Survival analysis, development & validation of tb-related death risk score of adult pulmonary tuberculosis in Selangor, Malaysia

Patients with Tuberculosis (TB) continue to die despite effective treatment. The study aimed to assess the 1 -year survival, develop and validate a TB-related death predictive risk score on adult Pulmonary TB (PTB) patients on treatment. A retrospective cohort study was conducted using data of eligible adult PTB patients registered in Selangor, Malaysia between 1st January 2013 and 31st December 2019 obtained from Selangor’s State Health Department. The survival function of the participants was assessed through the Kaplan-Meier method. Survival functions between and among groups within a variable were compared using Log-Rank tests. Out of 24570 enrolled participants, 595 (2.4%) died from TB. The mean survival time was 356.2 days and the 1-year survival rate was 96.9%. Sociodemographic and clinical TB-related death risk factors were determined through multivariable Cox Proportional Hazard model, TB-related death were significantly associated with far (adjusted Hazard Ratio (aHR) 9.98, 95%CI 4.28- 23.28) and moderately advanced (aHR 3.23, 95%CI 1.43-7.31) radiological findings at diagnosis; concurrent TB meningitis (aHR 7.67, 95%CI 4.53-12.98) and miliary TB (aHR 6.32, 95%CI 4.10-9.74) involvement; HIV positive at diagnosis (aHR 2.81, 95%C1 2.21-3.57); Hulu Selangor (aHR 1.95, 95%CI 1.29-2.93), Klang (aHR 1.53, 95%C1 1.18-1.98) and Hulu Langat (aHR 1.31,95%C1 1.03-1.68) residing districts; no formal education (aHR 1.70, 95%CI 1.23-2.35); unemployment (aHR 1.54, 95%CI 1.29-1.84), positive sputum smear AFB at diagnosis (aHR 1.51, 95%CI 1.22-1.85); rural residency (aHR 1.39, 95%CI 1.13-1.72); and advancing age (aHR 1.02, 95%CI 1.02-1.03). Subsequently, the data set was divided into development (registered patients between the years 2013 and 2017) and validation (patients between 2018 and 2019) data sets. A predictive model based on the determined TB-related death risk factors was developed using the development data set. The apparent model performance overall was good. The model was validated internally (using the simple bootstrap resampling method) and externally (using the validation data set). The model's predictive performance on internal validation (overall performance: R2D 0.38 (95%C1 0.33-0.42); discrimination: Harrell's C 0.75 (95%CI 0.73-0.78), Uno's C 0.76 (95%CI 0.73-0.78); calibration: observed-expected ratio 1.02 (95%CI 0.91-1.14), predictive index regression 1.12 (95%CI 1.09-1.15)) and external validation (overall performance: R2D 0.30 (95%CI 0.24-0.34); discrimination: Harrell's C 0.72 (95%CI 0.69-0.75), Uno's C 0.68 (95%CI 0.64-0.73); calibration (observed-expected ratio 0.97 (95%CI 0.87-1.08), predictive index regression 1.02 (95%CI 0.89-1.16)) were satisfactory. In summary, the study provides the latest information on TB-related death risk factors among local TB patients. Furthermore, the TB-Related Death Risk Score should assist clinicians in identifying high-risk TB patients for further management and improving patients's understanding of their own risk of TB-related death, encouraging their treatment adherence.