Metabolic & Cardiometabolic

PCOS increases type 2 diabetes

Strong support Metabolic & Cardiometabolic 💰 Industry COI noted

RefutedContestedStrong support

consensus score 1.00

📅 Last reviewed: 2026-07-15 ⓘ

Evidence ladder

How far up the ladder this claim has climbed. A high consensus on a low rung means "consistent so far," not "proven in people."

Top evidence so far: All trials, pooled (Meta-analysis)

MechanismIn-vitroAnimalObservationalRCTMeta-analysis

How the studies fall

6 support 0 contradict 0 tested null 0 mixed · 6 sources, 6 independent groups

What the evidence shows

PCOS markedly increases type-2-diabetes / impaired-glucose-tolerance risk (RR ~3-4 across four meta-analyses). The BMI-independence of that risk is now CONTESTED, not settled (re-read 2026-08-25): the only meta-analysis purpose-built to test it (Anagnostis 2021) found the risk elevated "only if they are obese/overweight" (non-obese RR 1.62, 95% CI 0.14-18.

The evidence (6)

Source	Grade	Stance	Quality	Finding
Wekker 2020 · Hum Reprod Update	meta-analysis	supports	moderate	SR+MA of longitudinal observational cohort studies (PROSPERO CRD42015023765, PRISMA, OVID MEDLINE+EMBASE from inception to 2 September 2019, dual independent selection/quality-assessment/extraction, Newcastle-Ottawa). Full-text re-read 2026-08-25 from the delivered CC-BY version of record plus the official supplementary tables. SUBSET DENOMINATOR: the headline RR is NOT over the paper's 23 included studies - 'Thirteen studies ... were included in the meta-analysis for T2D.' (4 prospective, 9 retrospective). Summing Supplementary Table SII gives that pool its own denominator, 55,124 women with PCOS vs 228,514 without (prospective 423 vs 3,655; retrospective 54,701 vs 224,859) - a total the paper itself never states. Result: T2D 5.9% vs 2.0%, RR 3.00 (95% CI 2.56-3.51), I2 83%. DIRECTION IS ROBUST across every cut the authors ran: prospective-only RR 2.26 (1.74-2.94, I2 0%), retrospective-only RR 3.29 (2.77-3.91), outlier-excluded RR 3.04 (2.77-3.35, I2 falls 83%->36%), high-quality-studies-only RR 3.07 (2.48-3.80), and visual inspection of the funnel plot 'did not indicate publication bias (Fig. 2d)'. QUALITY LOWERED high -> moderate on what the paper established about ITSELF, not on its conduct, which is good: the authors say their own pooled effect is likely inflated - 'The heterogeneity is likely to be based on an overestimation of the effect in the retrospective studies' - and the retrospective studies supply 54,701 of the 55,124 PCOS women in this pool (99.2%); and 'The most prevalent high risk of bias was because of studies not indicating whether the outcome of interest, such as HT and diabetes, was already present at the start of the study', i.e. prevalent T2D cases may be counted as incident. PCOS ascertainment across the pool is heterogeneous (Rotterdam in 7 studies, NIH in 3, ICD/READ codes in 5, symptom-based in 8) and the authors call this clinical heterogeneity 'a major limitation of our study'. SCOPE NOTE on the claim's ' independent of BMI': this paper does not support that clause and disclaims it - 'We were unable to assess the extent to which increased cardiometabolic risk among women with PCOS was independent of obesity'; they explicitly refrained from meta-regression on BMI. The single largest contributor to this pool, Morgan 2012 (21,734 PCOS vs 86,936 controls), reports RR 3.12 unmatched but aHR 1.12 [1.11-1.13] in a BMI-matched sample (Supplementary Table SII), and two further contributors lost significance on BMI adjustment (Wild 2000 aOR 2.2 [0.9-5.2]; Talbott 2007 aRR 2.4 [0.96-5.92]). Those are per-study data, not a pooled BMI-matched estimate by Wekker. STANCE HELD supports: the SPO's object is type 2 diabetes and every sensitivity cut stays above RR 2; the BMI material is a scope-clause question, not a disconfirmation on the object. DUPLICATE-POPULATION FLAG (flagged, not fixed - rubric axis 7): this pool contains the Northern Finland Birth Cohort 1966 study that also votes separately as ev-s28031324 (Table I gives its title, Human Reproduction, Finland 1966-2012, n PCOS 279 / n control 1577; Supplementary Table SII gives its contributed row, N PCOS 185 / N non-PCOS 1170, 23 vs 50 events, aOR 2.51 [1.38-4.56]). Wekker labels it 'Ollila et al., 2016' in the T2D study list while its own reference list gives Hum Reprod 2017;32:423-431, so an overlap check keyed on author-year would miss it. Funding: 'This work was supported by the Dutch Heart Foundation (2013T085).' COI: J.S.E.L. declares grants/consultancy from Ansh Labs, Danone, Euroscreen, Ferring, Merck Serono, Titus Healthcare and ZonMW; the UMCG department received an unrestricted educational grant from Ferring; 'All other authors have nothing to declare.'
Ollila M-M, West S, Keinanen-Kiukaanniemi S, Jokelainen J, Auvinen J, Puukka K, Ruokonen A, Jarvelin M-R, Tapanainen JS, Franks S, Piltonen TT, Morin-Papunen L 2017 · Hum Reprod	observational	supports	moderate	Prospective general-population birth cohort (Northern Finland Birth Cohort 1966, base n = 5889), abstract-grade re-read 2026-08-25 (efetch; PMC5260860 exists but the paper is NOT in the PMC open-access subset). PCOS was SELF-REPORTED — 'Women reporting both oligoamenorrhoea and hirsutism at age 31 years and/or diagnosis of PCOS by 46 years were considered as women with PCOS (n = 279). Women without any symptoms at 31 years and without PCOS diagnosis by 46 years were considered as controls (n = 1577).' Outcome ascertained at age 46 by 2-h OGTT in 2780 women plus national drug and hospital-discharge registers. SUBSET DENOMINATOR: the decisive estimate is a within-BMI-class comparison whose own n is NOT reported in the abstract — 279 and 1577 span both BMI strata and are not the denominator of the OR. Result: 'PCOS per se significantly increased the risk of T2DM in overweight/obese (BMI >= 25.0 kg/m2) women with PCOS when compared to overweight/obese controls (odds ratio: 2.45, 95% CI: 1.28-4.67).' The authors' condition is part of the finding: 'Normal weight women with PCOS did not present with an increased risk of prediabetes or T2DM' — reported with NO estimate, NO confidence interval and NO stratum n, so an informative null cannot be distinguished from an underpowered stratum; that is cannot-tell, not a demonstrated null, and is why this row reads supports rather than mixed. The authors' own summary is two-directional on MECHANISM, not on direction of risk: 'The risk of T2DM in women with PCOS is mainly due to overweight and obesity, although these two factors have a synergistic effect on the development of T2DM.' Grade held observational, quality held moderate: strong prospective design, objective OGTT+register outcome, no industry funding ('The authors have no conflicts of interests'), set against an exposure the authors admit is misclassified — 'The diagnosis of PCOS was based on self-reporting, and the questionnaire at 46 years did not distinguish between polycystic ovaries only in ultrasonography and the syndrome. Ovarian ultrasonography was not available to aid the diagnosis of PCOS.' NOTE ON SCOPE: this paper bears directly on the claim's ' independent of BMI' clause and cuts BOTH ways on it (OR 2.45 is labelled by the authors 'PCOS per se' with a 'synergistic effect', arguing for a residual PCOS effect; the normal-weight null and 'mainly due to overweight and obesity' argue against). Whether a clause binds stances is an unresolved rule question (followup 156); this stance grades the SPO only. An erratum exists (Hum Reprod 2017;32(4):968, doi 10.1093/humrep/dex030) whose content is cannot-tell from available text.
Moran LJ, Misso ML, Wild RA, Norman RJ 2010 · Hum Reprod Update	meta-analysis	supports	high	SR+MA, 35 studies selected from 2192 screened, PCOS vs non-PCOS. The endpoint is PREVALENCE, not incidence: DM2 OR 4.43 (95% CI 4.06-4.82) overall. The BMI-matched figure this claim rests on is a SUBSET analysis, OR 4.00 (1.97-8.10), and the BMI-matched DM2 subset's own study count is not stated in the abstract; 35 is the total pool across all three outcomes (IGT, DM2, metabolic syndrome), not the BMI-matched DM2 pool. The authors' conditions are part of the finding: the single incidence study found no significant difference in DM2 incidence, OR 2.07 (0.68-6.30); the single conversion study (normal glucose tolerance to IGT/DM2) gave OR 2.4 (0.7-8.0); and the conclusion is that 'further research is required' on incidence. Conduct is good (Newcastle-Ottawa + Cochrane risk-of-bias, dual independent abstraction, fixed- and random-effects Mantel-Haenszel); heterogeneity and publication-bias testing are not reported in the abstract, which settles nothing either way.
Glintborg 2024 · Hum Reprod	observational	supports	moderate	Nordic national register cohort across Denmark, Finland and Sweden; abstract-grade re-read 2026-08-25 (NCBI efetch, saved as fetched-efetch.txt; Hum Reprod 2024;39(8):1823-1834, doi 10.1093/humrep/deae124). Europe PMC core record probed and saved (fetched-europepmc.json): isOpenAccess=N, inEPMC=N, no PMCID, only a 'Subscription required' DOI link - this source is permanently abstract-grade, not a cache-miss artefact. DENOMINATOR CORRECTED: the prior extract's '99892 women' is the PCOS arm alone; the study is 99,892 women with PCOS AND 446,055 age-matched controls, 545,947 women in total, built from three national cohorts with their own denominators - 'The study cohort included women originating from Denmark (PCOS Denmark, N = 27 016; controls, N = 133 994), Finland (PCOS Finland, N = 20 467; controls, N = 58 051), and Sweden (PCOS Sweden, N = 52 409; controls, N = 254 010).' Median age at entry 28 years; median follow-up in PCOS women 8.5 (4.0-14.8) years in Denmark, 9.8 (5.1-15.1) in Finland, 6.0 (2.0-10.0) in Sweden. Outcome ascertainment: 'T2D was defined according to ICD-10 diagnosis codes and/or filled medicine prescriptions of anti-diabetic medication excluding metformin' - the metformin exclusion matters, since metformin is prescribed for PCOS itself and including it would manufacture cases in the exposed arm. NUMBERS, of which the prior extract carried none: crude HR 4.28 (3.98-4.60) in Denmark, 3.40 (3.11-3.74) in Finland, 5.68 (5.20-6.21) in Sweden; pooled, 'the crude HR for T2D in PCOS was 4.64 (3.40-5.87) and, after adjustment for BMI and education level, the HR was 2.92 (2.32-3.51).' GRADE HELD observational: the paper's 'combined meta-analysis' pools its OWN three national cohorts, it is not a synthesis of the published literature, so it does not earn a meta-analysis grade. 'AMPLIFIED BY OBESITY' RESTATED WITH ITS LIMITS: the prior gloss asserts effect modification the abstract does not quantify. What is reported is a main effect - 'In adjusted regression analyses, BMI >=30 vs <25 kg/m2 was associated with a 7.6- to 11.3-fold risk of T2D' - a range across the three countries whose stratum (within PCOS women, within controls, or pooled) is NOT stated; no PCOS-by-BMI interaction term appears in the abstract, so the size of any amplification is cannot-tell. The authors' own wording is 'the risk is aggravated in women with BMI >=30 kg/m2' and 'A diagnosis of PCOS and BMI >=30 kg/m2 is a high-risk phenotype for a prospective risk of T2D diagnosis across Nordic countries.' SCOPE-CLAUSE NOTE (the claim's ' independent of BMI'): this is one of the few rows in the vault carrying a BMI-adjusted estimate on this claim, and it points FOR residual risk - 4.64 falls to 2.92 on adjustment for BMI and education but stays far from 1 - which cuts against Anagnostis 2021 and with moran-2010. Read it against the authors' own caveat, 'BMI data were not available for all participants', and note that adjustment is not matching. Per followup 156 the scope clause does not bind stances; this stance grades the SPO only, and on the SPO's object (type 2 diabetes) the paper is unambiguously confirmatory. QUALITY LOWERED high -> moderate on what the paper established about ITSELF, not on its design, which is close to the best observational evidence obtainable (prospective, national registers, 546k women, age-matched controls, BMI- and education-adjusted, three countries agreeing in direction): the authors report a bias in their own risk estimate - 'Inclusion of more severe cases of PCOS in the present study design could have lead to an overestimation of risk estimates in our exposed population' - and the adjusted estimate that carries the scope question rests on incomplete data, 'BMI data were not available for all participants.' This is the same call already made on the sibling row ev-s32995872 (Wekker), lowered high -> moderate on the authors' own overestimation statement; all four rows on this claim now read moderate. A counterweight the authors state runs the other way and does not rescue high - 'However, some women in the control group would have undiagnosed PCOS, which would lead to an underestimation of T2D risk in women with PCOS.' Publication-bias testing is not applicable to a single cohort study; registration is 'TRIAL REGISTRATION NUMBER: N/A.' Funding: Region of Southern Denmark, Overlaegeraadet and Odense University Hospital (Denmark); Novo Nordisk Foundation, Finnish Research Council, Sigrid Juselius Foundation, the National Regional Fund, Sakari Alhopuro Foundation and Finnish Diabetes Research Foundation (Finland). COI: 'E.E. has received a research grant from Ferring Pharmaceuticals (payment to institution) and serves as medical advisor for Tilly AB, not related to this manuscript. The remaining authors declare no conflict of interest.' INDEPENDENCE (flagged, not fixed - rubric axis 7): group_id 'glintborg' is NOT an independent voice from group_id 'ollila' on this same claim - M.M. Ollila is second author here and first author of ev-s28031324 (Northern Finland Birth Cohort 1966), and T. Piltonen is on both papers; the vault's truncated author field '[Glintborg]' hides this entirely. Person-level overlap in the Finnish arm is possible though unquantifiable from the abstract, and ascertainment differs (register diagnosis here vs self-reported symptoms plus OGTT there). No duplicate-population problem against the two meta-analysis rows: Wekker searched to 2 September 2019 and Anagnostis to 31 October 2020, both before this 2024 paper, so it is genuinely new evidence on the claim.
Kakoly NS, Khomami MB, Joham AE, Cooray SD, Misso ML, Norman RJ, Harrison CL, Ranasinha S, Teede HJ, Moran LJ 2018 · Hum Reprod Update	meta-analysis	supports	moderate	SR+meta-regression, 40 studies of 4530 screened, NIH-defined PCOS vs non-PCOS, search to June 2016 (PROSPERO CRD42017056524). On meta-analysis of the good/fair-quality subset (the abstract gives no T2DM-specific study count or participant n), T2DM OR 2.87 (95% CI 1.44-5.72); the separate IGT outcome, which is NOT this claim's object, was OR 3.26 (2.17-4.90). Authors' conditions are part of the finding - prevalence was higher with obesity and DOUBLED among studies diagnosing diabetes by self-report or administrative data, and meta-regression found no clear contributor to heterogeneity. BMI-matched subgroups are reported only for the ethnicity difference in IGT, so this paper does not test the claim's BMI-independent scope for T2DM.
Anagnostis P, Paparodis RD, Bosdou JK, et al. 2021 · Endocrine	meta-analysis	supports	moderate	SR+MA of observational studies; PubMed, CENTRAL and Scopus searched to 31 October 2020. 23 studies met eligibility (319,780 participants; 60,336 PCOS, 8,847 T2DM cases). PCOS vs non-PCOS: RR 3.45 (95% CI 2.95-4.05, p<0.001), I2 81.6%; the risk 'remained significant both in studies matched or unmatched for participants' age'. SUBSET DENOMINATORS: the BMI strata are subsets of the 23 whose own study counts and participant numbers are NOT reported in the abstract - obese PCOS vs non-PCOS RR 3.24 (2.25-4.65, p<0.001; I2 30.9%), non-obese PCOS vs non-PCOS RR 1.62 (0.14-18.50, p=0.70; I2 89.9%). Read the non-obese figure honestly: with I2 89.9% and a CI spanning 0.14 to 18.50 it is UNINFORMATIVE, not a demonstrated null. Within PCOS, obese vs non-obese RR 3.85 (1.99-7.43, p<0.001; I2 46.2%), and the same for overweight vs lean. The authors' condition is part of the finding: they conclude PCOS raises T2DM risk 'only if they are obese/overweight'. This is the only meta-analysis in the vault that purpose-tests this claim's scope of BMI-independence, and it points AGAINST it, where moran-2010's BMI-matched OR 4.00 points for. Prevalence-vs-incidence design of the pooled studies, funding/COI, registration and publication-bias testing are not in the abstract and are cannot-tell, not absent.

Disagree, or know a study we missed?

We grade by evidence, not opinions. The way to weigh in is to point us to a study we haven't cited (check the evidence table above first), or to flag a problem with one we have. Every submission is reviewed; if it holds up, the grade updates and shows in Science Changes Its Mind.

📚 Suggest a study ⚑ Flag / request reclassification

Opens a short form. You'll sign in with Google so submissions are tied to a real account — we don't display your identity, and we only accept a link we can verify (PubMed, DOI, ClinicalTrials.gov).

Educational only, not medical advice. Grades and scores reflect published evidence weighted by study design and quality; see the methodology.