Bayesian adaptive randomization in a clinical trial to identify new regimens for MDR-TB: the endTB trial.

Abstract

Evidence-based optimization of treatment for multidrug-resistant tuberculosis (MDR-TB), including integration of new drugs, is urgent. Such optimization would benefit from efficient trial designs requiring fewer patients. Implementation of such innovative designs could accelerate improvements in and access to MDR-TB treatment. To describe the application, advantages, and challenges of Bayesian adaptive randomization in a Phase III non-inferiority trial of MDR-TB treatment. endTB is the first Phase III non-inferiority trial of MDR-TB treatment to use Bayesian adaptive randomization. We present a simulation study with assumptions for treatment response at 8, 39, and 73 weeks after randomization, on which sample size calculations are based. We show differences between Bayesian adaptive randomization and balanced randomization designs in sample size and number of patients exposed to ineffective regimens. With 750 participants, 27% fewer than required by balanced randomization, the study had 80% power to detect up to two (of five) novel treatment regimens that are non-inferior (margin 12%) to the control (70% estimated efficacy) at 73 weeks post randomization. Comparing Bayesian adaptive randomization to balanced randomization, up to 25% more participants would receive non-inferior regimens. Bayesian adaptive randomization may expose fewer participants to ineffective treatments and enhance the efficiency of MDR-TB treatment trials.