Adaptive Design with Bayesian Informed Interim Decisions: Application To a Randomized Trial of Mechanical Circulatory Support.

Background: Cardiovascular and oncology trials increasingly require large sample sizes and long follow-up periods. Several approaches have been developed to optimize sample size including sample size re-estimation based on the promising zone approach. With time-to-event endpoints, methods traditionally used to test for treatment effects are based on proportional hazards assumptions, which may not always hold. We propose an adaptive design wherein using interim data, Bayesian computation of Predictive Power (PP) guides the increase in sample size and/or the minimum follow-up duration.

Methods: PROTECT IV is designed to evaluate mechanical circulatory support device vs. standard of care during high-risk percutaneous coronary intervention with the initial enrolment of 1252 patients and initial minimum follow-up of 12 months. The primary endpoint is the composite rate of all-cause death, stroke, durable left ventricular assist device implant or heart transplant, myocardial infarction or hospitalization for cardiovascular causes. The study will employ an adaptive increase in sample size and/or minimum follow-up at the Interim analysis. The adaptations utilize simulations to choose a new sample size up to 2500 and new minimal follow-up time up to 36 months that provides PP of at least 90%.

Results: Via extensive simulations, we have examined the utility of the proposed design for situations like delayed treatment effect, early benefit only and in general crossing of survival curves. Separate Piece-wise Constant Hazard Models with non-influential (weakly-informative) Gamma-priors are fitted to the interim data for the two treatment arms, free from the proportional hazards assumptions, thus yielding more robust interim decision making. The Bayesian modeling facilitates sampling of future observations from the posterior predictive distributions with the predictive probability of trial success is computed via Monte-Carlo simulations. Simulation results show that the fitting Bayesian Piecewise Exponential models to the interim data along with the use of the posterior predictive distributions lead to more "specific" adaptation rules compared to the frequentist Conditional Power while the overall operating characteristics, type-I error and power, are similar.

Conclusion: For clinical trials with time-to-event endpoints and where crossing of survival curves might be anticipated at the planning stage, flexible modeling along with wholesome use of patient-level data such as the calculation of predictive power as proposed here, may be more robust and efficient in making interim decisions such as sample size increase than the traditional use of the conditional power based on summary statistics and proportional hazards assumption.