Cluster validity indices for mixture hazards regression models.

Area of Science:

• Biostatistics
• Survival Analysis
• Medical Statistics

Background:

• Competing risks analysis is crucial in medical studies where subjects can experience multiple failure types.
• Semiparametric mixture regression models offer flexibility and interpretability for competing risks.
• Existing methods lack robust approaches for determining the number of model components and estimating baseline hazards.

Purpose of the Study:

• To propose novel validity indices for selecting the optimal number of components in mixture Cox regression models.
• To introduce a kernel-based approach for smooth estimation of the baseline hazard function in mixture models.
• To address limitations in current methods for competing risks analysis.

Main Methods:

• Developed four validity indices based on posterior probabilities and residuals from an Expectation-Maximization (EM) algorithm.
• Implemented a kernel smoothing approach for baseline hazard estimation.
• Utilized mixture Cox regression models for competing risks data.

Main Results:

• The proposed validity indices effectively determine the optimal number of model components.
• The kernel approach provides smooth and accurate estimates of the baseline hazard function.
• Simulation studies demonstrate the superiority of the proposed cluster indices.

Conclusions:

• The new methods enhance the accuracy and reliability of mixture Cox hazard models in competing risks analysis.
• The approach is validated through simulation and a real-world prostate cancer dataset.
• This research provides valuable tools for biostatisticians and medical researchers.