Heart failure with preserved ejection fraction (HFpEF) is an increasingly widespread medical condition, with excessive morbidity and mortality. Recently, for the first time in HFpEF, a reduction in the primary composite outcome of cardiovascular death or HF hospitalization was shown with empagliflozin. The failure of previous clinical trials in HFpEF might have resulted from suboptimal patient selection and inclusion of patients without "true" or clinically significant HFpEF. Another important factor might be the heterogeneity of HFpEF, and thus there is a growing interest in HFpEF phenotyping. This phenotyping can be based on clinical presentation (e.g., subtypes with predominant atrial fibrillation, obesity, pulmonary hypertension and right ventricular failure, coronary artery disease (CAD), or noncardiac comorbidities), but also on HFpEF etiology. Specific therapies, such as tafamidis in transthyretin-related amyloidosis (ATTR) or mavacamten in hypertrophic cardiomyopathy, have demonstrated their efficacy. However, pathomechanisms leading to the development of different phenotypes of HFpEF seem more complex and subtle. Machine learning and neural network models might help identify specific subgroups within the HFpEF population that either cluster patients with similar genetic, biochemical, echocardiographic or clinical characteristics, or respond similarly to a given treatment. Herein, we review different approaches to HFpEF phenotyping and present some distinct HFpEF subtypes.
Keywords: artificial intelligence; diastolic dysfunction; heart failure with preserved ejection fraction; phenotype.