Predicting outcomes following open abdominal aortic aneurysm repair using machine learning.

Area of Science:

• Cardiovascular Surgery
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Open surgical repair of abdominal aortic aneurysms (AAA) carries a significant risk of post-operative complications.
• Currently, there is a lack of validated tools for predicting surgical risk in patients undergoing open AAA repair.

Purpose of the Study:

• To develop and validate automated machine learning (ML) algorithms for predicting 30-day adverse outcomes following open AAA repair.
• To identify pre-operative predictors of major adverse cardiovascular events (MACE) in this patient cohort.

Main Methods:

• Utilized the National Surgical Quality Improvement Program vascular database (2011-2021) for patients undergoing elective, non-ruptured open AAA repair.
• Trained six ML models using 35 pre-operative variables, with logistic regression as a baseline.
• Evaluated model performance using 10-fold cross-validation, focusing on 30-day MACE (myocardial infarction, stroke, or death).

Main Results:

• The XGBoost ML model demonstrated superior predictive performance, achieving an Area Under the Receiver Operating Characteristic curve (AUROC) of 0.90 (95% CI: 0.89-0.91).
• Logistic regression achieved a significantly lower AUROC of 0.66 (95% CI: 0.64-0.68).
• The best ML model showed good calibration, with a Brier score of 0.03, indicating strong agreement between predicted and observed event probabilities.

Conclusions:

• Automated ML algorithms can effectively predict 30-day MACE after open AAA repair.
• These ML tools offer a promising approach to guide pre-operative risk stratification and inform clinical decision-making.
• Implementation of such algorithms may facilitate targeted risk-mitigation strategies to improve patient outcomes.