Bayesian Survival Modeling: A Proportional Hazards Approach for Cervical Cancer Prognosis

Abstract

Cervical cancer continues to be a major global public health concern, with high incidence and mortality rates in many regions. In this study, a Bayesian Proportional Hazards (PH) model is employed to analyze survival outcomes in cervical cancer patients using data from the cBioPortal. Using Bayesian inference, regression coefficients are estimated while incorporating prior knowledge to improve model stability and interpretability. The study includes multiple genetic and clinical covariates, such as TP53, CDKN2A, CDKN1A, MDM2, and tumor staging variables, to identify significant prognostic factors. The posterior results reveal that TP53, MDM2, and CTLA4 have statistically significant negative effects on the hazard, indicating their potential protective role in survival. Conversely, clinical variables like Node Stage N1 and Tumor Stages T3 and T4 are significantly associated with increased mortality risk. In contrast, variables such as CDKN2A, CDKN1A, FANCA, FANCL, XRCC1, TGFB1I1, TNFAIP3, metastasis stage, and age exhibit 95% credible intervals that include zero, indicating a lack of statistically significant association with survival. These findings are further supported by survival probability plots at the median age of 46 years. In conclusion, the Bayesian PH model offers a robust, probabilistic approach to survival analysis, effectively identifying high-risk patients and aiding personalized treatment strategies in cervical cancer care.