Stratified proportional win-fractions regression analysis.

Area of Science:

• Biostatistics
• Clinical Trials
• Survival Analysis

Background:

• The proportional win-fractions (PW) model analyzes prioritized composite endpoints in regression, assuming covariate-specific win ratios are time-invariant.
• This proportionality assumption can be restrictive and may not hold for all covariates.

Purpose of the Study:

• To develop a stratified PW model that relaxes the proportionality assumption for certain prognostic factors.
• To provide a more robust and efficient analysis for prioritized composite endpoints in regression settings.

Main Methods:

• A stratified PW model is formulated using pairwise comparisons within strata.
• A common win ratio across strata is modeled multiplicatively with covariates.
• An estimating function based on an incomplete -statistic is constructed for regression coefficients.
• Two types of asymptotic variance estimators are developed, accommodating varying numbers of strata.

Main Results:

• Simulation studies demonstrate the stratified PW model's superior robustness and efficiency compared to the unstratified version.
• The model allows valid statistical inference even with very small strata, including matched pairs.
• Real data from a cardiovascular trial illustrate the practical benefits of stratification.

Conclusions:

• The stratified PW model provides a valuable extension to win ratio analysis, enhancing robustness and efficiency.
• It effectively handles situations where the proportionality assumption is violated.
• The methods are implemented in the R package WR, facilitating broader application.