A1315
Title: Penalized deep partially linear cox models with application to CT scans of lung cancer patients
Authors: Yuming Sun - College of William and Mary (United States) [presenting]
Abstract: Lung cancer is a leading cause of cancer-related death worldwide, underscoring the need to understand mortality risks for effective, personalized treatment. The National Lung Screening Trial (NLST) used CT texture analysis to quantify image-based risk factors. Partially linear Cox models are increasingly used in survival analysis for integrating both traditional (e.g., age, clinical covariates) and novel (e.g., imaging features) risk factors by combining parametric and nonparametric components. However, when the number of parametric covariates exceeds the sample size, model fitting becomes challenging, and nonparametric modeling suffers from the curse of dimensionality. A novel penalized deep partially linear Cox model (Penalized DPLC) is proposed, which integrates the SCAD penalty for selecting important texture features and uses a deep neural network to estimate the nonparametric component. Theoretical guarantees are established for the estimator, and its superior performance is demonstrated through simulations in both prediction and feature selection. Finally, the method is applied to NLST data to investigate how clinical and imaging features relate to survival, offering insights into the prognostic value of these multimodal risk factors.
