Methodological issues in cohort studies. II: Power calculations.

Abstract

A simple model is described for estimating power in cohort studies, in which the exposure is treated as a polytomous variable, with a known distribution in the population from which the sample is drawn. The model then requires the specification of the expected number of deaths which will occur in the cohort, calculated from the population rates, the dose-response relationship, and the size of the cohort. The model also allows for misclassification of exposure, the rule rather than the exception in epidemiological studies. The model is applied to a proposed study of saturated fat intake and risk of death from colorectal cancer in a male cohort drawn from the general population. It is demonstrated that this approach leads to an optimization of the power estimates, and in particular that maximization of power can be achieved by using a relatively small number of categories, eg four. It is also demonstrated that the effect of misclassification is less extreme if a polytomous dose-response model is used for analysis as compared to the usual simple dichotomous exposure model.