Tuberculosis (TB) remains a significant global health burden, necessitating robust statistical models to understand survival dynamics and inform interventions. Most survival analyses rely on the Cox Proportional Hazards (Cox PH) model, which may not adequately capture the survival time distribution. This study focuses on data from the National Tuberculosis and Leprosy Center (NTLC), Zaria, Kaduna State, Nigeria, to identify factors influencing TB survival and assess alternative parametric survival models. The study aims to identify factors associated with TB mortality, assess their impact on survival outcomes, and compare the performance of Weibull, and Log-Logistic Accelerated Failure Time (AFT) models to determine the most suitable model for TB survival data from NTLC Zaria. This study compares the performance of two Accelerated Failure Time (AFT) models, the Weibull, and the Log-Logistic in analyzing TB survival data. The analysis evaluates model fit using p-values, log-likelihood, and Akaike Information Criterion (AIC). Results indicate that the Weibull AFT model outperforms the Log-logistic, with the highest log-likelihood (-228.6) and the lowest AIC (485.11), and the Log-Logistic AFT model (AIC: 492.02, log-likelihood: -232.0). The p-values for both models demonstrate statistical significance, highlighting their effectiveness in modelling TB survival data. However, the Weibull model's higher performance suggests it better captures survival time variability in TB patients. These findings emphasize the importance of selecting appropriate survival models for TB data analysis and support the application of the Weibull AFT model for future studies. Further research should explore integrating advanced statistical techniques and machine learning approaches to enhance predictive accuracy and improve TB management strategies. This study contributes to this growing field by applying parametric survival models, to analyse TB survival data from the National Tuberculosis and Leprosy Centre (NTLC) in Zaria, Nigeria.
Read full abstract