Artificial Intelligence Identifies an Urgent Need for Peripheral Vascular Intervention by Multiplexing Standard Clinical Parameters

Abstract

Background: Peripheral artery disease (PAD) is a significant burden, particularly among patients with severe disease requiring invasive treatment. We applied a general Machine Learning (ML) workflow and investigated if a multi-dimensional marker set of standard clinical parameters can identify patients in need of vascular intervention without specialized intra–hospital diagnostics. Methods: This is a retrospective study involving patients with stable PAD (sPAD, Fontaine Class I and II, n = 38) and unstable PAD (unPAD, Fontaine Class III and IV, n = 18) in need of invasive therapeutic measures. ML algorithms such as Random Forest were utilized to evaluate a matrix consisting of multiple routinely clinically available parameters (age, complete blood count, inflammation, lipid, iron metabolism). Results: ML has enabled a generation of an Artificial Intelligence (AI) PAD score (AI-PAD) that successfully divided sPAD from unPAD patients (high AI-PAD in sPAD, low AI-PAD in unPAD, cutoff at 50 AI-PAD units). Furthermore, the probability score positively coincided with gold-standard intra-hospital mean ankle-brachial index (ABI). Conclusion: AI-based tools may be promising to enable the correct identification of patients with unstable PAD by using existing clinical information, thus supplementing clinical decision making. Additional studies in larger prospective cohorts are necessary to determine the usefulness of this approach in comparison to standard diagnostic measures.

Keywords

• artificial intelligence
• machine learning
• classification model
• Random Forest
• peripheral artery disease
• vascular disease