Machine learning-based models to predict 90-day non-excellent functional outcomes among patients with mild acute ischemic stroke after intravenous thrombolysis

Background Despite intravenous thrombolysis, a substantial proportion of patients with mild acute ischemic stroke fail to achieve an excellent functional outcome. Prediction models specifically developed for these patients remain limited. We aimed to construct and evaluate a machine learning model for identifying 90-day non-excellent functional outcomes. Methods We retrospectively collected data from patients with mild acute ischemic stroke treated with intravenous thrombolysis at Tianjin Fourth Central Hospital between January 2024 and June 2025. Demographic, clinical, laboratory, and neuroimaging variables were extracted from electronic medical records. LASSO regression was applied for feature selection, followed by the development of seven machine learning models. Model performance was assessed using discrimination, calibration, clinical utility, and classification metrics. We conducted SHAP analysis to elucidate the relative importance and direction of association of individual predictors with model predictions. Results A total of 507 patients with mild acute ischemic stroke undergoing intravenous thrombolysis were included, of whom 136 (26.8%) experienced a non-excellent functional outcome at 90 days. LASSO regression identified six predictors, including age, baseline National Institutes of Health Stroke Scale score, platelet count, monocyte count, blood glucose level, and neutrophil-to-lymphocyte ratio. In the validation cohort, XGBoost achieved the highest numerical AUC of 0.753 (95% CI, 0.660–0.840) and demonstrated adequate calibration and clinical usefulness. Conclusion A machine learning model was developed and internally validated to predict 90-day non-excellent functional outcomes in patients with mild acute ischemic stroke after intravenous thrombolysis. The XGBoost model showed promising predictive performance and potential utility for early risk stratification.