ABSTRACT
Sample size calculations for survival trials are based on the test statistics, assumption of the underlying survival model, accrual distribution, design parameters, and treatment effect size. The aim of this chapter is to make the reader aware of the non-robustness of the test statistics, parametric survival distribution and proportional hazards model assumptions, and the sensitivity to the design parameters and accrual distribution of the sample size calculation. The most popular assumption for survival trial design is the exponential distribution or proportional hazards model with uniform accrual and no loss to follow-up and cross-over. Usually, these simplified assumptions reflect the reality closely enough that the methods produce reasonably accurate sample sizes. Many times, however, the complexity of cancer survival trials leads to violations of the assumptions that result in unacceptably inaccurate sample size calculations and severe power loss for the trials (Wittes, Lakatos 2002). In the literature, some simple adjustments for loss to follow-up and cross-over treatments have been proposed. However, such simple adjustments are usually conservative and inaccurate. Formal sample size calculations for complex trial are available, e.g., the Lakatos Markov chain approach, and should be used for such survival trial design.
