An interpretable machine learning model for predicting all-cause in-hospital mortality in patients with ischemic stroke and coronary heart disease: A multicenter study in China

Timely and accurate prediction of clinical outcomes in patients with cardio-cerebral infarction (CCI) is essential for optimizing treatment strategies and improving prognosis. This study aimed to develop an interpretable machine learning model for predicting all-cause in-hospital mortality in patients with CCI. A total of 4105 patients with CCI from a multicenter database were retrospectively analyzed. Nine machine learning (ML) models were constructed to predict in-hospital mortality. Feature selection was performed using the recursive feature elimination algorithm. Model performance was evaluated using the F1 score, area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and accuracy. The model with the highest predictive performance was further interpreted using the SHapley Additive exPlanations (SHAP) method to provide individualized explanations for predictions. Of the 4105 patients, 2071 (50.5 %) were female, with a median age of 79 years (interquartile range IQR: 72, 85). After feature selection, eight of the initial 43 clinical variables were retained for model development. Among the nine models, the Random Forest (RF) model achieved the highest predictive performance, with an AUC of 0.88 in the test cohort. Feature importance analysis identified serum potassium (K), N-terminal pro-B-type natriuretic peptide (NT-proBNP), troponin T, cholinesterase, creatinine, age, albumin (ALB), and platelet count (PLT) as the most influential predictors. SHAP visualization was applied to enhance the interpretability of the RF model. The proposed ML-based risk prediction model demonstrated robust performance in predicting in-hospital mortality among patients with CCI. The integration of SHAP analysis provided transparent, patient-specific risk interpretation, thereby supporting clinical understanding of critical prognostic factors and enabling personalized decision-making.