Hemoglobin A_1c Genetics and Disparities in Risk of Diabetic Retinopathy in Individuals of Genetically Inferred African American/African British and European Ancestries

The Precision Medicine Initiative aims to enable a new era of medicine through research, technology, and policies to develop individualized care. Yet, it remains unclear how genetic information can be used in routine diabetes care (1,2). One application of precision medicine is to account for genetic variation that influences the performance of hemoglobin A_1c (HbA_1c), a widely used biomarker that measures the proportion of glycated hemoglobin to estimate ambient glycemia over the preceding 2–3 months. HbA_1c is a key modifiable risk factor for both macro- and microvascular complications (3–6) and has been accepted as the preferred diagnostic test for diabetes and measure of glycemic control, and as a clinical tool for managing complication risks.

Genetic variation that influences HbA_1c through nonglycemic mechanisms (e.g., differences in erythrocyte life span) can affect how accurately HbA_1c reflects underlying glycemia (7,8). In a large-scale, multiancestry genome-wide association study (GWAS) meta-analysis of HbA_1c, more than 200 genetic variants were reported to be associated with HbA_1c (9–15). The effect of being in the top 5%, relative to the bottom 5%, of a polygenic score comprising 60 genetic variants that influence HbA_1c independently of glycemia was 0.25% in individuals of European ancestry, but 0.8% in individuals of African ancestry (13,16). This large ancestral difference was due to a single African-specific missense variant in G6PD, rs1050828 (risk allele, T), which has a minor allele frequency of ∼12% in African Americans, lowers HbA_1c independently of glycemia, and causes glucose-6-phosphatase dehydrogenase (G6PD) deficiency, an X-linked disease (17,18).

Clinical practice guidelines have chosen the HbA_1c diagnostic threshold of 6.5% and the HbA_1c target of 7% for most adults, due to their associations with the risk of developing diabetes-related complications (19,20). However, people who carry genetic variants that lower HbA_1c independently of glycemia may be delayed in their diabetes diagnosis and undertreated for hyperglycemia, creating disparities in outcomes, especially among certain minority populations with a higher prevalence of large-effect variants, like the African-specific G6PD variant.

The objective of this study was to evaluate the nature and extent of such disparities by evaluating the effects of genetically driven nonglycemic variation in HbA_1c on the risk of retinopathy in people with type 2 diabetes. We hypothesized that individuals who carry genetic variants that lower HbA_1c independently of glycemia have higher real, but undetected, hyperglycemia compared with others who do not and, consequently, have a greater risk of developing diabetes-related complications, despite achieving similar HbA_1c targets. We chose retinopathy as the outcome because it is one of the earliest complications of diabetes and can develop even before a diabetes diagnosis if chronic hyperglycemia was undetected (6,21,22). We tested this hypothesis using data from 29,828 individuals with type 2 diabetes in the UK Biobank (UKBB) and All of Us (AoU) databases with genetically inferred African African/African British or European ancestry. We calculated ancestry-specific nonglycemic HbA_1c polygenic scores (ngA1cPS) using a weighted sum of genetic variants previously associated with HbA_1c in GWAS, but not with glucose. We evaluated whether individuals with a low ngA1cPS, indicating a genetic predisposition to lower HbA_1c independently of glycemia, had a greater risk of retinopathy, adjusting for measured HbA_1c and established retinopathy risk factors. In doing so, we evaluated disparities in retinopathy prevalence across the ngA1cPS and by ancestry to determine whether accounting for genetic effects could improve the clinical utility of HbA_1c in diverse populations or explain interindividual variation in outcomes despite seemingly achieving HbA_1c targets.

The UKBB is a prospective cohort study with genetic and phenotypic data collected on approximately 500,000 individuals from the U.K. who were between 40 and 69 years of age at recruitment (23). Using the UKBB array data, we applied preimputation quality control, performed phasing with SHAPEIT4 (https://odelaneau.github.io/shapeit4/), and imputed the phased haplotypes using the TOPMed reference panel freeze 8 (24).

The AoU Research Program is a U.S. biobank developed to leverage the diversity of the United States for facilitating and improving high powered genetic and epidemiological studies. Details of the recruitment methods, clinical sites, and data availability are described elsewhere (25). On 22 June 2022, the AoU Research Program released whole-genome sequencing data for 98,590 participants. Electronic health records and genetic information were extracted from the AoU, version 6, controlled tier data set using the AoU Researcher Workbench.

Type 2 diabetes, retinopathy, coronary artery disease, chronic kidney disease, and hypertension were defined using the ICD-9 and -10 codes, shown in Supplementary Table 1. Biological sex was inferred from genetic data. Diabetes duration was calculated as the number of years between the patient’s self-reported age of diagnosis and age at the time of enrollment. We used the mean value calculated across available measurements for diastolic and systolic blood pressure, HbA_1c, random glucose, triglycerides, LDL, and creatinine. Individuals with missing HbA_1c data were excluded from the analyses. For the remaining laboratory values, the mean missing rate was 7% (median, 8%), and missing values were imputed with the median of the nonmissing values (Supplementary Table 2).

For individuals of European ancestry, we calculated an ngA1cPS composed of 122 variants reported to be associated with HbA_1c at genome-wide significance in a multiancestry meta-analysis GWAS in people without diabetes by the Meta-Analysis of Glucose and Insulin-related traits Consortium (MAGIC) that had less than 25% probability of being assigned “glycemic” based on their association with glycemic traits, red blood cell traits, and iron metabolism in a soft clustering analysis (15). The threshold of 25% was selected to achieve a balance between attaining enough variants to form a polygenic score that explains genetically driven HbA_1c variation while excluding variants clearly associated with glycemic traits. In European ancestry, of the 185 HbA_1c variants, 63 were above this threshold due to their association with glycemia and were excluded, leaving 122 variants to construct the ngA1cPS (Supplementary Table 3). The ngA1cPS was applied with the score function in PLINK (https://www.cog-genomics.org/plink/1.9/), using the β values from the original GWAS as the weights with the effect allele defined as the HbA_1c-raising allele (i.e., all weights used in the ngA1cPS are positive; Supplementary Table 4). Because only 14 of the 122 variants discovered in the multiancestry GWAS meta-analysis were associated with HbA_1c at nominal significance in the African-ancestry GWAS (15), we tested the association of an ngA1cPS composed of only the 14 variants with HbA_1c in our study sample and found the association to be stronger than the 122-variant ngA1cPS in African ancestry (Supplementary Table 5). Thus, we opted to use the 14-variant ngA1cPS in the analysis of African ancestry. Because the effect of the G6PD variant on HbA_1c had a reported effect size that was many times larger than all the other variants combined (13), we excluded it from the ngA1cPS and examined the effect of the G6PD variant separately.

We plotted the proportion of retinopathy by ngA1cPS quintiles and by ancestry to visually inspect for differences. We fitted a logistic regression model to assess the effect being in the bottom quintile (ngA1cPS percentile <20), compared with the middle 60%, on retinopathy, and adjusted for age, genetic sex, and 10 genetic ancestry principal components, followed by measured HbA_1c and other retinopathy risk factors, including diabetes duration, chronic kidney disease, hypertension; random glucose, triglyceride, LDL, and creatinine levels; and current smoking. We calculated the effects in UKBB and AoU separately and then performed a fixed-effects meta-analysis within each ancestry. To determine whether the study had sufficient statistical power to detect meaningful effects, we performed a power calculation and showed that, in a case-control study with 498 cases in African American/African British ancestry and 1,862 cases in European ancestry and an equal number of controls, using α = 0.025, we had at least 80% power to detect an odds ratio (OR) as small as 1.20 in African American/African British ancestry and 1.10 in European ancestry per SD of the ngA1cPS. Due to the smaller sample of African American/African British ancestry, we acknowledge that subtle associations could be missed.

In European ancestry, we performed stratified analyses by measured HbA_1c categories of 0.5%. In African ancestry, due to the small number of G6PD carriers with retinopathy, particularly in the UKBB database, we restricted the stratified analysis by G6PD carrier status to AoU and compared individuals below the 50th percentile with those above the 50th percentile of the ngA1cPS instead of quintiles. Because the G6PD variant is located on the X chromosome, we reported results stratified by sex. The Mass General Brigham Institutional Review Board (study no. 2016P001018) approved this study. All participants in UKBB (National Research Ethics Committee reference no. 11/NW/0382) and AoU provided informed consent, and research was conducted according to the Declaration of Helsinki.

In the UKBB and AoU, we identified 25,895 and 11,165 individuals with type 2 diabetes, of whom 1,471 and 1,639, respectively, had retinopathy (Table 1). Of these 37,060 individuals, 32,350 were of European or African ancestry, and 29,828 had available HbA_1c data for downstream analysis. In both cohorts, individuals with retinopathy were slightly older (62 vs. 61 years, P = 3 × 10⁻¹⁹ in UKBB; 65 vs. 63 years old, P = 3 × 10⁻¹⁰ in AoU) and had longer diabetes duration (13.4 vs. 7.5 years, P = 5 × 10⁻¹⁴¹ in UKBB; 11.0 vs. 7.5 years, P = 9 × 10⁻¹⁴⁶ in AoU). Individuals with retinopathy had higher blood pressure, higher HbA_1c (7.7% vs. 7.0%, P = 1 × 10⁻⁹⁸ in UKBB; 8.0% vs. 7.0%, P = 4 × 10⁻⁸⁶ in AoU), higher random glucose levels, and higher proportions of chronic kidney disease, coronary artery disease, and hypertension. In both populations, we observed a smaller proportion of females with retinopathy compared with those without (32% vs. 38%, P = 1 × 10⁻⁵ in UKBB; 50.0% vs. 58.7%, P = 7 × 10⁻⁹ in AoU).

In both African American/African British and European ancestries, the bottom quintile of the ngA1cPS had the lowest retinopathy prevalence (Fig. 1A and B). The median measured HbA_1c in the bottom quintile, middle three quintiles, and top quintile of the ngA1cPS were 6.39% (interquartile range [IQR] = 5.79, 7.29), 6.47% (IQR = 5.9, 7.38), and 6.52% (IQR = 5.99, 7.30), and the corresponding median random glucose levels were 109.0 mg/dL (IQR = 91.4, 146.6), 107.9 mg/dL (IQR = 90.9, 143.1), 106.3 mg/dL (IQR = 90.6, 140.4), respectively. Creatinine, hemoglobin, reticulocyte percentage, and the proportion of individuals reported to be on metformin and insulin were similar across the ngA1cPS for both ancestries (Supplementary Table 6).

The distributions of the ngA1cPS by retinopathy status overlapped, indicating that the ngA1cPS cannot discriminate those with and without retinopathy, though the means of the ngA1cPS were lower in people with retinopathy compared with those without (Supplementary Fig. 1). Despite this lack of discrimination, the ngA1cPS was associated with retinopathy even after adjustment for measured HbA_1c and other retinopathy risk factors in the meta-analysis of UKBB and AoU. The meta-analytic OR for prevalent retinopathy in the bottom quintile of the ngA1cPS, adjusted for measured HbA_1c, was 1.45 (95% CI 1.13, 1.88; P = 3.5 × 10⁻³) in African American/African British ancestry (Fig. 1C) and 1.34 (95% CI 1.19, 1.51; P = 9.8 × 10⁻⁷) in European ancestry (Fig. 1D). The meta-analytic ORs after further adjusting for other retinopathy risk factors was 1.31 (95% CI 0.1.0, 1.73; P = 0.05) in African American/African British ancestry and 1.31 (95% CI 1.15, 1.50; P = 6.5 × 10⁻⁵) in European ancestry.

In the analysis stratified by measured HbA_1c categories in European ancestry, the meta-analytic ORs, adjusted for age, sex, and principal components and retinopathy risk factors, were largest in individuals with HbA_1c between 6% and 6.5% (OR 1.48; 95% CI 1.05, 2.09; P = 0.02; Fig. 2A and B), and HbA_1c between 7.5% and 8% (OR 1.59; 95% CI 1.07, 2.38; P = 0.02; Fig. 2A and 2B) The ngA1cPS was marginally associated with prevalent retinopathy in individuals with HbA_1c above 8% and below 6%.

There were no statistically significant differences in retinopathy prevalence by G6PD genotype and the median measured HbA_1c was similar across G6PD genotypes (6.8% in noncarriers; 6.9% in heterozygous females; 6.7% in affected males and homozygous female). However, among individuals with HbA_1c below 7%, retinopathy prevalence appeared to be higher in G6PD carriers compared with noncarriers (females: CT/TT 6.3% vs. CC: 6.5%; males: T: 13.6% vs. C: 8.8%; differences were not statistically significant; Supplementary Fig. 2). When stratified by ngA1cPS, retinopathy prevalence was higher for individuals below the 50th percentile compared with above the 50th percentile of the ngA1cPS, regardless of sex or G6PD genotype. Among individuals with HbA_1c greater than 7%, there were no clear differences in retinopathy prevalence across the ngA1cPS or G6PD genotype (Fig. 3).

This study provided a large, population-scale, multiancestry examination of polygenic nonglycemic HbA_1c effects on retinopathy risk among individuals with diabetes. We showed that individuals with diabetes who were genetically predisposed to having lower measured HbA_1c due to nonglycemic genetic effects had a higher prevalence of retinopathy. The higher prevalence of retinopathy in the bottom quintile of the ngA1cPS, compared with above the quintile, was observed despite similar measured HbA_1c across the ngA1cPS. The OR of the ngA1cPS bottom quintile did not attenuate when adjusting for known retinopathy risk factors, including measured HbA_1c, diabetes duration, hypertension, random glucose, triglycerides, LDL, creatinine, and smoking.

We found an equivalent increased risk of retinopathy among individuals in the bottom quintile of the ngA1cPS for both African American/African British and European ancestries. These findings were replicated in two populations that differed by continent, ancestral diversity, baseline comorbidities, and distribution of retinopathy risk factors. This suggests that precision diabetes care should be tailored based on an individual’s unique genetic profile and not only on discrete ancestry or geography. We emphasize that it would be inappropriate, and potentially harmful, to apply race-based or ancestry-based HbA_1c diagnostic thresholds or targets to an entire race or genetically inferred ancestral population without accounting for the full complement of genetic effects on nonglycemic variation of HbA_1c, as evidenced by the clinically meaningful differences in retinopathy prevalence across the ngA1cPS within ancestries.

Our results suggest that the clinical management of patients with diabetes could be improved by considering genetic effects on HbA_1c. These results differ from other disease prediction analyses involving genetic and clinical variables, whereby the marginal value of the genetic information declines as we consider relevant clinical measurements (26,27). We infer that the risk of other diabetes-related outcomes that depend on glycemic control could be better estimated if HbA_1c genetics are considered, and if ignored, could result in disparities in outcomes. Genetically determined nonglycemic variation in HbA_1c may partly explain why some individuals who achieve HbA_1c glycemic targets still develop complications. Although this study was focused on individuals with diagnosed type 2 diabetes, our findings have implications on the use of HbA_1c in the management of other types of diabetes, including type 1 diabetes. These results also raise the questions of whether the broad application of HbA_1c testing in diverse populations to diagnose diabetes without confirmation with glucose measurements is appropriate, and of the use of simple linear regression equations for translating HbA_1c measurements to average glucose (28,29) without considering genetic effects that may alter the HbA_1c-glycemia relationship.

Fructosamine and other glycated proteins (30,31) can be used in clinical practice in the presence of genetic or nongenetic factors that meaningfully interfere with HbA_1c measurement (e.g., hemoglobin variants and renal failure) or interpretation (e.g., pregnancy, anemia, recent blood loss). Yet, these glycated proteins have other limitations: their measurements are affected by albumin production, reflect glycemia over a short time, and have not been as extensively evaluated for their prediction of long-term complications of diabetes as has measured HbA_1c (32,33). Time in range from continuous glucose monitoring (CGM) is also highly correlated with HbA_1c and can be used as an outcome measure or predictor of diabetes-related complications (34–37). However, as the clinical use of CGM is often restricted to only patients requiring frequent glucose monitoring; CGM is unlikely to replace HbA_1c completely in diabetes screening or prevention of complications in general populations.

In the stratified analysis by measured HbA_1c categories in European ancestry, the excess risk conferred by the ngA1cPS was mostly driven by the individuals with HbA_1c between 6% and 6.5%, around the diagnostic threshold for diabetes, and between 7.5% and 8%, reflecting suboptimal glycemic control. HbA_1c genetics may have less of an impact on retinopathy risk in the nondiabetic range where its prevalence is low. In individuals with HbA_1c above 8%, lenient glycemic targets in the presence of comorbid conditions, psychosocial barriers to diabetes management, or use of home glucose monitoring may be more important determinants of complications risk than HbA_1c genetics.

Expectedly, the proportion of individuals on metformin or insulin use was three to four times higher among those with HbA_1c greater than 7% compared with those with HbA_1c less than 6%. Yet, the use of these medications was similar across the ngA1cPS regardless of HbA_1c category—a reflection of current clinical practice that does not consider genetic effects when using measured HbA_1c to guide treatment decisions. Individuals in the middle quintiles and top quintile of the ngA1cPS with HbA_1c between 6.5% and 7% had similar retinopathy prevalence as those in the bottom quintile with HbA_1c between 6% and 6.5%. If differences in retinopathy prevalence were due to undertreatment, individuals in the bottom quintile will need to have HbA_1c below 6.5% to have a similar risk for retinopathy as their counterparts above this quintile with HbA_1c between 6.5% and 7%.

Among individuals of African ancestry with HbA_1c below 7%, retinopathy prevalence was higher in G6PD carriers compared with noncarriers, though differences were not statistically significant. The large HbA_1c-lowering effect of the G6PD variant likely resulted in underdiagnosis of diabetes and its complications among carriers, reducing the power in our analysis. Indeed, the diabetes prevalence in G6PD carriers versus noncarriers was 8% versus 11% in males (P = 0.006) and 14% versus 16% in females (P = 0.02). Nevertheless, regardless of sex or G6PD carrier status, retinopathy prevalence was higher below the 50th percentile compared with those above the 50th percentile of the ngA1cPS. We concluded that the ngA1cPS, uniquely constructed for each ancestry, captured polygenic effects that represented a more comprehensive estimation of the genetic risk in people with diabetes. Still, we acknowledge that these ngA1cPS only included genetic variants reported in published GWAS and do not fully account for all genetic effects across the genome on nonglycemic variation in HbA_1c, such as rare or low-frequency variants.

Apart from undertreatment of hyperglycemia and delayed diabetes diagnosis, other factors could account for these differences in retinopathy prevalence across the ngA1cPS. The ngA1cPS could be associated with unmeasured variables or retinopathy risk factors that were unaccounted for in our analysis. Because genetic effects were modeled as polygenic scores, we were unable to distinguish among the various nonglycemic mechanisms, including a propensity for glycation, sometimes referred to the hemoglobin glycation index, which is associated with retinopathy and other outcome measures (38–40). Notably, a significant proportion of the contributing genetic variants affected erythrocytic or reticulocyte parameters (Supplementary Table 3), which suggests that the principal mechanism within ancestry was likely differences in erythrocytic turnover. By including genetic variants that had up to 25% probability of being assigned “glycemic” in the soft clustering analysis (15), the ngA1cPS may have included variants that were modestly associated with glycemic traits. Nonetheless, if the ngA1cPS included variants associated with glycemia, the observed association between lower ngA1cPS and higher risk of retinopathy would have been biased toward the null, because variants that raise HbA_1c through hyperglycemia are expected to increase, and not reduce, retinopathy risk. Future studies that involve multiple glucose measurements, such as continuous glucose monitoring, and treatment exposures over time could help clarify the mechanisms giving rise to these differences in retinopathy prevalence.

Similar to previous analyses of diabetic retinopathy, this study was limited by the precision with which diabetes and retinopathy could be defined by ICD coding. We recognize that the ngA1cPS was composed of variants reported to be associated with HbA_1c in a multiancestry GWAS meta-analysis of diabetes-free individuals, which could explain why genetic effects were attenuated in people with HbA_1c higher than 6.5%. Nevertheless, it is reasonable to assume that genetic effects on nonglycemic variation in HbA_1c do not vary by underlying glycemia or diabetes status. Although an analysis of incident retinopathy would be useful to evaluate the added value of ngA1cPS in the prediction of future retinopathy for people with newly diagnosed diabetes, we recognize the potential for detection bias in a time-to-event analysis. Compared with people with a lower ngA1cPS, those with a higher ngA1cPS are expected to have higher measured HbA_1c for the same average glucose and, therefore, are more likely to be diagnosed with diabetes, screened for diabetes-related complications, and be diagnosed with retinopathy, whereas people with a lower ngA1cPS are more likely to have undetected hyperglycemia and diabetes-related complications.

In a sensitivity analysis, we tested the association of a ngA1cPS composed of all 122 genetic variants with measured HbA_1c and retinopathy in African American ancestry, and effect estimates obtained were consistent with the ngA1cPS composed of only 14 genetic variants. Per SD of the ngA1cPS, the estimated odds of retinopathy were lower by 14% in African ancestry and by 9% in European ancestry (Supplementary Table 7). In people without the G6PD variant, the mean of the ngA1cPS composed of all 122 genetic variants was 0.26% higher in African ancestry compared with European ancestry, supporting the hypothesis that ancestral differences in mean HbA_1c may be explained, to some extent, by genetics (Supplementary Table 5). Given the poor transferability of the large number of genetic variants, and the different number of variants contributing to the ngA1cPS for each ancestry, we refrain from making any firm conclusions from ancestral differences. The lack of transferability could be explained by differences in linkage disequilibrium resulting in some of the causal variants tagged by lead variants in European ancestry but not tagged by the same variants in African ancestry. Furthermore, the discovery of nonglycemic HbA_1c variants in multiancestry GWAS meta-analysis is biased toward European ancestry because the contribution of African ancestry samples was much smaller. We also recognize that genetically inferring African American/African British ancestry does not capture the full genetic diversity of the Africa continent. We would expect that, as larger GWAS of HbA_1c are conducted in diverse populations, more HbA_1c variants will be discovered, enabling the construction of more comprehensive ngA1cPS in non-European ancestries and a fairer comparison of the impact of genetics on complication risks between ancestries.

In sum, our study showed that the aggregate effect of variants that lower HbA_1c independently of glycemia is associated with higher odds of retinopathy in individuals of African American/African British ancestry as well as European ancestry. Genetic effects need to be considered to define personalized HbA_1c targets, reduce disparities in diabetes-related outcomes, and promote equal care for people of all genetic backgrounds.

Funding. A.L. is supported by grant 2020096 from the Doris Duke Foundation and the American Diabetes Association grant 7-22-ICTSPM-23. J.M.M. is supported by American Diabetes Association Innovative and Clinical Translational Award 1-19-ICTS-068, American Diabetes Association grant 11-22-ICTSPM-16 and by National Human Genome Research Institute grant U01HG011723 and Medical University of Bialystok grant from the Ministry of Science and Higher Education (Poland).

The funders had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.

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

Author Contributions. R.M., P.H.S., and A.L. researched data, performed the analysis, and drafted the manuscript. J.M.M. contributed to the analysis and reviewed and edited the manuscript. J.C.F. reviewed and edited the manuscript. All authors approved the final version of the manuscript. A.L. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Part of this work was presented as an oral abstract at the American Diabetes Association Scientific Session, San Diego, CA, 24 June 2023.

Handling Editors. The journal editor responsible for overseeing the review of the manuscript was Stephen S. Rich.