A Bayesian approach for unplanned sample sizes in phase II cancer clinical trials

Abstract

Background Phase II cancer clinical trials commonly employ two-stage designs that incorporate a single interim analysis for lack of efficacy and are designed to achieve specified frequentist properties. The requirement to examine the outcome at a prespecified sample size (SS) can be problematic, because the attained SS often differs from the planned SS. Purpose We propose to address unplanned SSs achieved at either stage by means of a Bayesian approach that approximately preserves the original design’s properties. Methods Our approach translates the rejection rule of the original frequentist design into equivalent statements about the posterior distribution of the response rate and applies this Bayesian criterion to the analysis with any realized SS. Results The results demonstrate that our approach approximately maintains operating characteristics of the original frequentist design including type I and type II error rates, probability of early termination, and expected SS under the null hypothesis. Limitations Designs attained under this approach may not satisfy target limits for type I error rate and power. Conclusions Our method offers a coherent analysis plan when the attained SS at either stage deviates from that specified in the original design. The price of its flexibility, in terms of erosion of the desired frequentist properties, is modest.