Background: Social determinants of health affect health care utilization, access, and outcomes. When designing population health management strategies, phenotyping sub-populations of patients using both clinical and social risk factors is essential. Aims: Define sub-populations of heart failure (HF) patients using a novel machine learning method based on demographics, comorbidities, and social vulnerability. Compare days alive and out of hospital (DAOH) within 1-year post-discharge. Methods: Retrospective cohort study of patients admitted to 48 hospitals in Providence St. Joseph Health with a primary HF diagnosis from 02/01/2016 - 12/31/2022. CDC social vulnerability index (SVI) was linked to patients’ home address. Thirty-four demographic, clinical, and SVI variables were included in the model. Clusters were defined using hierarchical clustering of principal components (HCPC). Top drivers of cluster formation were assessed by v-test . Due to large sample size, 10 random samples were selected and clusters were assigned by vote. Cluster enriched subgroups (CES) were defined as patients having ≥6 cluster assignments and >80% cluster vote. Results: HCPC analysis grouped 61,126 hospitalizations into 3 clusters (Figure). CES-1 patients (n=25,873) were older, wealthier, White, with multiple comorbidities living in more educated communities. CES-2 patients (n=13,450) were younger with combined systolic/diastolic HF and fewer comorbidities living in crowded, urban communities. CES-3 patients (n=3,787) were non-White, Hispanic with higher cardiometabolic risk factors living in urban, crowded, low-income communities. CES-1 patients had fewer DAOH (230) and higher 1-year (22%) mortality. CES-2 patients had similar DAOH (274) to CES-3 (269), despite younger age and fewer comorbidities. Conclusions: In a large heart failure cohort, unique sub-populations based on clinical and social factors have distinct care needs, utilization, and outcomes.