Modeling Time to Death of Patients with Multidrug-Resistant Tuberculosis at Saint Peter's Specialized Hospital

Background: Currently, the worldwide prevalence and incidence of multidrug-resistant tuberculosis (MDR-TB) is drastically increasing. The main objective of this study was modeling the time-to-death of patients with MDR-TB at St. Peter's Specialized Hospital, Addis Ababa, Ethiopia, by using various parametric shared frailty models. Study Design: A retrospective study design was used. Methods: The study population was TB patients with MDR at St. Peter's Specialized Hospital from January 2016 through December 2019. Exponential, Weibull, and log-normal were used as baseline hazard functions with the gamma and inverse Gaussian frailty distributions. All the models were compared based on Akaike's Information Criteria. Results: The overall median time to death was 11 months and 123 (33.5%) patients died. Patients who lived in rural areas had shorter survival time than those who lived in urban areas with an accelerated factor of 0.135 (P=0.002). Patients with a history of anti-TB drug consumption had a short survival time than those without such a history with an accelerated factor of 0.02 (P=0.001). The variability (heterogeneity) of time to death of patients in the region for the selected model (Weibull-inverse Gaussian shared frailty model) was theta=0.144 (P=0.027). Conclusion: The MDR-TB patients with weight gain, khat and alcohol consumption, clinical complication of pneumothorax and pneumonia, extrapulmonary TB, and history of anti-TB drug consumption as well as those who lived in rural areas had a shorter survival time, compared to others. There was a significant heterogeneity effect in the St. Peter's Specialized Hospital. The best model for predicting the time to death of MDR-TB patients was Weibull-inverse Gaussian shared frailty model.