A Bayesian response-adaptive trial in tuberculosis: The endTB trial

Area of Science:

• Biostatistics
• Infectious Diseases
• Clinical Trial Design

Background:

• Multidrug-resistant tuberculosis (MDR-TB) presents a significant global health challenge.
• Evaluating novel treatments for MDR-TB requires efficient and effective clinical trial designs.
• Traditional randomization methods may not optimally allocate patients to promising experimental regimens.

Purpose of the Study:

• To assess the utility of Bayesian response-adaptive randomization in clinical trials for new MDR-TB treatments.
• To compare the efficiency and allocation properties of adaptive versus balanced randomization.

Main Methods:

• A response-adaptive randomization procedure was developed, incorporating preliminary outcomes (culture conversion at 8 weeks, treatment success at 39 weeks).
• The primary outcome was treatment success at 73 weeks.
• The adaptive design was compared to balanced randomization using hypothetical clinical trial scenarios.

Main Results:

• Bayesian adaptive randomization demonstrated increased statistical power and required fewer patients compared to non-adaptive designs in simulated scenarios.
• Adaptive designs consistently allocated a higher proportion of participants to more effective treatment arms.
• The observed advantages were contingent upon the hypothetical scenarios aligning with the adaptive algorithm's underlying assumptions.

Conclusions:

• Bayesian response-adaptive designs offer an attractive approach for accelerating the evaluation of multiple therapeutic regimens in MDR-TB trials.
• This methodology shows potential for improving patient allocation towards demonstrably effective treatments, thereby increasing trial efficiency.