ABSTRACT
This chapter focuses on procedures for sample size calculation based on exact tests for small samples. It discusses the procedures for obtaining sample sizes based on exact tests for comparing proportions such as the binomial test, negative binomial regression, and Fisher's exact test. The chapter provides procedures for sample size calculation under various optimal multiple-stage designs such as an optimal two-stage design, an optimal three-stage design, and a flexible optimal design for single-arm phase II cancer trials. It presents some practical issues and also provides procedures for sample size calculation under a flexible design for multiple armed clinical trials. The sample size can be obtained numerically by greedy search over the sample space. Chen and Ng indicated that the optimal two-stage design is similar to the Pocock sequence design for randomized controlled clinical trials where the probability of early termination is high, the total possible sample size is larger, and the expected size under the alternative hypothesis is smaller.
