A Practical Response Adaptive Block Randomization Design with Analytic Type I Error Protection

Response adaptive randomization is appealing in confirmatory adaptive clinical trials from statistical, ethical, and pragmatic perspectives, in the sense that subjects are more likely to be randomized to better performing treatment groups based on accumulating data. The Doubly Adaptive Biased Coin Design (DBCD) is a popular solution due to its asymptotic normal property of final allocations, which further justifies its asymptotic type I error rate control. As an alternative, we propose a Response Adaptive Block Randomization (RABR) design with pre-specified randomization ratios for the control and high-performing groups to robustly achieve desired final sample size per group under different underlying responses, which is usually required in industry-sponsored clinical studies. We show that the usual test statistic has a controlled type I error rate. Our simulations further highlight the advantages of the proposed design over the DBCD in terms of consistently achieving final sample allocations and of power performance. We further apply this design to a Phase III study evaluating the efficacy of two dosing regimens of adjunctive everolimus in treating tuberous sclerosis complex but with no previous dose-finding studies in this indication.