Cox regression analysis of multivariate failure time data: the marginal approach

Area of Science:

• Biostatistics
• Survival Analysis
• Epidemiology

Background:

• Multivariate failure time data arise from subjects experiencing multiple events or clustered data.
• Existing methods may not adequately handle the complex dependencies in such data.
• Accurate analysis is vital for drawing valid conclusions in clinical and epidemiological research.

Purpose of the Study:

• To present a general methodology for analyzing multivariate failure time data.
• To develop robust statistical methods that account for intra-class correlation.
• To provide a flexible approach analogous to longitudinal data analysis.

Main Methods:

• Formulating marginal distributions using Cox proportional hazards models.
• Leaving the dependence structure among related failure times unspecified.
• Developing simple estimating equations and robust variance-covariance estimators.

Main Results:

• The proposed methodology yields consistent and asymptotically normal estimators.
• Robust variance estimators effectively account for intra-class correlation.
• Simulation results confirm the adequacy of large-sample approximations and highlight risks of ignoring correlation.

Conclusions:

• The presented methodology offers a reliable approach for analyzing multivariate failure time data.
• Ignoring intra-class correlation can lead to misleading variance estimates.
• The method is implemented in user-friendly software and illustrated with real-world data.