ABSTRACT Sequential multiple assignment randomized trials (SMART) designs, which are used to evaluate adaptive treatment strategies (ATSs), involve multiple stages of patient randomization based on intermediate outcomes. While offering greater flexibility and personalization in treatment, these designs are often analytically complex and resource‐intensive—especially when survival outcomes are involved, given the extended follow‐up times and potential violations of proportional hazards assumptions. To address these challenges, we propose a statistical inference framework based on restricted mean survival time (RMST). RMST does not rely on the proportional‐hazards assumption and serves as a robust summary for survival data. The framework includes fixed‐ and dynamic‐weight RMST estimators, their variance–covariance structures, confidence intervals, and hypothesis tests for pairwise and global comparisons among ATSs. We also integrate interim analyses into SMART designs using RMST and develop a type I error‐control method that accommodates the lack of the independent‐increments property. Extensive simulations demonstrate good estimator performance and show that interim designs can reduce either sample size or trial duration. In summary, this study offers an efficient and practical framework for SMART trials with survival outcomes.
Pan et al. (Fri,) studied this question.