Comparative Survival Modeling of Low-Grade Glioma Patients Using Targeted Proteomic Data

Abstract

Low-grade gliomas (LGG) have long and highly variable survival times, making individualized survival prediction challenging. In this study, we analyzed targeted proteomic data from 422 LGG patients from The Cancer Genome Atlas (TCGA) to evaluate the association between protein expression and overall survival. After pre-processing, 217 proteins were summarized into 18 latent factors, and one representative protein per factor was selected based on univariate Cox regression. Multivariable survival models were then fit, including both the Cox proportional hazards model and parametric approaches, with proteins violating the proportional hazards assumption excluded. Model performance was evaluated using goodness-of-fit metrics and the concordance index. This study aims to determine which protein markers are most strongly associated with survival, and to compare the performance of semi-parametric and parametric models trained on proteomic data in the presence of heavy right- censoring.