Optimal two-stage designs for clinical trials with binary response

Abstract

We present a simple two-stage design for a randomized clinical trial with dichotomous outcomes. It is based on a design of Ellenberg and Eisenberger which permits early acceptance of the null hypothesis after the first stage. We have optimized the design by minimizing the average expected sample size subject to constraints on the type 1 and type 2 errors. The resulting designs have maximum sample size only slightly larger than that of single stage trial, with 40-45 per cent of this maximum allocated to the first stage, and null hypothesis probability of early acceptance in excess of 0.60.