HbA1c for the Diagnosis of Diabetes and Prediabetes: Is It Time for a Mid-Course Correction?

An estimated 6.2 million people in the United States have undiagnosed diabetes. The average time between onset and diagnosis of type 2 diabetes is 7 yr (1). Diagnosing diabetes is the first step in assuring that appropriate lifestyle, glycemic, and nonglycemic interventions are implemented (2) to reduce the toll that end-organ complications take on the life of the individual and on the health of the nation. The 2010 American Diabetes Association (ADA) standards of care for diabetes, based largely on the opinion of an international expert committee, added hemoglobin A1c (HbA1c) as diagnostic criteria for diabetes (≥6.5%) and prediabetes (5.7–6.4%) (3,4). In theory, wider application of this new approach should reduce the delay in diagnosing diabetes by adding a straightforward test to complement fasting glucose and oral glucose tolerance testing. However, if HbA1c is not sensitive, that is, if it does not identify individuals who truly have diabetes, and if its limitations are not fully appreciated by those implementing it, this new approach could fail to achieve this goal or further delay the diagnosis of diabetes. HbA1c has long been used as a marker of glycemic control in established diabetes. In affected patients, the rate of HbA1c formation is a direct function of the average blood glucose concentration. Compared with glucose measurements, the use of HbA1c as a diagnostic test has advantages, including convenience, less day-to-day variability, greater preanalytical stability, and international standardization (3,4). Disadvantages are: HbA1c is more costly than fasting plasma glucose (FPG), guidelines do not adequately reflect the accuracy of HbA1c measurements available across the nation at the current level of standardization, and more importantly perhaps, it may not correlate adequately with actual glucose levels (5,6). For more than a decade, it has been recognized that there may be a discordance between HbA1c and other measures of glycemia. The study in this issue of JCEM by Lipska et al. (7) highlights the limitations and strengths of HbA1c as a diagnostic test.

Lipska et al. (7) examine the performance of HbA1c in comparison to FPG in diagnosing diabetes and prediabetes in older adults. The participants in the study were between the ages of 70 and 79 yr, and 42% were black. Median HbA1c levels were higher among blacks compared with whites despite the fact that there was no difference in median FPG levels. When the authors compared the newly specified HbA1c cutoff of 6.5% or greater for diabetes to a prior reference standard, i.e. FPG of at least 126 mg/dl, the sensitivity was 57%, and the specificity was 98%. As the authors point out, this confirms that HbA1c misses many true positive cases of diabetes in older adults (3,5). Limitations are that only single measurements of HbA1c and FPG were performed and there was no measurement of 2-h post-glucose load glucose levels. The latter is important because post-challenge glucose and insulin levels deteriorate more rapidly than fasting levels as people age (8,9,10). Hence, when the standard is 2 h postprandial glucose, the diagnosis of diabetes based on HbA1c of 6.5% or greater might be expected to be more sensitive, rather than less sensitive, for diabetes when compared with FPG alone in an elderly cohort.

The authors report similar findings to others regarding racial/ethnic differences between HbA1c and FPG (11,12,13,14,15,16,17) and reinforce that these differences also occur in older adults (18). The use of HbA1c for diabetes diagnosis is based on the hypothesis that HbA1c is a true and consistent measure of mean blood glucose. The best published data describing the relationship were reported in the HbA1c-derived average glucose study (19). HbA1c, averaged across four methods calibrated to the recently developed International Federation of Clinical Chemists standard, was analyzed by regression against mean blood glucose, measured by both blood glucose meter and continuous glucose monitoring calibrated against the HemoCue analyzer to improve continuous glucose monitoring accuracy. Even with all these efforts to minimize measurement variability, a close examination of the relationship reveals substantial uncertainty in the HbA1c predicted from mean glucose or vice versa. This may be disturbing to some, but is not surprising to us. Our view is that the vast proportion of the variance in the HbA1c-mean glucose relationship results from interindividual differences in the biological determinants of hemoglobin glycation. That variation also explains why there is likely to be a trade-off between sensitivity and specificity for a simple HbA1c-based diagnostic cut-off compared with any fasting, 2 h post-glucose load or mean blood glucose determination. Greater precision may be possible if we can better define the biological determinants of hemoglobin glycation and use that understanding to adjust the cutpoints based on demographic, anthropometric, or laboratory measurements. Such approaches are routinely used in clinical medicine, for example, in diagnostic scoring for rheumatoid arthritis or in multivariate risk assessment in osteoporosis.

What are the biological determinants of hemoglobin glycation? We can consider them at the levels of epidemiology (age, race, intercurrent disease), genetics, and physiology. For the sake of this discussion, we will pass lightly over factors that affect red cell turnover such as hemoglobinopathies and thalassemia, and dramatic shifts in red blood cell (RBC) life span such as hemolytic disease and acute blood loss. Gould et al. (20) demonstrated that there are hematologically normal individuals who have a distinct relationship between HbA1c and glucose tolerance that is maintained for up to 4 yr, which they termed “high glycators” and “low glycators.” Although these descriptors are appealing, they imply a greater understanding of mechanism than justified either then or now. Quantification of the degree of difference between HbA1c and either glucose or glycated serum proteins has shown that this is widespread in populations without direct or indirect (e.g. mediated by chronic kidney disease) red cell disease (21,22,23). The strongest evidence for epidemiological differences was related to age (10) and race (11,12,13,14,15,16,17). Kirk et al. (24) demonstrated by meta-analysis 0.65% higher HbA1c levels in African-Americans compared with non-Hispanic whites but did not adjust for potential differences in blood glucose levels. They assumed that the differences in HbA1c levels were due to differences in glycemic control. In contrast, we observed differences in HbA1c by race in three multicenter clinical trials after adjusting for differences in glycemia. The first trial was in a population (n = 3819) with impaired glucose tolerance (11); the second, a population (n = 4360) with recent onset, drug-naive type 2 diabetes (12); and the third, a population (n = 2094) with long-standing, treated type 2 diabetes (13). Despite the differences in the three populations and the different stages of disease, the common finding was a consistent difference in the mean HbA1c-blood glucose relationship by race and ethnicity. These findings were confirmed by subsequent studies (14,15,16,17). Evidence that race introduces heterogeneity into the HbA1c-blood glucose relationship beyond the level of heterogeneity found in the A1c-derived average glucose dataset, which had only a 17% representation (19) of all races apart from Caucasian/white, makes it likely that a single set of HbA1c criteria for all will indeed change what we call diabetes.

The concept of race embodies a wide variety of genetic, environmental, social, and cultural factors, any number of which may be associated with variation in hemoglobin glycation. It will be difficult if not impossible to establish evidence-based, race-specific criteria for the diagnosis of diabetes until the impact of these factors on hemoglobin glycation is better understood. The importance of genetics was first established in studies demonstrating heritability of HbA1c in twins (25,26) and the concept that heritability was notably associated with nonglycemic determinants of HbA1c (27). There is growing literature on genetic determinants. A number of large population studies are providing important new findings on genetic linkage for both glycemic and nonglycemic determinants of HbA1c (28,29). Such studies, however, depend very heavily on the limitations to the methods of phenotyping that are applied.

Identifying and quantifying the physiological mechanisms underlying potential racial differences provides a more rational basis for expressing the determinants of the HbA1c-blood glucose relationship. It is well known that the formation of HbA1c is progressive throughout the life span of a RBC, such that younger cells have lower amounts and older cells have higher amounts of HbA1c. A single clinical measurement of HbA1c represents an average across the RBCs of all ages in a given individual. Therefore, RBC life span is a key determinant of HbA1c because this reflects the time of reaction for glucose and hemoglobin. It has been assumed that RBC life span varies little among individuals except in disease states such as hemolytic anemias. Recent studies using a precise biotin RBC label have documented substantial heterogeneity in RBC life span in hematologically normal people with and without diabetes, with sufficient variation to alter HbA1c by more than ±15% (30). Variation in the rate of hemoglobin glycation was detected (30) and will be of interest in relation to integrated glycemic control. Also, it has long been assumed that the rapid transit of glucose across the RBC membrane would result in a constant and common relationship between the glucose concentration outside the cell and that inside the cell. However, it appears that there is sufficient heterogeneity between people in the steady-state relationship between glucose inside and outside the RBC to produce measureable differences in glycation of hemoglobin compared with glycation of plasma proteins (31). Although methods for these determinations are complex, they should allow us to better define the basis for age, race, and other differences in hemoglobin glycation. It is likely that simple means for clinical translation to diagnostic criteria will then follow.

So where do we go from here? It is evident that we are asking more of HbA1c than it can do. We are now attempting to use it to make decisions beyond the limits of biological variability and measurement error. In this territory, it is likely that “one size does not fit all.” In the critical range, unexplained interindividual differences in hemoglobin glycation may have a disproportionately greater impact on the relationship with glycemia than in more advanced diabetes. The diabetes community has focused on improving the technical performance of the assay; however, the laboratory standardization acceptable for certification by the American College of Pathologists through 2012 (http://www.ngsp.org/news.asp) still will only require a quality control sample to give a result within 7%. For example, the lab’s result on a quality control sample with a reference value of 6.5 HbA1c percentage points would be acceptable within the range 6.0–7.0%. Hence, the laboratory could generate an individual patient report of 6.5% when the “true value” is as low as 6.0% or as high as 7.0%. There is concern with these laboratory sources of error, but now there is also increasing recognition of the interindividual physiological variability in HbA1c, whether on a racial or other basis. It is essential to further elucidate these determinants, to improve the sensitivity of HbA1c, and to facilitate its use as a diagnostic test for diabetes.

If our aim is to accurately recognize diabetes early to prevent or delay complications, it is imperative that we address these limitations of HbA1c more precisely than is currently defined in the ADA recommendations (6). We are not asking to “throw the baby out with the bath water”; rather we believe it is time to plot a mid-course correction to define at a more basic level the variables that determine HbA1c and work to include these in its use for diagnosis of diabetes. In the meantime, a “rule-in, rule-out approach” as presented at the 2010 European Diabetes meetings (32) represents one alternative that provides a practical strategy which can also take account of local variations in HbA1c assay performance. Defining HbA1c of at least 5.5 as normal and at least 7.0 as diabetes, and defining HbA1c of 5.6 to 6.9 as “impaired HbA1c” with a recommendation for follow-up fasting glucose or oral glucose tolerance testing both in that interval and when clinically indicated in any hemoglobinopathy or known disorder of red cell turnover would reduce the frequency of classification error while allowing the major benefits that HbA1c has to offer.