Piecewise cause-specific association analyses of multivariate untied or tied competing risks data

Area of Science:

• Biostatistics
• Epidemiology
• Survival Analysis

Background:

• Multivariate competing risks data analysis is crucial for understanding complex health outcomes.
• Existing methods for bivariate survival data do not adequately address multivariate competing risks, especially with tied events.

Purpose of the Study:

• To extend the bivariate hazard ratio to multivariate competing risks data.
• To propose and evaluate novel estimators for association measures in multivariate competing risks.
• To address the challenge of tied events in association analyses.

Main Methods:

• Extension of plug-in and pseudo-likelihood estimators for multivariate competing risks.
• Establishment of asymptotic properties using empirical processes.
• Development and application of a modified U-statistic to handle tied observations.

Main Results:

• The proposed estimators are equivalent to cause-specific cross hazard ratios.
• Extended plug-in and pseudo-likelihood estimators perform comparably to U-statistics without tied events.
• A modified U-statistic significantly outperforms other methods when dealing with tied events and rounding errors.

Conclusions:

• The modified U-statistic offers a reliable approach for association analysis in multivariate competing risks data with tied events.
• This work provides valuable tools for analyzing complex survival data in epidemiological studies.
• The findings are demonstrated through an application to the Cache County Study examining dementia associations.