Prediction model for death in patients with pulmonary tuberculosis accompanied by respiratory failure in ICU: retrospective study.

Background: The mortality rate of pulmonary tuberculosis (TB) patients with respiratory failure remains high. This study aimed to identify factors contributing to death in these patients, and develop a mortality prediction model for pulmonary TB patients with respiratory failure.

Methods: A retrospective study of patients admitted consecutively to the medical ICU of Beijing Chest Hospital, (Beijing, China), Chaoyang Fourth Hospital (Liaoning, China) and Hebi Third People's Hospital (Henan, China) from May 2018 to May 2019 was conducted. 153 patients with pulmonary TB accompanied by respiratory failure were enrolled. A multivariate analysis was performed to identify risk factors for death. A predictive fatality score was determined. The utility of the score for predicting death was evaluated using receiver operating characteristic (ROC) curve analysis.

Results: The patients' median age was 57.82±19.42 years (17.0-87.0 years) and there were 106 males (69.28%). The mortality rate was 39.22% (60 of 153). Independent predictive factors of mortality included the PaO2 (hazard ratio 0.928, 95% CI: 0.882 - 0.976, P=0.004), Albumin (hazard ratio 0.881, 95% CI: 0.792- 0.980, P=0.019), Apache II score (hazard ratio 1.120, 95% CI: 1.017-1.234, P=0.022) and C-reactive protein (hazard ratio 1.012, 95% CI: 1.004-1.019, P=0.003). Establishing a logistic model of the death risk grade of pulmonary TB with respiratory failure was Y=1.710 - 0.068*PaO2-0.163* albumin + 0.215* Apache II +0.012* C-reactive protein. The value was Y=-0.494. If the Y value was greater than or equal to -0.494, the patients belonged to the deceased group, and if less than -0.494 the patients belonged to the survival group. AUC=0.818, The sensitivity was 83.3%; specificity was 73.1%.

Conclusions: Pulmonary TB patients with respiratory failure have a high mortality rate and poor prognosis, particularly those with high Apache II scores, high C-reactive protein levels, low PaO2 admission to ICU and low albumin level. The prediction model will help assess the risk of death in patients with TB and respiratory failure.