Uncovering Distinct Phenotypes of Early Onset Heart Failure in Women
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Background
Heart failure in younger women is understudied and may reflect heterogeneous pathways and risk not adequately captured by conventional classifications. We aimed to identify clinically distinct phenotypes of early-onset heart failure in women, assess their prognostic significance, and evaluate their transportability in an independent national cohort.
Methods
We identified women aged 18-55 years with heart failure using electronic health record data from a large multi-hospital system. A variational deep embedding model incorporating 283 clinical features was used for unsupervised phenotyping. Associations with subsequent inpatient hospitalization, emergency department utilization, and mortality were evaluated using age-stratified Cox models. Phenotype transportability was assessed by both applying the institutionally trained model and by independent de novo clustering as a replication analysis in the All of Us cohort.
Results
Among 1,407 young women with heart failure (median age, 45 years), four phenotypes were identified: hypertensive pregnancy (n=494), cardiorenal (n=276), autoimmune (n=394), and cardiometabolic-ischemic (n=243), with a bootstrap adjusted Rand index of 0.65 and silhouette score of 0.45. Compared with the hypertensive pregnancy phenotype, risks of inpatient hospitalization were highest for the cardiorenal (HR 2.86 [95% CI, 2.35–3.49]) and autoimmune (HR 2.25 [95% CI, 1.87–2.70]) phenotypes; mortality risks were similarly elevated (HR 4.65 [95% CI, 3.17–6.82] and HR 4.38 [95% CI, 3.03–6.31], respectively) in these two groups. In All of Us participants (n=682), transferred phenotypes retained their principal clinical characteristics and continued to stratify hospitalization risk. Independent clustering most strongly reproduced the cardiorenal phenotype (Jaccard index, 0.744).
Conclusions
Early-onset heart failure in women comprises distinct cardio-obstetric, cardiorenal, autoimmune, and cardiometabolic-ischemic phenotypes with divergent clinical outcomes. These findings support a life-course approach to heart failure classification in women that complements conventional frameworks and incorporates pregnancy history and female-predominant systemic conditions.
Clinical Perspective
What Is New?
Unsupervised machine learning identifies distinct clinical phenotypes of heart failure among young women, highlighting heterogeneity in a population that remains understudied.
These phenotypes captured differing cardiometabolic, cardiovascular, and pregnancy-related characteristics, suggesting potentially distinct pathways contributing to heart failure development.
External validation in the All of Us cohort supported external validity and demonstrated the potential for identifying these patterns across diverse healthcare populations.
What Are the Clinical Implications?
Heart failure in young women is clinically diverse, underscoring the importance of comprehensive assessment beyond traditional heart failure classifications, including cardiometabolic risk factors and pregnancy history.
Recognition of distinct clinical phenotypes may help clinicians identify women with differing risk profiles and inform more individualized approaches to cardiovascular evaluation, risk factor management, and longitudinal follow-up.
Further studies are needed to determine whether phenotypes differ in treatment responses and whether phenotype-guided management can improve outcomes.
