Establishing and Validating a Risk Model for In-hospital Mortality Within 60 Days Under ICU Treatment for Tuberculosis in China

Abstract

Background Tuberculosis (TB) is the leading cause of death from a single infectious disease. Current studies on TB patient mortality risk factors in intensive care are old and scarce. We aimed to create a model to predict in-hospital mortality risk for TB patients in ICU and identify mortality risk factors. Methods TB patients' data from 2016 to 2020 admitted to the ICU were collected retrospectively and randomly split into derivation and validation groups at a 7:3 ratio. The main outcome was 60-day in-hospital mortality. Analyses included Cox, nomogram, decision curve, and Kaplan‒Meier methods. Results A total of 848 patients were included (594 in the derivation group and 254 in the validation group). A total of 106 (17.85%) patients died in the derivation group. Multivariate Cox regression analysis revealed that sputum smear, severe pneumonia, c-TnI, mold, age, diastolic blood pressure (DBP), and tracheotomy were independent risk factors for 60-day in-hospital mortality in ICU patients with TB, and the prognostic index (PI) was defined as follows: PI = 0.0084 × Age − 0.0026 × DBP + 2.1988 × Severe pneumonia1 + 0.9094 × Tracheotomy1 + 1.2253 × Sputum smear1 + 0.826 × Mold1 + 0.5147 × c-TnI. Decision curve analysis (DCA) diagrams showed that the diagnostic probabilities of the derivation and validation groups were 0–70% and 0–58% respectively, with high model application accuracy and net benefit. Receiver operating characteristic (ROC) curve analysis revealed that the PI could predict death with good sensitivity (0.830) and specificity (0.867), and the cutoff value was 0.195 (the area under the curve (AUC) was 0.894, 95% CI: 0.865 to 0.924). K‒M analysis revealed that the proportion of deaths was increased when the PI was ≥ 0.195. Conclusion The nomogram-based prediction model of mortality within 60 days in TB patients in the ICU showed good discrimination and accuracy, and is of great clinical value for screening patients at high risk of death to support the development of intervention strategies for ICU patients with TB and to reduce mortality.