Paper detail

Classification of syncope through data analytics

Objective: Syncope is a sudden loss of consciousness with loss of postural tone and spontaneous recovery; it is a common condition, albeit one that is challenging to accurately diagnose. Uncertainties about the triggering mechanisms and their underlying pathophysiology have led to various classifications of patients exhibiting this symptom. This study presents a new way to classify syncope types using machine learning. Method: we hypothesize that syncope types can be characterized by analyzing blood pressure and heart rate time series data obtained from the head-up tilt test procedure. By optimizing classification rates, we identify a small number of determining markers which enable data clustering. Results: We apply the proposed method to clinical data from 157 subjects; each subject was identified by an expert as being either healthy or suffering from one of three conditions: cardioinhibitory syncope, vasodepressor syncope and postural orthostatic tachycardia. Clustering confirms the three disease groups and identifies two distinct subgroups within the healthy controls. Conclusion: The proposed method provides evidence to question current syncope classifications; it also offers means to refine them. Significance: Current syncope classifications are not based on pathophysiology and have not led to significant improvements in patient care. It is expected that a more faithful classification will facilitate our understanding of the autonomic system for healthy subjects, which is essential in analyzing pathophysiology of the disease groups.