Diagnosis-time Risk Stratification for Tuberculosis Mortality in a Nationwide Public-Private Mix Program

Mortality remains a major barrier to tuberculosis (TB) control, as many deaths occur shortly after diagnosis. We aimed to identify determinants of mortality and develop diagnosis-time risk-prediction models for patients with TB.

We conducted a nationwide population-based cohort study using data from South Korea's public-private mix TB program, including patients diagnosed between January 2019 and December 2022. Two models were developed in the 2019-2021 cohort and validated in the independent 2022 cohort: TREAT-TB, incorporating demographic, clinical, radiologic, and microbiological variables, and SCREEN-TB, based only on demographic characteristics, symptoms, and comorbidity data. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), Brier score, and calibration analysis.

Among 24,745 patients, 2,667 (10.8%) died during treatment. Nearly half of the deaths occurred within two months of diagnosis. Older age, lower body mass index, and organ-specific comorbidities, particularly cardiovascular, neurological, and renal diseases, were associated with mortality. TREAT-TB and SCREEN-TB showed similar discrimination and good calibration in the validation cohort (AUCs 0.807 and 0.805, respectively), with mortality increasing across risk scores.

Diagnosis-time risk stratification is feasible without radiologic or microbiological results and may support early triage and targeted monitoring within TB control programs.