Predicting 30-day in-hospital mortality in ICU asthma patients: a retrospective machine learning study with external validation.

Abstract

Background: Asthma-related mortality in the intensive care unit (ICU) remains poorly characterized, with no existing predictive models specifically designed for this high-risk population. This study aimed to develop and externally validate a machine learning-based model to predict 30-day in-hospital mortality among ICU patients with asthma.

Methods: The model was developed using data from MIMIC-IV 2.2 and externally validated on a subset of MIMIC-IV 3.1. Clinical variables from the first 24 h of ICU admission were extracted. Feature selection was conducted using both LASSO regression and the Boruta algorithm. Seven machine learning algorithms were trained and evaluated using receiver operating characteristic (ROC) curves, calibration plots, and decision curve analysis. The best-performing model was identified based on internal and external validation results. SHapley Additive exPlanations (SHAP) were employed to interpret feature importance. The final model was deployed as an interactive web-based tool.

Results: A total of 4385 ICU asthma patients were analyzed. The final XGBoost model, using 12 features, achieved the highest AUROC in both internal (0.83) and external (0.80) validation, and demonstrated the best calibration and net clinical benefit. SHAP analysis identified age, respiratory rate, RDW, urine output, and anion gap as top predictors. The model outperformed conventional ICU scores and is available as a web-based tool.

Conclusions: We developed and externally validated a robust prediction model for 30-day mortality in ICU patients with asthma. The model offers strong performance, interpretability, and clinical utility, supporting its use for real-time risk stratification and decision-making in critical care settings.