Power calculation for cohort studies with improved estimation of expected numbers of deaths.

This paper focuses on improving the accuracy of sample size calculations for cohort studies by careful calculation of the expected number of deaths in the population, taking into account either prior information or realistic assumptions about variables which may affect the mortality or incidence. Sometimes small changes in the assumptions can dramatically alter the expected numbers and may necessitate modifications in the design of the study. Possible modification include extension of the follow-up time, and recognition that the real strength of the study may lie in the potential for pooling several similar studies. The problem will be discussed with reference to two examples of occupational cohort studies where differing prior information was available.