Hemoglobin A1c–Using Epidemiology to Guide Medical Practice: Kelly West Award Lecture 2020

Historical Perspective: Focus on Blood Glucose

Glucose has been central to the diagnosis of diabetes for centuries. The first systematic epidemiologic investigations of glucose in the 1960s demonstrated that a substantial portion of asymptomatic patients with diabetes had a high prevalence of complications at the time of screening (1). At the time, Dr. Kelly West stated, “Well designed long-range prospective studies of subjects who have had various kinds of tests for diabetes will be very helpful in determining the most appropriate criteria for interpreting these tests” (2). Landmark epidemiologic investigations, including the Whitehall study, subsequently established that fasting and 2-h glucose levels were associated with retinopathy, albuminuria, and future development of heart disease, stroke, and death (3–5).

A 1965 report by the World Health Organization established an early definition of diabetes in asymptomatic individuals based on elevated 2-h glucose. In the 1970s, optimal definitions were still being debated. In 1979, the National Diabetes Data Group (NGDP)—Dr. West was a member of the workgroup—established a single set of criteria for the diagnosis of diabetes with cut points at 140 mg/dL (7.8 mmol/L) for fasting glucose and 200 mg/dL (11.1 mmol/L) for 2-h glucose (6). These criteria were reevaluated in the mid-1990s, and new criteria were published in 1997 with the fasting glucose threshold for a diagnosis of diabetes lowered to 126 mg/dL (7.0 mmol/L) (7). These diagnostic cut points were largely based on cross-sectional associations of glucose measures with microvascular disease, particularly retinopathy (7).

Taking into account all available evidence, the most useful and appropriate short definition of diabetes mellitus is simple, “too much glucose in the blood.”

—Kelly West (1978), in Epidemiology of Diabetes and Its Vascular Lesions

HbA_1c for Management of Diabetes

Glycated hemoglobins were discovered in the late 1960s (8). In 1968, Dr. Samuel Rahbar conducted hemoglobin electrophoresis in blood samples from 1,200 patients and found that two individuals showed an “abnormal fast moving hemoglobin fraction” and that both of these patients were also found to have diabetes (9). This work subsequently led to the discovery of HbA_1c and the observation that HbA_1c was elevated in the setting of diabetes (10). Further research established the implications of this finding for the management of diabetes, fundamentally changing diabetes care. Drs. Ronald Koenig, Charles Peterson, and Anthony Cerami demonstrated that the HbA_1c molecule could be used to monitor glucose control in patients with diabetes. They showed that HbA_1c reflected average exposure to blood glucose over the life span of the erythrocyte and proved that HbA_1c was a valid and reliable measure of long-term glucose exposure in humans (10).

Epidemiologic evidence is crucial for the incorporation of a new biomarker into clinical practice. Large epidemiologic studies have demonstrated the association of HbA_1c with retinopathy and other diabetes complications (11–14), establishing its value as a prognostic marker. Randomized clinical trials established that interventions that lowered HbA_1c slowed or prevented complications in persons with type 1 and type 2 diabetes (15,16). This evidence is the basis for the use of HbA_1c treatment targets for diabetes control. Assays became widely available in the 1980s, and HbA_1c was rapidly adopted as the standard measure used in clinical practice to monitor glucose control in patients with diabetes.

HbA_1c for Diagnosis of Diabetes

Despite the strong epidemiologic evidence for its prognostic utility, it took several more decades before HbA_1c was recommended for the diagnosis of diabetes. A major barrier to the adoption of HbA_1c as a diagnostic test was a lack of standardization of the HbA_1c assays (17). The NGSP (ngsp.org) (formerly, the National Glycohemoglobin Standardization Program) was established in 1996 to implement a system of reference laboratories that would calibrate and standardized HbA_1c assessment methods and ensure comparability of results with the reference method established in the Diabetes Control and Complications Trial (DCCT) (15). As a result of the efforts of the NGSP, HbA_1c assessment was well standardized by ∼2008, removing this barrier to its use as a diagnostic test.

In 2009, an International Expert Committee convened by the American Diabetes Association, the European Association for the Study of Diabetes, and the International Diabetes Federation first recommended the use of HbA_1c for diagnosis of diabetes. Citing pooled data on the association of HbA_1c with prevalent retinopathy, the committee recommended an HbA_1c cut point of 6.5% for diagnosis (14). This recommendation was adopted in guidelines issued by the American Diabetes Association, World Health Organization, and other diabetes groups across the globe.

There are a number of advantages of HbA_1c for diagnosis of diabetes. First, HbA_1c has much lower biological (within-person) variability as compared with fasting glucose or 2-h glucose (18). Second, unlike glucose level, HbA_1c is an index of overall glycemic exposure, providing a window into past hyperglycemia over the prior 2–3 months. Third, HbA_1c does not need to be measured in the fasting state and HbA_1c assessment does not involve burdensome timed sampling like the oral glucose tolerance test, making it a convenient test for patients and providers. Fourth, for HbA_1c there are fewer preanalytical factors that can affect laboratory results, and it is relatively unaffected by physical activity, stress, or recent illness, which can alter glucose concentrations. Finally, HbA_1c is familiar to patients and providers, as it has been used for monitoring glucose control and guiding and adjusting diabetes treatment for decades.

Diagnostic cut points for diabetes have historically been based on epidemiologic studies demonstrating strong associations of biomarkers of hyperglycemia with prevalent retinopathy (11, 12). Population studies have also established HbA_1c as a potent marker of future risk of diabetes and major complications such as heart disease and kidney disease, even among individuals without a history of diabetes (19–21). We undertook one such investigation in a large community-based cohort, the Atherosclerosis Risk in Communities (ARIC) study (22). This work, published in 2010, demonstrated the importance of HbA_1c as a marker of future risk for diabetes, cardiovascular disease, and mortality, providing support for its use as a diagnostic test for diabetes.

Use and Interpretation of HbA_1c and Fasting Glucose as Diagnostic Tests

Diagnostic cut points for fasting glucose and HbA_1c will not always classify the same individuals as having diabetes. The cut point of 6.5% for HbA_1c has higher specificity as compared with fasting glucose 126 mg/dL (7.0 mmol/L); many people with elevated levels of fasting glucose will have an HbA_1c <6.5%. We provide here equivalent values of fasting glucose and HbA_1c based on percentile distributions in the U.S. adult population without diabetes (Table 1). An HbA_1c value of 6.5% will, on average, be roughly equivalent to a fasting glucose of 136 mg/dL (7.6 mmol/L) in the general adult population without a history of diabetes.

A simple and efficient approach to the diagnosis of diabetes is to measure fasting glucose and HbA_1c in a single blood sample. Until 2019, guidelines recommended that a second test be conducted in a new blood sample to confirm the diagnosis of diabetes. A more streamlined approach is to conduct two different diagnostic tests (e.g., fasting glucose and HbA_1c) in the same blood sample: if both tests are elevated, this confirms the diagnosis of diabetes (23,24). With this approach one can avoid the need for repeat bloodwork and potential delays in patient care. In the ARIC study, we examined the risk of a new diabetes diagnosis, kidney disease, cardiovascular disease, and mortality among individuals meeting this single-sample confirmatory definition of undiagnosed diabetes. We found that this definition had high positive predictive value for a future diagnosis of diabetes and identified adults at high risk for microvascular and macrovascular outcomes. This work demonstrated the efficiency and clinical utility of measuring HbA_1c and fasting glucose in a single blood sample and prompted changes in diagnostic guidelines in 2019.

In children and adolescents, fasting tests can be unduly burdensome and there has been controversy regarding optimal approaches to screening and diagnosis of diabetes. HbA_1c has practical advantages in this population as it does not require fasting and has low within-person variability. We recently demonstrated that HbA_1c measurement identifies children and adolescents with a high burden of cardiometabolic risk and is a useful screening test for prediabetes and diabetes for this population (25).

When diabetes diagnostic test results for HbA_1c and glucose in the same patient do not agree, health care providers must adjudicate this discordance. Because glucose is one of the most common laboratory tests in the practice of medicine, providers and scientists tend to be inured to its limitations (26–28). When laboratory measurements of glucose and HbA_1c are discordant, it is important to consider a potential problem with either test (28) (Table 2). For example, if a low glucose is observed in the setting of a high HbA_1c test result, a sample processing problem for glucose might be explored: when samples are not processed promptly, glycolysis will cause low glucose concentrations. Insufficient fasting (i.e., <8 h) is a common problem that can cause unexpectedly high glucose. Iron deficiency or other anemias can alter HbA_1c and might also be evaluated when glucose and HbA_1c test results are discordant.

As with all laboratory tests, HbA_1c and glucose results need to be viewed in full context of the patient. Most factors that interfere with laboratory results for HbA_1c are uncommon and many will be detected on other routine laboratory tests (e.g., anemia). Modern HbA_1c assays are unaffected or relatively unaffected by common hemoglobin variants (HbS, HbC, HbE, HbD), but some methods will give inaccurate results (especially for HbF) (ngsp.org). Hemoglobin variants arose from natural selection, most likely as a protective mechanism against malaria in carriers. The prevalence of abnormal hemoglobin variants globally is ∼5% but is higher in certain population subgroups (29). HbS may be as high as 25% in some parts of sub-Saharan Africa (30). The prevalence of HbS is ∼8% among Black persons in the U.S. (31,32). Because of potential interference, it is important that health care professionals know which method their laboratory is using. In patients with two alleles of abnormal variants (HbSS, HbCC, or HbSC, for example), the HbA_1c test should not be used due to altered erythrocyte turnover.

Controversies in the Interpretation of Racial Differences in HbA_1c

There is evidence for a small, but systematic, difference in HbA_1c (∼0.3%-points) according to race/ethnic ancestry that is independent of glucose (33–35). On the heels of the recommendation for the use of HbA_1c for diagnosis of diabetes in 2009, concerns were raised about the interpretation of observed race/ethnicity differences in HbA_1c.

Studies documenting race/ethnicity differences in HbA_1c have been widely misinterpreted to suggest that HbA_1c is a less valid test for certain race/ethnicity minority groups, especially Black adults. Differences in HbA_1c have also been used to promote the potentially harmful use of race-specific cut points for screening and diagnosis of diabetes. These claims have been made despite a large and robust literature linking HbA_1c with clinical outcomes in diverse populations (20,22,36) and a lack of evidence for racial differences in clinical trials of glucose-lowering interventions (37). Indeed, most studies show a higher risk of diabetes in Black adults and other race/ethnicity minority populations compared with White adults (38,39). There is no evidence for race/ethnicity differences in the correlations of HbA_1c with average glucose (assessed by continuous glucose monitoring [CGM]) or fasting glucose (40,41).

Evidence for small, glucose-independent differences in HbA_1c is not completely understood but likely arises from genetic variation (42,43). While some genetic variants may be more common in certain race/ethnicity groups, using race/ethnicity as a proxy for genetics or for poorly understood health-related factors is poor medical and scientific practice. Diabetes and its complications disproportionately affect race/ethnicity minority groups in the U.S. and other countries. These disparities primarily stem from a complicated mix of social factors including racism, historical factors (enslavement, segregation), opportunities for educational attainment and employment, environmental factors (the built environment, transportation, housing, food availability, pollutants, media exposure), health behaviors (physical activity, diet), and health care (access to care, health literacy, quality of care, communication). Race is a highly imprecise construct. Using race/ethnicity differences to justify differential approaches to diagnosis or treatment can do harm by legitimizing differences in treatment standards based on race/ethnicity rather than relying on objective, biological measures.

Potential Alternatives to HbA_1c: Serum Biomarkers of Hyperglycemia

In settings where HbA_1c is problematic (e.g., patients with altered red cell turnover or certain hemoglobinopathies), alternatives include fructosamine and albumin, which can be measured in serum or plasma (44). Fructosamine and glycated albumin are both ketoamines, formed by the reaction of glucose with proteins (nonenzymatic glycation). Fructosamine reflects the glycation of total serum proteins, predominately albumin but also globulins and lipoprotein. Glycated albumin is reported specifically as a proportion of total albumin. Serum proteins have a shorter half-life and undergo glycation at a higher rate as compared with hemoglobin. Thus, fructosamine and glycated albumin reflect short-term (2- to 3-week) glycemic control (45).

Fructosamine measurement is available from major laboratories in the U.S. Glycated albumin measurement is newly available in the U.S. (cleared for clinical use by the U.S. Food and Drug Administration in 2020) but has been used in Japan, Korea, China, and some other countries for a number of years. These biomarkers have been proposed for use in monitoring short-term or interim glycemic control as they will respond more quickly to changes in diabetes treatment as compared with HbA_1c.

The acceptance of new biomarkers is partly dependent on establishing their associations with clinically relevant outcomes. In our work we have established the prognostic value of fructosamine and glycated albumin measures, demonstrating robust associations with microvascular and macrovascular outcomes, with predictive values similar to HbA_1c (46–51). Statements from diabetes and laboratory organizations have suggested that these biomarkers may be useful but have not provided formal guidance on when and how they should be used in clinical practice (52). The results of our studies—particularly evidence of similar prediction relative to that of HbA_1c—suggest that fructosamine and glycated albumin may be useful as substitutes for HbA_1c or as complements for monitoring short-term glucose control.

Cut points are necessary for disease diagnosis, treatment monitoring and decision-making, and health care payment. Because there is no consensus on clinical cut points for fructosamine or glycated albumin, one approach is to use values that are roughly equivalent to those used for HbA_1c and fasting glucose. For example, our data from the ARIC study suggest that an HbA_1c of 7% is roughly equivalent to a fructosamine value of 280 μmol/L and a glycated albumin value of 17% (48) (Table 3).

HbA_1c for Population Surveillance: Accurately Estimating the Burden and Control of Diabetes

The widespread availability of HbA_1c testing has had a major effect on public health and diabetes surveillance. HbA_1c is routinely measured in large epidemiologic studies and national surveys. These data are used to estimate the burden of prediabetes and diabetes in the population and to evaluate trends in glucose control among patients with diabetes. National data on HbA_1c, such as those from the National Health and Nutrition Examination Survey (NHANES), allow us to monitor the population-level impact of diabetes, guiding allocation of public health resources.

We have used data from NHANES to evaluate trends in the prevalence of undiagnosed and diagnosed diabetes (53) and to document trends in diabetes control in U.S. adults (54,55). Modern point-of-care technology that can accurately and rapidly measure HbA_1c in a finger stick further opens up the opportunity for more wide-scale population screening and epidemiologic surveillance without the need for fasting or venous samples. In high-income countries, data on trends in undiagnosed diabetes and prediabetes are based on laboratory testing done as part of resource-intensive epidemiologic studies. Few data are available in the rest of the world.

Point-of-care HbA_1c testing is widely used, but there is substantial variability across devices, with some showing very poor performance and high bias (56). For this reason, HbA_1c point-of-care devices are not recommended for the diagnosis of diabetes. If methodological and standardization barriers can be overcome, it is possible that well-calibrated point-of-care HbA_1c testing could be used effectively in epidemiologic research, potentially offering an affordable alternative that could be implemented in low- and middle-income countries to fill a gap in global diabetes surveillance.

CGM and HbA_1c: Better Together

HbA_1c is invaluable for diagnosis and management of diabetes, but it does not provide information on hypoglycemic episodes or glucose variability. CGM is a novel technology that provides detailed information on glucose patterns and can detect hypoglycemia and short-term glucose variability. Recent studies have demonstrated the utility of CGM in the management of type 1 diabetes (57,58), and guidelines recommend the use of CGM technology for people with diabetes (of any type) who are on intensive insulin therapy (59).

CGM can add nuance to HbA_1c. However, there are a number of downsides to CGM that pose barriers to its widespread adoption for monitoring glucose control (Table 4). One issue in its interpretation is that CGM technology measures subcutaneous interstitial glucose. Interstitial glucose is determined by glucose diffusion from the plasma into the interstitial space and will be affected by blood flow and other factors (60). CGM devices have poor accuracy at the low (hypoglycemic) range (60–63), and CGM sensors from different manufacturers demonstrate discordance with each other (65–67). CGM technology generates huge amounts of data, with up to ∼1,000 to ∼5,000 measurements of glucose in one patient, typically over a 14-day period. For simplification of this information, summaries of these data are provided to patients including mean glucose, the coefficient of variation, and percentage time spent “in range” (typically 70–180 mg/dL). Even when this information is simplified, the amount of information can be overwhelming to patients and providers. It remains unclear how to use CGM data to optimize care, especially for patients who are not on insulin therapy.

To provide a summary measure of glucose control from CGM, some have suggested using an estimated HbA_1c, termed the “glucose management indicator” (GMI). However, this measure is unlikely to replace HbA_1c. The GMI is based on interstitial glucose measurements, is not standardized or validated, and will not necessarily align with laboratory HbA_1c. Studies have not yet demonstrated a clinical benefit of providing estimated GMI values to patients, and distinguishing “expected” from “unexpected” discordance between GMI and HbA_1c may be difficult for patients and health care providers.

Rigorous epidemiologic studies are needed to evaluate CGM as a useful adjunct measure to HbA_1c. The literature on CGM primarily comes from studies of populations with type 1 diabetes, often predominately White and educated patient populations being treated at academic medical centers. There are few large studies in diverse populations of adults with type 2 diabetes and sparse epidemiologic data linking CGM use and its metrics to long-term outcomes. Moving forward, we need rigorous studies in diverse populations that address how to use CGM and HbA_1c in a complementary manner to improve health outcomes for patients with diabetes.