Machine Learning Confirms Nonlinear Relationship between Severity of Peripheral Arterial Disease, Functional Limitation and Symptom Severity

Zulfiqar Qutrio Baloch; Syed Ali Raza; Rahul Pathak; Luke Marone; Abbas Ali

doi:10.3390/diagnostics10080515

Machine Learning Confirms Nonlinear Relationship between Severity of Peripheral Arterial Disease, Functional Limitation and Symptom Severity

Diagnostics (Basel). 2020 Jul 24;10(8):515. doi: 10.3390/diagnostics10080515.

Authors

Zulfiqar Qutrio Baloch¹, Syed Ali Raza², Rahul Pathak³, Luke Marone⁴, Abbas Ali⁴

Affiliations

¹ Department of Cardiology, Michigan State University/Sparrow Hospital, 1215 E Michigan Ave, Lansing, MI 48912, USA.
² Department of Neurology, Emory University, Atlanta, GA 30322, USA.
³ Heart of Florida Regional Medical Center, UCF College of Medicine, Orlando, FL 32827, USA.
⁴ Heart and Vascular Institute, West Virginia University, Morgantown, WV 26506, USA.

Abstract

Background: Peripheral arterial disease (PAD) involves arterial blockages in the body, except those serving the heart and brain. We explore the relationship of functional limitation and PAD symptoms obtained from a quality-of-life questionnaire about the severity of the disease. We used a supervised artificial intelligence-based method of data analyses known as machine learning (ML) to demonstrate a nonlinear relationship between symptoms and functional limitation amongst patients with and without PAD.

Objectives: This paper will demonstrate the use of machine learning to explore the relationship between functional limitation and symptom severity to PAD severity.

Methods: We performed supervised machine learning and graphical analysis, analyzing 703 patients from an administrative database with data comprising the toe-brachial index (TBI), baseline demographics and symptom score(s) derived from a modified vascular quality-of-life questionnaire, calf circumference in centimeters and a six-minute walk (distance in meters).

Results: Graphical analysis upon categorizing patients into critical limb ischemia (CLI), severe PAD, moderate PAD and no PAD demonstrated a decrease in walking distance as symptoms worsened and the relationship appeared nonlinear. A supervised ML ensemble (random forest, neural network, generalized linear model) found symptom score, calf circumference (cm), age in years, and six-minute walk (distance in meters) to be important variables to predict PAD. Graphical analysis of a six-minute walk distance against each of the other variables categorized by PAD status showed nonlinear relationships. For low symptom scores, a six-minute walk test (6MWT) demonstrated high specificity for PAD.

Conclusions: PAD patients with the greatest functional limitation may sometimes be asymptomatic. Patients without PAD show no relationship between functional limitation and symptoms. Machine learning allows exploration of nonlinear relationships. A simple linear model alone would have overlooked or considered such a nonlinear relationship unimportant.

Keywords: 6MWT (six-minute walk test); CLI (critical limb ischemia); Machine Learning; PAD (peripheral arterial disease); TBI (toe–brachial index).