LAUSR

Background To develop and validate a machine learning model for predicting subsequent vascular events (SVE) 6 months after mild ischemic stroke (MIS) in Chinese patients. Methods A retrospective analysis was performed on 495 newly diagnosed MIS patients by collecting their basic information, past medical history, initial NIHSS score, symptoms, obstruction sites of MIS, and MRI results. According to the ratio of 7:3, the dataset was divided into a training set (n=346) and a testing set (n=149) through stratified random sampling. In the training set, the recursive feature elimination (RFE) was used to select the optimal combination of features, and two machine learning algorithms, including the logistic regression (LR) and support vector machines (SVM), were used to build the prediction model, which was further validated by using 5-fold cross-validation. The receiver operating characteristic (ROC) curve was used on the testing set to evaluate the model’s performance, and the area under the curve (AUC), sensitivity, specificity, and accuracy were calculated. The calibration curve and decision curve of the two models were further compared. Results SVE occurred in 56 cases (11.3%) of 495 patients with MIS during the 6-month follow-up. Finally, the best 15 predictive features were selected, and the top three predictive features were diabetes, posterior cerebral artery lesion, and fasting blood glucose in order. In the testing set, the AUC of the LR model was 0.929 (95% CI: 0.875–0.964), and its accuracy, sensitivity, and specificity were 0.832, 0.765, and 0.841, respectively. The AUC of the SVM model was 0.992 (95% CI: 0.962–1.000), and its accuracy, sensitivity, and specificity were 0.966, 0.824, and 0.985, respectively. The SVM model’s discrimination, calibration, and clinical validity are better than those of the LR model. Conclusion The predictive models developed using machine learning methods can predict the risk of SVE after 6 months following MIS in Chinese patients.