A SHAP-based machine learning model for predicting prolonged hospitalization in AECOPD: a single-center study from China

Prolonged hospitalization in acute exacerbations of chronic obstructive pulmonary disease (AECOPD) is associated with adverse clinical and healthcare system outcomes. However, few studies have applied machine learning methods to predict this outcome. This study aimed to develop and internally evaluate an interpretable machine learning model for estimating the risk of prolonged length of hospital stay (LOHS ≥ 14 days) in hospitalized AECOPD patients using routinely available clinical data. We retrospectively analyzed 1,286 patients hospitalized for AECOPD between 2021 and 2023. Prolonged hospitalization was defined as LOHS ≥ 14 days. Twelve machine learning algorithms were developed and evaluated using selected clinical variables. Model performance was assessed using the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, calibration plots, Brier score, and decision curve analysis (DCA). SHapley Additive exPlanations (SHAP) was applied to improve model interpretability. Among the evaluated models, extreme gradient boosting (xgb) showed the most favorable overall performance in terms of AUC, Brier score, calibration, and DCA, achieving an AUC of 0.754 (95% CI: 0.713–0.799), specificity of 88.1%, and the lowest Brier score (0.110) in the testing set, although its sensitivity was only 31.0% and its F1 score was limited. SHAP analysis identified CRP, RDW, NLR, PLR, MLR, PaCO 2 , proBNP, albumin, and WBC as important contributors to model predictions. The xgb model demonstrated moderate discriminative performance for predicting prolonged hospitalization in patients hospitalized with AECOPD and showed good interpretability through SHAP analysis. These findings may provide auxiliary information for risk stratification in hospitalized patients with AECOPD, although further multicenter external validation is required before clinical application.