Sample size calculations for single group post-marketing cohort studies

In pharmacoepidemiology, single group cohort is the most frequently proposed design to determine if the incidence rate of an adverse drug reaction among the exposed differs from a reference value. In many situations, the number of events expected in the cohort is too small to conduct sample size calculations based on the normal distribution. This paper proposes, for a single group cohort study, calculations and tables derived from the Poisson distribution. The results are based on a one-sided test with a 0.05 significance level and a power of 0.9 and 0.8. Two parameters have to be specified a priori: the expected incidence of the event under the null hypothesis and the minimum risk ratio to be detected. The required sample size and the critical number of events to reject the null hypothesis are directly derived from the tables. Results show that the normal approximation may lead to an underestimation of the required sample size.