If it were a boxing match, the debate over whether hemoglobinshould be used to diagnose diabetes would place the odds-on favorite in the “Yes” corner. In the “No” corner would be the underdog. At least based on the mainstream consensus since 2010, HbA1c for diagnosis is well established as an alternative to measuring glucose.
At the July 2013 meeting of the American Association for Clinical Chemistry, where opposing sides on this question squared off in a debate, a quick vote beforehand showed “Yes” with a 20-to-one edge. But the speakers came armed with provocative data on comparative benefits and drawbacks, and both sides scored solid points.
Most of the debate centered on two questions: Which is more accurate, HbA1c or glucose? And with regard to diagnosing diabetes and predicting risk of complications, what are the most important things to measure anyway?
“Even though we’ve been measuring glucose in blood for 100 years, the question that comes up is: How accurate is a glucose result?” said moderator David Sacks, MB, ChB, FRCPath, senior investigator with the National Institutes of Health, in introductory remarks before the debate.
Not only is there the usual preanalytical and analytical variation, but there is also large biological variability among people and even within a single person, he pointed out. “That means for every one of you sitting here, if you are healthy, your fasting plasma glucose can have a coefficient of variation of up to 8.3 percent between today and tomorrow, even if everything’s the same.”
HbA1c is not a perfect alternative, but since 1993, when the NGSP (formerly the National Glycohemoglobin Standardization Program) launched efforts to standardize HbA1c, there has been a huge improvement in the assay, Dr. Sacks said. “The means are much closer to the target, and the standard deviations have been reduced considerably. So while HbA1c is not perfect, it’s getting progressively better over time.”
In Dr. Sacks’ view, there are substantial deficiencies in the glucose criteria for diagnosing diabetes. Hemoglobin A1c standardization has significantly improved the measure. “Of course there are deficiencies in hemoglobin A1c. But hemoglobin A1c can be measured accurately in the vast majority of subjects with hemoglobin variants.”
The FDA’s approval of Roche’s Tina-quant HbA1c Dx blood test last spring showed just how far the HbA1c assay has come, said Robert A. Vigersky, MD, director of the Diabetes Institute at Walter Reed National Military Medical Center. In contrast with coefficients of variation typical of HbA1c 20 years ago, “the coefficient of variation in this assay was 1.5 percent. And I think that actually makes a very good case for the use of HbA1c for diagnosis, because it has by far the least analytic variability or overall variability.”
The American Diabetes Association made a big shift in 1997 when it lowered the recommended cut point for diagnosing diabetes. “If you have a fasting plasma glucose of 126 mg/dL or greater, you have diabetes. And prior to 1997, it had been 140. And the reason for that change was that it became obvious from the data that complications, particularly the microvascular complications like retinopathy and nephropathy, were appearing in people who had blood sugars much lower than 140.”
Still, Dr. Vigersky said, the variability of fasting glucose is high. “Of all the tests, the oral glucose tolerance test has the most variability, besides which it’s the most cumbersome to perform and our patients don’t like doing it.”
What are the implications of this lack of reproducibility? He pointed to a Swedish study of 2,200 women of whom 241 were diagnosed as having diabetes based on the oral glucose tolerance test. But when the test was repeated, 19 percent of them had normal glucose tolerance, 32 percent had impaired glucose tolerance, and only 49 percent had the diagnosis confirmed.
Findings like these were what led the American Diabetes Association to convene an International Expert Committee in 2009 to weigh whether HbA1c should be used for diagnosing diabetes. “They came down on the side of yes.” In 2010, the ADA added an HbA1c of 6.5 percent or greater as a diagnostic cut point.
“And needless to say, this wasn’t accepted by everyone, and we’re still having this debate four years later,” Dr. Vigersky pointed out. But he thinks it’s significant that when the recommendation to use HbA1c for diagnosis was adopted, the typical CV was much higher than it is now. “This new FDA approval is such a game changer for the field, because we now have an assay that when the CVs were 10 percent the International Expert Committee already thought it was acceptable to use, and now we’re down to a real CV of 1.5 percent.”
The diagnosis of diabetes is very different depending on which parameter you use, Dr. Vigersky said. The reason: “We’re measuring very different aspects of dysglycemia.” On a Venn diagram showing prevalence of diabetes by different measures, the largest circle would be the two-hour post-glucose load and the smallest would be the circle of HbA1c.
How many people have diabetes, diagnosed or undiagnosed? “By fasting plasma glucose of 126 there are 21.5 million people based on the latest data from 2005–2006. Using just HbA1c, there would be fewer. If you used either glucose or HbA1c, there would be 22.4 million. If you used fasting plasma glucose or a glucose tolerance test there would be more. So depending on the diagnostic criteria, there are somewhere between 20 and 27 million people in the U.S. who have diabetes.”
Depending on what you want to believe, “you might want to pick, for political reasons, the criterion that gives you the highest number of people,” Dr. Vigersky said. “And I’ve heard people in important places say they’ll be damned if they’ll allow criteria to exist that show there are fewer people with diabetes.”
Interestingly, the data show a much greater increase in the prevalence of diabetes based on HbA1c over the past 10 years compared with fasting plasma glucose, he said. “In the diagnosis of prediabetes, if you have the cut point of an HbA1c of 5.7, in 1999 the prevalence was 10 percent and now the prevalence is almost 20 percent, almost a doubling in the last decade of prediabetes based on HbA1c.”
“But prediabetes affected 25 percent of the population if you used fasting glucose in 1999–2000 and only a slight increase to now—27.5 percent. So we’re measuring different things. We’re measuring the pathophysiology of a different aspect of dysglycemia when we throw in the HbA1c as a measure for the diagnostic criteria.”
It’s true that glycation rates affect interpretation of HbA1c, Dr. Vigersky said. Studies have shown for a mean glucose over a three-month period, “there’s a significant splay in the data of HbA1cs. And I think you would conclude that some people are high glycators and some people are low glycators at the same blood glucose levels.”
But these differences have important implications, he pointed out. “If you look at the complications of retinopathy and nephropathy, those with high glycation rates had much higher percentages, by a factor of six, compared to those with low glycation rates. So I would argue that even if we understand that there’s biologic variability and genetic variability in glycation rates, it may be beneficial to overestimate and pick up those with high glycation rates, because they in fact are at most risk of developing complications of diabetes.”
A powerful study (Selvin E, et al. Glycated hemoglobin, diabetes, and cardiovascular risk in nondiabetic adults. N Engl J Med. 2010; 362:800–811) looked at the prognostic value of HbA1c and fasting glucose done in patients during 1990–1992 and did a followup in 2010, some 20 years later, to see which of these two parameters, HbA1c or fasting glucose, might be more predictive of cardiovascular events and death. “And it turns out that fasting glucose was not predictive at all in patients without diabetes,” Dr. Vigersky said.
“If you had a fasting glucose in the prediabetes range of 100 to 125 mg/dL, it had no predictive value for coronary heart disease. But if you had an HbA1c between six and 6.4 percent, you had an 88 percent risk of developing coronary heart disease. Similar effects were found in comparing HbA1c with glucose for ischemic stroke and all-cause mortality. “So does HbA1c predict cardiovascular disease as well as glucose? I’d say actually better, since it reflects other aspects of dysglycemia.”
In Dr. Vigersky’s view, the advantages of HbA1c in diagnosing diabetes are many. “Biochemically, it has greater analytic accuracy and better reproducibility. Clinically, it’s more convenient and fasting isn’t required, and there is less day-to-day variation during periods of stress and illness. Epidemiologically, it’s a better predictor of cardiovascular events and death than fasting. And there are some interesting pathophysiologic aspects related to the different glycation rates that may better predict complications.”
In conclusion, he said, “I would say that HbA1c should be used to diagnose diabetes. But—and this may need to be modified going forward—if you only use an HbA1c it should be seven percent or greater, or in combination with a fasting glucose of 126 or greater if the HbA1c is between 6.5 and 6.9.”
“Should hemoglobin A1c be used to diagnose diabetes? I would answer no to that question,” said William Herman, MD, MPH, director of the Michigan Center for Diabetes Translational Research, in staking out his position in the debate.
He agreed that HbA1c has advantages. “HbA1c is certainly convenient. It can be drawn anytime. It’s less susceptible to short-term lifestyle modification. Patients are not going to game you or, if they fast for three days before their appointment, will not fool you in terms of their average level of glycemia. Compared to glucose, HbA1c certainly has a smaller intra-individual biologic variation with about a four percent coefficient of variation, and it has benefited tremendously from international assay standardization.”
“But HbA1c has a major disadvantage for diagnosis of diabetes, and that is the potential for bias that consistently lowers or raises the measured HbA1c value relative to the true level of glycemia.” It’s a systematic error, he added, and it creates the most problem in the near-normal glucose range.
Many factors may alter hemoglobin A1c test results—for example, hemoglobinopathies, thalassemia syndromes, or decreased red cell age—and the recommendation is that if the patient has any of these factors, either a different assay or an alternative measure of glycemia should be used, Dr. Herman said. But possibly more problematic may be the less well-known factors that may alter HbA1c results.
If the patient has uremia, hyperbilirubinemia due to liver failure, iron deficiency—which occurs in about 20 percent of women of reproductive age—excessive use of vitamin C and vitamin E, hypertriglyceridemia, or excessive use of opiates, “these can affect the accuracy of HbA1c, and if you know about them you can use an alternative measure,” he said. “But if you don’t know about them, you may be fooled by the HbA1c test result. And if that’s not bad enough, there are a number of yet unknown factors that may alter HbA1c levels relative to the true level of glycemia.”
He believes the push to use HbA1c to diagnose diabetes has to do with familiarity bias and ambiguity aversion. “Clinicians are used to using HbA1c to monitor glycemic control in patients with established diabetes, and they question, well, why not just use the same test for the diagnosis? But I think this puts us on a slippery slope.”
The problem he sees is that at the lower end of the HbA1c range, which is where the diagnostic numbers are, HbA1c levels vary considerably among individuals, but within individuals they change little over time. Only about one-third of the variance in HbA1c can be explained on the basis of measures of glycemia.
“Glucose is glucose, yet an HbA1c of 6.5 corresponds to very different average glucose levels across individuals. So why may HbA1c not be an accurate measure of glycemia? I think to answer this, one needs to think of the ration-ale or the assumptions we make as to why HbA1c is a good measure of glycemia,” Dr. Herman said.
It’s commonly assumed that, first, erythrocyte lifespan is constant; second, erythrocytes are freely permeable to glucose; and third, HbA1c is formed slowly and nonenzymatically. But studies have shown all three of those assumptions may not be true, he said. “In the hematologically normal population, mean red cell age has been found to be about 50 days, but there is physiologic variation from about 40 days to 60 days. And if one looks at the impact of a 10-day shorter mean red cell age, it will in fact lower HbA1c about one point.”
Hemoglobin lives in the intracellular environment, while glucose is in the extracellular environment, Dr. Herman said. “The differences between intracellular and extracellular measures of glycemia are common in clinical practice and largely unexplained. This difference is measured as the glycation gap, and within individuals it is a characteristic of the individual and is stable over time.”
Interestingly, he added, there seems to be high heritability of glycation gap in nondiabetic twins. “If one twin has a high glycation gap, the other does too, and this leads us into the question of genetics. Here, I think the data are showing associations between genetic polymorphisms and hemoglobin glycation independent of glycemia.”
As people age, there’s very little change in fasting glucose, very little change in fructosamine, but a largely unexplained increase in HbA1c, such that older people have higher HbA1c levels than younger people. Smoking appears to affect hemoglobin glycation, alcohol consumption seems to affect it inversely (lifelong abstainers have HbA1c levels .41 percent higher than people who have three or more drinks a day), and an increase in total daily energy intake from dietary fat is associated with an increase in HbA1c, Dr. Herman said.
“In addition, there are well-described racial and ethnic differences in hemoglobin A1c. There is a consistently higher HbA1c in racial and ethnic minority groups compared with whites, despite identical glucose profiles. So we come then to the question, should hemoglobin A1c be used to diagnose diabetes?”
“If the clinical laboratory-based hemoglobin A1c assay—not a point-of-care assay—is available, it could be performed. Or if the glucose testing is not convenient. If you have a high index of suspicion and a patient is coming after lunch for a visit, then sure, do a hemoglobin A1c level,” Dr. Herman advised. “But if you’re suspicious of type 1 diabetes with a relatively recent onset of severe hyperglycemia, then HbA1c testing is not appropriate. You’re certainly much better off measuring glucose, and you can measure HbA1c when there are no known patient factors that preclude interpretation of the HbA1c.”
He would argue, however, that that is almost never. “As we’ve already pointed out, there are a number of patient factors that preclude the interpretation of hemoglobin A1c. And I would argue that most clinicians screening a patient for type 2 diabetes are not going to know when these things are going on. You certainly could do a hemoglobin electrophoresis and a complete blood count, do a reticulocyte count, do a stool guaiac for blood loss, do a BUN-creatinine-liver function study, a lipid profile, iron studies, drug screens, and assess all of these other factors, and do your hemoglobin A1c and feel relatively confident that it’s accurate.”
“But if you don’t want to do those things, you could also consider measuring a glucose level. If hemoglobin A1c is measured without glucose, undetected, systematic bias may exist and repeated measurements of hemoglobin A1c will not reveal the true level of glycemia. If an individual has low-grade hemolysis and the clinician keeps measuring hemoglobin A1c, it’s always going to be low. If glucose is measured, random error may occur, but repeated measurements will reveal the true level of glycemia,” Dr. Herman said.
An audience member asked Dr. Herman why he was so confident in the ability of the laboratory to measure glucose accurately, given all the problems of getting specimens to the lab, and standardization issues involving whole blood versus plasma. Dr. Herman conceded the point. “It’s true—I’ve seen tubes sitting out on desktops for hours waiting to get transported to the labs. And these are all real issues. But they are addressable issues as well, and I’d like to see us spend as much effort on that as we’ve spent on hemoglobin A1c standardization.”
His conclusion: “Hemoglobin A1c is potentially a systematically biased measure of glycemia, particularly in the near-normal range. Some reasons are known; some reasons remain unknown. And glucose may be less precise, but it is certainly an unbiased measure of glycemia and remains the preferred diagnostic test.”
In the end, the debate over HbA1c versus glucose couldn’t be said to produce a knockout, a split decision, or a draw. Unlike a prizefight, this debate will go into overtime. When Dr. Sacks polled the audience again, clearly a few had been convinced to switch to “no on HbA1c.” But the majority were still proponents—leaving plenty of grounds for a reprise of the HbA1c debate, which the AACC and the Endocrine Society plan to hold at this year’s meeting in July.