Development and Validation of Clinical Prediction Score of Mortality in Tuberculosis Patients

OBJECTIVES: Tuberculosis (TB) is one of public health problems both globally and nationally. Death is the main serious problem for tuberculosis (TB) treatment programs. Therefore, this study aimed to develop a simple risk scoring for predicting mortality in TB patients.

METHODS: Data were collected from TB patients registry in Phichit Hospital and Community Hospitals in Phichit Province, Thailand between January 1, 2017 - December 31, 2020. Eligible patients were over 18 years old and have completed treatment or death. All TB patients were diagnosed by clinicians using the standard procedure, i.e. chest x-ray, sputum acid-fast bacillus (AFB) and/or GeneXpert test. To develop and internal validate the risk score system, we used multivariable logistic regression, and weighted point using regression coefficient. The scoring scheme was applied in validation cohort to test the diagnostic performances.

RESULTS: A total of 2,157 patients were included in the study and randomly divided into 2 groups; derivation cohort (n=1,585), and validation cohort (n=572). The risk score consists of 4 predictors: Age ≥ 65 years, liver disease, HIV status positive and TB meningitis. The risk score showed area under the receiver operating characteristic curve (AuROC) 78.53% with good calibration (Hosmer-Lemeshow χ²=2.47; P=.29). The positive likelihood ratio of death in TB patients with low risk (scores ≤ 2.5) and high risk (scores > 4.5) were 1.22 (95% CI:1.15-1.29) and 25.63 (95% CI:7.56-86.87) respectively. When applied in validation cohort, the score showed good performance with AuROC 76.93%, and illustrated 85.19%, and 77.78% certainty in low-and high-risk groups respectively.

CONCLUSIONS: The simple risk score with four predictors developed from routine data collection might be used to predict mortality in TB patients care. This risk score may help clinicians in terms of planning a proper care of TB patients.