Cross-validation approaches for penalized Cox regression.

Area of Science:

• Biostatistics
• Survival Analysis
• Machine Learning in Medicine

Background:

• Penalized regression models are widely used for high-dimensional data.
• Cross-validation is standard for tuning parameter selection in penalized regression.
• Application of cross-validation to penalized Cox models is less explored due to partial likelihood complexities.

Purpose of the Study:

• To propose and evaluate a novel cross-validation approach for penalized Cox models.
• To compare the new method with existing cross-validation strategies for Cox models.
• To assess the performance and numerical stability of different cross-validation techniques.

Main Methods:

• Development of a new cross-validation method based on linear predictors of the Cox model.
• Comparison with previously proposed cross-validation approaches for penalized Cox models.
• Validation using simulated datasets and a real-world high-dimensional gene expression dataset.

Main Results:

• The proposed cross-validation approach demonstrates a favorable balance of performance and numerical stability.
• The method effectively handles parameter selection in penalized Cox models.
• Successful application to a high-dimensional lung cancer gene expression and survival study.

Conclusions:

• The novel cross-validation strategy offers a practical and robust solution for penalized Cox models.
• This approach enhances the reliability of tuning parameter selection in survival data analysis.
• The findings are relevant for high-dimensional survival data analysis, particularly in genomics and clinical research.