Restricted mean survival time to estimate an intervention effect in a cluster randomized trial

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Epidemiology

Background:

• Restricted mean survival time (RMST) is a valuable measure for intervention effects in time-to-event analyses.
• Existing RMST estimation methods are primarily designed for independent data.
• Cluster randomized trials (CRTs) present unique analytical challenges due to data dependency within clusters.

Purpose of the Study:

• To extend existing RMST estimation approaches for independent data to clustered data within CRTs.
• To evaluate and compare the statistical performance of these extended RMST methods via simulation.
• To provide practical guidance for analyzing time-to-event outcomes in CRTs using RMST.

Main Methods:

• Extended Kaplan-Meier curve integration and pseudo-values regression for RMST estimation in clustered data.
• Simulation studies to assess performance under various CRT scenarios (cluster number/size, clustering degree, effect size, hazard assumptions).
• Implementation of a permutation test for pseudo-values regression in CRTs with few clusters (<50).

Main Results:

• Extended RMST methods accurately estimated variance and controlled Type I error with sufficient clusters (≥50) under proportional and non-proportional hazards.
• Permutation tests corrected Type I error for CRTs with limited clusters, yielding adequate confidence interval coverage.
• Pseudo-values regression offered covariate adjustment, a benefit over other methods.

Conclusions:

• The extended RMST estimation methods are statistically sound for CRTs, providing a robust alternative to hazard ratios.
• The choice of method may depend on the number of clusters, with permutation tests recommended for smaller trials.
• These methods facilitate a more comprehensive analysis of intervention effects in clustered trial settings, as demonstrated in an asthma education program trial.