Explainable machine learning for predicting 30-day readmission in acute heart failure patients

We aimed to develop a machine-learning based predictive model to identify 30-day readmission risk in Acute heart failure (AHF) patients. In this study 2232 patients hospitalized with AHF were included. The variance inflation factor value and 5-fold cross-validation were used to select vital clinical variables. Five machine learning algorithms with good performance were applied to develop models, and the discrimination ability was comprehensively evaluated by sensitivity, specificity, and area under the ROC curve (AUC). Prediction results were illustrated by SHapley Additive exPlanations (SHAP) values. Finally, the XGBoost model performs optimally: the greatest AUC of 0.763 (0.703-0.824), highest sensitivity of 0.660, and high accuracy of 0.709. This study developed an optimal XGBoost model to predict the risk of 30-day unplanned readmission for AHF patients, which showed more significant performance compared with traditional logistic regression (LR) model.