An EM-based semi-parametric mixture model approach to the regression analysis of competing-risks data

Area of Science:

• Biostatistics
• Survival Analysis
• Medical Statistics

Background:

• Competing risks data present unique challenges in regression analysis.
• Understanding factors influencing both event probability and hazard rates is crucial.

Purpose of the Study:

• To develop and evaluate a semi-parametric mixture model for analyzing competing risks.
• To jointly estimate logistic and regression coefficients for event probabilities and hazard rates.

Main Methods:

• A semi-parametric mixture model combining logistic and proportional hazards models.
• Maximum likelihood estimation using an Expectation-Conditional Maximization (ECM) algorithm.
• Comparison with a fully parametric mixture approach via simulation studies.

Main Results:

• The semi-parametric method offers comparable performance to the fully parametric approach when baseline hazards are monotonic.
• For non-monotonic baseline hazards, the semi-parametric method yields less biased estimates.
• Efficiency is comparable across all censoring levels for the semi-parametric method.

Conclusions:

• The proposed semi-parametric mixture model is a flexible and reliable tool for competing risks regression.
• It demonstrates superior performance in scenarios with non-monotonic hazard functions.
• The method is validated using prostate cancer patient data.