A Flexible Adaptive Lasso Cox Frailty Model Based on the Full Likelihood

Area of Science:

• Biostatistics
• Survival Analysis
• Statistical Modeling

Background:

• Cox frailty models are widely used for survival data analysis.
• Existing methods often struggle with time-varying covariates and coefficients.
• There is a need for robust regularization techniques in complex survival models.

Purpose of the Study:

• To propose a novel method for regularizing Cox frailty models.
• To accommodate time-varying covariates and coefficients using a full likelihood approach.
• To enable smooth, semiparametric modeling of the baseline hazard.

Main Methods:

• Utilizes a full likelihood framework instead of partial likelihood.
• Employs lasso and group lasso penalties for variable selection.
• Incorporates a second penalty for smoothing time-varying coefficients and baseline hazard.
• Includes adaptive weights for estimation stabilization.
• Implemented in the R function coxlasso within the PenCoxFrail package.

Main Results:

• The proposed method effectively regularizes Cox frailty models.
• Accommodates complex scenarios with time-varying parameters.
• Allows for smooth estimation of the baseline hazard function.
• Demonstrates stability and accuracy in estimation.

Conclusions:

• The new regularization method provides a flexible and powerful tool for Cox frailty models.
• It enhances the analysis of survival data with time-varying factors.
• The PenCoxFrail package offers a valuable resource for researchers in survival analysis.