Comparing Cox Proportional Hazards (Cox PH) and Log-Logistic Survival Regression (LLSR) Models for Predicting Survival in Cancer-Associated Thrombosis (CAT)

Abstract

This study addresses the heightened risk of Cancer-associated Thrombosis (CAT), a significant cause of morbidity and mortality among cancer patients, particularly focusing on venous thromboembolism (VTE). Although VTE risk is perceived as lower in Asian populations, evidence suggests that factors like existence of chemotherapy treatment, advanced cancer staging, and immobility can negate this assumption. This study compares two survival analysis models: Cox Proportional Hazards (Cox PH) and Log-Logistic Survival Regression (LLSR) to evaluate their predictive capabilities in modelling CAT survival. A prospective cohort of 249 adult cancer patients from Hospital Canselor Tuanku Muhriz (HCTM) and Hospital Kuala Lumpur (HKL) was analysed. Survival time was defined as the duration between cancer diagnosis to death or censoring. Predictors included demographic and clinical variables. RStudio was used to fit the Cox PH and LLSR models. Evaluation metrics, including classification accuracy, sensitivity, specificity, and AUROC, are applied to real CAT datasets. Both models were successful in identifying significant survival predictors. The findings underscore the model’s effectiveness in predicting CAT risk, thereby advancing the precision of early detection and personalized treatment strategies for cancer patients. The model evaluation showed that Coc PH model performs slightly better compared to LLSR model which may be due to its flexibility to model hazard function without assuming specific distribution. This study's findings are poised to advance public health by optimizing CAT management, leading to improved patient care and clinical decision-making.