Background: Using machine learning for the phenotyping of patients with heart failure with preserved ejection fraction (HFpEF) has emerged as a novel approach to understanding the pathophysiology and stratifying the patients. Our objective is to perform phenotyping of patients with HFpEF in stable phase and to investigate the phenotypic trajectory from acute worsening phase to stable phase.

Methods: The present study is a post hoc analysis of the PURSUIT-HFpEF (Prospective Multicenter Observational Study of Patients with Heart Failure with Preserved Ejection Fraction) study. We applied the latent class analysis to the discharge data of patients hospitalized for acute decompensated heart failure.

Results: We finally included patient data of 1100 cases and 63 features in the latent class analysis. All patients were subclassified into 5 phenogroups as follows: Phenotype 1, characterized by better renal function and lower NT-proBNP (N-terminal pro-B-type natriuretic peptide) level [N=325 (29.5%)]; Phenotype 2, higher blood pressure, sinus rhythm, and poor renal function. [N=242 (22.0%)]; Phenotype 3, higher prevalence of atrial fibrillation, higher tricuspid pressure gradient, and lower tricuspid annular plane systolic excursion [N=214 (19.5%)]; Phenotype 4, higher C-reactive protein level and higher tricuspid pressure gradient [N=245 (22.3%)]; and Phenotype 5, poor nutritional status, poor renal function, and higher NT-proBNP level [N=74 (6.7%)]. A particular phenotype observed at the time of discharge was correlated with a distinct phenotype of acute worsening.

Conclusions: We identified 5 distinct stable phase phenotypes of the patients with HFpEF from the data at discharge. A specific phenotype at discharge was associated with a particular phenotype of acute worsening. This grouping can be a basis for future precision medicine of patients with HFpEF. Background: URL: https://www.umin.ac.jp/ctr/; Unique identifier: UMIN000021831.