Kernel Cox partially linear regression: Building predictive models for cancer patients' survival

Area of Science:

• Bioinformatics
• Computational Biology
• Cancer Research

Background:

• Cancer patient survival shows wide heterogeneity, necessitating accurate predictive models linking molecular profiles to outcomes.
• High-dimensional molecular data presents challenges for nonparametric survival modeling and simultaneous irrelevant predictor removal.

Purpose of the Study:

• To develop a novel method for accurately predicting cancer patient survival by integrating molecular data.
• To address the challenge of simultaneously modeling complex relationships and removing irrelevant predictors in high-dimensional survival analysis.

Main Methods:

• Proposed a kernel Cox proportional hazards semi-parametric model.
• Introduced a novel Regularized Garrotized Kernel Machine (RegGKM) method incorporating a LASSO penalty.
• Developed an efficient high-dimensional algorithm for the RegGKM method.

Main Results:

• The RegGKM method demonstrated superior predictive accuracy compared to competing methods in simulations.
• The method effectively models complex relationships between molecular predictors and survival.
• Applied to a multiple myeloma dataset, it predicted patient death burden based on gene expression.

Conclusions:

• The RegGKM method offers a powerful tool for accurate cancer survival prediction using molecular data.
• It aids in classifying patients into distinct death risk groups for tailored treatment strategies.
• This approach facilitates improved clinical outcomes by enabling risk-stratified patient management.