Design of Censored Survival Studies

Censored survival studies create unusual problems in experimental design. Likelihood methods of analysis depend primarily on the number of uncensored observations. Therefore, the estimation of required sample size calls for additional assumptions about the censoring mechanism. In this paper the exact distribution of the time on study with Poisson arrivals and exponential survival is derived and used for sample-size determination. The design criterion equates the expected number of accruals with the required number of observations. Within this framework, it is demonstrated that power is lost unless accrual stops before the end of the study.