Response-adaptive randomization with imperfect intermediate endpoints

Area of Science:

• Clinical Trials Methodology
• Biostatistics
• Infectious Disease Research

Background:

• Response-adaptive randomization is debated, especially with long-term primary outcomes.
• Intermediate endpoints are used to update randomization in trials requiring extended follow-up.
• Tuberculosis trials serve as the context for evaluating adaptive designs with imperfect data.

Purpose of the Study:

• To assess the impact of response-adaptive randomization utilizing an imperfect intermediate endpoint.
• To evaluate the effectiveness of adaptive designs in allocating participants to superior treatment arms.
• To examine the influence of intermediate endpoint accuracy and time-trends on trial outcomes.

Main Methods:

• Simulated a response-adaptive randomization design for a three-arm superiority trial.
• Used culture conversion at 8 weeks as an intermediate endpoint for a 73-week primary outcome (treatment success).
• Varied sensitivity, specificity, and true treatment efficacy to analyze adaptive randomization performance and type I error.

Main Results:

• Even with perfect intermediate endpoint accuracy, adaptive randomization did not consistently favor the better arm.
• Lower accuracy of the intermediate endpoint significantly reduced the goal of allocating more patients to the superior arm.
• Time-trends increased type I error; stratification corrected this but reduced statistical power.

Conclusions:

• Response-adaptive randomization is appealing for evaluating multiple regimens efficiently.
• However, it necessitates highly accurate intermediate endpoints, which do not guarantee reliable patient allocation.
• The trustworthiness of response-adaptive randomization is questionable with imperfect intermediate endpoints.