Introduction
Glycated haemoglobin (HbA1c) is used for assessing diabetes management and more recently also for the diagnosis of diabetes [
1‐
3]. However, it is associated with some limitations [
4‐
7] and its use in many countries, including the USA [
8], is limited. The reasons are mainly financial, but logistical reasons, like no access to laboratories, may also play some role. It is also worth noting that HbA1c does not inform patients about blood glucose values on a daily basis and some patients do not understand what HbA1c means.
Therefore, in 2009, shortly after the publication of the A1c-Derived Average Glucose (ADAG) study [
9] that provided data allowing the estimation of average blood glucose (ABG) values from HbA1c values, the American Diabetes Association (ADA) incorporated those data while framing the guidelines [
10]. This practical algorithm, shown in the form of a table, may serve “as a check on the accuracy of the patient’s meter” [
11]. However, in practice, it can also be used for reverse calculation, i.e. for estimating the HbA1c value from the ABG values and for assessing whether the therapeutic targets have been met in those patients for whom HbA1c measurement is not possible.
In clinical practice, however, there may be differences between the actual, measured HbA1c and the calculated value. In such cases, it is often argued that patients do not record high glucose values in their diaries, and therefore the calculations may be biased. Although it may be true in many cases, this explanation would apply only in the situation where a high HbA1c level is matched with a low ABG, whereas the opposite situation is also observed. It seems, therefore, that there is some biological variability that may influence glycated haemoglobin. Results of some studies suggest that a substantial portion of HbA1c is determined by nonglycemic factors [
12]. Further, certain conditions, e.g. anaemia, rheumatoid arthritis and chronic kidney disease, could have influenced the results of HbA1c.
Studies evaluating the relationship between blood glucose and haemoglobin A1c levels in subjects with diabetes have some limitations. The Irish study was performed in a relatively small group of patients [
13]. A similar study, based on the Diabetes Control and Complications Trial (DCCT) data, was performed only in type 1 diabetes patients and the results differ from the results of the ADAG study in that higher ABG values were calculated from the respective HbA1c values, and the results had a broader distribution, at least at higher HbA1c and glucose values [
14]. In the ADAG study [
9] only 159 had type 2 diabetes. Of the remaining subjects, 80 had normal glucose tolerance (NGT) with low glucose and HbA1c values. The data from the subjects with NGT (low HbA1c and low average glucose) could affect the slope of the regression line. Additionally type 2 diabetes patients were an average of 15 years older than those with NGT and 11 years older than patients with type 1 diabetes and certain conditions in elderly, multimorbid patients with type 2 diabetes conditions could have influenced the results. Moreover, the average glucose value in the ADAG study was calculated by combining weighted results from the continuous glucose monitoring (CGM) performed every 4 weeks (over at least a 48-h period, with glucose levels assessed every 5 min) with 8-point daily self-monitoring of capillary (fingerstick) glucose performed at least every 2 days, or in patients not on CGM, 7-point daily monitoring for 3 days a week during a 12-week study period. This is definitely not a real-life situation. Normally the number of records does not exceed 4 values per day, whereas nocturnal values are only rarely recorded. Thus, in common practice, the calculations are based on a few values captured each day rather than many values captured over greater time intervals. Another study used only fasting plasma glucose values, achieving results similar to those of the ADAG study [
15].
A new analysis of the ADAG Study Group report empirically links target blood glucose values with HbA1c levels [
16]. The authors calculated the average fasting, postprandial and bedtime self-monitored blood glucose concentrations associated with specified HbA1c levels achieved by subjects in the ADAG study. The relationship between HbA1c and fasting glucose may be different from its relationship with postprandial glucose [
13,
17]. A recently published systematic review and meta-analysis found that postprandial glucose has a closer association with HbA1c than fasting glucose (although the heterogeneity score was high) [
18]. In a widely cited study by Monnier, the relative contribution of postprandial glucose excursions was predominant in fairly controlled patients, whereas the contribution of fasting hyperglycaemia increased gradually with diabetes worsening [
19].
Taking into account the above facts, as the reviewed trials show some high-quality evidence but have also some limitations, we have only moderate confidence that a relationship between HbA1c is adequately established. It is, for example, not clear whether the results are equally reliable in patients with type 2 as with type 1 diabetes. Additionally, studies on larger groups are needed. Thus this study aimed to assess the relationship between HbA1c and average glucose concentrations in a large number of type 2 diabetes patients treated with premixed insulin.
Discussion
This study was aimed at assessing the relationship between HbA1c and average glucose concentrations in a large number of type 2 diabetes patients, as this is the most prevalent type of diabetes, and the calculation of ABG from HbA1c (and vice versa) is more useful in type 2 diabetes (in intensively treated type 1 diabetes patients usually both HbA1c and blood glucose are measured and accessible; therefore, such calculation is not necessary). The results were compared with the results of the ADAG study [
9], which is the most widely cited and most widely used for the calculation of ABG from HbA1c. We found that use of HbA1c as a tool to enable average glucose estimations or to check the accuracy of a glucometer may be not as reliable as suggested by the ADAG study and ADA guidelines.
The comparisons of ABG calculation from HbA1c (Table
2) using both the ADAG formula [
9] and the formula derived from the present study begin and end at different values. This is because the ADAG study included healthy subjects with low HbA1c and ABG values.
In the ADAG study the data from the subjects with NGT (low HbA1c and low average glucose) could affect the slope of the regression line. Additionally, it was not clear whether and how the HbA1c values in those groups overlapped; in other words, whether and how many subjects with NGT had HbA1c values higher than the upper limit of the normal range, or how many diabetic subjects had well-controlled disease with low HbA1c. The PROGENS study included only type 2 diabetes patients, which explains the absence of low and the presence of higher ABG and HbA1c values. The comparison shows that both formulas give similar results at low HbA1c values (although the 95% confidence interval is broader in the PROGENS study), but differ at higher HbA1c values. However, the slope of the regression lines would not be as steep. This means that even though the regression formulas in the ADAG and PROGENS studies give similar results at the lower range of HbA1c values (6.0–8.0%) (42–64 mmol/mol), the predicted ABG values at the higher range of HbA1c (9.0–12.0%) (75–108 mmol/mol) will be overestimated with the ADAG formula for our population. In fact, the predicted ABG values for our population in that range were more than 60 mg/dL higher when calculated by the ADAG formula (Table
2). This result is probably caused by differences in HbA1c measurements between the point-of-care and HPLC methods. It also calls into question the reliability of using the ADAG formula in patients with type 2 diabetes, in whom point-of-care devices are used, at least in the higher range of HbA1c. Because higher HbA1c values are common in patients with type 2 diabetes, this discrepancy may be a significant practical issue.
The distribution of the results in the present study is broader than in the ADAG study. This is due to a much higher number of participants and other reasons discussed in the “
Introduction”. The broader distribution and the wider 95% confidence interval may influence the reliability of the calculation. Unfortunately, the ADA guidelines do not show the 95% confidence intervals in the table showing the ABG values calculated from the given HbA1c values [
11], which suggests to practitioners that the calculation is entirely reliable.
In many countries and centres where the access to reliable HbA1c measurement is limited, it is much more common to use the reverse calculation, i.e. from ABG to HbA1c, as shown in Table
3. Translating average glucose values to HbA1c may be a practical solution for those physicians and endocrinologists. However, this study raises some concerns regarding the use of ABG as a tool for patient assessment. For example, the 95% confidence intervals for the calculation of HbA1c from various glucose levels (as shown in Table
3) are broad over the entire range of glucose values. The probability of a real patient’s HbA1c being over the treatment target for a given ABG value is shown in the last column of Table
3. As can be seen, relying only on the estimated HbA1c value makes the probability of inaccurate judgements rather high. In other words, there is a high probability that a physician trying to assess HbA1c from ABG (and therefore trying to assess whether the patient’s diabetes is well controlled) may be in error. An ABG value as high as 220 mg/dL or more is required to achieve a 95% probability that the patient’s actual HbA1c value is over the treatment target (i.e. greater than 7.0%, 53 mmol/mol). At the same time, there is only a 72% probability that a patient with an ABG of 110 mg/dL has an actual HbA1c value of 7.0% (53 mmol/mol) or less. In other words, in almost one third of the patients with an ABG of 110 mg/dl, the actual HbA1c value is over the general therapeutic target. These data highlight the unreliability of this type of assessment. Unfortunately, similar evaluations were not performed in the other studies assessing the relationship between ABG and HbA1c.
The differences between the results obtained in our study and the ADAG study may be due to various reasons. The ADAG study have used standard A1C estimation, whereas in the present study point-of-care A1C is used. The differences may also be due to the different accuracies of the glucometers. It also could be due to the different number of values that had been done in fasting or prandial state. However, it should be emphasized that such differences are also present in real life.
We tested whether the HbA1c values measured by the point-of-care device were reliable, using a direct side-by-side comparison with the gold standard HPLC method. Some studies comparing HbA1c point-of-care test results with those from laboratory analyses suggest no significant differences between the methods [
23‐
25], but others suggest that proficiency testing should mandatory for users of HbA1c point-of-care assays to ensure quality [
26]. The HbA1c analyser A1cNow+ method is included in the NGSP list, and the method is continuously validated. Our study in the local setting also confirmed the correlation between this method and the gold standard HPLC method (Fig.
1); however, the correlation line is shifted in the higher range of haemoglobin values. This difference implies that the real HbA1c may be higher than that measured by the point-of-care device. This is not of importance from a practical viewpoint, as regardless of whether the real HbA1c is 10% or 12%, it is well above the target and the patient requires intensification of treatment. However, it seems that if the line had not been shifted, the slope of the regression line in the ADAG and PROGENS studies would have been similar.
The study has some limitations. First, point-of-care devices were used for assessing glucose and HbA1c, and the results may therefore be different had golden standard methods been used. However, point-of-care devices or other non-laboratory methods are used for HbA1c estimation in clinical practice. Use of point-of-care devices reflects the practice in real-life settings. In addition, only one kind of glucometer was used at all sites, ensuring the comparability of the results. With regard to HbA1c estimation, as described above and shown in Fig.
1, the correlation with HPLC estimations is very good.
The examined population (although numerous, which may be regarded as a strength of the study) is limited to premixed insulin-treated patients with type 2 diabetes. This may be regarded as another limitation of the study. Indeed, it is not known whether and how the treatment may influence the relationship between HbA1c and ABG. Therefore caution is required when translating the results to other populations.
There are multiple mechanisms underlying the variations in HbA1c, blood glucose and the relationship between them [
5,
6,
27]. Genetics are thought to play a significant role in the determination of HbA
1c [
28]. HbA
1c levels are genetically determined, with heritability of 47–59% [
29]. In Jansen’s study [
12] age, gender, BMI, FPG, mean corpuscular haemoglobin, mean corpuscular haemoglobin concentration, current smoking and alcohol consumption were independent predictors of HbA1c, together explaining 26.2% of the variance in HbA1c, with FPG contributing 10.9%. Rather than assuming that it is possible to make precise calculations based on the results of this study, the ADAG study or other formulas, we suggest that such calculations should be seen as only rough assessments of glycaemic control. As the ABG is used more and more frequently, especially in those countries or regions where HbA1c tests are too expensive, inaccessible or both, it is important that physicians be aware of the limitations of this method, in terms of calculation the HbA1c from ABG. In parallel, use of HbA1c for assessing ABG or checking accuracy of a glucometer may also be unreliable.