Adaptive treatment allocation and selection in multi-arm clinical trials: a Bayesian perspective

Area of Science:

• Clinical Trials Methodology
• Biostatistics
• Pharmaceutical Research

Background:

• Adaptive clinical trial designs offer flexibility, including dynamic patient allocation and early stopping for success or futility.
• Frequentist adaptive designs, like group sequential methods, can inflate Type 1 error rates or reduce statistical power during interim analyses.

Purpose of the Study:

• To demonstrate the utility of the Bayesian approach for designing and conducting adaptive randomized clinical trials in phase II and III.
• To present a Bayesian framework for sequential comparison of treatment arms based on joint posterior probabilities.

Main Methods:

• Sequential evaluation of treatment arm performance using joint posterior probabilities.
• Implementation of control actions (treatment allocation or selection) based on pre-specified critical thresholds.
• Development focused on binary outcomes, with extensions for time-to-event data, including vaccine trials.

Main Results:

• Extensive simulation experiments validated the proposed Bayesian methodology.
• Numerical results and graphical illustrations are provided in supplementary materials.
• A freely available R package, 'barts', implements the discussed methods.

Conclusions:

• The proposed Bayesian methods offer a compelling alternative to traditional frequentist approaches for adaptive clinical trial design.
• The Bayesian framework provides a robust and flexible strategy for optimizing clinical trial conduct and decision-making.