Predicting Outcomes Following Endovascular Abdominal Aortic Aneurysm Repair Using Machine Learning

Area of Science:

• Vascular Surgery
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Endovascular aneurysm repair (EVAR) for abdominal aortic aneurysms (AAA) has risks, but lacks established outcome prediction tools.
• Accurate prediction of perioperative risks is crucial for patient management.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting 30-day outcomes after EVAR.
• To compare ML model performance against traditional logistic regression.

Main Methods:

• Utilized the National Surgical Quality Improvement Program database (2011-2021) for patients undergoing EVAR for infrarenal AAA.
• Trained 6 ML models using 36 preoperative variables and evaluated using area under the receiver operating characteristic curve (AUC).
• Assessed model robustness with calibration plots and Brier scores, including subgroup analyses.

Main Results:

• The XGBoost ML model achieved the highest AUC of 0.95 (0.94-0.96), significantly outperforming logistic regression (0.72 [0.70-0.74]).
• The best model demonstrated good calibration (Brier score of 0.06) and robust performance across diverse patient subgroups.
• A 2.4% rate of 30-day major adverse cardiovascular events was observed in 16,282 patients.

Conclusions:

• Developed accurate ML models for predicting 30-day post-EVAR outcomes using preoperative data.
• These models offer superior performance compared to logistic regression for EVAR risk prediction.
• Automated algorithms can inform risk mitigation strategies for patients considered for EVAR.