6 Latent Menopausal Phenotypes Identified by Machine Learning in 10-Year SWAN Follow-Up

A machine learning (ML) and artificial intelligence (AI) approach identified 6 clinically meaningful menopausal phenotypes in a 10-year follow-up of the Study of Women’s Health Across the Nation (SWAN) cohort, highlighting the heterogeneity of menopause and potential avenues for personalized care. The findings were presented during The Menopause Society 2025 Annual Meeting, October 21=25, in Orlando, FL.

The authors noted that menopause presents with diverse symptoms—including vasomotor disturbances, sleep disruption, and mood changes—superimposed on variable metabolic and cardiovascular risk profiles. Traditional single-factor stratification, such as grouping women solely by vasomotor symptom severity, fails to capture this complexity. The study leveraged multi-factor analysis to integrate symptoms, cardiometabolic health, and hormone levels, offering a more comprehensive view of midlife women’s health trajectories.

Study Design

• ~3,000 women in midlife from a the multiethnic US cohort, followed annually from baseline (Visit 1) through 10-year follow-up (visit 10)
• Unsupervised ML techniques (K-means clustering and latent class analysis) grouped participants based on:
  • Menopausal symptom severity: vasomotor, sleep, mood
  • Anthropometric and cardiometabolic measures: BMI, blood pressure, fasting glucose, lipids
  • Sex hormone levels

The researchers reported that 6 optimal phenotypic clusters emerged based on validation metrics that included silhouette analysis. They specifically assessed phenotype stability, transitions over time, hormone therapy (HRT) use, and family medical history associations.

The 6 Phenotypes

1. High multisymptom: multiple severe menopausal symptoms with moderate metabolic elevations; had the poorest quality of life and persistent symptoms into late menopause
2. Metabolic risk: obesity, hypertension, adverse glucose/lipid profile, few menopausal symptoms; 10-year incidence of type 2 diabetes and metabolic syndrome hazard ratio ~1.8 vs Asymptomatic group
3. Asymptomatic: minimal symptoms, healthy metabolic and hormonal profile; fewest chronic health issues
4. Mood-dominant: elevated depression/anxiety, minimal vasomotor complaints, moderate metabolic status
5. Healthy obese: high BMI but metabolically healthy, low symptom burden; no excess metabolic events
6. Mild symptoms: generally mild menopausal symptoms with average health metrics

Phenotype Stability and Transitions

• ~65% of women remained in the same phenotype from baseline to visit 10
• Women on HRT were more likely to shift from high-symptom to milder phenotypes (odds ratio ~2.0)
• Women not on HRT more often persisted in their initial phenotype

The patterns were influenced by family history, according to the study, including metabolic or cardiovascular conditions associated with remaining in the Metabolic risk group and a history of mood disorders linked to the Mood-dominant phenotype.

Additional Insights

Using multifactor ML analysis, the researchers identified more distinct clusters than prior symptom-only latent class analyses, indicated by higher silhouette scores. These new phenotypes better predicted long-term health outcomes, highlighting the potential for subgroup-specific interventions, they noted.

The authors concluded that AI/ML can uncover latent menopausal phenotypes that traditional approaches overlook. The "data-driven phenotypes point toward more personalized menopause care via subgroup-specific interventions. Our findings underscore the potential of AI in menopause research and warrant further validation of subgroup-specific clinical treatment response trajectories," the authors said.

Source: Grygoryan O, Liu J, Schwab E. Machine learning uncovers six latent menopausal phenotypes in SWEAN (10-year follow-up). Poster presented at: The Menopause Society’s 2025 Annual Meeting; September 21-25, 2025; Orlando, FL. Accessed October 21, 2025.