Randomization-based covariance analysis for hypothesis testing of treatment comparisons based on restricted mean...

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Survival Analysis

Background:

• Restricted Mean Survival Time (RMST) offers a clinically meaningful measure of treatment effect, independent of proportional hazards assumptions.
• Existing methods for RMST analysis may lack robust hypothesis testing frameworks, especially when accounting for covariates.

Purpose of the Study:

• To introduce a novel randomization-based analysis of covariance (RB-ANCOVA) for hypothesis testing of RMST differences.
• To provide a method that adjusts for baseline covariates and controls Type I error under the strong null hypothesis.

Main Methods:

• The proposed RB-ANCOVA method partitions the follow-up period into intervals to approximate RMST.
• It utilizes the known asymptotic covariance structure of RMST and covariate means under the strong null hypothesis to construct a chi-squared test statistic.
• The approach allows for exact p-values through re-randomization and accommodates stratified trial designs.

Main Results:

• The RB-ANCOVA method enables hypothesis testing of RMST differences, adjusting for baseline covariate imbalances.
• It demonstrates reduced variance and clear control of Type I error compared to methods without covariate adjustment.
• The method was illustrated using data from an amyotrophic lateral sclerosis (ALS) clinical trial.

Conclusions:

• RB-ANCOVA provides a robust framework for hypothesis testing of RMST differences in time-to-event data.
• This method enhances statistical power and reliability in clinical trials by incorporating covariate adjustment.
• It offers a valuable tool for analyzing treatment effects without imposing restrictive statistical assumptions.