Longitudinal Changes in Sex Hormone Binding Globulin (SHBG) and Risk of Incident Diabetes: The Study of Women’s Health Across the Nation (SWAN)

Low levels of sex hormone binding globulin (SHBG) and high levels of testosterone (T), indicative of serologic hyperandrogenism, have each been positively associated with incident diabetes in women (1–4). The menopause transition (MT) is accompanied by hormonal and metabolic changes and an increased risk of cardiovascular disease and diabetes (5,6). The increased risk of diabetes may be driven in part by the longitudinal changes in SHBG and T and the relative androgen excess occurring after menopause (7). A meta-analysis of nine cohort studies found that low levels of SHBG were associated with increased risk of diabetes in women (4); however, most of the cohort studies to date were conducted among postmenopausal women (2,8,9) and measured sex hormones at only one time point. Whether longitudinal changes in SHBG and T over the MT affect the risk of developing diabetes is unclear.

One study found that Black women had higher levels and Chinese women had lower levels of SHBG compared with White women (10), two subgroups that are also at higher risk of diabetes compared with White women (11). However, whether the associations between sex hormones and type 2 diabetes (T2D) risk vary by race/ethnicity is unclear. The Study of Women’s Health Across the Nation (SWAN), a multisite, multiracial/ethnic longitudinal study of the natural history of the MT includes approximately annual or biannual hormone measurements over a 17-year period as well as annual longitudinal follow-up for incident diabetes. We undertook the present analysis to evaluate the following hypotheses: 1) increases in SHBG and decreases in T over time since baseline (indicative of a less androgenic profile) would be associated with a decreased risk of diabetes; and 2) these relations may differ by race/ethnicity.

SWAN is a longitudinal cohort study of a racially and ethnically diverse cohort of women traversing from midlife to late adulthood. A total of 3,302 premenopausal and early perimenopausal women who were 42–52 years old in 1996–1997 were recruited from seven geographic sites across the U.S.: Boston, MA, Chicago, IL, Detroit area, MI, Los Angeles, CA, Newark, NJ, Oakland, CA, and Pittsburgh, PA. Women were eligible if they had an intact uterus and at least one ovary, reported at least one menstrual period and no exogenous hormone use in the 3 months prior to recruitment, were not currently pregnant or lactating, and identified their primary race/ethnicity as Black (at the Boston, Chicago, Detroit, and Pittsburgh sites), Japanese (at the Los Angeles site), Hispanic (at the Newark site), Chinese (at the Oakland site), or White (at all sites). The sampling and recruitment strategies have been previously described in greater detail (12). All sites used common protocols, which were approved by the Institutional Review Board at each of the participating institutions. All participants provided written informed consent.

Women completed nearly annual follow-up study visits through 2015 (follow-up visit 15). Women with diabetes at baseline (n = 152), using hormones at baseline (n = 5), and with no follow-up visits (n = 193) were excluded. The final analytic sample thus included 2,952 women (Fig. 1). The New Jersey site had a hiatus in data collection, unrelated to the scientific integrity of SWAN, for visits 6–8 and 10 and 11; therefore, a sensitivity analysis excluding the New Jersey site was performed (described below in Statistical Analyses)

Standardized questionnaires were used to collect information on participants’ age, race/ethnicity (13), sociodemographic characteristics (educational attainment) (13), health behaviors (smoking status) (13), and medical characteristics (MT status) (14), menopause hormone use (13–17), and lipid-lowering medication use (14,16,17). BMI (kg/m²) was calculated using height, measured by stadiometer, and weight, measured using a calibrated scale. BMI was categorized using standard adult BMI cut points for Black and White women (18) as underweight: <18.5 kg/m²; normal weight: 18.5–24.9 kg/m²; overweight: 25–29.9 kg/m²; and obese: ≥30 kg/m², and using Asian-specific BMI cut points for Chinese and Japanese women (19) as underweight: <18.5 kg/m²; normal weight: 18.5–22.9 kg/m²; overweight: 23–27.4 kg/m²; and obese: ≥27.5 kg/m².

Phlebotomy was performed in the morning following an overnight fast. Participants were scheduled for one venipuncture on days 2–5 of a spontaneous menstrual cycle within 60 days of recruitment. Two attempts were made to obtain a day 2–5 sample. If a timed sample could not be obtained, a random fasting sample was taken within 90 days of recruitment. Blood was refrigerated prior to centrifugation 1 to 2 h after phlebotomy, and the serum was aliquoted, frozen, and batched for shipment to the central laboratory. Samples were cataloged and assayed continuously upon arrival.

The total T assay was a modified rabbit polyclonal anti-T ACS-180 immunoassay (Bayer Diagnostics Corp., Norwood, MA), with a lower limit of detection of 2.19 ng/dL. The interassay coefficients of variation for the total T assay are 11.34% (53.3 ng/dL, n = 37) and 9.7% (250.2 ng/dL, n = 37). The SHBG assays were developed on site at the Central Ligand Assay Satellite Services Laboratory at the University of Michigan using rabbit anti-SHBG antibodies, with a lower limit of detection of 1.95 nmol/L. T was only measured up to visit 10. E2 assays were conducted in duplicate using an ACS-180 automated analyzer. E2 concentrations were measured with a modified, off-line ACS-180 (E2–E6) immunoassay. Inter- and intraassay coefficients of variation averaged 10.6 and 6.4%, respectively, over the assay range, and the lower limit of detection was 1 pg/mL.

Serum glucose levels were measured at all clinic visits, except follow-up visits 9 and 10, using a hexokinase-coupled reaction (Boehringer Mannheim Diagnostics, Indianapolis, IN). Women were defined as having diabetes if they reported use of glucose-lowering medication at any time during the study, had two consecutive visits with fasting glucose ≥126 mg/dL while not on steroids, or had any two visits with self-reported diabetes and a visit with fasting glucose ≥126 mg/dL. The SWAN visit at which diabetes may have first been observed/detected (visit for incident diabetes) was defined as among women who used glucose-lowering medication or the first visit with serum glucose ≥126 mg/dL before first use of glucose-lowering medication; otherwise, the first visit with self-reported diabetes before first use of glucose-lowering medication; otherwise, the first visit at which the participant reported use of glucose-lowering medication. Among women who did not use glucose-lowering medication, the visit at which diabetes was first observed was defined as the first visit with fasting serum glucose ≥126 mg/dL while not on steroids. For 41 participants (8%) who had more than two missing visits with no fasting glucose or self-reported diabetes data between the first visit at which they used glucose-lowering medication or had high fasting glucose ≥126 mg/dL and the assigned visit as described above, the first diabetes visit was imputed as the mean visit between these two visits. The type of diabetes (i.e., type 1 or type 2) was not determined, but most of the cases of diabetes in this life stage can be assumed to be T2D.

Sociodemographic and medical characteristics at baseline were summarized for all women. Mean and SD were used to describe continuous variables. Frequency and percentage were used to describe categorical variables. To test hypothesis 1, complementary log-log discrete time survival models were used to model the association between time-varying SHBG and T and incident diabetes. Two approaches were used to capture the time-varying nature of SHBG and T. In the first, time-varying SHBG or T was the exposure, and the estimated effect was the adjusted hazard ratio (HR) and 95% CI for incident diabetes associated with a 5-unit increase in time-varying SHBG or T. Since exposures and outcomes were collected at annual visits, we used a discrete-time survival analysis framework. Participants were followed from the baseline visit to the first occurrence of the incident diabetes diagnosis, last observed visit, or end of study. Time-varying exposures and covariates were updated at each annual visit, and only values that were measured during a participant’s follow-up time were used in the analysis (Fig. 2). In the second, average change in SHBG or T relative to baseline was defined by the difference in SHBG or T values at the last observed visit (up to and including the visit where incident diabetes was captured or the end of follow-up) and the baseline divided by time-in-study. The average change in SHBG or T relative to the baseline (categorized by quartiles) was then included in the model as a time-invariant exposure.

All models were adjusted for study clinical site, race/ethnicity, baseline age, baseline SHBG or T, baseline MT status, and baseline BMI. Models for time-varying hormone levels additionally adjusted for cycle day of the blood draw, time-varying MT status, and BMI. Missing time-varying exposure and covariate values were imputed using the last value carried forward approach.

To test hypothesis 2, effect modification of time-varying and average change in SHBG by race/ethnicity was examined by testing the interaction term and then fitting separate models relative to baseline for each racial/ethnic stratum if the P value for the interaction term was <0.05. Parallel analyses were performed using T as the exposure. Multivariable models including both SHBG and T were fit to assess the association of time-varying SHBG with incident diabetes independent of T and vice versa.

In sensitivity analyses, the impact of the following were considered: log₂-transformation of SHBG; adjusting for waist circumference in lieu of BMI to better account for adiposity; additionally adjusting for education, baseline alcohol use, physical activity score (baseline and time-varying), and smoking (baseline and time-varying) to examine additional covariates; using only data through visit 10 was performed because T was only measured up to that visit; fitting models without baseline SHBG, T, or BMI; excluding the imputed data; removing data from the New Jersey site to assess the impact of the hiatus in data collection at that site, stratifying by menopausal status at baseline, and lagging the exposure variables to be limited to one visit before the outcome assessment.

SWAN participants were, on average, 46 (SD 2.7) years old at baseline (Table 1). By race/ethnicity, 27% of participants were Black, 8% Chinese, 7% Hispanic, 9% Japanese, and 48% White. The average BMI was 27.8 (SD 7.0) kg/m². Women were followed for an average of 13 years, and 376 (12.7%) developed diabetes during follow-up (Fig. 1). The median levels at baseline for SHBG were 41.6 nmol/L (interquartile range [IQR] 28.5–58.0), for T were 41.2 ng/dL (IQR 29.5–56.0), and for E2 were 55.2 pg/mL (IQR 33.2–88.5). The median rate of change for sex hormones was −0.2 nmol/L/year (IQR −1.5 to 1.3) for SHBG and −0.3 ng/dL/year (IQR −1.7 to 1.0) for T, suggesting overall levels were fairly stable or slightly decreasing over time.

A 5-unit higher concentration of time-varying SHBG was associated with an ∼12% lower risk of incident diabetes (adjusted [a]HR 0.88; 95% CI 0.85–0.91) adjusting for baseline age, race/ethnicity, cycle day of blood draw, study site, MT status (baseline and time-varying), and BMI (baseline and time-varying) (Table 2; model 1). The association between time-varying SHBG and incident diabetes persisted when further controlling for baseline and time-varying T and E2 (Table 2; model 2). In the fully adjusted models, a 5-unit higher concentration of time-varying T was associated with a slight increase in incident diabetes (aHR 1.01; 95% CI 1.00–1.03) (Table 2).

Compared with the lowest quartile for change in SHBG (−92.3 to −1.5 nmol/L), all other quartiles were associated with a decreased risk of diabetes (quartile [Q] 2 [> −1.5 to −0.2 nmol/L] aHR 0.16, 95% CI 0.12–0.22; Q3 [> −0.2 to 1.3 nmol/L] aHR 0.13, 95% CI 0.10–0.18; Q4 [>1.3–82.0 nmol/L] aHR 0.18, 95% CI 0.13–0.25) adjusting for baseline covariates and baseline SHBG (Table 2; model 1); the findings were attenuated but consistent when further adjusting for baseline T and E2 and the time-invariant change in T and E2 (Table 2; model 2). The aHR for incident diabetes comparing the highest quartile (>1.0 to 151.7 nmol/L increase per year) versus the lowest quartile (−91.8 to −1.7 nmol/L decrease per year) for rate of change in T was 0.59 (95% CI 0.42–0.84), adjusting for baseline covariates and baseline T (Table 2; model 1); the association was similar upon further adjusting for SHBG and E2 levels (Table 2; model 2).

We found statistically significant effect modification between time-varying SHBG levels and diabetes by race/ethnicity (P-interaction = 0.01). Across all racial-ethnic groups, a 5-unit increase in SHBG was associated with a decreased risk of developing T2D; however, the magnitude of the decreased risk of diabetes associated with increasing SHBG appeared slightly stronger among Hispanic and Chinese women and lower among Black and Japanese women (Fig. 3). No significant effect modification by race/ethnicity was observed between time-varying or rate of change in SHBG or T and BMI or MT status.

Results of all the following sensitivity analyses were similar (Supplementary Tables 1–5): log transformation of SHBG; adjusting for waist circumference in lieu of BMI; additionally adjusting for education, baseline alcohol use, physical activity score (baseline and time-varying), or smoking (baseline and time-varying); administrative censoring at visit 10; fitting models without baseline SHBG, T, or BMI; excluding the imputed data; excluding data from the New Jersey site; stratifying by menopausal status at baseline and lagging the exposure variables to be limited to one visit before the outcome assessment

We found that among midlife women, increasing levels of SHBG measured longitudinally were strongly associated with a decreased risk of incident diabetes, independent of other known diabetes risk factors. The association between increasing SHBG and decreased incidence of diabetes was also independent of changes in levels of T, suggesting SHBG may affect glucose tolerance independently of androgenicity. Women whose levels of SHBG and T increased over time had a decreased risk of incident diabetes compared with women who experienced a negative rate of change in SHBG and T levels indicative of decreasing levels over time.

Our study was uniquely able to examine longitudinal changes in sex hormones among midlife women during and after the MT in relation to incident diabetes. Menopause is an important transition in a woman’s life that is accompanied by an increased risk of cardiovascular disease and T2D (1,2). Changes in hormonal patterns in menopause, including the relative androgen excess, contribute to an increase in visceral adiposity that is associated with glycemic traits and therefore may influence the risk of diabetes (3,4). Our findings are consistent with other studies that have found low SHBG levels in women are associated with an increased risk of diabetes. A meta-analysis conducted by Muka et al. (4) revealed that in 13 population‐based studies with 1,912 incident cases of T2D, low SHBG was associated with increased risk of T2D in women, irrespective of MT status. However, prior studies to date only measured SHBG at a single point in time in relation to diabetes risk. We add to this literature by demonstrating that, among midlife women, longitudinal increases in SHBG and T were associated with a decreased risk of incident diabetes.

We observed wide variation in changes in SHBG over the MT, with some women experiencing increasing levels over the MT and others experiencing decreasing levels. Further research is needed to clarify which factors affect changes in sex hormones over the MT and whether modifying levels of SHBG can reduce diabetes risk long-term. In the Diabetes Prevention Program (DPP), randomization to an intensive lifestyle intervention increased SHBG levels in postmenopausal women and attenuated the decline in SHBG levels in premenopausal women; however, changes in SHBG levels were not independently associated with reductions in diabetes incidence over the 3 years of follow-up, after adjusting for adiposity (20). The DPP was conducted among individuals at high risk for diabetes, which may have limited the ability to see an additional impact of SHBG on diabetes risk. Future studies are needed to determine whether modifying SHBG levels in a representative sample of midlife women during the MT impacts diabetes risk.

The complex biological mechanisms that explain the association between circulating SHBG levels and the risk for diabetes are not fully understood. The primary function of SHBG is thought to be the binding of circulating hormones to regulate free sex hormone bioavailability to target tissues (21). The majority of circulating T is bound to SHBG, such that only the unbound or “free” fraction is capable of exerting effects in target tissues (22). Therefore, reduced SHBG levels in women are a surrogate marker of increased circulating active androgens. We found the association between SHBG and diabetes was independent of T; however, whether SHBG acts directly to affect the pathophysiology of metabolic diseases, such as diabetes, is unclear. SHBG is primarily produced in the liver. Low circulating SHBG may reflect an increased amount of liver fat content; androgen excess is an independent risk factor for nonalcoholic fatty liver disease in women (23). Further, circulating SHBG has been correlated more strongly with liver fat content than with total body and visceral fat mass or whole-body insulin sensitivity (24,25), supporting the novel concept that hepatokines are very important for the regulation of metabolism (26).

We found no association between time-varying T and incident diabetes overall; however, a positive rate of change in T, indicative of increasing T over time, was associated with a decreased risk of incident diabetes (4). Prior studies in postmenopausal women have shown that higher T is associated with an increased risk of diabetes. However, what factors drive the variation in annual rate of change in T over the MT is unclear. It is possible that the protective effect of a positive rate of change in T over the MT is due in part to corresponding changes in estradiol levels. However, we found no evidence of effect modification by MT status.

The strengths of our study include repeated measurements of serum SHBG, T, glucose, and insulin with up to 17 years of follow-up in a large, diverse cohort of midlife women followed over the MT.

However, the study also had several limitations. Results are not generalizable to women younger or older than the age range of our participants, from other racial/ethnic backgrounds, or to men. Also, we were unable to determine whether the associations between SHBG and incident diabetes were driven by differences in liver fat content or visceral adiposity.

In conclusion, increasing levels of SHBG and a positive rate of change in SHBG or T over the MT were associated with a decreased risk of incident diabetes. The association between SHBG and diabetes was independent of BMI and T levels, suggesting SHBG may have glucogenic properties beyond androgenicity. The association appeared to be stronger among Chinese women. It is unclear what factors were driving the changes in sex hormones during and after the MT. Future research is needed to determine whether modifiable risk factors are associated with changes in SHBG and T during and after the MT.

Acknowledgments. The authors thank the study staff at each site and all the women who participated in SWAN.

Funding. The Study of Women’s Health Across the Nation (SWAN) has grant support from the U.S. Department of Health and Human Services, National Institutes of Health (NIH), through the National Institute on Aging (U01AG012505, U01AG012535, U01AG012531, U01AG012539, U01AG012546, U01AG012553, U01AG012554, U01AG012495, and U19AG063720), the National Institute of Nursing Research (grant U01NR004061), and the NIH Office of Research on Women’s Health.

The content of this paper is solely the responsibility of the authors and does not necessarily represent the official views of the National Institute of Aging, National Institute of Nursing Research, Office of Research on Women’s Health, or the NIH.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. M.M.H. conceived and designed the analysis, interpreted the data, and drafted and critically revised the manuscript for important intellectual content. L.A.H., J.L., S.E.B., E.B.G., S.D.M., and S.R.E.K. interpreted data and drafted and critically revised the manuscript for important intellectual content. A.C. and C.L. conducted the statistical analyses. M.M.H. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at the 79th Scientific Sessions of the American Diabetes Association, San Francisco, CA, 7–11 June 2019.

Clinical Centers. The clinical centers and principal investigators (PI) of SWAN are as follows: University of Michigan, Ann Arbor—Carrie Karvonen-Gutierrez, PI 2021–present, Siobán Harlow, PI 2011–2021, and MaryFran Sowers, PI 1994–2011; Massachusetts General Hospital, Boston, MA—Sherri‐Ann Burnett‐Bowie, PI 2020–present, Joel Finkelstein, PI 1999–2020, and Robert Neer, PI 1994–1999; Rush University, Rush University Medical Center, Chicago, IL—Imke Janssen, PI 2020–present, Howard Kravitz, PI 2009–2020, and Lynda Powell, PI 1994–2009; University of California, Davis, Davis, CA/Kaiser—Elaine Waetjen and Monique Hedderson, PIs 2020–present, Ellen Gold, PI 1994–2020; University of California, Los Angeles, Los Angeles, CA—Arun Karlamangla, PI 2020–present, and Gail Greendale, PI 1994–2020; Albert Einstein College of Medicine, Bronx, NY—Carol Derby, PI 2011–present, Rachel Wildman, PI 2010–2011, and Nanette Santoro, PI 2004–2010; University of Medicine and Dentistry–New Jersey Medical School, Newark, NJ—Gerson Weiss, PI 1994–2004; and the University of Pittsburgh, Pittsburgh, PA—Rebecca Thurston, PI 2020–present, and Karen Matthews, PI 1994–2020.

NIH Program Office. SWAN investigators at the NIH Program Office are as follows: National Institute on Aging, Bethesda, MD—Rosaly Correa-de-Araujo, 2020–present, Chhanda Dutta, 2016–present, Winifred Rossi, 2012–2016, Sherry Sherman, 1994–2012, and Marcia Ory, 1994–2001; and National Institute of Nursing Research, Bethesda, MD—program officers.

Central Laboratory. Laboratory support for SWAN is as follows: University of Michigan, Ann Arbor—Daniel McConnell (Central Ligand Assay Satellite Services).

Coordinating Centers. The coordinating centers and PIs for SWAN are as follows: University of Pittsburgh, Pittsburgh, PA—Maria Mori Brooks, PI 2012–present, and Kim Sutton-Tyrrell, PI 2001–2012; and New England Research Institutes, Watertown, MA—Sonja McKinlay, PI 1995–2001.

Steering Committee. The SWAN Steering Committee comprises Susan Johnson, current chair, and Chris Gallagher, former chair.