Using machine learning to predict outcomes following suprainguinal bypass.

Area of Science:

• Vascular Surgery
• Machine Learning Applications
• Outcome Prediction

Background:

• Suprainguinal bypass for peripheral artery disease (PAD) has significant surgical risks.
• Existing tools for predicting outcomes after this procedure are limited.
• There is a need for improved risk stratification and management strategies.

Purpose of the Study:

• To develop and evaluate machine learning (ML) algorithms for predicting outcomes following suprainguinal bypass.
• To compare the performance of ML models against traditional logistic regression.
• To identify key predictors of adverse events after suprainguinal bypass.

Main Methods:

• Utilized the Vascular Quality Initiative database (2003-2023) for patient data.
• Developed six ML models (including XGBoost, random forest, logistic regression) using preoperative, intraoperative, and postoperative variables.
• Primary outcome: major adverse limb events (MALE) or death at 1 year; evaluated using Area Under the Receiver Operating Characteristic Curve (AUROC).

Main Results:

• The XGBoost model demonstrated superior predictive performance with an AUROC of 0.92 (preoperative), significantly outperforming logistic regression (AUROC 0.67).
• XGBoost performance remained high across all stages: 0.93 (intraoperative) and 0.98 (postoperative).
• Key predictors included chronic limb-threatening ischemia, prior procedures, comorbidities, and functional status.

Conclusions:

• Developed accurate ML models for predicting 1-year MALE or death after suprainguinal bypass.
• These ML algorithms offer improved performance over logistic regression.
• The models show potential utility in guiding perioperative risk mitigation for better patient outcomes.