On sample size determination for restricted mean survival time-based tests in randomized clinical trials

TL;DR

This paper develops a new sample size calculation method for augmented RMST-based tests in clinical trials, allowing for more flexible and efficient study designs without full survival curve specifications.

Contribution

It introduces an approximated sample size formula and a recalculation approach for augmented RMST tests that handle baseline covariates and do not require full survival curve assumptions.

Findings

Provides a practical sample size formula for augmented RMST tests.

Enables power maintenance with blinded data recalculations.

Improves flexibility in trial design without full survival curve info.

Abstract

Restricted mean survival time (RMST) is gaining attention as a measure to quantify the treatment effect on survival outcomes in randomized clinical trials. Several methods to determine sample size based on the RMST-based tests have been proposed. However, to the best of our knowledge, there is no discussion about the power and sample size regarding the augmented version of RMST-based tests, which utilize baseline covariates for a gain in estimation efficiency and in power for testing the no treatment effect. The conventional event-driven study design based on the log-rank test allows us to calculate the power for a given hazard ratio without specifying the survival functions. In contrast, the existing sample size determination methods for the RMST-based tests relies on the adequacy of the assumptions of the entire survival curves of two groups. Furthermore, to handle the augmented test, the correlation between the baseline covariates and the martingale residuals must be handled. To address these issues, we propose an approximated sample size formula for the augmented version of the RMST-based test, which does not require specifying the entire survival curve in the treatment group, and also a sample size recalculation approach to update the correlations between the baseline covariates and the martingale residuals with the blinded data. The proposed procedure will enable the studies to have the target power for a given RMST difference even when correct survival functions cannot be specified at the design stage.