The following is a summary of “Predicting Major Adverse Cardiovascular Events Following Carotid Endarterectomy Using Machine Learning,” published in the October 2023 issue of Cardiology by Li et al.

Carotid endarterectomy (CEA) stands as a vital surgical intervention aimed at preventing strokes, yet it poses considerable perioperative risks with limited predictive tools available for outcomes. To address this gap, the researchers devised machine learning algorithms specifically tailored to forecast post-CEA results.

Patients undergoing CEA were identified using data from the National Surgical Quality Improvement Program’s vascular database spanning 2011 to 2021, and 36 preoperative demographic and clinical variables were considered as input features. The primary focus was on predicting 30‐day major adverse cardiovascular events, encompassing stroke, myocardial infarction, or mortality. Splitting the dataset into training (70%) and test (30%) subsets, the team employed 10-fold cross-validation to train six machine learning models based on preoperative characteristics. Model performance was predominantly assessed by the area under the receiver operating characteristic curve, with additional evaluation via calibration plots and Brier score to ensure model robustness. 

Throughout the study period, 38,853 patients underwent CEA, and 4.3% (1683 individuals) experienced major adverse cardiovascular events within 30 days. Notably, the XGBoost model emerged as the most effective, achieving an area under the receiver operating characteristic curve of 0.91 (95% CI, 0.90–0.92). In stark comparison, logistic regression yielded an area under the receiver operating characteristic curve of 0.62 (95% CI, 0.60–0.64), while existing literature tools showcased values between 0.58 and 0.74. The calibration plot depicted a strong alignment between predicted and observed event probabilities, underscored by a Brier score of 0.02. The carotid symptom status was a pivotal feature in their algorithm’s predictive strength.

In summary, the machine learning models developed in this study exhibited high accuracy in predicting 30‐day outcomes post-CEA, outperforming existing tools. These models hold promise for aiding clinicians in devising risk-mitigation strategies, ultimately enhancing the outcomes for patients undergoing evaluation for CEA.

Source: ahajournals.org/doi/10.1161/JAHA.123.030508