Thyroid Autoimmunity in Polycystic Ovary Syndrome: Phenotype Distribution, HDL-Cholesterol, and Data-Driven Clusters in a Retrospective Cohort Study

Background and Objectives: Autoimmune thyroiditis (AIT) is often reported patients with PCOS, and may co-occur with altered metabolic risk markers. The aims of this study were to assess baseline differences according to thyroid autoimmunity status, evaluate adjusted associations between thyroid autoimmunity and metabolic parameters, examine associations with PCOS phenotype distribution, and perform k-means clustering to explore data-driven subgroups and their autoimmune enrichment. Materials and Methods: We performed a retrospective cohort study of 651 women with PCOS, comparing those without AIT (n = 506) versus with AIT (n = 145). Associations between AIT and continuous outcomes (HDL; composite metabolic score) were evaluated using robust linear regression with HC3 standard errors and age modeled with a natural cubic spline (3 knots). The association between AIT and phenotype A was assessed via logistic regression with exponentiated coefficients (odds ratios). Unsupervised phenotyping used k-means clustering with silhouette analysis across k = 2…6. Results: Patients with AIT were older (median 40 vs. 35 years; p = 0.021). Phenotype distribution differed by AIT status (overall p = 0.029), with phenotype A less frequent among AIT-positive women (27% vs. 40%). In adjusted robust regression, AIT was associated with lower HDL by β = −4.34 mg/dL (95% CI −9.18 to 0.51; p = 0.081), while obesity (−7.04 mg/dL; p < 0.001) and diabetes (−6.47 mg/dL; p = 0.004) were associated with lower HDL. AIT was not associated with the composite metabolic score (β = −0.005; 95% CI −1.22 to 1.21; p = 0.994), whereas obesity was associated with higher score (β = 1.76; p = 0.003) and urban residence with lower score (β = −0.94; p = 0.011). In logistic regression, AIT was associated with lower odds of phenotype A (OR 0.63; 95% CI 0.41–0.97; p = 0.038), and hypertension was associated with higher odds of phenotype A (OR 1.91; 95% CI 1.20–3.04; p = 0.006). Silhouette analysis supported k=3 clusters (silhouette 0.349), and AIT prevalence was highest in cluster 3 (26.4%) versus clusters 1 (19.9%) and 2 (18.3%). Conclusions: AIT was associated with lower odds of phenotype A, and showed a borderline association with lower HDL-cholesterol but not with a composite metabolic score. Data-driven clustering identified a subgroup with higher autoimmune burden.