Development and validation of an interpretable machine learning model to predict malignant cerebral edema after endovascular treatment in acute anterior circulation large vessel occlusion stroke

Background: Malignant cerebral edema (MCE) is a life-threatening complication following endovascular treatment (EVT) in patients with acute anterior circulation large vessel occlusion (LVO) stroke. This study aimed to develop and validate a machine learning (ML)-based predictive model for early risk assessment of MCE in this population. Methods: We retrospectively collected data of 364 acute ischemic stroke patients with acute anterior circulation large vessel occlusion from a comprehensive stroke center in Shanghai, between August 2018 and December 2024. Eighty percent of patients were randomly assigned to the training set, and the remaining 20 % to the internal validation set. Additional 162 patients from the One Pass Tirofiban In Management of Ischemic Stroke Thrombectomy In China (OPTIMISTIC) trial were included as an external validation set. Six machine learning models were developed, and the model with the highest area under the receiver operating characteristic curve (AUC) was selected as the optimal model. Its performance was evaluated in both internal and external validation sets. Decision curve analysis (DCA) and calibration curves were plotted to assess clinical utility. SHapley Additive exPlanations (SHAP) was employed to perform interpretative analysis of the model. Results: In this study, a total of 79 patients developed malignant cerebral edema (MCE), including 45 out of 291 (15.46%) patients in the training set, 13 out of 73 (17.81%) patients in the internal validation set, and 21 out of 162 (12.96%) patients in the external validation set. The random forest model performed best, achieving an AUC of 0.901 (95% CI: 0.858-0.943) in the training set, 0.849 (95% CI: 0.700-0.970) in the internal validation set, and 0.724 (95% CI: 0.606-0.841) in the external validation set. Conclusion: This study developed and externally validated an interpretable machine learning model to predict the risk of MCE in patients with acute anterior circulation LVO stroke following EVT.