Deep learning-based phenotyping and assessment of treatment responses in heart failure with preserved ejection fraction

Abstract Background Heart failure with preserved ejection fraction (HFpEF) accounts for over half of heart failure cases, yet effective treatments remain limited due to the syndrome’s clinical heterogeneity. Purpose This study aimed to identify distinct HFpEF phenotypes within a well-defined cohort of Chinese HFpEF patients, and to examine clinical characteristics, prognosis, and treatment responses across different phenogroups. Methods The study included 2,147 hospitalized patients with HF with left ventricular ejection fraction (LVEF) ≥50% at our hospital (2012–2023). A two-stage DeepCluster model based on fully-connected neural network was used to identify HFpEF phenogroups based on 107 EMR-based demographic and clinical variables. Kaplan-Meier curves and Cox proportional hazard models assessed prognosis and treatment responses. The phenotyping model was validated externally with data from the TOPCAT clinical trial and a well-characterized HFpEF patient cohort from University of Michigan Health System (UMHS). Results Three distinct HFpEF phenogroups were identified. Phenogroup 1 (n=815) had the highest burden of metabolic comorbidities, along with left ventricular hypertrophy, and both systolic and diastolic dysfunction. Phenogroup 2 (n=608) comprised predominantly lean, older females with atrial fibrillation and structural abnormalities in the atria and right ventricle, with mainly diastolic dysfunction. Phenogroup 3 (n=724) included younger men with unhealthy lifestyles, and exhibited higher burdens of hyperlipidemia and liver dysfunction. Overall, phenogroup 1 had the highest risk for all-cause mortality. We didn’t observe significant survival benefits from major HF therapies overall. However, after phenotyping, we observed that phenogroup 1 experienced a 56% reduced risk of HF-rehospitalization from prescription of sodium-glucose cotransporter 2 (SGLT2) inhibitors (HR = 0.44, 95%CI 0.20–0.98) and a 59% risk reduction of all-cause mortality (HR = 0.41, 95%CI 0.18–0.94) from angiotensin receptor–neprilysin inhibitors (ARNIs). In comparison, calcium channel blockers (CCBs) were associated with 38% reductions in the risk of all-cause mortality (HR = 0.62, 95%CI 0.39–0.99) and 42% reductions in the risk of HF rehospitalization (HR = 0.58, 95%CI 0.38–0.87) in phenogroup 2, and CCBs were also associated with a lower risk of all-cause mortality in phenogroup 3. Furthermore, our DeepCluster model was external validation in the UMHS cohort with well-characterized hemodynamic features also showed consistent pathophysiologic characteristics for the three identified phenogroups. Conclusions Deep learning-based phenotyping identifies HFpEF phenogroups with unique clinical features and treatment responses. SGLT2 inhibitors and ARNI improve the outcomes in patients with cardiometabolic comorbidities and both systolic and diastolic impairment, while CCBs are most likely to benefit HFpEF patients with atrial fibrillation or mild symptoms.Graphic Abstract two-stage DeepCluster model