BayCAR: A Bayesian based Covariate-Adaptive Randomization method for multi-arm trials

Area of Science:

• Clinical Trials Methodology
• Biostatistics
• Bayesian Inference

Background:

• Randomization is crucial for controlled clinical trials to prevent confounding.
• Existing methods like restricted randomization and minimization have limitations, especially with many covariates.
• Minimization methods require further theoretical justification for their adaptive randomization probability.

Purpose of the Study:

• To propose a novel Bayesian covariate-adaptive randomization method.
• To provide meaningful interpretations for adaptive randomization probabilities.
• To achieve balanced distributions of numerous categorical and continuous covariates across treatment arms.

Main Methods:

• Development of a Bayesian framework for covariate-adaptive randomization.
• Incorporation of adaptive randomization probabilities with clear interpretations.
• Application to scenarios requiring the balance of a large number of covariates.

Main Results:

• The proposed method demonstrates desirable marginal and overall covariate balance.
• Effective balancing is achieved for both categorical and continuous covariates.
• The method is particularly advantageous when dealing with a large number of covariates.

Conclusions:

• The Bayesian covariate-adaptive randomization method offers a robust solution for complex clinical trial designs.
• It provides interpretable adaptive probabilities and superior covariate balancing.
• This method enhances the reliability and validity of controlled clinical trials with many covariates.