Polycystic ovary syndrome (PCOS) is among the most prevalent endocrine disorders affecting 12.1% of reproductive-aged women, yet it remains one of the least precisely defined.1 Despite its high global prevalence, PCOS continues to be diagnosed and managed through broad syndromic criteria that fail to capture its underlying biological diversity. This mismatch between clinical heterogeneity and diagnostic uniformity has long constrained mechanistic discovery, therapeutic innovation and individualised care. As artificial intelligence (AI) becomes increasingly integrated into biomedical research, it is now reshaping how complex, multifactorial diseases such as PCOS can be conceptualised and classified.2–4
Recent data-driven subtyping efforts, particularly the landmark clustering analysis published in Nature Medicine by Gao et al, demonstrate how machine learning can refine complex disease classification and move the field closer to precision care.5 6 Using unsupervised machine learning in a discovery cohort of more than 10 000 Chinese women with PCOS, Gao et al identified four reproducible subtypes—hyperandrogenic PCOS, obesity-related PCOS, sex hormone-binding globulin (SHBG)-elevated PCOS and luteinising hormone (LH)/anti-Müllerian hormone (AMH)-elevated PCOS—based on nine routinely measured clinical variables: body mass index, follicle-stimulating hormone, LH, dehydroepiandrosterone sulphate, testosterone, SHBG, AMH, fasting insulin and glucose.
Longitudinal follow-up and treatment outcome analyses within the cohort further revealed that routinely measured hormonal and metabolic variables can resolve PCOS into reproducible subgroups with distinct trajectories of metabolic risk, reproductive outcomes and treatment responsiveness. Importantly, these AI-derived subtypes are not limited to specific populations, showing consistency across independent cohorts and ancestries, a prerequisite for clinical translation. Such findings resonate with parallel progress in other complex metabolic and endocrine disorders, where AI-enabled stratification has uncovered clinically meaningful subphenotypes previously obscured by traditional diagnostic frameworks.7–9
The significance of AI-driven subtyping extends beyond improved classification. By linking baseline clinical features to long-term outcomes, this approach introduces a prognostic dimension that has been largely absent from PCOS diagnostics. It enables the identification of women at elevated risk for adverse cardiometabolic outcomes, suboptimal fertility treatment responses or treatment-related complications, thereby redefining PCOS as a condition with predictable life-course trajectories rather than a static reproductive diagnosis. From a research perspective, this stratification provides a robust framework for investigating disease aetiology, aligning with emerging genetic and multi-omics studies that suggest distinct molecular architectures underlying different PCOS presentations.10–12
Equally transformative are the implications for clinical research design. Traditional PCOS trials, typically conducted in heterogeneous populations, often yield diluted or conflicting results that impede guideline development.13–17 AI-defined subtypes provide a rational basis for subtype-stratified or enrichment trial designs, enabling more precise evaluation of therapeutic efficacy and safety. This paradigm mirrors advances in oncology and cardiometabolic medicine, where patient stratification has been instrumental in translating biological insights into effective therapies. For PCOS, such an approach could accelerate the development of targeted interventions while reducing unnecessary treatment exposure.
Looking ahead, the integration of AI-driven subtyping into routine clinical care will require coordinated efforts across disciplines. Future clinical decision-support systems may dynamically incorporate longitudinal clinical data, genetic susceptibility, lifestyle factors and treatment history to refine risk prediction and guide individualised management.18 Achieving this vision will depend not only on methodological rigour and external validation but also on the establishment of integrated care models that address reproductive, metabolic and long-term health outcomes across the lifespan.
In this context, AI-driven subtyping should be viewed not as a final destination, but as a foundational step toward precision medicine in PCOS. By translating a heterogeneous syndrome into clinically actionable subgroups, this approach opens new avenues for mechanistic discovery, trial innovation and personalised care. If thoughtfully implemented, it has the potential to transform PCOS from a diagnosis defined by variability into one guided by predictability, prevention and precision.