An Explainable Two-Stage Machine Learning Model for Predicting the Post-Thrombolysis Complications in Stroke...

Area of Science:

• * Neurology and Artificial Intelligence
• * Clinical Decision Support Systems
• * Stroke Management and Thrombolysis Therapy

Background:

• * Current stroke thrombolysis risk prediction tools show limited accuracy for early hemorrhagic events.
• * There is a significant unmet need for improved stroke management strategies and risk stratification.
• * Existing methods struggle to accurately predict post-thrombolysis complications.

Purpose of the Study:

• * To develop and validate an explainable 2-stage machine learning model for stroke risk stratification.
• * To predict the risk of bleeding, composite complications, and all-cause death in patients before and after thrombolysis.
• * To enhance the accuracy of predicting adverse events in stroke patients receiving thrombolysis.

Main Methods:

• * Development of a 2-stage machine learning model integrating LightGBM, XGBoost, random forest, decision tree, and logistic regression.
• * Training the model on data from 5,333 stroke patients at Tongji Hospital.
• * External validation performed on data from 526 patients across two additional hospitals.

Main Results:

• * The model demonstrated improved predictive accuracy in the post-thrombolysis stage compared to the pre-thrombolysis stage.
• * Achieved higher Area Under the Curve (AUC) values for bleeding, composite complications, and death prediction post-thrombolysis.
• * Identified key predictors including temperature, vital signs, and demographic factors, with a validated prototype.

Conclusions:

• * The developed machine learning model significantly enhances thrombolysis risk prediction accuracy for stroke patients.
• * The model supports personalized patient care management and offers potential for clinical decision support integration.
• * This approach represents a substantial advancement in stroke management and patient safety.