Analysis of Strategies for Reconstructing Exposures

Abstract

Many retrospective epidemiologic studies require the reconstruction of exposures over long periods of time on the basis of incomplete or absent data. The modern requirements of acceptably accurate, quantitative exposure estimations for the construction of exposure-response relationships and extrapolation of estimated risks to untested regions suggest the existence of a conflict between the study requirements and the more or less qualitative reconstruction methodologies.

While the importance of exposure reconstruction is accepted, there is no structured analysis of the logical basis of the reconstruction process. In an attempt to analyze exposure reconstruction strategies (ERS) mathematically, a formal set of properties for ERSs are developed. These properties are presented in the form of three necessary and six desirable/undesirable general attributes of any exposure reconstruction process. Even a cursory analysis of these formal attributes suggests that a careful analysis of the exposure reconstruction method utilized in an epidemiologic study is necessary. Without such an analysis, the reconstructed exposures may lead to exposure-response relationships which are more hypothetical than real. Since no analyses of reconstruction strategies have been reported, the frequency of erroneous procedures for the exposure reconstructions reported in the literature is unknown. It is also shown that a significant amount of fundamental methodological research is needed to develop tests for the reliability, accuracy, and sensitivity of exposure reconstruction methods.