Ethnicity Affects A1C Levels in Patients With Diagnosed Type 2 Diabetes in Southern Israel Free

This was a retrospective, cross-sectional, epidemiological study based on computerized medical records. The study period was 1 August 2015 to 30 July 2020. In Israel, a compulsory national health insurance law that passed in 1994 provides a diverse basket of health care services and drugs for all Israeli citizens. Health care services are provided through four major sick funds. The largest sick fund, which serves >52% of the citizens of Israel and >54% of the population in the southern district of Israel, is Clalit Health Services (CHS) (23). CHS has an advanced centralized electronic medical registration system that served as the database for this study. The system was first activated in 1998 but began to be used universally in CHS in 2003 (24). Thus, the study data could be considered equally representative of all ethnicities and of the overall Israeli population.

The study population comprised type 2 diabetes patients insured in the southern district of the CHS who were ≥40 years of age and who underwent fasting glucose and A1C tests during the study period. The threshold age of 40 years was chosen for two reasons: 1) type 2 diabetes prevalence is higher after this age and 2) the immigration of Ethiopian Jews to Israel began in 1984. Using an earlier age threshold could have led to difficulties in the identification of Ethiopian Jewish immigrants based on country of birth in the computerized medical records.

Type 2 diabetes patients undergo routine laboratory testing in the outpatient setting, including fasting glucose and A1C tests, once in a 3- to 12-month period, depending on the severity of their diabetes and the success of their type 2 diabetes control. Thus, most individuals had more than one laboratory value during the study period. To prevent clustering, we used the earliest set of simultaneous measurements taken for each patient.

We used an epidemiological approach, similar to that used in another study (16), which relied on one set of simultaneous glucose and A1C measurements in a very large number of individuals. Although each correlation may be not completely accurate, the inaccuracy is reduced to a minimum in this large sample of individuals. All laboratory tests for this study were performed at a single site, the Soroka University Medical Center. Since 2011, five centrifugation centers have been established, resulting in a shorter pre-analytic stage and better test accuracy (25).

The following data were collected from the medical records: sociodemographic information, including age, sex, ethnicity, country of birth, and socioeconomic status; laboratory data, including A1C, fasting glucose, hemoglobin, non-HDL cholesterol, and microalbumin-to-creatinine ratio; comorbid conditions, including stroke, ischemic heart disease, and thalassemia; purchase of antidiabetic medications; disease duration; and BMI.

The study was approved by the Helsinki Committee of the Soroka Medical Center (approval #0273-20), which exempted it from the requirement to obtain informed consent from the participants.

Diabetes diagnoses were recorded in medical records in accordance with International Classification of Diseases, 9th revision, codes 250.x0 and 250.x2. The three ethnicity categories (i.e., Arabs, Ethiopian Jews, and non-Ethiopian Jews) were defined based on ethnicity and country of birth. Socioeconomic status was classified at three levels based on the site of residence as it appeared in the database. Medication use was based on actual purchases from pharmacies of at least one of the following antidiabetic drug groups: biguanides, sulfonylureas, meglitinides, α-glucosidase inhibitors, thiazolidinediones, sodium–glucose cotransporter 2 inhibitors, glucagon-like peptide 1 receptor agonists, dipeptidyl peptidase 4 inhibitors, insulin, or combination medications 3 months before the defining blood test. Based on their purchase of drugs, participants were categorized as combination therapy users if they purchased more than one type of medication concomitantly or combination medications or as insulin users if they purchased insulin alone or in combination with other drugs. Disease duration was calculated from the earliest of the first date the diagnosis of type 2 diabetes appeared in the medical record, the first A1C test result ≥6.5%, or the first date the patient purchased an antidiabetic medication. The highest BMI measurement over the 5-year study period was extracted from the medical record.

Statistical analyses were conducted using R, v. 4.0.5, software. Categorical variables are presented as numbers and percentages, and continuous variables are described as mean ± SD or median (interquartile range [IQR]), as appropriate for the distribution. Statistical differences were tested by the χ² test for categorical variables and ANOVA/Kruskal-Wallis tests for continuous variables.

A normal-error multiple linear regression model was used to assess differences in associations between ethnicity and A1C, adjusted for the effects of other covariates on A1C. The non-Ethiopian Jewish group served as the reference ethnic group in the model. Colinearity diagnostics indicated that there was no statistical issue with the simultaneous use of all tested variables. To ensure a better model fit, influential observations (outliers and leverages) were extracted from the final model using Cook’s distance. The normal-errors assumption was verified using the diagnostic quantile-quantile, residuals-versus-fitted, and residuals-versus-leverage plots. The strength of the effect of the covariate is expressed using semipartial R², which is a proportion decrease in the total sum of squares when the covariate is removed from the regression model containing all other covariates ([N − 1] × variance). All statistical tests were two-sided, with P <0.05 set as the threshold for significance.

A total of 59,432 patients with type 2 diabetes of Ethiopian Jewish (n = 1,804), non-Ethiopian Jewish (n = 49,296), or Arab (n = 8,332) ethnicity underwent at least one blood test with simultaneous measurements of fasting glucose and A1C during the study period. Table 1 presents the characteristics of the study participants. There was a statistical difference in all socioeconomic, laboratory, and disease-specific variables. Females comprised ∼49% of the study population, and the mean age was 64.7 ± 12.1 years. Compared with non-Ethiopian Jews, participants in the other ethnic groups were younger (P <0.001). The percentage of females was significantly higher in the Arab group compared with the other groups. The majority of the population were in the middle socioeconomic class, but significantly more Arabs and Ethiopian Jews were in lower socioeconomic classes (P <0.001). The median (IQR) diabetes duration was 8.4 (3.4–13.4) years, which was shorter in the Arab group and shortest in the Ethiopian Jewish group (P <0.001). The median (IQR) fasting glucose was 133 (108–158) mg/dL and the median (IQR) A1C was 51 (41.7–60.3) mmol/mol (6.8 [6.0–9.7]%), which was lowest for non-Ethiopian Jews (P <0.001). The median (IQR) for BMI was 29.7 (26.2–33.3) kg/m². Arabs had higher BMI measurements, and Ethiopian Jews had lower measurements than non-Ethiopian Jews (P <0.001). Approximately 38% of the patients had non-HDL cholesterol levels >130 mg/dL, and these levels were higher among Arabs and lower among Ethiopian Jews compared with non-Ethiopian Jews (P <0.001). Additionally, statistically significant differences were observed for the prevalence of microalbuminuria, stroke, and ischemic heart disease (P <0.001 for all). About 15% were treated with insulin, and 28% took more than one medication (or combination therapy), all significantly higher among Arabs and lower among Ethiopian Jews compared with non-Ethiopian Jews (P <0.001).

In the univariate analysis, there were significant associations with A1C for all the variables presented in Table 1 (P <0.001), with the exception of stroke. Table 2 presents a multiple regression model of the combined association of all of the variables listed in Table 1 with A1C. Compared with non-Ethiopian Jewish ethnicity, Ethiopian Jewish ethnicity raised A1C by 1.24 mmol/mol (0.113%), and Arab ethnicity increased A1C by 2.46 mmol/mol (0.255%) (P <0.001) after adjusting for other variables that could influence A1C. The model is stable, significant (P <0.001), and explained 74.6% of the variance of the dependent variable, A1C.

Figure 1 shows the contribution of various predictors of the variance of A1C. Fasting glucose showed the strongest association with A1C, explaining 65% of its variance (P <0.001). Ethnicity was also a strong predictor, explaining 0.6% of the variance in A1C (P <0.001).

Figure 2 presents the results generated by this model, showing mean A1C values over a range of fasting plasma glucose levels after adjustment for all other significant predictor variables. An A1C of 53 mmol/mol (7%) correlated with fasting glucose levels of 141, 136, and 126 mg/dL in non-Ethiopian Jews, Ethiopian Jews, and Arabs, respectively. An A1C of 64 mmol/mol (8%) correlated with fasting glucose levels of 184, 180, and 169 mg/dL in non-Ethiopian Jews, Ethiopian Jews, and Arabs, respectively.

Our study showed that the relationship between fasting glucose and A1C significantly differs among patients with type 2 diabetes from different ethnic groups in Israel, with the mean A1C for the same fasting glucose level higher among minority Arab and Ethiopian Jewish ethnicities than for the non-Ethiopian Jewish population. Ethnicity was among the five most significant factors that explained the variance in A1C among study participants. In fact, it contributed to A1C variance even more (0.6%) than microalbuminuria (0.2%) and socioeconomic level (0.4%), which are well known factors that affect glycemic control in patients with diabetes (8). The considerable effect of treatment characteristics such as insulin treatment (4.9%) and combination therapy (0.8%) can be explained easily by the fact that more intensive treatment is usually required in patients with higher A1C levels.

In none of the other studies in which multivariable models were built to assess the impact of ethnicity (11,12,16–18) were many sociodemographic and clinical factors included. The current study reinforces previous reports that A1C differs in various ethnic groups after adjusting for other factors that could affect A1C.

When comparing our findings to the findings of previous studies on this topic, differences in aims, methodology, and study population must be taken into account. One study (16), conducted in patients with diagnosed type 2 diabetes, compared A1C levels with random blood glucose values. In this study, A1C was higher by 0.3% in African Americans than in Whites, at a glucose level of 8.3 mmol/L. In the same study, an A1C of 7% correlated with a random blood glucose value of 7.5 mmol/L in African Americans and 8.4 mmol/L in Whites. Another study (11) also focused on African American and White races, but the study participants were diagnosed with impaired fasting glucose, and the assessed correlation was between A1C and 2-hour plasma glucose after a 75-g oral glucose load. This study also found that A1C levels were higher in African Americans than in Whites, on average by 0.3%, through the plasma glucose range of 7.8–11.0 mmol/L after adjustment for glucose levels and other important predictors. Another study (17) showed even higher A1C levels in African American compared with White patients with diabetes, with an average difference of 0.47% after adjustment for plasma glucose and other covariates.

Interestingly, two studies from Singapore, one in patients with known type 2 diabetes (12) and the other in patients without known diabetes (18), showed similar results. For a given fasting glucose value, A1C was higher in Malays and Indians by 0.9–1.1 mmol/L, at a fasting plasma glucose level of 5.6 mmol/L, and by 2.1–2.6 mmol/L at a fasting plasma glucose level of 7.0 mmol/L relative to Chinese people. In both studies, an interaction between ethnicity and fasting blood glucose was found, with a greater difference at higher glucose levels (12,18).

In other words, in some ethnicities, A1C has higher sensitivity for both the detection of new cases of prediabetes and diabetes and the monitoring of diabetes control. Indeed, in studies that aimed to detect the sensitivity of A1C in different ethnicities, different optimal thresholds were found. In a Danish population, for example, the Inuit had significantly higher A1C levels than the Danes at any given level of fasting and 2-hour postprandial glucose. The prevalence of diabetes diagnosed by A1C versus glucose load was >10–20% higher among Inuit residents in Greenland and Inuit migrants compared with a few percentage point difference in native Danish individuals (19). One study (26) showed that A1C had a 13% higher sensitivity in detecting prediabetes in individuals of Middle East ancestry than in those of Swedish ancestry in Sweden. In Canada, the sensitivity of A1C for the diagnosis of both diabetes and dysglycemia was much higher in Caucasians than in First Nations and Metis individuals (29 vs. 22% and 47 vs. 24–33%, respectively) (13). An A1C cutoff of 6.3%, as compared with 6.5%, was more sensitive for the diagnosis of type 2 diabetes in Sri Lankan adults (17). These studies demonstrate that different thresholds of A1C can be optimal for different ethnicities for the diagnosis of prediabetes and diabetes and for follow-up of diagnosed patients with type 2 diabetes.

In our study, an A1C of 53 mmol/mol (7%), which is a treatment goal for type 2 diabetes control for most patients, correlated with fasting glucose levels of 141 mg/dL in non-Ethiopian Jews, 136 mg/dL in Ethiopian Jews, and 126 mg/dL in Arabs. An A1C of 64 mmol/mol (8%), which is a treatment goal for patients with limited life expectancy or at risk for the potential harms of treatment (27), correlated with fasting glucose levels of 184 mg/dL in non-Ethiopian Jews, 180 mg/dL in Ethiopian Jews, and 169 mg/dL in Arabs. Thus, while, in Arab patients, the recommended A1C goal of 53 mmol/mol (7%) may reflect controlled type 2 diabetes based on fasting glucose level, this level correlates with a fasting glucose level that is higher than the recommended value for Ethiopian Jews and much higher than the recommended value for non-Ethiopian Jews.

Several explanations have been proposed for the ethnic/racial differences in A1C. Lifestyle customs and habits such as dietary fat intake (28), alcohol consumption (29), and smoking habits (30) may affect A1C independent of glycemia. Several single-nucleotide polymorphisms that operate through nonglycemic mechanisms and affect erythrocyte parameters rather than glucose metabolism have been associated with A1C (31–33). Additionally, enzymes that deglycate A1C have been described in several populations (34,35).

Most guidelines, including those from Israel, do not take ethnic differences in A1C into consideration, but rather use the same thresholds for the assessment of type 2 diabetes in all patients. Our study findings indicate that, while in the Arab minority, the relationship between fasting glucose and A1C among people with type 2 diabetes is in line with ADA recommendations, higher fasting glucose levels correlate with the same A1C levels among Ethiopian Jews and especially non-Ethiopian Jews.

This finding implies that making timely adjustments to treatment plans for type 2 diabetes may be a more complex task for treating physicians than is usually assumed. Clinically, continuing the present approach of using uniform thresholds of A1C for all ethnicities in Israel could affect optimal glycemic control. Lower thresholds and goals are appropriate for Ethiopian and non-Ethiopian Jews with diabetes. The results of our study can help clinicians from other countries with similar populations, such as neighboring Arab countries and Ethiopia, as well as physicians treating similar small communities in more distant countries, in adapting treatment goals for patients with type 2 diabetes.

Knowledge and awareness of the effects of ethnicity on A1C could help clinicians determine more appropriate A1C targets and assist in tailoring treatment to individual patients. This knowledge also could assist in the education of patients living with diabetes, empowering them to make more informed decisions and achieve successful diabetes self-management. Because a patient-oriented approach is one of the cornerstones of optimal glycemic management, it is crucial to individualize glycemic targets. Given the high prevalence of diabetes and the potential ramifications of poor control, every tool that can help medical teams and patients achieve optimal blood glucose control would be valuable.

In the broader sense, our findings signify that different A1C thresholds must be developed for different ethnicities, even within the same country. We encourage physicians and researchers, especially in multiethnic communities, to invest in assessment of the relationship between A1C and blood glucose in different ethnicities to inform better glucose monitoring and treatment adjustments. Increasing immigration rates and globalization make this issue even more relevant.

A potential weakness of this study is that our database included fasting glucose only. Postprandial glucose levels affect A1C to a greater extent (7), so it is possible that there are no ethnic differences in the correlation between A1C and postprandial glucose or even that the correlation is in the opposite direction. Another limitation is that we could not determine whether these ethnic differences in A1C reflect differences in the risk for microvascular and macrovascular complications of type 2 diabetes. However, because most guidelines recommend fasting glucose for diagnosis, self-monitoring, and even treatment adjustments, our findings are clinically relevant. An additional limitation is the retrospective nature of the study, with possible missing and/or inaccurate information in patients’ records. Some important factors that influence A1C, such as self-management engagement, could not be assessed in this study, which was based on data from computerized records.

The main strength of this study is the size of the study population. It is the first study in which multiple clinical and laboratory covariates that could affect A1C levels were taken into account when exploring the association between ethnicity and A1C. This strategy enabled us to develop a model with excellent goodness of fit that explained 74.6% of the variance of A1C.

This study reinforces and further expands on previous reports of the existence of ethnic differences in A1C in patients with type 2 diabetes independent of other factors that could influence A1C level. These differences should be considered by the clinicians in patient care, by researchers in future studies, and by policymakers in guideline development.

The authors thank Miss Bracha Cohen of the Soroka Clinical Research Center at Soroka University Medical Center in Beer-Sheva, Israel, for her assistance with data extraction.

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

Y.T.-G. wrote the manuscript and researched data. I.F.L. contributed to discussion and reviewed and edited the manuscript. R.P. researched data. Y.T.-G. 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.