Machine learning-based prediction model for post-stroke cerebral-cardiac syndrome: a risk stratification study

Cerebral-cardiac syndrome (CCS) is a severe cardiac complication following acute ischemic stroke, often associated with adverse outcomes. This study developed and validated a machine learning (ML) model to predict CCS using clinical, laboratory, and pre-extracted imaging features. A retrospective cohort of 511 post-stroke patients was analyzed. Data on demographics, laboratory results, imaging findings, and medications were collected. CCS diagnosis was based on cardiac dysfunction occurring after stroke, excluding pre-existing cardiac diseases. Five machine learning models, including Logistic Regression, Random Forest, Support Vector Machine (SVM), XGBoost, and Deep Neural Network, were trained on 80% of the data and tested on the remaining 20%. Discrimination was assessed by AUC (95% CI), calibration by Hosmer-Lemeshow test and Brier score, and thresholds by Youden's index. Model interpretability was evaluated using SHAP. On the test set, XGBoost achieved the highest discrimination (AUC 0.879; 95% CI 0.807-0.942), accuracy 0.825, precision 0.844, recall 0.675, and F1 score 0.750. Random forest followed closely (AUC 0.866; accuracy 0.845; precision 0.962; recall 0.625; F1 0.758). SVM and logistic regression yielded AUCs of 0.853 and 0.818, respectively. Calibration was optimal for SVM (HL p > 0.05; Brier 0.126) and random forest (HL p > 0.05; Brier 0.131). SHAP analysis identified D-dimer, ACEI/ARB use, HbA1c, C-reactive protein, and prothrombin time as top predictors. ML-based models accurately predict early CCS in ischemic stroke patients. XGBoost offers superior discrimination, while SVM and random forest demonstrate better calibration. Incorporation of these models into clinical workflows may enhance risk stratification and guide targeted preventive strategies.