Evaluating Bayesian adaptive randomization procedures with adaptive clip methods for multi-arm trials

Abstract

Bayesian adaptive randomization is a heuristic approach that aims to randomize more patients to the putatively superior arms based on the trend of the accrued data in a trial. Many statistical aspects of this approach have been explored and compared with other approaches; yet only a limited number of works has focused on improving its performance and providing guidance on its application to real trials. An undesirable property of this approach is that the procedure would randomize patients to an inferior arm in some circumstances, which has raised concerns in its application. Here, we propose an adaptive clip method to rectify the problem by incorporating a data-driven function to be used in conjunction with Bayesian adaptive randomization procedure. This function aims to minimize the chance of assigning patients to inferior arms during the early time of the trial. Moreover, we propose a utility approach to facilitate the selection of a randomization procedure. A cost that reflects the penalty of assigning patients to the inferior arm(s) in the trial is incorporated into our utility function along with all patients benefited from the trial, both within and beyond the trial. We illustrate the selection strategy for a wide range of scenarios.

Identification of an optimal dose of intravenous ketamine for late-life treatment-resistant depression: a Bayesian adaptive randomization trial

“…This allows statements about the probability that treatment shows benefit of some magnitude. This data is used as the basis by the Bayesian Adaptive algorithm to change allocation of subjects or to close conditions with low probability of efficacy of a certain magnitude per pre-defined stopping rules [ 23 ]. Bayesian adaptive randomization begins with specification of a prior distribution that formalizes the available information regarding the anticipated effect and its associated uncertainty.…”