not been answered by scientific studies. In any case, the
    statistical relationship between biological toxicant levels and
    average toxicant levels in settled dust levels across large areas in
    which a person may be exposed are likely to be better than the
    relationship between biological levels and a potential high-dose
    source of toxicant exposure for a short period of time. Davies et al.
    (1990) used this assumption to design a sampling strategy that
    collected settled dust "taken over all the exposed floor surface in
    the rooms concerned" (thus, the average level was measured in a room)
    rather than from small areas in the room, and found a relatively high
    statistical relationship with children's blood lead levels
    ( r = 0.46).

          Possibly the best measures of toxicants in settled dust for
    exposure assessment purposes are averages of dust measurements taken
    repeatedly over time. If one were to repeat sampling over time,
    averages across space and time could be obtained. However, most
    sampling strategies used in previous studies collected settled dust at
    only one point in time. An obvious advantage to cross-sectional (one

    time) studies is that they are less expensive than longitudinal
    (repeated measures) studies, which require repeated visits to a
    dwelling, greater occupant burden, and higher laboratory analysis
    costs.

          One possible, but untested, approach to strengthening estimates
    of time-weighted average dust levels in cross-sectional studies may be
    to measure exposure-weighted average levels based on the activity of
    the person. This may be done by listing indoor locations where the
    person spends time, then roughly estimating the percent of time spent
    actively in each location, rounded to a convenient percentage. Samples
    can then be composited from the specific areas by adjusting the
    subsample areas to be proportional to the percent of time spent in
    each area. An exposure-weighted average toxicant dust level could then
    be estimated from the result.

          Finally, laboratories performing the chemical analysis should be
    consulted before settled dust samples are collected. This is
    particularly true when collecting composite wipe samples. An excess of
    towelette material may present problems during the laboratory
    digestion phase of analysis, requiring more reagents and larger
    beakers than normally used, and potentially reducing the toxicant
    recoveries owing to matrix effects. Similarly, vacuum sampling may
    collect more dust than is required for analysis. If this is the case,
    techniques need to be employed by the laboratory to ensure that the
    fraction of dust analysed represents the whole. Another potential
    source of error in the results lies in how the dust is handled after
    sampling and prior to analysis. If measurements of lead concentration
    in dust are important for the objectives of the study, sampling
    methods that present the dust to the laboratory in an easy-to-handle
    form should be considered over alternate methods. These issues and
    others should be well thought out before the commencement of a settled
    dust sampling effort.

    8.5  Summary

          Human contact with soil and settled dust can be an important
    source of exposure to chemical contaminants, especially for children.
    Although many sampling methods have been developed, no single approach
    has been demonstrated to be superior to the others. As a consequence,
    it is difficult to compare results from studies that utilize different
    sampling methods. Important factors to consider when selecting a
    sampling method include collection efficiency, differences in human
    activity patterns, physical variability of soil and dust levels over
    space and time, surface and substrate sampled, timing of sample
    collection and analytical methods used to measure toxicants in the
    laboratory.

    9.  MEASURING BIOLOGICAL HUMAN EXPOSURE AGENTS IN AIR AND DUST

    9.1  Introduction

          Microbiological organisms have long played an important role in
    human ecology. Fungi are critical to the production of cheese and the
    fermentation of beer, and in some cases are a direct source of
    nourishment. In the first half of the 20th century,  Penicillium 
     chrysogenum colonies were discovered to inhibit growth of other
    organisms. Today pharmaceutical companies, among others, are exploring
    fungal enzymes for a variety of reasons including new drugs,
    non-chemical pesticides, biodegradation of waste and possible
    catalysis of chemical reactions.

          However, natural does not mean benign. Human exposures to
    microorganisms have resulted in allergic, toxic and infectious
    disease. As humans have modified the environment through cultivation,
    landscaping and building structures, ecological balances have been
    disturbed. The distribution of moisture and nutrients has been altered
    to a point where it is quite common to encounter reservoirs of fungi,
    bacteria and algae, and infestations of mites and cockroaches.

          Through airborne dispersion, ingestion or direct contact, humans
    confront components of microorganisms continuously. We may be affected
    through an immune reaction requiring sensitization. Predisposed
    individuals may not experience a reaction for some time after they
    have been exposed. Once an individual is sensitized, a reaction such
    as an asthmatic attack might be delayed hours following the exposure
    event. However, there are many infectious diseases induced by fungi
    and bacteria that require no period of sensitization before illness
    develops. There is yet another route whereby microorganisms can evoke
    irritation and health effects: some metabolites from moulds are
    carcinogenic (e.g., aflatoxin B) or immunosuppressors; some cause
    dermatoxic effects; others cause annoyance and irritation by the VOCs
    they release.

          Table 31 provides basic categories for the microorganisms of
    primary interest and some possible sources. Assessing exposures to
    microorganisms is very different in some aspects from assessing
    exposures to physical or chemical agents. For virtually all
    microorganisms, exposure-response or dose-response information is
    currently limited. Nevertheless, exposures to allergens, fungal
    spores,  Legionella, and tuberculosis, among many others, are being
    inferred from sampling. And, particularly for assayable antigens and
    endotoxin, dose-response data are accumulating rapidly. Observed
    increases in tuberculosis and asthma as well as atopy have brought a
    resurgence of epidemiology and expanded interest in exposure
    assessment.


        Table 31. Common bioaerosols, related diseases and typical sources

                                                                                                                        
    Bioaerosol                Examples of diseases                 Common sources
                                                                                                                        

    Pollens                   hay fever                            plants, trees, grasses, ferns harvesting, cutting, 
    Spores                    allergic rhinoconjunctivitis         shiploading
    Plant parts               asthma
                              upper airway irritation

    Fungi                     asthma, allergic diseases            plant material, skin, leather, oils; bird, bat and 
                              infection                            animal droppings; feathers, soil nutrients, 
                              toxicosis                            glues, wool
                              tumours

    Bacteria                  endotoxicosis                        humans, birds and animals (e.g., saliva, blood, 
                              tuberculosis                         dental secretions, skin, vomit, urine, faeces)
                              pneumonia, respiratory and wound     water sprays and surf, humidifiers, hot tubes, 
                              infections, legionellosis, Q and     pools, drinking water, cooling towers
                              pontiac fever

    Other allergen sources    asthma                               mite excreta, insect parts (cockroach, spiders, 
    Arthropods                dermatitis                           moths, midge)
    Vertebrates               hypersensitivity                     dander and saliva from cats, dogs, rabbits, 
                              pneumonitic                          mice and rats, bird serum, farm animal dander

    Virusesa                  respiratory infections, colds,       infected humans, animal excreta, 
                              measles, mumps, hepatitis A,         insect vectors, protozoab
                              influenza, chicken pox, Hanta virus
                                                                                                                        

    a  Viruses are included in table for completeness but are not covered in this chapter.
    b  Protozoa in the form of free-living amoebae can be direct acting pathogens or allergens; they can also 
       interact with bacteria (e.g.,  Legionella growth within amoebae).
       Source: developed from Burge (1995).
    

          This chapter discusses the strategy and methodology for exposure
    assessment of five major categories of biological particles:

    *   house dust mites and their faeces

    *   allergens from pets and cockroaches

    *   allergens and/or toxins derived from

        -   fungi
        -   bacteria

        -   pollen

          For each category information will be presented regarding
    sampling methods, methods of analysis, and advantages and drawbacks of
    the different methods. Seasonal variations in mite allergen and fungi
    are illustrated by showing the summary results of an extensive survey
    conducted in Australia. Mite and pollen antigen as well as fungal
    organisms can vary substantially within homes and buildings, as
    illustrated in the figures in this chapter. The reader is referred to
    texts such as ACGIH (1989) and Burge (1995) for details on
    instrumentations, specific information relevant to the allergenic,
    infectious and toxigenic properties of many microorganisms and their
    constituents and metabolic by-products.

          There are three different basic approaches for the exposure
    assessment of biological particles: observational sampling, reservoir
    sampling (dust, surfaces, water) and air sampling.

    *    Observational sampling means that one uses sensory perception to
        collect data about potential sources of exposure to biological
        particles (e.g., visible fungal growth).

    *    Reservoir sampling refers to the collection of bulk material
        (e.g., surface contact, bulk material, water sample or dust sample)
        to estimate the potential exposure.

    *    Air sampling is the most likely to be representative of human
        exposure.

          This chapter will emphasize reservoir (primarily indoor dust) and
    air sampling of bioaerosols and not gaseous metabolic products.

          Designing a specific sampling programme requires consideration of
    the aim of the sampling, the nature of the biological particles
    (including size and expected concentrations) and parameters that
    influence the actual exposure to these particles. These parameters
    determine the choice of the sampling and quantification method, the
    sampling strategy (e.g., location, season, duration and frequency),
    and approaches for statistical analysis and interpretation of the
    data. For most situations, the exposure route of interest is
    inhalation. Therefore, ideally, the exposure should be assessed by

    personal air monitoring. As will become clear from the remainder of
    this chapter, however, no single sampler fulfils the characteristics
    of the ideal sampler to measure the total exposure to biological
    particles. Many of the methods used for estimating environmental
    concentrations of biological particles are not truly representative of
    an individual's exposure to these particles. As stated earlier, this
    is, in part, because the exposure measure of biological importance is
    not well understood. In addition, the field of environmental
    aeromicrobiology developed from a laboratory biology base that
    borrowed sampling techniques and equipment from other fields. Until
    recently there had been little convention or need for uniformity of
    methods. It is not surprising, therefore, to find a general lack of
    data regarding the validity of the methods used to estimate the
    exposure to biological particles. This situation has certainly changed
    as those investigating exposure assessment aspects of aerobiology have
    cooperated with environmental epidemiologists.

          Useful reference texts with regard to sampling and analysis of
    biological particles include those by the American Conference of
    Governmental Industrial Hygienists (ACGIH, 1995), the European
    Commission (EC, 1993), Hamilton et al. (1992), Pope et al. (1993),
    Burge (1990, 1995), and Burge & Solomon (1987), Reponen (1994), and
    Verhoeff (1994a,b).

    9.2  House dust mites

          House dust mites are members of the arachnid family having eight
    legs and an exoskeleton. They can be up to 300 µm in length and live
    off organic debris found in house dust (e.g., skin flakes, hair
    follicles and fungi) (Colloff, 1991). Because mites absorb water
    vapour they are critically dependent on the absolute humidity.
    Survival in the adult stage requires environmental moisture conditions
    be sustained not lower than 7-8 g/m3 (Korsgaard & Iversen, 1991;
    Fernandes-Caldas et al., 1994). This is equivalent to a relative
    humidity of about 50% at 20°C.

          Mite antigen is mainly found in the faecal pellets which may be
    10-20 µm in diameter and will not remain suspended for very long.
    Feather et al. (1993) identified enzymes derived from the mite gut as
    the source of allergens. These enzymes might remain as potent
    allergenic material in bedding, mattresses, carpets and furnishings
    long after the mite population has diminished, further complicating
    exposure determination. 

          Two different approaches, the sampling of air and of settled
    dust, are available to measure the presence of house dust mites and
    their allergens as indicators of environmental exposure. The latter is
    the most commonly used approach. 

    9.2.1  Air sampling for house dust mites

          Several techniques exist for volumetric sampling of airborne mite
    allergens, using cascade impactors or high- and low-volume samplers in
    combination with membrane filters (Swanson et al., 1985; Price et al.,
    1990; Sakaguchi et al., 1993; Oliver et al., 1995). These techniques
    have the advantage that they sample airborne allergens and might
    therefore be more representative of the true exposure than assays of
    settled dust. The literature is limited, however, on the validity of
    air sampling as measure of exposure to house dust mite allergens
    (Swanson et al., 1985; Price et al., 1990; Sakaguchi et al., 1993),
    and further research is needed.

          Mites themselves are not seen in air samples. Furthermore, in
    undisturbed rooms amounts of airborne mite allergens are small and
    difficult to detect, even after prolonged sampling. Most of the mite
    allergens bind to faecal pellets, which become airborne only as a
    result of disturbance, and little allergen is associated with
    particles that remain airborne for more than a few minutes. Therefore,
    practical disadvantages of airborne sampling of mite allergen are the
    requirements for long sampling periods (2-24 h) and very sensitive
    assays (Thien et al., 1994). Price et al. (1990) used a low-volume air
    sampler (2 litre/min) for 3 h to sample suspended dust mite allergen
    in homes. They reported that the airborne allergen levels correlated
    better with sensitization to mites among children than the levels in
    dust. Further, the air and dust antigen levels were not correlated.
    Although this is the only study linking atopy to airborne mite
    allergens, it does suggest potential limitations of using dust
    sampling as a surrogate exposure measure. In a small number of
    studies, air sampling and dust sampling were carried out in parallel
    (Price et al., 1990; Sakaguchi et al., 1993; Oliver et al., 1995). In
    only one study were significant correlations found between the levels
    of house dust mite allergens in air and dust (Oliver et al., 1995).
    Allergenic responses to dust mite allergens may be induced by
    short-duration high-concentration exposure events. Therefore, the
    clinical importance of integrated air samples may be more relevant in
    predicting prevalence of atopy to mites rather than predictive of
    acute health effects.

          At present no reliable information is available that will support
    adoption of a standardized method for air sampling of house dust mite
    allergens. According to an international workshop held in 1987
    (Platts-Mills & De Weck, 1989) airborne sampling has not been shown to
    be better than dust sampling to measure the level of mite infestation
    in homes or schools. This was confirmed by a second international
    workshop in 1990 (Platts-Mills et al., 1992). It was also stated that
    there are few or no data showing a relationship between airborne
    measurements and sensitization to house dust mites or symptoms. In
    contrast, a relationship is apparent between the concentrations of
    mite allergens in settled house dust and sensitization or symptoms.
    Therefore, air sampling was not recommended (Platts-Mills et al.,
    1992).

    9.2.2  Dust sampling for house dust mites

          Dust sampling for measurement of the level of mite infestation is
    accepted and recommended as the best-validated "index of exposure" to
    house dust mite allergens. The approach assumes that the quantity of
    allergens released into the air is a function of what is present in
    settled dust, or, conversely, that the measurement of allergen in
    settled dust is related to both the long-term dose a person receives
    and to the short-term airborne levels experienced during events that
    raise dust.

          Standardized sampling procedures to measure house dust mites and
    their allergens in house dust have been proposed (Platts-Mills & De
    Weck, 1989; Platts-Mills et al., 1992; EC, 1993; Dreborg et al.,
    1995). Sampling sites should be consistent throughout the study and
    preferably include the upper mattress surface and the floors of the
    living room and bedroom. Sampling can be conducted with vacuum
    cleaners equipped with a special attachment to collect dust on a paper
    filter. Vacuuming 1 m2 of surface in 2 min is a commonly used
    sampling method. Depending on experiences with the amount of dust
    recovered in specific situations, investigators may have to modify the
    sampling procedures. Samples can also be obtained from upholstered
    furniture, soft toys and clothing. Alternative techniques for
    collecting dust samples include shaking blankets in a plastic bag and
    scraping flat surfaces higher than floor level with a piece of firm
    card. However, these techniques are less effective than collection by
    vacuum cleaner and not standardized. The dust samples may be sieved
    before analysis to obtain a sample of fine dust that can be weighed
    accurately. Unfortunately, dust samples may still vary in density
    after sieving. An alternate method for sampling airborne mite
    allergens is to collect settling dust on large Petri dishes over a
    period of 14 days (Tovey et al., 1992; Oliver et al., 1995). Brown
    (1994) developed a variation on the integrated settling method. A
    100-cm2 piece of sticky tape is placed in contact with the surface
    for 24 h. Under low-power magnification (36×), the trapped mites are
    counted. Using an empirically derived collection efficiency of 30%,
    the number of live mites per area is estimated. However, this does not
    reflect the true extent of exposure to mite allergens (see section
    9.2.3.1).

    9.2.3  Available methods of analysis for house dust mites

          There are three types of method for estimating the concentrations
    of house dust mites or their allergens in (airborne) dust samples:
    mite counts, immunochemical assays of mite allergen and guanine
    determinations. The choice of a particular method depends on the
    specific purpose of a study.

    9.2.3.1  Mite counts

          The prevalence of mites in settled house dust can be determined
    by counting under a microscope after separation from the dust sample
    by flotation or suspension. This technique permits the identification

    of the predominant species and the recognition of live, dead, larval
    or adult types. The disadvantages of this method include:

    *   the need for training and development of skill in determining
        different mite species

    *   the failure to quantify faecal pellets and disintegrated mite
        bodies and therefore to reflect the true extent of exposure to mite
        allergen levels

    *   the unsuitability for large-scale (epidemiological) studies owing
        to the time-consuming nature of the work (Platts-Mills & De Weck,
        1989; EC, 1993).

          A further limitation of this method is variation among the actual
    extraction techniques. Bischoff et al. (1992) estimates that less than
    10% of the mites are removed from the carpet by typical vacuuming
    techniques, but this number varies with the type of surface, the type
    of vacuum used and the vacuuming technique.

    9.2.3.2  Immunochemical assays of dust mite allergens

          Immunochemical assays are widely used to measure the
    concentrations of house dust-mite allergens. The dust mite germ is
     Dermatophogoides and allergens have been identified for three
    species. The conventional labelling of these allergens are denoted by
    the prefix "Der" followed by a letter indicating the species. These
    assays are possible because the major allergens produced by house dust
    mites, i.e., the group 1 allergens (Der p I, Der f I, Der m I) and the
    group 2 allergens (Der p II, Der f II, Der m II) are well
    characterized and purified. For immunochemical analysis, the dust
    sample is extracted (e.g., in a buffered saline solution), and then
    stored frozen until analysis.

          Total mite allergen content can be assessed by
    radioallergosorbent tests (RAST). This method provides a good estimate
    of the relative potency of different allergen extracts, but cannot be
    used for absolute quantification of mite allergen levels. An advantage
    of the method is that it measures "relevant" antigenic determinants
    that have elicited a response in allergic subjects, since human IgE is
    used. Results vary with the composition of the extract used on the
    solid phase and with the composition of the serum pool used for
    detecting bound allergen. However, RAST inhibition results are
    difficult to reproduce over an extended period of time.

          Individual mite allergens can be measured with enzyme-linked
    immunosorbent assays (ELISA) or radioimmunoassays (RIA). Sandwich
    radio- or enzyme immunoassays employ either rabbit polyclonal or mouse
    monoclonal antibody for capture, and a second monoclonal antibody for
    detection (see Fig. 25). These assays are more sensitive than RAST.
    Those using monoclonal antibodies in particular have also the great
    potential advantage of long-term reproducibility. Furthermore, ELISA
    assays have been shown to be highly reproducible (e.g., Munir et al.,

    FIGURE 25

    1993; Van Strien et al., 1994) and can quantify antigen levels to less
    than 1 ng/mg dust.

          Immunochemical assays are highly specific and the results
    obtained with these assays can be expressed in absolute units of a
    defined protein by unit weight of dust or by unit area sampled. They
    are suitable for large-scale surveys because they can be automated.
    However, a sophisticated laboratory is required.

    9.2.3.3  Guanine determination

          The third possibility is the measurement of guanine, which is a
    nitrogenous excretory product of arachnids, found in house dust. Since
    mites are predominant among arachnids in house dust, determination of
    guanine content in the dust is an indirect method for assessing mite
    allergen levels. Analysis of guanine content is based on a colour
    reaction between guanine and an azo compound (Le Mao et al., 1989;
    Hoyet et al., 1991). The amounts of guanine can be measured
    quantitatively on a weight/weight basis using a spectrophotometer, or
    semiquantitatively using a commercially available test kit (Pauli et
    al., 1995). The quantitative assay has been reported to demonstrate a
    good correlation with the assay of Group 1 allergens (Platts-Mills et
    al., 1992), whereas the semiquantitative test was found to be less
    sensitive (Lau et al., 1990).

    9.2.4  Mite allergens

          Sampling strategies may vary depending on objectives but most
    studies collect vacuum samples using a protocol that, at least
    internally, standardizes equipment, area, duration and location. Mites
    are typically found in higher concentration in bedding. Typical areas

    would include mattresses, pillows, blankets and bedroom floors.
    Because of spatial variability, mixed floor samples can be used. Other
    areas of high use include living room, upholstered chairs and couches,
    and covered floors. Bischoff et al. (1992) describes an approach used
    to avoid depletion of the dust reservoir during routine and repeated
    sampling.

          Mite-antigen levels have been shown to vary with season,
    reflecting the moisture and temperature dependency controlling mite
    development stages. Garrett (1996) conducted a yearlong study in 80
    homes in eastern Australia. Fig. 26 reveals the temporal variation in
    Der p I, the prominent allergen. The allergen levels in dust collected
    from the bedroom and living room are higher during the warmer and more
    humid months of the year. Garrett (1996) has shown that the allergen
    level for Der p I is consistently higher in dust collected directly
    from the bedding. The between-home variation is quite apparent,
    ranging over almost two orders of magnitude. Examining Fig. 27 offers
    an explanation for the higher levels of greater variability in the
    allergen levels recovered from the bedding dust. Mites survive better
    in mattresses with spring cones than in foam rubber. Presumably, less
    moisture is retained in the hydrophobic foam material. Also, wool
    sheets and blankets favour the growth and retention of mite antigens
    more than alternative bedding material. Other studies on mites in wool
    rugs suggest that the thermal properties of wool help mites to survive
    fluctuations in temperature and moisture and, perhaps, inhibit their
    removal.

    9.3  Allergens from pets and cockroaches

          For estimating the exposure to allergens derived from pets (e.g.,
    cats and dogs), and cockroaches, the same approaches are available as
    for house dust mites and their allergens (i.e., the sampling of air
    and dust). The major allergens of the cat (Fel d I), dog (Can f I),
    the German cockroach (Bla g I, Bla g II), and the American cockroach
    (Per a I), have been characterized and purified (Chapman et al., 1988;
    Pollart et al., 1991a; Schou et al., 1991, 1992). Research is still in
    progress to further unravel the structure of the allergens derived
    from pets and roaches (and house dust mites) using techniques for
    allergen cloning and sequencing.

    9.3.1  Air sampling for allergens from pets and cockroaches

          Cockroaches are year-round inhabitants of homes. They need access
    to both food and water, so they are often found in kitchens and
    bathrooms. Unlike mites, where the antigen source is in faecal matter,
    cockroaches are thought to secrete their allergen on to their bodies
    and on to surfaces (Vailes et al., 1990). This means that body parts,
    egg shells, faecal particles and saliva might contain allergens
    (Lehrer et al., 1991).

          Similarly, a wide range of materials derived from mammals contain
    potentially allergenic material, including hair, dander, serum,
    saliva, urine and faecal matter. Direct contact as well as inhalation

    and ingestion can cause allergic reactions (Burge, 1995). Because of
    the popularity of cats and dogs as domestic pets, they have been the
    subject of much of the work on mammalian allergenic reactions. Cat
    allergens from saliva and sebaceous gland secretions reside on
    particles less than 2.5 µm in size. Much of the dog allergen is
    believed to be associated with dander and hair, but saliva and serum
    are also important sources.


    FIGURE 26

    FIGURE 27

          There are only limited data on the size ranges for airborne
    allergen particles from dogs, rabbits, rats and other animals. In
    general, however, saliva sources tend to be small (<2 µm) whereas
    dander and urine particles are larger (10 µm).

          For the sampling of airborne allergens derived from cats, dogs
    and cockroaches, the same methods can be used as for the sampling of
    airborne mite allergens (see section 9.2.1). These allergens have been
    sampled using cascade impactors (Luczynska et al., 1990; De Blay et
    al., 1991), high-volume samplers in combination with fibreglass
    filters (Swanson et al., 1985; Sakaguchi et al., 1993) and liquid
    impingers (Luczynska et al., 1990).

          As is the case for house dust mite allergens, only limited data
    have been published on the validity of air sampling as a measure of
    exposure to allergens derived from pets and cockroaches. At present
    there is no reliable information to support adoption of a standard
    method for air sampling of these allergens. Airborne sampling has not
    yet been shown to be a better estimation of the exposure to these
    allergens than dust sampling. Therefore, further research to compare
    the usefulness of air and dust sampling is needed.

    9.3.2  Dust sampling for allergens from pets and cockroaches

          The sampling of house dust to investigate the presence of
    allergens derived from pets and cockroaches can be conducted exactly
    as for house dust mites and their allergens (see section 9.2.2).

    9.3.3  Available methods of analysis

          Immunochemical assays (ELISA) are available for detection of the
    allergens derived from cats (Chapman et al., 1988), dogs (Schou et
    al., 1992) and cockroaches (Pollart et al., 1991b; Schou et al., 1991)
    in (airborne) dust samples. The allergens of the American and German
    cockroach (i.e., Per a I and Bla g I) were demonstrated to be
    immunologically cross-reactive proteins and can be measured in the
    same assay (Schou et al., 1991). For immunochemical analysis, the dust
    sample is extracted (e.g., in a buffered saline solution), and then
    stored frozen until analysis.

          The ELISA assays for Fel d I and Can f I were found to be highly
    reproducible (Chapman et al., 1988; Schou et al., 1992). For the
    Bla g I and Bla g II ELISA assays the intra- and interassay
    variability were also found to be small (Pollart et al., 1991b).

    9.3.4  Typical allergen concentrations

          Cat and dog allergens have been reported more often than
    allergens from other mammals. Homes with cats have dust levels of Fel
    d I exceeding 10 µg/g, whereas homes without cats have typically less
    than 1 µg/g. A provisional value of 8 µg/g of dust has been proposed
    as indicating significant exposure. Cat antigen has been found in dust
    samples collected in theatres, offices, aeroplanes, schools and homes
    without a cat. Because of its small particle size, cat antigen can
    stick to clothing and be transported to other locations. Dog allergens
    have not been as extensively examined for non-residential sites.
    Dybendal et al. (1989) has reported that dog allergen was present in
    homes and schools where dogs were not kept.

    9.4  Fungi

          Fungi are a large and diverse class of microorganisms. They live
    on organic nutrients and have no chlorophyll or internal organs. The
    cells that make up fungal colonies contain complex carbohydrate
    macromolecules. Fungi must produce spores or conidia for their
    reproduction. Spores are usually 2-20 µm in size and oblong in shape.
    In the appropriate setting, spores reproduce new organisms.

          The two different approaches to assess the exposure to fungal
    particles are air sampling and dust sampling. For completeness, other
    approaches to "dust" sampling include lifting spores from a surface
    with sticky tape or direct contact with culture agar. The most
    commonly used approach is air sampling of culturable (viable) fungal
    particles.

    9.4.1  Air sampling for fungi

          Several techniques have been described for volumetric sampling of
    fungi in outdoor and indoor environments. Table 32 presents an
    overview of the techniques most commonly used for the sampling of
    fungal particles. Detailed information on the different sampling

    devices can be found in ACGIH (1995). Some of the techniques give
    total counts of all airborne particles, viable and non-viable, whereas
    others only give counts of viable fungal particles (e.g., propagules
    or colony forming units (CFU)). A few methods are discussed that
    provide not only total counts, but also viable counts (e.g., filter
    samplers). The sampling efficacy of a bioaerosol sampler is both a
    physical and a biological problem. For air sampling of fungal
    particles the following physical sampling principles may be
    distinguished: impaction on to a solid or semi-solid surface (e.g., a
    culture medium or an adhesive), centrifugal impaction, filtration and
    liquid impingement.

          Impaction on to a culture medium (e.g., for culturable fungi) is
    the most widely used technique, particularly in non-industrial indoor
    environments. This process depends on the inertial properties of the
    particles, such as size, density and velocity, and on the physical
    parameters of the impactor, such as inlet-nozzle dimensions and
    airflow paths. Because of differences in characteristics, samplers
    differ in cut-off size ( d50) (e.g., the particle size above which
    50% or more of the particles are collected). As most impactors have
    very sharp cut-off characteristics, almost all particles larger than
    the  d50 are collected and  d50 is therefore assumed to be the size
    above which all particles larger than that size are collected
    (Nevalainen et al., 1992). No sampler collects all particles with
    equal efficiency, and it is therefore not surprising that different
    quantitative and qualitative results are obtained using different
    sampling devices for culturable fungi (Verhoeff et al., 1990). The
    choice of the collection (culture) medium also affects the kinds and
    levels of fungi recovered (Verhoeff et al., 1990). No single
    collection medium will enable the entire range of viable fungi in the
    air to be isolated. Media which are generally accepted for
    aerobiological studies include malt extract agar (MEA), V8 juice agar
    and dichloran 18% glycerol agar (DG18) (EC, 1993; ACGIH, 1995). MEA
    and V8 agar are broad spectrum media, whereas DG18 is intended to be a
    selective medium for xerophilic fungi, but many of the common fungal
    species in air can also be isolated (Verhoeff et al., 1990).

          Few published data are available on the validity (accuracy and
    precision) of the measurement of fungi in air as estimate of exposure.
    All commonly used cultural air samplers use short sampling periods,
    typically 30 seconds to several minutes (Table 32). The
    reproducibility of parallel duplicate samples and sequential duplicate
    samples is only moderate, both in terms of CFU/m3 and in terms of
    species isolated (Verhoeff et al., 1990). More importantly, repeated
    sampling within weeks has demonstrated that variation in time within
    homes is much higher than the variation between homes (Verhoeff et
    al., 1992). This means that a single air sample has only a low
    predictive value for exposure over time. Furthermore, the use of
    cultures for quantifying fungal particle concentrations in air samples
    will give an underestimate of the actual particle concentrations, and
    may cause significant fungal contamination to be missed altogether.
    The culturable fungal particles may comprise only a few percent of the


        Table 32.  Overview of sampling techniques for airborne fungal particlesa

                                                                                                                               
    Method with examples                    Sampling rate and time               Remarks
                                                                                                                               

    Non-viable, non-volumetric
    - settling surface, adhesive-coated     undefined, minutes to days           semi-quantitative, over-representation of 
                                                                                 larger particles, microscopic identification

    Non-viable, volumetric
    - rotating tape/slide impactors
      Burkard trap                          10 litre/min, 7 days                 cut-off 2.5 or 5.2 µm, depending on slot

    - rotating arm impactors
      Rotorod sampler                       47 litre/min, intermittent           cut-off unknown

    - filter methods
      cassette filters                      1-4 litre/min, hours                 viable counts possible by plating washings 
      high-volume filters                   150-2000 litre/min, hours            from the filters

    Viable, non-volumetric
    - settlement plates                     undefined, hours                     semi-quantitative, over-representation of 
                                                                                 larger particles

    Viable, volumetric
    - multiple hole impactors
      Andersen 6-stage sampler              28.3 litre/min, 1-30 min             cut-off 0.65-0.70 µm, size separation
      Andersen 2-stage sampler              28.3 litre/min, 1-30 min             cut-off 0.65-0.70 µm, size separation
      Andersen 1-stage (N6)                 28.3 litre/min, 1-30 min             cut-off 0.65-0.70 µm
      Surface Air System sampler            90/180 litre/min, 20 sec-6 min       cut-off depends on number of holes and flow
      Eight-stage personal impactor         2 litre/min, 5-30 min                cut-off 5.2 µm, size separation
      Burkard portable sampler              10/20 litre/min, 1-9 min             cut-off 4.1/2.9 µm (10/20 litre/min)

    - centrifugal impactors
      Reuter Centrifugal sampler (RCS)      ca. 40 litre/min, 20 sec-8 min       cut-off 3.8 µm
      Reuter Centrifugal Plus (RCS-Plus)    ca. 50 litre/min, 30 sec-8 min       cut-off unknown
                                                                                                                               

    Table 32.  (continued)

                                                                                                                               
    Method with examples                    Sampling rate and time               Remarks
                                                                                                                               

    - rotating slit-to-agar impactors
      Mattson-Garvin air sampler            28 litre/min, 5-60 min               cut-off 0.5 µm

    - liquid impingers
      single-stage all glass impingers      12.5 litre/min                       cut-off 0.3 µm
      three-stage impingers                 20 litre/min                         cut-off <4 µm, size separation
                                                                                                                               

    a For detailed information see ACGIH (1995).
    

    total number of fungal particles (Horner et al., 1994). Thus, in order
    to optimize the information available from air sampling, both types of
    particle should be sampled. However, even using the best available
    method, a large number of airborne spores will not grow in culture and
    cannot be visually identified with available methods.

          At present, there is no standardized method for the sampling of
    airborne fungi, although the American Conference of Governmental
    Industrial Hygienists (ACGIH, 1989) and the European Commission (EC,
    1993) have given recommendations. An outline for selecting a
    bioaerosol sampler is presented by the American Conference of
    Governmental Industrial Hygienists (ACGIH, 1995). Selection criteria
    include sampling location, form of recovered particles (intact or
    dispersed), the need for size separation and the expected
    concentrations of the particles.

    9.4.2  Settled dust for fungi

          Settled house dust can be sampled for viable fungi in exactly the
    same way as for house dust mites and their allergens (see section
    9.2.2). The dust samples can be stored at room temperature but the
    analysis should be performed within a few days.

          Few published data are available on the validity of the
    measurement of culturable fungi in settled dust as estimate of
    exposure. The results, both quantitatively and qualitatively, depend
    on the method of inoculation of the dust and on the culture medium
    used (Verhoeff et al., 1994a). The reproducibility of duplicate
    analyses in terms of CFU/g dust is acceptable, but in terms of species
    isolated only moderate. However, as is the case for air sampling, a
    single dust sample is a poor estimate of exposure to fungi over time
    (Verhoeff et al., 1994a).

    9.4.3  Available methods of analysis for fungi in air

          Air samples obtained with sampling devices collecting total
    fungal particles can be analysed by direct examination to obtain total
    counts of fungal particles. Samples collected on culture media have to
    be incubated to obtain counts of viable fungal particles. Dust can be
    plated either directly on to a culture medium or suspended and diluted
    prior to plating. Total counts of fungal particles in dust can also be
    obtained by partitioning into an aqueous two-phase system followed by
    epifluorescence microscopy (Strom et al., 1987).

          Samples are incubated for at least 4 days; up to 7 days is the
    typical time needed for spores to generate identifiable col