Explainable machine learning model for predicting functional outcomes in posterior circulation stroke after...

Area of Science:

• Neurology
• Medical Informatics
• Artificial Intelligence

Background:

• Predicting functional outcomes in posterior circulation stroke (PCS) is vital for timely interventions.
• Machine learning (ML) models can aid in predicting outcomes for patients undergoing endovascular thrombectomy (EVT).
• This study focuses on developing and validating ML models for 3-month functional outcome prediction in PCS patients post-EVT.

Purpose of the Study:

• To develop and validate machine learning models for predicting functional outcomes in PCS patients after EVT.
• To identify key predictive features for functional outcomes.
• To provide an accessible tool for clinical decision support.

Main Methods:

• A derivation cohort of 202 PCS patients undergoing EVT was used for training and internal validation.
• An external dataset of 54 patients was used for external validation.
• Seven ML models were trained on preoperative features, with the best performing model further trained using intraoperative and postoperative data. Model interpretability was assessed using SHAP.

Main Results:

• The Random Forest model showed the best performance.
• The preoperative model achieved an AUC of 0.83 (test set) and 0.81 (external validation).
• Incorporating intraoperative and postoperative features improved AUCs to 0.84/0.90 (test) and 0.83/0.90 (external validation). A web calculator is available.

Conclusions:

• Interpretable ML models accurately predict functional outcomes in PCS patients post-EVT.
• These models offer valuable insights for personalized risk stratification and perioperative management.
• The models have potential for integration into clinical workflows to optimize patient care.