Prospective Association of Serum Androgens and Sex Hormone-Binding Globulin with Subclinical Cardiovascular Disease in Young Adult Women: The “Coronary Artery Risk Development in Young Adults” Women’s Study

Abstract

Context: The role of endogenous androgens and SHBG in the development of cardiovascular disease in young adult women is unclear.

Objective: Our objective was to study the prospective association of serum androgens and SHBG with subclinical coronary and carotid disease among young to middle-aged women.

Design and Setting: This was an ancillary study to the Coronary Artery Risk Development in Young Adults (CARDIA) study, a population-based multicenter cohort study with 20 yr of follow-up.

Participants: Participants included 1629 women with measurements of serum testosterone and SHBG from yr 2, 10, or 16 and subclinical disease assessment at yr 20 (ages 37–52 yr).

Main Outcome Measures: Coronary artery calcified plaques (CAC) and carotid artery intima-media thickness (IMT) were assessed at yr 20. The IMT measure incorporated the common carotid arteries, bifurcations, and internal carotid arteries.

Results: SHBG (mean of yr 2, 10, and 16) was inversely associated with the presence of CAC (multivariable adjusted odds ratio for women with SHBG levels above the median = 0.59; 95% confidence interval = 0.40–0.87; P = 0.008). SHBG was also inversely associated with the highest quartile of carotid-IMT (odds ratio for women with SHBG levels in the highest quartile = 0.56; 95% confidence interval = 0.37–0.84; P for linear trend across quartiles = 0.005). No associations were observed for total or free testosterone with either CAC or IMT.

Conclusion: SHBG levels were inversely associated with subclinical cardiovascular disease in young to middle-aged women. The extent to which low SHBG is a risk marker or has its own independent effects on atherosclerosis is yet to be determined.

In a long follow-up study, sex-hormone binding globulin, but not testosterone, is inversely associated with subclinical cardiovascular disease in young to middle-aged women.

Studies of the association between androgens and cardiovascular disease (CVD) in young women have been conducted mainly in the setting of polycystic ovarian syndrome (PCOS), a condition with prevalence of 3–7%, characterized by hyperandrogenicity. PCOS has been associated with cardiovascular risk factors including obesity, insulin resistance, and lipid abnormalities (1,2). Several studies suggest that women with PCOS have a higher risk of clinical and subclinical CVD (3,4,5,6,7,8,9,10).

Few studies have examined the association of androgens and SHBG with atherosclerosis among women from the general population, particularly in the absence of clinically evident PCOS. Studies of cardiovascular risk factors yielded associations of free testosterone (FT) with body mass index (BMI) among pre- and postmenopausal women (11,12,13). Among postmenopausal women, FT levels were associated with waist circumference (11), plasma insulin levels (14), and insulin resistance (15), and high testosterone levels were associated with an increased risk of type 2 diabetes (16).

Recent evidence suggests that SHBG has an independent, possibly causal, role in the pathogenesis of type 2 diabetes mellitus among postmenopausal women (17,18). This adds to the previously described inverse associations of SHBG with BMI, both in pre- and postmenopausal women (11,12,13,19).

Studies of the association between androgens and/or SHBG and subclinical CVD in women are mainly limited to postmenopausal women and to cross-sectional or case-control analyses and provide inconsistent information (20,21,22,23,24,25,26,27). We aimed to study the prospective associations of SHBG and endogenous androgens with subclinical CVD, namely coronary artery calcified plaques (CAC) and carotid intima-media thickness (IMT), in a population-based, multicenter, prospective study of young and middle-aged women.

Subjects and Methods

The Coronary Artery Risk Development in Young Adults (CARDIA) Women’s Study (CWS)

The CWS, an ancillary study to the CARDIA study, was designed to assess the associations of endogenous androgens and SHBG with coronary risk factors and subclinical disease development in young adult women. The CARDIA study was initiated in 1985–1986 and included 5115 adults aged 18–30 yr recruited from four centers: Birmingham, AL; Chicago, IL; Minneapolis, MN; and Oakland, CA (28,29). The study cohort was balanced at baseline for age (45% aged 18–24 yr), sex (54% women), race (52% Black, 48% White), and education (40% completed up to 12 yr of education). Consecutive examinations took place at baseline and at yr 2, 5, 7, 10, 15, and 20. The overall response rate at baseline was 51%, and the retention rates were 90, 86, 81, 79, 74, and 72% in yr 2, 5, 7, 10, 15, and 20, respectively.

Eligibility criteria for the CWS included participation in the yr-15 CARDIA exam, residence within 50 miles of a clinical center, not being pregnant at the time of contact, and having at least one intact ovary. Stored blood samples from yr 2 and 10 of 1629 potentially eligible women were tested for androgens and SHBG. Of these, 1227 women were invited at yr 16 to participate in a study protocol consisting of self-administered reproductive health questionnaires and blood draw for serum SHBG and androgens.

Study protocols were approved by the Institutional Review Board committees of the participating institutions, and all participants provided written informed consent.

SHBG, total testosterone (TT), and FT

SHBG, TT, and FT were measured in a single batch on serum specimens collected at yr 2, 10, and 16, when available. Hormone measurements for yr 2, 10, and 16 were obtained for 84, 90, and 71% of the 1629 women, respectively. For 86.3% of the study population, there were at least two measurements, and 54.4% had measurements from all three time points. Assays were conducted by the Obstetrics and Gynecology Research and Diagnostic Laboratory at the University of Alabama, Birmingham (UAB), a Clinical Laboratory Improvement Amendments-certified reproductive endocrinology laboratory. SHBG was determined using equilibrium dialysis on Sephadex G-25 as described by Pearlman (30). This method estimates the amount of testosterone capable of being bound by SHBG. The interassay coefficients of variation (CV) in males and pregnant females were 5.2 and 5.3%, respectively, for the UAB laboratory. TT was measured with a competitive immunoassay (Bayer Diagnostics, Tarrytown, NY) that employed direct chemiluminescent technology on the ACS:180 automated chemiluminescent system (Beckman Coulter, Fullerton, CA). The CV for the quality control TT sample (80 ng/dl) was 5.9% for the UAB laboratory. The manufacturer, Beckman Coulter, reports that the CV for this assay is less than 10% for TT of 50 ng/dl or higher. The College of American Pathologists distributed 30 ng/dl TT samples to multiple laboratories and found a 13.4% interassay CV for the Beckman Access system. This was the lowest CV of 16 systems surveyed (31). In the current analyses, TT levels below 10.0 ng/dl, the lower detection limit of the assay, were set to 5. FT was calculated based on measured TT and SHBG levels, using the method described by Pearlman (30).

We included only measures obtained from women who reported that they were not pregnant in the analyses. For most analyses, we used the mean value of yr 2, 10, and 16 of SHBG, TT, and FT levels to represent cumulative exposure and minimize the effect of measurement error. Spearman coefficients for the correlations of the individual measurements in yr 2, 10, and 16 ranged 0.46–0.65 for SHBG, 0.54–0.65 for TT, and 0.50–0.64 for FT.

CAC

CAC was assessed at the CARDIA yr-15 and -20 examinations by computed tomography (CT) of the chest. Electron beam CT (Imatron C-150; Imatron Corporation, San Francisco, CA) or multidetector CT scanners [GE Lightspeed (GE Medical Systems, Milwaukee, WI) or Siemens VZ/Siemens Biograph 16 (Siemens Medical Solutions, Erlangen, Germany)] obtained consecutive 2.5- to 3-mm-thick transverse images from the root of the aorta to the apex of the heart in two sequential electrocardiogram-gated scans. Experienced image analysts measured calcified plaques in the epicardial coronary arteries (left main, left circumflex, left anterior descending, and right) at a central reading center (Wake Forest University Health Sciences, Winston-Salem, NC). A total calcium score, using a modified Agatston method to account for slice thickness, was calculated on a U.S. Food and Drug Administration-approved workstation (TeraRecon Aquarius Workstation, San Mateo, CA) for each of the two sequential scans and averaged (29). Scoring parameters were specified as a minimum lesion size based on four adjacent pixels (an area of at least 1.87 mm²) with a density greater than 130 Hounsfield units. Review and adjudication by an expert physician in cardiovascular imaging was performed for all participants with discordant scan pairs, a change in calcium status between yr 15 and 20, indications of possible surgical intervention, or concerns identified by the reader. An additional review was performed of all exams with calcium scores higher than 200. CT analysts were blinded to participant information. The inter- and intra-observer reliabilities of this test in this population have been found to be between 0.77 and 0.95 (32).

For these analyses, the CAC score was used as a dichotomous variable (CAC negative stands for a score of zero, whereas any value above zero was considered as CAC positive), as previously suggested to predict CVD in low-risk women (33).

IMT

IMT measures of the common carotid artery (CCA), the carotid bulb (CB), and the internal carotid artery (ICA) were obtained at the CARDIA yr-20 examination. Carotid B-mode ultrasound examinations were conducted by trained sonographers at each field center employing a standard protocol using the GE Logiq 700 device (GE Medical Systems). The carotid ultrasound procedures were performed in the supine position with the participant’s head rotated 45° away from the side of study. Magnified longitudinal images in gray-scale of the far and near wall of the distal CCA, the CB, and the proximal ICA were obtained on the right and left sides. Images were read at the ultrasound reading center (Tufts Medical Center, Boston, MA). The maximum IMT of each segment was defined as the mean of the maximal IMT of the near and far wall of both the left and right sides. To create a composite IMT measure, the arithmetic average of all CB and ICA measurements were first combined into a single variable for each subject (CB/ICA). A composite IMT measure [standardized (STD)-IMT] combining the maximal CCA-IMT and CB/ICA IMT was calculated by averaging these two measurements after standardization (subtraction of the mean and division by the sd for each measurement). The number of measurements available for averaging ranged from one to four for the CCA and one to 16 for the CB/ICA.

Data analysis

Logistic regression models were fit to examine the associations of SHBG, TT, or FT with CAC at yr 20. Initially, the associations of serum SHBG and androgens with CAC were examined across quartiles of SHBG and androgens. For SHBG, the odds ratio (OR) derived from the logistic regression models were similar for the two higher quartiles compared with the lower quartile. Given evidence for a threshold at the median, SHBG levels were treated as a dichotomous variable (values above vs. below the median). Further analyses used the natural logarithmic transformation of the continuous levels.

Initial models were adjusted for age and BMI at baseline. Multivariable models were adjusted in addition for race (Black vs. White), use of oral contraceptives (yes, no, or missing; separate variables for yr 2, 10, and 16), and smoking at yr 15 visit (ever vs. never). Additional models controlled for baseline measurements of systolic blood pressure, low-density lipoprotein (LDL)-cholesterol, and the homeostasis model assessment for insulin resistance (HOMA-IR) (26) (calculated as fasting glucose × fasting insulin/22.5). We further estimated interactions between SHBG, TT, or FT as dichotomous variables and race or oral contraceptive use in these models. Separate models were constructed taking into account only SHBG/androgen measurements obtained when women were not using oral contraceptives and excluding measurements taken when women were using oral contraceptives. Additional models controlling for ever use of oral contraceptives (as estimated at yr 20) yielded practically the same results and are not presented. Information on self-reported oligomenorrhea at age 20–30 yr was obtained at yr 15 and was evident for 4.8% of the study population. Models were constructed adjusting for oligomenorrhea.

Because CAC was evaluated both in yr 15 and 20, we also assessed the association between SHBG, TT, or FT and incidence of CAC. This was estimated by women who had negative CAC at yr 15 and positive CAC at yr 20 (n = 84). Women who were CAC positive at yr 15 were excluded from this analysis.

Logistic regression models were used to assess the OR associated with androgen and SHBG levels for the highest quartile of STD-IMT compared with the lower three quartiles of STD-IMT. Models were adjusted for similar covariates as the analyses for CAC. General linear models were used to assess the associations of SHBG with STD-IMT as continuous variables. Models yielded the parameter estimates (β) for a 1-sd increment in the natural logarithmic transformation of SHBG, controlling for the above-mentioned covariates.

Of the 146 women who reported not having menstrual periods in the year before the CWS interview, only 28 (19%) had not undergone a previous hysterectomy. Therefore, postmenopausal status was defined according to yr 16 FSH levels (above 40 mIU/ml). Sensitivity analyses were conducted restricting the study population to the 1552 (95.3%) premenopausal women, based upon the serum FSH. Analyses controlling for FSH as a continuous variable, and using self-definition of menopause at yr 15, yielded practically the same results and thus are not shown.

We present OR for logistic regression models, parameter estimates (β) for general linear model regression models, 95% confidence interval (CI), and two-sided P values

Results

Table 1 shows the baseline characteristics of the study population.

Table 1.

Baseline (yr 0) characteristics of the study population, CWS

	Baseline characteristics of study population (n = 1629)
	Mean ± sd	n	%
Age (yr)	25.3 ± 3.64
Race
Black		795	48.8
White		834	51.2
Education ≤12 yr		555	34.1
Smoking
Never		955	58.6
Ex-smoker		241	14.8
Current smoker		422	25.9
Use of oral contraceptives		520	31.9
BMI (kg/m2)
<25		1057	64.9
25–29		332	20.4
>29		229	14.1
Blood pressure (mm Hg)
Systolic blood pressure	106.5 ± 9.88
Diastolic blood pressure	66.9 ± 8.92
Hypertension		53	3.3
Diabetes		12	0.7

CAC

Of the 1363 women with CAC measurements at yr 20, 146 (10.7%) were CAC positive. Table 2 shows associations of SHBG, TT, and FT with CAC. SHBG levels were inversely related to CAC; SHBG levels above the median were less common among CAC-positive women (37.7%) than among women with no CAC (52.3%), yielding a multivariable OR of 0.59 (95% CI = 0.40–0.87; P = 0.017) comparing those above vs. below the median of SHBG. The logarithmic transformation of SHBG levels as a continuous variable was associated with CAC (multivariable adjusted OR of 0.78, 95% CI = 0.64–0.96; P = 0.017 per sd increase in log-transformed SHBG). Restricting the analysis to women who were premenopausal at yr 16 did not alter the estimate of effect (adjusted OR of 0.56, 95% CI = 0.37–0.85). The results were similar when the analysis was restricted to women with at least two measurements of SHBG (n = 1119; OR = 0.58; 95% CI = 0.37–0.90; P = 0.015) or based on the median of yr 2 (n = 1080; OR = 0.54; 95% CI = 0.34–0.86; P = 0.009), the medians of the mean of yr 2 and 10 (n = 1295; OR = 0.58; 95% CI = 0.39–0.87; P = 0.009), or the medians of mean SHBG from yr 10 and 16 (n = 1242; OR = 0.67; 95% CI = 0.45–0.99; P = 0.044). Additional adjustments for LDL-cholesterol, systolic blood pressure, and HOMA-IR did not materially change the association (OR = 0.57; 95% CI = 0.38–0.85; P = 0.007). Exclusion of measurements taken when women were using oral contraceptives did not materially affect the results (OR = 0.64; 95% CI = 0.43–0.95; P = 0.026). Similarly, controlling for ever use of oral contraceptives or oligomenorrhea did not change the results. No interaction was found between levels of SHBG and use of oral contraceptives or race (data not shown).

Table 2.

Association between SHBG, FT, and TT (means of yr 2, 10, and 16) and CAC at yr 20, CWS

Analyte	CAC negative, n = 1217 (%)	CAC positive, n = 146 (%)	Model 1 (age and BMI)			Model 2 (adjusting for age, BMI, race, OC use, and smoking)
			OR	95% CI	P for trend	OR	95%CI	P for trend
SHBG (nmol/liter)
Q1 (10.50–21.50)	23.6	30.8	1			1
Q2 (21.67–28.67)	24.1	31.5	1.01	0.65–1.58		0.97	0.62–1.53
Q3 (29.00–37.67)	25.6	19.2	0.62	0.37–1.04		0.59	0.35–0.99
Q4 (38.00–82.00)	26.8	18.5	0.59	0.35–1.00	0.017	0.57	0.32–1.00	0.017
Serum levels above the median (29.0–82.0 nmol/liter)	52.3	37.7	0.61	0.42–0.87	0.008	0.59	0.40–0.87	0.008
Excluding OC use			0.64	0.43–0.94	0.023	0.64	0.43–0.95	0.026
TT (ng/dl)
Q1 (5.00–20.50)	26.3	25.3	1			1
Q2 (20.51–30.67)	24.5	24.7	1.03	0.63–1.69		0.90	0.55–1.49
Q3 (30.68–42.50)	25.1	25.3	1.04	0.64–1.70		0.88	0.53–1.47
Q4 (42.51–175.33)	24.1	24.7	1.08	0.66–1.78	0.757	0.95	0.56–1.59	0.823
Serum levels above the median	49.2	50.0	1.04	0.74–1.48	0.809	0.96	0.67–1.38	0.830
Excluding OC use			1.02	0.69–1.50	0.936	0.96	0.65–1.42	0.825
FT (unbound) (ng/dl)
Q1 (0.01–0.13)	26.4	24.7	1			1
Q2 (0.13–0.21)	25.2	22.6	0.88	0.53–1.45		0.78	0.47–1.32
Q3 (0.21–0.31)	24.6	26.7	1.15	0.70–1.87		1.03	0.61–1.73
Q4 (0.31–1.10)	23.8	26.0	1.08	0.65–1.79	0.541	0.94	0.55–1.61	0.901
Serum levels above the median	48.4	52.7	1.13	0.77–1.65	0.530	1.19	0.83–1.70	0.340
Excluding OC use			1.11	0.76–1.62	0.590	1.15	0.79–1.68	0.456

OC, Oral contraceptive; Q, quartile. 

Among women without CAC at yr 15, 84 women had new onset of CAC at yr 20. The OR for the association between SHBG (above vs. below the median) and incident CAC, adjusted for age, BMI, race, oral contraceptive use, and smoking was 0.50 (95% CI = 0.30–0.85; P = 0.010). Additional adjustment for systolic blood pressure, LDL-cholesterol, and HOMA-IR at baseline did not affect the association (OR = 0.48; 95% CI = 0.29–0.81; P = 0.006). Restricting the analysis to women with at least two SHBG measurements, the OR was 0.43 (95% CI = 0.24–0.76; P = 0.004). Analyzing the association of SHBG in yr 2 and 10, the OR was 0.54 (95% CI = 0.32–0.91; P = 0.020), but it did not reach statistical significance when only SHBG in yr 10 and 16 were included (OR = 0.66; 95% CI = 0.39–1.09; P = 0.105).

No associations were found between levels of TT or FT and CAC in yr 20, including incident CAC (Table 2).

IMT

In our population, 1381 women had STD-IMT measurements at yr 20. In logistic regression models comparing the highest quartile of STD-IMT with the lower three as the outcome, SHBG quartiles showed a monotonic inverse association, with OR of 0.77, 0.72, and 0.56 for the second, third, and fourth (highest) quartiles of SHBG, respectively (P for trend = 0.005) (Table 3). SHBG levels above the median were associated with an OR of 0.74 (95% CI = 0.56–0.98; P = 0.033) for the highest quartile of STD-IMT. Additional adjustments for systolic blood pressure, LDL-cholesterol, and HOMA-IR at baseline had minimal effect on this association (OR = 0.71; 95% CI = 0.53–0.95; P = 0.023). Results were similar, but nonsignificant, when only yr-2 SHBG levels were entered into the models (OR = 0.74; 95% CI = 0.52–1.05; P = 0.090). Restricting the population to premenopausal women had no effect on the association (OR = 0.72; 95% CI = 0.53–0.98); neither did the exclusion of samples obtained when women were using oral contraceptives or controlling for oligomenorrhea (not shown). Including only those with at least two SHBG measurements did not alter the results (OR = 0.73; 95% CI = 0.53–0.99; P = 0.044).

Table 3.

Association between SHBG, FT, and TT (means of yr 2, 10, and 16) and carotid IMT (highest quartile vs. lower three quartiles), CWS

	Model 1 (adjusted for age and BMI)			Model 2 (model 1 and race, smoking, and birth control pill usea)
	OR	95% CI	P for trend	OR	95% CI	P for trend
SHBG (ng/dl)
Q1 (10.50–21.50)	1.00	Reference	<0.001	1.00	Reference	0.005
Q2 (21.67–28.67)	0.76	0.54–1.08		0.77	0.54–1.09
Q3 (29.00–37.67)	0.66	0.46–0.95		0.72	0.50–1.04
Q4 (38.00–82.00)	0.51	0.35–0.75		0.56	0.37–0.84
Serum levels above the median	0.67	0.52–0.88	0.004	0.74	0.56–0.98	0.033
Excluding oral contraceptives use	0.67	0.51–0.89	0.006	0.74	0.55–0.98	0.037
TT (ng/dl)
Q1 (5.00–20.50)	1.00	Reference	0.534	1.00	Reference	0.254
Q2 (20.51–30.67)	1.04	0.73–1.48		1.01	0.69–1.47
Q3 (30.68–42.50)	0.86	0.60–1.23		0.86	0.59–1.26
Q4 (42.51–175.33)	0.94	0.66–1.35		0.83	0.55–1.23
FT (ng/dl)
Q1 (0.01–0.13)	1.00	Reference	0.243	1.00	Reference	0.513
Q2 (0.13–0.21)	1.29	0.88–1.90		1.16	0.78–1.73
Q3 (0.21–0.31)	1.54	1.05–2.28		1.30	0.86–1.94
Q4 (0.31–1.10)	1.23	0.82–1.84		0.84	0.54–1.29

IMT is a composite, standardized, IMT measurement of the CCA, ICA, and bifurcation (STD-IMT). 

^aWhere relevant. 

Neither FT nor TT were associated with carotid STD-IMT (Table 3). Results remained similar after exclusion of oral contraceptive use, restricting the study population to premenopausal women, or adjusting in addition for LDL-cholesterol, systolic blood pressure, and HOMA-IR (data not shown).

We further explored the association between SHBG and carotid IMT as continuous variables. In general linear models, a significant inverse association was found for SHBG with IMT (β_lnSHBG, per sd increment = −0.057; 95% CI = −0.098 to −0.016; P = 0.006). Additional adjustment for baseline systolic blood pressure, LDL-cholesterol, and HOMA-IR did not alter the association (β = −0.063; 95% CI = −0.104 to −0.022; P = 0.002). Similar estimates were observed when the study population was restricted to premenopausal women (β = −0.052; 95% CI = −0.094 to −0.010; P = 0.015) and to women with at least two SHBG measurements (β = −0.057; 95% CI = −0.104 to −0.011). Analysis of yr-2 SHBG in relation to IMT yielded a nonsignificant association (β = −0.032; 95% CI = −0.082–0.018; P = 0.211).

Discussion

In this population-based cohort study of young adult women followed for 18 yr, SHBG levels measured in young adulthood were negatively associated with two markers of subclinical CVD, CAC and carotid IMT. These associations were independent of BMI and HOMA-IR. In contrast, testosterone (either total or free) levels were not associated with either CAC or carotid IMT.

Previous studies have not assessed prospectively the associations of SHBG and serum androgens with CAC in young adult and middle-aged women. In a large cross-sectional study of 1947 postmenopausal women, no association was found between androgens and SHBG and presence of CAC, but among women with CAC, SHBG was positively associated with the extent of CAC (27). Other studies of androgens and CAC include small case-control studies in which associations were measured cross-sectionally among women with PCOS (5,6,7). In one of these studies, SHBG was associated with CAC only in the univariate analysis but not after controlling for BMI (5). Two other studies did not show any statistically significant associations of FT, TT, or SHBG with CAC, although this might be due to lack of power (6,7). Differences in design and study populations could account for the differences in results; the present study used a prospective design and included mostly premenopausal women with low prevalence of CAC. This low prevalence precluded further analyses related to the extent of subclinical disease.

In our study, SHBG was inversely associated with the STD-IMT, a combined measure of all three carotid locations combined. In a population-based, cross-sectional study of 483 middle-aged women (of whom 62% were pre- or perimenopausal), no association was found between testosterone or free androgen index and CCA-IMT, and the inverse association of SHBG disappeared in multivariable analysis that included, in addition to age and day of the cycle, ethnicity, site, smoking status, anthropometric measurements, systolic blood pressure, serum lipid levels, glucose, and insulin (34). Similarly, a longitudinal study, nested within a clinical trial, of 180 postmenopausal women (21), found no associations for FT and TT with progression of CCA-IMT and an inverse association for SHBG in univariate analysis that disappeared with adjustment for cholesterol levels. A recent large population-based cross-sectional study of postmenopausal women demonstrated a positive association of total and bioavailable testosterone, but not SHBG, with CCA-IMT (27). A cross-sectional study of 101 pre- and postmenopausal women in Italy (20) suggested an inverse association for FT with CCA-bulbar IMT and with a combination of all IMT measurements. In a nested case-control study within the Atherosclerosis Risk in Communities cohort, postmenopausal women who had TT levels at the highest quartile had an OR of 0.34 (95% CI = 0.16–0.70) for the highest 5th percentile of a combined IMT measure (22). In small case-control studies comparing women with and without PCOS, PCOS was associated with increased carotid IMT measurements in most (8,10,35), but not all (36), studies. These studies yielded mixed results for androgens/SHBG, with suggested associations for carotid IMT with SHBG (8,35), TT (8,10), and FT (10) among women with PCOS.

Although SHBG was believed to be only a transport glycoprotein, there is growing evidence suggesting that SHBG may have an independent biological function. A possible direct effect of SHBG on diabetes mellitus was suggested by a recent report based on data from the Women’s Health Study and the Physicians’ Health Study II. Low levels of SHBG were associated with increased risk of diabetes both in postmenopausal women and men. In accordance, the study demonstrated association of SHBG gene single-nucleotide polymorphisms with risk of diabetes in directions that corresponded to the single-nucleotide polymorphism-associated SHBG levels (17). Mendelian randomization analysis supported a causal effect for SHBG in the incidence of diabetes among postmenopausal women. Another recent study also supports an independent role of SHBG in diabetes among postmenopausal women (18). In a study of coronary artery disease, long repeats in the SHBG gene promoter (the (TAAAA)n) were associated with low SHBG and high testosterone and with increased severity of coronary artery disease on angiography (37). SHBG has been suggested to act through the steroid signal transduction system of cell membranes (38). Although there is no study to our knowledge examining SHBG receptors in cell membranes of the heart or blood vessels, a recent study showed the presence of SHBG receptors in smooth muscles of the uterus (39).

One of the limitations of this study is the lack of evaluation of all steroid sex hormones, in particular estradiol. SHBG is a glycoprotein that binds to testosterone and with weaker affinity to estradiol. The association of SHBG with subclinical disease may represent a protective effect of estrogens on CVD. SHBG levels are positively correlated with serum estrogens (19), partly through a positive feedback of serum estrogens on the synthesis of SHBG in the liver. Endogenous estradiol levels have been negatively associated with CCA adventitial diameter in a cross-sectional study of pre- and postmenopausal women (34). Another limitation of our study lies in the indirect method of estimating FT, suggesting perhaps that the sensitivity of SHBG measurements in representing the levels of bioavailable androgens in women is higher than the actual measurements of serum TT and FT. Nevertheless, in a previous study using the same setting and laboratory methods, we have demonstrated a positive association for FT with BMI (12).

Comparisons between studies are complicated by lack of standardization of laboratory methods across studies. Interpretation of comparisons is further complicated by differences in design (prospective vs. cross-sectional and case-control studies) and by differences in study populations, especially with regard to menopausal status. Nevertheless, these limitations in previous studies contribute to the main strengths of the current study: the prospective evaluation of a large, young, largely premenopausal population, with a population-based recruitment, and a long-term follow-up. Additional prospective studies are needed to confirm our results and to shed light on whether low SHBG is merely a risk marker or a risk factor for CVD in women.