Development and External Validation of a Machine Learning Model for 90-Day Readmission in Hospitalized Older Patients with AECOPD: A Two-Center Study

Guibin Zhang,1 Dan Wang,1 Hang Chen,1 Wenchao Dai,1 Xingfu Fan,2 Yulian Liu,3 Li Jiang1 1Department of Respiratory and Critical Care Medicine, Affiliated Hospital of North Sichuan Medical College, Nanchong, Sichuan, 637000, People’s Republic of China; 2Department of General Medicine, Affiliated Hospital of North Sichuan Medical College, Nanchong, Sichuan, 637000, People’s Republic of China; 3Department of Respiratory and Critical Care Medicine, Dazhou Hospital of Integrated Traditional and Western Medicine, Dazhou, Sichuan, 635000, People’s Republic of ChinaCorrespondence: Li Jiang, Email [email protected]: Short-term readmission after hospitalization for acute exacerbation of chronic obstructive pulmonary disease (AECOPD) is common in older adults, yet early risk stratification remains limited. We aimed to develop and validate a 90-day readmission model using early admission data.Methods: This retrospective two-center study used a development cohort from the Affiliated Hospital of North Sichuan Medical College and an external validation cohort from Dazhou Integrated Traditional Chinese and Western Medicine Hospital. Predictors were limited to early admission variables harmonized across sites. The neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), and systemic immune-inflammation index (SII) were derived, with component blood counts removed to reduce collinearity. Feature selection used stability selection with Elastic Net regularization. Five models were trained and compared: multivariable logistic regression, naïve Bayes (NB), linear discriminant analysis (LDA), gradient boosting machine (GBM), and extreme gradient boosting (XGBoost). Discrimination was assessed by area under the receiver operating characteristic curve (AUC), with internal validation using bootstrap-derived optimism-corrected AUC and Brier score for overall error. Interpretability was examined with Shapley additive explanations (SHAP).Results: A total of 692 patients were included (development, n=513; external validation, n=179). Six predictors were retained: NLR, SII, D-dimer, frequent exacerbations (FE), body mass index (BMI), and albumin (ALB). No strong multicollinearity was detected (|r|< 0.90; variance inflation factors (VIFs) < 5). XGBoost showed the best discrimination in the development cohort (AUC=0.892) and remained stable after internal validation (optimism corrected AUC=0.864). In the external cohort, XGBoost again achieved the highest AUC (0.847) with a lower Brier score than alternative models. SHAP analyses indicated D-dimer, NLR, and FE as major contributors and suggested non-linear effects.Conclusion: Using early admission data, we developed and externally validated a 90-day readmission prediction model for older adults hospitalized with AECOPD. XGBoost showed stable performance and clinically interpretable risk patterns, supporting its potential for early risk stratification.Keywords: acute exacerbation of chronic obstructive pulmonary disease, 90-day readmission, extreme gradient boosting, external validation, shapley additive explanations, multicohort study