Abstract 17308: Unsupervised K-means Clustering to Identify Signatures of Right Ventricle Dysfunction in Patients With Pulmonary Hypertension

Abstract

Introduction: Right ventricular (RV) function is a determinant of mortality in patients with pulmonary hypertension but current assessments are based on individual variables. Unsupervised k-means clustering can be used as an unbiased approach to determine distinct patient groups based on RV dysfunction. The aim of the study was to use unsupervised clustering to identify clinically meaningful RV phenotypes. Methods: Participants were identified from the University of Arizona PH registry (n=213). Patients with missing hemodynamic RV function data were excluded (n=23). Unsupervised clustering (k-means) was applied to variables of systolic and diastolic ventricular function including: max dP/dt/RVSP, Ees, Ea, Ees/Ea, max dP/dt, min dP/dt, RV EDP, β and Eed. Linear discriminant analysis (LDA) was performed to determine the weight of variables in cluster assignment. Results: Participants without PH (n=41) and World Symposium Pulmonary Hypertension (WSPH) Groups 1 (n=104), 2 (n=22), 3 (n=28), 4 (n=12), 5 (n=2) were assigned to four clusters based on the elbow method (Figure 1A). Two major determinants of cluster assignment were arterial elastance (Ea) and end systolic elastance (Ees). Participants with WSPH Group 1 were distributed across clusters (Figure 1B). Degree of RV dysfunction was highest in cluster 4 with increased Ees and Ea compared to other clusters (Figure 1C). Eed, Ees/Ea and mPAP (61 ± 22 mmHg) were increased compared to clusters 1 and 3. Sex, Ethnicity, BMI and cardiac output were not significant between clusters. Cluster 4 had increased PVR (11.1 ± 5.5 WU), and decreased PA compliance (1.2 ± 0.6 mmHg/ml) compared to other clusters. Conclusions: Unsupervised k-means clustering based on hemodynamic measures of RV function identified four distinct RV dysfunction phenotypes ranging from mild to severe.