7 ExposureAssessment

Markf. Nieuwenhuijsen Centrefor Researchin Environmental (CREAL), Barcelona, Spain

7.1. INTRODUCTION 7.2. INITIATCONSIDERATIONS OF AN 'Exposure' is a substanceor factor affecting human EXPOSUREASSESSMENT STRATEGY , either adversely or beneficially. Exposure A major aim of epidemiologicalstudies is to deter- variablesused in practicein epidemiologyusually mine whether or not there is an associationbetween 'true' have to be regardedas approximationsto the a particular substanceof interest,the exposureand exposure of the subjects who are being studied. morbidity and/ormortality. If there is an association, 'true' The accuracy and precision with which it is desirableto be ableto showan exposure-response exposure is being approximated may vary widely (or dose-response)relationship, i.e. a relationshipin 'surrogate' from one exposurevariable to the next. which the rate of diseaseincreases when the level Exposure misclassification and/or measurement of exposure(or dose) increases.This will aid in the error can lead to attenuationin health estimates interpretationof such studies. and/or a loss of power. A simple example illustrates Over recentyears there hasbeen increasinginter- the potential effects of exposure misclassification est in the field of exposureassessment causing it to and shows how important a good exposure assess- develop rapidly. We know more than ever to what, ment is. Let's assumea hypotheticalexample of where and how people are exposed and improve- a study with 2000 subjects,of whom 1000 are ments havebeen made to methodsfor assessingthe exposed to an environmental pollutant and the levels of exposure,its variability and the determi- other 1000 are unexposed.In the exposed group, nants.New methods have been developedor newly 200 have developednthedisease, while only 100 applied throughout this field, including analytical, have done so in the unexposedgroup (Table 7.1). measurement, modelling and statistical methods. The relative risk in this caseis 2. However, the risk This has led to a considerable improvement in estimatewould be only 1.5 if 20Voof the subjects exposure assessmentin epidemiological studies, were misclassified (20Vo of the exposed become and therefore improvement in the epidemiological unexposedand vice versa; see numbers in brack- studiesthemselves. ets in Table 7.1). This exampleshows that even if In epidemiology,there are different study designs the exposure misclassification is relatively minor (e.g.cohort, case-control) to assessthe association (20Vo),the effect on the relative risk is consider- between exposure and disease (see Chapter 2). able. Of course,this is a fairly simple example, and All the study designs require exposure estimates the effect of exposure misclassification/measure- or exposure indices to be able to estimate the ment error dependson other factors, including the risk associatedwith the substanceof interest, but type of error model (Classical or Berkson error; they may differ depending on the study design. Nieuwenhuijsen,2003). The design and interpretation of epidemiological

Molecular Epidemiology of Chrcnic Diseases, Edited by C. P. Wild, P. Vineis, and S. Garte O 2008 John Wiley & Sons, Ltd 84 EXPOSUREASSESSMENT

Thble 7.1 Hypothetical example of the effect of The basic premise of Armstrong's considera- the relativ, exposuremisclassification. The numbers in brackets tions is that it is better, but more expensive, to main studl 'true' 'approxi- are after 20Eoexposure misclassification measure exposurethan to measure Besides mate' exposure.When the correlation between the ti

)onsidera- the relative allocation of budget for pilot and the ation. A representativesample of members from ensive, to main study. each exposuregroup can be personally monitored, 'approxi- Besides sample size and costs, other considera- either once or repeatedly.If the aim is to estimate rtweenthe tions to be taken into accountin designinga specific mean exposure,the averageof the exposuremeas- rariable is exposureassessment strategy for an epidemiological urements is then assigned to all the members in rroximate' study are, for example, accessibilityto the subjects that particular exposuregroup. Alternatively, other : is small. and availability of tools and measurementmethods, exposure estimatescan be assignedto the groups, rf measur- particularly for historical assessments.For exam- e.g. data from ambient air pollution monitors in er than the ple, many environmental epidemiological studies the area where the subjects live. Ecological and y using the use routinely collected health outcome data from individual estimatescan be combined,e.g. in the I be more subjectsfor whom only postcode(or zip code) loca- case of chlorination by-products, where routinely :compared tion of residenceis available and no contact with collected trihalomethane measurementsproviding ic relative the subject can be made. Questionnaires,personal ecologicalestimates are sometimescombined with the sample or cannot be used or conducted in individual estimateson actualingestion, showering ) detect an this case and one is often restricted to modelling and bathing (Nieuwenhuijsenet al.2O00). is the cor- of environmental concentrationsat the location of Intuitively, it is expected that the individual 'true' I the residence.A further disadvantageof this is that no estimatesprovide the best exposure estimatesfor Lga subject or little information is available where the subject an epidemiological study. This is often not true, of disease spends hisArer time outside the residence, which however, because of within-subject variability in measuring may be important for obtaining information on total exposureand the limited number of sampleson each :ctively, Per exposure. Estimating past exposure of subjects is individual. In general, in epidemiological studies, 'gssedas: difficult and the tools availableare sparce.Subjects individual estimates lead to attenuated, although often tend to recall only limited information and per- more precise, health risk estimatesthan ecological rJl sonal monitoring is not possible.Biomonitoring is estimates. The ecological estimates, in contrast, often not informative becausethe biological half-life result in less attenuationof the risk estimates,albeit 'approxi- e for many substancesis too short to provide helpful with some loss of precision (Kromhout et al. 1996; is equal to information on pastexposures, and thereforeone has Seixas and Sheppard 1996; Heederik et al. 1996; rproximate' to rely on some form of environmentalmodelling. Nieuwenhuijsen 1997). These differences can be 'true' rf the A further consideration is whether to obtain explainedby Classicaland Berkson-typeerror mod- ing subjects individual or group estimates. In the individual els. The betweengroup, betweensubjects and within ble is more approach, exposure estimates are obtained at the subject variance,can be estimatedusing analysisof rf including individual level, e.g. every memberof the popula- variance(ANOVA) models and this information can 'true'exPo- tion is monitored either once or repeatedly.In the be usedto optimize the ability to detectan exposure- ris or even a group approach,the group is first split into smaller responserelationship, e.g. by changingthe distribu- 'exposure n the choice subpopulations,more often referred to as tion of exposure groups (Kromhout and Heederik easurements groups', based on specific determinants of expo- 1995; Nieuwenhuijsen 1997; van Tongeren et al. information sure, and group or ecologicalexposure estimates 1997).In this casethe aim is to increasethe contrast by question- are obtained for each exposuregroup. In environ- of exposurebetween exposuregroups, expressed as luding studY mental epidemiological studies, exposure groups the ratio between the between group variance and to the cost of may be defined, e.g. on the basis of presence or the sum of the between-and within-group variance, ?. absenceof an exposuresource (such as gas cooker while maintaining a reasonable precision of the lposurevari- or smoker in the house),distance from an exposure exposureestimates of the exposuregroups. edge of the source(such as roads or factories) or activity (such 'true' nd the as playing sport or not); in occupational studies 7.3. EXPOSUREPATHWAYS AND ROUTES on needs to exposure groups are often defined by title. ling the full The underlying assumptionis that subjects within Generally there are a number of different pathways r total studY, each exposure group experience similar exposure of exposure to a given contaminant, e.g. food, guidancefor characteristics,including exposurelevels and vari- indoor and outdoor air, water, soil, workplace, EXPOSUREASSESSMENT

and these may all need to be considered to obtain mostly be metabolized rapidly in the liver and 7.5. EXPO estimates of total exposure for subjects. not appear in blood, uptake through inhalation MEASURE Furthermore, there are three possible exposure and skin increasesthe blood levels substantially. Different to, routes for substances:inhalation through the res- Furthermore, metabolic polymorphisms may lead and statisticz piratory system; ingestion through the gastrointes- to different dose estimatesunder similar exposure ing exposurt tinal system; and absorption through the skin. The conditions. direct and in exposureroute(s) of a substancedepends on e.g. The main the biological, chemical and physical characteris- 7.4. EXPOSUREDIMENSIONS erally to obl tics of the substances.the location and the activi- relevant exp ties of the person. Inhalation and deposition of The exposure to a substancecan depend upon the and cost-eff particles through the depends following factors: costofthe e; on the particle diameter and the charac- with increas teristics of the person. Smaller particles are more Duration (e.g. in hours or days) or amount (kg/ fore the asse often inhaled and penetrate deeper into the lungs. day ingested) cost on one Furthermore, inhalation depends on the breath- (e.g. in mg/m3 in air or mgA in other (Arms ing rate of the subject; those carrying out heavy water) method dePr work may inhale much more air and breathe more Frequency(e.g. times/week) often, on tht deeply (20 Umin for light work vs. 60 l/min for Subjects heavy work). It is important to recognize that there may be classified tc Skin absorption can play an important role for considerable variability in these factors, both tem- scale, for ex uptake of substancessuch as solvents, pesticides porally and geographically,which can be exploited and trihalomethanes.Trihalomethanes are volatile by epidemiological studies. Any of these factors Yes/no compounds that are formed when water is chlorin- can be used as an exposure index in an epide- No, low, t ated and the chlorine reacts with organic matter in miological study, but they can also be combined to the water. In this context there are a number of pos- obtain a new exposureindex, e.g. by multiplying This can be sible exposurepathways and routes (Figure 7.1). duration and concentration to obtain an index of cumulative exposure.The choice of index depends The main pathway of ingestion is generally drink- ' Expert a ing on the health effect of interest.For substancesthat tap water or tap water-baseddrinks (e.g. tea, team dec coffee cause acute effects, such as ammonia (initation), and squash). Swimming, showering, bath- the subje ing, the short-term concentration is generally the most and dish washing may all result in consider- e.g. livel able relevant exposure index, while for substancesthat uptake through inhalation and skin absorption soil or n cause chronic effects, such as (cancer), and, for the former three, ingestion to minor extent. cal studi Water standing or flushing in the toilet may lead to long-term exposure indices, such as cumulative uptake by inhalation, through volatilization of the exposure, may be a more appropriate exposure chloroform. The total uptake of trihalomethanes index. may be assessedusing the concentration measured t f in exhaled breath or serum. Exposwepathways Exposureroutes Biomarker Direcl met I In the human body, the uptake, distribution, -t- transformation and excretion of substancessuch Drinkingtap water/ _ ,_r- ,- Biolocical as trihalomethanes can be modelled using physi- I I Tapwarer based drinki): rng".tion- _---- I monitoringI ologically-based pharmacokineric (PBPK) models -T- -, ---7 ,.ru_ (Nieuwenhuijr"n zoo3y.These models are becom- :n::"""* S ,/t Skinabsorption 14 Exhated'breath I*-r-- PBPK modelling I ing more sophisticated, although they are still l"*t"t =*i rwrmmrns<-rK rarely used in environmental epidemiology. They // Dishwashing Inhalation/ I can be usedto estimatethe contributionof various - F"--"d"r'l expoSurepathwaysandroutestothetotaluptakeToilet and model the dose of a specific target organ. For Figure 7.1 Examplesofexposure pathways, routes and Figure 7.2 example, where ingested trihalomethanes may biomarkersfor trihalomethanes assessment. EXPOSURECTASSIFICATION, MEASUREMENT OR MODETI-ING 87 liver and 7.5. EXPOSURECLASSIFICATION, tance from a (point) source, such as a factory inhalation MEASUREMENTOR MODELLING (Dolk et al. 1999), radio and TV transmitters (Dolk et al. 1997), incinerators (ElIiott et al. stantially. Different tools, such as questionnaires, monitors 1996), emissionsfrom roads (Livingstone et al. may lead and statisticaltechniques, are available for classify- 1996;English et al. 1999;Hoek e/ aL.2002)or exposure ing exposures.The methods are often divided into landfill (Elliott et aI.20Ol), while othershave direct and indirect methods (Figure 7.2). categorized exposure by industrial sources, The main aim of an exposureassessment is gen- land use or urban zone (Barbone et al. 1995). erally to obtain accurate,precise and biologically . Self-assessmentby questionrnire. The subject relevant exposure estimates in the most efficient I upon the in the study is asked to fill out a questionnaire and cost-effective way, As discussed before, the in which he/sheis askedabout a particular sub- cost of the exposureassessment generally increases stance,e.g. pesticides.Questionnaires are often with increasing accuracy and precision, and there- nount (kg/ usedto ask a subjectwhether he/she is exposedto fore the assessmentis often a balancing act between a particular substanceand also for an estimation cost on one side and accuracyand precision on the or mg/l in of the duration of exposure. They are also often other (Armstrong 1996).The choice of a particular used in nutritional epidemiology,in the form of method dependson the aim of the study and, more food frequencyquestionnaires or diaries. often. on the financial resourcesavailable. Subjects in an epidemiological study tT. b: re may be Questionnairescan be used not only to ask the classified to a particular substanceon an orqrnal both tem- subject to estimate his/trer exposure but also to scale, for example as exposed: a exploited obtain information related to the exposure, such gse factors as where people spend their time (time micro- Yes/no an epide- environment diaries), work history, including the No, low, medium,high rmbined to and tasks they carry out, what they eat and nultiplying drink, and where they live. These variables could This achievedby: n index of can be be used as exposure indices in epidemiological ex depends studies or translated into a new exposure index, . Expert assessment.A member of the research ;tancesthat e.g. by multiplying the amount of tap water people team decides,based on prior knowledge,whether (), drink and the contaminant level in the tap water to the subjectin the study is exposedor unexposed, Lythe most obtain the total ingested amount of the substance. e.g. lives in an area with highly contaminated itances that Expert and self-assessmentmethods are generally soil or not. Some environmental epidemiologi- rs (cancer), the easiestand cheapest,but can suffer of a lack of cal studies have used simple proxies, e.g. dis- cumulative objectivity and knowledge and may therefore bias e exposure the exposureassessment. Both experts and subjects may not know exactly what the subjectsare exposed to or at what level, and therefore misclassify the exposure, while diseasedsubjects may recall cer- tain substancesbetter than subjectswithout disease (recall bias) and causedifferential misclassification, Environmenhl leading to biasedhealth risk estimates. Monitoring/ A more objective way to assessthe exposure is modelling Serum by measuring the level of the contaminant in air, Exhaled breath water or food. Here are some examples of such measurements:

. Levels of outdoor air pollution can be measured by ambient air monitors (Dockery et al. 1993; ys, and Figure 7.2 Different approachesto humanexposure routes Katsouyanniet al. 1995;Dockeryand Pope 1997). assessment.PBPK, physiological pharmacokinetic 88 EXPOSUREASSESSMENT

These monitors are placed in an area and meas- substance of interest, rather than the exposure. Modelling ure the particular substanceof interest in this Biomonitoring can be very informative, particu- area.Subjects living within this area are consid- larly for substancesthat have multiple pathways Determin ered to be exposedto the concenffationsmeas- and routes. A major drawback is often the fairly the mode ured by the monitoring station. This may or may short biological half-life of many substances, variables not be true, depending on, for example, where which makesthis methodonly useful for estimat- edge of t the subject in the study lives, works or travels. ing exposures/doses.However, infectious cal mech The advantageof this method is that it could diseaseepidemiology has extensivelyemployed (Brunekrr provide a range of exposure estimates for a large exposure biomarkers. In the case of infections Stochasti population. Nowadays there are many monitors and cance! developmental work to establish which th that are routinely monitoring air pollutants for validated laboratory assays for antibodies to elled bet' regulatory purposes,particularly in cities in the viral or bacterial antigens,e.g. hepatitis viruses, necessari developedworld but also elsewhere. human papilloma virus and Helicobacter pylori, of the u lzvels of air pollution can be measured by hasbeen central to understandingthe aetiological biologica personal exposure monitors (Magnus et al. role of these agentsin epidemiological studies. 1998; Kriimer et a\.2000; Magari et al. 2002; Furthermore, the more recent emphasis on DNA Brant et al. 2OO5). These monitors are light- and protein adducts has undoubtedly contributed Box 7.1 D devices that are worn by the subject in substantiallyto establishingthe biological plausi- Hodgsonei the study. They are often used in occupational bility of exposure-diseaseassociations. Examples persion in I studies and are becoming more frequently used include investigationsof the associationbetween and diamet in environmental studies.The advantageof this genotypesfor carcinogen-metabolizingenzymes local geogl method is that it is likely to estimate the sub- and adduct levels; or the use of biomarkers as pollutants r ject's exposure better than, for example, ambi- modifiable end points in short-term interven- an associat ent air monitoring. The disadvantageis that this tion studies (Groopman et al. 1999). [n contrast, proxy for e method is often labour intensive and expensive the application of exposure biomarkers of this and can often only be used for relatively small natureto aetiological studiesis far more limited. populations. However, personal monitoring is Aflatoxin is perhaps the prime example in which ideal for validation studies, e.g. of modelled the exposure biomarker permitted categorization exposureestimates. of this environmental agent as a human car- cinogen (Qian et al. 1994: Wang et al. 1996), lzvels of water pollutants and soil contaminants othersinclude polycyclic aromatichydrocarbon- can be estimated by taking water samples and DNA adducts in lung cancer (Tang et al.200l) soil samples,respectively, and analysing these and arylamine- haemoglobinadducts in bladder for a substanceof interest in the laboratory. Often cancer (Gan et al. 20A$. Further developments these need to be combined with behavioural fac- are envisaged(Wild er al.2Cf.5). tors, such as water intake, contaminated food intake or hand-to-month contact. to obtain a level The measurementof exposureis generally expen- of exposure (Nieuwenhuij sen et al. 2000). sive, particularly for large populations, and, as Levels of uptake of the substance into the body mentioned above, can be restricted due to inac- can be estimatedby biomonitoring, e.g. for lead cessibility to subjects or the need for a historical (Bellinger and Schwartz 1997; Nieuwenhuijsen assessmentof exposure,rather than assessmentof 2003). Biomonitoring consistsof taking biologi- current exposure.It may be very useful for valida- cal samples,such as urine, exhaledbreath, hair, tion purposes. adiposetissue or nails, e.g. the measurementof Modelling of exposurecan be carried out, pref- lead in serum. The samples are subsequently erably in conjunction with exposuremeasuremenrs, 4 analysed in a laboratory for the substance of either to help to build a model and/or to validate a interest or a metabolite. Biomonitoring is expect- model. It is in particularly important that the model Figure 7.3 ed to estimate the actual uptake (dose) of the estimatesare validated. distanceas €,' t i - EXPOSURECLASSIFICATION, MEASUREMENT OR MODELLINC 89 , rxposure. i. Modelling can be divided into: ables. Examples are regression and Bayesian ,, particu- modelling. e (i.e. pathwaYs Deterministic modelling physical), in which between For regression modelling, a statistical regression the fairlY ?, the models describe the relationship knowl- model can be constructed,expressed in the form: rbstances, variablesmathematically on the basis of and/or biologi- )r estimat- i edge of the physical, chemical : p2var1* Ln(C1;) Fo * ppari * E infectious I cal mechanisms governing these relationships (Brunekreef expo- employed 1999;see Box 7.1). in which ln(C11)denotes the log-transformed (i.e. in pe level, var, infections i Stochastic modelling statistical), sure concentration, the background mod- p, the establish g,' which the statistical relationships are the potential determinant of exposure, do not providing the mag- bodies to elled between variables. These models regression coefficient of var, i knowledge variable with is viruses, necessarily require fundamental nitude of the effect, and E a random (for 'ter pylori, of the underlying physical, chemical and/or mean 0, often called the error term examples, the vari- etiological biological relationships between seeBoxes 7.2 and7.3). al studies. Lson DNA :ontributed Box 7.1 Deterministicmodelling plausi- ical Hodgsoner cl. (2006)used an atmospheric dispersion modelling system (ADMS) to assessmercury dis- .Examples persionin Runcornin north-westernEngland. ADMS usesalgorithms that take account of stackheight rn between anddiameter, volume flow rate, and emission rates of pollutants,as well asmeteorology, enzymes Lg local geography,atmospheric boundary layer and deposition parameters, to calculateconcentrations of as markers pollutantsat groundlevel. Three authorized processes were included in the model,a chloralkaliplant, r interven- an associatedmulti-fuel power station, and a coal-firedpower station. Compared to usingdistance as a ln contrast, proxy for exposure,the modelidentified a muchsmaller exposed population (Figure 7.3). iers of this ore limited. rle in which Fgorization numan car- t al. 1996), rdrocarbon- et al.20Ol) :s in bladder :veloPments rally expen- Ins, and, aS lue to inac- a historical rsessmentof Hg (ng/m3 (Average 1998-2001) rl for valida- <3 planland multifuel powerslation E Chloratkali Effi >3-<4 I 4-10 ed out, Pref- I <10 0 1 2 3 4 5kilometers easurements, ----- Point sources to validate a Figure 7.3 Comparisonof modelledexposure output (average1998-2001) to exPosureanalysis based on rat the model distanceas a proxy for exposurefor a studyofmercury in the north-westofEngland 90 EXPOSUREASSESSMENT

Box7.2 Regression modelling rypes,ar among ! Harris et al (2002) measured 2,4-o (2,4-dichlorophenoxyacetic acid), mecopop [2-(4-chloro-2 specihc methylphenoxy)propionic acid, MCPPI and dicamba (3,6-dichloro-o-anisic acid) in urine (two con- with the secutive 24 h periods) collected from a group of 98 professional turf applicators from 20 companies were tak across south-westernOntario. The group also filled out questionnairesto acquire information on all zonessu known variablesthat could potentially increaseor decreasepesticide exposureto the amount handled, sons (Fi to build models for epidemiological studies. They used linear regression to assessthe relationship than wh, between the of the substancesin urine and the questionnairedata. They found that the volume of pesticide (active ingredient) applied was only weakly related to the total dose of 2,4-o absorbed(R2 : 0.21). TWo additional factors explained a large proportion of the variation in measured pesticide exposure,the type of spray nozzleused and the use of gloves while spraying.Individuals who useda fan-typenozzle had significantly higher dosesthan those who useda gun-typenozzle. Glove use was associatedwith significantly lower doses.Job satisfactionand current smoking influenced the dose but were not highly predictive. In the final multiple regressionmodel, it was concluded that approxi- mately 64Voof the variation in doses could be explained by the small number of variables identified (Table 7 .2). Biological monitoring in this casewas important to be able to determine the true effect of wearing protective equipment, such as gloves. This study provided extremely useful information for epidemiological and health risk assessmentstudies, which could focus on obtaining information on theseparticular variables in a larger population.

Table 7.2 Regressionmodels predicting the log of total dose of 2, 4 d in 94 volunteers(l = 0.64) Variable Estimate SE p Value Partial I Intercept - 1.09 0.01 0.29 Log spray 0.96 0.r2 0.001 o.M Nozzle r.37 0.23 0.001 0.29 Glove wear - 1.50 0.25 0.001 0.29 Satisfaction -0.39 0.17 0.021 0.06 Smoke 0.51 0.22 0.02 0.06 Reprinted by permission from Macmillan Publishers Ltd:Haris et al. (2002) Joumal of ExposureScience and Erwirorvrwntal Epidemiology.

Box 7.3 Modelling using sparse data

Trihalomethane(THM) concentrationswere used as the marker for chlorination by-products in a study of chlorination by-products and birth outcomes.In the UK, where the study was conducted,water sam- ples are routinely collected and analysedfrom each water zone (population up to 50 000 people), using random samplesat the tap (an averageof four measurementsper zone). Becauseof the small number of THM measurementsin some water zones,the need for quarterly (3-monthly) estimates(to allow for trimester-weightedexposure estimates) and the problem of measurementsbelow the limit of detection, it was necessaryto model the raw THM data to obtain more robust estimatesof the mean THM con- centration in each zone. This was done using a hierarchical mixture model in the software WinBUGS (Bayesianinference using Gibbs sampling) (Spiegelhalteret at. 1996),as describedin detail elsewhere (Whitaker et aL.2005). A three-componentmixture model was f,rtted,in which zoneswere assumedto 'ground', 'lowland belong to one or some mixture of three components,which were labelled surface' 'upland and surface' waters (the componentsmay not strictly correspondto thesethree water source EXPOSURECLASSlFICATION, MEASUREMENT OR MODETTINC 91

types, and simply aimed to group waters with similar THM prof,rles,which are more likely to be shared among water of the samesource type; Figure 7.4).T}:,Lehierarchical model was assignedover the zone- 'borrow' rhloro-2 specific mean individual THM concentrations,enabling zonesto information from other zones vo con- with the same water source type. This resulted in more stable estimatesfor zones where few samples npanies were taken. Seasonalvariation was taken into account by estimating a quarterly effect common to all n on all zonessupplied by the samesource type. The modelling provided estimatesof exposuresfor various sea- randled, sons (Figure 7.5). The modelled exposureestimates provided a better exposure-responserelationship tionship than when using estimatesbased on the mean of the raw THM concentrationsfor each zone. rnd that Seasonaleffect of 2,4-o reasured ralswho ro'Lowland love use Region: |-eo,,'.,dl l- surface waters t-**r-lsurfacewaters the dose lwatersllsurface I I approxi- /\ I lentified /\ I of effect Water ation for zones: all4 hieh I nion I I | | I I on ation I rHnltsrw | | bromoformI I chloroformI il t lil\ iltl

Tap lll//l\//1\//t\ samples: XXX XXXX XXXX XXXX Figure 7.4 Hierarchical mixture model to estimate the water zone means of THMs by water source, using tap water samplesand applying a common seasonaleffect

SevernTrent Water

January- March THMExposure score April- June THM Exposure score ;## i. lLow(0-<30) :1.. Low(0 - < 30) iFi Medium(>= 30 - < 60) SIJ Modium(>= 30. < 60) r Hish ( '= 60) r Hish ( >= 60) &"'i

tEt ,#,a *,r #

lff,"*

study Oclober - December Lna July- September THMExposure score THMExposure score ater sam- :-,Ld(0-<30) I : Low(0-<30) le), using Li lr€dm (>= 30 - < @) ffi Medium(>= 30 " < 80) ll number - High ( >= 60) I Hish ( '= 60) allow for detection, 'HM con- [inBUGS elsewhere tsumed to lmperial college London d surface' Figure 7.5 Modelled THM levels by water zone, using Bayesian mixture modelling r source EXPOSUREASSESSMENT

A problem in exposure assessmentis that often spatial form and to analyse the data geographi greatly faci few routinely collected measurementsare available cally. It is these capabilities that give GISs their (Bayer-Ogl to model exposure estimates and therefore more special power in relation to exposure assessment. The aPP sophisticatedstatistical techniquesneed to be used, However, there are often some problems in acquir- where ther as was demonstratedin a study of chlorination by- ing the data neededto carry out geographic meth- pollution a products(see Box 7.3). ods of exposure assessment.Also, a GIS requires ing networ In recent years, in environmental epidemiology that all data be georeferenced.Various GIS and fulfilled. Ir many of the modelling methods have been greatly geostatisticaltechniques have been used to model plement th' strengthenedby the use of geographical informa- local pollution patterns, on the basis of the moni- use of covi tion system(GIS) techniques(Nuckols et a|.2004: tored data, e.g. using inverse distance weighting, monitored Briggs 2005). Looked at simply, GISs are com- kriging or focal sum methods. These essentially obtained tl puterized mapping systems. GISs, however, can fit a surface through the available monitored data, Oglesby 2 do more than simply map data. They also provide in order to predict pollutant concentrations at methods et the capability to integrate the data into a common unmeasuredsites. It is an approach that has been People and this m importanct Box 7.4. GlS-basedregression modelting Helsinki (J sion mode Regression techniques were used as part of the (SAVIAH) study, for example, to model exposures centrations to NO2 (as a marker for traffic-related air pollution) in four study cities (Briggs et at. l99i). Data day. These from 80 monitoring sites were used to construct a regression equation, using information on road where peo traffic (e.g. road network, road type, traffic volume), land cover/use, altitude and monitored NO2 build up a data. The results showed that the maps produced extremely good predictions of monitored pollution out the d: levels, both for individual and for the mean annual concentrations,with * = 0.794.87 across 8-10 area is tha reference points, although the accuracy of the predictions for individual periods was more variable their envir, (Figure 7.6). Subsequentlyit was shown that regressionmodels developed in one location could be (GPS) witl applied successfully,with local calibration using only a small number of sites, to other study areasor type of stu periods (Briggs et al.2000). More recently, the same approach has been further developedto assess 2003). Th exposuresto particles in a number of different cities as part of the TRAPCA study (Brauer et al. 2003), and to model traffrc-related air pollution in Munich (Carr et a|.2002). the limitat the satellit shielding 1 C = 38.5+ .003705Traff+ 0.232'Land- 5.673.lo91OAlt wnere: crete, stee. Traff= 18"Tvol0-40+ Tvol40-300 extent vet Land= 8*HDH0-300+ Ind0-300 pollution r spendthei estimates ogy can br All thes and often ' index. It it the expost therefore estimated, in the US data and r such as lat Figure 7.6 ModelledNO2 levels in Huddersfield,using the SAVIAH approach of the hie fairly got RETROSPECTIVEEXPOSURE ASSESSMENT 93 ographi- greatly facilitated through the developmentof GIS subjects, while those at the bottom may not be lSs their (Bayer-Oglesby 2004). helpful in the interpretation of exposure levels in essment. The approach appears to work well in areas an epidemiological study. Of course, there are still n acquir- where there is relatively gentle variation in air other issuesthat are important in the ranking, e.g. Lic meth- pollution and/or where the density of the monitor- many area measurementsmay still be better than a requires ing network is high; conditions that are often not few personal exposure measurementsif exposure GIS and fulfilled. In other situations, it is helpful to sup- arises from a general source in the area. to model plement the available monitoring data through the use of covariates,i.e. variables that correlate with he moni- 7.6. RETROSPECTIVEEXPOSURE 'eighting, monitored concentrationsand can be more readily ASSESSMENT ssentially obtained than quantitative measurements(Bayer- rred data, Oglesby 2004). Both cokriging and regression A special challenge in epidemiological studies ations at methodsenable this (seeBox 7.4). is studying diseasewith a long latency time, e.g. has been People typically move about during the day cancer; in such casesit is not the current exposures and this mobility can greatly affect exposure.The that are of most interest, but those that occurred importance of this was clearly shown by a study in in the past. A reconstruction of historical expo- Helsinki (Kousa et al. 2002), which used disper- sure, often referred to as retrospective exposure sion modelling to predict nitrogen dioxide con- assessment,is therefore needed,and often involves iposures centrations acrossthe city at different times of the some extensive modelling and specific exper- 7). Data day. These were then overlaid onto data showing tise. Retrospectiveexposure assessment is difficult on road where people were at different times, in order to because there are often many changes occurring red NO2 build up a picture of exposure variations through- over time. nllution i.t out the day. An interesting development in this A good recent example ofretrospective exposure rss 8-10 area is that it is possible to track people through assessmentis provided by the study of air pollution variable T. their environment using global positioning systems and lung cancer in Stockholm, where the investi- could be fi'l (GPS) with enough resolution to be useful for this gators used emission data, dispersion models and areasor type of study (Phillips et al.200l; Elgethun et al. GIS to assesshistorical exposuresto air pollutants to assess 2003). There are some restrictions as a result of and compared these estimates with actual meas- ,1.2003), the limitations in the technology, e.g. reception of urements (Bellander et al. 2001). For NO2 they the satellite signals can be adversely impacted by used a detailed regional database,which included shielding from buildings of certain materials (con- information on approximately 4300 line sources crete, steel), electrical power stations, and to some related to traffic and 500 point sources,including extent vehicle body panels. However, combining major industries and energy plants as well as small pollution maps with information on where people industry and ferries in ports. Limited diffuse emis- spendtheir time may greatly improve the exposure sion sources, e.g. air traffic and merchant vessels estimates if further improvements to the technol- in commercial routes, were treated as arca sources, ogy can be made. and several population density-related sources, ii All these different approachesare not exclusive such as local heating, were mapped as grid sources, and often are combined to obtain the best exposure as were work machine emissions. They collected :r:;! index. It is often difficult or impossible to measure information on the growth of urban areas, the the exposureto the actualsubstance ofinterest and developmentof district heating, and the growth and 'exposure therefore exposure to an surrogate' is distribution of the road traffic over time. They used estimated.The National ResearchCouncil (NRC) the Airviro model, togetherwith population data, to in the USA came up with a ranking of exposure derive population-weighted averageexposures dur- data and surrogate measuresaround point sources, ing 1955-1990. In the case of traffic-related NO2, such as landfill sites Clable 7.2). The data at the top exposures were seen to increase over this period of the hierarchy shown in this table provide some from about 15 pglm3 in 1955 to about 24 1s"g/m3 fairly good information on the exposure of the in 1990, showing the effect of increasing traffrc 94 EXPOSUREASSESSMENT

Thble7.3. Hierarchy ofexposuredata and surrogatesfor fixed sourcecontaminants Referenc Type of data Approximation Armstrong, to actual exposure resoruces study.An 1. Quantified personal measurement Best BarboneF, 2. Quantified area measurementsin the vicinity of the residenceor sites of activity tion and 3. Quantified surrogatesofexposure (e.g. estimatesofdrinking water use) l4l: 116 4. Distance from the site and duration ofexposure Bayer-Ogle 5. Distance or duration of residence Exposure 6. Residenceor employment in the geographical area in reasonableproximity to the site iras.uu.nl where exposure can be assumed Bellingert 7. Residenceor employment in a defined geographical area (e.g. a county) of the site Worst adults. I Steenlan Adaptedfrom NRC (1991). New Yor BellanderI geograPl volumes. In contrast, modelled SO2 exposuresfell 7.8. QUALTTYCONTROL ISSUES historica from ) 90 pglm3 to < 20 pg/m3 as a result of househe A well-designed and well-thought-out exposure 633 improvements in fuel technology, emission con- 109: assessmentstrategy carried out by well-trained per- BrantA, B trols and a shift to district heatins. sonnel is essentialfor a successfulexposure assess- distribut. ment. Issues such as cost, feasibility, accuracy, piratory precision, validity, sample size, power, sensitivity, markett 7.7. V ALIDATIONSTUDIES specifrcity, robustnessand reproducibility always BrauerM, In epidemiological studies it is often not pos- need to be addressed(e.g. during sampling, storage long-ten atr sible to obtain detailed exposure information on and analysis), while feasibility and pilot studies tions: informat each subject in the study.For example,in a large always need to take place before the actual study. Briggs DJ cohort study it is generally not feasible to take Any form of bias (e.g. bias in sampling, selection, urban a measurementson each subject and administer a participation, monitoring, information, measure- approaci detailed exposurequestionnaire. In this caseit is ment error and exposure misclassification) should BriggsDJ. desirableto carry out a small validation study on be avoided where possible, or if it takes places it sion-bas a subset of the population that is representativeof should be clearly described. pollutior the larger population. Ideally this will be carried Clear protocols for sampling, storageand analy- urbaner out before the main study starts and can make use sis, including quality control should be written BriggsD. of information from the literature. Questions in and be available at any time and researchersin the time in the questionnairecould be validated with meas- study should be properly trained. Potential sources Environ Brunekree urements and exposure models could be con- ofbias should be addressedat every stage. Epidemt structed. The exposure assessmentin the whole Control measurement,e.g. air pollution frlters Health t population could focus on key questions that that are not exposed but are otherwise treated as Carr D, have a large influence on the exposure estimates (S-lOVo exposedfilters, should be included oftotal Modelli and thereby reduce the length of the question- samples) particularly where measurements are concent naire. Information on key determinants will also closeto the detectionlimit. tics. En provide a better understanding of the exposure Samplers can measure with different accuracy, DolkH,I and how it may affect exposure-response rela- e.g. over- or under-samplingthe true level, and this dencer tionships in epidemiological studies. Besides should be addressedwhen different samplers are GreatB assessingthe validity of the exposure surrogates, used in order to reduce or avoid bias. This can be 145:l0 the reproducibility of the surrogates can also be easily done by comparativesampling and adjusting Dolk H, T evaluated in the subsample. for any difference observed. residenl Environ REFERENCES

References Dockery DW, Pope A III, Xu X et al. 1993. An asso- ciation between air pollution and mortality in six US power in allocating ion Armstrong, B. 1996. Optimizing cities.NEngl J Med329:1753-1759. an epidemiologic tosure resourcesto exposure assessmentin Dockery DW, Pope CA. 199'7. Outdoor air I: par- study. Am J Epidemiol 144: 192-19'1 . ticulates. ln Tbpics in Ewironmental Epidemiology, Air pollu- Barbone R Boveni M, Cavallieri F et al. 1995. SteenlandK, Savitz D (eds). Oxford University Press: J Epidemiol tion and lung cancer in Trieste, Italy. Am New York; 119-166. l4l: 1161-1169. Elliott P, Arnold R, Cockings S et al. 2000 Risk of Aimet Bayer-OglesbyL, Briggs D, Hoek G et a|.2004. mortality, cancer incidence, and stroke in a population Exposure Assessment Workgroup Report: http://airnet. potentially exposed to cadmium. Occup Environ Med Aas.uu.nVproducts/pdf/airnet-wg 1-exposure-report.pdf 57:64'7448. and Bellinger D, Schwartz J. Effects of lead in children Elliott P. Shaddick G. Kleinschmidtl et al. 1996. Cancer adults. In Topics in Environmental Epidemiology, incidence near municipal solid waste incinerators in Press: SteenlandK, Savitz D (eds). Oxford University GreatBritain. Br J CancerT3:702-710. New York; 314-349. Ellion R Briggs DJ, Morris S et aL 2@1. Risk of adverse BellanderT, BerglindN, GustavssonP et al.200l.Using birth outcomesin populations living near landfill sites. individual geographic information systems to assess Br Med J 323:363-368. and historical exposure to air pollution from trafEc English P, Neutra R, Scalf R et aI. 1999. Examining Perspect house heating in Stockholm. Environ Health associations between childhood asthma and traffrc xposure 109: 633-639. flow using a geographic information system. Environ nedper- Brant A, Berriman J, Sharp C et a\.2005. The changing Health Perspect 107; 76l-7 67. res- ) ASSCSS- distribution of : work-related Gan JP, Skipper PL, Gago-Dominguez M et al. 2004. ccuracy, piratory symptoms and specific sensitisationin super- Alkylaniline-hemoglobin adducts and risk of non- rsitivity, market bakery workers. Eur Resp J 25: 303-308. smoking-related bladder cancer. J Natl Cancer Inst Estimating ' always Brauer M, Hoek G, van Vliet P et a\.2003. 96:1425-1431. long-term averageparticulate air pollution concentra- end of , storage Groopman JD, Kensler TW. 1999.The light at the tions: application of traffic indicators and geographic : studies the tunnel for chemical-specific biomarker: daylight or information systems.Epidemiology 14: 228-239. 20: l-l l. al study. headlight? Carcinogenesis Briggs DJ, Collins S, Elliott P et al. 1997. Mapping Harris SA, Sass-Kortsak AM, Corey PN et al. 2002. election, urban air pollution using GIS: a regression-based Development of models to predict dose of pesticides neasure- 18. approach.Int J Geogr Inform Sci ll: 699-:7 in professional turf applicators. J Expos Anal Environ regres- L)should Briggs DJ, de Hoogh C, Gulliver J et al. 2000' A Epidemiol t2: l3Fl44. air placesit sion-based method for mapping traffic-related Heederik D, Kromhout H, Braun W. 1996.The influence pollution: application and testing in four contrasting of random exposureestimation error on the exposure- 151-16'7. rd analy- urban environments.Sci Total Environ2S3: responserelationship when grouping into homegene- space and : written Briggs D. 2005. The role of GIS: coping with ous exposurecategories. Occup Hygiene 3:229-241. J Toxicol )rs in the time in air pollution exposure assessment' Hoek G, Brunekreef B, Goldbohm S et al. 2002. Environ Health A 6E:. 1243-1261. Associations between mortality and indicators of I sources ggg.Exposureassessment. BrunekreefB.1 InEnvironmental traffic-related air pollution in The Netherlands: a Epidemiology: A Textbookon Study Methods and Public rn filters cohort study. Lancet 360: 1203-1209. Health Applications. WHO: Geneva. Hodgson S, Nieuwenhuijsen MJ, Colvile R et al. \n reated as 2002. Can D, von Ehrenstein O, Weiand S et al. press.Identifying populations at risk of mercury expo- Voof total and soot Modelling annual benzene, toluene, NO2, sure from industrial emissions. Occup Erwiron Med- rents are concentrationson the basis of road traffic characteris- D, Spix C et al. 1995. Short- 'tics. Katsouyanni K, Zmirou Environ ResA90: 111-118. term effects of air pollution on health - a European inci- accuracy, Dolk H, Elliott P, Shaddick G et aI. 1997. Cancer approach using epidemiologic time-series data - the in l, and this dence near high power radio and TV transmitters Aphea project - background, objectives, design. Eur Epidemiol rplers are Great Britain: II. All transmitter sites.Am J RespJ 8: 1030-1038. ds can be 145: 10-17. KousaA, Monn C, Rotko T er al. 2001. Personalexposwes H. Thakrar B, WallsP et al. 1999' Mortality among adjusting Dolk to NO2 in the E)GOLIS study: relation to residential residents near cokeworks in Great Bitan. Occup indoor, outdoor and workplace concentrations in Basel, Erwiron Med 56:3440. Helsinki and hague. Atmos Environ 35:. 3405-3412. 96 EXPOSUREASSESSMENT

Kriimer U, Koch T, Ranft U et a1.2000. Traff,rc-related cancer risk in Shanghai, People's Republic of China. air pollution is associatedwith atopy in children living Cancer Epidemiol Biomarkers Prevent 3: 3-10. in urban areas.Epidemiology ll 64-70. Ranft U. Miskovic P. PeschB et a|.2003. Association Kromhout H, Heederik D. 1995. Occupational epidemi- between arsenic exposure from a coal-burning power ology in the rubber industry; implications of exposure plant and urinary arsenic concentrationsin Priedvidza variability. Am J Indust Med 2i7: I7I-185. District, Slovakia. Environ Health Perspect 111: 889- Cart Kromhout H, Tielemans E, Preller L et al. L996. 894. Estimates of individual dose from current measure- Seixas NS, Sheppard L. 1996. Maximazing accuracy ments of exposure. Occup Hyg 3: 23-29. and precision using individual and grouped expo- Livingstone AE, Shaddick G, Grundy C et al. 1996. Do sure assessments.Scand J Environ Work Health 22: people living near inner city main roads have more 94-tOL. asthmaneeding treatment?Case control study.Br Med SpiegelhalterD, Thomas A, Best N et al. 1996.BUGS J 312: 676477. 0.5; Bayesian inference using Gibbs sampling. Magari SR, Schwartz J, Williams PL et al. 2002. The Manual, Version ii. Available at: http://www.mrc-bsu. association between personal measurementsof envi- cam.ac.uk/bugs 8.1. INTR( ronmental exposure to particulates and heart rate vari- Tang DL, Phillips DH, Stampfer M et al. 2001. Why would o ability. Epidemiology 13:.305-3 10. Association between carcinogen-DNA adducts in as biomarker Magnus P, Nafstad P, @ie L et al. 1998. Exposure to white blood cells and lung cancerrisk in the Physicians cer? To inves nitrogen dioxide and the occurrence of bronchial Health Study. Cancer Res 61: 6708-5712. some terrns. . obstruction in children below 2 years.Int J Epidemiol van Tongeren M, Gardiner K, Calvert I et al. 1997. or vi 27:995-999. Effrciency of different grouping schemes for dust physical National Research Committee (NRC). 1991. exposure in the European carbon black respiratory incidence of Environmental Epidemiology I. morbidity stldy. Occup EnvironMed 54:7l4-719. will consider and Hagardous Waste. National Academy Press: Wang LY, Hatch M, Chen CJ et al. 1996.Aflatoxin expo- is a substanct Washington,DC. sure and risk of hepatocellular carcinoma in Taiwan. the carcinogt Nieuwenhuijsen MJ (ed.). 2003. Exposure Assessment Int J Cancer 67:620425. in this case in Occupational and Environmental Epidemiology. Ward MH, Nuckols JR, Weigel SJ et a1.2000. Identifying tive biologic Oxford University Press:New York. populationspotentially exposedto agriculturalpesticides (as a metabc Nieuwenhuijsen MJ, Toledano MB, Ellion P. 2000. using remote sensing and a Geographical Information (Merriam-W Uptake of chlorination disinfection by-products; a System.Environ Heabh Perspeu 108: 5-12. Carcinoge review and a discussion of its implications for epide- Whitaker H, Best N, Nieuwenhuijsen MJ et al. 2005. most widelY miological studies. J Expos Anal Environ Epidemiol Hierarchical modelling of trihalomethane levels in advanl 10: 586-599. drinking water. ,I Expos Anal Environ Epidemiol 15: major Nuckols JR, Ward M, Jarup L.2004. Using geographic 138-146. many of the ( information systemsfor exposureassessment in envi- Wild CP. 2005. Complementing the genome with an assessment 'exposome': ronmental epidemiological studies. Environ Health the outstanding challenge of environ- urements of Perspect112: 1007-1015. mental exposure measurement in molecular epide- generalenvil Qian GS, Ross RK, YUMC et al. 1994.A follow-up study miology. Cancer Epidemiol Biomarkers Prevent l4'. vidual expos of urinary markers of aflatoxin exposure and liver 1847-1850. factors. Diel and maY not assessment, specific tYPt to carcinoge on how an iI may be quit by smoking

Molecular EPidet O 2008 John Wil