A Machine Learning Tool to Predict Survival After First Surgery in Peripheral Artery Disease Patients

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Vascular Surgery Outcomes

Background:

• Peripheral artery disease (PAD) poses significant mortality risks, especially after surgical intervention.
• Accurate prediction of long-term survival is crucial for optimizing patient management and treatment strategies in PAD.
• Existing prognostic tools may not fully leverage the potential of machine learning for PAD survival prediction.

Purpose of the Study:

• To develop and validate a machine learning tool for predicting survival in patients with PAD undergoing surgical treatment.
• To identify key predictors of mortality in PAD patients post-surgery.
• To assess the performance of the developed predictive model.

Main Methods:

• Utilized data from 1,615 patients who underwent PAD surgery between 2005 and 2020.
• Employed Gradient Boosted Decision Trees (GBDTs) for mortality prediction at one, three, and five years post-surgery.
• Assessed predictor importance using SHAP values.

Main Results:

• Achieved Area Under the Curve (AUC) values of 0.86 (1-year), 0.84 (3-year), and 0.80 (5-year) for the prediction models.
• Identified disease stage, age, chronic kidney disease, hospital length-of-stay, and comorbidities as primary predictors.
• Dyslipidemia showed a slight predictive value for one- and three-year mortality.

Conclusions:

• Simple clinical and demographic parameters are sufficient to train effective GBDT models for PAD survival prediction.
• The developed machine learning tool demonstrates strong predictive performance for PAD patient mortality.
• This tool can aid clinicians in forecasting long-term outcomes for PAD patients following surgical treatment.