An Interpretable AdaBoost Model for 1-Year Readmission Risk Prediction in AECOPD Patients with Hypertension

Xinxin Zhang,1 Maolang He,1 Jingyi Zhang,2 Luna Zhao,1 Dong Liu1 1Department of Pulmonary and Critical Care Medicine, The First Affiliated Hospital of Shihezi University, Shihezi, People’s Republic of China; 2Shihezi University School of Medicine, Shihezi, People’s Republic of ChinaCorrespondence: Dong Liu, Department of Pulmonary and Critical Care Medicine, The First Affiliated Hospital of Shihezi University, No. 107 North second Road, Hongshan Street, Shihezi, 832008, People’s Republic of China, Email [email protected]: Chronic obstructive pulmonary disease (COPD) complicated by hypertension imposes a substantial global health burden, with acute exacerbation of chronic obstructive pulmonary disease (AECOPD) significantly increasing 1-year readmission risk. This study aimed to develop and validate an interpretable machine learning (ML) model that predicts 1-year readmission risk in AECOPD patients complicated by hypertension using real-world data.Methods: This retrospective cohort study enrolled 2042 patients with AECOPD complicated by hypertension from the First Affiliated Hospital of Shihezi University between 2015 and 2024. The data were split into training and test sets at a 7:3 ratio. Feature selection was performed based on machine learning methods. Eight ML models were trained and tested to construct predictive models. Model performance was evaluated by the area under the receiver operating characteristic curve (AUC), accuracy, recall, specificity, and F1-score. The Shapley additive explanation method (SHAP) was used to rank the feature importance and explain the final model. An online risk prediction tool was developed based on the optimal model to facilitate clinical application.Results: The 1-year readmission rate of patients with AECOPD complicated by hypertension was 37.5%. Seven independent predictors, including times of inhospitalization, procalcitonin, total protein, international normalized ratio (INR), prothrombin time, D-dimer, and hypoproteinemia, were identified as the most valuable features for establishing the models. The AdaBoost model showed optimal performance, with an AUC of 0.884 in the test set and an average AUC of 0.889 in 5-fold cross-validation. SHAP analysis confirmed that times of inhospitalization were the strongest predictor, followed by INR and total protein. An online calculator was deployed (https://fast.statsape.com/tool/detail?id=17) for clinical use.Conclusion: This study developed an interpretable AdaBoost-based online calculator for 1-year readmission risk assessment in AECOPD patients by hypertension. The tool highlight the importance of addressing hypercoagulability and nutritional status to reduce readmission risk. Further external multi-center validation is needed to enhance its generalizability.Keywords: acute exacerbation of chronic obstructive pulmonary disease, hypertension, 1-year readmission, machine learning, web calculator