In this week's blog, Dr. Kert Viele discusses his thoughts on the RAR literature and the current state of RAR in practical clinical trials. He focuses on general principles that have emerged in the literature, the nuts and bolts of constructing and tuning RAR designs, and operational issues.

Longitudinal modeling and predictive modeling are the foundation of learning efficiently in adaptive clinical trials, allowing real-time learning from early data. Bayesian models enable faster, more informed decisions even when primary endpoints are distant.

REMAP-CAP, ATTACC, and ACTIV-4a unified under a multi-platform randomized clinical trial with a joint Bayesian adaptive design, rapidly determining that therapeutic anticoagulation benefits moderate, but not severe, COVID-19 patients.

ICH-E20’s regulatory caution towards adaptive designs is often misapplied, resulting in inefficient or unrealistic alternatives for sponsors and patients. Operational casework demonstrates that so-called “complexity” in adaptive design is frequently misunderstood and that regulatory “false choices” undermine trial effectiveness.

Bayesian borrowing in Phase 3 trials formally combines prior evidence with the new trial data to enhance development efficiency and regulatory decision making. This approach requires rigorous statistical modeling, careful selection of historical information, and detailed regulatory dialogue.

A rigorous review of ordinal endpoint analyses, showing every approach—utility weighting, proportional odds, dichotomization, or non-parametric—inevitably assigns relative weights to outcome states. Berry Consultants’ mathematical demonstration reveals how proportional odds analysis embeds prevalence-based weights, underscoring the need for transparency and clinical input in trial design.