Intended for healthcare professionals


Is haemoglobin A1c a step forward for diagnosing diabetes?

BMJ 2009; 339 doi: (Published 10 November 2009) Cite this as: BMJ 2009;339:b4432
  1. Eric S Kilpatrick, consultant in chemical pathology1,
  2. Zachary T Bloomgarden, clinical professor of medicine2,
  3. Paul Z Zimmet, director emeritus3
  1. 1Department of Clinical Biochemistry, Hull Royal Infirmary and Hull York Medical School, Hull HU3 2JZ
  2. 2Department of Medicine, Mount Sinai School of Medicine, New York, USA
  3. 3Baker IDI Heart and Diabetes Institute, Caulfield South 3162, Australia
  1. Correspondence to: Eric S Kilpatrick eric.kilpatrick{at}

    Eric Kilpatrick, Zachary Bloomgarden, and Paul Zimmet question proposals to diagnose diabetes by raised glycated haemoglobin concentration rather than glucose testing

    Currently, diabetes is diagnosed by measuring plasma glucose concentration fasting (threshold ≥7 mmol/l) or after a oral glucose tolerance test (≥11.1 mmol/l). However, an international expert committee comprising members appointed by the American Diabetes Association, the European Association for the Study of Diabetes, and the International Diabetes Federation recently recommended replacing these tests with one for glycated haemoglobin A1c (HbA1c).1 The committee stated that type 2 diabetes should be diagnosed in anyone with a confirmed HbA1c value ≥6.5% (48 mmol/mol) without glucose testing, although the glucose criteria will continue to be used in people in whom measurement of HbA1c may be inappropriate. Measurement of HbA1c has several advantages over glucose but its exclusive use could present problems, as we discuss here.

    Advantages of haemoglobin A1c

    The expert committee document gave no specific reasons for dispensing with glucose criteria in favour of HbA1c, but it did highlight many of the advantages of using HbA1c (table). These include the undoubted benefits of requiring a single blood sample and being able to test in the non-fasting state. The day to day variability of HbA1c within an individual is also smaller than that of fasting glucose and considerably less than glucose concentration after a glucose tolerance test (coefficient of variation 3.6% v 5.7% v 16.6% in one study2), so repeated measurements should be more consistent with HbA1c. There is also the argument that, by giving an estimate of glycaemia over the preceding few weeks or months, HbA1c could provide a more complete view of glycaemia than a single fasting glucose measurement or the “artificial” conditions of a glucose tolerance test, which the expert committee refers to as the “common sense” rationale for its use. HbA1c measurement is also the most common means of guiding management of diabetes and adjusting treatment, so its use for diagnosis would simply be an extension of this role.

    View this table:

    A previous hurdle to even considering use of HbA1c for diagnosis has been the lack of standardisation in the assay, meaning that results could vary depending on the laboratory method used. This is being overcome through standardisation led by the International Federation of Clinical Chemistry and Laboratory Medicine (IFCC) and should now be less of a problem.3


    The case to move to HbA1c for diagnosis seems compelling. However, there are some real difficulties. Indeed, the expert committee accepts many of the inherent problems in using HbA1c for diagnosis but do not discuss the likely practical effects of these limitations. For example, the presence of any abnormal haemoglobin may affect HbA1c measurement and give misleading results. How and if it will affect the result depends on the analyser used. One instrument might identify and account for certain haemoglobinopathies but not others, and a different analyser could pick up (or miss) a completely different spectrum of abnormal haemoglobins.4

    The prevalence of haemoglobinopathies varies widely between countries and races, but an indication of the magnitude of the problem is given by looking at people in the United States with either sickle cell or haemoglobin C trait. At least 10% of the 26 million African American citizens have these traits,5 which can cause incorrect results in at least a third of HbA1c analysers routinely used there.4 Just how, therefore, are we going to be sure that someone does not have a haemoglobinopathy that is causing them to be diagnosed with diabetes (or not) inappropriately? In patients already known to have diabetes, the National Institutes of Health in the United States recommends that this possibility should be considered in people of African, Mediterranean, or South East Asian heritage when glucose measurements are discrepant with HbA1c results, when the HbA1c result is unexpected, when the result is greater than 15%, or when a value changes drastically after a change in laboratory method.6 But if HbA1c is the sole means of diagnosis and there is encouragement not to self monitor glucose concentrations until diabetes requires insulin treatment, we cannot identify wrongly diagnosed patients without the additional expense, interpretation, and consent required for concurrent haemoglobinopathy screening.

    Iron deficiency anaemia (a condition which affects 3.3 million US women7) can also cause HbA1c to be 1-1.5% higher than usual before iron treatment.8 9 Should we be testing for this as well as haemolytic anaemia and any other condition that can alter red cell survival in anyone with suspected diabetes? What about patients with renal failure, which can have a variable effect on HbA1c concentrations (through haemolytic and iron deficient processes as well as the formation of carbamylated haemoglobin), and conditions such as HIV, where HbA1c appears 1% lower in patients taking antiretroviral drugs?

    The committee mentions the effect of ageing (HbA1c is 0.4% higher in 70 year olds than 40 year olds with the same glucose tolerance)10 and ethnicity (0.4% higher in Afro-Caribbeans than Europids)11 but thinks that their “etiology and significance are unclear.” So in the meantime we do not know if we will wrongly identify older and non-Europid people as having diabetes. Indeed, the committee fails to consider whether there is important within-patient variability in the relation between HbA1c concentration and mean glycaemia.12 13

    Even the move to standardisation of HbA1c measurement, although necessary, will not instantly improve the performance of assays. If we make an analogy between laboratory HbA1c analysers and wristwatches, then international standardisation is the equivalent to setting our watches to an atomic clock rather than Big Ben. This will not, in itself, make our watches more accurate. This was shown in June 2009 when the UK National External Quality Assessment Service found that 251 laboratories sent standard samples with an HbA1c concentration of 6.5% (48 mmol/mol) reported values varying between 5.8% and 7.2% (40 mmol/mol and 55 mmol/mol). Variation could be even higher in countries with fewer resources to measure HbA1c and less experience with harmonisation of methods. These issues, and the potential list of tests required in addition to HbA1c to ensure its accuracy, make the idea of simply fasting overnight for a glucose test much more appealing.

    The aim of diagnosing type 2 diabetes is to identify people at high risk of developing the microvascular complications of the disease. As HbA1c will identify diabetes in different groups of people than would glucose tests, there remains concern about how well HbA1c compares with glucose in predicting microvascular risk, even after people in whom measuring HbA1c is likely to pose a problem are excluded. The main figure in the expert committee’s report shows three studies (in Pima Indian, Egyptian, and US populations) that found that the risk of retinopathy increases with rising fasting plasma glucose, two hour glucose, and HbA1c levels in roughly the same decile, implying that the tests are interchangeable. However, this would be expected within a population no matter how poorly one of the tests predicts risk compared with another. What is not mentioned is that in all three studies receiver operating characteristic curves show fasting or two hour glucose measurement to be better than HbA1c.14 One study cited to justify the 6.5% HbA1c cut-off even found that random glucose testing provided “similar results.”15 The report presents new data showing HbA1c to be at least as predictive of retinopathy as glucose measurement, but these data have yet to be fully published.

    Before considering any change, we also need to know how the current WHO recommendation of measuring two hour glucose in patients with impaired fasting glucose results,16 as practised in many countries, compares to measuring only fasting plasma glucose or HbA1c. Also, in large population studies HbA1c is often measured by the same method, or even the same instrument, in a central laboratory. This removes the between laboratory differences in HbA1c results already mentioned and may give a false impression of how well the test will perform routinely.

    Lastly, there is the HbA1c cut-off of 6.5% (48 mmol/mol) itself. Any threshold is likely to be arbitrary to some extent, but a proposed aim of using HbA1c is to help reduce the time between onset of diabetes and diagnosis and to pick up the third of patients who have diabetes but do not know it.17 However, according to data from the US National Health and Nutrition Examination Survey, 50-60% of patients with a fasting plasma glucose concentration above the current threshold of 7 mmol/l will have an HbA1c concentration below 6.5%.17 Thus, use of a 6.5% HbA1c threshold will add patients to the missing third. What this presumably means is that in people for whom HbA1c measurement is known to be unreliable, the use of glucose criteria will make them two to three times as likely to be diagnosed with diabetes as someone in whom HbA1c testing can be used. We also have the problem of whether to diagnose diabetes in someone with a fasting glucose of 10 mmol/l and an HbA1c of 6.4% (47 mmol/mol).


    When new diagnostic criteria for any disease are proposed it is assumed they will be demonstrably superior to existing criteria. There are advantages and disadvantages to using either plasma glucose or HbA1c to diagnose diabetes, as summarised in the table. However, measuring glucose may present less of a risk of complete misdiagnosis than measuring HbA1c alone. Rather than being a step forward, a move to HbA1c only diagnosis could be a step too far.


    Cite this as: BMJ 2009;339:b4432


    • Contributors and sources: ESK is honorary professor in clinical biochemistry with a 20 year clinical interest in diabetes and assessment of glycaemic control. ZTB is a clinician treating diabetic patients. PZZ has clinical, molecular, and epidemiological research interests. All authors drafted and amended the article. ESK is the guarantor.

    • Competing interests: None declared.

    • Provenance and peer review: Not commissioned; externally peer reviewed.


    View Abstract