Toward Making Epidemiologic Data More Useful for Quantitative Risk Assessment

Abstract

This paper discusses issues faced in using epidemiologic data to develop quantitative estimates of risk from specified patterns of exposure to a toxicant. We focus on use of data from cohort studies with binary endpoints (occurrence or non-occurrence of disease). Relative advantages of Cox regression and Poisson regression are presented. A general form of exposure metric is presented, and criteria for selecting an appropriate metric are discussed. Advantages and disadvantages of various dose-response models are discussed. It is argued that, unless low-dose linearity of the dose response can be ruled out on non-statistical grounds, then bounds for low-dose risk should incorporate low-dose linearity; a sequential procedure for computing such bounds is illustrated. Limitations in exposure data and their impact on risk assessments are discussed. Issues arising when using meta-analytic techniques to combine data from multiple epidemiologic studies are discussed. Limitations in risk assessments resulting from reliance upon published results alone are described. Methods for converting from measures of risk used in epidemiologic studies (e.g., relative risk) to measures appropriate for a risk assessment (e.g., additional lifetime probability of disease occurrence resulting from a specific exposure pattern) are described in detail. Several examples from the asbestos epidemiologic literature are presented to illustrate these issues.