Competing risks regression for clustered data with covariate-dependent censoring.

Area of Science:

• Biostatistics
• Clinical Trials
• Epidemiology

Background:

• Competing risks data in clinical studies often exhibit cluster effects (e.g., center effects, matched pairs).
• The proportional subdistribution hazards (PSH) model is standard for competing risks, but existing methods for clustered data lack covariate-dependent censoring and stratification.
• Real-world data frequently involve covariate-dependent censoring and non-proportional hazards structures.

Purpose of the Study:

• To propose a novel marginal stratified PSH model for clustered competing risks data.
• To address limitations of existing methods by incorporating covariate-dependent censoring and stratification.
• To evaluate the model's performance and apply it to leukemia patient data.

Main Methods:

• Developed a marginal stratified PSH model with covariate-adjusted censoring weight for clustered data.
• Utilized a marginal stratified proportional hazards model to estimate censoring probabilities, accounting for clusters and non-proportional hazards.
• Conducted simulation studies to assess parameter estimation and coverage rates.

Main Results:

• The proposed method yields unbiased parameter estimates in the presence of covariate-dependent censoring.
• Simulation results demonstrate approximate 95% coverage rates, indicating good performance.
• The method was successfully applied to stem cell transplant data to analyze HLA matching effects on graft-versus-host disease.

Conclusions:

• The proposed marginal stratified PSH model effectively handles clustered competing risks data with covariate-dependent censoring and non-proportional hazards.
• This approach offers a more robust and accurate analysis for complex clinical datasets.
• The findings have implications for understanding donor-recipient matching in leukemia transplantation.