Futility analysis is a critical component of phase III oncology trials, enabling early termination of ineffective treatments to save resources and time. However, these interim analyses are often challenged by insufficient data for long-term primary endpoints, such as overall survival. A comprehensive simulation study was conducted to model various phase III trial scenarios. This simulation study investigates the benefits of utilizing mixture models to incorporate short-term endpoint information for enhancing futility stopping decisions across various scenarios. When the short-term endpoint was strongly predictive of the long-term endpoint, its inclusion via mixture models increases trial power and the probability of correctly stopping for futility, leading to reduced expected trial duration and sample size. In scenarios with delayed treatment effects, simple mixture models perform poorly; however, by employing mixture segmented hazards models to account for the delays, the benefits of integrating short-term endpoints remain evident. Importantly, when the short-term endpoint was non-predictive or contradictory, its inclusion did not inflate the Type I error rate or otherwise negatively impact the trial’s integrity. Incorporating short-term endpoint information into futility analysis through appropriately specified mixture models is a robust and efficient strategy. Our findings provide evidence-based guidance for the design of adaptive clinical trials.
Liu et al. (Sun,) studied this question.