Machine learning and decision making in aortic arch repair

Area of Science:

• Cardiovascular Surgery
• Medical Artificial Intelligence
• Health Informatics

Background:

• Aortic arch surgery requires careful decisions on cannulation and temperature to minimize risks.
• Individualized, data-driven strategies are needed to optimize patient outcomes.
• Machine learning (ML) offers a promising approach to model surgical risks.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting mortality and stroke risk in elective aortic arch surgery.
• To identify key patient characteristics and intraoperative decisions influencing adverse events.
• To compare the predictive performance of ML models against traditional logistic regression.

Main Methods:

• A cohort of 1323 patients undergoing elective aortic arch procedures was analyzed.
• Logistic regression and XGBoost ML models were trained using 69 variables.
• Shapely additive explanations (SHAP) were used to assess the importance of intraoperative decisions.

Main Results:

• XGBoost models showed superior discrimination for death (AUC 0.77) and stroke (AUC 0.87) compared to logistic regression.
• Intraoperative decisions were among the top predictors for both mortality and stroke.
• SHAP analysis revealed patient-specific predictor weights, highlighting the personalized nature of risk.

Conclusions:

• Machine learning accurately identifies patients at high risk of death and stroke after aortic arch surgery.
• ML models enable tailored operative decisions for personalized risk reduction.
• This data-driven approach surpasses traditional prediction models in optimizing patient care.