Background: There are little data on clinically meaningful heart failure (HF) phenogroups, which are associated with treatment response across the wide spectrum of left ventricular (LV) ejection fraction.
Objectives: The authors aimed to identify the phenotypes of patients with HF with different prognoses and responses to medical therapies.
Methods: We examined consecutive 2,301 chronic HF patients from the ELMSTAT-HF (EpidemioLogical Multicenter Study for Tailored Treatment in Heart Failure) registry, a prospective multicenter cohort in which 2,317 patients were enrolled between January 2020 and September 2024. Latent class analysis was performed using 99 clinical features. The primary outcome was a composite of all-cause death and hospitalization for worsening HF.
Results: The analysis subclassified the patients into 8 phenogroups: group 1, characterized by younger age with obesity; 2, less structural abnormality and comorbidity; 3, younger age with LV dilation; 4, LV hypertrophy; 5, older age with small LV and diastolic dysfunction; 6, ischemic cardiomyopathy; 7, advanced LV remodeling and ventricular arrhythmias; and 8, atrial myopathy. During a median follow-up of 597 (IQR: 302-932) days, the incidence of the primary outcome significantly differed between the phenogroups (P < 0.001). In phenogroup 5, patients taking beta-blockers or sodium-glucose cotransporter 2 inhibitors had a significantly higher rate of hospitalization for worsening HF (HR: 2.20; 95% CI: 1.04-4.68; HR: 4.27; 95% CI: 2.02-9.05, respectively).
Conclusions: We identified 8 phenogroups with distinct clinical outcomes in patients with HF. This phenotyping provides appropriate risk stratification and may aid clinical decision-making in patients with HF.
Keywords: heart failure; machine learning; medical therapy; phenotyping; prognosis.
Copyright © 2025 The Authors. Published by Elsevier Inc. All rights reserved.