Development and Validation of an Interpretable Machine Learning Model for Prediction of the Risk of Clinically Ineffective Reperfusion in Patients Following Thrombectomy for Ischemic Stroke

Xiaolong Hu,1,2,* Dayong Qi,2,* Suya Li,2 Shifei Ye,2 Yue Chen,2 Wei Cao,2 Meng Du,2 Tianheng Zheng,2 Peng Li,2 Yibin Fang1– 3 1Tongji University School of Medicine, Tongji University Affiliated Shanghai 4th People’s Hospital, Shanghai, People’s Republic of China; 2Department of Neurovascular Disease, Tongji University Affiliated Shanghai 4th People’s Hospital, Shanghai, People’s Republic of China; 3Translational Research Institute of Brain and Brain-Like Intelligence, Shanghai Fourth People’s Hospital, School of Medicine, Tongji University, Shanghai, People’s Republic of China*These authors contributed equally to this workCorrespondence: Yibin Fang, Tongji University Affiliated Shanghai 4th People’s Hospital, Sanmen Road 1279, Hongkou District, Shanghai, People’s Republic of China, Email [email protected]: Despite successful recanalization after thrombectomy in patients with acute ischemic stroke, poor prognosis often persists. This study aimed to investigate the factors contributing to clinically ineffective reperfusion (CIR), develop and validate a machine-learning model to predict CIR, and provide guidance for future clinical treatments.Methods: We collected data from patients undergoing thrombectomy at Shanghai Fourth People’s Hospital between December 2021 and June 2024. The clinical variables were compared between the clinically ineffective and effective recanalization groups using univariate analysis. Four machine learning models were developed: random forest (RF), support vector machine (SVM), decision tree (DT), and k-nearest neighbor (KNN). Model performance was evaluated using receiver operating characteristic (ROC) curves and heatmap visualization. The SHAP method rank the feature importance and provided interpretability for the final model.Results: Among the four machine learning models, the RF model showed the best performance, with an area under the curve (AUC) of 0.96 (95% CI: 0.91– 1.0), accuracy of 0.93, and specificity of 0.97 on the test dataset. The SHAP algorithm identified the number of endovascular thrombectomy (EVT) attempts as the key factor influencing CIR. Based on the RF model, a web-based calculator for CIR prediction is available at https://ineffectivereperfusion.shinyapps.io/calculate/. The final model included ten parameters: EVT attempts, diabetes mellitus, previous ischemic stroke, National Institutes of Health Stroke Scale (NIHSS score), preoperative infarction in the basal ganglia, baseline diastolic blood pressure, clot burden score (CBS)/basilar artery on computed tomography angiography (BATMAN) score, stroke cause, collateral grade, and MLS.Conclusion: We developed and validated the first interpretable machine learning model for CIR prediction after EVT, surpassing traditional methods. Our CIR risk prediction platform enables early intervention and personalized treatment. The number of EVT attempts has emerged as a key determinant, underscoring the need for optimized procedural timing to improve outcomes.Keywords: machine learning, clinically ineffective reperfusion, predictive model, acute ischemic stroke, online predictive platform