Study design for exposure assessment in epidemiological studies.

Abstract

We consider the implication, for study efficiency, of choice of method of exposure assessment in epidemiological studies, and in particular the optimal allocation of resources that should be devoted to improving the accuracy of exposure assessments. Useful for this purpose is a general result that the efficiency of a study based on approximate exposures relative to one based on exact exposures is equal to the square of the correlation between the true exposure and the approximate measurement in the study base (called the validity coefficient). This implies that to maximize study power, investment in increased precision is worthwhile up to the point at which proportional increase in total costs per subject exceed the proportional gain in the square of the validity coefficient. This result does not hold if exposure measurement error depends on disease status (is differential), or if important confounders are measured with error. 'Classical' exposure measurement error (uncorrelated with true exposure) or misclassification usually biases estimates of effect. Information from validity or reliability sub-studies can be used to correct this bias, but not substantially recover lost efficiency. There are several papers on the optimal allocation of resources to a validity sub-study.