Machine learning models for predicting futile recanalisation after endovascular treatment in patients with large core infarction.

BACKGROUND: Predicting futile recanalisation following endovascular treatment (EVT) in patients with large core infarctions is crucial for guiding clinical decisions, optimising perioperative management and improving healthcare resource allocation. This study aimed to compare four machine learning (ML) algorithms and identify the most effective model for preinterventional prediction of futile recanalisation. METHODS: Patients achieving successful reperfusion (expanded Thrombolysis in Cerebral Infarction Score≥2b) from the EVT in Acute Anterior Circulation Large Vessel Occlusive Patients With a Large Infarct Core trial were stratified into two groups: no-futile recanalisation (90-day modified Rankin Scale (mRS) 0-3) and futile recanalisation (mRS 4-6). The least absolute shrinkage and selection operator regression method was used for feature selection, and four ML algorithms, including logistic regression, support vector machine (SVM), decision tree and random forest, were applied. Model performance was evaluated using receiver operating characteristic curves, calibration plots and decision curve analysis. Feature importance was ranked using SHapley Additive exPlanation (SHAP) values. RESULTS: Among 146 patients, 74 experienced futile recanalisation. Eight predictors were identified and ranked by SHAP analysis from highest to lowest importance: sex, age, National Institutes of Health Stroke Scale, glucose, systolic blood pressure, neutrophil-to-lymphocyte ratio, fibrinogen and occlusion site. Among the four models, the SVM model achieved the highest area under the curve of 0.891 (95% CI 0.837 to 0.945), along with good calibration (Hosmer-Lemeshow test, p=0.103) and clinical utility. CONCLUSION: The SVM model emerges as the optimal predictive tool for futile recanalisation following EVT in patients with large core infarction. Nevertheless, external validation is required to confirm its performance before clinical application. TRIAL REGISTRATION NUMBER: NCT04551664.