Optimal timing for an accelerated interim futility analysis incorporating real world data

Abstract

Background: Randomized controlled trials include interim monitoring guidelines to stop early for safety, efficacy, or futility. Futility monitoring facilitates re-allocation of limited resources. However, conventional methods for interim futility monitoring require a trial to accrue nearly half of the outcome data to make a reliable early stopping decision, limiting its benefit. As early stopping for futility will not inflate type-I error, these analyses are an appealing venue for incorporating external data to improve efficiency.

Methods: We propose a Bayesian approach to futility monitoring leveraging real world data using Semi-Supervised MIXture Multi-source Exchangeability Models, which accounts for both measured and unmeasured differences between data sources. We implement futility monitoring using predictive probabilities and investigate the optimal timing with respect to the expected sample size under the null hypothesis. Because we only incorporate external data during the interim futility analysis the proposed design is not limited by type-I error inflation.

Results: When the external and trial data are exchangeable, the proposed method provides a roughly 70 person reduction in expected sample size under the null. Under scenarios where exchangeability does not hold, our approach still provides a 10-20 person reduction in expected sample size under the null with about 80% power.

Conclusions: External data borrowing in interim futility monitoring is a promising venue to improve trial efficiency without type-I error inflation. Approaches that are acceptable to regulatory authorities and leverage the complementary strengths of real world and trial data are vital to more efficiently allocate limited resources amongst clinical trials.

Keywords: Bayesian adaptive design; Interim futility monitoring; Randomized controlled trial; Real world data.