Development and Risk Factors of Type 2 Diabetes in a Nationwide Population of Women With Polycystic Ovary Syndrome

Abstract

Objective

Polycystic ovary syndrome (PCOS) is associated with insulin resistance and obesity. Prospective population-based data regarding development and possible predictors of type 2 diabetes (T2D) in PCOS are limited.

Design

National Patient Register–based study.

Methods

Patients with PCOS [PCOS Denmark and embedded cohort, PCOS Odense University Hospital (OUH)] and a control population with no previous diagnosis of T2D. PCOS OUH (N = 1,162) included premenopausal women with PCOS and standardized clinical and biochemical examination. PCOS Denmark (N = 18,477) included women with PCOS in the Danish National Patient Register. Three age-matched controls were included per patient (N = 54,680).

Main outcome

T2D events according to diagnosis codes and filled medicine prescriptions.

Results

The median (quartiles) follow-up was 11.1 (6.9 to 16.0) years. The hazard ratio (HR) with 95% confidence interval (CI) for development of T2D in PCOS Denmark was HR = 4.0 (95% CI, 3.7 to 4.3; P < 0.001), and the total event rate of T2D was 8.0 per 1000 person years in PCOS Denmark vs 2.0 per 1000 person years in controls (P < 0.001). The median age at diagnosis of T2D was 31 (26 to 37) years in PCOS Denmark vs 35 (27 to 44) years in controls (P < 0.001). In multiple regression analyses, body mass index, glycated hemoglobin, fasting blood glucose, 2-hour blood glucose, homeostasis model assessment of insulin resistance, and triglycerides were positively associated with development of T2D, whereas higher number of births was negatively associated with development of T2D.

Conclusion

The event rate of T2D was higher in PCOS compared with controls, and T2D was diagnosed at a younger age.

Polycystic ovary syndrome (PCOS) is most often defined according to the Rotterdam criteria, which include irregular ovulation, biochemical/clinical hyperandrogenism, and/or polycystic ovaries when other etiologies are excluded (1, 2). Insulin resistance is part of the pathogenesis of PCOS (3) and insulin resistance is associated with increased risk of type 2 diabetes (T2D) in PCOS (4–7). Insulin resistance is closely associated with obesity (3, 8). As recently reviewed, evidence regarding the increased risk of T2D in PCOS is based on observational studies with high risk of bias (4), and data from population-based studies are limited (9, 10). High body mass index (BMI), increasing age, consecutive pregnancies, and treatment with oral contraceptives (OCP) could modify the risk of development of T2D in PCOS, but no data are available from prospective studies. In women with T2D, a diagnosis of PCOS was associated with higher BMI and earlier onset of diabetes (11). Increasing age is associated with loss of beta cell function (12) and age >40 years is considered an additional risk factor for T2D in PCOS in some guidelines (13–15). The prevalence of gestational diabetes mellitus (GDM) is increased in PCOS (16) and cumulative weight-gain during successive pregnancies would increase the risk of developing T2D in PCOS (17). Fasting insulin levels were unchanged during treatment with OCP in PCOS (18). Use of OCP could, however, be associated with increased risk of T2D due to weight gain (19, 20) and use of OCP could lead to higher levels of glucose and insulin levels during oral glucose test (OGTT) (20, 21).

The aim of the present register-based study was to investigate the risk of development of T2D in women with PCOS and possible modifying effects of age, number of births, and prescription of OCP. We extracted diagnosis codes in the Danish National Patient Register (NPR) and medical prescriptions from the National Prescriptions Registry. Possible associations between baseline clinical and biochemical characteristics and later development of T2D were investigated in a well-described representative subgroup of patients with hyperandrogenism and/or PCOS.

Material and Methods

The study design and baseline data for this study have recently been reported in detail (10). Briefly, the study used an observational register-based cohort drawn from Danish national health registers including two patient populations with PCOS and one control population (Fig. 1). The PCOS cohort included all women in Denmark aged 12 to 60 years, who were diagnosed through a hospital contact with PCOS (E282) and/or hirsutism (L680) between 1 January 1995 and the end of 2012 (PCOS Denmark).

Figure 1.

Flowchart of included women with PCOS and controls. Flowchart of included women in PCOS Denmark, PCOS OUH, and controls. Women with the diagnoses E221 (hyperprolactinemia), E220 (acromegaly), E24 (Cushing’s syndrome), E25 (adrenogenital syndrome), and Q96 (Turner syndrome) were excluded from the cohort. T2D was defined as: (1) ICD-10: E11 (noninsulin dependent diabetes mellitus), E14 (unspecified diabetes mellitus), or O24 (diabetes mellitus in pregnancy); (2) prescription of drugs for treatment of diabetes excluding prescription of metformin (A10BA02); or (3) HbA1c ≥6.5% (48 mmol/mol) and/or blood glucose ≥11.1 mmol/L (fasting or during a 2-hour OGTT) in PCOS OUH.

For each patient with PCOS, three control women born in the same year as the patient were randomly drawn from the civil population register. Controls were assigned the index date (date of first PCOS diagnosis) of their matched PCOS cases. Within PCOS Denmark, we embedded a local subcohort of women with PCOS and/or hirsutism treated at Odense University Hospital (PCOS OUH). For the current study, estimated glomerular filtration rate (eGFR) and glycated hemoglobin (HbA1c) measured at baseline were included in the dataset (10) along with prospective results of HbA1c and blood glucose (BG) samples analyzed at our hospital. These samples included samples drawn by in- and outpatient clinics at the hospital and samples sent to analysis by general practitioners on Funen.

Assays in PCOS OUH subcohort

Serum total testosterone was analyzed using a specific radioimmunoassay after extraction as previously described (22) and sex hormone–binding globulin (SHBG) was analyzed by an autoDelfia assay. The intra-assay coefficient of variation (CV) for the total testosterone assay was 8.2% and 5.2% for SHBG. The interassay CV for the total testosterone assay was 13.8% and 7.5% for SHBG. Insulin was analyzed by a time-resolved fluoroimmunoassay using a commercial kit (AutoDelfia; Wallac Oy, Turku, Finland) with an intra-assay variation of 2.1% to 3.7% and interassay variation of 3.4% to 4.0%. Plasma creatinine, total cholesterol, high-density lipoprotein cholesterol, and triglycerides were analyzed by enzymatic colorimetric reactions (Modular P; Roche, Hvidovre, Denmark), while low-density lipoprotein cholesterol was calculated using the Friedewald equation. BG was measured on capillary ear blood using Hemo Cue. We calculated the homeostasis model assessment of insulin resistance (HOMA-ir = fasting insulin × fasting BG / 22.5) (23). HbA1c was measured by high-performance liquid chromatography as fraction of total hemoglobin A0 using Tosoh G8 (Medinor, Broendby, Denmark) with reagents as recommended by the supplier. The analytical CV was 0.9%. eGFR was calculated from plasma creatinine values using the simplified four-variable “modification of diet in renal disease” equation without correction for race (24).

During 1997 to 2003, OGTT was part of the routine evaluation program for newly referred women with PCOS (25). P-insulin and capillary BG were measured at fasting and 30, 60, and 120 minutes after oral ingestion of a glucose load containing the equivalent of 75 g anhydrous glucose dissolved in water.

The Danish health registries

All individuals in Denmark are assigned a unique personal identification number and data from national registers can therefore be linked at an individual level. We retrieved information about hospital contacts and medicine prescriptions filled, along with dates of death if applicable, in women with PCOS and control subjects from NPR, the National Prescription Registry, and the National Cause of Death Register from 1995 to 31 December 2015.

Exposure

Use of OCP, use of metformin and number of births were used as exposures. Uses of OCP (Anatomical Therapeutic Chemical codes G03AA, G03AB, and G03HB01) and metformin (Anatomical Therapeutic Chemical code A10BA02) were defined as two or more dispenses of medicine prescriptions in the National Prescription Registry. Number of births [International Classification of Diseases (ICD)-10 codes O80-O84] were categorized as 0, 1, 2, or ≥3 and extracted from NRP. The covariates were defined before the index date in baseline tables and if present during the time span of the study from 1995 to 31 December 2015 or to an outcome event occurred.

Outcome

The primary outcome was T2D, defined as at least one of the following criteria. (1) The presence of a diabetes diagnosis in NPR according to ICD-10: E11 (noninsulin dependent diabetes mellitus), E14 (unspecified diabetes mellitus), or O24 (diabetes mellitus in pregnancy); (2) prescription of drugs for treatment of diabetes according to the National Prescriptions Registry database: A10 (antidiabetics), excluding prescription of metformin (A10BA02); or (3) in PCOS OUH, the occurrence of HbA1c ≥6.5% (48 mmol/mol) and/or BG ≥11.1 mmol/L (fasting or during 2-hour OGTT) was defined as incident diabetes. In regression analyses including number of births as predictors of T2D, O24 was excluded from T2D events. Women with known T2D and women with newly diagnosed T2D during baseline evaluation for PCOS were excluded, and therefore, participants with T2D events occurring before and up until 3 months after the index date were excluded.

Statistical analyses

Descriptive analyses for categorical variables were presented as frequencies and difference between PCOS and control group was evaluated by χ² test. Continuous variables were tabulated as medians (with quartiles 1 and 3) and nonparametric test on the equality of medians were used to test for differences between groups. P values below 0.05 were considered statistically significant.

Cox proportional hazard models were used to calculate incidence rates per 1000 person years (PY), hazard ratios (HRs) and 95% confidence intervals (95% CIs), and corresponding P values for outcomes (T2D with GDM included or excluded). The analysis including the total cohort (PCOS Denmark and controls) was carried out as crude HR, only including the exposure, and adjusted for OCP in analysis when the outcome was T2D including GMD. In the analyses where outcome was T2D excluding GDM, adjusted analyses were reported with OCP and number of births. The matching of PCOS and controls was taken into account by estimating stratified baseline hazard for each matching set. Analyses were conducted using STATA 14 through a remote virtual private network access to Statistics Denmark.

Ethics

The core study was an open register-based cohort study. The study did not need approval from the local ethics committee or institutional review board by Danish law. The study was approved by the Data Protection Agency and by Statistics Denmark (project no. 704175).

Results

The flowchart of included women is summarized in Fig. 1. A total of 18,477 women with PCOS (PCOS Denmark and the embedded cohort PCOS OUH, N = 1,162) and 54,680 controls were included in the study.

Baseline characteristics

PCOS Denmark vs controls

Women in PCOS Denmark had higher prevalence of ICD-10 codes and medicine prescriptions related to the metabolic syndrome occurring before the index date compared with controls, higher prescription of OCP and drugs for fertility treatment and higher number of births before the index date (18% vs 13% had ≥ 1 births) than controls (Supplemental Tables 1 and 2).

Table 1.

Event Rates of T2D in PCOS OUH, PCOS Denmark, and Controls

	PCOS OUH (N = 1,162)		PCOS Denmark (N = 18,477)		Controls (N = 54,680)		Pa
	N (%)	Incidence Rate per 1000 PYs	N (%)	Incidence Rate per 1000 PYs	N (%)	Incidence Rate per 1000 PYs
Total event rates of T2D (GDM included)	115 (10)	9.0	1621 (9)	8.0	1274 (2)	2.0	<0.001
Total event rates of T2D (GDM excluded)	89 (8)	6.9	1120 (6)	5.4	996 (2)	1.6	<0.001
ICD-10 diabetes, total	97 (8)	7.5	1,394 (8)	6.8	1007 (2)	1.6	<0.001
E11 (T2D)	60 (5)	4.6	790 (4)	3.8	663 (1)	1.1	<0.001
E14 (diabetes)	25 (2)	1.9	149 (1)	0.7	140 (0.3)	0.2	<0.001
O24 (GDM)	42 (4)	3.1	688 (4)	3.3	330 (0.6)	0.5	<0.001
Antidiabetic medicine,b total	46 (4)	3.5	764 (4)	3.7	765 (1)	1.2	<0.001
Insulin (A10A)	24 (2)	1.8	310 (2)	1.5	402 (1)	0.6	<0.001
Other (A10B)	36 (3)	2.7	665 (4)	3.2	577 (1)	0.9	<0.001
HbA1c ≥6.5%	47 (4)	3.6	n/a	n/a	n/a	n/a	n/a
BG ≥11.1 mmol/L	1 (0.1)	0.1	n/a	n/a	n/a	n/a	n/a

^a

χ² test between PCOS Denmark and controls.

^b

Antidiabetic medicine: Prescription of metformin excluded (medicine code A10BA02).

Table 2.

Baseline Clinical and Biochemical Characteristics According to Development of T2D (GDM Included) in PCOS OUH

Baseline Characteristics	Development of T2D (GDM included)				Pa
	Yes (N = 115)		No (N = 1,047)
	N (%)	Median (Q1 to Q3)	N (%)	Median (Q1 to Q3)
Age, y	115 (100)	31 (26 to 36)	1047 (100)	28 (22 to 34)	0.045
BMI, kg/m2	111 (97)	32.3 (27.9 to 36.5)	971 (92)	26.3 (22.6 to 31.2)	<0.001
Waist, cm	65 (56)	102 (91 to 112)	664 (63)	87 (77 to 101)	<0.001
HbA1c, mmol/mol	59 (51)	37.7 (34.4 to 42.1)	506 (48)	33.3 (31.2 to 36.6)	<0.001
Fasting BG, mmol/L	65 (57)	4.8 (4.5 to 5.3)	466 (45)	4.6 (4.2 to 5.0)	0.001
2-h BG, mmol/L	66 (57)	7.3 (6.1 to 9.3)	454 (43)	6.0 (5.2 to 6.8)	<0.001
Fasting insulin, pmol/L	68 (59)	104 (68 to 158)	501 (48)	51 (35 to 82)	<0.001
HOMA-ir, pmol mmol l−2	64 (56)	23.1 (15.3 to 36.7)	475 (45)	11.1 (7.4 to 17.0)	<0.001
LDL cholesterol, mmol/L	82 (71)	2.9 (2.3 to 3.7)	765 (73)	2.7 (2.2 to 3.3)	0.23
HDL cholesterol, mmol/L	84 (73)	1.2 (1.0 to 1.5)	766 (73)	1.4 (1.1 to 1.6)	0.001
Cholesterol, mmol/L	84 (73)	5.0 (4.4 to 5.8)	777 (74)	4.6 (4.1 to 5.2)	0.005
Triglycerides, mmol/L	84 (73)	1.6 (1.0 to 2.3)	765 (73)	1.0 (0.7 to 1.4)	<0.001
Total testosterone, nmol/L	87 (76)	1.7 (1.1 to 2.2)	743 (71)	1.8 (1.3 to 2.4)	0.59
SHBG, nmol/L	111 (97)	36 (24 to 51)	965 (92)	46 (32 to 68)	<0.001
Free testosterone, nmol/L	86 (75)	0.033 (0.021 to 0.053)	730 (70)	0.033 (0.021 to 0.048	0.98
eGFR, mL/min	84 (73)	110 (98 to 122)	802 (77)	114 (102 to 125)	0.25

Abbreviations: HDL, high-density lipoprotein; LDL, low-density lipoprotein; Q, quartile.

^a

Nonparametric test on the equality of medians.

Baseline clinical and biochemical data in PCOS OUH

The prevalence of women in PCOS OUH with BMI <25 kg/m² was 40% (430/1082), BMI between 25 and 29.9 kg/m² was 25% (273/1082), and BMI ≥30 kg/m² was 35% (379/1082). The majority of women in PCOS OUH fulfilled the Rotterdam criterion of hyperandrogenism (biochemical and/or clinical; 1074/1156 = 93%), and 33% (387/1159) women fulfilled all three Rotterdam criteria.

T2D event rates

The median (quartiles 1 to 3) follow-up duration was 11.1 (6.9 to 16.0) years (Table 1). The incidence rate of T2D was 9/1000 PYs in PCOS OUH, 8/1000 PYs in PCOS Denmark, and 2/1000 PYs in controls (P < 0.001 PCOS Denmark vs controls). Ten events of T2D in PCOS OUH were defined by HbA1c ≥6.5% (N = 9) or BG ≥11.1 mmol/L (N = 1) in women with no ICD-10 diagnosis of T2D and no prescription of antidiabetic medicine.

The incidence rates for T2D were comparable in the four different PCOS phenotypes: phenotype A (all three Rotterdam criteria present, N = 387), 9.8/1000 PYs; phenotype B (hyperandrogenism and oligomenorrhea, N = 232), 9.7/1000 PYs; phenotype C (hyperandrogenism and polycystic ovary, N = 106), 9.5/1000 PYs; and phenotype D (oligomenorrhea and polycystic ovary, N = 49), 13.7/1000 PYs, P = 0.85.

Baseline characteristics According to development of T2D in PCOS OUH

Women in PCOS OUH who developed T2D were significantly older, more obese (higher BMI and waist circumference), had higher glucose (fasting and 2-hour), insulin, HOMA-ir, more adverse lipid profile (low-density lipoprotein, high-density lipoprotein, cholesterol, triglycerides), and lower SHBG upon baseline evaluation compared with women in PCOS OUH and no development of T2D, whereas the two groups were comparable regarding testosterone levels (total and free testosterone) and renal function (Table 2). The prevalence of HbA1c <42 mmol/L at baseline was significantly lower in women in PCOS OUH with later development of T2D compared with women in PCOS OUH that did not develop T2D (43/115 = 37% vs 495/1047 = 47%), whereas the prevalence of fasting BG <6.1 mmol/L and 2-hour BG <7.8 mmol/L was comparable between the two groups (data not shown).

Characteristics According to development of T2D in PCOS Denmark and controls

PCOS Denmark vs controls

The prescription of OCP (71% PCOS vs 34% controls) and metformin (26% PCOS vs 2% controls) occurring after the index date and the number of births (46% PCOS vs 26% controls had ≥1 birth) was significantly higher in PCOS Denmark vs controls (Table 3).

Table 3.

Characteristics According to Development of T2D in PCOS Denmark and Controls

	Development of T2D in PCOS			Development of T2D in Controls			PCOS vs Controls	P A vs B	P A vs C
	Total	Yes: A	No: B	Total	Yes: C	No: D
Development of T2D (GDM included)	18,477	1621	16,856	54,680	1274	53,406
Age at diagnosis, y	29 (24 to 36)	31 (26 to 37)	29 (23 to 36)	29 (24 to 36)	35 (27 to 44)	29 (23 to 36)	0.49	<0.001	<0.001
Median (Q1, Q3)
	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)
Age <40 y	15,477 (84)	1323 (82)	14,154 (84)	45,909 (84)	836 (66)	45,073 (84)	0.53	0.01	<0.001
Use of OCPa	13,201 (71)	1056 (65)	12,145 (72)	18,833 (34)	400 (31)	18,433 (35)	<0.001	<0.001	<0.001
Development of T2D (GDM excluded)	18,477	1120	17,357	54,680	996	53,684
Age at diagnosis, y	29 (24 to 36)	33 (27 to 40)	29 (23 to 36)	29 (24 to 36)	38 (31 to 46)	29 (23 to 36)	0.49	<0.001	<0.001
Median (Q1, Q3)
Age <40 y	15,477 (84)	827 (74)	14,650 (84)	45,909 (84)	559 (56)	45,350 (85)	0.53	<0.001	<0.001
Characteristics before T2D (GDM excluded)
	N (%)	N (%)	N (%)	N (%)	N (%)	N (%)
Use of OCPa	13,209 (71)	653 (58)	12,556 (72)	18,841 (34)	163 (16)	18,678 (35)	<0.001	<0.001	<0.001
Use of metformina	4833 (26)	680 (61)	4153 (24)	961 (2)	438 (44)	523 (1)	<0.001	<0.001	<0.001
Number of births
0	8992 (49)	791 (71)	8201 (47)	40,470 (74)	897 (90)	38,573 (73)	<0.001	<0.001	<0.001
1	3966 (21)	208 (19)	3758 (22)	5324 (10)	62 (6)	5262 (10)
2	4106 (22)	90 (8)	4016 (23)	6494 (12)	24 (2)	6470 (12)
≥3	31 (3)	31 (3)	1382 (8)	2392 (4)	13 (1)	2379 (4)

P values obtained with χ² test for categorical variables and nonparametric test on the equality of medians for continuous variables.

Abbreviation: Q, quartile.

^a

ICD 10 codes OCP: G03AA, G03AB, G03HB01; metformin: A10BA02.

Women with development of T2D in PCOS Denmark vs controls

The median age at T2D diagnosis was significantly lower in PCOS Denmark vs controls (31 vs 35 years, P < 0.001), and 82% vs 66% were aged <40 years at T2D diagnosis, respectively (P < 0.001). When GDM was excluded from T2D diagnosis, the median age at T2D diagnosis was 33 vs 38 years in PCOS vs controls (P < 0.001). Prescription of metformin and OCP and number of births (30% PCOS vs 9% controls had ≥1 birth) were significantly higher in women in PCOS Denmark with development of T2D compared with controls with development of T2D.

PCOS Denmark and development of diabetes yes vs no

Women in PCOS Denmark that developed T2D were significantly older than women in PCOS Denmark without development of T2D (31 vs 29 years, P < 0.001), the prevalence of ever users of OCP was lower, the prevalence of ever users of metformin was higher and the number of births was lower (30% vs 53% had ≥ 1 births).

Regression analyses

The HR for development of T2D was 4.0 (95% CI, 3.7 to 4.3) in PCOS Denmark vs controls when GDM was included, and the HR for T2D in PCOS was 3.5 (95% CI, 3.2 to 3.8) when GDM was excluded (Tables 4 and 5). In regression models, where GDM was included as a T2D outcome, prescription of OCP was statistically associated with a higher risk of T2D (HR = 1.4; 95% CI, 1.3 to 1.6), whereas higher number of births decreased the risk of T2D. However, use of OCP was not associated with increased risk of T2D when GDM was not part of the diabetes outcome (HR = 1.0; 95% CI, 0.9 to 1.2). We found a significant interaction between PCOS diagnosis and age (HR = 0.96; 95% CI, 0.96 to 0.97, P < 0.001).

Table 4.

Crude and Adjusted HRs in PCOS Denmark (N = 18,477) and Controls (N = 54,680) and Development of T2D (With or Without GMD Excluded)

	Crude HR (95% CI)	Adjusted HR HR (95% CI)	Adjusted HR HR (95% CI)
Outcome: T2D (GDM included)
PCOS (yes/no)	4.0 (3.7 to 4.3)	3.6 (3.4 to 4.0)
	P < 0.001	P < 0.001
OCP (yes/no)		1.4 (1.3 to 1.6)
		P < 0.001
Outcome: T2D (GDM excluded)
PCOS (yes/no)	3.5 (3.2 to 3.8)	3.5 (3.2 to 3.9)	3.6 (3.2 to 3.9)
	P < 0.001	P < 0.001	P < 0.001
OCP (yes/no)		1.0 (0.9 to 1.2)	1.1 (0.9 to 1.3)
		P = 0.73	P = 0.06
Number of births
0			1
1			0.9 (0.7 to 1.0)
			P = 0.13
2			0.5 (0.4 to 0.6)
			P < 0.001
≥3			0.8 (0.5 to 1.1)
			P = 0.20

Predictors for development of T2D in PCOS Denmark and controls. HRs are presented for crude models and models corrected for use of OCP and number of births.

Table 5.

Cox Regression Models: Crude and Adjusted HRs in PCOS OUH and Development of T2D (GDM Included)

	N	Crude HR (95% CI)	N	Age- and BMI Adjusted HRa (95% CI)
HbA1c	565	1.2 (1.2 to 1.3)	521	1.2 (1.1 to 1.2)
		P < 0.001		P < 0.001
Fasting BG	531	2.8 (2.1 to 3.6)	508	2.6 (2.0 to 3.4)
		P < 0.001		P < 0.001
2-h BG	520	1.5 (1.4 to 1.6)	496	1.4 (1.3 to 1.5)
		P < 0.001		P < 0.001
HOMA-ir	539	1.0 (1.0 to 1.1)	509	1.0 (1.0 to 1.0)
		P < 0.001		P < 0.001
Triglycerides	849	1.5 (1.4 to 1.6)	807	1.4 (1.2 to 1.6)
		P < 0.001		P < 0.001
SHBG	1076	1.0 (1.0 to 1.0)	1005	1.1 (1.0 to 1.0)
		P = 0.011		P = 0.57
Age	1162	1.0 (1.0 to 1.0)	1082	1.0 (1.0 to 1.0)
		P = 0.042		P = 0.09
BMI	1082	1.1 (1.1 to 1.1)	1082	1.1 (1.1 to 1.1)
		P < 0.001		P < 0.001

Baseline characteristics in PCOS OUH and risk of development of T2D. HRs are presented for crude models and models corrected for age and BMI.

^a

Except age, which is adjusted for BMI alone, and BMI, which is adjusted for age alone.

In PCOS OUH, HbA1c, fasting BG, 2-hour BG, HOMA-r, triglycerides, SHBG, age, and BMI upon baseline were predictors of development of T2D (Table 5). When models were corrected for age and BMI, fasting BG, 2-hour BG, and triglycerides were the best predictors of development of T2D.

Discussion

In the current study we found an odds ratio of 4.0 for the development of T2D and GDM in a nationwide population-based cohort of women with PCOS, and T2D was diagnosed on average 4 years earlier in PCOS. Higher number of pregnancies was associated with lower risk of T2D among women with PCOS. In a representative subgroup of women with PCOS from our outpatient clinic, the risk of T2D was adversely affected by higher BMI, lipids, insulin and glucose levels upon PCOS diagnosis. The PCOS OUH cohort was relatively lean (median BMI = 26.9 kg/m²) upon PCOS diagnosis, but 10% women developed T2D during a median follow-up of 11.1 years.

To our knowledge, this is the first nationwide study to describe prospective risk of T2D in PCOS and the modifying effect of several risk factors. First, we could confirm results from a recent meta-analysis where the odds ratio for T2D was 4.4 in women with PCOS compared with controls (4).The meta-analysis included data from smaller observational studies in selected populations of women with PCOS (4), whereas few nationwide data set regarding risk for T2D in PCOS are available (9, 10). We recently reported that 1.5% women had been diagnosed with T2D at the time of their diagnosis of PCOS compared with 0.4% of non-PCOS controls, which corresponded to a five times increased likelihood of T2D in PCOS at presentation (10). The risk for T2D is closely associated with BMI, and T2D was very uncommon in normal weight women at the time of diagnosis of PCOS (25, 26). A recent population-based Finnish study reported a synergistic effect of overweight/obesity and PCOS for the risk of development of T2D, whereas the risk of T2D was not increased in normal weight women with PCOS (7). Two previous observational studies from the United States and one from Australia reported T2D prevalence of 10% (mean age = 26.5 years, BMI = 37.1 kg/m²) (27), 7.5% (mean age = 28 years, BMI = 32.9 kg/m²) (27), and 5.8% (mean age = 20.5 years, BMI = 28.0 kg/m²) (28), respectively. These findings support a relatively low baseline risk of T2D in Danish women with PCOS compared with other nationalities. Only hospital diagnoses are included in NPR and patients receiving no diabetes medications and needing no hospital appointments for their diabetes would not be identified as having T2D in our dataset. However, the current study design could favor patients with a more severe PCOS phenotype, which would affect our risk estimates. In comparison, the HR for development of T2D was 3.0 in women with PCOS compared with controls (event rates = 5.7/1000 PYs vs 1.7/1000 PYs, P < 0.001) during a median follow-up of 4.5 years in a British study using data from the General Practice Research Database (9). The study included 21,740 women with PCOS and mean BMI of 28.7 kg/m² and age of 27.1 years at baseline (9). The lower event rate for T2D in the British study compared with our data could, however, also be explained by a relatively short follow-up duration in a young study cohort. Prospective studies are needed in more obese study populations with higher absolute risk of T2D, but available data support that the RR for T2D is relatively conserved in women with PCOS compared with controls independent of baseline risk profile.

We found that the median age at T2D diagnosis was significantly lower in PCOS Denmark vs controls and 82% vs 66% was aged <40 years at T2D diagnosis. These data suggest that the majority of T2D cases in PCOS were diagnosed already before the age of 40 years and during multiple regression analysis, age was not a strong predictor of T2D development. Furthermore, age and PCOS showed a significant interaction, supporting that PCOS was stronger associated with development of T2D in young than in older age. BMI and age are closely associated (29, 30), and in accordance, the significant association between age and T2D development became nonsignificant after correcting for BMI. Increasing age was considered an important predictor for T2D in PCOS and guidelines suggested that screening for T2D was especially relevant in women with PCOS aged >40 years (13–15). Overall, our data support that the inclusion of increasing age as a separate risk parameter is not relevant in future algorithms regarding screening for T2D in PCOS.

Testosterone levels (free and total) measured at baseline did not predict later development of T2D in PCOS OUH and the incidence rate of T2D was comparable between different PCOS phenotypes. The importance of individual Rotterdam criteria, especially hyperandrogenism, for metabolic risk in PCOS is currently debated. Women in PCOS OUH were relatively lean and 93% had clinical and/or biochemical hyperandrogenism, which could have affected study results. However, we recently reported that the presence of individual Rotterdam criteria was not associated with cardiometabolic diagnoses upon diagnosis in PCOS OUH despite a more adverse metabolic risk profile in women with polycystic ovaries and irregular menses (10). The present results were in accordance with a Chinese study in 2436 women with PCOS reporting no impact of PCOS phenotype on glucose tolerance (31). Importantly, mean BMI varies between different PCOS phenotypes (32, 33), and studies regarding metabolic risk should adjust for BMI. Different results could be found in more obese study populations predominantly consisting of other PCOS phenotypes. Further prospective studies are needed to determine if the long-term metabolic risk is affected by PCOS phenotype independent of BMI or lipid levels.

OCP are often used to treat hyperandrogenism and irregular menses in PCOS. In accordance, 71% women in PCOS Denmark vs 34% controls were treated with OCP. We found that treatment with OCP increased the risk of T2D in PCOS Denmark, but this became nonsignificant when GDM was excluded from T2D diagnosis and number of births was added as predictor of T2D development. These results could suggest a confounding effect of pregnancies for effect of OCP treatment on development of T2D. In pregnancy, maternal insulin sensitivity changes to accommodate fetal glucose needs, with insulin resistance prevailing in the third trimester (34). The risk of GDM was more than six times increased in PCOS Denmark and PCOS OUH compared with controls in the current study, but in multiple regression analyses, higher number of births decreased the risk of T2D. The most straightforward explanation could be that multiple pregnancies indicate a less severe PCOS phenotype and therefore also a lower risk of later developing T2D. The possible metabolic risk of OCP is discussed. We and others reported increased insulin levels during OGTT in women with PCOS treated with OCP (21, 35) along with weight gain (19), which could increase the risk of development of T2D. The risk of venous thromboembolism is increased in PCOS and increases further during use of OCP (36). Therefore, OCP should not be prescribed in high-risk individuals, and this could affect study results. The number of births was higher in PCOS Denmark vs controls, which supported that PCOS was often diagnosed as part of fertility treatment. The number of births was reduced in obese study populations with PCOS (37), but our study confirms that fertility is not reduced in relatively lean study populations with PCOS.

The inclusion of PCOS OUH allowed us to investigate associations between baseline metabolic screening and later development of T2D. Glucose levels at fasting and during OGTT, fasting insulin, and HOMA-ir levels were higher upon baseline evaluation in women with later development of T2D compared with women without development of T2D. Furthermore, the lipid profile was more unfavorable and SHBG levels were lower. In multiple regression analyses, fasting BG, 2-hour BG, and triglycerides were the best predictors of development of T2D after correcting for age and BMI. We found that the regression model including fasting BG showed the highest HR for T2D development. In accordance, fasting BG is applied in many prediction models for risk of developing T2D in non PCOS populations (38), but we are not aware of such models in PCOS study populations. HbA1c is a measure of average glucose levels and could be a better predictor of cardiovascular disease and overall mortality than fasting or 2-hour glucose (39). However, the current study did not support that HbA1c levels at baseline were superior to other glucose measures as indicator for increased risk of T2D in PCOS.

The strength and limitations of the current study have in part been described recently (10). Our study was nationwide and the embedded cohort of PCOS OUH with available clinical and biochemical data allowed us to test hypotheses that could not be tested in the national cohort. PCOS OUH and PCOS Denmark were comparable regarding presence of cardio-metabolic diagnoses before the index date (10) and regarding their likelihood of development of T2D. Some limitations may however apply to our study as we only included women with PCOS diagnosed during hospital contacts. Therefore, some women in the control group may have undiagnosed PCOS, which could lead to underestimation of T2D risk in PCOS. An ICD-10 diagnosis of type 1 diabetes was excluded from T2D definition, but some patients with type 1 diabetes could have been included in the study population due to prescriptions of antidiabetic medicine. Women were included after the diagnosis of PCOS, whereas type 1 diabetes is most often diagnosed in young age. Hence, inclusion of patients with type 1 diabetes is unlikely to have taken place to any major degree.

In conclusion, the risk of development of T2D was significantly increased in PCOS and T2D was diagnosed at a younger age. BMI and fasting BG were the best predictors of development of T2D in PCOS, whereas our data supported that increasing age should not be included in future guidelines as an isolated risk marker of T2D development in PCOS. Future studies are needed to evaluate the effect of OCP and number of births on risk of T2D in PCOS.

Abbreviations

 
• BG

  blood glucose

 
• BMI

  body mass index

 
• CI

  confidence interval

 
• CV

  coefficient of variation

 
• eGFR

  estimated glomerular filtration rate

 
• GDM

  gestational diabetes mellitus

 
• HbA1c

  glycated hemoglobin

 
• HR

  hazard ratio

 
• HOMA-ir

  homeostasis model assessment of insulin resistance

 
• ICD

  International Classification of Diseases

 
• NPR

  National Patient Register

 
• OCP

  oral contraceptives

 
• OGTT

  oral glucose tolerance test

 
• OUH

  Odense University Hospital

 
• PCOS

  polycystic ovary syndrome

 
• PY

  person year

 
• SHBG

  sex hormone–binding globulin

 
• T2D

  type 2 diabetes.

Acknowledgments

Disclosure Summary: The authors have nothing to disclose.