Machine learning to define phenotypes and outcomes of patients hospitalized for heart failure with preserved ejection fraction: Findings from ASCEND-HF

Abstract

Heart Failure with Preserved Ejection Fraction (HFpEF) is a heterogenous disease with few therapies proven to provide clinical benefit. Machine learning can characterize distinct phenotypes and compare outcomes among patients with HFpEF who are hospitalized for acute HF. We applied hierarchical clustering using demographics, comorbidities, and clinical data on admission to identify distinct clusters in hospitalized HFpEF (ejection fraction >40%) in the ASCEND-HF trial. We separately applied a previously developed latent class analysis (LCA) clustering method and compared in-hospital and long-term outcomes across cluster groups. Of 7141 patients enrolled in the ASCEND-HF trial, 812 (11.4%) were hospitalized for HFpEF and met the criteria for complete case analysis. Hierarchical Cluster 1 included older women with atrial fibrillation (AF). Cluster 2 had elevated resting blood pressure. Cluster 3 had young men with obesity and diabetes. Cluster 4 had low resting blood pressure. Mortality at 180 days was lowest among Cluster 3 (KM event-rate 6.2 [95% CI: 3.5, 10.9]) and highest among Cluster 4 (18.8 [14.6, 24.0], P < .001). Twenty four-hour urine output was higher in Cluster 3 (2700mL [1800, 3975]) than Cluster 4 (2100mL [1400, 3055], P < .001). LCA also identified four clusters: A) older White or Asian women, B) younger men with few comorbidities, C) older individuals with AF and renal impairment, and D) patients with obesity and diabetes. Mortality at 180 days was lowest among LCA Cluster B (KM event-rate 5.5 [2.0, 10.3]) and highest among LCA Cluster C (26.3 [19.2, 35.4], P < .001). In patients hospitalized for HFpEF, cluster analysis demonstrated distinct phenotypes with differing clinical profiles and outcomes.