Penalized variable selection for cause-specific hazard frailty models with clustered competing-risks data.

Area of Science:

• Biostatistics
• Clinical Trials
• Survival Analysis

Background:

• Competing risks data are common in clustered clinical studies, where events preclude others.
• Existing frailty models for clustered competing risks data lack variable selection methods.
• Cause-specific hazard frailty models are crucial for analyzing such complex data.

Purpose of the Study:

• To develop and evaluate a novel variable selection procedure for fixed effects in cause-specific competing risks frailty models.
• To address the gap in existing literature regarding variable selection for these complex models.
• To compare the performance of different penalty functions for variable selection.

Main Methods:

• Proposed a variable selection procedure using a penalized h-likelihood (HL) approach.
• Investigated three penalty functions: LASSO, SCAD, and HL.
• Conducted simulation studies to assess the performance of the proposed methods.
• Applied the method to real-world clustered competing-risks cancer data.

Main Results:

• The proposed variable selection procedure using the HL penalty demonstrated superior performance.
• The HL penalty achieved a higher probability of selecting the true model compared to LASSO and SCAD.
• Prediction accuracy was maintained while improving model selection.

Conclusions:

• The penalized h-likelihood approach provides an effective method for variable selection in cause-specific competing risks frailty models.
• The HL penalty is recommended for its balance of model selection accuracy and prediction performance.
• This method offers a valuable tool for analyzing complex clustered competing risks data in clinical research.