StrokeENDPredictor-19: Setting New Prediction Model in Neurological Prognosis in Acute Ischemic Stroke.

Abstract

Early Neurological Deterioration (END) following intravenous thrombolysis (IVT) highlights potential risks in current management strategies for acute ischemic stroke. Early identification of at-risk patients could enhance treatment efficacy. This study aims to develop an advanced AI predictive model that improves accuracy in forecasting END while ensuring interpretability for clinical application. This prospective cohort study included 970 patients with acute ischemic stroke who underwent IVT. Data from 365 patients were used for model development and internal validation, while data from 605 patients were utilized for external validation. Five machine learning models were developed and compared using evaluation metrics such as accuracy and AUC. Feature selection and model optimization were performed using the XGBoost algorithm and SHapley Additive exPlanations (SHAP) method, resulting in the StrokeENDPredictor-19 model. Among the five models, XGBoost demonstrated superior performance with an internal validation accuracy of 91% (AUC = 0.96) and external validation accuracy of 90% (AUC = 0.95). Notably, this study established cutoff values for critical clinical features, providing quantifiable reference standards for practical applications. The StrokeENDPredictor-19 model offers neurologists a valuable tool for forecasting the likelihood of END in patients receiving IVT therapy, thereby supporting more precise clinical decision-making.