Glycemic Control, Atherosclerosis, and Risk Factors for Cardiovascular Disease in Individuals With Diabetes

RESEARCH DESIGN AND METHODS

The Atherosclerosis Risk in Communities (ARIC) study is a community-based cohort study of nearly 16,000 people aged 45–64 years at baseline (1987–1989). All participants were selected from four U.S. communities: suburban Minneapolis, Minnesota; Forsyth County, North Carolina; Washington County, Maryland; and Jackson, Mississippi. The cohort in Jackson, Mississippi, was sampled and recruited to have an all-black study population. Details on the design and conduct of the ARIC are available elsewhere (8). For the present cross-sectional study, we analyzed data from the second examination (visit 2) of ARIC participants, the only visit for which A1C data are available, which took place from 1990 to 1992. All individuals with diagnosed or undiagnosed diabetes at visit 1 or 2 were eligible for this study. We excluded 51 people who were missing data on A1C, four people who were nonwhite or nonblack, and 222 additional people who did not fast for ≥8 h before visit 2. After exclusions, 2,060 subjects were included in this study.

Diabetes status

Participants were asked to fast for 12 h before the ARIC clinic visits and to bring all current medications to determine mediation use. Glucose was measured using the hexokinase method (9). Individuals were defined as having diagnosed diabetes if they self-reported a physician diagnosis of diabetes or were currently taking diabetes medication. Individuals were classified as having undiagnosed (unreported) diabetes if they had a fasting glucose level ≥126 mg/dl or a nonfasting glucose level ≥200 mg/dl at the first or second ARIC examination but did not self-report a physician diagnosis of diabetes and were not taking diabetes medication. We further classified participants with diagnosed diabetes into the following treatment categories: 1) no pharmacologic treatment, 2) insulin treatment, or 3) sulfonylurea treatment (oral hypoglycemic medication). Data for this study were collected in the early 1990s before metformin and thiazolidinediones were widely available.

A1C assay

Frozen whole blood samples from ARIC visit 2 were thawed and assayed for A1C using a Tosoh high-performance liquid chromatography instrument. The within-batch coefficient of variation for the Tosoh assay was 2.4%. We found that measurements from these long-term stored samples were extremely reliable when compared with measurements from the same samples conducted before long-term storage (n = 336, r = 0.97) (10,11).

Other variables of interest

Details have been previously described for measurement of lipids (12–14), determination of BMI (weight in kilograms divided by the square of height in meters), waist-to-hip ratio (15), and systolic and diastolic blood pressures (16). Smoking, hormone use (women only), and alcohol consumption (current, former, or never) were assessed by interview. Prevalent cardiovascular disease was based on self-report, ARIC clinical examination, or hospital records (17).

To assess carotid IMT, B-mode carotid ultrasound (Biosound 2000 II SA; Biosound, Indianapolis, IN) evaluations were completed on bilateral segments of the extracranial carotid arteries. The carotid artery was divided into three 1-cm regions and readers measured IMT within these regions. If the participants had missing IMT information from any carotid artery site, values were imputed for missing sites based on sex and race. Mean far wall IMT was also adjusted for reader differences and downward drift. Further details are described in previous articles (18,19).

Variables of interest were categorized using clinically relevant cut points wherever possible. BMI was categorized according to the classification system established by the National Institutes of Health (<25, 25.0–29.9, and ≥30.0 kg/m²) (20). Hypertension was defined as a systolic blood pressure of ≥140 mmHg, a diastolic blood pressure of ≥90 mmHg, or current use of antihypertensive medication. Carotid IMT and waist-to-hip ratio were categorized according to quartiles.

Statistical analysis

All analyses were stratified by undiagnosed (unrecognized) and diagnosed diabetes status. Pearson’s correlation coefficients and scatterplots were used to examine and present the association between A1C and fasting serum glucose.

Mean A1C levels were compared across categories of variables of interest after adjusting for age, sex, and race using a linear prediction model with each adjustment variable set to its mean value. In people with diagnosed diabetes, we also examined the association between A1C level and pharmacologic treatment status (none, sulfonylurea, or insulin). Mean A1C levels by presence of prevalent cardiovascular disease and quartile of IMT were compared separately in individuals with undiagnosed and diagnosed diabetes after adjustment for all other cardiovascular risk factors.

Multivariable linear regression models were used to assess the independent relationship between variables of interest and A1C after adjustment for relevant covariates. Diabetes treatment was assessed in a separate model because glucose-lowering treatment directly affects A1C levels and should not strictly be considered a “confounding” factor. Because information on duration of diabetes was missing for >30% of people with diagnosed diabetes, we did not include it in our final models. We did, however, conduct a subgroup analysis of people with diagnosed diabetes for which information on duration of diabetes was available. We also used a multivariable logistic regression model to compare the odds of being in the highest quartile of IMT (“thick IMT”) versus the lowest by quartile of A1C after adjustment for potential confounding factors. These results were displayed graphically (Fig. 2). Because the distributions of both A1C and IMT differed in individuals with undiagnosed and diagnosed diabetes, diagnosis-specific quartiles were used in this analysis. All statistical analyses were conducted using Stata 8.2 (Stata, College Station, TX).

RESULTS

In this community-based study of people with diabetes, the mean level of A1C in the total study population (diagnosed and undiagnosed diabetic patients combined) was 6.82 ± 2.10%. When stratified by diagnosis status, the average levels of A1C were 5.95 ± 1.41 and 7.56 ± 2.31% in individuals with undiagnosed (n = 949) and diagnosed (n = 1,111) diabetes, respectively.

Fasting glucose and A1C

As expected, the correlation between fasting glucose and A1C was high. Correlations were 0.84 overall and 0.87 and 0.81 for individuals with diagnosed diabetes and undiagnosed diabetes, respectively (Fig. 1). In people with diagnosed diabetes who were not receiving pharmacologic treatment, the correlation was 0.90. The correlations were 0.56 and 0.82 in individuals receiving insulin and sulfonylurea treatment, respectively.

Age-, sex-, and race-adjusted analyses

In age-, sex-, and race-adjusted analyses (Table 1), race, measures of adiposity (BMI and waist-to-hip ratio), smoking status, LDL and HDL cholesterol concentrations, plasma triglyceride levels, and glucose-lowering medications (diagnosed diabetes only) were all significantly associated with glycemic control as measured by A1C (P < 0.01). Current smoking and current alcohol consumption were significantly associated with lower A1C levels in individuals with diagnosed diabetes only. Mean A1C levels were also strongly associated with IMT after adjustment for age, sex, and race in individuals with diagnosed and undiagnosed diabetes (Table 2).

Multivariable analyses

Table 2 displays the results of the adjusted multivariable linear regression models in people with undiagnosed and diagnosed diabetes. All models were simultaneously adjusted for age, sex, race, and all other variables in the table except for current hormone use. Current hormone use was examined in a separate model of women only, which included all other variables in the table. Correlations between A1C and BMI and waist-to-hip ratio were similar, so only waist-to-hip ratio was included in the multivariable models.

African-American race was strongly associated with A1C levels even after adjustment for all cardiovascular disease risk factors in individuals with undiagnosed and diagnosed diabetes. In people with diagnosed diabetes, African-American race was associated with ∼1% higher mean A1C level compared with whites in this study, even after adjustment for potential confounding factors. HDL cholesterol (P < 0.01) was associated inversely with A1C in individuals with undiagnosed diabetes. LDL cholesterol was positively associated with A1C in individuals with diagnosed diabetes (P < 0.01). In both diagnosed and undiagnosed diabetic individuals, waist-to-hip ratio was strongly positively associated with A1C (P < 0.01). In people with diagnosed diabetes, being in the upper quartile of waist-to-hip ratio was associated with ∼0.8% higher A1C level. Plasma triglycerides were associated positively with A1C in individuals with diagnosed and undiagnosed diabetes in a graded fashion. In people with diagnosed diabetes, glucose-lowering medication use was strongly associated with glycemic control (P < 0.01). Individuals currently taking sulfonylurea treatment had an ∼2% higher mean A1C level than those with diagnosed diabetes who were not receiving any pharmacologic treatment (P < 0.01). People with diagnosed diabetes who were currently receiving insulin therapy had a 1.7% higher mean A1C level compared with individuals not taking medication (P < 0.01). There was some evidence that current alcohol consumption was associated with lower A1C, but this result was not statistically significant in any of the multivariable models.

In women, current hormone use was associated with lower A1C levels. Women with diagnosed diabetes who were currently taking hormones had ∼0.8% lower mean A1C levels compared with women not currently taking hormones (P < 0.01). In women with undiagnosed diabetes, those currently taking hormones had 0.4% lower mean A1C levels (P < 0.01).

In the subgroup analysis of people with diagnosed diabetes for which information on duration of diabetes was available (n = 696), we did not find any association between A1C and diabetes duration after adjusting for cardiovascular risk factors (analysis not shown). We also did not find any evidence that duration of diabetes was an important confounder in the relationship between A1C and any other variables of interest, and therefore, it was not included in any of the final models.

Although A1C was not significantly independently associated with prevalent cardiovascular disease in this study (data not shown), A1C was associated with higher carotid IMT in both undiagnosed and diagnosed diabetes. Figure 2 displays the odds ratios (ORs) of being in the highest quartile of IMT (“thick IMT”) versus the lowest quartile of IMT by quartile of A1C after adjustment for cardiovascular risk factors. In people with diagnosed and undiagnosed diabetes, the highest quartile of A1C was significantly associated with having a thick IMT. In individuals with undiagnosed diabetes, the adjusted ORs of thick IMT were 2.42 (95% CI 1.16–5.03), 2.32 (1.14–4.71), and 2.46 (1.16–5.22) comparing the second, third, and fourth quartiles of A1C with the lowest, respectively (P = 0.051). For individuals with diagnosed diabetes, a strong trend was observed. In a model adjusted for the same covariates, the ORs of thick IMT were 1.09 (0.58–2.04), 2.04 (1.08–3.87), and 2.62 (1.36–5.06) for the second, third, and fourth quartiles of A1C, respectively (P = 0.001). In model 2 (diagnosed diabetes), which included all covariates plus diabetes medication use, the ORs of being in the highest quartile of IMT were 1.18 (0.60–2.33), 2.17 (1.06–4.46), and 2.83 (1.34–5.96) for the second, third, and fourth quartiles of A1C compared with the first (P = 0.002).

CONCLUSIONS

African-American race, cholesterol levels, plasma triglycerides, and waist-to-hip ratio were all associated with A1C levels independently of each other and known cardiovascular risk factors. As reported in previous studies (21–24), African-American race was strongly associated with higher A1C levels even after adjustment for potentially confounding factors. Understanding why African Americans are more likely to have poor glycemic control in undiagnosed and diagnosed diabetes warrants further investigation in future epidemiologic and clinical studies.

A1C was strongly associated with atherosclerosis as measured by carotid IMT. Carotid IMT is a widely accepted marker of atherosclerosis. It closely reflects the atherosclerotic process and is strongly related to LDL cholesterol and other cardiovascular risk factors and predicts coronary heart disease events (25–28). Several recent studies have also shown that improvements in glycemic control can slow progression of atherosclerotic disease in individuals with type 1 (5,29,30) and type 2 diabetes (31). Our study provides evidence to suggest that glycemic control as measured by A1C is related to carotid IMT in middle-aged adults with diabetes independent of traditional cardiovascular risk factors. A previous case-control study in people without diabetes in the ARIC suggested that IMT case status was associated with A1C after controlling for age, sex, race, field center, examination date, smoking, BMI, hypertension, LDL and HDL cholesterol concentrations, education, fasting glucose, and fasting insulin (OR 1.88 [95% CI 0.9–4.1]) for the highest quartile of A1C compared with the lowest) (10). Previous studies have proposed that the glycation and oxidation of lipids and other proteins may contribute to the development of atherosclerosis in individuals with diabetes via the formation of advanced glycation end product and other related mechanisms (32–36).

A1C was not associated with prevalent cardiovascular disease in this study. Because this study was cross sectional, it is possible that individuals with prevalent cardiovascular disease (“survivors”) in this cohort had better risk factor profiles (including lower A1C) and less severe disease than people who died of cardiovascular disease before baseline and therefore could not be included in the study. Thus, associations observed here may be affected by “survival bias” and biased toward the null. Nonetheless, age-stratified analyses did not show a stronger relation between A1C and prevalent cardiovascular disease in younger compared with older subjects.

In this middle-aged community-based cohort, there was some evidence of an association between decreased A1C and increasing age in individuals with diagnosed diabetes. In previous studies of people with type 2 diabetes, associations between glycemic control and age have been mixed. Several studies have shown that younger age is associated with higher A1C levels (37–39), whereas other studies have shown a positive association (40) or no association (41).

Similar to our results, previous studies have shown a positive association between glycemic control and adiposity (39,42,43) and cholesterol levels (21,44). Studies in individuals without diabetes have also shown similar correlations between lipids and A1C (45–48).

In this cross-sectional analysis, the strong association between pharmacologic treatment and elevated A1C levels in individuals with diagnosed diabetes most likely reflects severity of diabetes and/or poor medical compliance in people taking sulfonylurea and insulin medication compared with individuals not receiving pharmacologic treatment.

Several previous studies suggested possible beneficial effects of moderate alcohol consumption on glycemic control (40,42,49–51). In our study, there was some evidence of an association between current alcohol consumption and decreased A1C, but this relationship did not persist after adjustment for cardiovascular risk factors.

Similar to our results, previous studies of postmenopausal women with diabetes have shown that hormone use is associated with better glycemic control (22,52–56). Although this could be a direct effect of hormone therapy on glucose metabolism (57,58), it could also reflect a “healthy user” effect because previous studies have shown that women taking hormone replacement therapy are more likely to have a higher education level, have better health status, and engage in health-promoting behaviors compared with women not taking hormones (59–61).

This study benefited from the rigorous methodology of the ARIC study and the availability of a wide range of risk factor data from a large community-based sample of people with diabetes. Because ARIC is a community-based cohort study, we were able to identify and examine individuals with both undiagnosed and diagnosed diabetes. Previous studies have examined associations between A1C and individual cardiovascular risk factors but few studies have had sufficient sample size or covariate information to be able to examine possible independent relationships after adjustment for multiple risk factors. To our knowledge, no previous study has compared the relationship between A1C and cardiovascular disease risk factors separately in people with diagnosed and undiagnosed diabetes. The availability of information on IMT also allowed us to assess the relationship between A1C and a measure of atherosclerosis, providing further evidence that elevated glucose levels may be contributing to the development of atherosclerosis and possibly subsequent cardiovascular events. There have been few studies of the relationship between A1C and atherosclerosis in individuals with type 2 diabetes.

Nonetheless, the cross-sectional design limited our ability to draw conclusions regarding the temporality of these associations. It is also important to note that during the time of the ARIC examinations for which data in this study were obtained (1990–1992), the criterion for the diagnosis of diabetes was a fasting glucose level ≥140 mg/dl. Thus, a number of people (∼20%) classified as diabetic in the present study would not have been classified as diabetic under the clinical criteria at the time. However, similar relationships were observed in an analysis using a cut point of 140 mg/dl to define diabetes (analysis not shown). We were also unable to distinguish between the effects of race and geography in this analysis because most African-American participants (90%) were drawn from a single study center in Jackson, Mississippi. However, as mentioned earlier, previous studies have also shown that African-American race is associated with glycemic control (21–23). Furthermore, we did not have information on fasting insulin at ARIC visit 2, and we cannot eliminate the possibility of a direct effect of hyperinsulinemia. Nonetheless, when measurements of fasting insulin from ARIC visit 1 were included in our analyses (data not shown), no changes were observed.

We know from the results of clinical trials that interventions that decrease A1C levels, even by 1% (e.g., from 8 to 7%) can make an enormous difference in the health and lives of people with diabetes. This study suggests that A1C levels are also related to carotid IMT, race/ethnicity, adiposity, lipid levels, hormone therapy, and diabetes treatment. These results provide useful information for future observational and clinical studies. Understanding which factors are related to A1C is important for the development of appropriate models of glycemic control and risk of clinical events in epidemiologic studies. The results presented here provide evidence that A1C is cross-sectionally related to carotid IMT independent of other risk factors. Chronically elevated glucose levels may contribute to the development of atherosclerosis in individuals with diabetes.