Standardizing the haemoglobin glycation index

Abstract Aims A high haemoglobin glycation index (HGI) is associated with greater risk for hypoglycaemia and chronic vascular disease. Standardizing how the HGI is calculated would normalize results between research studies and hospital laboratories and facilitate the clinical use of HGI for assessing risk. Methods The HGI is the difference between an observed HbA1c and a predicted HbA1c obtained by inserting fasting plasma glucose (FPG) into a regression equation describing the linear relationship between FPG and HbA1c in a reference population. We used data from the 2005–2016 U.S. National Health and Nutrition Examination Survey (NHANES) to identify a reference population of 18,675 diabetes treatment–naïve adults without self‐reported diabetes. The reference population regression equation (predicted HbA1c = 0.024 FPG + 3.1) was then used to calculate the HGI and divide participants into low (<−0.150), moderate (−0.150 to <0.150) and high (≥0.150) HGI subgroups. Diabetes status was classified by OGTTs. Results As previously reported in multiple studies, a high HGI was associated with black race independent of diabetes status, and with older age, higher BMI and higher CRP in normal and prediabetic but not diabetic participants. The mean HGI was 0.6% higher in self‐reported diabetic adults. The HGI was not associated with plasma insulin, HOMA‐IR or 2 h OGTT in participants classified as normal, prediabetic or diabetic. Conclusions The regression equation derived from this demographically diverse diabetes treatment–naïve adult NHANES reference population is suitable for standardizing how the HGI is calculated for both clinical use and in research to mechanistically explain population variation in the HGI and why a high HGI is associated with greater risk for chronic vascular disease.

consistently lower or higher than average, respectively, than other people with similar blood glucose concentrations. Multiple clinical studies confirm that a high HGI (i.e., higher HbA1c than predicted by blood glucose) is associated with greater risk for chronic vascular disease in normal, 2-5 prediabetic, 6 type 1 diabetic 7 and type 2 diabetic 8-11 study populations. A high HGI has also been repeatedly associated with greater risk for iatrogenic (caused by medical intervention) hypoglycaemia in people with diabetes. 8,9,[12][13][14] HGI results between studies is confounded, however, by large interstudy variation in the slopes and intercepts of the linear regression equations used to calculate HGI; a consequence of differences in how blood glucose was measured and in the demographic composition of the study populations.
The HGI could have a practical clinical role in personalizing both hypoglycaemia prevention in diabetes patients and guiding treatment to limit chronic vascular disease in both nondiabetic and diabetic people. Because the HGI measures bias in the quantitative relationship between HbA1c and blood glucose, it could also have a clinical role in diagnosing diabetes when diagnoses were based on HbA1c and blood glucose disagree. 15 Lack of a standardized way to calculate the HGI in diverse human populations poses a significant barrier to both HGI research and the possible clinical use of the HGI.
Widespread standardization would require (1) using assays that give the same HbA1c and glucose results for the same blood sample, and (2) a demographically diverse reference population. We accept that national and international glycated haemoglobin standardization programmes make analytical variation in HbA1c measurement a minor concern. And although there are more comprehensive blood glucose metrics, fasting plasma glucose (FPG) is the simplest, lowest cost and most clinically practical way to assess blood glucose status, especially in economically disadvantaged areas of the world.
Furthermore, there is no evidence that the HGI calculated using mean blood glucose, glycated albumin or other measures of blood glucose status provides superior information about risk compared with the HGI calculated using FPG. 1,15 This report provides rationale for widespread adoption of a simple linear regression equation that can be used to calculate HGI for both research and clinical use. To develop this equation, we first se-

| Data source
NHANES is an ongoing national survey directed by the Centers for Disease Control that uses a stratified multistage probability sampling design to represent the noninstitutionalized U.S. civilian population. 16 The National Centers for Health Statistics (NCHS) Ethics Review Board approved the NHANES study protocol, and each participant provided written informed consent.

What's New?
• The lack of standardization in calculating the haemoglobin glycation index (HGI) in demographically diverse populations is a significant barrier to its clinical use as a measure of bias in the quantitative relationship between HbA1c and blood glucose concentration.
• A simple linear regression equation using fasting plasma glucose (FPG) and HbA1c derived from a demographically diverse diabetes treatment-naïve adult NHANES reference population is proposed to standardize HGI calculation.
• Standardizing how the HGI is calculated will facilitate research into understanding why some people have consistently higher or lower than average HbA1c levels than other people with similar blood glucose concentrations, and why a high HGI phenotype (i.e., higher HbA1c than predicted by FPG) is associated with greater risk for chronic vascular disease.
To assess whether other NHANES participants should be ex- Statin users were identified as participants taking any of the following cholesterol-lowering drugs in the past 30 days for which they needed a prescription: simvastatin, lovastatin, atorvastatin, rosuvastatin, pravastatin or pitavastatin. Glucocorticoid users were identified as participants taking any of the following anti-inflammatory corticosteroids in the past 30 days for which they needed a prescription: prednisone, hydrocortisone, prednisolone, methylprednisolone or dexamethasone.

| HGI classification
The diabetes treatment-naïve adult population was divided into tertile (33.3%) subgroups with a low (<−0.150%), moderate (−0.150% to <0.150%) or high (≥0.150%) HGI. We then compared means of selected biochemical, clinical and demographic variables in low-, moderate-and high-HGI participants in the population as a whole and after further subdivision by diabetes classification based on 2-h OGTT cut points recommended by the ADA for classification as normal (<140 mg/dl), prediabetic (140 to <200 mg/dl) or diabetic (≥200 mg/dl). The reference population regression equation was also used to calculate HGI for participants who were excluded from the reference population including youth 12-19 years of age and selfreported diabetic adults.

| Statistical analysis
SAS software (Windows version 9.4; SAS Institute) and R (Version

3.3.2; R Core Team) was used for all statistical analyses and data
representations. Descriptive statistics were used to characterize the study population. Categorical variables were summarized as frequencies, whereas continuous variables were summarized using means and standard deviations. Log transformations were performed for CRP. Analysis of variance was used to compare means among low-, moderate-and high-HGI participants. Because statistical significance may not reflect a meaningful biological or clinical significance when applied to very large study populations, 22 our interpretation of the results also considered whether (1) there was an expected progressive stepwise increase or decrease in the mean of a variable going from the low-to the moderate-to the high-HGI subgroup; (2) the magnitude of the difference between low-and high-HGI subgroups was biologically or clinically important based on our best judgement; and (3) an association between a variable and HGI had been previously reported.   to the 1999 or later NHANES cohorts because HbA1c assays were not all standardized in earlier cohorts.

| Selection of the diabetes treatment-naïve adult reference population
The first issue we addressed was whether any NHANES partici-  Tables 3 and 4 which show clear trends towards older age with a higher mean HGI in the diabetes treatment-naïve adult population as a whole and in normal and prediabetic participants, but not in diabetic participants. These observations are consistent with multiple reports of an association between higher HGI and older age in nondiabetic study populations 2,3,16 but not diabetic study populations where age either did not differ 10,11,14,[27][28][29][30] or was higher in the low-HGI subgroup. 8,12,13 We also excluded participants with a self-reported history of diabetes or taking diabetes medications because the present study and published research conclusively show that the quantitative relationship between HbA1c and FPG in people treated for diabetes differs from that typically observed in people without diagnosed diabetes. Table 1 and Figure 1 show that the regression equation  Table 2 shows that obesity was associated with a small but progressive increase in the mean HGI going from normal (−0.028) to overweight (−0.011) to obese (0.049) subgroups (range 0.077%). Although 44.5% of diabetes treatment-naïve adult participants were classified as insulin-resistant based on HOMA-IR, the mean HGI was not different from the population as a whole. Diabetes treatment-naïve adult participants diagnosed as diabetic based on an FPG ≥ 126 mg/ dl had a mean HGI that was not different from that observed in the population as a whole. In contrast, participants diagnosed as diabetic based on HbA1c ≥ 6.5% had a + 0.760% higher mean HGI. That HGI is higher in untreated adults with diabetes is graphically supported by Figure 1A which shows that nearly all diabetes treatment-naïve adults with FPG over 175 mg/dl had HbA1c above the population regression line: the hallmark of a high HGI phenotype. Haemolytic anaemia shortens the RBC lifespan, and the amount of time HbA1c has to accumulate inside RBCs. 33 Consequently, with all other conditions equal, a shorter RBC lifespan will naturally lower both HbA1c and HGI. In the present study, NHANES participants with low haemoglobin levels diagnostic of anaemia represented 8.0% of the diabetes treatment-naïve adult population (Table 2) and had a mean HGI that was −0.082% lower than that observed in the overall population. In contrast however, mean haemoglobin concentration was higher in low-HGI participants than high-HGI participants (Tables 3 and 4 HbA1c is reportedly higher than normal in people with iron deficiency 34 or asthma 35 independent of blood glucose concentration, which should produce a higher than normal HGI.

| Clinical implications
Longitudinal studies have shown that people with diabetes tend to have an HGI that is significantly different between individuals but relatively consistent within individuals over time and over the physiological range of blood glucose concentrations. 1,8,38 And there is little doubt that genetic variation is a major source of person-toperson variation in HbA1c. 39,40 The present study adds to a growing list of biological, clinical and demographic factors associated with bias in the quantitative relationship between HbA1c and blood glucose concentration. It is important to note that although the contribution of any one factor to an individual's HGI may be relatively small, combinations of factors that promote low or high HGI could collectively produce individuals with the range of HGIs observed in human populations.