Bayesian Response-Adaptive Randomization for Cluster Randomized Controlled Trials.

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Public Health Research

Background:

• Cluster randomized controlled trials (CRCTs) are vital when individual randomization is impractical or interventions are group-based.
• Standard CRCT randomization can be inefficient, leading to resource strain and ethical concerns if subjects are assigned to suboptimal arms.
• Adaptive randomization offers a potential solution to optimize treatment allocation in CRCTs.

Purpose of the Study:

• To propose a novel Bayesian response-adaptive randomization design for CRCTs using Thompson sampling.
• To incorporate early stopping rules for efficacy and futility based on posterior probabilities.
• To evaluate the performance and compare the proposed adaptive design against standard CRCT designs.

Main Methods:

• A Bayesian response-adaptive randomization design employing Thompson sampling and Markov chain Monte Carlo (MCMC) sampling.
• Dynamic allocation of clusters to treatment arms based on interim posterior distributions of treatment effects.
• Implementation of early stopping rules for efficacy and futility using prespecified posterior probability thresholds.

Main Results:

• The proposed adaptive design demonstrated improved efficiency and ethical considerations compared to standard CRCT designs in simulation studies.
• The method preferentially assigned more clusters to the more efficacious intervention.
• Robust statistical power and controlled false positive rates were maintained across various settings, including different intra-cluster correlation coefficients, cluster sizes, and effect sizes.

Conclusions:

• Bayesian response-adaptive randomization using Thompson sampling offers an efficient and ethical approach for CRCTs.
• This adaptive design optimizes resource allocation by dynamically assigning clusters to superior treatments.
• The methodology shows promise for improving the conduct of cluster randomized trials, as evidenced by simulations based on an HIV behavioral trial.