Introduction: Tuberculosis (TB) is one of major causes of death and active public health problems in Thailand. The current TB mortality rate is still high.

Objective: This study aimed to identify prognostic factors which can predict mortality and develop a risk score system to predict the mortality rate of new TB cases.

Methods: A retrospective cohort study was conducted at Phrae hospital. Newly diagnosed TB patients who were registered between 1 October 2013 to 30 June 2020 were enrolled. The relationship between death and prognostic factors including age, body mass index (BMI), extra-pulmonary TB, initial hospitalization at diagnosis time, HIV infection, COPD, CKD, diabetes, and cancer were analyzed by the hazard ratio. The prognostic prediction model was developed under the cox proportional hazard model. The goodness-of-fit test to trial the final prediction model by the Hosmer and Lameshow test were applied. The potential of the risk score level was tasted by the likelihood ratio positive (LR+)

Results: 791 new TB patients were enrolled. Five prognostic factors, including initial hospitalization at diagnosis time, cancer, age (>60), BMI (<18.5 kg/m²), and sputum AFB smear positive were associated with mortality of new TB cases. The mathematical equation of the final prediction model showed the c-stat was 0.82. The risk scores were set from 0 to 13 and classified into 3 levels: low risk (0-3), moderate risk (4-7), and high risk (8-13). The level of LR+ were 1.00(0.63-1.60), 2.22(1.84-2.26), and 5.59(4.3-7.28) respectively.

Conclusion: This prognostic prediction model can effectively predict mortality in the high-score group of new TB cases. Consequently, these patients should be closely followed up.