Disparities in HbA_1c Levels Between African-American and Non-Hispanic White Adults With Diabetes: A meta-analysis

We conducted a Medline search in PubMed, the Cumulative Index to Nursing and Allied Health, the Combined Health Information Database, the Cochrane Library, and the Web of Science using Medical Subject Heading (MeSH) and free text forms for the period 1993 through 2005. We used the search terms, “Diabetes Mellitus”[MeSH] AND “U.S.”[MeSH] AND “Hemoglobin A, Glycosylated”[MeSH] OR glycemia OR glycemic control OR A1C with Limits: All Adult: 19+ years and English language. We initially retrieved 1,596 abstracts. The search was further limited to articles containing race or ethnicity (305 total articles). The literature accepted had to include patients with diabetes and contain comparative data for both African Americans and non-Hispanic whites. We rejected abstracts that included patients with gestational diabetes or pre-diabetes. We collected additional references from bibliographies of reviews, original research articles, and other articles of interest.

To conduct as broad an analysis as possible, we included any study design that was statistically valid. For retrospective chart review studies, we evaluated the most recent A1C data. We included observational data on A1C control. For four studies (20–23) in which sample size and A1C summary statistics were not provided in the original publication, we were able to retrieve this information from the authors. We only accepted author-reported data if we received written validation that the information was obtained from the original computerized dataset. If the SD of the A1C was not reported or otherwise obtained, we did not include the study in the meta-analysis. We did not exclude studies that failed to categorize the type of diabetes, as the diabetic population primarily consists of people with type 2 diabetes; this is especially true among African Americans.

Two investigators (J.K.K. and R.A.B.) independently reviewed each study for the following data: 1) sample size, 2) mean (±SD) participant age, 3) number of male and female subjects, 4) mean (±SD) A1C, 5) geographic location of the research, and 6) study design.

We initially retrieved 1,581 abstracts of English-language studies conducted in the U.S. from PubMed and found 15 additional abstracts in other data sources, resulting in a total of 1,596 abstracts. We rejected 1,151 abstracts because they did not meet the inclusion criteria. Of the 445 articles initially considered, review of the full citation resulted in 55 being excluded for not meeting the inclusion criteria. Of the 391 articles that were ultimately evaluated, 78 studies contained glycemia data for minority populations. Eighteen of the studies contained glycemia data for African Americans and non-Hispanic whites. If mean (±SD) glycemic information was not available, the authors were contacted to provide these data. Because of lab variability, we only analyzed A1C values, resulting in elimination of three studies that reported glycosylated hemoglobin (24–26). Four of the remaining studies reported comparative A1C data for African Americans but were not included for a variety of reasons: not providing SD for A1C (27), using pooled A1C data from multiple studies (28), and not defining the breakdown of an ethnic group (29); an additional intervention trial was excluded due to the possibility of patient selection bias in participant recruitment (30). Eleven studies that reported A1C were represented in our final analysis, including 3 prospective cohort studies (9,20,31) and 8 cross-sectional studies (12,21–23,32–35). Only a minor portion of patients in the studies reviewed had type 1 diabetes, with 15 reported in one study (9), and another study did not discuss any breakdown of patients by diabetes type (21).

A primary meta-analysis was conducted on the 11 studies (9,12,20–23,31–35) that met the inclusion criteria (Table 1). To judge the sensitivity of the results and justify the conclusions of the primary analysis, individual meta-analyses were conducted on subsets including study type (cohort and cross-sectional studies), managed care or nonmanaged care, and method of data collection for A1C (chart review or blood draw). Characteristics of the 11 studies are summarized in the table. An effect size (mean difference in A1C divided by the pooled SD) was calculated for the difference in A1C measurements between African Americans and non-Hispanic whites. For each study, a 95% CI for the effect size was also calculated.

Homogeneity of the effect sizes across studies was first assessed using a χ² test to determine whether a fixed- or random-effects approach should be implemented. A fixed-effects approach considers the set of studies as homogenous and representing all potential studies of interest, whereas the random-effects approach treats the studies as heterogeneous and considers them to be a sample from a population of comparable studies. With the exception of the managed care subset, the homogeneity test results led to the use of a random-effects model to pool effect-size estimates and compute a treatment effect. All tests of effect were two sided, and P < 0.05 was considered statistically significant. Results are reported for the entire dataset and then stratified by sex, study design, and data collection type for A1C.

Variability existed in the age of participants across studies, but most studies included patients aged >50 years. The sample sizes also varied widely across studies; that variability, however, was accommodated through the analysis for heterogeneity/homogeneity. Of the 11 studies in our meta-analysis, significant differences between African Americans and non-Hispanic whites were originally reported in 5 studies (12,31–33,35), no significant differences were reported in 1 study (34), and 1 study did not test for significant differences between ethnic groups (9). For four of the studies included in this meta-analysis, we contacted the authors for A1C data and tests for differences between African Americans and non-Hispanic whites that were not provided (20–23).

Through the meta-analysis, 10 of 11 studies indicated significantly higher A1C levels in African Americans than in non-Hispanic whites (Fig. 1). The summary effect size was −0.32 (P < 0.0001), which indicated that African Americans had A1C values that were, on average, 0.32 SD above those of non-Hispanic whites. This corresponds to an estimated 0.65% A1C difference. To evaluate the potential bias in the results due to sex, a meta-analysis conducted to compare the A1C levels between men and women, independent of race, had an estimate effect of 0.0024 (P = 0.9882).

The effects were similar regardless of study design. The cross-sectional studies had an estimated effect of 0.30 (P < 0.0001), and the prospective cohort studies had an estimated effect of −0.42 (P < 0.0010). Similarly, when studies were divided into two groups according to data collection type, the effects were consistent with the results from the summary analysis. Studies in which the A1C values were collected from chart reviews had an estimated effect of −0.31 (P < 0.0001), and studies in which the values were obtained from baseline blood draws had an estimated effect of −0.34 (P = 0.0041). When studies were divided into managed care or nonmanaged care, the effects were again consistent with the primary results. Studies in which the patients had managed care insurance had an estimated effect of −0.27 (P < 0.0001), and studies in which the patients did not have managed care insurance had an estimated effect of −0.38 (P < 0.0001).

A meta-analysis conducted for the comparison between the two ethnic groups among men had an estimated effect of −0.29 (P = 0.0026). Individually, four of the six studies providing sex-specific information would not have found a strong significant difference in men. However, the combination of the six studies through meta-analytic techniques demonstrates the significant difference between African Americans and non-Hispanic whites (Fig. 2). Likewise, the meta-analysis conducted for women showed a significant difference (P < 0.0001) between the ethnic groups with an estimated effect of −0.36 (Fig. 3).

This meta-analysis shows that African Americans have elevated A1C compared with non-Hispanic whites. The effect size estimate for this difference translates into an ∼0.65% difference (pooled SD for A1C across studies of 2.1) in A1C levels between African Americans and non-Hispanic whites. Our findings confirm those of Trivedi et al. (36), who, in a recent analysis of 1.8 million individual-level observations from 183 health plans, reported disparities between African Americans and non-Hispanic whites in glucose control.

Poor glycemic control is an important risk factor for diabetes complications, particularly microvascular complications (10). The difference between racial groups found in this meta-analysis represents a potentially significant increase in the micro- and macrovascular complications associated with diabetes. The U.K. Prospective Diabetes Study found a 21% reduction in any outcome for every 1% reduction in A1C (37). Thus, the difference found in this meta-analysis represents an ∼15% reduction in risk for vascular complications among non-Hispanic whites. However, this does not explain the magnitude of difference in diabetes complications between African Americans and non-Hispanic whites. Although A1C control among African Americans likely contributes to their elevated risk of complications, it accounts for only a portion of their excess risk. The need to control A1C, blood pressure, and cholesterol risk factors among patients with diabetes has been previously reviewed (19).

The consistency of the findings is notable: 10 of 11 studies showed significance such that the range of effect sizes did not include zero. Furthermore, all analyses provided the same general results and therefore strongly support the conclusion that there are significant differences in A1C between African Americans and non-Hispanic whites. In addition, two of the largest studies indicated similar point estimates for the standard effect size (Fig. 1) (20,22).

The strengths of this analysis are its inclusion of a variety of study designs, the ability to examine A1C differences by sex and study type, and the use of previously unpublished data (20–23). The results of this meta-analysis, however, depend in part on the accuracy and reliability of the A1C measurement across studies. Variability in tests of glycemia has been an issue in the field, and standardization of A1C was not widespread until the last decade (38). In more recent years, the International Federation of Clinical Chemistry and the National Glycohemoglobin Standardization Program have been working toward a certification of diagnostic equipment as being traceable to the Diabetes Control and Complications Trial reference method (38,39). In addition, persistence of hereditary fetal hemoglobin may cause a spurious elevation of A1C (40). However, type 1 diabetes and insulin treatment seem to be more associated with fetal hemoglobin production (41).

Another limitation of the analysis is publication bias; however, we performed numerous searches on this topic and contacted multiple investigators to retrieve the data for A1C means and SDs. The heterogeneity of the studies adds to the limitations of the analysis. Nevertheless, results are likely generalizable to African-American and non-Hispanic white adult patients with type 2 diabetes because the data included a broad range of patient ages, geographic settings, and study types. An additional limitation of this meta-analysis is that despite the comprehensive review of abstracts, the potential for omission exists if an abstract initially reviewed through our search process did not specifically address racial disparities.

The cause of the disparity in glucose control is multifactorial. Previous studies that have examined differences by ethnicity have found significant variation in rates of medical insurance coverage (42). Other studies of patients with diabetes have found that most have some form of medical insurance and that African Americans and non-Hispanic whites have similar rates of coverage (43). We were unable to obtain data on insurance status for eight studies included in the meta-analysis; however, two of the largest studies were conducted in managed care settings where all subjects were insured (9,23). In these studies, all of the patients had access to health care, but, even then, disparity in A1C levels was seen for minority groups.

Differences in the intensity of treatment may also account for a portion of the variation in A1C control. Determining bias is difficult, and we were unable to examine it in this study. Patient preferences for a type of treatment or comprehension of the disease and its treatment may account for some of the differences. Studies examining acceptance of cardiac procedures and renal transplants have found variation by ethnicity and race; however, differences in preference probably contribute only to a small portion of the variation in care (44,45).

This is the first meta-analysis of racial and ethnic differences in glycemic control among patients with diabetes. Although the studies included used a variety of designs, a consistency in the degree of disparity of glycemic control was found regardless of study type. Multiple separate meta-analyses were conducted across study types and sex with the same outcome. African Americans with diabetes have an ∼0.65% higher A1C than non-Hispanic whites; this difference may explain a portion of the excess microvascular complications in this population nationally. We need to understand more fully why this disparity in glycemic control exists and act to eliminate the factors that are modifiable (e.g., improved access to and delivery of care). Further research is needed to elucidate why African Americans with diabetes have poorer glycemic control than non-Hispanic whites and to identify interventions that might prevent or reduce the disparities.

The authors thank Carol Hildebrandt for her expertise in assembling the references and editing the manuscript.