Statistical Evaluation of Exposure Assessment Strategies

Abstract

It is well known that exposure measurement error or misclassification can bias an epidemiologic risk assessment. Accordingly, it is useful to have methods of assessing the degree of error in exposure assessments. This paper proposes statistical evaluation techniques for two types of exposure assessment strategies: categorization of exposure and estimation of actual personal exposure levels (with particular attention to the latter).

Categorization of exposure is generally done in one of the two ways: ordinal but nonquantitative grouping or quantitative exposure intervals. Estimation of actual exposure levels is most often done in the form of job-by-year exposure matrices and somewhat less often by exposure prediction models. The nature of uncertainty inherent with each of these approaches is described.

Quantitative estimates of the magnitude of error in an exposure assessment are possible when exposure matrices or prediction models are employed. Three types of statistical approaches are explored for estimation of the degree of error in quantitative exposure assessments, with particular attention to methods associated with exposure prediction models. The methods described are sensitivity analysis, cross-validation techniques, and field testing. Advantages and disadvantages of each method are discussed, as well as the appropriate situation for their application. Cross-validation and sensitivity analysis are shown to be particularly well-suited to the validation of exposure prediction models. Field testing is a more general technique that can be used to assess the validity of exposure matrices. Hornung, R.W.: Statistical Evaluation of Exposure Assessment Strategies. Appl. Occup. Environ. Hyg. 6:516–520; 1991.