Clinical Biochemistry 48 (2015) 514–518
Contents lists available at ScienceDirect
Clinical Biochemistry journal homepage: www.elsevier.com/locate/clinbiochem
Relationship between measured average glucose by continuous glucose monitor and HbA1c measured by three different routine laboratory methods Tze Ping Loh a,⁎, Sunil Kumar Sethi a, Moh Sim Wong b, E. Shyong Tai c, Shih Ling Kao c a b c
Department of Laboratory Medicine, National University Hospital, Singapore Department of Laboratory Medicine, Khoo Teck Puat Hospital, Singapore Department of Medicine, National University Hospital, Singapore
a r t i c l e
i n f o
Article history: Received 26 December 2014 Received in revised form 3 February 2015 Accepted 23 February 2015 Available online 1 March 2015 Keywords: Glycated hemoglobin A1c Glucose Continuous glucose monitoring Correlation Regression
a b s t r a c t Objectives: The relationship between glycated hemoglobin A1c (HbA1c) and average glucose has been described by the empirically derived estimated average glucose (eAG) equation in the A1c-Derived Average Glucose (ADAG) study, with extensive calibration efforts in four secondary reference HbA1c methods. It is not known if this relationship is preserved when HbA1c is measured by routine laboratory methods under routine conditions. Design and methods: We measured average glucose (mAG) by six days of continuous glucose monitoring in 45 adults with stable HbA1c (b1% HbA1c change in the preceding two months). Subjects with medical conditions that may confound HbA1c measurement, including anemia and hemoglobinopathy, were excluded. HbA1c was measured using Bio-Rad Variant II (cation-exchange HPLC), Bio-Rad in2it (boronate affinity HPLC) and Roche Tina-quant (immunoassay) methods. Results: The average differences between eAG derived from the routine HbA1c methods and mAG were 10.4% (Variant II), 6.0% (Tina-quant) and 1.0% (in2it). The regression coefficients between the mAG and HbA1c are different between in2it (mAG, mmol/L = 0.58 × %HbA1c + 2.3), Tina-quant and Variant II (both mAG, mmol/L = 0.66 × %HbA1c + 1.9). However, the 95% confidence intervals of the slope and bias of these methods overlap. The correlation between mAG and HbA1c was greatest when measured using the Variant II (r2 = 0.84), followed by Tina-quant (r2 = 0.82) and in2it (r2 = 0.71). Conclusions: The relationship between HbA1c and measured average glucose is method-dependent despite improved HbA1c standardization. The differences in relationship may reflect as discrepant eAG and home glucose monitoring results. © 2015 The Canadian Society of Clinical Chemists. Published by Elsevier Inc. All rights reserved.
Introduction Glycated hemoglobin A1c (HbA1c) is formed from non-enzymatic glycation of hemoglobin. HbA1c concentrations are highly correlated with average glycemia, and reflect the glycemic control of a patient over the past two to three months. The relationship between average glucose and HbA1c has been described by the empirically derived estimated average glucose (eAG) equation in the seminal A1c-Derived Average Glucose (ADAG) study [1]. Several other groups have also examined the relationship between average glucose by continuous glucose monitoring (CGM) and HbA1c Abbreviations: HbA1c, glycated hemoglobin A1c; NGSP, National Glycemic Standardization Program; HPLC, high performance liquid chromatography; eAG, estimated average glucose; mAG, measured average glucose; CGM, continuous glucose monitoring; CVa, analytical coefficients of variation. ⁎ Corresponding author at: National University Hospital, 5 Lower Kent Ridge Road, 119074 Singapore, Singapore. Fax: +65 67771613. E-mail address: [email protected] (T.P. Loh).
in other populations. The linear relationship between average glucose and HbA1c has been demonstrated in children and adult with type 1 diabetes [2–4], and patients with concomitant advanced chronic kidney disease and type 2 diabetes [5], albeit with different correlation and regression characteristics. A selection of these publications is summarized in Table 1. The ADAG equation is the most widely used eAG equation in routine clinical practice to translate HbA1c values into average glucose. The eAG equation in the ADAG study was derived by correlating the mean of four HbA1c values, which were measured using four different International Federation of Clinical Chemistry and Laboratory Medicine (IFCC) secondary reference laboratory methods with rigorous calibration efforts [6], and with average glucose obtained from capillary blood and CGM. Such rigorous effort in HbA1c measurement is unlikely to be achieved in routine practice. Therefore, it is important to understand the impact of using HbA1c measured by routine laboratory methods under routine laboratory conditions on the relationship with average glucose. Such evidence is
http://dx.doi.org/10.1016/j.clinbiochem.2015.02.012 0009-9120/© 2015 The Canadian Society of Clinical Chemists. Published by Elsevier Inc. All rights reserved.
Medtronic Minimed Gold Medtronic Minimed
Medtronic Enlite sensor
Siemens DCA 2000+
Tosoh A1c 2.2 and Adams Arkray
Bio-rad Variant II, Roche Tina-quant, Bio-rad in2it
+5.95 (4.72 to 7.19) 0.20
−1.16
1.9 (1.2 to 2.5, Variant II); 1.9 (1.1 to 2.7, Tina-quant); 2.3 (1.5 to 3.1 in2it) 0.66 (0.58–0.72, Variant II); 0.66 (0.56 to 0.76, Tina-quant); 0.58 (0.46 to 0.70, in2it) 6 days Chinese
1.38 + [0.57 × CKD status] 2 days/month for 3 consecutive months
This study
Lo et al. [5]
Not specified
234 subjects with type 1 diabetes, aged b19 years 43 subjects with stages 3–5 chronic kidney disease and type 2 diabetes, aged 18-80 years 45 subjects with (n = 36) and without (n = 9) type 2 diabetes, aged 21–80 years O'Riordan et al. [4]
Not specified
Not specified +2.22 1.00 (0.78 to 1.22)
Subjects encouraged to performed CGM as often as possible for 6 months 5 days 48 subjects with type 1 diabetes, aged 93% non-Hispanic 4–b18 years whites, 7% non-whites DirecNet study [3]
JDRF-CGM study [2]
0.66 (CKD 3–5); 0.79 (CKD 3); 0.34 (CKD 4–5) 0.84 (Variant II); 0.82 (Tina-quant); 0.71 (in2it)
Medtronic MiniMed, Abbott FreeStyle Navigator Abbott FreeStyle Navigator Tosoh HbA1c 2.2 Plus 0.63 −0.90 (−1.83 to 0.03) 1.35 (1.22 to 1.48) 3 months 94% whites, 6% non-whites
At least 2 days performed 1.59 4 times in 3 consecutive months 83% white, 8% black, 8% Hispanic, 2% others
507 subjects that comprised of 268 patients with type 1 diabetes, 159 patients with diabetes and 80 subjects, aged 18–70 years 252 subjects with type 1 diabetes, aged 8–74 years ADAG study [1]
0.49 (0.35 to 0.62)
Medtronic MiniMed Mean of 4 assays were used. Roche Tina-quant, Primus Ultra-2, Tosoh G7, Roche A1c 0.84 −2.59
Siemens DCA 2000+
CGM HbA1c assay Linear correlation, r2 Intercept (95% CI) Slope (95% CI) Duration of CGM Race Population Reference
Table 1 A selection of previously published studies that examined the relationship between HbA1c and average glucose. The regressions are expressed as average glucose (mmol/L) = slope × HbA1c (% unit) + intercept. CGM = continuous glucose monitor, CKD = chronic kidney disease.
T.P. Loh et al. / Clinical Biochemistry 48 (2015) 514–518
515
lacking in the literature. To fill this gap, we compared HbA1c measured using three National Glycohemoglobin Standardization Program (NGSP)-certified methods that employed different analytical principles, and estimated the discrepancy between measured average glucose (mAG) and eAG derived from the different HbA1c measurements. The relationship between HbA1c measured by the three routine laboratory methods and mAG was also examined.
Materials and methods This study was performed as part of a larger study examining glycemic and non-glycemic determinants of HbA1c in a Chinese population in Singapore. The inclusion criteria were as follows: age 21–80 years, Chinese ethnicity, and a screening HbA1c between 4.8 and 10%. Previous studies have shown that the relationship between average glucose and HbA1c is different between types 1 and 2 diabetes [7,8]. In this study, we have chosen to include examine patients with type 2 diabetes as type 1 diabetes is uncommon in our population [9]. Patients with known factors that may alter the relationship between average glucose and HbA1c including pregnancy, use of erythropoietin, treatment with hemodialysis, history of recent surgery, blood donation or transfusion in the past six months, were excluded. To ensure a stable metabolic state, patients with a history of recent hospitalization or steroid use in the past six months as well as patients who had a weight change of greater than 3% in the month preceding recruitment were also excluded. Forty-six consecutive subjects of the larger study were enrolled into this sub-study. Participants with diabetes were on a stable dose of diabetic medications for the last three months, and had stable glycemic control as evidenced by two sequential HbA1c values in the last six months, which were within 1% of each other. The study was approved by the ethics committee of National University Hospital, Singapore. Informed consent was obtained from all participants. Data on demographics and medical history was obtained through an interviewer-administered questionnaire. Height and weight were measured. All participants underwent six days of CGM (Enlite sensor, Medtronic, Northridge, USA), in which interstitial glucose concentrations were measured every 5 min. This CGM uses the glucose oxidase method to convert interstitial glucose into hydrogen peroxide that reacts with the sensor surface to generate electrical signal that is proportional to the changes in interstitial glucose. The participants performed capillary blood glucose monitoring during CGM for calibration purposes (One Touch Ultra, glucose oxidase biosensor method, Lifescan, Milipitas, USA). Capillary blood glucose readings were reported in plasma equivalent units. The participants who had less than 72 h of CGM readings were excluded from analysis. CGM data with a mean absolute difference compared to capillary blood glucose of 18% or greater was also excluded, as recommended by the manufacturer. The measured average glucose (mAG) was the arithmetic mean of all interstitial glucose values obtained from the CGM. Whole blood was collected into K2 EDTA blood tubes for HbA1c measurement and hemoglobin capillary zone electrophoresis (Sebia Capillarys, Norcross, GA) examination at the end of the CGM period, immediately after removal of the sensor from the patient. HbA1c was measured in an NGSP level I-certified clinical laboratory, using the Variant II Dual Program in HbA1c mode (Bio-Rad Laboratories, Hercules, CA), within 4 h of sample collection. The sample was then immediately stored at −70 °C. The archived samples were gradually thawed in iceslurry in a single cycle and kept chilled until measurement by the in2it (II) A1c (Bio-Rad Laboratories, Hercules, CA) and Tina-quant HbA1c Gen 3 (applied on c501 analyzer, Roche Diagnostics, Mannheim, Germany) methods within six months of sample collection. HbA1c concentrations in whole blood samples are stable for at least twelve months when stored at −70 °C. The correlation coefficient (r) is N0.99 between fresh and freeze–thaw samples [10,11].
516
T.P. Loh et al. / Clinical Biochemistry 48 (2015) 514–518
The Variant II method utilizes the cation-exchange principle to separate the hemoglobin species. The total analytical coefficients of variation (CVa) of this analyzer at HbA1c values of 5% and 10% are less than 2%. The Tina-quant method is a latex-enhanced competitive turbidimetric immunoassay and has CVa of 2.1% at similar HbA1c concentrations. The in2it method uses the boronate affinity chromatography principle to measure total glycated hemoglobin, which is then converted to HbA1c values. Its CVa is approximately 4% at HbA1c of 6.0%–10.4% [12]. All the HbA1c methods are NGSP-certified. Hemoglobin concentration was measured using the XE-5000 analyzer (Sysmex Corp, Kobe, Japan). All HbA1c measurements were performed according to manufacturer's recommendation under routine laboratory conditions. Statistical analysis The eAG were derived from the HbA1c results using the ADAG equation [eAG (expressed in mmol/L) = 1.59 × HbA1c (expressed in % unit) − 2.59] [1]. Bland–Altman plots were drawn to compare the difference between the eAG calculated from the results of the three HbA1c methods and mAG. Simple linear regression was performed with mAG as the dependent variable and HbA1c as the explanatory variable. The statistical analyses were performed using Microsoft Office Excel version 2010 (Microsoft Corporation, Redmond, WA) and SPSS version 19 (SPSS IBM Inc., Chicago, IL, USA). Data are shown as means (SD), and a p-value of b 0.05 was considered statistically significant. Results A total of 46 subjects were included in this study. One subject was regarded as an outlier by visual inspection of the scatter plot of mAG and HbA1c, and was excluded from further analysis. The mean age was 56.8 years (SD: 9.8) and mean BMI was 25.0 kg/m2 (4.3). Twentyseven of these subjects (60%) were male. Thirty-six subjects (80%) had type 2 diabetes. None of the subjects had anemia or hemoglobinopathy when examined by hemoglobin electrophoresis. The results of the laboratory tests are summarized in Table 2. The eAG derived from the HbA1c measured by the three routine laboratory methods were higher than mAG (Table 2). These differences were reflected in the Bland–Altman plots (Fig. 1), which showed the percentage difference between the eAG and mAG (y-axis) as a function of the average of eAG and mAG (x-axis). There was no proportional bias among all three HbA1c methods. The systematic bias, expressed as average percentage differences between eAG and mAG, was 10.4% (95% confidence interval: 7.7 to 13.2), 1.0% (− 1.7 to 3.6) and 6.0% (4.3 to 7.7) for Variant II, in2it and Tina-quant methods respectively. The relationships between the mAG and HbA1c measured by the three laboratory methods are shown in Fig. 2 together with the regression and correlation coefficients. While numerically different, the slope
Table 2 Summary of laboratory results in this study. SD, standard deviation; HbA1c, glycated hemoglobin A1c; eAG, estimated average glucose; CGM, continuous glucose monitoring. Laboratory tests, unit
Mean (SD)
Reference interval
Hemoglobin, g/dL
13.0 (0.8) (female) 14.6 (1.3) (male) 88.9 (4.0) 30.1 (1.6) 7.0 (1.1) 6.8 (1.1) 7.1 (1.1) 8.5 (1.7) 8.2 (1.7) 8.6 (1.8) 7.7 (1.5) 1701 (57)
10.9–15.1 (female) 12.9–17.0 (male) 80.0–95.0 27.0–33.0 4.5–5.6 4.5–5.6 4.5–5.6 – – – – –
Mean corpuscular volume, fL Mean corpuscular hemoglobin, pg HbA1c, Variant II, % HbA1c, in2it, % HbA1c, Tina-quant, % eAG, Variant II, % eAG, in2it, mmol/L eAG, Tina-quant, mmol/L Measured average glucose, mmol/L Number of CGM readings per subject
and intercept of the three laboratory methods have overlapping standard errors. The correlation between mAG and HbA1c was greatest when measured using the Variant II (r2 = 0.84), followed by Tina-quant (r2 = 0.82) and in2it (r2 = 0.71). Discussion The reporting of average glucose is thought to be more intuitive for a patient to understand his glycemic control. The eAG equation derived from the ADAG study provides a convenient tool for converting HbA1c results to estimated average glucose. In certain regions, clinical laboratories are encouraged to report eAG together with HbA1c to help patients better understand and manage their glycemic control [13,14]. Nevertheless, there are several limitations of the ADAG equation that have been well argued [15]. Firstly, the eAG assumes that relationship between HbA1c and average glucose is mathematically consistent between subjects. However, the extent to which hemoglobin is glycated at a given concentration of glucose also depends on factors such as red blood cell lifespan and turnover, rate of glucose uptake into the blood cells, rate of glycation and deglycation. These intricate factors are different between individuals and contribute to discordant eAG and selfmonitored mean blood glucose seen in routine practice [16]. Secondly, subjects included in the ADAG study were not sufficiently representative to allow reliable expression of HbA1c as eAG in all patients. The subjects included in the ADAG study were mostly nonpregnant white adults without renal failure or hemoglobin variants [1]. The narrow inclusion criteria meant that this equation should not be used beyond these populations without exercising caution or additional studies (Table 1). There are racial and ethnic differences in the relationship between HbA1c and blood glucose [17]. It would be of interest to evaluate the relationship between mean glucose and HbA1c in other ethnic populations. Finally, the wide scatter of mAG around the regression line meant that a high proportion of patients are likely to have significantly different true average glucose when compared to the eAG. By some estimates, up to 100% difference may be seen in some patients when the analytical bias in HbA1c assay is considered [15]. In this study, the application of the ADAG-derived eAG equation was associated with an over-estimation of average glucose compared to mAG. The degree of over-estimation of eAG varied with the laboratory method used to measure HbA1c. The over-estimation was greatest for Variant II (10%), followed by Tina-quant (6%) and in2it (1%). The overestimation of average glucose implied that for the same degree of glycemia, the HbA1c is higher in our cohort compared to those in the ADAG study. One possible explanation for this may be the ethnicity of our study population. Asian and black populations are known to have higher HbA1c values compared with Caucasians at the same level of fasting glycemia [17]. The subjects included in the ADAG study are predominantly Caucasians and people of African descent, whereas our cohort is entirely Chinese. The eAG equations (regression coefficients shown in Fig. 2) derived from this study for each method were different from the ADAG study, which could also explain the reason for the over-estimation of average glucose seen in our study subjects. This may be related to betweenmethod biases in the HbA1c instruments, which remain present despite the manufacturer receiving NGSP certification (http://www.ngsp.org/ CAPdata.asp). The NGSP has significantly reduced the bias and imprecision of most routine laboratory methods since its inception [18] and continued tightening of the NGSP performance criteria may reduce these differences. Of note, the ADAG equation has a residual HbA1c of 1.63% when eAG is 0 mmol/L. This suggests that the equation includes some non-glycemic factors. A recent in-silico semi-mechanistic modeling study has found that red blood cell lifespan and non-specificity of the routine (NGSP) HbA1c assays were able to explain the non-
T.P. Loh et al. / Clinical Biochemistry 48 (2015) 514–518
517
Fig. 1. The Bland–Altman plots showing the difference between measured average glucose and estimated average glucose (eAG) derived from (panel A) in2it, (panel B) Tina-quant and (panel C) Variant II. The average percentage difference (solid line), its 95% confidence intervals (dotted lines) and 1.96 standard deviations (dashed lines) are also displayed.
proportional relationship between HbA1c and average glucose [19]. Alternatively, the discordance may be related to the difference in the calibration characteristics, and hence bias, of the CGM and glucometers used between the studies. The correlation between the mAG and HbA1c in this study was good, with r2 values ranging from 0.71 to 0.84. The degree of correlation is inversely related to the CVa of the HbA1c laboratory method. The high CVa of the in2it instrument may increase the variability of each HbA1c measurement, leading to poorer correlation between in2it and mAG. The
Fig. 2. The relationships between the measured average glucose and HbA1c measured by the three routine laboratory methods. B = slope, SE = standard error, C = intercept, Adj r2 = adjusted r2.
correlation coefficients in this study are comparable to the ADAG study, where an r2 value of 0.84 was achieved [1]. The high degree of correlation was achieved in this study despite the relatively short period of intensive glucose monitoring of six days compared to three months in the ADAG study. This may be attributable to the use of precise laboratory methods (Variant II and Tina-quant), the inclusion of patients with stable glycemia (b 1% unit change in HbA1c in the preceding two months before enrollment), and the high number of CGM readings achieved in this study (Table 2). One potential limitation of this study is the use of samples that have been stored at − 70 °C and subjected to a single freeze–thaw cycle. Under this condition, a high degree of correlation (r2 = 0.82) was obtained for the Tina-quant immunoassay method, while the in2it boronate affinity method has poorer correlation (r2 = 0.71). Besides the difference in CVa between the methods, the difference may be caused by degradation of HbA1c molecule during the freeze–thaw cycle in a manner that affects certain method more than others. We have sought to minimize any potential degradation by slowly thawing the samples in ice slurry and keeping the samples chilled until testing. Of note, the ADAG study also subjected their samples to a single freeze–thaw cycle prior to HbA1c measurement [1]. A direct comparison between the different methods on freshly collected samples would address this limitation. In summary, the relationship between HbA1c and measured average glucose is method-dependent despite improved standardization of HbA1c. The differences in relationship may reflect as discrepant eAG and readings on glucometers. Patients who routinely perform selfmonitoring of blood glucose may observe a significant discordance of their results with the eAG reported by the laboratory using the ADAG equation, which may confuse as well as undermine their confidence in the ability of the laboratory to produce accurate results. Laboratory and clinical practitioners should thus be aware of such differences when considering the use of the ADAG equation in their reports.
518
T.P. Loh et al. / Clinical Biochemistry 48 (2015) 514–518
Method-specific eAG equations may reconcile some of these differences but require further study. Conflict of interest The authors have no competing interest or disclosure to declare. Acknowledgment We would like to thank the staff of the Department of Laboratory Medicine, National University Hospital and Khoo Teck Puat Hospital for the kind technical assistance. We also gratefully acknowledge the funding from the National Medical Research Council (Singapore) NMRC//NIG/1050/2011, awarded to Dr Shih Ling Kao. References [1] Nathan DM, Kuenen J, Borg R, Zheng H, Schoenfeld D, Heine RJ, et al. Translating the A1C assay into estimated average glucose values. Diabetes Care 2008;31(8):1473–8. [2] Juvenile Diabetes Research Foundation Continuous Glucose Monitoring Study Group. Hemoglobin A1c and mean glucose in patients with type 1 diabetes: analysis of data from the Juvenile Diabetes Research Foundation continuous glucose monitoring randomized trial. Diabetes Care 2011;34(3):540–4. [3] Diabetes Research in Children Network (DirecNet) Study Group, Wilson DM, Kollman. Relationship of A1C to glucose concentrations in children with type 1 diabetes: assessments by high-frequency glucose determinations by sensors. Diabetes Care 2008;31(3):381–5. [4] O'Riordan SM, Danne T, Hanas R, Peters CJ, Hindmarsh P. Paediatric estimated average glucose in children with type 1 diabetes. Diabet Med 2014;31(1):36–9. [5] Lo C, Lui M, Ranasinha S, Teede HJ, Kerr PG, Polkinghorne KR, et al. Defining the relationship between average glucose and HbA1c in patients with type 2 diabetes and chronic kidney disease. Diabetes Res Clin Pract 2014;104(1):84–91.
[6] Lenters-Westra E, Slingerland RJ. Hemoglobin A1c determination in the A1CDerived Average Glucose (ADAG) study. Clin Chem Lab Med 2008;46(11):1617–23. [7] Borg R, Kuenen JC, Carstensen B, Zheng H, Nathan DM, Heine RJ, et al. Associations between features of glucose exposure and A1C: the A1C-Derived Average Glucose (ADAG) study. Diabetes 2010;59(7):1585–90. [8] Kuenen JC, Borg R, Kuik DJ, Zheng H, Schoenfeld D, Diamant M, et al. Does glucose variability influence the relationship between mean plasma glucose and HbA1c levels in type 1 and type 2 diabetic patients? Diabetes Care 2011;34(8):1843–7. [9] Todd AL, Ng WY, Lui KF, Thai AC. Low prevalence of autoimmune diabetes markers in a mixed ethnic population of Singaporean diabetics. Intern Med J 2004;34(1–2): 24–30. [10] Mercadanti M, Caleffi A, Targher G, Lippi G. Sample stability for the measurement of hemoglobin A1c. Diabetes Res Clin Pract 2010;90(1):e13–4. [11] Liotta L, Di Franco A, Pazzagli M, Luconi M. Glycated hemoglobin (HbA1c) measurement in frozen whole blood depends on baseline values of fresh samples. Anal Bioanal Chem 2013;405(1):429–34. [12] Yeo CP, Tan CH, Jacob E. Haemoglobin A1c: evaluation of a new HbA1c point-of-care analyser Bio-Rad in2it in comparison with the DCA 2000 and central laboratory analysers. Ann Clin Biochem 2009;46(Pt 5):373–6. [13] Kahn R, Fonseca V. Translating the A1C assay. Diabetes Care 2008;31(8):1704–7. [14] American Association for Clinical Chemistry. AACC position on reporting and use of estimated average glucose in diabetes. http://www.aacc.org/gov/gov_affairs/ positions/pos_stat_09/pages/default.aspx#; May 09, 2014. [Last accessed]. [15] Young IS. Counterpoint: the reporting of estimated glucose with hemoglobin A1c. Clin Chem 2010;56(4):547–9. [16] Hempe JM, Soros AA, Chalew SA. Estimated average glucose and self-monitored mean blood glucose are discordant estimates of glycemic control. Diabetes Care 2010;33(7):1449–51. [17] Herman WH, Cohen RM. Racial and ethnic differences in the relationship between HbA1c and blood glucose: implications for the diagnosis of diabetes. J Clin Endocrinol Metab 2012;97(4):1067–72. [18] Little RR, Rohlfing CL. The long and winding road to optimal HbA1c measurement. Clin Chim Acta 2013;418:63–71. [19] Lledó-García R, Mazer NA, Karlsson MO. A semi-mechanistic model of the relationship between average glucose and HbA1c in healthy and diabetic subjects. J Pharmacokinet Pharmacodyn 2013;40(2):129–42.
Contents lists available at ScienceDirect
Clinical Biochemistry journal homepage: www.elsevier.com/locate/clinbiochem
Relationship between measured average glucose by continuous glucose monitor and HbA1c measured by three different routine laboratory methods Tze Ping Loh a,⁎, Sunil Kumar Sethi a, Moh Sim Wong b, E. Shyong Tai c, Shih Ling Kao c a b c
Department of Laboratory Medicine, National University Hospital, Singapore Department of Laboratory Medicine, Khoo Teck Puat Hospital, Singapore Department of Medicine, National University Hospital, Singapore
a r t i c l e
i n f o
Article history: Received 26 December 2014 Received in revised form 3 February 2015 Accepted 23 February 2015 Available online 1 March 2015 Keywords: Glycated hemoglobin A1c Glucose Continuous glucose monitoring Correlation Regression
a b s t r a c t Objectives: The relationship between glycated hemoglobin A1c (HbA1c) and average glucose has been described by the empirically derived estimated average glucose (eAG) equation in the A1c-Derived Average Glucose (ADAG) study, with extensive calibration efforts in four secondary reference HbA1c methods. It is not known if this relationship is preserved when HbA1c is measured by routine laboratory methods under routine conditions. Design and methods: We measured average glucose (mAG) by six days of continuous glucose monitoring in 45 adults with stable HbA1c (b1% HbA1c change in the preceding two months). Subjects with medical conditions that may confound HbA1c measurement, including anemia and hemoglobinopathy, were excluded. HbA1c was measured using Bio-Rad Variant II (cation-exchange HPLC), Bio-Rad in2it (boronate affinity HPLC) and Roche Tina-quant (immunoassay) methods. Results: The average differences between eAG derived from the routine HbA1c methods and mAG were 10.4% (Variant II), 6.0% (Tina-quant) and 1.0% (in2it). The regression coefficients between the mAG and HbA1c are different between in2it (mAG, mmol/L = 0.58 × %HbA1c + 2.3), Tina-quant and Variant II (both mAG, mmol/L = 0.66 × %HbA1c + 1.9). However, the 95% confidence intervals of the slope and bias of these methods overlap. The correlation between mAG and HbA1c was greatest when measured using the Variant II (r2 = 0.84), followed by Tina-quant (r2 = 0.82) and in2it (r2 = 0.71). Conclusions: The relationship between HbA1c and measured average glucose is method-dependent despite improved HbA1c standardization. The differences in relationship may reflect as discrepant eAG and home glucose monitoring results. © 2015 The Canadian Society of Clinical Chemists. Published by Elsevier Inc. All rights reserved.
Introduction Glycated hemoglobin A1c (HbA1c) is formed from non-enzymatic glycation of hemoglobin. HbA1c concentrations are highly correlated with average glycemia, and reflect the glycemic control of a patient over the past two to three months. The relationship between average glucose and HbA1c has been described by the empirically derived estimated average glucose (eAG) equation in the seminal A1c-Derived Average Glucose (ADAG) study [1]. Several other groups have also examined the relationship between average glucose by continuous glucose monitoring (CGM) and HbA1c Abbreviations: HbA1c, glycated hemoglobin A1c; NGSP, National Glycemic Standardization Program; HPLC, high performance liquid chromatography; eAG, estimated average glucose; mAG, measured average glucose; CGM, continuous glucose monitoring; CVa, analytical coefficients of variation. ⁎ Corresponding author at: National University Hospital, 5 Lower Kent Ridge Road, 119074 Singapore, Singapore. Fax: +65 67771613. E-mail address: [email protected] (T.P. Loh).
in other populations. The linear relationship between average glucose and HbA1c has been demonstrated in children and adult with type 1 diabetes [2–4], and patients with concomitant advanced chronic kidney disease and type 2 diabetes [5], albeit with different correlation and regression characteristics. A selection of these publications is summarized in Table 1. The ADAG equation is the most widely used eAG equation in routine clinical practice to translate HbA1c values into average glucose. The eAG equation in the ADAG study was derived by correlating the mean of four HbA1c values, which were measured using four different International Federation of Clinical Chemistry and Laboratory Medicine (IFCC) secondary reference laboratory methods with rigorous calibration efforts [6], and with average glucose obtained from capillary blood and CGM. Such rigorous effort in HbA1c measurement is unlikely to be achieved in routine practice. Therefore, it is important to understand the impact of using HbA1c measured by routine laboratory methods under routine laboratory conditions on the relationship with average glucose. Such evidence is
http://dx.doi.org/10.1016/j.clinbiochem.2015.02.012 0009-9120/© 2015 The Canadian Society of Clinical Chemists. Published by Elsevier Inc. All rights reserved.
Medtronic Minimed Gold Medtronic Minimed
Medtronic Enlite sensor
Siemens DCA 2000+
Tosoh A1c 2.2 and Adams Arkray
Bio-rad Variant II, Roche Tina-quant, Bio-rad in2it
+5.95 (4.72 to 7.19) 0.20
−1.16
1.9 (1.2 to 2.5, Variant II); 1.9 (1.1 to 2.7, Tina-quant); 2.3 (1.5 to 3.1 in2it) 0.66 (0.58–0.72, Variant II); 0.66 (0.56 to 0.76, Tina-quant); 0.58 (0.46 to 0.70, in2it) 6 days Chinese
1.38 + [0.57 × CKD status] 2 days/month for 3 consecutive months
This study
Lo et al. [5]
Not specified
234 subjects with type 1 diabetes, aged b19 years 43 subjects with stages 3–5 chronic kidney disease and type 2 diabetes, aged 18-80 years 45 subjects with (n = 36) and without (n = 9) type 2 diabetes, aged 21–80 years O'Riordan et al. [4]
Not specified
Not specified +2.22 1.00 (0.78 to 1.22)
Subjects encouraged to performed CGM as often as possible for 6 months 5 days 48 subjects with type 1 diabetes, aged 93% non-Hispanic 4–b18 years whites, 7% non-whites DirecNet study [3]
JDRF-CGM study [2]
0.66 (CKD 3–5); 0.79 (CKD 3); 0.34 (CKD 4–5) 0.84 (Variant II); 0.82 (Tina-quant); 0.71 (in2it)
Medtronic MiniMed, Abbott FreeStyle Navigator Abbott FreeStyle Navigator Tosoh HbA1c 2.2 Plus 0.63 −0.90 (−1.83 to 0.03) 1.35 (1.22 to 1.48) 3 months 94% whites, 6% non-whites
At least 2 days performed 1.59 4 times in 3 consecutive months 83% white, 8% black, 8% Hispanic, 2% others
507 subjects that comprised of 268 patients with type 1 diabetes, 159 patients with diabetes and 80 subjects, aged 18–70 years 252 subjects with type 1 diabetes, aged 8–74 years ADAG study [1]
0.49 (0.35 to 0.62)
Medtronic MiniMed Mean of 4 assays were used. Roche Tina-quant, Primus Ultra-2, Tosoh G7, Roche A1c 0.84 −2.59
Siemens DCA 2000+
CGM HbA1c assay Linear correlation, r2 Intercept (95% CI) Slope (95% CI) Duration of CGM Race Population Reference
Table 1 A selection of previously published studies that examined the relationship between HbA1c and average glucose. The regressions are expressed as average glucose (mmol/L) = slope × HbA1c (% unit) + intercept. CGM = continuous glucose monitor, CKD = chronic kidney disease.
T.P. Loh et al. / Clinical Biochemistry 48 (2015) 514–518
515
lacking in the literature. To fill this gap, we compared HbA1c measured using three National Glycohemoglobin Standardization Program (NGSP)-certified methods that employed different analytical principles, and estimated the discrepancy between measured average glucose (mAG) and eAG derived from the different HbA1c measurements. The relationship between HbA1c measured by the three routine laboratory methods and mAG was also examined.
Materials and methods This study was performed as part of a larger study examining glycemic and non-glycemic determinants of HbA1c in a Chinese population in Singapore. The inclusion criteria were as follows: age 21–80 years, Chinese ethnicity, and a screening HbA1c between 4.8 and 10%. Previous studies have shown that the relationship between average glucose and HbA1c is different between types 1 and 2 diabetes [7,8]. In this study, we have chosen to include examine patients with type 2 diabetes as type 1 diabetes is uncommon in our population [9]. Patients with known factors that may alter the relationship between average glucose and HbA1c including pregnancy, use of erythropoietin, treatment with hemodialysis, history of recent surgery, blood donation or transfusion in the past six months, were excluded. To ensure a stable metabolic state, patients with a history of recent hospitalization or steroid use in the past six months as well as patients who had a weight change of greater than 3% in the month preceding recruitment were also excluded. Forty-six consecutive subjects of the larger study were enrolled into this sub-study. Participants with diabetes were on a stable dose of diabetic medications for the last three months, and had stable glycemic control as evidenced by two sequential HbA1c values in the last six months, which were within 1% of each other. The study was approved by the ethics committee of National University Hospital, Singapore. Informed consent was obtained from all participants. Data on demographics and medical history was obtained through an interviewer-administered questionnaire. Height and weight were measured. All participants underwent six days of CGM (Enlite sensor, Medtronic, Northridge, USA), in which interstitial glucose concentrations were measured every 5 min. This CGM uses the glucose oxidase method to convert interstitial glucose into hydrogen peroxide that reacts with the sensor surface to generate electrical signal that is proportional to the changes in interstitial glucose. The participants performed capillary blood glucose monitoring during CGM for calibration purposes (One Touch Ultra, glucose oxidase biosensor method, Lifescan, Milipitas, USA). Capillary blood glucose readings were reported in plasma equivalent units. The participants who had less than 72 h of CGM readings were excluded from analysis. CGM data with a mean absolute difference compared to capillary blood glucose of 18% or greater was also excluded, as recommended by the manufacturer. The measured average glucose (mAG) was the arithmetic mean of all interstitial glucose values obtained from the CGM. Whole blood was collected into K2 EDTA blood tubes for HbA1c measurement and hemoglobin capillary zone electrophoresis (Sebia Capillarys, Norcross, GA) examination at the end of the CGM period, immediately after removal of the sensor from the patient. HbA1c was measured in an NGSP level I-certified clinical laboratory, using the Variant II Dual Program in HbA1c mode (Bio-Rad Laboratories, Hercules, CA), within 4 h of sample collection. The sample was then immediately stored at −70 °C. The archived samples were gradually thawed in iceslurry in a single cycle and kept chilled until measurement by the in2it (II) A1c (Bio-Rad Laboratories, Hercules, CA) and Tina-quant HbA1c Gen 3 (applied on c501 analyzer, Roche Diagnostics, Mannheim, Germany) methods within six months of sample collection. HbA1c concentrations in whole blood samples are stable for at least twelve months when stored at −70 °C. The correlation coefficient (r) is N0.99 between fresh and freeze–thaw samples [10,11].
516
T.P. Loh et al. / Clinical Biochemistry 48 (2015) 514–518
The Variant II method utilizes the cation-exchange principle to separate the hemoglobin species. The total analytical coefficients of variation (CVa) of this analyzer at HbA1c values of 5% and 10% are less than 2%. The Tina-quant method is a latex-enhanced competitive turbidimetric immunoassay and has CVa of 2.1% at similar HbA1c concentrations. The in2it method uses the boronate affinity chromatography principle to measure total glycated hemoglobin, which is then converted to HbA1c values. Its CVa is approximately 4% at HbA1c of 6.0%–10.4% [12]. All the HbA1c methods are NGSP-certified. Hemoglobin concentration was measured using the XE-5000 analyzer (Sysmex Corp, Kobe, Japan). All HbA1c measurements were performed according to manufacturer's recommendation under routine laboratory conditions. Statistical analysis The eAG were derived from the HbA1c results using the ADAG equation [eAG (expressed in mmol/L) = 1.59 × HbA1c (expressed in % unit) − 2.59] [1]. Bland–Altman plots were drawn to compare the difference between the eAG calculated from the results of the three HbA1c methods and mAG. Simple linear regression was performed with mAG as the dependent variable and HbA1c as the explanatory variable. The statistical analyses were performed using Microsoft Office Excel version 2010 (Microsoft Corporation, Redmond, WA) and SPSS version 19 (SPSS IBM Inc., Chicago, IL, USA). Data are shown as means (SD), and a p-value of b 0.05 was considered statistically significant. Results A total of 46 subjects were included in this study. One subject was regarded as an outlier by visual inspection of the scatter plot of mAG and HbA1c, and was excluded from further analysis. The mean age was 56.8 years (SD: 9.8) and mean BMI was 25.0 kg/m2 (4.3). Twentyseven of these subjects (60%) were male. Thirty-six subjects (80%) had type 2 diabetes. None of the subjects had anemia or hemoglobinopathy when examined by hemoglobin electrophoresis. The results of the laboratory tests are summarized in Table 2. The eAG derived from the HbA1c measured by the three routine laboratory methods were higher than mAG (Table 2). These differences were reflected in the Bland–Altman plots (Fig. 1), which showed the percentage difference between the eAG and mAG (y-axis) as a function of the average of eAG and mAG (x-axis). There was no proportional bias among all three HbA1c methods. The systematic bias, expressed as average percentage differences between eAG and mAG, was 10.4% (95% confidence interval: 7.7 to 13.2), 1.0% (− 1.7 to 3.6) and 6.0% (4.3 to 7.7) for Variant II, in2it and Tina-quant methods respectively. The relationships between the mAG and HbA1c measured by the three laboratory methods are shown in Fig. 2 together with the regression and correlation coefficients. While numerically different, the slope
Table 2 Summary of laboratory results in this study. SD, standard deviation; HbA1c, glycated hemoglobin A1c; eAG, estimated average glucose; CGM, continuous glucose monitoring. Laboratory tests, unit
Mean (SD)
Reference interval
Hemoglobin, g/dL
13.0 (0.8) (female) 14.6 (1.3) (male) 88.9 (4.0) 30.1 (1.6) 7.0 (1.1) 6.8 (1.1) 7.1 (1.1) 8.5 (1.7) 8.2 (1.7) 8.6 (1.8) 7.7 (1.5) 1701 (57)
10.9–15.1 (female) 12.9–17.0 (male) 80.0–95.0 27.0–33.0 4.5–5.6 4.5–5.6 4.5–5.6 – – – – –
Mean corpuscular volume, fL Mean corpuscular hemoglobin, pg HbA1c, Variant II, % HbA1c, in2it, % HbA1c, Tina-quant, % eAG, Variant II, % eAG, in2it, mmol/L eAG, Tina-quant, mmol/L Measured average glucose, mmol/L Number of CGM readings per subject
and intercept of the three laboratory methods have overlapping standard errors. The correlation between mAG and HbA1c was greatest when measured using the Variant II (r2 = 0.84), followed by Tina-quant (r2 = 0.82) and in2it (r2 = 0.71). Discussion The reporting of average glucose is thought to be more intuitive for a patient to understand his glycemic control. The eAG equation derived from the ADAG study provides a convenient tool for converting HbA1c results to estimated average glucose. In certain regions, clinical laboratories are encouraged to report eAG together with HbA1c to help patients better understand and manage their glycemic control [13,14]. Nevertheless, there are several limitations of the ADAG equation that have been well argued [15]. Firstly, the eAG assumes that relationship between HbA1c and average glucose is mathematically consistent between subjects. However, the extent to which hemoglobin is glycated at a given concentration of glucose also depends on factors such as red blood cell lifespan and turnover, rate of glucose uptake into the blood cells, rate of glycation and deglycation. These intricate factors are different between individuals and contribute to discordant eAG and selfmonitored mean blood glucose seen in routine practice [16]. Secondly, subjects included in the ADAG study were not sufficiently representative to allow reliable expression of HbA1c as eAG in all patients. The subjects included in the ADAG study were mostly nonpregnant white adults without renal failure or hemoglobin variants [1]. The narrow inclusion criteria meant that this equation should not be used beyond these populations without exercising caution or additional studies (Table 1). There are racial and ethnic differences in the relationship between HbA1c and blood glucose [17]. It would be of interest to evaluate the relationship between mean glucose and HbA1c in other ethnic populations. Finally, the wide scatter of mAG around the regression line meant that a high proportion of patients are likely to have significantly different true average glucose when compared to the eAG. By some estimates, up to 100% difference may be seen in some patients when the analytical bias in HbA1c assay is considered [15]. In this study, the application of the ADAG-derived eAG equation was associated with an over-estimation of average glucose compared to mAG. The degree of over-estimation of eAG varied with the laboratory method used to measure HbA1c. The over-estimation was greatest for Variant II (10%), followed by Tina-quant (6%) and in2it (1%). The overestimation of average glucose implied that for the same degree of glycemia, the HbA1c is higher in our cohort compared to those in the ADAG study. One possible explanation for this may be the ethnicity of our study population. Asian and black populations are known to have higher HbA1c values compared with Caucasians at the same level of fasting glycemia [17]. The subjects included in the ADAG study are predominantly Caucasians and people of African descent, whereas our cohort is entirely Chinese. The eAG equations (regression coefficients shown in Fig. 2) derived from this study for each method were different from the ADAG study, which could also explain the reason for the over-estimation of average glucose seen in our study subjects. This may be related to betweenmethod biases in the HbA1c instruments, which remain present despite the manufacturer receiving NGSP certification (http://www.ngsp.org/ CAPdata.asp). The NGSP has significantly reduced the bias and imprecision of most routine laboratory methods since its inception [18] and continued tightening of the NGSP performance criteria may reduce these differences. Of note, the ADAG equation has a residual HbA1c of 1.63% when eAG is 0 mmol/L. This suggests that the equation includes some non-glycemic factors. A recent in-silico semi-mechanistic modeling study has found that red blood cell lifespan and non-specificity of the routine (NGSP) HbA1c assays were able to explain the non-
T.P. Loh et al. / Clinical Biochemistry 48 (2015) 514–518
517
Fig. 1. The Bland–Altman plots showing the difference between measured average glucose and estimated average glucose (eAG) derived from (panel A) in2it, (panel B) Tina-quant and (panel C) Variant II. The average percentage difference (solid line), its 95% confidence intervals (dotted lines) and 1.96 standard deviations (dashed lines) are also displayed.
proportional relationship between HbA1c and average glucose [19]. Alternatively, the discordance may be related to the difference in the calibration characteristics, and hence bias, of the CGM and glucometers used between the studies. The correlation between the mAG and HbA1c in this study was good, with r2 values ranging from 0.71 to 0.84. The degree of correlation is inversely related to the CVa of the HbA1c laboratory method. The high CVa of the in2it instrument may increase the variability of each HbA1c measurement, leading to poorer correlation between in2it and mAG. The
Fig. 2. The relationships between the measured average glucose and HbA1c measured by the three routine laboratory methods. B = slope, SE = standard error, C = intercept, Adj r2 = adjusted r2.
correlation coefficients in this study are comparable to the ADAG study, where an r2 value of 0.84 was achieved [1]. The high degree of correlation was achieved in this study despite the relatively short period of intensive glucose monitoring of six days compared to three months in the ADAG study. This may be attributable to the use of precise laboratory methods (Variant II and Tina-quant), the inclusion of patients with stable glycemia (b 1% unit change in HbA1c in the preceding two months before enrollment), and the high number of CGM readings achieved in this study (Table 2). One potential limitation of this study is the use of samples that have been stored at − 70 °C and subjected to a single freeze–thaw cycle. Under this condition, a high degree of correlation (r2 = 0.82) was obtained for the Tina-quant immunoassay method, while the in2it boronate affinity method has poorer correlation (r2 = 0.71). Besides the difference in CVa between the methods, the difference may be caused by degradation of HbA1c molecule during the freeze–thaw cycle in a manner that affects certain method more than others. We have sought to minimize any potential degradation by slowly thawing the samples in ice slurry and keeping the samples chilled until testing. Of note, the ADAG study also subjected their samples to a single freeze–thaw cycle prior to HbA1c measurement [1]. A direct comparison between the different methods on freshly collected samples would address this limitation. In summary, the relationship between HbA1c and measured average glucose is method-dependent despite improved standardization of HbA1c. The differences in relationship may reflect as discrepant eAG and readings on glucometers. Patients who routinely perform selfmonitoring of blood glucose may observe a significant discordance of their results with the eAG reported by the laboratory using the ADAG equation, which may confuse as well as undermine their confidence in the ability of the laboratory to produce accurate results. Laboratory and clinical practitioners should thus be aware of such differences when considering the use of the ADAG equation in their reports.
518
T.P. Loh et al. / Clinical Biochemistry 48 (2015) 514–518
Method-specific eAG equations may reconcile some of these differences but require further study. Conflict of interest The authors have no competing interest or disclosure to declare. Acknowledgment We would like to thank the staff of the Department of Laboratory Medicine, National University Hospital and Khoo Teck Puat Hospital for the kind technical assistance. We also gratefully acknowledge the funding from the National Medical Research Council (Singapore) NMRC//NIG/1050/2011, awarded to Dr Shih Ling Kao. References [1] Nathan DM, Kuenen J, Borg R, Zheng H, Schoenfeld D, Heine RJ, et al. Translating the A1C assay into estimated average glucose values. Diabetes Care 2008;31(8):1473–8. [2] Juvenile Diabetes Research Foundation Continuous Glucose Monitoring Study Group. Hemoglobin A1c and mean glucose in patients with type 1 diabetes: analysis of data from the Juvenile Diabetes Research Foundation continuous glucose monitoring randomized trial. Diabetes Care 2011;34(3):540–4. [3] Diabetes Research in Children Network (DirecNet) Study Group, Wilson DM, Kollman. Relationship of A1C to glucose concentrations in children with type 1 diabetes: assessments by high-frequency glucose determinations by sensors. Diabetes Care 2008;31(3):381–5. [4] O'Riordan SM, Danne T, Hanas R, Peters CJ, Hindmarsh P. Paediatric estimated average glucose in children with type 1 diabetes. Diabet Med 2014;31(1):36–9. [5] Lo C, Lui M, Ranasinha S, Teede HJ, Kerr PG, Polkinghorne KR, et al. Defining the relationship between average glucose and HbA1c in patients with type 2 diabetes and chronic kidney disease. Diabetes Res Clin Pract 2014;104(1):84–91.
[6] Lenters-Westra E, Slingerland RJ. Hemoglobin A1c determination in the A1CDerived Average Glucose (ADAG) study. Clin Chem Lab Med 2008;46(11):1617–23. [7] Borg R, Kuenen JC, Carstensen B, Zheng H, Nathan DM, Heine RJ, et al. Associations between features of glucose exposure and A1C: the A1C-Derived Average Glucose (ADAG) study. Diabetes 2010;59(7):1585–90. [8] Kuenen JC, Borg R, Kuik DJ, Zheng H, Schoenfeld D, Diamant M, et al. Does glucose variability influence the relationship between mean plasma glucose and HbA1c levels in type 1 and type 2 diabetic patients? Diabetes Care 2011;34(8):1843–7. [9] Todd AL, Ng WY, Lui KF, Thai AC. Low prevalence of autoimmune diabetes markers in a mixed ethnic population of Singaporean diabetics. Intern Med J 2004;34(1–2): 24–30. [10] Mercadanti M, Caleffi A, Targher G, Lippi G. Sample stability for the measurement of hemoglobin A1c. Diabetes Res Clin Pract 2010;90(1):e13–4. [11] Liotta L, Di Franco A, Pazzagli M, Luconi M. Glycated hemoglobin (HbA1c) measurement in frozen whole blood depends on baseline values of fresh samples. Anal Bioanal Chem 2013;405(1):429–34. [12] Yeo CP, Tan CH, Jacob E. Haemoglobin A1c: evaluation of a new HbA1c point-of-care analyser Bio-Rad in2it in comparison with the DCA 2000 and central laboratory analysers. Ann Clin Biochem 2009;46(Pt 5):373–6. [13] Kahn R, Fonseca V. Translating the A1C assay. Diabetes Care 2008;31(8):1704–7. [14] American Association for Clinical Chemistry. AACC position on reporting and use of estimated average glucose in diabetes. http://www.aacc.org/gov/gov_affairs/ positions/pos_stat_09/pages/default.aspx#; May 09, 2014. [Last accessed]. [15] Young IS. Counterpoint: the reporting of estimated glucose with hemoglobin A1c. Clin Chem 2010;56(4):547–9. [16] Hempe JM, Soros AA, Chalew SA. Estimated average glucose and self-monitored mean blood glucose are discordant estimates of glycemic control. Diabetes Care 2010;33(7):1449–51. [17] Herman WH, Cohen RM. Racial and ethnic differences in the relationship between HbA1c and blood glucose: implications for the diagnosis of diabetes. J Clin Endocrinol Metab 2012;97(4):1067–72. [18] Little RR, Rohlfing CL. The long and winding road to optimal HbA1c measurement. Clin Chim Acta 2013;418:63–71. [19] Lledó-García R, Mazer NA, Karlsson MO. A semi-mechanistic model of the relationship between average glucose and HbA1c in healthy and diabetic subjects. J Pharmacokinet Pharmacodyn 2013;40(2):129–42.
