Bayesian Adaptive Randomization Design

The Bayesian adaptive randomization design is an extension of adaptive research designs, which modify the randomization ratio during a study in order to maximize individual participant outcomes. Bayesian adaptive randomization designs use Bayesian data analysis to estimate the probability that each intervention maximizes the outcome of interest, and the randomization ratio is adjusted according to these estimates. In adjusting the randomization ratio based on which intervention is estimated to produce the most favorable outcome, Bayesian adaptive randomization designs maximize the likelihood of participants experiencing positive outcomes and minimize the likelihood of participants experiencing poor outcomes.

Bayesian adaptive randomization designs follow a four-step procedure. In Step 1, participants are randomly allocated to one of the interventions with the probability of allocation into each intervention being specified by the randomization ratio. In Step 2, participants’ outcome data are collected. In Step 3, the probability that each intervention produces the most favorable outcome is estimated with Bayesian data analysis. In Step 4, the randomization ratio is updated. Steps 1–4 are then repeated until either the desired number of participants has been recruited or one of the interventions has shown convincing evidence of producing the most favorable outcome.

This entry goes on to provide an examination of adaptive randomization ratios, two primary assumptions made by Bayesian adaptive randomization design, and logical considerations. Various types of data used in Bayesian data analysis are introduced, and the steps of early and late randomizations are then explained. The entry concludes by listing some advantages and limitations of Bayesian adaptive randomization design, followed by an example of a hypothetical study implementing a two-group Bayesian adaptive randomization design.