Multimodal machine learning predicts type 2 respiratory failure in COPD exacerbations: a multicenter XGBoost model...

Acute exacerbations of chronic obstructive pulmonary disease (AECOPD) frequently lead to life-threatening type 2 respiratory failure (T2RF). Existing predictive models rely on single biomarkers or linear methods and lack rigorous external validation. This study aimed to develop a multimodal machine learning framework to predict in-hospital T2RF risk with temporal-geographic external validation.

We employed a two-source design. A development cohort of 6,954 AECOPD patients from a single tertiary hospital (2023-2025) was randomly divided into training (n = 4,867) and internal test (n = 2,087) sets. A temporal external validation cohort included 1,252 patients from seven hospitals (2016-2020). Eighteen admission predictors were evaluated. Missing values were imputed using missForest. Hybrid feature selection (LASSO plus XGBoost ranking) identified key variables. Six algorithms-logistic regression, SVM, random forest, GBDT, LightGBM, and XGBoost-were compared. Performance was assessed by AUROC, sensitivity, specificity, calibration, decision curve analysis, and SHAP values. A logistic nomogram was constructed.

In the internal test set, XGBoost achieved an AUROC of 0.660 (95% CI: 0.631-0.689). In the external validation set, XGBoost achieved an AUROC of 0.699 (95% CI: 0.661-0.738), with 45.9% sensitivity and 79.0% specificity. LightGBM performed comparably (AUROC 0.700). Seven predictors were selected: lymphocyte count, eosinophil count, COPD duration, RDW-CV, age, hypertension, and sex. SHAP analysis identified low lymphocyte count and long COPD duration as dominant risk drivers. The logistic nomogram achieved an external AUROC of 0.666.

This externally validated framework enables early T2RF risk stratification at admission using routine blood counts and demographics. Future work should integrate dynamic monitoring and prospective multicenter validation.