Analysis of clustered competing risks data using subdistribution hazard models with multivariate frailties

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Survival Analysis

Background:

• Multi-center randomized clinical trials often exhibit variations in treatment effects and baseline risks across different centers.
• Competing risks data are common in such trials, necessitating specialized analytical approaches.

Purpose of the Study:

• To propose a subdistribution hazard regression model incorporating multivariate frailty to investigate treatment effect heterogeneity in multi-center clinical trials.
• To develop an efficient inference method using hierarchical likelihood (h-likelihood) to handle complex frailty models.

Main Methods:

• Development of a subdistribution hazard regression model with multivariate frailty.
• Application of a hierarchical likelihood (h-likelihood) method for tractable inference, avoiding complex integrations.
• Extension of the h-likelihood to weighted partial likelihood for subdistribution hazard frailty models.

Main Results:

• The profile likelihood from the h-likelihood is shown to be equivalent to the partial likelihood.
• The proposed method effectively models and quantifies heterogeneity in treatment effects among centers.
• Demonstration of presenting center-specific treatment effect heterogeneity using confidence intervals for frailty.

Conclusions:

• The proposed h-likelihood approach provides a statistically sound and computationally feasible method for analyzing competing risks in multi-center trials with treatment effect heterogeneity.
• The method allows for robust investigation and presentation of variations in treatment effects across different study sites.