Diabetes: A1c Questions/Diagnosis
A Report of the Diabetes Mellitus Interagency Coordinating Committee
On March 11, 2011, the Diabetes Mellitus Interagency Coordinating Committee convened a meeting on the uses and relative merits of hemoglobin A1c (A1c) tests, in particular for the diagnosis of diabetes and prediabetes. The speakers provided expert overviews of various aspects of scientific research on the A1c test. Federal panelists then provided perspectives from their agencies on the use of A1c tests. This summary document is intended to combine pertinent elements from all of these presentations, and to be made available to interested parties in the scientific community and the public.
Dr. David Nathan: Historic and Clinical Overview of A1c
Dr. David Nathan began by reviewing the history of HbA1c, from its discovery to its use as a biomarker for diabetes control and diagnosis. His own work began in the early 1980s, when he characterized the relationship between percent A1c and average capillary blood glucose, which demonstrated a potential clinical role for the assay, later confirmed by clinical trials such as the Diabetes Control and Complications Trial and the United Kingdom Prospective Diabetes Study. These trials clearly demonstrated for both type 1 and type 2 diabetes that blood glucose control that reduces A1c levels confers dramatic reductions in the risk of disease complications. HbA1c subsequently became the universally endorsed measurement of chronic glycemia with standardized goals. The importance of the HbA1c assay led to efforts to standardize and improve quality of the tests, as will be discussed in the talk by Dr. Randie Little.
Dr. Nathan then reviewed the large body of evidence correlating A1c with average glucose levels. Although this work has led to the development of reliable equations to convert A1c values to average blood glucose levels, the correlation is imperfect, owing in part to the inaccuracy and imprecision of the methods to measure glucose levels. Moreover, although reporting HbA1c as an estimated average glucose level, in the same units as the self-monitoring of blood glucose results, has intuitive appeal, there is currently little evidence indicating that such reporting leads to improvements in care compared to reporting A1c levels themselves. Dr. Nathan reviewed the data showing that A1c rises with age, as do other relevant metabolic measures including fasting blood glucose and oral glucose tolerance tests, in parallel with the overall increase in risk for diabetes over a person’s lifetime.
Improvements in standardization of the assay world-wide, the ease of measuring A1c compared with fasting measures of glucose or oral glucose tolerance testing, and the confirmation of HbA1c as a relevant clinical biomarker that correlates with risk for disease complications in all racial and ethnic groups suggested that it might be useful as a diagnostic test. An A1c level greater than or equal to 6.5 percent has been recommended by an International Expert Committee as a means of diagnosing diabetes and has now been adopted by the American Diabetes Association and World Health Organization. An important caveat to this approach is that certain assay methods may provide erroneous results in the presence of hemoglobin variants and assay methods that are “blind” to these interfering factors must be selected. Medical conditions that result in altered red blood cell turnover can render HbA1c levels inaccurate as a measure of average blood glucose, regardless of the assay method. In this setting, measurements of glucose are necessary for the diagnosis and management of diabetes.
Dr. Randie Little: Current Status of A1c Measurement
Dr. Little, network coordinator of the National Glycohemoglobin Standardization Program (NGSP), reviewed organization of the NGSP, and the processes of certification for new A1c measurement methods, as well as proficiency testing for the laboratories that perform clinical A1c tests. Although methods used to measure A1c have changed considerably over the years, the DCCT and UKPDS remain reference standards, so that a given percent A1c found in a modern test would correspond to the same percent found in those clinical trials. She noted that there are now about 100 NGSP certified methods and about the same number of NGSP certified laboratories. In addition to the certification process, the College of American Pathologists (CAP), using samples characterized by the NGSP, performs twice annual proficiency tests of each laboratory to monitor for consistent test quality. Dr. Little presented data from these CAP surveys which demonstrated a dramatic improvement in accuracy and consistency over the period from 1993 to 2010, for a range of A1c values.
To address the question of whether A1c testing as it is currently performed is “good enough,” Dr. Little referred to DCCT and UKPDS data showing that very small reductions of HbA1c conferred very significant reductions in risk for complications. She noted that the ADA and other international groups consider an absolute change in A1c of 0.5 percent to be clinically significant. Therefore, the question is whether clinical laboratory testing is accurate to within less than that value. CAP criteria for proficiency testing now require results to be within ±7 percent of the NGSP assigned value. In the normal and treatment target range (<7 percent A1c), if a lab consistently passes the CAP limit, then the lab is likely to give results within 0.5 percent A1c of the target. More than 90 percent of laboratories currently pass the CAP survey. Dr. Little noted that further improvements in method performance can be achieved by tightening both NGSP manufacturer certification criteria and CAP proficiency criteria. CAP and NGSP have gradually increased stringency of these criteria, and further increases are expected in 2011 and 2012.
Dr. David B. Sacks: Global Harmonization of A1c
Dr. Sacks noted first that although the United States and some other countries continue to use NGSP numbers (i.e,, percent A1c) for clinical reporting of A1c levels, many other countries are changing to only the International Federation of Clinical Chemistry and Laboratory Medicine (IFCC) numbers. IFCC values differ significantly from those used by the NGSP, but the two can be easily calculated from one another by a simple equation. Note, however, that IFCC numbers will generally be reported as mmol/mol, rather than as a percent. The IFCC/IDF/EASD/ADA consensus statement from 2010 recommends that scientific literature report both IFCC A1c units (in mmmol/mol) and derived NGSP units (in percent). Despite the consensus conference, true “global harmonization” has not occurred to date.
Dr. Sacks then expanded on Dr. Little’s discussion of the impact of hemoglobin variation on A1c tests, pointing out that people who completely lack the normal version of the gene for hemoglobin A cannot be meaningfully tested for A1c by any test. However, it is much more common to need to test A1c in people who have one normal and one variant copy of the gene, and in these patients, as noted by Dr. Little, many methods are accurate. He also discussed POC testing, noting that they are convenient, and are approved by the U.S. Food and Drug Administration (FDA) for home use. However, the FDA has also placed these devices in a category that is “waived” for ongoing proficiency testing. Indeed, some POC test devices have been found not to meet analytical performance criteria, and even a device that is accurate when used by trained personnel when it leaves the factory might later lose calibration or be used improperly, and therefore provide erroneous A1c readings in clinical use. For this reason, Dr. Sacks agreed with Dr. Little and the ADA consensus statement that POC A1c testing is inappropriate for use in diagnosis of prediabetes and diabetes.
Dr. Sacks also explored the question of whether the changes in diagnostic criteria are having an impact in the way diabetes is being diagnosed. He showed data from an individual laboratory from March 2009 and May 2010 showing that the proportion of tests with values below the definition for diabetes increased significantly over that time, suggesting that more clinicians are using the A1c test on patients who have either not previously been diagnosed with diabetes, or who have controlled their blood glucose to a remarkable degree. He noted that over that time period, the total number of tests had increased by 28 percent.
Dr. Lawrence Phillips: Screening with A1c-based Criteria
Since identifying people who have early diabetes or prediabetes is desirable to allow preventive care, it is worth considering how best to screen for these problems. Key issues:
- Early diabetes and prediabetes are asymptomatic.
- Most U.S. healthcare systems do not screen.
- The oral glucose tolerance test (OGTT) provides sensitivity to identify early disease, but findings vary with day to day metabolic status, the test is not widely used, and it is inconvenient – it takes 2 hours, and it has to be done in the morning after a prior fast.
- A1c is widely used, consistent from day to day, and convenient. But is it accurate enough in all people?
The OGTT is the current standard for diagnosis around the world, plasma glucose measurements are precise and accurate, varying only 2 to 3 percent from lab to lab, and OGTT glucose levels are thought to be an accurate measure of metabolism; glucose levels vary from day to day because of underlying differences in insulin secretion and insulin resistance. In contrast, A1c measurement varies much more from lab to lab, and individuals with the same glucose levels may have different A1c levels. Such individual to individual variation is a problem in using A1c to screen.
Compared to OGTT testing, use of A1c to screen with the American Diabetes Association (ADA) diagnostic criteria would miss about 70 percent of the people who have diabetes – close to 6 million Americans, and 70 percent of those with prediabetes – 43 million Americans, and label another 6 million who have normal OGTTs as being high risk or diabetic. These problems with A1c screening are due to individual to individual variation, and to the ADA diagnostic criteria being specific but insensitive. A1c is also affected by racial/ethnic background, and by age. People with the same glucose levels have higher A1c if they are African-American, or if they are older.
Because of these issues, A1c testing will have some use in screening for diabetes, but be of limited benefit in screening for prediabetes.
Dr. Robert Cohen: Biological and Clinical Variability in A1c
Dr. Cohen began by asking to what extent observed differences between A1c and blood glucose are due to biologic variability versus measurement methods or inadequate time integration. He used the A1c-Derived Average Glucose Study data previously shown by Dr. Nathan as an example of variation too broad to be explained solely by measurement variation despite the best techniques available. By inference, that has to reflect biologic variation. The A1c observed in people with an average or mean blood glucose (MBG) level of 150 mg/dl varied from 5.9 to 7.5 percent while in those with an A1c at 7 percent, MBG ranged from 130 to 200 mg/dl. A1c levels that are higher or lower than expected for a given MBG in people with normal red blood cells and hemoglobin could result from physiological differences in fundamental mechanisms determining A1c.
He discussed quantifying this degree of variation with a “glycation gap” defined as the difference between an individual’s measured A1c and the A1c that would be predicted by another measure such as that individual’s glycated serum proteins or MBG. Glycation gap is highly reproducible as is A1c, further evidence that this effect is biological. This glycation gap can then be used to quantify the mismatch and test specific mechanisms for their contribution to the variation.
Using this glycation gap strategy on data from 157 individuals, Dr. Cohen found that 17 percent of these people had an A1c level more than 1 percent below the A1c value predicted by the other test, and 23 percent had an A1c value more than 1 percent above their expected A1c level. Because a difference in A1c level of 0.5 percent or more is considered clinically significant, such differences could have real impact on care. He offered the view that the gold standard should be consistency between multiple independent measures.
He then reviewed data on physiologic mechanisms to explain this biologic variability. While it is commonly taken for granted that glucose gets into all people’s red blood cells equally, Dr. Cohen showed evidence that there are differences in this property between people affecting the level of glucose inside red blood cells, where it has the potential to react with hemoglobin and form A1c. Recent data from his laboratory demonstrated that synthesis of A1c in vivo in people with and without diabetes is linear over time and therefore precisely dependent on the amount of time the red cell is in the circulation. In the same studies, they found that the lifespan of red blood cells can vary from person to person, whether or not they have diabetes, by wide enough margins to affect diagnoses and clinical decision-making.
In summary, Dr. Cohen showed that although A1c is an excellent marker of glycemic control on a population basis, it cannot necessarily be taken at face value for every individual. It is likely that simple techniques can be worked out to overcome that limitation in the coming years.
Dr. William Herman: Impact of Race and Ethnicity
Dr. Herman reviewed the impact of race and ethnicity on A1c levels independent of average blood glucose. After reviewing the impact of hemoglobinopathies, which are not randomly distributed in the global population, he provided an overview of studies comparing A1c among different racial and ethnic groups. In general, most racial and ethnic groups had higher levels of A1c than were observed in non-Hispanic whites. These findings have been reproduced using data on people with type 2 diabetes from managed care organizations, adjusted for specific measures of care. For example, the Translating Research Into Action for Diabetes study found that A1c levels of white participants averaged 7.7percent, for African-American participants 7.9 percent, and for Hispanic, Asian/Pacific Islander, and other ethnic/racial groups, 8.1 percent. Differences were also found in young people without diabetes, adults with impaired glucose tolerance, and adults with both recent onset and long-standing type 2 diabetes. He noted one study that found that markers of glycemia both inside and outside of red blood cells were higher in African-Americans than whites, suggesting that true differences in glycemia may drive differences rather than the effects of glucose transport noted in Dr. Cohen’s talk. This suggests an important question: if differences in hemoglobin glycation among racial and ethnic groups are real, do they indicate differences in susceptibility to diabetic complications? The answer to this question remains unknown.
Dr. Elizabeth Selvin: A1c: Reliability and Risk Prediction
Dr. Selvin began by reviewing the ADA’s 2010 recommended diagnostic guidelines for diabetes: A1c>6.5 percent; or a FPG>126 mg/dl; or a 2 hour plasma glucose during an OGTT>200 mg/dl; or in a patient with classic symptoms of hyperglycemia or hyperglycemic crisis. As with most diagnostic tests, re-testing should be performed to rule out laboratory error, unless the diagnosis is clinically clear. The reliability of the various tests recommended is illustrated by data from the Third National Health and Nutrition Examination Survey (NHANES III). Of the 3 recommended blood tests, OGTT had the highest variability on retest, while A1c exhibited the lowest. Further, while 70 percent of individuals with a FPG>126 mg/dl retested as having diabetes using FPG, 80 percent of those with A1c>6.5 percent retested as having diabetes using A1c. She noted that the ADA developed these recommendations based on A1c values associated with subsequent diabetes related complications—largely retinopathy, as demonstrated in large scale clinical trials and epidemiological studies. However, few data exist on A1c and long-term outcomes in persons without a prior diagnosis of diabetes. To address this problem, she compared the ability of A1c and FPG to predict risk for diabetes, coronary heart disease, stroke, and all-cause mortality over 15 years in participants from the Atherosclerosis Risk in Communities (ARIC) study who were not known to have diabetes before the study began. She found that predictive models based on FPG were improved by addition of A1c, but that predictive models based on A1c were not improved by addition of FPG. These observations illustrate that A1c is superior to fasting glucose for assessment of long-term prognosis of the likeliest causes of death in people with diabetes. Importantly, she found that despite significant differences between blacks and whites in A1c at baseline, race did not modify the associations between A1c value and cardiovascular outcomes and death.
Panel Discussion: FDA, VA, CDC, and CMS
Dr. Courtney Harper is the Director of the Division of Chemistry and Toxicology Devices in the FDA Office of in Vitro Diagnostic Device Evaluation and Safety, which regulates both in-home and laboratory diagnostic tests including those on the market for A1c. She emphasized the importance of making clinicians and patients aware of what assays are most appropriate for various diagnostic and disease monitoring purposes, and of encouraging use of the most accurate assays that are practically available. She noted that information about specific assays and their approved uses and limitations (such as reliability of A1c testing in the presence of variant forms of hemoglobin) are available in the product labeling and also on the FDA website (http://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfPMN/pmn.cfm - search by assay name, manufacturer, or using the product code LCP). Dr. Harper emphasized the desire of her office to improve regulation of assays used for clinical diagnostic purposes, and solicited the input of the DMICC on how FDA can better achieve this goal. DMICC members inquired about what would be required to reverse the FDA decision to waive POC A1c devices for proficiency testing, as discussed during Dr. Sacks’ presentation. Dr. Harper responded that it is difficult but not impossible, and recommended that people interested in petitioning FDA to do so should provide to FDA in their petition the clinical safety evidence that would support such an action.
Dr. Leonard Pogach, is the Veterans Health Administration (VHA) National Program Director of Endocrinology and Diabetes (Office of Patient Care Services/Office of Specialty Care) and representative to the DMICC. He spoke on “VHA-DoD Guidelines for Use of A1c in Clinical Practice: The Importance of Laboratory Precision and Accuracy.” Dr. Pogach began his presentation by noting the known variation between A1c results obtained from the same laboratory or across different laboratories as well as emerging evidence of racial differences in A1c results. He highlighted the fact that the 2010 VHA Diabetes Clinical Practice Guidelines address these issues and emphasize the importance that clinicians be aware of test accuracy and precision.
He noted that the VHA diagnostic guidelines for prediabetes and diabetes are different from those recommended by the American Diabetes Association (ADA) and the Endocrine Society. The VHA considers a patient to have prediabetes if FPG <126 mg/dL but ≥100mg/dL on two occasions, or A1c readings ≥5.7 percent but <6.5 percent, confirmed with a FPG ≥ 100 mg/dL and <126 mg/dl. The VHA considers a patient to have diabetes if A1c ≥7 percent on two occasions using a clinical laboratory methodology standardized to the NSGP (not POC testing); or if A1c ≥6.5 percent, confirmed with a FPG ≥126 mg/dL. These tests can be done on the same or different days; or FPG ≥126 mg/dL on at least two occasions. He emphasized that in addition to the racial differences, the major reason for the difference between the VA and ADA diagnostic recommendations is that the VA guidelines accommodate for the known variation in A1c results in clinical practice whereas the ADA does not. He cited recent NGSP Clinical Advisory Committee minutes that even an A1c assay with high precision (less than 2 percent coefficient of variation) is still incapable of discriminating a difference between two A1c tests of <0.5 percent A1c as statistically significant. Also, despite standardization to the NGSP, two results from different clinical laboratories cannot be directly compared.
Dr. Pogach went on to express concern about the use of A1c in management of diabetes without taking laboratory variation into account, and the possibility that the improper use of A1c thresholds in standards of care or performance measurement might lead to inappropriate intensification of treatment leading to hypoglycemia or other harms, especially for patients receiving insulin. This concern is particularly acute in populations such as those treated by the VA. Elderly patients with multiple chronic conditions, often including decreased renal function and impaired cognitive function, are an example of the types of individuals for whom a low target A1c level could confer unnecessary risk of serious hypoglycemic episodes. He suggested that the DMICC and NDEP should collaborate to disseminate this important information about A1c test accuracy and other related issues more widely known to clinicians and patients.
Dr. Edward Gregg, an epidemiologist in the Division of Diabetes Translation at the CDC, discussed dilemmas and challenges in the application of risk stratification tools. He noted that the relationship between morbidity and a particular risk factor tends to be continuous—there is generally no precise threshold which determines precisely whether or not a person is likely to develop a particular medical problem. Defining a target threshold for such a risk factor therefore yields both false positives and false negatives. Lowering the threshold for diagnosis of prediabetes using A1c, for example, will identify more of the population which will ultimately go on to develop diabetes, but will also identify more people who will not. Dr. Gregg identified several unanswered questions, including:
- Does the discordance between glycemic markers matter?
- Does incidence of outcomes for a given A1c level differ by race/ethnicity and people in discordant groups?
- Do the optimal risk indicators in adults also apply to youth?
- Should A1c be used in combination with other glycemic markers and other measures of risk for pre-diabetes?
Dr. Elizabeth Koller, of the Centers for Medicare and Medicaid Services (CMS) Office of Clinical Standards and Quality, noted that while CMS has received requests to use A1c as a diagnostic test, there are some statutory issues and CMS staff do not think that there is sufficient evidence for a specific diagnostic cut-point and that there is insufficient precision in such a test-even at the best laboratories. CMS coverage is also based on “reasonable and necessary”. If there are other better tests, it may not be reasonable and necessary to pay for an inferior test—that may require subsequent follow-up. (This is even outside the relative costs for a fasting, random, or 2 hour postprandial glucose test.) Furthermore, the patients in whom the acute diagnosis is most critical, i.e., those with high glucose levels that could result in urinary loss of glucose and nutrients or leukocyte dysfunction will likely be diagnosed with the fasting, random, or postprandial glucose given a likely A1c >8 percent. There may be, indeed likely is, cardiac risk at low A1c levels but that is because the underlying metabolic condition preceded the elevated glucose level (which is only a marker of the metabolic problem). Intervention with glycemic control will not improve the cardiovascular outcomes for these people-especially those >65. The intervention that is needed is blood pressure and lipid control—things which are diagnosed with other than an A1c test.