Predicting intravenous thrombolysis outcomes in acute ischemic stroke using machine learning

Jiang Mm, Qiong Yue, 洲光 王, Qian Cui

Objective Intravenous thrombolysis remains a cornerstone intervention for improving clinical outcomes in patients with acute ischemic stroke (AIS). Accurate prediction of poor functional outcomes following thrombolysis is essential for optimizing individualized treatment and guiding clinical decision-making. This study aimed to develop machine learning-based models for early post-treatment reassessment of post-thrombolysis outcomes in AIS patients, thereby providing a reliable tool for early prognostic reassessment after thrombolysis. Methods A total of 383 AIS patients who received intravenous thrombolysis between November 2024 and November 2025 were retrospectively enrolled and randomly assigned to a training set ( n = 268) and a validation set ( n = 115) in a 7:3 ratio. Univariate analysis was initially conducted to identify indicators associated with thrombolysis outcomes ( p < 0.05). Least Absolute Shrinkage and Selection Operator (LASSO) regression was subsequently applied for feature selection. Based on the selected variables, three machine learning models—Random Forest (RF), Gradient Boosting Machine (GBM), and Support Vector Machine (SVM)—were developed, with logistic regression established as a benchmark. Model performance was assessed using the area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis. Model interpretability was further evaluated using SHapley Additive exPlanations (SHAP) values. Results Baseline characteristics were well-balanced between groups (all p > 0.05). Multivariate logistic regression identified a higher admission National Institutes of Health Stroke Scale (NIHSS) score, elevated admission blood glucose, a higher 24-h post-thrombolysis NIHSS score, increased infarct core volume, higher glycated hemoglobin and an elevated Neutrophil-to-Lymphocyte Ratio as significant risk factors for poor outcomes, whereas a larger ischemic penumbra volume emerged as a protective factor ( p < 0.05). The RF model demonstrated superior predictive performance, achieving an AUC of 0.784 in the training set, along with improved calibration and greater clinical net benefit within the 0.0–0.60 risk threshold range compared to the GBM and SVM. SHAP analysis showed the importance ranking of core predictors. Conclusion The proposed prediction model demonstrates satisfactory performance in estimating functional outcomes following thrombolysis in AIS patients. The identified core variables may offer valuable insights for clinical prognosis and the design of individualized thrombolytic strategies.