83 A Bayesian framework for efficacy-and-effectiveness too randomized clinical trials with dynamic borrowing

Objectives/Goals: Efficacy-and-Effectiveness Too (EE2) randomized clinical trials pair a controlled efficacy cohort with a pragmatic effectiveness cohort. We will deliver a Bayesian EE2 framework with interim go/no-go rules and adaptive cross-cohort borrowing for time-to-event endpoints. Methods/Study Population: We formalize three approaches: 1) simultaneous: both cohorts start together, and interim looks apply predictive success probability (PSP) to enable early efficacy or futility stopping within each cohort; 2) staggered: the trial begins with the efficacy cohort and, at 50% information, initiates the effectiveness cohort only if the PSP under a retained-effect assumption (e.g., ≥70% of the efficacy log-HR retained) exceeds a prespecified threshold; 3) sequential: the efficacy trial is completed first, and the effectiveness trial is designed using the efficacy posteriors. Across approaches, the Bayesian method relies on piecewise-exponential modeling with Gamma-Poisson conjugacy, Negative-Binomial predictive counts, and power or commensurate priors with robust discounting. Results/Anticipated Results: Operating characteristics (type I error, power, bias/coverage, borrowing Effective Sample Size) will be evaluated through simulation studies across heterogeneity scenarios and prior-data conflict. We will re-analyze the RUBY and NRG-GY018 endometrial trials (endpoints: Progression-Free Survival and Overall Survival) as EE2 case studies (efficacy: mismatch repair-deficient tumors; effectiveness: mismatch repair-proficient tumors) using dynamic borrowing and the appropriate interim logic to demonstrate feasibility, robustness, and gains in precision. We anticipate improved precision for each cohort and for the combined effect and stable decisions under partial retention. Discussion/Significance of Impact: This work provides the first end-to-end Bayesian EE2 toolkit spanning simultaneous, staggered, and sequential implementations, with interim rules, principled borrowing, and software, to accelerate the generation of generalizable evidence while safeguarding against prior-data conflict, advancing efficient, patient-centered translation.