Does Genetic Ancestry Explain Higher Values of Glycated Hemoglobin in African Americans?

We included 2,294 African American participants without known diabetes and with a complete set of covariates of interest presenting to visits 1 and 2 of the Atherosclerosis Risk in Communities (ARIC) Study, an ongoing prospective cohort study of adults from four U.S. communities who were 45–64 years of age at the baseline visit in 1987–1989 (13).

We included single nucleotide polymorphisms (SNPs) whose frequencies differ significantly between Caucasian and African ancestral populations, ancestry-informative markers (AIMs), to estimate the PEA among African Americans, an admixed population (14). The race-specific frequency of each SNP in the AIM panel was estimated using West African and European samples to provide a Bayesian prior for ancestral allele frequencies (14).

Genotyping methods for estimating genetic ancestry in African Americans in the ARIC Study have been described previously (14). In brief, genotyping was performed on stored DNA from visit 1 using the Illumina BeadLab platform (15) at the Center for Inherited Disease Research (Johns Hopkins University). Evaluation of 218,461 blind duplicate genotypes yielded a mismatch rate of 0.1% (14). We used standard filters for quality control of individual SNP genotyping. Samples were excluded for duplicity, low call rate, lack of sex concordance, or excess heterozygosity. We used ANCESTRYMAP software to estimate PEA for each participant (14). Additional detail regarding study population and genotyping is provided in the Supplementary Data.

HbA_1c was measured from frozen, whole-blood samples from visit 2 using ion-exchange, high-performance liquid chromatography (Tosoh A1c 2.2 Plus Glycohemoglobin Analyzer in 2003–2004 and Tosoh G7 Instruments in 2007–2008; Tosoh Corp., San Francisco, CA). These methods are aligned to the Diabetes Control and Complications Trial assay (11). We classified participants according to guidelines from the American Diabetes Association for HbA_1c: <5.7%, “lowest risk of diabetes (normoglycemia)”; 5.7–6.4%, “increased risk for diabetes (prediabetes)”; and ≥6.5%, “undiagnosed diabetes” (8).

Participants self-reported race, sex, education (<high school, high school, or >high school), combined family income (<$35,000, ≥$35,000, or unknown), employment (unemployed, employed, or retired), family history of diabetes, and physical activity (using the Baecke questionnaire [16]) at visit 1. Participants self-reported age, smoking status (current/former or never), and alcohol use (current or former/never) at visit 2.

Methods for measurement of glucose (11), lipids (17), BMI, waist-to-hip ratio (18), and blood pressure (19) at visit 2 were described previously. Hypertension was defined by current use of antihypertensive medication or mean of two blood pressure measurements ≥140/90 mmHg.

We evaluated for trends across quartiles of PEA using linear regression for means, Cuzick extension of the Wilcoxon rank-sum test for medians (20), and Goodman and Kruskal γ for proportions (21). We used Spearman rank correlation coefficients to evaluate unadjusted linear relationships and Kendall partial τ (ranked partial correlation coefficient) to estimate adjusted correlations (22). We estimated variance in HbA_1c, explained by each variable from the square of these correlation coefficients.

For the analyses that examined the impact of covariates, including PEA and fasting glucose, on reclassification of diabetes status, we obtained the adjusted values of HbA_1c from several regression models. In Model 1, HbA_1c was adjusted for fasting glucose, representing the variation in HbA_1c in the population that is independent of hyperglycemia. Model 2 further adjusted for PEA (Model 1 + PEA); Model 3 further adjusted for demographic and socioeconomic factors (Model 2 + age, sex, site, education, income, and employment); and Model 4 included other metabolic risk factors (Model 3 + BMI, waist-to-hip ratio, hypertension status, LDL and HDL cholesterol, triglycerides, family history of diabetes, alcohol use, physical activity, and smoking status). Using the predicted (adjusted) HbA_1c values from each model, we obtained the proportion of participants classified into the three HbA_1c categories. The base, unadjusted model indicates the classification of diabetes status using measured HbA_1c; Model 1 accounts for fasting glucose measured at the same visit; and Model 2 displays the effect of genetic ancestry (PEA) on diabetes status independent of fasting glucose. Recognizing that Model 1 is not directly relevant to clinical practice (as is often the case), as clinicians do not adjust for variables such as fasting glucose in the interpretation of other measures, we use Model 1 to estimate the proportion of variation in HbA_1c that is independent of fasting glucose levels and to illustrate whether genetic ancestry can still contribute to HbA_1c independent of fasting glucose levels.

We calculated the percentage of participants reclassified for each category as (N_i – N₀)/N₀, where N₀ is the number of participants in a given HbA_1c category for fasting glucose–adjusted HbA_1c (Model 1) and N_i is the number of participants in a given HbA_1c category for the i^th model (Models 2, 3, or 4). We then compared classification into and out of each of the three categories of diabetes status between the nested models, Models 2 (HbA_1c adjusted for fasting glucose and PEA) and 1 (HbA_1c adjusted for fasting glucose), with unadjusted (measured) HbA_1c as a reference (23): p_l = [n_i,k + n_{i,k + 1} – n_j,k – n_{j,k + 1}]/N_o (i = number shifted into a given category, j = number shifted out of the category, k = model, N_o = number in category from reference model, and l = HbA_1c category). We then calculated the overall net reclassification index: . HbA_1c values were log transformed.

We used Stata version 11.1 (StataCorp LP, College Station, TX) for statistical analyses. Institutional review boards approved the study protocol at each study site, and written informed consent was obtained from each participant.

PEA was right-skewed with a median (interquartile range [IQR]) of 14% (8–22) (Supplementary Fig. 1). Participants with lower PEA were more likely to have less than a high school education and were more likely to have an annual combined family income <$35,000 (Table 1). Median HbA_1c and mean BMI were higher in lower quartiles of PEA, and the prevalence of hypertension was highest in the lowest quartile of PEA (Table 1). HbA_1c was significantly and positively associated with age, family history of diabetes, prevalence of hypertension, BMI, glucose, LDL cholesterol, and triglycerides and inversely associated with education, income, HDL cholesterol, and current alcohol use (results not shown).

PEA was significantly correlated with HbA_1c but explained only 0.5% of the variance in HbA_1c on adjusted analyses (Table 2). On the other hand, a single fasting glucose explained the largest fraction of variance in HbA_1c (10%). Fasting glucose and PEA were not significantly correlated (r = 0.01; P = 0.63).

Two hundred forty-four participants (11%) had diabetes defined by HbA_1c ≥6.5%. Accounting for fasting glucose (Model 1) shifted higher HbA_1c values downward and lower HbA_1c values upward. Consequently, diabetes prevalence decreased to 4.4%, and the prevalence of prediabetes and normal HbA_1c increased (Table 3 and Supplementary Fig. 2). Further adjustment for PEA (Model 2) did not reclassify diabetes status substantially (Table 3 and Fig. 1). Normal HbA_1c status and prediabetes status changed negligibly with net reclassification of −1% (P = 0.21) and 4% (P < 0.001), respectively, and there was no net reclassification of diabetes status. Taking these individual category-level changes into account, the overall classification of normal HbA_1c, prediabetes, and diabetes status did not change with adjustment of fasting glucose–adjusted HbA_1c for PEA (net reclassification index 0.034; P = 0.94). Additional adjustment for demographic and socioeconomic factors (Model 3) and traditional metabolic risk factors (Model 4) did not substantially affect reclassification (Table 3 and Fig. 1).

In our study, we show that there is not a substantial (<1%) contribution of genetic ancestry to HbA_1c among African Americans; thus, ancestral genetic differences are unlikely to significantly explain the observed black–white difference in HbA_1c. Furthermore, a recent admixture scan failed to identify an HbA_1c locus among African Americans (C.-Y.C., N.M.M., personal communication), suggesting the absence of a genetic variant that varies enough in frequency between whites and African Americans (i.e., a variant of African ancestral origin) to explain the observed racial disparity in HbA_1c values.

The hypothesis that HbA_1c values are higher among African Americans for reasons unrelated to hyperglycemia stems from evidence that HbA_1c values are higher in African Americans when adjusted for fasting glucose and/or postload glucose measurements (1–4). However, HbA_1c reflects average glucose over 2–3 months (24) and may reflect racial disparities in socioeconomic and biologic factors. In our study, we show that hyperglycemia, as indicated by a fasting glucose value, accounts for the largest proportion of HbA_1c, and that adjustment for fasting glucose results in substantial reclassification of diabetes status. Moreover, after accounting for fasting glucose, additional adjustment for PEA does not significantly reclassify diabetes status.

We used of a large number of AIMs (1,350) to estimate PEA in an admixed population well described by AIMs, thus enhancing our ability to accurately assign percentage of genetic ancestry. Nonetheless, in this African American population, PEA was skewed toward lower values; the contribution of PEA to HbA_1c might be further clarified in a population with a more uniform distribution of PEA.

To evaluate the effect of PEA on classification of diabetes status by HbA_1c independent of fasting glucose, we adjust HbA_1c for fasting glucose. While not relevant to clinical practice, our approach directly addresses questions surrounding the interpretation of higher HbA_1c values in African Americans as representing hyperglycemia because of observed racial differences in HbA_1c even after glucose adjustment (6,7).

By estimating the global percentage of European ancestry for each individual, we may overlook local ancestry that could explain the observed racial differences in HbA_1c. A recent admixture scan to evaluate for specific regions that associate with HbA_1c did not reveal a locus in African Americans (C.-Y.C., N.M.M., personal communication).

In this large community-based population of African Americans, PEA was associated with, but explained little of, the variance in HbA_1c. Importantly, after accounting for a single fasting glucose measurement, PEA did not significantly affect classification of diabetes status using HbA_1c.

Our results suggest that elevated HbA_1c values in African Americans are likely not determined by ancestral genetic differences in the biology of HbA_1c but instead may reflect a true disparity in glycemia likely mediated by socioeconomic and other downstream factors. These findings support the use of the currently recommended HbA_1c categories for diagnosis of diabetes and classification of future diabetes risk (8) in African Americans. Consistent with this, previous studies have demonstrated that HbA_1c is equally predictive of cardiovascular disease and all-cause mortality in African Americans and whites (11). Future research should address the environmental reasons for disparities by PEA among African Americans and focus on methods to eliminate these disparities.

This research was supported by National Institutes of Health (NIH)/National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) Grant R21-DK-080294. N.M.M. was supported by NIH/National Center for Research Resources Grant 1KL2-RR-025006-01. E.S. was also supported by NIH/NIDDK Grant K01-DK-076595. The ARIC Study is carried out as a collaborative study supported by National Heart, Lung, and Blood Institute (NHLBI) contracts HHSN268201100005C, HHSN268201100006C, HHSN268201100007C, HHSN268201100008C, HHSN268201100009C, HHSN268201100010C, HHSN268201100011C, and HHSN268201100012C with support for this analysis by NIDDK grants R21-DK-073482 and K01-DK-067207 (W.H.L.K.). Genotyping services for this analysis in the ARIC Study were provided by the Johns Hopkins University under federal contract number N01-HV-48195 from the NHLBI and by the Center for Inherited Disease Research (CIDR). CIDR is fully funded through a federal contract from the NIH to the Johns Hopkins University, Contract N01-HG-65403.

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

N.M.M. refined the study design, analyzed data, and wrote the manuscript. W.H.L.K. refined the study design and reviewed and edited the manuscript. J.M.C. and F.L.B. reviewed and edited the manuscript. C.-Y.C. analyzed data and reviewed the manuscript. J.S.P. reviewed and edited the manuscript. E.S. refined the study design and reviewed and edited the manuscript.

Parts of this study were presented in an oral form at the 70th Scientific Sessions of the American Diabetes Association, Orlando, Florida, 25–29 June 2010, and in the Kelly West Award Lecture at the 71st Scientific Sessions of the American Diabetes Association, San Diego, California, 24–28 June 2011.

The authors thank the ARIC staff and participants for their important contributions.