Hemoglobin A1c Is Associated With Increased Risk of Incident Coronary Heart Disease Among Apparently Healthy, Nondiabetic Men and Women

Background Hemoglobin A1c (HbA1c), a time‐integrated marker of glycemic control, predicts risk of coronary heart disease (CHD) among diabetics. Few studies have examined HbA1c and risk of CHD among women and men without clinically elevated levels or previously diagnosed diabetes. Methods and Results We conducted parallel nested case–control studies among women (Nurses' Health Study) and men (Health Professionals Follow‐up Study). During 14 and 10 years of follow‐up, 468 women and 454 men developed incident nonfatal myocardial infarction (MI) and fatal CHD. Controls were matched 2:1 based on age, smoking, and date of blood draw. For these analyses, participants with a history of diabetes or HbA1c levels ≥6.5% at baseline were excluded. Compared with HbA1c of 5.0% to <5.5%, those with an HbA1c of 6.0% to <6.5% had a multivariable‐adjusted relative risk (RR) of CHD of 1.90 (95% CI 1.11 to 3.25) in women and 1.81 (95% CI 1.09 to 3.03) in men. The pooled RR of CHD was 1.29 (95% CI 1.11 to 1.50) for every 0.5%‐increment increase in HbA1c levels and 1.67 (95% CI 1.23 to 2.25) for every 1%‐increment increase, with the risk plateauing around 5.0%. Furthermore, participants with HbA1c levels between 6.0% and <6.5% and C‐reactive protein levels >3.0 mg/L had a 2.5‐fold higher risk of CHD compared with participants in the lowest categories of both biomarkers. Conclusions Our findings suggest that HbA1c is associated with CHD risk among apparently healthy, nondiabetic women and men and may be an important early clinical marker of disease risk.

D iabetes is a well-known risk factor for cardiovascular disease (CVD), 1 and although the exact mechanisms are unknown, angiopathy from dysglycemia is likely a primary cause. Glycated hemoglobin, assessed clinically by hemoglobin A1c (HbA 1c ), is a time-integrated marker of average blood glucose concentration increasingly used in the screening for and management of diabetes, and it is more closely related to the risk of complications than are single or episodic measures of glucose levels. [2][3][4] Despite the clinical utility of HbA 1c , several recent large randomized trials of intensive HbA 1c lowering among high-risk diabetics have not demonstrated lower risk of macrovascular events. [5][6][7][8] Currently, the degree to which mild elevations of HbA 1c not in the diabetic range are associated with risk of coronary heart disease (CHD) is unclear.
The association between dysglycemia and risk of CVD may start at levels that are only modestly elevated, well below the glucose threshold for diabetes. 9,10 The majority of studies conducted among nondiabetics have included participants with HbA 1c ≥6.5%, 11,12 the threshold set for diabetes diagnosis by the American Diabetes Association 2 and World Health Organization. 4 In several studies, the association between HbA 1c and CHD is attenuated and no longer significant after adjustment for well-known cardiovascular risk factors. 11,13 Thus, it remains unclear whether HbA 1c is a clinically relevant marker of CHD risk in apparently healthy, nondiabetic populations, or if it is a proxy for dyslipidemia, hyperinsulinemia, or other metabolic pathways. In the present study, we examined the association between HbA 1c and risk of incident CHD in 2 prospective nested case-control studies of health professionals without diabetes or without HbA 1c concentrations ≥6.5%.

Study Population
The Nurses' Health Study (NHS) and the Health Professionals Follow-Up Study (HPFS) are prospective cohort investigations among 121 700 female US registered nurses aged 30 to 55 years at baseline in 1976 (NHS) and 51 529 male US health professionals aged 40 to 75 years at baseline in 1986 (HPFS). Information about health and disease is assessed biennially, and information about diet is assessed every 4 years via self-administered questionnaires. 14, 15 Between 1989 and 1990, a blood sample was requested from all participants in the NHS, and it was returned from 32 826 women. Similarly, between 1993 and 1995, a blood sample was returned from 18 225 men in the HPFS. Participants who provided blood samples were similar to those who did not, albeit somewhat younger. In the NHS, among women without CVD or cancer before 1990, we identified 468 women with incident nonfatal myocardial infarction (MI) or fatal CHD between the date of blood draw and June 2004. In the HPFS, we identified 454 men with incident nonfatal MI or fatal CHD between the date of blood draw and return of the 2004 questionnaire. Using risk-set sampling, 16 controls were randomly selected on a 2:1 basis matched for age, smoking, date of blood draw, and fasting status (women only) from participants free of CVD at the time the case was diagnosed. To assess men and women without diabetes, participants with a history of diabetes or HbA 1c levels ≥6.5% at baseline (153 women; 123 men) were excluded from further analyses. We used 6.5% as the cutoff because this value has been recommended as a level defining diabetes by the American Diabetes Association 2 and the World Health Organization 4 and is the level at which complications of diabetes such as retinopathy arise.
The study protocol was approved by the Institutional Review Board of the Brigham and Women's Hospital and by the Harvard School of Public Health Human Subjects Committee Review Board.

Assessment of CHD
MI was confirmed by study physicians blinded to participants' exposure status if it met World Health Organization criteria (symptoms plus either diagnostic electrocardiographic changes or elevated levels of cardiac enzymes). 17,18 Deaths were identified from state vital records and the National Death Index or reported by the participant's next of kin or the postal system. Fatal CHD was confirmed by use of hospital records or on autopsy or if CHD was listed as the cause of death on the death certificate, if it was the underlying and most plausible cause, and if evidence of previous CHD was available.

Assessment of Other Factors
Anthropometric data, lifestyle behavior, and diet were derived from the questionnaire administered in 1990 to women and 1994 to men, with missing information substituted from previous questionnaires. Body mass index (BMI) was calculated as body weight divided by height squared. Average nutrient intake was computed using the semiquantitative food frequency questionnaire. Physical activity was expressed as metabolic equivalent-hours. The questionnaires and the validity and reproducibility of measurements have been previously reported. 15,19

Measurement of Biochemical Variables
Blood samples were collected in liquid sodium heparin blood tubes for women and EDTA tubes for men, placed on ice packs, stored in polystyrene foam containers, returned to our laboratory via overnight courier, and centrifuged and aliquoted for storage in liquid nitrogen freezers (À130°C or colder). The first nested case-control set included cases and controls selected from 1990 to 1998 for NHS and 1994 to 2000 for HPFS and were previously documented. 20 These samples were assayed in 2002. Since then, additional nested case-control sets with follow-up through 2004 have been completed for both NHS and HPFS and were sent for assaying using similar methods in 2007. All analyses were performed in a laboratory certified by the Centers for Disease Control and Prevention/ National Heart, Lung, and Blood Institute Lipid Standardization Program with commercially available analytic systems.
HbA 1c was measured by turbidimetric immunoinhibition using packed red cells (Roche Diagnostics), which is a standard approved by the US National Glycohemoglobin Standardization Program and the Food and Drug Administration for clinical use, and traceable to the Diabetes Controls and Complications Trial. The day-to-day variability at HbA 1c values of 5.5 and 9.1 were 1.9% and 3.0%, respectively. In a substudy of 83 participants measured 3 years apart, HbA 1c was highly correlated between draws (Spearman=0.88) and had excellent intraclass correlations (0.73). C-reactive protein (CRP) concentrations were determined with an immunoturbidimetric high-sensitivity assay using reagents and calibrators from Denka Seiken with assay day-to-day variability between 1% and 2% and were previously shown to be largely unaffected by transport conditions and to be reproducible within persons over time. 21,22 Total, high-density lipoprotein (HDL), and directly assayed low-density lipoprotein cholesterol and triglycerides were measured using standard methods with reagents from Roche Diagnostics and Genzyme. Study samples were sent to the laboratory for analysis in randomly ordered batches, and the laboratory was blinded to case-control status. The coefficients of variation for all the assays were <10%.

Statistical Analyses
We analyzed the 2 cohorts separately. HbA 1c levels were categorized into quintiles based on the sex-specific distributions among the controls as well as according to categories (<5.0%, 5.0% to <5.5%, 5.5% to <6.0%, and 6.0% to <6.5%). In the categorical analyses, the HbA 1c 5.0% to <5.5% category was used as the reference because there were more participants than in the <5.0% category. Because of this study design, the odds ratio derived from the logistic regression directly estimates the incidence rate (hazard) ratio and, thus, the RR. 16,23 We analyzed the association between HbA 1c levels and risk of CHD using both conditional and unconditional logistic regression with adjustment for matching factors and laboratory batch. Because both analyses provided essentially the same results, we present unconditional logistic regression, which parallels the subgroup analyses. In our multivariable model, we further adjusted for parental history of MI before the age of 60 (yes/no), alcohol intake (nondrinker; 0.1 to 4.9, 5.0 to 14.9, 15.0 to 29.9, or ≥30.0 g/d; or missing), physical activity (quintiles), total cholesterol:HDL cholesterol ratio (quintiles), BMI in categories (<20, 20 to 24.9, 25 to 29.9, 30 to 34.9, and ≥35 kg/m 2 ), and postmenopausal hormone therapy (yes/no, women only). Additional adjustment for hypertension (yes/no) at baseline, triglyceride levels (quintiles), and CRP levels (quintiles) was also conducted. Correlation coefficients were calculated using age-adjusted Spearman partial correlation coefficients. To test for linear trend, we used the median HbA 1c levels in the control categories as a continuous variable. In addition, we conducted analyses per 0.5%-increment increase in HbA 1c levels. We examined the possibly nonlinear relation between HbA 1c and CHD nonparametrically with restricted cubic splines. 24 Tests for nonlinearity used the likelihood ratio test, comparing the model with only the linear term to the model with the linear and the cubic spline terms. To pool the RR estimates for men and women, we used the weighted average of estimates using the DerSimonian and Laird random effects model. 25 Because we are using a single measure of HbA 1c to predict future disease over long-term follow-up, fluctuations in baseline biological risk factors over time may lead to an underestimation of the true association in the form of "regression dilution bias." 26,27 To account for potential changes in our baseline measure over time, we also present RR estimates corrected for potential regression dilution bias using the model described by Clarke et al. 28 In brief, the b coefficients and standard errors for the entire cohort were divided by the regression dilution ratio, based on the within-person correlation coefficient calculated from the subset of participants with repeated measurements, to obtain the corrected risk estimates for the usual relation of HbA 1c levels and CHD.
All P values presented are 2-tailed, and P values <0.05 were considered statistically significant. All analyses were performed using SAS version 9 (SAS Institute).

Results
Cardiovascular risk factors were higher or more frequent among the cases compared with the controls in both men and women ( Table 1). As expected, cases were more likely to have parental history of early MI, history of hypercholesterolemia, and history of hypertension. Also, cases were more likely to have higher CRP levels and adverse lipid profiles compared with controls for both women and men. Age-adjusted Spearman partial correlation coefficients are presented among controls for women and men ( Table 2). HbA 1c percentage was positively correlated with CRP levels, triglycerides:HDL cholesterol ratio, triglycerides, and BMI and inversely correlated with HDL cholesterol and alcohol intake.
Among participants with HbA 1c levels <6.5% and without history of diabetes at baseline, greater HbA 1c levels were significantly associated with an increased risk of CHD in both women and men. Comparing extreme quintiles of HbA 1c levels, the multivariable-adjusted RR of CHD was 1.70 (95% CI 1.22 to 2.36) in the pooled analysis, with a P for trend of <0.001 (Table 3). Because additional adjustment for history of hypertension, triglycerides, and CRP levels did not significantly attenuate the risk estimates, we refer to model 3 in multivariable-adjusted and pooled analyses in Table 3.
We further examined the association using fixed cut points. Compared with HbA 1c category 5.0% to <5.5% (reference), the pooled multivariable-adjusted RRs for CHD were 1.14 (95% CI 0.82 to 1.59) for <5.0%, 1.34 (95% CI 1.08 to 1.68) for 5.5% to <6.0%, and 1.85 (95% CI 1.28 to 2.69) for 6.0% to <6.5% (Figure 1). A significantly elevated risk of CHD was observed among nondiabetic women and men with HbA 1c levels of ≥5.5%. Even in these populations restricted to lower HbA 1c levels, we found that for every 0.5%-increment increase in HbA 1c levels, the RR of CHD was 1.32 (95% CI 1.05 to 1.65) in women and 1.27 (95% CI 1.04 to 1.55) in men, and 1.29 (95% CI 1.11 to 1.50) when pooled. This association did not differ by health and lifestyle factors such as age, smoking status, BMI <25 and ≥25 kg/m 2 , history of hypertension, or parental history of early MI (Table S1). We further examined the joint effect of CRP levels based on clinical cut points for CRP and categories of HbA 1c . Of note, in the lower risk group with CRP levels ≤3.0 mg/L, a 0.5%-increment increase of HbA 1c was significantly associated with risk of incident CHD among women (RR 1.42, 95% CI 1.04 to 1.94) and men (RR 1.28, 95% CI 1.03 to 1.60). In pooled analyses, participants in the highest CRP and HbA 1c categories had an RR of CHD of 2.49 (95% CI 1.32 to 4.71) compared with participants in the lowest of both biomarkers (Figure 2).
A single biomarker at baseline may be a better proxy for exposure over the first several years and possibly become less correlated with exposure over a longer period. When we stratified according to the midpoint of follow-up, the association was strongest among cases and controls in the first 7 years of follow-up. For every 0.5%-increment increase in HbA 1c levels in the earlier follow-up (<7 years) only, the multivariable-adjusted RR of incident CHD was 1. 39

Discussion
In these prospective nested case-control studies in 2 cohorts of US health professionals, we found that HbA 1c was predictive of CHD risk in participants without diagnosed diabetes and with HbA 1c concentrations <6.5%. Men and women with an HbA 1c of 6.0% to <6.5% had an 85% higher risk of CHD than those with levels of 5.0% to <5.5%. The positive association persisted when adjusted for traditional cardiovascular risk factors, including hyperlipidemia, anthropometric variables, and CRP. Our results suggest that HbA 1c is an important marker of risk across a broad range of levels and that the causal etiologic pathway may be complex and include mechanisms other than lipids or inflammation.
The relationship of HbA 1c with CHD has been studied in several other nondiabetic populations 11,12,[29][30][31][32] with inconsistent results, possibly due to diverse study protocols, a wide range of length of follow-up, inconsistent exclusion criteria, and differences in cohort demographics and outcome definitions. Two previous studies in particular have examined HbA 1c in relation to risk of CHD in participants without diabetes or clinically elevated HbA 1c levels. 13   (368 cases of incident MI, stroke, or coronary revascularization) comparing the highest (≥5.50%) with the lowest (<5.23%) quartile of HbA 1c . 13 However, this association was attenuated (RR 1.00; 95% CI 0.65 to 1.54) after adjustment for BMI, blood pressure, CRP, low-density lipoprotein and HDL cholesterol, and triglycerides. 13 The authors note that the same results were observed when the end point of CHD was examined alone, although with limited power because the CHD analysis was limited to 136 cases. In a recent analysis from the Atherosclerosis Risk in Communities (ARIC) study, the risk factor-adjusted hazard ratios for CHD in participants with HbA 1c of 5.5% to <6.0% and of 6.0% to <6.5% were 1.23 (95% CI 1.07 to 1.41) and 1.78 (95% CI 1.48 to 2.15), respectively, compared with participants with baseline HbA 1c concentrations of 5.0% to <5.5%. 29 Our results were similar in magnitude to this study. Furthermore, we showed a significant joint effect of mildly elevated HbA 1c and CRP levels on the risk of incident CHD among nondiabetic men and women. Interestingly, we observed a joint association similar to that of Sander et al, 33 who reported that among nondiabetics, the combination of elevated HbA 1c and CRP was associated with advanced early carotid atherosclerosis progression and a 3-fold increased risk of major vascular events. This association between elevated HbA 1c and early atherosclerosis progression supports the hypothesis that HbA 1c may be causally related to CHD even at low levels. 33 Hyperglycemia, in addition to diabetes, can lead to the formation of advanced glycosylated end products, which have been shown to induce production and secretion of inflammatory cytokines. 34 Indeed, HbA 1c has been reported to predict CRP in established diabetes. 35 In addition, nitric oxide bioavailability, a marker of intimal reactivity, has been shown to be reduced by both CRP 36 and HbA 1c . 37 This joint effect was observed at even mildly elevated levels of both biomarkers and may influence CHD risk through mechanisms other than diabetes, atherosclerosis, and inflammation.
A recent meta-analysis 38 examined the association between HbA 1c and risk of CHD in persons without diabetes, although most studies included participants with HbA 1c well beyond the established clinical threshold of 6.5%. Using data from 9 cohorts (1639 cases), the meta-analysis reported an RR of 1.20 (95% Cl 1.10-1.31) for each 1%-increment increase in levels of HbA 1c . 38 This risk estimate was not adjusted for established CHD risk factors and was weaker than our own. These differences may be due to the inherent weakness of meta-analyses, where pooling results from many studies may lead to random error from differences in lab methods, study design, and residual confounding. In our nested prospective studies, the pooled RR of CHD for men and women was 1.67 (95% CI 1.23 to 2.25) per 1%-increment increase in HbA 1c , even after detailed adjustment for other risk factors. among women and men according to categories of HbA 1c levels. *Adjusted for age, smoking status, date of blood draw, fasting status (in NHS only), laboratory batch, family history of myocardial infarction before age of 60 years, alcohol intake, physical activity, NSAID use, total cholesterol:HDL cholesterol ratio, BMI, and hormone replacement therapy use among women. CHD indicates coronary heart disease; HbA 1c , hemoglobin A1c; NHS, Nurses' Health Study; NSAID, nonsteroidal antiinflammatory drug; HDL, high-density lipoprotein; BMI, body mass index. Figure 2. Multivariable-adjusted* relative risk of incident CHD among women and men according to categories of HbA 1c and clinical cut points of CRP levels. *Adjusted for age, smoking status, date of blood draw, fasting status (in NHS only), laboratory batch, family history of myocardial infarction before age of 60 years, alcohol intake, physical activity, NSAID use, total cholesterol:HDL cholesterol ratio, BMI, and hormone replacement therapy use among women.
In the present study, even a 0.5%-increment increase of HbA 1c was significantly associated with risk of incident CHD among both women and men. Although a 0.5%-increment decrease in HbA 1c in patients undergoing treatment for diabetes 39 and morbid obesity 40 has proven health benefits, there are no published experimental trials that show a similar decrease in HbA 1c leads to a reduction in CHD or CVD end points in the general population. Nevertheless, the most effective ways of reducing HbA 1c , namely dietary changes, increased physical activity, and weight loss, have been shown to have beneficial effects on other cardiovascular risk factors. HbA 1c in nondiabetic individuals may be successfully reduced through nonpharmacologic lifestyle interventions, as demonstrated in the Finnish Diabetes Prevention Study, 41,42 Diabetes Prevention Program 43 and Outcomes Study, 44 and Look AHEAD study. 45 In a small intervention study of nondiabetic obese males, the mean HbA 1c levels were significantly lower after the weight loss intervention compared with the beforeintervention levels (4.9% versus 5.1%), 46 and in a randomized controlled trial of male construction workers, lifestyle intervention aimed at energy balance was significantly associated with improved cardiovascular risk factors, including body weight, blood pressure, and lipid and HbA 1c levels. 47 Although whether lowering HbA 1c lowers risk of future CHD among nondiabetic individuals has not been directly tested, working to optimize cardiovascular health in individuals with mildly higher HbA 1c seems prudent and clinically important.
Strengths of the present analysis include the long duration of follow-up, the prospective nature of the data collection, replication in a second cohort, and comprehensive lifestyle and health factor data used for detailed multivariable adjustment. A few limitations should be addressed. First, the potential exists for measurement error in lifestyle data ascertained from self-reported questionnaires. However, the reproducibility and validity of the self-reported questionnaire have been welldocumented in these 2 populations of health professionals. 15,48 Second, our study was conducted in 2 cohorts of middle-aged, predominantly white health professionals and may not be generalizeable to other populations. However, the relationship between HbA 1c and CHD risk was similar in a recent community-based study among whites and African Americans. 29 Third, we only had a single measurement of HbA 1c at baseline and did not have glucose measurements. However, one advantage of using HbA 1c is that it is stable in stored samples and reflects glycemic control over an average of 3 months rather than measures such as glucose, which are susceptible to daily fluctuations. 49 In addition, we observed excellent within-person reproducibility in repeated samples measured 3 years apart and conducted additional analyses correcting for potential changes in the baseline measurement over time.
In summary, we found that HbA 1c is a predictor of CHD risk among apparently healthy women and men without diabetes and with HbA 1c <6.5%. Further examination is warranted to determine whether reducing HbA 1c among otherwise intermediate-and low-risk individuals without diagnosed diabetes will prevent incident CHD events.

Sources of Funding
This study was supported by grants (HL35464, CA55075, HL34594, CA87969, and CA49449) from the National Institutes of Health. Dr Pai was supported in part by National Institutes of Health grant HL088372. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.