Sample size for cohort studies in pharmacoepidemiology

Object—Cohort studies in pharmacoepidemiology can result in a unique type of study, where subjects have complex types of exposure to drugs (with periods of non-exposure as well). The object of this paper is to explain how to calculate the sample size of such a study. Method—It is assumed that adverse events follow Poisson distributions in the two study groups. The null hypothesis is that the two groups have equal rates of disease. Formulae are provided to calculate the sample size required to significantly reject the null hypothesis. Sample size is given as the number of events, rather than the number of subjects entered. In a Poisson study, it is the ratio of the amount of person-years exposure in the two groups that is important to calculate sample size, rather than the actual amounts of exposure (or number of subjects in the study). Some examples are included. © 1997 John Wiley & Sons, Ltd.