Optimal cost-effective designs of randomized phase II proof of concept trials and associated go-no go decisions

Abstract

14669 Background: In oncology, the number of possible combinations of therapeutic agents far exceeds available patients for clinical proof of concept (POC) studies. We have investigated optimal cost-effective designs for randomized (POC) trials. Unlike a Phase III confirmatory registration trial, a randomized POC trial is exploratory in nature, allowing flexibility in type I/II error rates. Methods: We have derived an efficiency score function representing the estimated probability of identifying a truly active treatment per patient exposure in the POC trial and in subsequent phase III trials. Optimization of the score function leads to type I/II error rates (and therefore sample size) for POC trials that are most cost-effective. This in turn leads to optimal cost-effective decision criteria for continuing to Phase 3 development and optimal allocation of fixed resources. The analysis has been done first assuming multiple investigational hypotheses of equal estimated likelihood of ultimate Phase 3 success and equal value, and then in the case where these parameters are unequal. Results: The optimal POC programs for equal hypotheses involve smaller randomized POC trials. Viewed as hypothesis testing, the type I error rates for optimal trials would be in the 0.05–0.1 range, but type II error rates would be 0.4 or higher. This translates into relatively high hurdles for continuation to Phase 3 for any individual trial. For unequal hypotheses, we can show when it is optimal to investigate only the more likely or valuable hypothesis. Conclusions: Generally, it is more efficient to investigate more of the credible hypotheses, even if at lesser statistical accuracy. No significant financial relationships to disclose.