Hemodynamic‐Driven Staging of Heart Failure With Preserved Ejection Fraction Using Unsupervised Cluster Analysis

Invasive exercise hemodynamics are used to diagnose heart failure with preserved ejection fraction (HFpEF), based on pulmonary artery wedge pressure (PAWP) or left atrial (LA) pressure elevations. We hypothesized that applying unsupervised cluster analysis to comprehensive hemodynamic characterization might provide data‐driven phenotypes, with pathophysiological and prognostic implications. Eighty consecutive HFpEF patients underwent right heart catheterization at rest, during passive leg raise, and at peak exercise. We performed unsupervised k ‐means clustering analysis, using eight hemodynamic variables that were not strongly correlated (Pearson correlation coefficient < 0.80). Hemodynamics and clinical characteristics, as well as event‐free survival, were assessed. k = 5 clusters were identified. Hemodynamic severity increased from Cluster 1 to Clusters 4–5 ( p < 0.01 for most of the hemodynamic variables), mirrored by different event‐free survival (log‐rank test p < 0.001). Clusters 1 and 2 presented with either steep PAWP rise or LA hypertension and pulmonary hypertension (PH) only during exercise. Cluster 3 presented with LA hypertension and PH already at rest, as well as with tall PAWP V waves during exercise. Cluster 4 presented with post‐ and precapillary PH, tall PAWP V waves, right atrial (RA) hypertension, dynamic tricuspid regurgitation (TR), and low cardiac output (CO) reserve. Cluster 5 presented with TR and RA hypertension, low CO, and a lack of decrease in PVR. Data‐driven unsupervised cluster analysis of advanced invasive hemodynamics allowed for the identification of distinct HFpEF phenotypes across the spectrum of disease severity. We found a progressive involvement of the pulmonary circulation and of the right heart, coupled with a worse prognosis.