An explainable machine learning model for predicting the outcome of ischemic stroke after mechanical thrombectomy

Area of Science:

• Neurology
• Medical Informatics
• Biostatistics

Background:

• Significant variability exists in clinical outcomes for patients with acute ischemic stroke (AIS) following mechanical thrombectomy (MT).
• Predicting functional independence after MT is crucial for patient management and treatment planning.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting functional independence in AIS patients treated with MT.
• To identify key clinical and laboratory features that predict outcomes after mechanical thrombectomy.

Main Methods:

• Analysis of 217 consecutive patients with anterior circulation large vessel occlusion undergoing MT.
• Training and testing of 7 ensemble ML models on derivation and temporal validation cohorts.
• Utilized the SHapley Additive exPlanations (SHAP) framework for model interpretability.

Main Results:

• A 9-item predictive score (PFCML-MT) was developed, incorporating age, NIH Stroke Scale, collateral status, and specific postoperative laboratory indices.
• The model demonstrated strong predictive performance with an area under the curve of 0.87 (test set) and 0.84 (temporal validation cohort).
• An online calculator was created for public access to the prediction model.

Conclusions:

• A machine learning model utilizing readily available features can accurately predict outcomes for AIS patients undergoing MT.
• The developed model shows potential for real-time clinical application to guide treatment decisions.
• The PFCML-MT score offers a valuable tool for prognostication in mechanical thrombectomy for acute ischemic stroke.