Predicting the Severity of Infrapopliteal Artery Lesions in Patients With Peripheral Artery Disease Using Interpretable Machine Learning

Abstract

Purpose:

Infrapopliteal arterial disease represents a complex subtype of peripheral artery disease (PAD). This study aimed to develop an interpretable machine learning model to assess the severity of infrapopliteal artery lesions.

Methods:

Clinical data from patients with PAD treated at our institution were obtained between July 2019 and October 2024. Based on the angiographic results, patients were categorized into a mild lesion group (n=584) and a severe lesion group (n=478). Data from 2019 to 2023 were used for model development, with 70% allocated to the training set and 30% to the validation set. Clinical data from patients in 2024 served as the external test set. Feature selection was performed using 3 distinct machine learning algorithms. Subsequently, 10 different predictive models were developed and compared. The optimal model was interpreted and deployed. Finally, we conducted subgroup analyses.

Results:

A total of 1062 patients were included in the study. Six predictors were identified through feature selection and used for model construction. Among the 10 models, the Gradient Boosting Machine (GBM) demonstrated the best predictive performance, achieving area under the curve (AUC) values of 0.891 in the validation set, indicating high discriminative ability. The calibration curve showed good agreement with the ideal line. Decision curve analysis demonstrated that the model provided superior net benefit within a threshold probability range of approximately 15% to 90%. Model interpretation was performed using Shapley additive explanations. Additionally, a nomogram was developed, and the model was deployed as an interactive web-based tool. The GBM model maintained robust performance across all subgroups.

Conclusions:

The GBM model, developed using 6 clinically relevant variables, enables accurate prediction of infrapopliteal artery disease severity and demonstrates its significant potential to support clinical decision-making and improve risk stratification for patients with PAD.

Clinical Impact

This study introduces a Gradient Boosting Machine model to predict the severity of infrapopliteal arterial disease using six readily available variables. For clinicians, this offers a non-invasive, rapid decision-support tool. By accurately predicting lesion severity, the model enables surgeons to better stratify patients, optimize treatment planning, and potentially improve the suboptimal outcomes currently associated with infrapopliteal interventions. This innovation shifts practice towards a more personalized, data-driven approach in the initial assessment of peripheral artery disease.