Data-driven endocrine–metabolic phenotypes in young women with polycystic ovary syndrome and associations with cardiometabolic risk markers

Polycystic ovary syndrome (PCOS) is one of the most common endocrine disorders in women of reproductive age and represents a leading cause of ovulatory dysfunction, hyperandrogenism, and reproductive morbidity [ 1 ]. Beyond its gynecologic manifestations, PCOS is increasingly recognized as a systemic condition associated with metabolic abnormalities, including dyslipidemia, impaired glucose tolerance, and an elevated long-term risk of type 2 diabetes and cardiovascular disease [ 2 , 3 ]. Importantly, PCOS is not a uniform disorder but rather a highly heterogeneous syndrome, with substantial variability in endocrine profiles, metabolic risk burden, and clinical trajectories across affected individuals [ 4 ]. This heterogeneity may contribute to differences in endocrine and metabolic trajectories among women with PCOS and complicates efforts toward biologically informed risk characterization [ 5 ]. Current clinical classification relies primarily on the Rotterdam diagnostic framework, which defines PCOS based on combinations of oligo-anovulation, hyperandrogenism, and polycystic ovarian morphology [ 6 ]. Although useful for diagnosis, these traditional phenotypes (A–D) are limited in their ability to capture the full multidimensional biological diversity of PCOS [ 7 , 8 ]. In particular, Rotterdam-based subtypes are criterion-driven rather than biology-driven and do not explicitly integrate broader endocrine axes or metabolic variation. As a result, biologically relevant endocrine–metabolic variation may remain underrecognized within standard diagnostic categories [ 9 ]. In recent years, there has been growing interest in applying data-driven subtyping approaches in endocrinology and metabolic medicine to refine disease classification beyond symptom-based frameworks [ 10 ]. Unsupervised machine learning methods, including principal component analysis and clustering algorithms, have been used to identify data-driven subgroup structures and biological variation in complex disorders, offering the potential to improve characterization of endocrine–metabolic heterogeneity [ 11 , 12 ]. Within PCOS, emerging studies have suggested the presence of subgroups characterized by differential metabolic dysfunction, androgen excess, or inflammatory profiles; however, most work has focused on adult populations or has not comprehensively integrated multiple endocrine axes alongside cardiometabolic markers. Notably, in young women with PCOS, early endocrine–metabolic phenotypes that jointly reflect gonadal, thyroid, and adrenal hormone regulation together with lipid and glucose metabolism remain poorly characterized. Crosstalk between ovarian steroidogenesis, thyroid function, adrenal hormone production, and metabolism signaling pathways is increasingly recognized as a key determinant of phenotypic expression in PCOS [ 13 , 14 ]. Studying these relationships in adolescence and early adulthood is particularly relevant, as metabolic complications often evolve gradually and may not yet be clinically overt [ 15 , 16 ]. Early endocrine–metabolic subgroup patterns potentially associated with adverse metabolic profiles therefore remains an important clinical challenge, given that conventional screening strategies frequently rely on abnormalities that emerge later in the disease course [ 17 ]. Importantly, the application of data-driven phenotyping in the young PCOS population remains limited, despite the potential clinical value of identifying early endocrine-metabolic constellations that may precede overt cardiometabolic disease. Integrating biomarkers across multiple hormonal axes together with metabolic parameters may provide a more biologically grounded framework for characterizing endocrine–metabolic heterogeneity than traditional diagnostic classifications alone. Improved characterization of endocrine–metabolic variability may help refine understanding of early biological heterogeneity in PCOS and support future hypothesis-generating research.