THU629 Cluster Analysis Identifies Distinct Subtypes Of PCOS

Disclosure: K. van der Ham: None. L.M. Moolhuijsen: None. K. Brewer: None. R. Sisk: None. Y.V. Louwers: None. J.A. Visser: None. A.E. Dunaif: None. J.S. Laven: None. Polycystic ovary syndrome (PCOS) is a common endocrine disorder in women of reproductive-age. The different criteria that are used to diagnose PCOS reflect the heterogeneity of the syndrome. However, PCOS diagnosed by NIH or Rotterdam criteria have a similar genetic architecture. Using Hierarchical Clustering (HC) in cohort of ∼900 NIH PCOS cases from European ancestry, we have previously identified discrete and stable PCOS clusters, which were designated “reproductive subtype” (high LH, FSH and SHBG) and “metabolic subtype” (high BMI, insulin and glucose); cases that did not belong to these clusters were designated “indeterminate subtype”. The subtypes appeared to capture biologically meaningful differences because assessment by a genome-wide association study indicated that they were characterized by distinct and novel genome-wide significant loci. In the current study, we tested the hypothesis that these subtypes would be present in an independent PCOS cohort diagnosed with Rotterdam criteria using the traits drivers (BMI, LH, FSH, DHEAS, SHBG, testosterone, fasting insulin and fasting glucose). We also compared two clustering methods, HC and K-means. We assessed whether the following additional traits not used for clustering differed between the subtypes thus identified: total follicle count (TFC), modified Ferriman-Gallwey score, and levels of anti-Müllerian hormone (AMH), estradiol, TSH, DHEA, cortisol, androstenedione, prolactin, LDL, HDL, triglyceride (TC) and cholesterol. European ancestry PCOS cases, n=2502, fulfilling Rotterdam criteria, aged 13-45 years, were included; n=1067 cases also fulfilled NIH criteria. In the Rotterdam cases, the reproductive subtype (n=450) had significantly (all P<0.001) higher TFC and levels of AMH and HDL compared to the metabolic (n=1067) and indeterminate (n=1026) subtypes. These findings suggest that clustering identifies a discrete reproductive subtype of cases with alterations in folliculogenesis, without using AMH and TFC to define the subtype. In contrast, the metabolic subtype had significantly (all P<0.001) higher TC and LDL levels compared to the other subtypes providing further evidence that this subtype identifies cases at higher cardiometabolic risk. The results were similar when the analysis was limited to NIH PCOS cases only. Comparing both clustering methods showed that there was consensus for 50% (metabolic) and 57% (reproductive) of PCOS cases. The cases in the non-consensus group showed a less pronounced metabolic and reproductive phenotype, both in the trait drivers and in the additional traits. Overall, our findings suggest that clustering algorithms capture etiologically distinct subtypes of PCOS diagnosed by both NIH and Rotterdam criteria. Further, our findings provide an example of the power of modern disease classification based on objective biologic differences rather than expert opinion. Presentation: Thursday, June 15, 2023