A new machine learning study published in the eClinicalMedicine journal has reported three distinct forms of heart failure with preserved ejection fraction (HFpEF). HFpEF accounts for half of the heart failure cases worldwide, and has diverse clinical presentations, which makes diagnosis difficult. Previous studies had poor response due to their one-size-fits-all approach.

In this study, researchers from Peking University Third Hospital employed a deep learning model, known as DeepCluster, to analyze clinical and imaging data from over 2,000 hospitalized patients with HFpEF. The model consistently identified three patient subgroups. These three subgroups reappeared consistently in two other cohort trials, including the well-known TOPCAT trial. 

The first group, metabolic HFpEF, included older, obese patients with diabetes and kidney disease.  This group showed the highest mortality risk but responded well to SGLT2 inhibitors and ARNI therapy. The second group, atrial fibrillation/right heart HFpEF, had the highest rates of rehospitalization and responded well to calcium channel blockers. The third, younger lifestyle HFpEF, often associated with smoking and alcohol use, had the best prognosis but still improved with calcium channel blocker therapy.

In the metabolic subgroup, SGLT2 inhibitors reduced heart failure hospitalizations by 55%, while ARNI therapy lowered mortality risk by nearly two-thirds. Patients with atrial fibrillation–driven HFpEF saw a 38% drop in death rates and a 42% reduction in rehospitalizations when treated with calcium channel blockers.
Owing to these differences, the study demonstrates that HFpEF is not a single disease but a spectrum of distinct conditions that require targeted therapies.