Original editorial illustration, not a figure from the study. Source: DOI 10.1186/s12902-026-02551-0.
Does a broader hormone panel lead to a better diagnosis? Intuitively, more results seem to mean more knowledge. Yet learning about an additional marker, changing a classification and improving treatment are different things. A new study of androgen assessment in PCOS illustrates that distinction. Its findings are worth discussing with a clinician, but should not become a shopping list of tests that every patient is expected to order independently.
What was studied—and what was not?
Ayşe Merve Ok Kurt and Seda Erem Basmaz reviewed records of women assessed for PCOS. Their final analysis included 438 women aged 18–35 years. This was a retrospective, single-centre study: the researchers used existing records rather than randomly assigning patients to different diagnostic pathways and following the consequences. The article appeared in BMC Endocrine Disorders on 11 September 2026.
The same women were classified using different definitions of androgen excess. The starting definition used documented hirsutism and/or elevated total testosterone. The researchers then added elevated androstenedione, DHEAS, or both. The question was how broadening the available information changed classification. The study did not test whether ordering a broader panel improves quality of life or treatment effectiveness.
This matters when interpreting the title. “Incremental classification yield” is not another name for “greater diagnostic accuracy”. The former describes a change produced by applying a set of rules. The latter would require evidence that the new assessment more often reaches the correct diagnosis. Without an appropriate reference, counting additional classifications cannot settle that question.
Two numbers that should not be confused
Under the broader definition, the number of women classified as having PCOS changed from 319 to 331: an increase of 12. Separately, 24 women changed phenotype assignment—the combination of features used to classify the syndrome. That does not mean 24 new diagnoses. Some changes concerned women who had already been classified as having PCOS.
Phenotype reassignment affected 5.5% of the analysed group, with a reported 95% confidence interval of 3.7–8.0%. These figures describe this particular cohort. They are not the population frequency of “missed cases”, or the probability that an additional test will change an individual reader’s diagnosis. The underlying results appear in Tables 2 and 3 of the full paper.
The safest reading is literal: the definition was broadened, and the researchers counted how many classifications changed. We should not add a story in which previous clinicians necessarily missed a disease. Nor should we assume the opposite—that every new classification represents overdiagnosis. The study provides no independent resolution between those interpretations. Responsible communication keeps that boundary visible.
The central limitation: an incomplete starting assessment
SHBG and free testosterone data were unavailable. The baseline model therefore cannot be presented as a complete first-line biochemical assessment. This is central to the comparison, not a minor footnote. Adding information to a limited dataset may produce a different gain from adding the same information to a complete starting assessment.
This does not mean androstenedione or DHEAS results are irrelevant. It means the experiment cannot answer every practical question about their place in diagnosis. In particular, it does not establish their additional value beyond a complete assessment of total and free testosterone. It should not be described as a direct comparison between two fully equivalent diagnostic pathways.
For a patient, this is a prompt for discussion, not an instruction to build a personal testing panel. Ask what information has already been collected and which specific uncertainty remains. That is more useful than comparing the number of items on a laboratory request. A longer list, by itself, does not show that the diagnostic plan better matches your situation.
Why is a larger classification count not enough?
Consider a general classification system that assigns a category when a condition is met. Adding another way of meeting that condition can move some people into the category. This can follow from the structure of the rule without automatically validating the rule. This example explains the logic of interpretation; it is not another study finding or a model for calculating patient benefit.
To call a classification more accurate, we need a reliable way to establish which assessment is correct. To claim clinical utility, we must also demonstrate its relevance to decisions and treatment outcomes. Those questions may have different answers. A study can usefully describe how classification varies without establishing that everyday clinical practice should change.
Statistical significance does not bridge this gap. A statistical test can provide evidence against a specified no-difference hypothesis within the analysis. It does not, on its own, tell us whether the observed change benefits an individual. Discussions of results should therefore distinguish detecting a difference from establishing its clinical value.
Limitations beyond the headline result
The study used immunoassays rather than LC–MS/MS. Hirsutism scoring was not consistent, and ovarian morphology assessment relied on routine transabdominal ultrasound reports. There was no independent external validation or longitudinal evaluation of clinical outcomes. These details limit transfer to other settings. They do not automatically invalidate the work, but they define how far its conclusions can extend.
In practice, this calls for caution around claims that a new panel “finds more true cases”. That wording contains a conclusion the study did not demonstrate. A phenotype shift should not be presented as proof of a worse prognosis, greater complication risk or a need for more intensive treatment. A new category and a new health situation are not interchangeable.
The retrospective design also reminds us that existing records were not created specifically to answer every question now being asked. Where a key element is missing, we cannot reconstruct it from what seems most likely. Patient-facing communication is better served by saying “this was not assessed” than by filling an evidence gap with a promising conclusion.
Questions to bring to a consultation
The following questions are suggestions for preparing a discussion, not a diagnostic algorithm evaluated in this study. First: what is the purpose of this additional test? The answer should identify a particular uncertainty. Second: is my initial assessment complete for my circumstances? The aim is not to order every available test, but to establish what the clinician needs to make a decision.
Third, ask about consequences: what would change if the result were elevated, and what would change if it remained within the reference range? Also ask whether reclassification would alter management or simply refine the description. This distinction helps prevent every new term in a medical record from being interpreted as a new problem requiring treatment.
Fourth: what are the limitations of this result? That question applies both to an individual measurement and to a publication cited in support of testing. Ask whether the study population resembles your circumstances and whether the research examined treatment benefits or classification alone. Asking about the boundaries of evidence is part of an informed conversation, not a challenge to the clinician’s competence.
This article should not be used to stop medication, change doses or conclude that a previous diagnosis was wrong. It cannot resolve an individual case. If a new result causes concern, discuss it alongside the existing records instead of interpreting it in isolation or comparing it with someone else’s result. The objective is a coherent assessment, not the largest possible collection of laboratory measurements.
What research would help us take the next step?
A useful next assessment would start from a complete baseline and then examine the contribution of additional information. We need to know not only whether a category changes, but whether decisions become more appropriate and lead to meaningful outcomes. From a patient’s perspective, the next step is precisely that: explaining what changes in care, rather than only in a table.
Until then, the message should remain balanced. It is not “never test additional androgens”. Nor is it “every patient needs an expanded panel”. The study shows how the information included affects classification and why evaluating laboratory testing requires close attention to the comparator. It offers a useful methodological lesson, not a ready-made universal recommendation.
The main point to remember is that more people meeting broader criteria does not automatically mean more correctly diagnosed disease. Before ordering another test, establish its purpose and potential effect on management. An increased classification count should not replace that conversation or become a promise of better treatment.
This principle can also help readers assess online testing offers. When a panel description cites research, check whether the promise matches what was actually measured. Did the investigators assess patients’ health after management changed, or reapply rules to existing results? Was the baseline assessment complete? You do not need to resolve every medical detail yourself to recognise when a cautious finding has become a much broader marketing claim. If uncertain, bring the source to a consultation and discuss it alongside your symptoms and previous findings.
Source and transparency
Kurt AMO, Basmaz SE. Incremental classification yield of androstenedione and DHEAS in women evaluated for polycystic ovary syndrome: a retrospective study. BMC Endocrine Disorders, 2026. DOI: 10.1186/s12902-026-02551-0. The reviewed version is an Article in Press. No specific grant was received; the authors declare no competing interests. The journal’s quartile remains unverified. The source article is available under CC BY-NC-ND 4.0. No source figures or tables have been reproduced. This independently written educational commentary is not individual medical advice.