Methodologic Issues in Epidemiologic Risk …

1 Methodologic Issues in Epidemiologic Risk AssessmentMarkku Nurminen,' Tuula Nurminen, ' and Carlos F. Corvalan2 This paper reviews Methodologic Issues pertinent to theapplication of epidemiology in risk assessment and discussesconcerns in the presentation of results from such an of the health risks associated with occupational andenvironmental exposures involves four phases: hazardidentification, ie, the detection of the potential for agents tocause adverse health effects in exposed populations; exposureassessment, ie, the quantification of exposures and theestimation of the characteristics and sizes of the exposedpopulations; dose-response assessment , ie, the modeling for riskrealization.

2 And risk characterization, ie, the evaluation of theimpact of a change in exposure levels on public health risk assessment process involves limitations of exposuredata, many assumptions, and subjective choices that need to beconsidered when using this approach to provide guidance forhealth policy or action. In view of these uncertainties, wesuggest that the provision of estimates of individual risk anddisease burden in a population must be accompanied by thecorresponding estimates of precision; risks should be presentedin a sufficiently disaggregated form so that populationheterogeneities are not lost in the data aggregation; and differentscenarios and risk models should be applied. The methods areillustrated by an assessment on the health impacts of exposure tosilica.

3 (Epidemiology 1999;10:585-593)Keywords: Epidemiologic methods, exposure estimation, health risk, silica, quantitative risk bodies at the local, national, and international levelsface difficult decisions that would ideally be based on weighingthe health and environmental cost of a technology against itseconomic and social benefits.' This requires that health effects ofenvironmental exposures be quantified, yet data for thisquantification are often limited. Quantitative risk assessment (QRA) must nevertheless be carried out for regulatory fact that the result is often presented as a single number (forexample, excess number of exposed disease cases) may implycertainty, which has obvious appeal among regulators and decisionmakers.

4 Despite its apparent objectivity, QRA is dependent on aseries of assumptions and subjective choices that can have criticaleffects on the resulting risk estimates. Thus, it is of primeimportance that QRA be founded on solid scientific risks in human populations are increasingly beingassessed by the use of empirical data from Epidemiologic studies.'Although usually coupled with retrospective exposureassessments, Epidemiologic studies canFrom the 'Department of Epidemiology and Biostatistics, Finnish Institute ofOccupational Health, Helsinki, Finland, and 2 Department of Protection of the HumanEnvironment, World Health Organization, Geneva, correspondence to.

5 Markku Nurminen, Department of Epidemiology andBiostatistics, Finnish Institute of Occupational Health, Topeliuksenkatu 41 a A,FIN-00250 Helsinki, paper is based on material that appeared in the unpublished document, "Linkageanalysis for environmental health analysis-technical guidelines," released by theWorld Health Organization in 1997 ( 1).The views expressed in this article are those of the authors and do not necessarilyreflect the position of the World Health Organization.(D 1999 by Epidemiology Resources more defensible estimates of likely human health risks thanthose obtained from biologic models based on animal work that evaluated and improved the accuracy of priorestimates of exposure to benzene for a rubber worker cohortfurnishes an The approach quantitatively accounted formultiple relevant factors, such as uptake of benzene due toshort-term, high-level exposure to vapors; backgroundconcentrations in the manufacturing building; and con, tact withthe skin.)

6 The predicted levels of exposure for the process workers,combined with morbidity and mortality data, were used toestimate the carcinogenic potency of nonexperimental Epidemiologic methods, it is fairly easyto demonstrate that occupational groups sustain excess risks athigh exposure levels. However, detecting health risks in relationto nonoccupational environmental exposures is far more disease occurrence is often rare in the low,exposure rangeand, for example, in data analysis, the assumptions of the appliedmodel cannot be tested. Thus, it is not surprising that differentmodels may result in considerably disparate risk estimates whenapplied to the same data.

7 In a review of the current methods formodeling Epidemiologic studies for QRA, Stayner et a14 suggestedthat a reasonable approach in choosing the appropriate model is toconsider alternative (empiric and biologic) models and to presentrisk estimates from the models that fit the data the actual health effects are subtle, extremely largepopulations need to be studied to establish an exposure,diseaserelation, and the cost of cohort studies can be prohibitive. The useof the ecologic approach585 Hazard identificationExposure assessment Dose-response assessmentRisk characterizationDescriptive epidemiologyAnalytic epidemiologyInterventionepidemiology586 Nurminen et alMolecular epidemiologyFIGURE 1.

8 The phases of the quantitative risk assessmentprocess (left) and the associated Epidemiologic strategies(right).permits the study of very large populations at a decidedlyreduced cost, and the advances of Epidemiologic studydesigns have brought about ways of improving the validityof such It may be feasible to supplementaggregate data with sampled data at the individual level ina multilevel design, and the method proposed by Prenticeand Sheppard6 can be a cost-effective alternative to thestudy of entire Risk- assessment ProcessFocusing on epidemiology, the different phases of theQRA process vis- 1-vis the types of epidemiologicstrategy may be arranged as in Figure 1.

9 Risk assessmentconventionally involves the following four steps (cf Ref 7):1. Hazard identification aims at answering the ostensiblysimple question, "Does the available evidence point to thepotential for a risk agent to cause harmful health effects inexposed populations?"2. Exposure assessment proposes to describe the expo-sure patterns and processes and to estimate the intensityand duration of exposure, as well as the characteristics andnumber of persons actually or potentially Dose-response assessment is modeling for a relationbetween exposure to an identified hazard at different doselevels and the disease risk it Risk characterization seeks to provide answers to twoquestions: "What are the health consequences of exposureto hazards at current dose levels?

10 " and "What would be thehealth benefits of risk reduction by lowering the doselevels?" Epidemiologic approaches are used in the variousphases of QRA. In cancer risk assessment , for example,techniques of "molecular" epidemiology's such as bio,logic marker? of exposure, effect, and susceptibility, andbiologically based pharmacokinetic models,10 have beenused to estimate doses in target organs and tissues, as wellas to elucidate mechanisms of dose-response rela-Epidemiology September 1999, Vol. 10 observed in classical Epidemiologic studies. Thesestudies bridge the gap between laboratory experimentationand population-based has remarked that "descriptive relations bearon such passive matters as prognosis setting and riskassessment, whereas knowledge of causal relations is thebasis for interventions, that is, for willful alterations of theoutcome through perturbations of the determinant.