ADVANCED SURVIVAL MODELING OF TUBERCULOSIS PATIENTS: INSIGHTS FROM EXPONENTIAL AND WEIBULL AFT MODELS

Abstract

Tuberculosis is a significant public health issue in high-burden countries like Nigeria, causing increased disability and claiming many lives. The Cox Proportional Hazards model is commonly used in survival studies, but it fails to define the distribution of survival time. This study uses data from the National Tuberculosis and Leprosy Center (NTLC), Zaria, Kaduna State, Nigeria, to determine Tuberculosis survival and compare alternative parametric survival models. The objectives include determining predictors of TB mortality, evaluating the effect of these predictors on survival probability, and comparing Exponential and Weibull AFT models on NTLC Zaria TB survival data. The results show that the Weibull AFT model is most effective in modelling TB survival rates, with the lowest AIC score of 485.1 and the highest log likelihood of -228.6. Major factors associated with mortality include age over 55 years, pulmonary tuberculosis, family history of tuberculosis, alcohol and smoking history, and BMI less than 18.5 kgs/m2. The study emphasizes the need for region-specific survival models to reveal major directions for successful interventions and TB policies. Future studies should consider translating the highly-parametric approach into next-generation non-parametric models/machine learning for more accurate prognoses for implementing state-of-the-art public health interventions.