Vasomotor Symptoms and Insulin Resistance in the Study of Women's Health Across the Nation

Abstract

Context:

Emerging research suggests links between menopausal hot flashes and cardiovascular disease risk. The mechanisms underlying these associations are unclear, due to the incomplete understanding of the physiology of hot flashes.

Objective and Main Outcome Measures:

We examined the associations between hot flashes/night sweats and glucose and insulin resistance over 8 yr, controlling for cardiovascular risk factors and reproductive hormones.

Design, Setting, and Participants:

Participants in the Study of Women's Health Across the Nation (SWAN) (n = 3075), a longitudinal cohort study, were ages 42–52 yr at entry. Women completed questionnaires (hot flashes, night sweats: none, 1–5 d, ≥6 d, past 2 wk), physical measures (blood pressure, height, weight), and a fasting blood draw [serum glucose, insulin, estradiol (E2), FSH] annually for 8 yr. Hot flashes/night sweats were examined in relation to glucose and the homeostasis model assessment (HOMA) in mixed models, adjusting for demographics, cardiovascular risk factors, medications, and E2/FSH.

Results:

Compared to no flashes, hot flashes were associated with a higher HOMA_log index [vs. none; hot flashes, 1–5 d: % difference (95% confidence interval), 2.37 (0.36–4.43), P = 0.02; and ≥6 d: 5.91 (3.17–8.72), P < 0.0001] in multivariable models that included body mass index. Findings persisted adjusting for E2 or FSH, and were similar for night sweats. Findings were statistically significant, yet modest in magnitude, for the outcome glucose.

Conclusions:

Hot flashes were associated with a higher HOMA index, an estimate of insulin resistance, and to a lesser extent higher glucose. Metabolic factors may be relevant to understanding the link between hot flashes and cardiovascular disease risk.

Vasomotor symptoms (VMS) are classic symptoms of the menopausal transition, experienced by upwards of 70% of women living in the United States (1). VMS have important quality of life implications because women reporting VMS consistently show poorer sleep quality (2), more negative mood (3), and impaired quality of life (4). However, many questions remain about the basic physiology of VMS and their association with health outcomes.

An area of active inquiry is the relation between VMS and cardiovascular disease (CVD) risk. Data from both the Women's Health Initiative hormone therapy trial (5) and the Heart and Estrogen Replacement Study (6) showed elevated risk for clinical CVD with hormone use among older women with moderate/severe VMS at baseline relative to women with no/mild VMS. In the Study of Women's Health Across the Nation (SWAN), VMS were associated with higher subclinical CVD (7, 8). However, findings have been mixed (9) or observed only among certain women (8, 10). Thus, much remains to be learned about links between VMS and CVD risk.

Although several investigations have examined associations between VMS and CVD risk factors such as blood pressure (11, 12), relations between VMS and metabolic abnormalities such as insulin resistance have received less attention. One cross-sectional study of a Swedish midlife sample showed symptoms of “sweats” to be associated with elevated (nonfasting) blood glucose (11). However, associations did not persist after adjustment for body mass index (BMI). Considering BMI in relation between insulin resistance and VMS is particularly important given that higher BMI is a potent risk factor for insulin resistance and is associated with greater VMS reporting in perimenopausal and early postmenopausal women (1). Other work postulates alterations in glucose transport across the blood-brain barrier as a trigger of VMS (13), although this hypothesis has limited empirical support. No work has examined the relation between VMS and fasting blood glucose and insulin resistance in a community sample of women followed over time.

The aim of this investigation was to test whether VMS would be associated with higher fasting blood glucose and a higher homeostasis model assessment (HOMA) index, an estimate of insulin resistance (14). We tested hypotheses controlling for key risk factors, including BMI, and with adjustment for reproductive hormones such as estradiol (E2) and FSH. In an exploratory fashion, we additionally examined any variations in associations between VMS and glucose/HOMA by BMI, race/ethnicity, or menopausal stage, given that these factors may modify relations between VMS and other CVD risk factors (8, 15). We examined these associations in SWAN, a multisite longitudinal study of midlife women followed through the menopausal transition, a unique cohort in which to address these questions.

Subjects and Methods

Sample

SWAN is a multiracial/ethnic cohort study designed to characterize biological and psychosocial changes over the menopausal transition. Details of SWAN design and procedures have been reported elsewhere (16). Briefly, each SWAN site recruited non-Hispanic Caucasian women and women belonging to a predetermined racial/ethnic minority group (African-American women in Pittsburgh, Boston, Michigan, Chicago; Japanese women in Los Angeles; Hispanic women in New Jersey; Chinese women in the Oakland area of California). Los Angeles, Pittsburgh, and New Jersey sites used random-digit-dialed sampling from banks of telephone numbers, and Boston, Chicago, Michigan, and Oakland sites selected randomly from lists of names or household addresses. Select sites supplemented primary sampling frames to obtain adequate numbers of racial/ethnic minority women. SWAN protocols were approved by the institutional review boards at each site. Each participant provided written informed consent.

Baseline eligibility criteria for SWAN included being aged 42–52 yr, having an intact uterus and at least one ovary, not being pregnant or lactating, not using oral contraceptives or hormone therapy, and having a menstrual cycle in the 3 months before the interview. A total of 3302 women were enrolled, and annual clinic assessments began in 1996–1997. This investigation was an analysis of associations between VMS, glucose, and HOMA from baseline through the seventh annual visit.

Of the 3302 women enrolled in SWAN, 46 women were excluded due to missing data for VMS (n = 5) or glucose and insulin (n = 41), 100 due to baseline use of insulin or a glucose-lowering medication, and 81 women due to reporting having had a stroke, angina, or myocardial infarction at baseline. Data were censored during the follow-up period at the time of hysterectomy/oophorectomy (due to potential differing pattern of VMS), stroke/myocardial infarction, or initiation of insulin or glucose-lowering medication. Data from visits in which pregnancy or hormone use (hormone therapy, oral contraceptives) was reported in the previous year were excluded. Women with self-reported diabetes were included in primary models but were censored at the time of reported diagnosis in sensitivity analyses. The 227 women excluded from primary models at baseline had a higher BMI, more hot flashes and night sweats, higher anxiety and depressive symptoms, lower physical activity, higher glucose and HOMA, less education, and less alcohol use, and were more likely to be perimenopausal (vs. premenopausal), use cardiovascular medication (heart, blood pressure-lowering, blood-thinning medications), or to be a smoker than women included (P < 0.05). These differences were expected given sample exclusions listed above. Primary models included 3075 women.

Design and procedures

Vasomotor symptoms

VMS were assessed via questionnaire at each SWAN visit on the same day as the blood draw. Women responded to two questions that asked separately how often they experienced: 1) hot flashes and 2) night sweats in the past 2 wk (not at all, 1–5 d, 6–8 d, 9–13 d, every day; categorized as none, 1–5 d, ≥6 d for analysis due to small cell sizes in the highest symptom categories). Hot flashes and night sweats were considered separately due to their differential pattern of associations with the outcomes considered here.

Blood assays

Phlebotomy was performed in the morning after overnight (minimum of 10 h) fast. Participants were scheduled for venipuncture on d 2–5 of a spontaneous menstrual cycle. Two attempts were made to obtain a d 2–5 sample. If a timed sample could not be obtained (as menstrual cycles became less regular, samples tied to the early follicular phase were less feasible), a random fasting sample was taken within 90 d of the annual visit. Blood was maintained up to 1 h at 4 C until separated, frozen (−80 C), and sent on dry ice to the Clinical Laboratory Improvement Amendments-certified CLASS laboratory at the University of Michigan (Ann Arbor, MI; E2) and Medical Research Laboratories (Highland Heights, KY; glucose, insulin) for analysis. For budgetary reasons, glucose and insulin assays were completed at SWAN baseline and annual visits 1, 3, 4, 5, 6, and 7.

Glucose was measured using a hexokinase-coupled reaction on a Hitachi 747-200 (Roche Molecular Biochemicals Diagnostics, Indianapolis, IN). Serum insulin was measured using a RIA (DPC Coat-a-Count; Diagnostic Products Corporation, Los Angeles, CA) procedure and monitored as part of the monthly quality assurance program by the Diabetes Diagnostic Laboratory at the University of Missouri. HOMA, an index derived from glucose and insulin measures, reflected insulin resistance [(fasting insulin*fasting glucose)/22.5] (17).

E2 assays were performed on the ACS-180 automated analyzer (Bayer Diagnostics Corporation, Tarrytown, NY) using a double-antibody chemiluminescent immunoassay with a solid phase anti-IgG immunoglobulin conjugated to paramagnetic particles, anti-ligand antibody, and competitive ligand labeled with dimethylacridinium ester. The E2 assay modifies the rabbit anti-E2-6 ACS-180 immunoassay to increase sensitivity, with lower limit of detection = 6.6 pg/ml and inter- and intraassay coefficients of variation of 10.6 and 6.4%, respectively (18). Duplicate E2 assays were conducted, and results were reported as the arithmetic mean. FSH assays were performed using a two-site chemiluminometric immunoassay, with inter- and intraassay coefficients of variation of 11.4 and 3.8%, respectively, and lower limit of detection = 1.1 mIU/ml.

Covariates

Race/ethnicity (determined in response to: “How would you describe your primary racial or ethnic group?”) and education (categorized as less than vs. college completion or greater) were determined from the SWAN screening interview. Age, smoking status (current vs. past/never) (19), depressive/anxious symptoms, physical activity, alcohol use, menopausal status, and medication use/health conditions were derived from questionnaires and interviews administered during annual visits. Age, smoking, alcohol use, physical activity, BMI, menopausal status, and medication use/health conditions were considered as time-varying covariates. Physical activity was assessed at baseline and annual visits 3, 5, and 6 via a modified Kaiser Permanente Health Plan Activity Survey (20), with values for visits 1, 4, and 7 carried forward from the last completed observation. Alcohol use was the reported average weekly number of servings of beer, wine, liquor, or mixed drinks. Menopausal status was obtained from self-reported bleeding patterns over the year preceding the visit, categorized as premenopausal (bleeding in the previous 3 months with no change in cycle predictability in the past year), early perimenopausal (bleeding in the previous 3 months with decrease in cycle predictability in the past year), late perimenopausal (<12 and >3 months of amenorrhea), or postmenopausal (≥12 months of amenorrhea) at each visit. Consistent with our prior work (7, 21), baseline depressive symptoms (assessed via the Center for Epidemiologic Studies Depression scale) (22) and anxiety symptoms (sum score of number of days in the past 2 wk, 0 = no days, 4 = every day; reporting irritability or grouchiness, feeling tense or nervous, heart pounding or racing, or feeling fearful for no reason) were considered as covariates. Anxiety, rather than depressive symptoms, was included in final models due to its more consistent associations with study outcomes. BMI was derived from annual physical measures and considered as a continuous variable in analyses. Participants were censored from analyses at the time of reported insulin/glucose-lowering medication use or of reported heart attack or stroke. Reported hypertension/use of blood pressure-lowering medications, use of lipid-lowering medications, or use of other cardiovascular medications (reported use of medication for a heart condition or for blood thinning) were covaried.

Data analyses

Glucose, HOMA, E2, and FSH values were natural log-transformed for analysis. Baseline differences between included/excluded participants were tested using Wilcoxon rank-sum and χ² tests. Univariate associations between covariates and each outcome were evaluated at baseline using linear regression. Associations between hot flashes/night sweats and each outcome were estimated in a random intercepts linear mixed model. An autoregressive error correlation structure was selected based upon standard model fit statistics. Due to the skewed nature of the outcomes, effect sizes are reported as the percentage of difference in the outcome for a given level of VMS relative to no VMS. Models were adjusted for age and site, and next additionally for covariates selected based upon previously documented and present associations with outcomes at P < 0.05. Serum E2 was added to covariate-adjusted models with blood draw timing (in vs. out of cycle d 2–5 window). FSH was considered instead of E2 in secondary models. Because cycle day of blood draw and menopausal status were collinear (only early peri- and premenopausal women had menstrual cycles to provide a timed sample), they were considered as a composite variable (pre- or perimenopausal timed sample, pre- or perimenopausal untimed sample, late perimenopausal, or postmenopausal). For time-varying covariates, values concurrent with the outcome measure time point were used. An α = 0.05 was adopted for analyses of hypothesized relations. Interactions between hot flashes/night sweats and race/ethnicity, menopausal status, and BMI were examined together as cross-product terms in multivariable models, with a Bonferroni correction applied to exploratory analyses of multiple interactions (α = 0.008 for interactions). Residual analysis and diagnostic plots were conducted to verify model assumptions of normality. Analyses were performed with SAS version 9.2 (SAS Institute, Cary, NC). Models were two-sided.

Results

At baseline, the participants were on average 46 yr old, overweight, and nonsmoking, with half of the sample in the early perimenopause and half in the premenopause. Approximately half of the sample was comprised of Caucasian women, one fourth were African-American, and the remainder of the sample consisted of Hispanic, Chinese, or Japanese women (Table 1).

Table 1.

Baseline characteristics of analytic sample by hot flash status, SWAN

	Frequency of hot flashes
	None	1–5 d	≥6 d
n	2262	575	227
Age, yr (mean ± sd)c	45.6 ± 2.6	46.2 ± 2.8	46.9 ± 2.8
Racec
African-American	539 (23.8)	191 (33.2)	93 (41.0)
Chinese	192 (8.5)	35 (6.1)	13 (5.7)
Hispanic	170 (7.5)	57 (9.9)	18 (7.9)
Japanese	230 (10.2)	38 (6.6)	12 (5.3)
Caucasian	1131 (50.0)	254 (44.2)	91 (40.1)
Educationc
≤High school	496 (22.1)	168 (29.5)	66 (29.3)
Some college	666 (29.7)	204 (35.8)	96 (42.7)
≥College graduate	1078 (48.1)	198 (34.7)	63 (28.0)
Menopausal statusc
Premenopausal	1323 (60.1)	221 (39.0)	78 (34.8)
Early perimenopausal	879 (39.9)	346 (61.0)	146 (65.2)
BMI, kg/m2 (mean ± sd)c	27.4 ± 6.9	28.9 ± 7.1	30.6 ± 7.4
Current smokerc	333 (14.9)	115 (20.3)	54 (23.8)
Alcohol consumption
<1 time/month	948 (50.8)	227 (55.0)	87 (52.4)
≥1 time/month to <2 times/wk	524 (28.1)	102 (24.7)	47 (28.3)
≥2 times/wk	393 (21.1)	87 (20.3)	32 (19.3)
Physical activity score (mean ± sd)a	7.7 ± 1.8	7.6 ± 1.7	7.5 ± 1.7
Hypertension/antihypertensive medication usec	364 (16.4)	137 (24.3)	66 (29.5)
Use of heart medications or anticoagulants	36 (1.6)	12 (2.1)	6 (2.7)
Use of lipid-lowering medicationsc	10 (0.4)	3 (0.5)	8 (3.5)
Depressive symptoms, median (IQR)c	7.0 (3.0, 14.0)	10.0 (5.0, 18.0)	13.0 (6.0, 24.0)
Anxious symptoms, median (IQR)c	2.0 (1.0, 3.0)	2.0 (1.0, 4.0)	4.0 (2.0, 6.0)
E2 (pg/ml), median (IQR)b	58.1 (34.2, 89.8)	46.5 (31.8, 84.2)	50.5 (27.5, 88.4)
FSH (mIU/ml), median (IQR)c	15.1 (10.5, 24.5)	17.2 (11.5, 29.0)	22.1 (12.4, 46.9)
Glucose (mg/dl), median (IQR)c	90.0 (86.0, 97.0)	93.0 (86.0, 101.0)	95.0 (87.0, 103.5)
HOMA index, median (IQR)c	1.78 (1.28, 2.75)	2.25 (1.46, 3.65)	2.30 (1.52, 4.27)

Data are presented as number (percentage), unless specified otherwise. Statistical comparisons were conducted using ANOVA or Kruskal Wallis for continuous variables, and χ² and Fisher exact test for categorical variables; numbers may not add up to reflect the frequency of VMS due to different missing values at baseline.

^aP < 0.05;

^bP < 0.01;

^cP < 0.0001.

For HOMA, both hot flashes and night sweats were associated with a higher HOMA index (Table 2 and Fig. 1). These associations remained after adjustment for multiple covariates, including BMI. Associations also remained after additional adjustment for E2. Findings were similar with adjustment for FSH in multivariable models (for hot flashes: 1–5 d (% difference [95% confidence interval (CI)]), 1.66 (−0.33–3.87), P = 0.5; ≥6 d, % difference (95% CI), 4.40 (1.69–7.19), P = 0.007; relative to none; and for night sweats: 1–5 d, % difference (95% CI), 2.73 (0.73–4.78), P = 0.06; ≥6 d, % difference (95% CI), 3.47 (0.42–6.60), P = 0.04; relative to none). For hot flashes, associations were most pronounced for women reporting hot flashes of at least 6 d in the prior 2 wk.

Table 2.

Relation between vasomotor symptoms and HOMA

	HOMA index
	Model 1	Model 2	Model 3
Hot flashes
None‡	—	—	—
1–5 d	4.31 (2.21, 6.44)c	2.37 (0.36, 4.43)a	1.66 (−0.33, 3.70)
6+ d	9.66 (6.80, 12.59)c	5.82 (3.09, 8.63)c	4.31 (2.47, 7.99)b
Trend P value	<0.0001	<0.0001	0.002
Night sweats
None‡	—	—	—
1–5 d	4.51 (2.42, 6.65)c	3.11 (1.09, 5.16)b	2.78 (0.78, 4.82)b
6+ d	9.51 (6.26, 12.86)c	4.67 (1.59, 7.84)b	3.33 (0.30, 6.46)a
Trend P value	<0.0001	0.0003	0.004

Data are expressed as percentage difference (95% CI). HOMA index was log-transformed. Regression coefficients were back-transformed using the formula (100 * (exp^(β) − 1) to calculate percentage difference and 95% CI in level of HOMA for VMS group relative to no VMS. Model 1: adjusted for age and site; model 2: age, site, race, education, menopausal status, alcohol consumption, physical activity, anxiety, hypertension/antihypertensive medication use, lipid-lowering medication use, use of heart or anticoagulant medication, BMI; and model 3: model 2 plus E2_log.

^‡Reference group.

^aP < 0.05;

^bP < 0.01;

^cP < 0.0001.

Fig. 1.

HOMA index by hot flash status across SWAN visits.

For glucose, significant associations were also observed between hot flashes and glucose [1–5 d, % difference (95% CI), 0.34 (−0.14–0.82), P = 0.2; and ≥6 d, % difference (95% CI), 1.22 (0.57–1.87), P = 0.002; relative to none], and between night sweats and glucose [1–5 d, % difference (95% CI), 0.58 (0.10–1.06), P = 0.02; and ≥6 d, % difference (95% CI), 1.06 (0.33–1.79), P = 0.004; relative to none] in multivariable models (covariates: age, site, race, education, menopausal status, alcohol consumption, physical activity, anxiety, hypertension status/blood pressure-lowering medication use, use of medications for a heart condition or blood thinning, lipid-lowering medication use, BMI). Findings also persisted with additional adjustment for E2 or FSH (data not shown). However, it is notable that the magnitude of associations for glucose was quite small.

We next tested for interactions by race, BMI, or menopausal stage. There were no interactions between these factors and either VMS variable for HOMA. However, for glucose, there was some evidence of significant interactions between hot flashes and race, hot flashes and menopausal stage, and night sweats and BMI in multivariable models (P < 0.05). Associations between VMS and glucose were slightly more pronounced among women who were African-American, obese, or in the early or late perimenopause (compared with women who were Caucasian, lean, or in the postmenopause; data not shown). However, the magnitude of the overall associations for VMS and glucose, as well as the variations in these relations by BMI, race, and stage was very small, and therefore these interactions should be regarded with caution.

We conducted several sensitivity analyses. Although all analyses censored women at the time of using a glucose-lowering medication, we additionally censored women at the time of self-reported diabetes, and significant associations remained (data not shown). In addition, we included women with heart disease or stroke in primary models, and results were unchanged (data not shown). Furthermore, instead of E2 or FSH, we adjusted for SHBG or testosterone; findings were largely unchanged (data not shown).

Discussion

In this study of women transitioning through the menopause, VMS were associated with elevated HOMA and to a lesser extent fasting blood glucose over a period of approximately 8 yr. These associations were not explained by potential confounders, including BMI, nor by E2 or FSH. Associations were most apparent for women reporting VMS for 6 d or more in the prior 2 wk.

The present study found elevations in both glucose and HOMA with VMS in a community-based cohort study. Although the magnitude of the effect sizes for relations between VMS and glucose were small, relations between VMS and HOMA were more clinically meaningful. For example, the median HOMA value in the high hot flash group would meet the threshold for insulin resistance established by Radikova et al. (23). This study importantly advances the existing literature on these relations. One other epidemiological study showed that “sweats” were cross-sectionally associated with elevated blood glucose, and whereas associations persisted controlling for many confounders and E2, they did not persist controlling for BMI (11). Lee et al. (24) found the somatic symptom subscale (which encompassed several symptoms including VMS) of the Menopause Rating Scale was positively associated with the metabolic syndrome. Although early work suggested slight acute elevations in glucose during hot flashes (25), a more recent study of 10 women indicated the opposite (13). Thus, the literature surrounding the nature of the relation between VMS and glucose/insulin resistance has been conflicting, and acute changes in glucose during VMS may vary from tonic levels among women with VMS. The present study, with its large multiracial/ethnic community sample and examination of VMS and an estimate of insulin resistance assessed over time adds to the literature on relations between VMS and levels of glucose and HOMA.

The mechanisms underlying associations between VMS and insulin resistance are not entirely clear, due in part to the limited understanding of the underlying physiology of VMS. First, VMS and insulin resistance have shared risk factors, most notably elevated BMI—important given the importance of overweight/obesity in elevated glucose and insulin resistance as well as positive associations observed between BMI and VMS in SWAN (1). However, neither BMI, other shared risk factors, nor E2, FSH, testosterone, or SHBG explained the associations observed here. Although residual confounding cannot be ruled out in this observational study, it is possible that other physiological systems might be at play. For example, VMS have been linked to an autonomic nervous system balance favoring sympathetic predominance/parasympathetic withdrawal (26–28). This profile is associated with elevated glucose availability and decreased pancreatic insulin production, is observed among individuals with diabetes (29), and may in fact precede the development of diabetes (30). A related system, the hypothalamic-pituitary-adrenal axis, has also been implicated as a potential link between VMS and insulin resistance (31), although the hypothalamic-pituitary-adrenal axis requires further characterization in both conditions. Thus, several pathways requiring further elucidation may explain the observed relations between VMS and glucose/insulin resistance.

This study can be interpreted in the context of the larger literature about VMS and CVD risk. In the SWAN Heart Study, women with VMS had evidence of poorer endothelial function, elevated aortic calcification, and elevated intima media thickness (7, 8). In the full SWAN cohort, VMS are also associated with a procoagulant hemostatic profile (21). Furthermore, in the Women's Health Initiative and the Heart and Estrogen/Progestin Replacement hormone trials, older women with moderate/severe VMS at baseline appeared to be particularly vulnerable to an adverse cardiovascular impact of hormone therapy (5, 32). However, much remains to be learned about any association between VMS and CVD risk, such as variations by clinical characteristics of the individual (8), or the timing of VMS in a woman's lifespan (10). Understanding potential mechanisms underlying any links between VMS and CVD risk is critical to better understanding the nature of any relations between VMS and CVD risk.

We observed some evidence of interactions of relations between VMS and glucose for BMI, race/ethnicity, and menopausal stage, with somewhat stronger associations observed among the higher BMI, African-American, and perimenopausal women. The magnitudes of the association between VMS and glucose, as well as of variations in these associations by subject characteristics (race, BMI, menopausal stage) were small and not consistently observed across independent and dependent variables. These interactions should therefore not be over-interpreted. However, higher BMI and African-American women tend to have higher levels of glucose, VMS, and risk for diabetes and metabolic syndrome (1, 33); the late perimenopause may be an important time with respect to vascular remodeling (34); and relations between VMS and subclinical CVD may be most pronounced among higher BMI women (8). Future work may consider whether relations between VMS and glucose/insulin resistance are most relevant to women with some existing CVD risk.

This study had several limitations. First, VMS were measured via a brief symptom inventory in which women recall their VMS over the prior 2 wk. Although feasibility issues in large epidemiological investigations prohibits more detailed measures, these questionnaire measures do not have the precision of diary or physiological measures and are thereby subject to more error. Moreover, blood samples and corresponding measures of glucose, HOMA, and E2 were obtained approximately annually, and may not have fully captured the women's hormonal state, particularly with respect to the highly fluctuating reproductive hormonal states of midlife menopausal women. Measures of insulin resistance were obtained via blood sample rather than gold standard measures (e.g. euglycemic hyperinsulinemic clamp), which are not feasible in epidemiological investigations such as this one. Dietary information was also not included in this analysis [although few dietary factors appear related to VMS in SWAN (1, 35)]. Furthermore, the effect sizes for glucose were small, indicating that other factors beyond VMS are important in determining these levels. Moreover, the temporal or causal nature of relations between VMS and glucose/HOMA cannot be determined from this analysis aimed at examining the overall relations between VMS and glucose/HOMA. The temporal and causal nature of these relations should be examined in future work. Finally, although these analyses adjusted for many shared risk factors, issues of residual confounding or underlying third variables related to both VMS and glucose/insulin resistance cannot be ruled out in this observational study.

This study had several strengths. This is the first study to examine associations between VMS and glucose/HOMA over time. Measures were obtained annually over approximately 8 yr. This sample was large, community-based, and included multiple ethnic groups. A wide range of demographic, physical, endocrine, and psychosocial confounders were assessed and controlled. In short, this analysis represents the largest and strongest test of associations between VMS and glucose/insulin resistance to date.

In summary, VMS were associated with insulin resistance, as measured by the HOMA index, over a period of approximately 8 yr. These associations were not explained by BMI or by other potential confounders. These findings may contribute to ongoing efforts to better understand any mechanisms linking hot flashes to cardiovascular health.