Data Driven Endocrine Metabolic Phenotypes in Young Women With Polycystic Ovary Syndrome and Associations With Cardiometabolic Risk Markers

In a large cohort of young women with polycystic ovary syndrome, data-driven clustering identified two distinct endocrine-metabolic phenotypes, including a rare thyroid/autoimmune-enriched subgroup associated with a higher prevalence of atherogenic lipid profiles, highlighting the value of integrated profiling for early cardiometabolic risk stratification beyond traditional diagnostic criteria.

The Gist

The Big Picture: PCOS isn't just one thing

Imagine Polycystic Ovary Syndrome (PCOS) not as a single, uniform disease, but more like a giant, chaotic orchestra. In this orchestra, some musicians are playing too loud (hormones), some are out of tune (metabolism), and some are missing entirely.

For a long time, doctors have tried to sort these musicians into just four standard groups (called "phenotypes") based on a simple checklist: Do they have irregular periods? Do they have high male hormones? Do they have cysts on their ovaries?

But the researchers in this paper asked a better question: "What if we stop looking at the checklist and start listening to the actual music?" They wanted to see if there were hidden patterns in how the body's chemistry works that the old checklist missed.

The Experiment: Listening to the "Bio-Symphony"

The team studied 1,032 young women (ages 16–25) in Poland who had been diagnosed with PCOS. Instead of just checking their ovaries, they took a massive "snapshot" of their entire internal chemistry.

Think of this like taking a 360-degree photo of their bodies, capturing:

• Reproductive hormones (the "music" of the ovaries).
• Thyroid hormones (the body's "thermostat" and energy regulator).
• Adrenal hormones (the "stress" system).
• Metabolic markers (sugar and fat levels in the blood).

They used a computer program (a type of artificial intelligence) to find patterns in this massive amount of data, looking for groups of women whose bodies were "playing the same tune."

The Discovery: Two Distinct "Tribes"

The computer found that these young women naturally fell into two main groups, rather than the four standard ones doctors usually use.

Group 1: The "Standard" Mix (92% of the women)

• The Analogy: Think of this group as a jazz band. It's a bit of everything. They have the typical PCOS mix of hormonal ups and downs and metabolic quirks. It's the "default" setting for most people with PCOS.
• The Reality: This is the large majority of the women studied. Their bodies showed a "mixed" profile without one single dominant issue.

Group 2: The "Thyroid & Autoimmune" Tribe (8% of the women)

• The Analogy: Imagine a specialized choir within the orchestra. These women aren't just playing the standard PCOS notes; they have a very specific, loud section playing the "Thyroid" and "Immune System" instruments.
• The Reality: This smaller group had significantly higher levels of thyroid antibodies (signs the immune system is attacking the thyroid) and slightly higher thyroid-stimulating hormone (TSH).
• The Surprise: Even though this group was small, they had a distinct "signature." Their bodies were fighting a different battle than the rest of the group.

Why Does This Matter? (The "Heart" of the Issue)

The researchers wanted to know: Does belonging to one of these groups change your risk for future heart or diabetes problems?

• The Result: The "Thyroid Tribe" (Group 2) had a slightly higher risk of having an "atherogenic lipid profile."
• The Analogy: Think of your blood vessels as pipes. "Atherogenic lipids" are like grease and sludge building up in the pipes. This group had a slightly higher tendency for their blood to get "sludgy" (high triglycerides relative to good cholesterol).
• The Takeaway: Even though these women were young and didn't have diabetes yet, their specific "Thyroid" type of PCOS might be a warning sign that their pipes are getting clogged earlier than the others.

The "Aha!" Moment

The most important lesson from this paper is that PCOS is not a one-size-fits-all disease.

By using data to listen to the "bio-symphony," the researchers found that about 1 in 13 young women with PCOS has a specific subtype driven by thyroid and immune issues.

Why is this a big deal?

1. Better Diagnosis: Currently, a doctor might treat all PCOS patients the same. This study suggests that if a young woman has this specific "Thyroid Tribe" signature, she might need extra attention on her thyroid and heart health, not just her ovaries.
2. Early Warning: These differences show up in young women (teens and early 20s). Catching this early means doctors can prevent heart trouble before it starts, rather than waiting until the patient is older and sicker.
3. Personalized Medicine: It moves us away from a "cookie-cutter" approach to a "tailor-made" approach. Just like you wouldn't wear a suit that fits everyone, you shouldn't treat every PCOS patient exactly the same.

In a Nutshell

This study is like finding a hidden map inside a complex city. For years, we thought everyone with PCOS lived in the same neighborhood. Now, we know there's a specific, smaller neighborhood (the Thyroid/Immune group) that has different risks and needs different care. By identifying this group early, doctors can help these young women stay healthier for longer.

Technical Summary

1. Problem Statement

Polycystic Ovary Syndrome (PCOS) is a heterogeneous endocrine disorder characterized by reproductive dysfunction and long-term cardiometabolic risks. Current clinical classification relies on the Rotterdam criteria (oligo-anovulation, hyperandrogenism, and polycystic ovarian morphology), which creates four phenotypes (A–D). However, these criteria are criterion-driven rather than biology-driven, failing to capture the multidimensional biological variability across different endocrine axes (gonadal, thyroid, adrenal) and metabolic pathways.

There is a critical gap in understanding the early endocrine-metabolic constellations in young women (adolescents and early adults) with PCOS. Traditional screening often misses subtle heterogeneity that may precede overt cardiometabolic disease. The authors hypothesize that unsupervised, data-driven approaches can identify biologically coherent subtypes (endotypes) that offer better risk stratification than traditional diagnostic categories.

2. Methodology

Study Design and Population

• Design: Cross-sectional study conducted at a tertiary Gynecological Endocrinology Clinic in Poland (Jan 2018 – May 2025).
• Participants: Initially 1,300 young women (16–25 years) diagnosed with PCOS via Rotterdam criteria.
• Analytic Cohort: 1,032 participants retained after applying a complete-case approach to ensure data integrity across core biomarkers.
• Exclusions: Women with alternative causes of hyperandrogenism (e.g., Cushing's, tumors), diabetes, or recent hormonal/metabolic therapy.

Data Collection & Features

The study integrated biomarkers across four key physiological axes:

1. Gonadal Axis: LH, FSH, LH/FSH ratio, Total Testosterone, SHBG.
2. Thyroid Axis: TSH, Free Thyroxine (FT4), Anti-Thyroid Peroxidase (anti-TPO) antibodies.
3. Adrenal Axis: DHEA-S, Morning/Evening Cortisol, Diurnal Cortisol Ratio.
4. Metabolic Parameters: Fasting glucose, 2-hour OGTT glucose, Lipid profile (TG, HDL-C, LDL-C, Total Cholesterol, Non-HDL-C).

• Note: Anti-TPO was log-transformed due to skewness.

Analytical Pipeline

1. Preprocessing: Z-score normalization of continuous variables.
2. Dimensionality Reduction: Principal Component Analysis (PCA) was applied to reduce collinearity. The first 10 principal components (PCs) were retained, explaining 81.9% of the total variance.
   • PC1: Driven by lipid markers (Dyslipidemic-Metabolic axis).
   • PC3: Gonadotropin-related reproductive axis.
   • PC5: Thyroid/Autoimmune features.
3. Clustering: Gaussian Mixture Models (GMM) were applied in the reduced PCA space.
   • Model selection (K=2 to 10) utilized Silhouette coefficients, AIC/BIC, and stability metrics (Adjusted Rand Index).
   • Final Model: A 2-cluster solution was selected based on stability (ARI = 0.842) and clinical interpretability.
4. Validation:
   • Internal: High membership certainty (mean posterior probability = 0.989).
   • External (Post-hoc): Anti-Müllerian Hormone (AMH) was analyzed as an external validator; it showed no significant difference between clusters, confirming the separation was not driven by ovarian reserve markers.
5. Statistical Testing: Associations with cardiometabolic risk endpoints were tested using Chi-square/Fisher's exact tests and Logistic Regression (adjusted for age).

3. Key Results

Identified Phenotypes

The unsupervised clustering revealed two distinct, stable phenotypes:

• Cluster 0 (Reference, n=954; 92.4%): A "mixed" endocrine-metabolic profile without a single dominant axis. This represents the majority of the cohort.
• Cluster 1 (Enriched, n=78; 7.6%): A distinct Thyroid/Autoimmune-enriched phenotype.
  • Key Characteristics: Significantly higher levels of Anti-TPO antibodies (median 20.1 vs 10 IU/mL) and TSH (median 2.24 vs 1.77 µU/mL).
  • Other Markers: Higher evening cortisol and cortisol ratio; slightly higher LH levels.

Cardiometabolic Risk Associations

The study evaluated four risk endpoints: Impaired Glucose Tolerance (2h-OGTT ≥140 mg/dL), Atherogenic Lipid Profile (TG/HDL-C >3.50), Elevated Non-HDL-C (≥130 mg/dL), and a composite "Any Abnormality."

• Atherogenic Lipid Profile: Cluster 1 showed a significantly higher prevalence of an atherogenic lipid profile (TG/HDL-C >3.50) compared to Cluster 0 (12.8% vs 6.1%, OR = 2.27, 95% CI 1.11–4.64, p=0.024).
• Glucose Intolerance & Non-HDL-C: Differences were observed (Cluster 1 had higher rates) but were not statistically significant after multiple testing correction (p > 0.05).
• Composite Outcome: Cluster 1 had a higher prevalence of any cardiometabolic abnormality (42.3% vs 32.2%), though this did not reach statistical significance (p=0.069).

4. Key Contributions

1. Data-Driven Subtyping: Successfully applied unsupervised machine learning (PCA + GMM) to move beyond the symptom-based Rotterdam criteria, identifying a biologically distinct thyroid/autoimmune-enriched endotype within young PCOS patients.
2. Early Detection of Heterogeneity: Demonstrated that significant endocrine heterogeneity (specifically thyroid autoimmunity) is detectable in young women (16–25 years), potentially preceding the onset of severe metabolic complications.
3. Integration of Multi-Axis Data: Unlike previous studies focusing solely on reproductive or metabolic axes, this study integrated gonadal, thyroid, adrenal, and metabolic data, revealing that thyroid status is a relevant dimension of PCOS heterogeneity.
4. Clinical Stratification Potential: Identified a small but distinct subgroup (7.6%) that may require different monitoring strategies, specifically regarding thyroid function and lipid metabolism, even in the absence of overt metabolic syndrome.

5. Significance and Implications

• Precision Medicine: The findings support a shift toward multidimensional phenotyping in PCOS. The identification of a thyroid-enriched subgroup suggests that thyroid autoimmunity is not merely a comorbidity but a core component of a specific PCOS endotype.
• Risk Stratification: While metabolic differences were modest in this young cohort, the enrichment of atherogenic lipids in the thyroid-enriched group suggests that early endocrine divergence may predict future cardiometabolic trajectories.
• Clinical Practice: The study advocates for systematic thyroid evaluation (TSH and anti-TPO) in young women with PCOS, particularly for those who may belong to this specific phenotype, to enable earlier intervention and longitudinal monitoring.
• Limitations & Future Directions: The cross-sectional design limits causal inference. The lack of BMI data and insulin resistance markers (HOMA-IR) in the clustering features is a limitation. Future longitudinal studies are required to determine if these phenotypes predict the development of Type 2 diabetes or cardiovascular disease over time.

Conclusion: The study provides robust evidence that PCOS in young women is not a monolithic disorder but contains reproducible, data-driven phenotypes. Specifically, a distinct thyroid/autoimmune-enriched subgroup exists, characterized by higher TSH and anti-TPO levels and a trend toward atherogenic dyslipidemia, offering a new target for early risk stratification and personalized management.