Introduction : Heart failure with preserved ejection fraction (HFpEF) is a heterogeneous condition with distinct phenotypes. Prior studies have used cluster analysis to group patients with chronic ambulatory HFpEF by shared characteristics. Data are limited regarding different HFpEF phenotypes in the hospitalized setting. Hypothesis We hypothesized that distinct clusters of patients hospitalized for HFpEF would have unique baseline characteristics and distinct clinical outcomes. Methods : We applied an unsupervised machine learning analysis to group patients hospitalized with HFpEF in the ASCEND-HF trial into distinct phenotypes. Groups were made via hierarchical clustering after variable reduction on a set of 39 baseline variables including demographics, medical comorbidities, and clinical data at the time of admission. Cox regression models were used to evaluate the association between HFpEF phenotypes and 180-day mortality. Results : Overall, 812 patients with EF >40% were included. Hierarchical clustering identified 4 phenotypes that had distinct baseline characteristics and outcomes (Figure). Cluster 1 was older, predominantly White/Asian, female, and had high rates of atrial fibrillation. Cluster 2 had high rates of hypertension, the highest systolic blood pressure, and lowest heart rate. Cluster 3 was the youngest, predominantly male, had a high proportion of Black patients, and had the highest BMI and diabetes prevalence. Cluster 4 had a high the lowest systolic blood pressure, a high heart rate, and high natriuretic peptide concentrations. Cluster 4 (chosen as the reference group) demonstrated the highest 180-day mortality, followed by Cluster 1 (HR= 0.68 [95% CI: 0.42-1.09]), Cluster 2 (HR=0.45 [95% CI: 0.25-0.79]), and Cluster 3 (HR=0.31 [95% CI: 0.16-0.59]) (Overall p-value =0.007). Conclusions : In this clinical trial cohort of patients hospitalized for HFpEF, cluster analysis incorporating demographic, vital sign, laboratory, and comorbidity data demonstrated distinct patient phenotypes with differing clinical profiles and outcomes. These clusters may inform patient selection for future clinical trials and may facilitate matching investigational therapies to patients more likely to benefit. : Heart failure with preserved ejection fraction (HFpEF) is a heterogeneous condition with distinct phenotypes. Prior studies have used cluster analysis to group patients with chronic ambulatory HFpEF by shared characteristics. Data are limited regarding different HFpEF phenotypes in the hospitalized setting. We hypothesized that distinct clusters of patients hospitalized for HFpEF would have unique baseline characteristics and distinct clinical outcomes. : We applied an unsupervised machine learning analysis to group patients hospitalized with HFpEF in the ASCEND-HF trial into distinct phenotypes. Groups were made via hierarchical clustering after variable reduction on a set of 39 baseline variables including demographics, medical comorbidities, and clinical data at the time of admission. Cox regression models were used to evaluate the association between HFpEF phenotypes and 180-day mortality. : Overall, 812 patients with EF >40% were included. Hierarchical clustering identified 4 phenotypes that had distinct baseline characteristics and outcomes (Figure). Cluster 1 was older, predominantly White/Asian, female, and had high rates of atrial fibrillation. Cluster 2 had high rates of hypertension, the highest systolic blood pressure, and lowest heart rate. Cluster 3 was the youngest, predominantly male, had a high proportion of Black patients, and had the highest BMI and diabetes prevalence. Cluster 4 had a high the lowest systolic blood pressure, a high heart rate, and high natriuretic peptide concentrations. Cluster 4 (chosen as the reference group) demonstrated the highest 180-day mortality, followed by Cluster 1 (HR= 0.68 [95% CI: 0.42-1.09]), Cluster 2 (HR=0.45 [95% CI: 0.25-0.79]), and Cluster 3 (HR=0.31 [95% CI: 0.16-0.59]) (Overall p-value =0.007). : In this clinical trial cohort of patients hospitalized for HFpEF, cluster analysis incorporating demographic, vital sign, laboratory, and comorbidity data demonstrated distinct patient phenotypes with differing clinical profiles and outcomes. These clusters may inform patient selection for future clinical trials and may facilitate matching investigational therapies to patients more likely to benefit.