We consider phase II trials randomizing patients between a prospective control and an experimental therapy. Proposed are two-stage designs allowing for early termination of the study when the experimental arm does not show promising efficacy at the interim analysis. By using exact binomial distributions, the design characteristics, such as type I error and power, are exactly calculated. Given response probabilities for two arms, we define minimax and optimal designs that satisfy a prespecified restriction on type I and II error probabilities. These designs are randomized phase II trial analogs of Simon's designs that were proposed for single-arm phase II trials. The methods for two-arm trials are easily extended to multi-arm trials with one control and K (> or =2) experimental arms. Some phase II trials are taken as real examples.
2007 John Wiley & Sons, Ltd