Bayesian interim analysis and efficiency of phase III randomized trials.

Abstract

Background: Improving efficiency of phase III trials is paramount for reducing costs, hastening approvals, and mitigating exposure to disadvantageous randomizations. Compared to standard frequentist interim analysis, Bayesian early stopping rules may improve efficiency by the flexibility of differential priors for efficacy and futility coupled with evaluation of clinically meaningful effect sizes.

Methods: Individual patient-level data from 184,752 participants across 230 randomized two-arm parallel oncology phase III trials were manually reconstructed from primary endpoint Kaplan-Meier curves. Accrual dynamics, but not patient outcomes, were randomly varied. Bayesian Cohen's κ assessed agreement between the original analysis and the Bayesian interim analysis.

Results: Trial-level early closure was recommended based on the Bayesian interim analysis for 82 trials (36%), including 62 trials which had performed frequentist interim analysis and 33 which were already closed early by the frequentist interim analysis. Bayesian early stopping rules were 96% sensitive for detecting trials with a primary endpoint difference, and there was a high level of agreement in overall trial interpretation (κ, 0.95). Moreover, Bayesian interim analysis was associated with reduced enrollment.

Conclusions: Bayesian interim analyses seem to improve trial efficiency by reducing enrollment requirements without compromising interpretation.