Polycystic ovary syndrome is the most common endocrine disorder in women of reproductive age, and its name is one of the least helpful in medicine.
It is not really about cysts. The structures seen on ultrasound are not cysts but small immature follicles — the consensus definition counts follicles measuring 2 to 9 mm [1] — present in larger numbers than usual — and the ovarian appearance is neither necessary nor sufficient for the diagnosis. What defines the condition is a triad: irregular or absent ovulation, signs of excess androgens (hirsutism, acne, or a raised blood measurement), and often insulin resistance sitting underneath both.
That third element is the part most often missed. PCOS is filed under gynaecology and presented as a fertility problem, but the metabolic disturbance is present in lean women as well as heavy ones [2], carries a substantial risk of type 2 diabetes [3], raises the risk of fatty liver [4], and does not go away when a woman stops wanting to conceive. It also carries a real psychological burden that professional societies have had to issue a position statement about [5] [6].
Prevalence depends entirely on which definition you use — a systematic review put it at 6 percent under the strict 1990 NIH definition and 10 percent under both the broader Rotterdam criteria and the Androgen Excess Society criteria [7]. On any of those numbers this is a very common condition, and it is widely under-recognised.
Start here: the two halves of the syndrome
The reproductive half is straightforward to describe. Follicles begin to develop but do not complete the process; ovulation is irregular or absent; menstrual cycles are long or missing. The ovary accumulates a large number of small antral follicles, which is what the ultrasound sees.
The metabolic half is the one that explains the rest. Insulin resistance in PCOS was demonstrated with the euglycaemic clamp — the reference method — in 1989, and the crucial finding was that insulin-stimulated glucose utilisation was significantly reduced in both obese and non-obese women with PCOS compared with body-composition-matched controls [2]. Obesity worsens it, but it is not the cause of it. That result has been revisited and refined repeatedly since [8] [9], and it is why a woman of normal weight can have the full syndrome.
The link between the two halves is the subject of the model below, and it is unusually clean: insulin raises free androgen — not by making the ovary produce more, but by removing the protein that keeps androgen bound and inactive.
Pillar 1: measurement and diagnosis
The Rotterdam criteria
The 2003 Rotterdam consensus requires two of three features: oligo- or anovulation; clinical or biochemical hyperandrogenism; and polycystic ovarian morphology on ultrasound [10] [11]. "Two of three" is the source of most of the confusion around this diagnosis, because it means four different phenotypes all carry the same label, and they do not carry the same metabolic risk [12]. The Androgen Excess and PCOS Society subsequently argued that androgen excess should be required, not optional [13], and the debate has never fully closed [14].
Ultrasound morphology has its own consensus definition — at least 12 follicles measuring 2–9 mm, or an ovarian volume above 10 cm³, in either ovary [1]. That threshold was set before high-resolution transvaginal probes became routine, and it is a soft criterion: a great many women without PCOS meet it, and polycystic ovaries were documented decades ago in women with anovulation or hirsutism alone [15].
In adolescents the ultrasound criterion is dropped altogether, because multifollicular ovaries and irregular cycles are both normal features of puberty [16] [17]. This is a genuine diagnostic hazard in the other direction: over-diagnosing a teenager with a lifelong metabolic condition on findings that are developmentally normal.
It is a diagnosis of exclusion
Rotterdam explicitly requires excluding conditions that mimic the picture [10]. The main ones are thyroid disease, hyperprolactinaemia, and non-classic congenital adrenal hyperplasia — most often 21-hydroxylase deficiency, which produces androgen excess and menstrual disturbance by a completely different route and has its own management [18] [19] [20]. A prospective series of 1,281 consecutive patients presenting with possible androgen excess put numbers on how often the alternatives turn up: among the 873 with confirmed androgen excess, non-classic 21-hydroxylase-deficient adrenal hyperplasia accounted for 1.6 percent, classic congenital adrenal hyperplasia 0.6 percent, and an androgen-secreting tumour 0.2 percent [21]. Individually rare, collectively worth excluding, and each one mismanaged if assumed to be PCOS.
What to measure, and why the metabolic screen matters
For androgens, a total testosterone and an SHBG allow calculation of the free androgen index — and as the model below shows, SHBG is not a technicality but the mechanism. For hirsutism the assessment is clinical [22].
The metabolic screen is the part most often skipped. An oral glucose tolerance test in 122 well-characterised women with PCOS found glucose tolerance abnormal in 45 percent of them: 35 percent had impaired glucose tolerance and 10 percent already had diabetes [3]. That study also found something that argues directly against the cheaper test: among the women with impaired glucose tolerance, fasting glucose was a poor predictor of the two-hour level (r = 0.25, not significant), even though across the whole cohort the two correlated strongly (r = 0.76) [3]. A normal fasting glucose is not reassurance in this population. Independent cohorts reached the same conclusion about diabetes risk [23], and the associated lipid and metabolic-syndrome burden is well documented [24] [25] [26].
Centerpiece: a simple simulatable model of the insulin-androgen loop
Most androgen in the circulation is not free. It is bound to sex hormone-binding globulin, made by the liver, and only the unbound fraction acts on tissue. So free androgen is set by two numbers — how much androgen is produced, and how much SHBG is available to bind it:
free androgen ≈ total androgen / SHBG
The question is what sets SHBG. Two independent experiments answer it, and together they make the loop quantitative.
The cell line. In human hepatoma (Hep G2) cells, insulin at 10⁻⁸ mol/L reduced SHBG production from 65.0 to 46.8 nmol per 10⁶ cells — a 28 percent suppression — and also blocked the stimulation of SHBG by oestradiol and thyroxine [27]. Insulin directly turns the liver's SHBG production down.
The clinical experiment. Six obese women with PCOS were studied in two steps [28]. First, a GnRH agonist shut down ovarian steroid production: total testosterone fell from 1.72 to 0.32 nmol/L, non-SHBG-bound testosterone from 104 to 19 pmol/L — and SHBG did not change, 18.8 to 17.8 nmol/L. Then, with the steroids still suppressed, diazoxide was used to suppress insulin release (glucose-tolerance-test insulin area under the curve fell from 262 to 102 nmol·min/L), and SHBG rose by 32 percent.
The design is what makes it convincing. Step one removes androgen and SHBG does not move; step two removes insulin and SHBG moves substantially. Androgen is not setting SHBG. Insulin is.
Two checks, and one of them uses a number the arithmetic was never given.
The first is cross-experiment. If insulin suppresses SHBG production by 28 percent, then removing insulin should raise it by 65.0/46.8, or 38.9 percent. The clinical answer was 32 percent. A hepatoma cell line in 1988 and six women given diazoxide in 1991 are entirely independent, and the cell line was never fitted to the clinical result — yet they agree in direction and to within seven percentage points, with the in-vitro effect the larger and cleaner one, as you would expect of an isolated system.
The second is internal, and stronger. The 1991 study measured SHBG directly and reported it unchanged during the GnRH-agonist step. But it also reported total and non-SHBG-bound testosterone at the same two timepoints — and if SHBG really was unchanged, free testosterone must have fallen in exact proportion to total testosterone, a constraint the SHBG assay never enters. Total testosterone fell 81.4 percent; free testosterone fell 81.7 percent. The free-per-total ratio was 60.5 before and 59.4 after, a difference of 1.8 percent. The steroid measurements independently confirm the SHBG measurement.
The consequence is the right-hand panel. Holding total androgen production completely constant, a 32 percent rise in SHBG lowers free androgen by 1 − 1/1.32, or 24 percent. Nothing about the ovary changed. The hyperbola is steep at low SHBG, which is why the same absolute gain in SHBG matters far more to a woman who starts low — that is, to a woman who is insulin resistant.
The teaching point is that this closes a loop. Insulin resistance produces hyperinsulinaemia; hyperinsulinaemia lowers SHBG and so raises free androgen; and hyperandrogenism is itself associated with worse metabolic outcomes, including fatty liver [4] [29]. The two halves of the syndrome reinforce each other, which is why improving insulin sensitivity — by weight loss, by exercise, by metformin — untangles much of the picture at once rather than treating one symptom.
Three honest limits. Six women is a small experiment, however elegant its design, and they were obese; the SHBG effect in lean PCOS is not established by this study. Diazoxide is a research tool, not a treatment, and it worsened glucose tolerance while it suppressed insulin — the experiment isolates a mechanism, it does not model a therapy. And insulin's effect on SHBG is only one arm of the loop: insulin also acts directly on the ovary, stimulating testosterone biosynthesis by thecal cells from women with PCOS through its own receptor [30] and functioning as what one review calls a co-gonadotropin modulating ovarian steroidogenesis [9] — an arm this figure does not represent at all, so the 24 percent shown is a lower bound on what insulin-lowering does to free androgen, not the whole of it.
Pillar 2: treatment, which depends on what the woman wants
There is no single treatment for PCOS, because the syndrome presents three different problems and a given woman may care about one, two or all three. The international evidence-based guideline runs to 166 recommendations for exactly this reason [31] [32] [33], as does the Endocrine Society's [34]. The rational way to organise it is by goal.
Foundation: metabolic
Lifestyle modification is the recommended first step, and a randomised trial actually tested it against the alternative. Overweight and obese infertile women with PCOS were assigned to 16 weeks of a continuous oral contraceptive, a lifestyle programme targeting 7 percent weight loss, or both, before standardised ovulation induction [35]. Trials of that design are rare and this is why the recommendation has any evidence behind it at all.
GLP-1 receptor agonists are the most substantial recent addition. Exenatide and metformin were compared alone and in combination for menstrual cyclicity in PCOS [36]; liraglutide combined with metformin produced significant weight loss [37]; exenatide has been studied for weight, metabolic parameters and pregnancy in overweight women with PCOS [38]; a systematic review and meta-analysis has compared the class directly against metformin [39]; and the rationale for the class in PCOS and infertility has been set out explicitly [40]. Given that the core lesion is insulin resistance, a drug class that treats insulin resistance and obesity together is well matched to the mechanism — though the evidence base in PCOS specifically remains far thinner than in diabetes or obesity, and the trials to date are small and short.
Goal: regular cycles and less androgen
Combined oral contraceptives are first-line for menstrual regulation and hyperandrogenism [17] [41]. They also raise SHBG — oestradiol stimulates hepatic SHBG production in the same cell-line experiment that showed insulin suppressing it [27] — which is the model's lever pulled from the liver rather than from insulin. Regular withdrawal bleeding additionally protects the endometrium, which matters because chronic anovulation leaves it under unopposed oestrogen [34].
Anti-androgens, principally spironolactone, are added for hirsutism when the contraceptive alone is insufficient, and are used alongside reliable contraception [17] [41] [42].
Goal: insulin resistance
Metformin was shown thirty years ago to reduce hyperinsulinaemia, insulin resistance and hyperandrogenaemia in PCOS [43] — which is the loop above, run backwards. It is a metabolic drug used for a metabolic problem, and the evidence for it is best understood in that frame rather than as a fertility drug, for reasons the next section makes plain.
Goal: pregnancy
This is where the evidence is strongest and where practice has actually changed twice.
First, metformin was displaced. A trial randomised 626 infertile women with PCOS to clomiphene, extended-release metformin, or both. Live-birth rates were 22.5 percent with clomiphene, 7.2 percent with metformin, and 26.8 percent with the combination — clomiphene clearly superior to metformin, and the combination not significantly better than clomiphene alone [44]. Metformin also had a lower conception rate among women who ovulated (21.7 versus 39.5 percent), so its weakness was not merely a failure to induce ovulation.
Then clomiphene itself was displaced. A double-blind trial randomised 750 women with PCOS to letrozole or clomiphene for up to five cycles. Letrozole produced more cumulative live births — 27.5 versus 19.1 percent, a rate ratio of 1.44 (95% CI 1.10–1.87) — and a higher cumulative ovulation rate, 61.7 versus 48.3 percent of treatment cycles [45]. Letrozole is now first-line for ovulation induction in PCOS. There were four major congenital anomalies in the letrozole group against one with clomiphene, a difference that was not statistically significant (P = 0.65) but which the trial reported plainly and which is worth stating rather than omitting.
Beyond ovulation induction the pathway continues to gonadotropins and IVF [46].
One thing worth telling patients: cycles tend to become more regular with age. In 205 previously oligo- or amenorrhoeic women followed to age 30 and beyond, there was a highly significant trend towards shorter cycle length with increasing age, independent of BMI, weight loss, hirsutism, prior treatment or ethnicity [47].
Long-term risk is the part that outlasts the fertility question
Type 2 diabetes [3] [23], metabolic syndrome [24], adverse lipids [25] [26], subclinical cardiovascular disease and coronary calcification [48] [49] [50], and non-alcoholic fatty liver disease [51] all cluster with PCOS. The fatty liver association has been quantified at scale: in 63,120 women with PCOS matched to 121,064 controls on age, BMI and location, the hazard ratio for NAFLD was 2.23 (95% CI 1.86–2.66), and in a separate cohort a testosterone above 3.0 nmol/L was itself associated with increased NAFLD risk [4]. Matching on BMI is the important detail — this is not simply obesity. A professional consensus exists on cardiovascular risk assessment in this population [52], and the honest reading is that the reproductive complaint that brings a woman in is frequently the least of her long-term problems.
Pillar 3: what is unresolved
Which definition is right. Rotterdam's "two of three" produces phenotypes with materially different metabolic risk [12], and whether androgen excess should be mandatory remains argued [13] [14]. The prevalence of the condition changes by nearly a factor of two depending on the answer [7].
Anti-Müllerian hormone as a substitute for ultrasound. AMH is markedly raised in PCOS — 47.1 against 20.8 pmol/L in one study [53] — and it is produced by granulosa cells [54]. But the same study found the AMH-per-follicle ratio essentially identical in PCOS and controls (4.8 versus 4.8), so AMH was tracking the number of small follicles rather than a per-follicle abnormality, which is precisely what the ultrasound criterion counts. It is a blood test standing in for an imaging count, not an independent marker — and it is lowered by obesity, which complicates its use in the women who most need assessing [55].
Cause. Insulin resistance, ovarian steroidogenic abnormality and neuroendocrine dysfunction are all present, and which is primary is unresolved [9] [14] [56]. The condition was described in 1935 [57] and the argument is not settled ninety years later.
Care that matches the disease. PCOS is diagnosed in gynaecology, driven by metabolism, and carries psychological consequences that a position statement had to be written about [5] [6]. Very little of the care system is organised to deliver all three at once, and that — rather than any missing drug — is probably the largest available improvement.
Dig deeper in lmmol
The metabolic engine of PCOS is shared with the rest of this collection. Insulin resistance is the common thread to obesity, where the GLP-1 evidence is far more developed than it yet is in PCOS [40] [39], and to type 2 diabetes and glycemic control, which nearly half of women with PCOS are already on the path towards [3]. The liver appears twice over in this review — as the organ that makes SHBG and so sets free androgen [27], and as the organ at more than double the risk of fatty change even after matching for BMI [4] — which makes fatty liver disease the closest sibling of the three. Thyroid disease and hyperprolactinaemia must be excluded before the diagnosis is made at all [10], so thyroid disease is the differential rather than the comorbidity. The cardiovascular clustering connects to hypertension [52], and the psychological burden to depression [5]. The full collection is at health.