The study drew on an existing single-center database of 495 women evaluated at the hospital’s Reproductive Medicine Center, a cohort defined by its association with infertility care. From that pool, 431 women had complete records for age, body mass index, anti-Müllerian hormone, inhibin B, and a validated eight-steroid panel measured by mass spectrometry. The researchers sorted the women into categories based on historical antral follicle count, the number of small resting follicles visible on ultrasound, combined with menstrual-cycle function. A reference group with low follicle counts and regular cycles comprised 27 women, an isolated high-follicle-count group comprised 69, and a combined group with both high counts and menstrual dysfunction comprised 333, yielding a three-group analysis cohort of 429 women.

Anti-Müllerian hormone, a growth factor secreted by small growing follicles, showed a dramatic and statistically unambiguous gradient across the three categories. Median values climbed from 0.89 nanograms per milliliter in the reference group to 3.85 in the isolated high-count group and 5.76 in the combined group, a difference with a p-value of 2.51 × 10⁻¹⁷. When the researchers adjusted for age, body mass index, and the hormone itself in multinomial regression, each standard deviation increase in anti-Müllerian hormone raised the odds of belonging to the isolated high-count group more than fivefold, and the odds of belonging to the combined group nearly ninefold compared with the reference.

The classical androgens told a different story. Testosterone and androstenedione, the two workhorse androgens traditionally measured in suspected cases of androgen excess, did differ between the groups in raw, unadjusted comparisons. Yet once the models accounted for anti-Müllerian hormone, age, and body mass index, those differences evaporated. The residual distributions of testosterone and androstenedione, conditioned on those covariates, no longer separated the groups, meaning that whatever signal these hormones carried was largely redundant with the information already captured by anti-Müllerian hormone and body characteristics.

Perhaps the most technically ambitious part of the analysis involved the 11-oxygenated androgens, a relatively recently appreciated family of steroid metabolites, including 11β-hydroxytestosterone and 11-ketotestosterone, that are produced through an adrenal-centric pathway and have been proposed as culprits in the androgen excess of polycystic ovary syndrome. Using the mass spectrometry panel, the researchers quantified four of these metabolites and asked whether they distinguished the three ovarian-menstrual categories. The answer was no: none of the four 11-oxygenated androgens differed significantly across the groups, a result that tempers the hypothesis that these novel androgens are central markers of the polycystic ovary phenotype itself.

The story shifted, however, when the team turned to metabolism. In analyses that were not prespecified, an unsupervised k-means clustering of the cohort revealed a subgroup characterized by higher insulin resistance, measured by the homeostatic model assessment of insulin resistance, together with higher body mass index. In that cluster, 11β-hydroxytestosterone and 11-ketotestosterone were significantly elevated, with p-values of 6.78 × 10⁻⁴ and 8.51 × 10⁻⁴ respectively. In an extreme comparison between women with polycystic ovary syndrome and high body mass index versus those without either feature, only 11β-hydroxytestosterone remained significantly different.

To gauge whether a small hormone panel could classify patients, the researchers built a four-marker model and evaluated it with receiver operating characteristic analysis. The apparent area under the curve reached 0.724, but when they applied leave-one-out cross-validation, a stricter test that asks how well the model performs on patients it has never seen, the value dropped to 0.664. That gap between apparent and cross-validated performance is a familiar caution in exploratory biomarker work, a reminder that models fitted to a single dataset tend to overstate their own discriminative power.

The authors are careful about the limitations embedded in their design, and those caveats matter for interpretation. The follicle-count categories came from a historical database in which per-ovary counts were not uniformly available, so the study-specific designation of elevated counts is not equivalent to the current criterion for polycystic ovarian morphology, which requires at least 20 follicles in one ovary. Reliable reclassification to modern definitions was not possible. The isolated menstrual dysfunction group contained only two women and was retained for descriptive purposes alone. Crucially, the metabolic findings were exploratory, were not adjusted for multiple comparisons, and have not been externally validated, so they should be read as hypothesis-generating rather than settled fact.

Those constraints aside, the central message is clinically resonant. In a cohort of women seeking fertility care, anti-Müllerian hormone emerged as the single strongest adjusted correlate of ovarian and menstrual categorization, while the androgens that dominate diagnostic checklists contributed little independent information. If the findings hold up in prospective, appropriately powered studies using current morphology criteria, they could nudge the field toward a reordering of priorities, in which follicular output, read through anti-Müllerian hormone, becomes the anchor biomarker, and androgens, whether classical or 11-oxygenated, are reinterpreted mainly through their relationship with metabolic health rather than ovarian anatomy.

For patients, the practical takeaway is one of measured expectations. This study does not rewrite diagnostic criteria, and its retrospective, single-center design limits generalizability. But it adds weight to a growing view that polycystic ovary syndrome’s heterogeneity is not just clinical noise, it reflects distinct biological axes, with the follicular compartment and the metabolic compartment each carrying their own hormonal signatures. Disentangling those axes, the Beijing team suggests, may ultimately be the key to matching patients with the treatments most likely to help them.

Subject of Research: Hormonal profiling across ovarian and menstrual categories in polycystic ovary syndrome

Article Title: Exploratory analysis of anti-Müllerian hormone and androgen profiles across historical ovarian–menstrual categories defined by antral follicle count in an infertility-associated reproductive endocrine cohort

Article References: Zhang, X., Xu, H., Xiong, X., Zhao, R., Zhao, L., Li, R., & Yang, L. (2026). Exploratory analysis of anti-Müllerian hormone and androgen profiles across historical ovarian–menstrual categories defined by antral follicle count in an infertility-associated reproductive endocrine cohort. Journal of Ovarian Research. https://doi.org/10.1186/s13048-026-02301-9

Image Credits: AI Generated

DOI: 10.1186/s13048-026-02301-9

Keywords: anti-Müllerian hormone, polycystic ovary syndrome, 11-oxygenated androgens, testosterone, androstenedione, antral follicle count, steroid profiling, reproductive endocrinology, insulin resistance, infertility, ovarian morphology, mass spectrometry

A new retrospective study from Kocaeli City Hospital in Türkiye now offers one of the clearest quantitative answers to that question. Writing in BMC Endocrine Disorders, endocrinologists Ayşe Merve Ok Kurt and Seda Erem Basmaz analyzed the records of 438 women aged 18 to 35 who had been evaluated for suspected PCOS or related clinical features and who had complete measurements of total testosterone, androstenedione, and DHEAS. Rather than simply observing which patients had abnormal values, the researchers designed their analysis as a formal experiment in classification. They constructed four nested definitions of hyperandrogenism and asked, at each step, how many additional women crossed the threshold for the biochemical criterion and how many saw their overall diagnosis or Rotterdam phenotype change as a result.

The four models formed a deliberate hierarchy. Model 1, the pragmatic baseline, defined hyperandrogenism as documented hirsutism, elevated total testosterone, or both—precisely the elements most clinicians can reliably capture in routine practice. Model 2 added elevated androstenedione to the baseline, Model 3 added elevated DHEAS, and Model 4 combined both additional androgens. Under each definition, the team recalculated whether each woman met the hyperandrogenism criterion, whether she fulfilled the criteria for a PCOS classification, and which of the Rotterdam phenotypes she would be assigned. Statistical comparisons relied on Cochran’s Q test for the overall differences across models and exact McNemar tests with Holm adjustment for pairwise comparisons, a rigorous approach that guards against false positives when multiple related tests are performed on the same patients.

The headline finding is a tale of two magnitudes. Broadening the biochemical definition with androstenedione and DHEAS raised the proportion of women classified as hyperandrogenic from 86.8 percent under the baseline definition to 93.2 percent under the fully expanded definition—a substantial jump in a criterion that sits at the center of the syndrome’s identity. Yet the effect on the overall diagnosis of PCOS was far more modest, climbing from 72.8 percent to 75.6 percent. That asymmetry makes biological sense: hyperandrogenism is only one of several routes into a PCOS diagnosis, and many women who lack the biochemical marker still qualify through oligo-ovulation and polycystic ovarian morphology. The additional androgens, in other words, confirm and sharpen the androgen excess dimension of the syndrome more than they expand its overall footprint.

Where the expanded testing did matter materially was in phenotype assignment. Under the most inclusive definition, 24 women—5.5 percent of the cohort, with a 95 percent confidence interval of 3.7 to 8.0 percent—were assigned to a different Rotterdam phenotype than they would have been under the baseline definition. Because Rotterdam phenotypes are increasingly used to stratify patients in research and to tailor clinical counseling, a shift of this size in roughly one woman in eighteen is not trivial. The two additional hormones, however, were not equal contributors. Adding androstenedione alone changed phenotype assignment in 4.3 percent of women, compared with just 1.8 percent for DHEAS alone, a difference that reached statistical significance at p equals 0.027. Consistent with that pattern, isolated elevation of androstenedione was markedly more common in the cohort than isolated elevation of DHEAS, occurring in 14.4 percent versus 4.1 percent of participants.

Perhaps the most provocative results emerged from an exploratory subgroup analysis. The researchers identified 58 women who had neither documented hirsutism nor elevated total testosterone—women who, under conventional biochemical thinking, would show no laboratory evidence of androgen excess at all. Yet nearly half of them, 48.3 percent, had elevated androstenedione or DHEAS when those hormones were measured. In this subgroup, the expanded definition reclassified 20.7 percent of women from non-PCOS to PCOS, with a confidence interval spanning 12.3 to 32.8 percent, and changed phenotype assignment in 41.4 percent. The authors are careful to frame these figures as exploratory: the subgroup was small, the analyses were not powered for definitive inference, and the findings need confirmation in larger, prospective cohorts. Even so, they point to a population of women whose androgen excess is invisible to the standard testosterone-based screen but detectable with a broader panel.

The technical backbone of the study deserves attention in its own right. Measuring androgens accurately is notoriously difficult: testosterone circulates at low concentrations in women, and immunoassays—the workhorses of many clinical laboratories, including the chemiluminescence and electrochemiluminescence platforms referenced in the study—are known to lose precision at exactly those concentrations. Liquid chromatography–tandem mass spectrometry, or LC–MS/MS, offers superior specificity but is not universally available. Sex hormone-binding globulin further complicates the picture, since it binds testosterone in the circulation and makes total testosterone an imperfect proxy for biologically active hormone, which is precisely why guidelines increasingly favor free testosterone when it can be measured reliably. The authors acknowledge that their pragmatic baseline deliberately used documented hirsutism and total testosterone rather than a full guideline-concordant workup, a choice that maximizes real-world applicability but also limits how directly the results map onto ideal diagnostic practice.

That limitation shapes the study’s most important caveat: quantifying reclassification is not the same as proving diagnostic benefit. The 5.5 percent of women whose phenotypes changed, and the women in the exploratory subgroup who moved from non-PCOS to PCOS, may have been more accurately classified—or they may simply have been labeled differently without any improvement in predicting their actual clinical outcomes, from metabolic risk to fertility. The authors state this explicitly, noting that their findings quantify the classification impact of broader androgen assessment but do not establish improved diagnostic accuracy. They call for prospective studies that incorporate guideline-recommended free testosterone assessment and follow clinical outcomes over time, so that the field can learn whether androstenedione- and DHEAS-driven reclassifications identify women with genuinely different disease trajectories or treatment responses.

For clinicians, the practical message is a measured one. The study suggests that androstenedione, more than DHEAS, carries real informational weight beyond testosterone in women being evaluated for PCOS, and that in a minority of patients—particularly those with suspected hyperandrogenic features but normal testosterone—measuring additional androgens may meaningfully alter both diagnosis and phenotype. For researchers, the work provides a template: nested, prespecified definitions combined with paired statistical testing offer a clean way to measure the incremental value of any proposed addition to a diagnostic workup. And for the millions of women navigating a condition whose boundaries have been debated since the Rotterdam criteria were first articulated, the study is a reminder that diagnostic categories in endocrinology are not fixed walls but thresholds drawn through evidence—thresholds that careful, quantitative studies like this one continue to refine, one hormone and one reclassified patient at a time.

Subject of Research: The incremental diagnostic value of androstenedione and DHEAS testing in the classification of polycystic ovary syndrome in women

Article Title: Incremental classification yield of androstenedione and DHEAS in women evaluated for polycystic ovary syndrome: a retrospective study

Article References: Kurt, A. M. O., & Basmaz, S. E. (2026). Incremental classification yield of androstenedione and DHEAS in women evaluated for polycystic ovary syndrome: a retrospective study. BMC Endocrine Disorders. https://doi.org/10.1186/s12902-026-02551-0

Image Credits: AI Generated

DOI: 10.1186/s12902-026-02551-0

Keywords: polycystic ovary syndrome, hyperandrogenism, androstenedione, DHEAS, testosterone, Rotterdam criteria, PCOS diagnosis, hirsutism, endocrinology, phenotype reclassification, reproductive hormones, retrospective study