Comprehensive Summary
In this study, Li et al. (2025) developed StrokeENDPredictor-19, an artificial intelligence-based tool designed to forecast early neurological deterioration (END) following intravenous thrombolysis (IVT). The model analyzed 970 patients with acute ischemic stroke who underwent IVT, where 365 patients were used for the development of the model and for internal validation, while the other 605 patients were used for external validation. Five machine learning models were created and tested to compare evaluation metrics, such as accuracy. Of the five models, the XGBoost model showed the best performance with an internal accuracy of 91%, an internal AUC of 0.96, an external accuracy of 90%, and an external AUC of 0.95. Also, the researchers created cutoff values for 19 clinical features (baseline blood glucose, uric acid, neuroimaging findings, etc.), allowing for the model to be easily interpretable for clinicians. Overall, these results showed that the StrokeENDPredictor-19 outperformed existing risk scores for predicting END, offering better diagnostic and prognostic accuracy for clinicians.
Outcomes and Implications
This research highlights the growing role of AI in stroke care, providing a more reliable method to identify patients at risk of worsened outcomes after IVT. In clinical care, this model can help neurologists triage patients, have a stronger prognosis at earlier timepoints, and refine treatment strategies to provide better outcomes for their patients. By using the cutoff values the model provides, clinicians can easily interpret the model’s data to make treatment decisions. While additional validation is needed before being widely used in clinical formats, this study highlights how AI and ML models can soon be widely used to enhance stroke management and prognosis.