Related Experiment Video
Updated: Sep 18, 2025
A Thrombotic Stroke Model Based On Transient Cerebral Hypoxia-ischemia
Published on: August 18, 2015
A two-stage machine learning-based risk assessment model for intravenous thrombolysis in acute ischemic stroke (AIS):
Shudan Zhu1, Qing Ye2, Hong Wu3
1School of Medicine and Health Management, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, China; Union Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, China.
Objective:
Develop a two-stage, machine learning-based thrombolysis risk stratification model from existing medical datasets and electronic health records to predict the risk of early hemorrhagic transformation(HT) and in-hospital mortality(IM) following thrombolysis in patients with acute ischemic stroke (AIS), as well as to monitor changes in risk post-thrombolysis, thereby facilitating clinical decision-making and enhancing the recovery rate of AIS.
Materials And Methods:
Patients with AIS admitted to a Grade III Class A hospital between October 2001 and October 2022 were included. We extracted 48 clinical features from the datasets and categorized model features into "pre-thrombolysis" and "post-thrombolysis" stages based on the thrombolysis timing point. Utilizing 5 distinct machine learning algorithms, we separately conducted model training to predict the risk of HT and IM both before and after thrombolysis. By establishing a combined model, we further explored the correlation between pre- and post-thrombolysis risks, with external validation performed on 1,777 patients from geographically distinct hospitals. Model performance was assessed according to a suite of learning metrics, including AUC, accuracy, precision, recall, and F1 score.
Results:
In the first stage model, the Random Forest model demonstrated the highest performance in predicting HT outcomes (AUC = 0.73), with an external validation cohort AUC of 0.69. For predicting IM outcomes, the XGBoost model performed the best (AUC = 0.78), albeit with an external validation AUC of 0.81. Moving to the second stage, XGBoost excelled in predicting HT outcomes, achieving an AUC of 0.83 in the internal validation cohort and 0.75 in the external validation queue. Regarding IM outcomes in the second stage, XGBoost again proved optimal, yielding an AUC of 0.91 in internal validation and 0.88 in the external validation set.
Conclusions:
The two-stage machine learning-based model for predicting thrombolysis treatment risks in AIS patients is feasible and effective.
