Patient Clustering in HFpEF: Guiding Future Trial Design and Drug Therapy

Computer illustration of a heart.
Cardiopulmonary adverse events with carfilozmib treatment included dyspnea, hypertension, peripheral edema, cough, pneumonia, and heart failure.
The researchers aimed to derive and validate clinically useful clusters of patients with heart failure with preserved ejection fraction of 50% or greater.

Clustering of patients with heart failure with preserved ejection fraction (HFpEF) confirmed the heterogeneity of HFpEF, providing researchers with future direction for tailoring trial design and individualized drug therapy. This is according to research published in the European Journal of Heart Failure.

Using data from the Swedish Heart Failure Registry (SwedeHF), researchers sought to both derive and validate clinically useful clusters of people with HFpEF, using advanced analytic techniques, that distinguish easily accessible clinical characteristics and outcomes to create clusters applicable across multiple healthcare settings.

The study included 6909 patients from SwedeHF with left ventricular ejection fraction of 50% or greater who were registered with the cohort between 2013 and 2016 as a derivation cohort. The external validation cohort included patients from the Chronic Heart Failure ESC-guideline-based Cardiology Practice Quality project registry (CHECK-HF), a cross-sectional registration of patients with a chronic HF diagnosis from 24 Dutch hospitals.

Median age in the SwedeHF cohort was 80 years (interquartile range [IQR], 72-86; 52% of participants were women. Comorbidities included hypertension, atrial fibrillation, and ischemic heart disease in 82%, 68%, and 48% of participants, respectively. Beta-blockers were the most commonly prescribed HF medication, followed by diuretics and renin-angiotensin system (RAS) inhibitors (83%, 81%, and 73%, respectively).

In the CHECK-HF cohort, patient median age was 77 years (IQR, 69-84 years); 55% were women, and both comorbidities and medication use were distributed similarly to the SwedeHF cohort. Implantable devices, including implantable cardioverter defibrillators and/or cardiac resynchronization therapy, mineralocorticoid receptor antagonist (MRA), digoxin, and statin use were more prevalent in CHECK-HF (9%, 39%, 18%, and 82% of participants, respectively).

Five distinct clusters were identified based on latent class analysis; each cluster had 694, 2066, 1709, 1069, and 1371 patients, respectively. High-distinctive discrimination was noted between clusters and confirmed via pairwise comparisons between cluster 1 as the reference and clusters 2 through 5.

Cluster classifications are as follows:

  • Cluster 1: young-low comorbidity cluster. Median age, 59 years (youngest); 58% men. Fewer comorbidities vs other clusters, New York Heart Association (NYHA) class I/II, low N-terminal pro-brain natriuretic peptide (NT-proBNP) values, normal estimated glomerular filtration rate.
  • Cluster 2: atrial fibrillation-hypertension cluster. Median age, 77 years; 54% men. Atrial fibrillation and hypertension were common; only 2% had diabetes.  
  • Cluster 3: older-atrial fibrillation cluster. Median age, 88 years; 64% women. Highest NT-proBNP values and lowest body mass index (BMI).
  • Cluster 4: hypertensive-diabetic cluster. Median age, 71 years; 67% men. High BMI value; 97% of patients had hypertension, and 95% had diabetes.
  • Cluster 5: cardiorenal cluster. Median age, 82 years; 68% women. NYHA class III/IV, ischemic heart disease, atrial fibrillation, higher NT-proBNP, and high BMI.

When the cluster model was applied to CHECK-HF for external validation, more patients were assigned to clusters 1, 2, and 4; median probabilities for each cluster demonstrated similar distribution between the two cohorts.

For the cohort overall, median follow-up time was 1.37 years (95% CI, 0.60-2.39). Investigators assessed the relationship between clusters in terms of composite outcome, all-cause mortality, cardiovascular mortality, noncardiovascular mortality, and HF hospitalization. Outcomes demonstrated differences in survival between clusters (P <.0001). Cluster 1 demonstrated the lowest 3-year event rate across outcomes (14.8%, 10.0%, 3.3%, 6.8%, and 13.0%, respectively). For the composite outcome measure, event rates were similar with age- and sex-adjusted hazard ratios (HRs) for clusters 2 and 4; event rates were 35.1% and 44.8%, respectively. Clusters 3 and 5 had the highest event rates and adjusted HRs: 61.3% and 59.3%, respectively.

HF hospitalization differences between clusters were small, with an event rate of 26.6% in cluster 2, 35.7% in cluster 3, 35.4% in cluster 4, and 42.9% in cluster 5. Adjusted HRs for clusters 3 and 4 were comparable; adjusted HR for cluster 5 was the highest (42.9%; 95% CI, 39.5-46.3).

Study limitations include missing indications for prescriptions, a lack of follow-up data in the validation cohort, the inclusion of only a small proportion of patients available in the SwedeHF cohort, and the use of a data-driven approach to phenotypic clustering, which is “highly influenced by the cohort.”

“This study demonstrates that phenotype clustering may result in clinically meaningful clusters of HFpEF patients,” the researchers concluded. “Clinical characteristics of patients between clusters varied considerably….These results signify the heterogeneity in the HFpEF population and may form a basis for tailoring trial design.”

Disclosure: Several study authors declared affiliations with the pharmaceutical industry. Please see the original reference for a full list of authors’ disclosures.


Uijl A, Savarese G, Vaartjes I, et al. Identification of distinct phenotypic clusters in heart failure with preserved ejection fraction. Eur J Heart Fail. Published online March 29, 2021. doi:10.1002/ejhf.2169