Using machine learning to predict outcomes following carotid endarterectomy.

Area of Science:

• Vascular Surgery
• Medical Informatics
• Machine Learning Applications

Background:

• Predicting outcomes after carotid endarterectomy (CEA) is challenging due to a lack of standardized tools.
• Effective perioperative management strategies are needed to mitigate risks associated with CEA.

Purpose of the Study:

• To develop and validate machine learning (ML) algorithms for predicting outcomes following CEA.
• To assess the performance of ML models compared to traditional methods and existing literature tools.

Main Methods:

• Utilized the Vascular Quality Initiative (VQI) database (2003-2022) with 166,369 CEA patients.
• Trained six ML models (including XGBoost) on preoperative, intraoperative, and postoperative features.
• Evaluated models using area under the receiver operating characteristic curve (AUROC), calibration plots, and Brier scores.

Main Results:

• The XGBoost model achieved an AUROC of 0.90 for preoperative prediction, significantly outperforming logistic regression (0.65) and existing tools (0.58-0.74).
• Models demonstrated sustained high performance across intraoperative (AUROC 0.90) and postoperative (AUROC 0.94) stages.
• Top predictors included preoperative comorbidities, functional status, and previous procedures, with robust performance across patient subgroups.

Conclusions:

• Developed accurate ML models for predicting CEA outcomes, surpassing current prediction tools.
• These algorithms offer significant potential for guiding perioperative risk mitigation strategies.
• The findings support the utility of ML in enhancing patient safety and optimizing care following CEA.