Simultaneous variable selection and estimation for survival data via the Gaussian seamless- L 0 $$ {L}_0 $$ penalty.

Area of Science:

• Biostatistics
• Statistical modeling
• Survival analysis

Background:

• Variable selection and estimation are critical in survival analysis for building accurate predictive models.
• Existing methods like best subset selection (BSS) and regularization have limitations.
• There is a need for integrated approaches that leverage the strengths of both BSS and regularization.

Purpose of the Study:

• To propose a novel simultaneous variable selection and estimation procedure using the Gaussian seamless (GSELO) penalty.
• To apply the GSELO penalty within the framework of Cox proportional hazard and additive hazards models.
• To enhance the performance of existing variable selection techniques in survival analysis.

Main Methods:

• Development of the Gaussian seamless (GSELO) penalty for simultaneous variable selection and estimation.
• Implementation using an efficient iterative algorithm with established convergence properties.
• Proposal of an extended Bayesian information criteria (EBIC) for parameter tuning.
• Validation through simulated and real data studies.

Main Results:

• The GSELO procedure effectively integrates strengths from both best subset selection (BSS) and regularization.
• The developed iterative algorithm ensures computational efficiency.
• Theoretical analysis confirms the convergence and asymptotic properties of the GSELO procedure.
• The EBIC parameter selector optimizes GSELO performance.

Conclusions:

• The proposed GSELO procedure offers a powerful new tool for variable selection and estimation in survival models.
• The method demonstrates superior prediction performance compared to state-of-the-art techniques.
• The GSELO procedure, coupled with EBIC tuning, provides an effective and computationally efficient solution for survival data analysis.