Acute respiratory failure (ARF) is a common complication of bronchial asthma (BA). ARF onset increases the risk of patient death. This study aims to develop a predictive model for ARF in BA patients during hospitalization.
This was a retrospective cohort study carried out at two large tertiary hospitals. Three models were developed using three different ways: (1) the statistics-driven model, (2) the clinical knowledge-driven model, and (3) the decision tree model. The simplest and most efficient model was obtained by comparing their predictive power, stability, and practicability.
This study included 398 patients, with 298 constituting the modeling group and 100 constituting the validation group. Models A, B, and C yielded seven, seven, and eleven predictors, respectively. Finally, we chose the clinical knowledge-driven model, whose C-statistics and Brier scores were 0.862 (0.820–0.904) and 0.1320, respectively. The Hosmer-Lemeshow test revealed that this model had good calibration. The clinical knowledge-driven model demonstrated satisfactory C-statistics during external and internal validation, with values of 0.890 (0.815–0.965) and 0.854 (0.820–0.900), respectively. A risk score for ARF incidence was created: The A 2-BEST 2 Risk Score (A 2 (area of pulmonary infection, albumin), BMI, Economic condition, Smoking, and T 2(hormone initiation Time and long-term regular medication Treatment)). ARF incidence increased gradually from 1.37% (The A 2-BEST 2 Risk Score ≤ 4) to 90.32% (A 2-BEST 2 Risk Score ≥ 11.5).