Optimal two-stage design based on error rates under a Bayesian perspective

In phase II clinical trials, two-stage designs allowing early stopping for lack of efficacy are frequently used. We present a Bayesian two-stage design that ensures high posterior probabilities that the response rate of the experimental drug exceeds a desirable level, when the decision is to proceed with treatment evaluation. Moreover, the design exploits the distinction between analysis and design prior distributions, to control the predictive prob- ability of Type I and II errors, while minimizing the expected sample size under the null hypothesis.