All Times EDT
Keywords: play-the-winner rule, Neymann allocation, adaptive designs
Response-adaptive randomization (RAR) designs were introduced due to the captivating characteristics of tilting the randomization probability to assign more subjects to the better-performing treatment group based on the interim data. However, the substantial disagreement regarding bias due to time-trends in adaptive randomization is not fully recognized. The type-I error is inflated in the traditional frequentist and Bayesian RAR approaches when a time-trend is present. Blocked RAR approaches where the subjects are assigned in blocks and the randomization ratio is altered by blocks have been proposed to avoid bias and control type I error due to time trends. We provide a theoretical framework for optimizing blocked RAR for binary responses using the Cochran-Mantel-Haenszel test, especially in a non-inferiority trial setting. We discuss how the framework is useful in the design of non-inferiority trials using blocked RAR for both additive and multiplicative margins. We also provide a generalized relationship for multiple treatments with a control. We evaluated our blocked RAR designs and compared them with the traditional 1:1 design for two treatment arms. We concluded that the blocked RAR design is occasionally more efficient than the traditional 1:1 design.