Differences in A1C by Race and Ethnicity Among Patients With Impaired Glucose Tolerance in the Diabetes Prevention Program

RESEARCH DESIGN AND METHODS—

The DPP was a 27-center randomized controlled clinical trial designed to evaluate the safety and efficacy of interventions to delay or prevent the development of diabetes in people at increased risk for type 2 diabetes. The baseline characteristics of the cohort have been described elsewhere (11). In brief, eligibility required age ≥25 years, BMI ≥24 kg/m² (≥22 kg/m² for Asians), and plasma glucose 2 h after a 75-g oral glucose load of 140–199 mg/dl (7.8–11.1 mmol/l) plus a fasting plasma glucose (FPG) 95–125 mg/dl (5.3–6.9 mmol/l) (or any fasting glucose ≤125 mg/dl [6.9 mmol/l] for American Indians).

The data reported here were obtained before randomization and are based on the 3,819 participants screened and found to be eligible to participate in the DPP. Standardized interviewer-administered questionnaires were used to obtain data on race/ethnicity, education, and marital status. Seated blood pressures were measured twice with a mercury sphygmomanometer. Standing height and weight were determined in duplicate with stadiometers and calibrated balance beam scales by certified clinic staff. Waist circumference was measured at the midpoint between the iliac crest and the costal margin in the midaxillary line. The oral glucose tolerance test (OGTT) was preceded by instructions to consume a usual diet with adequate carbohydrates and was initiated between 0700 and 1100 h after an overnight fast. Blood was sampled from a vein before (0 min) and after 75 g oral glucose (Trutol 75; Custom Laboratories, Baltimore, MD). Blood was drawn during the fasting state for hematocrit, plasma glucose, and insulin; at 30 min for plasma glucose and insulin; and at 120 min for plasma glucose. The 120-min glucose AUC (in millimoles per liter per 120 min) was computed using the trapezoidal rule. Dividing the AUC by 120 min yields the corresponding AUC mean glucose in millimoles per liter. β-Cell function was measured as corrected insulin response (CIR), where CIR = [100 × 30-min insulin (microunits per milliliter)]/[30-min glucose (milligrams per deciliter)] × [30-min glucose (milligrams per deciliter) − 70 mg/dl)]. Insulin resistance was measured as homeostasis model assessment of insulin resistance (HOMA-IR), where HOMA-IR = fasting insulin [microunits per milliliter] × [fasting glucose (milligrams per deciliter)/18.01]/22.5.

Blood samples were collected and processed following the DPP Standardized Manual of Operations. Serum and plasma samples were stored at −20°C for a few days and shipped on dry ice in batches. Whole-blood samples for A1C analysis were shipped by overnight express within 24 h of sample collection. Fasting specimens for A1C were obtained in eligible participants immediately before randomization. The average time interval between OGTT specimens and A1C was 60 ± 20 days. The time interval ranged from 52 days for Hispanics to 60 days for Asians and blacks to 62 days for whites and 69 days for Native Americans. Hematocrit was performed locally. All other analytical measurements were performed at the Central Biochemistry Laboratory at the University of Washington, Seattle, Washington. Plasma glucose was measured on a chemistry autoanalyzer by the glucokinase method. Insulin measurements were performed by a radioimmunoassay method using an anti–guinea pig antibody that measures total immunoreactive insulin. A1C was measured by a dedicated ion-exchange high-performance liquid chromatography instrument (Variant; BioRad, Hercules, CA). The intra-assay coefficient of variation (CV) was 1.36%, and the interassay CV was 1.70%.

A normal-errors multiple linear regression model was used to assess differences between groups and the effects of covariates on levels of A1C (12). Collinearity diagnostics indicated that no statistical problems existed with the use of all the covariates simultaneously in a single multiple-regression model. The normal-errors assumption was verified using the Shapiro-Wilks test of the residuals (13). However, White's test of homoscedasticity was significant (14). Thus, models were refit using White's asymptotic (consistent) robust information sandwich estimate of the covariance matrix of the estimates and these robust estimates used to construct a robust large sample test of the significance of each effect in the models. The results were unchanged; thus, the model-based tests of significance are presented. The strength of the effect of a covariate is expressed using the semipartial R², which is the proportion decrease in the total sum of squares [(N-1) × variance] when that covariate is removed from the regression model containing all other covariates. For multiple comparisons among ethnic groups versus whites, the Holm step-down Bonferroni method was used to adjust P values for multiple tests (15). All analyses were performed using SAS (SAS Institute, Cary, NC).

RESULTS—

Table 1 presents the characteristics of DPP participants at randomization by self-reported race and ethnic group. Of the 3,819 participants, 55% were white, 20% were black, 16% were Hispanic, 5% were American Indian, and 4% were Asian. Mean age was 51 years. Two-thirds were women. Approximately two-thirds were married, and one-quarter had college or higher educations. Mean blood pressure was 124/78 mmHg. Among women, mean BMI was 35 kg/m² and mean waist circumference 104 cm. Among men, mean BMI was 32 kg/m² and mean waist circumference 108 cm. Mean hematocrit was 41.1%. Mean fasting glucose was 106 mg/dl (5.9 mmol/l) and fasting insulin 184.5 pmol/l. Mean CIR was 0.6 and mean HOMA-IR 7. Mean A1C was 5.91%.

Compared with whites, blacks, Hispanics, American Indians, and Asians tended to be younger. Compared with whites, blacks, Hispanics, and American Indians were more likely to be women and Asians more likely to be men. Compared with whites, Asians and Hispanics were more likely to be married and Asians more likely to be college graduates. Compared with whites, blacks and Asians had higher and American Indians lower blood pressure levels. Compared with whites, blacks had slightly higher BMIs and Hispanics and American Indians slightly lower BMIs. Asians had substantially lower BMIs, reflecting that Asians with IGT and BMIs ≥22 kg/m² were eligible to participate. A similar pattern was observed with respect to waist circumference, with the exception that black and American Indian women had the highest waist circumferences. Compared with whites, blacks had lower and Asians higher hematocrits. FPG levels were quite similar across groups except for American Indians, who had lower fasting glucose levels, reflecting that American Indians with IGT had no lower eligibility limit for FPG to participate in the DPP, a protocol variation based on their known high rate of conversion to diabetes regardless of FPG level. Compared with whites, Hispanics, Asians, blacks, and American Indians tended to have lower 30-min post–glucose load glucose values. Because all DPP participants were required to have IGT, 2-h plasma glucose levels did not differ among groups. Whites and Hispanics had similar glucose AUCs, blacks and American Indians lower glucose AUCs, and Asians higher glucose AUC means. Whites and Asians had lower CIRs and were less insulin resistant. Compared with whites, all racial and ethnic groups had significantly higher A1C levels.

Table 2 presents a multiple regression model of the joint association of all covariates in Table 1 with A1C. The racial and ethnic differences in A1C persisted after adjusting for variables that might be related to the differences in A1C. The covariate-adjusted mean A1C levels estimated from the model were 5.78% for whites, 6.18% for blacks, 5.93% for Hispanics, 6.12% for American Indians, and 6.00% for Asians. The values for all racial and ethnic groups were significantly higher than for whites (each P < 0.0001 adjusted for multiple comparisons). Nonwhite race, older age, female sex, lower systolic blood pressure, higher diastolic blood pressure, greater BMI, higher FPG, greater glucose AUC, lower CIR, and higher HOMA-IR were all independently associated with higher A1C and, together, explained 22% of the variance.

Because the DPP used race- and ethnic group–specific cut points for BMI (BMI ≥24 kg/m² except ≥22 kg/m² for Asians) and FPG (95–125 mg/dl except ≤125 mg/dl for American Indians), we reran the analyses using common criteria for all groups (i.e., BMI ≥25 kg/m² and FPG 95–125 mg/dl [5.3–6.9 mmol/l]). The results were not substantially changed, and all differences between racial and ethnic groups remained statistically significant (P < 0.0001 adjusted for multiple comparisons).

Because there was a 60-day time interval between the OGTT and A1C, we reran the full model for the 2,022 participants who had not developed diabetes at 1 year of follow-up and had OGTTs and A1Cs performed on the same day. Compared with whites, the covariate-adjusted A1C values for all other racial and ethnic groups remained significantly higher (P < 0.0001 adjusted for multiple comparisons).

CONCLUSIONS—

In this cohort of adults with IGT enrolled in the DPP, blacks, Hispanics, American Indians, and Asians had higher A1C levels than whites. This effect persisted after adjusting for differences among groups in age, sex, education, marital status, blood pressure, BMI, hematocrit, fasting and post–glucose load glucose levels, glucose AUC, β-cell function, and insulin resistance. Taken together, these factors explained 22% of the variance in A1C.

Previous studies have demonstrated higher A1C levels among blacks and Hispanics, but these results have been attributed to poorer glycemic control among racial and ethnic minority groups (3). However, studies of nondiabetic populations and studies that have compared A1C levels among racial and ethnic groups within organized systems of health care and have carefully adjusted for processes of care have still demonstrated persistent differences in mean A1C (4–8). Our findings that factors that differed among racial and ethnic groups or were likely to affect glycemia did not explain differences in A1C suggest that hemoglobin glycation or red cell survival may differ among racial and ethnic groups.

Previous studies in nondiabetic individuals have shown that A1C levels in the same individual change little over time but that levels vary markedly between individuals (16–18). Additional variation in A1C levels between individuals has been shown to be related to factors independent of glycemia such as female sex (19), sex hormones (20), differences in visceral fat (21), and biologic variation in hemoglobin glycation or red cell survival. Recent studies have suggested that interindividual differences in intra-erythrocyte 2,3-diphosphoglycerate, which catalyzes the production of A1C, may in part account for the variability of A1C observed in nondiabetic subjects (22). Similarly, interindividual variation in intra-erythrocyte fructosamine 3-kinase, which deglycates intracellular fructosamines, might partially explain nonglucose-mediated interindividual variation in A1C (23). Evidence from diabetic twin studies have suggested that A1C levels are genetically determined (24,25). Interindividual variation in A1C may also be explained by differences in erythrocyte survival. Studies in both type 1 (26) and type 2 (27) diabetes have, for example, demonstrated that hyperglycemia is associated with decreased erythrocyte survival. In the DPP population, there were significant differences in hematocrit among racial and ethnic groups. Hemoglobinopathies were not systematically assessed but are generally more common in nonwhites and are associated with decreased erythrocyte survival and decreased glycohemoglobin percentages.

In conclusion, our analyses demonstrate that A1C levels are higher among U.S. racial and ethnic minority groups with IGT after adjustment for differences among groups in age, sex, education, marital status, blood pressure, adiposity, hematocrit, fasting and post–glucose load glucose levels, β-cell function, and insulin resistance. Thus, the racial and ethnic differences in A1C are not explained by differences in these factors. We appreciate that these glucose levels may not be a robust reflection of the 24-h glucose profile and that other unmeasured or suboptimally measured risk factors may explain some of these racial and ethnic differences in A1C. Clearly, further studies are needed to confirm our observation. It is not known whether these racial and ethnic differences in A1C lead to differences in the risk of microvascular, neurologic, or macrovascular complications. Our results raise the possibility that A1C may not be valid for assessing and comparing glycemic control across racial and ethnic groups or as an indicator of health care disparities. They also raise the important question of whether A1C can be used as a diagnostic test for diabetes.