Metabolically Distinct Phenotypes of Polycystic Ovary Syndrome Identified from Seven Routinely Measured Variables: A Cross-Sectional Study in Real-World Clinical Practice

Polycystic ovary syndrome (PCOS) is clinically heterogeneous, and analysis of incomplete clinical data may reveal clinically meaningful patterns. We investigated whether routinely measured variables could identify interpretable PCOS phenotypes in 1735 women diagnosed according to the Rotterdam criteria. Unsupervised clustering used body mass index (BMI), fasting glucose, fasting insulin, total testosterone, follicle-stimulating hormone (FSH), luteinizing hormone (LH), and sex hormone-binding globulin (SHBG) across multiply imputed datasets. Phenotypes were characterized using additional reproductive, androgenic, and metabolic traits. Metabolic analyses were adjusted for age and BMI, with sensitivity analyses for missing postload outcomes. Two phenotypes were identified, Metabolic-PCOS in 657 women (37.9%) and Reproductive-PCOS in 1078 (62.1%). Reproductive-PCOS showed a gonadotropin-predominant profile, whereas Metabolic-PCOS showed a less favorable metabolic profile. Separation between these profiles was modest, supporting their interpretation as predominant clinical patterns rather than sharply distinct categories. Metabolic-PCOS also had higher 2-h glucose, 120-min insulin, triglycerides, low-density lipoprotein cholesterol (LDL-C), and uric acid, which were not used for phenotype derivation. After adjustment, 2-h glucose was 16.4% higher, 120-min insulin was 111.6% higher, and the odds of elevated 2-h glucose (≥7.8 mmol/L) were greater (odds ratio [OR] 3.98, 95% confidence interval [CI] 2.62 to 6.04). Findings were consistent across sensitivity analyses. This parsimonious biomedical data mining approach identified clinically interpretable PCOS phenotypes that warrant external and prospective validation for metabolic risk stratification.