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

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.