Clinical Chemistry 50:1

166 –174 (2004) Endocrinology and

Metabolism

IFCC Reference System for Measurement of

Hemoglobin A1c in Human Blood and the
National Standardization Schemes in the
United States, Japan, and Sweden:
A Method-Comparison Study
Wieland Hoelzel,1 Cas Weykamp,2 Jan-Olof Jeppsson,3 Kor Miedema,4* John R. Barr,5
Ian Goodall,6 Tadao Hoshino,7 W. Garry John,8 Uwe Kobold,1 Randie Little,9
Andrea Mosca,10 Pierluigi Mauri,11 Rita Paroni,12 Fransiscus Susanto,13
Izumu Takei,14 Linda Thienpont,15 Masao Umemoto,16 and Hsiao-Mei Wiedmeyer,9 on
behalf of the IFCC Working Group on HbA1c Standardization

Background: The national programs for the harmoni- (DCMs). The future basis for international standardiza-
zation of hemoglobin (Hb)A1c measurements in the US tion will be the reference system developed by the IFCC
[National Glycohemoglobin Standardization Program Working Group on HbA1c Standardization. The aim of
(NGSP)], Japan [Japanese Diabetes Society (JDS)/Japa- the present study was to determine the relationships
nese Society of Clinical Chemistry (JSCC)], and Sweden between the IFCC Reference Method (RM) and the DCMs.
are based on different designated comparison methods Methods: Four method-comparison studies were per-
formed in 2001–2003. In each study five to eight pooled
blood samples were measured by 11 reference laborato-
1
ries of the IFCC Network of Reference Laboratories, 9
Roche Diagnostics GmbH, Penzberg, Germany.
2
SKZL Queen Beatrix Hospital, Winterswijk, The Netherlands.
Secondary Reference Laboratories of the NGSP, 3 refer-
3
Department of Clinical Chemistry, Malmo University Hospital, Lund ence laboratories of the JDS/JSCC program, and a Swed-
University, Malmo, Sweden. ish reference laboratory. Regression equations were
4
Department of Clinical Chemistry, Isala Klinieken, NA 8000 GM Zwolle,
The Netherlands.
determined for the relationship between the IFCC RM
5
Centers for Disease Control and Prevention, Atlanta, GA. and each of the DCMs.
6
Department of Laboratory Medicine. Austin & Repatriation Medical Results: Significant differences were observed between
Centre, Melbourne, Australia.
7
Institute of Biopathological Medicine, Ono, Japan.
the HbA1c results of the IFCC RM and those of the
8
Department of Clinical Biochemistry, Norfolk and Norwich University DCMs. Significant differences were also demonstrated
Hospital, Norwich, Norfolk, United Kingdom. between the three DCMs. However, in all cases the
9
University of Missouri School of Medicine, Columbia, MO.
10
Department of Science and Biomedical Technology, University of Milan,
relationship of the DCMs with the RM were linear.
Milan, Italy. There were no statistically significant differences be-
11
Institute of Biomedical Technology, Consiglio Nazionale Delle Ricerche, tween the regression equations calculated for each of
Milan, Italy.
12
Istituto di Ricovero e Cura a Carattere Scientifico, Hospital San Raffaele,
the four studies; therefore, the results could be com-
Milan, Italy. bined. The relationship is described by the following
13
German Diabetes Research Institute, Duesseldorf, Germany. regression equations: NGSP-HbA1c ⴝ 0.915(IFCC-
14
Department of Laboratory Medicine, School of Medicine, Keio Univer-
sity, Tokyo, Japan.
HbA1c) ⴙ 2.15% (r2 ⴝ 0.998); JDS/JSCC-HbA1c ⴝ
15
Laboratory of Analytical Chemistry, Faculty of Pharmaceutical Sciences, 0.927(IFCC-HbA1c) ⴙ 1.73% (r2 ⴝ 0.997); Swedish-
University of Gent, Ghent, Belgium. HbA1c ⴝ 0.989(IFCC-HbA1c) ⴙ 0.88% (r2 ⴝ 0.996).
16
Standard Reference Center, Kawasaki, Japan.
Conclusion: There is a firm and reproducible link
*Author for correspondence. Fax 31-38-4242676; e-mail [email protected].
Received July 21, 2003; accepted October 1, 2003. between the IFCC RM and DCM HbA1c values.
Previously published online at DOI: 10.1373/clinchem.2003.024802 © 2004 American Association for Clinical Chemistry

166
 Clinical Chemistry 50, No. 1, 2004 167

The measurement of hemoglobin A1c (HbA1c)17 in human N-terminal valine of the ␤-chain of Hb. Primary reference
blood is the most important marker for long-term assess- materials of pure HbA1c and HbA0 have been prepared
ment of the glycemic state in patients with diabetes, and (18 ), and a reference method that specifically measures
goals for therapy are set at specific HbA1c target values HbA1c has been developed (19, 20 ). The reference method
(1– 4 ). There are many commercial methods available for has been approved by all member national societies of the
the routine measurement of HbA1c. These methods are IFCC, and a global network of reference laboratories has
based on different analytical principles, such as immuno- been established (20, 21 ).
assays, ion-exchange chromatography, and affinity chro- When the IFCC Reference Method for the calibration of
matography. Several surveys (5– 8 ) have demonstrated HbA1c routine methods is used, laboratorians must con-
that different results are produced by these different sider that the current clinical interpretation of HbA1c
method principles. However, therapeutic targets for pa- results is based on data from tests that were calibrated to
tients with diabetes require method-independent values DCMs that were less specific than the IFCC Reference
for HbA1c. Furthermore, if in the future HbA1c is to be Method and therefore generated HbA1c values that are
used for the diagnosis of diabetes or impaired glucose higher than those that will be obtained if the calibration is
tolerance, it is essential that the different HbA1c methods traced back to the IFCC Reference Method. To ensure the
used routinely in medical laboratories provide compara- proper clinical use of the tests, it is important to under-
ble results. stand the numeric relationship between the IFCC Refer-
Studies have shown that harmonization of test results ence Method and the DCMs used in the national stan-
obtained by different HbA1c assays is feasible if all of the dardization schemes. This study was designed to
methods are calibrated with the same set of calibrators investigate this relationship.
(9, 10 ) and/or are adjusted to a Designated Comparison
Method (DCM) (11 ). These principles have been used in Materials and Methods
national initiatives for the harmonization of HbA1c re- ifcc reference method
sults. In the US, the National Glycohemoglobin Standard- The reference method was developed as candidate refer-
ization Program (NGSP) has been established (12 ). The ence method by Kobold et al. (19 ) on behalf of the IFCC
Japanese Diabetes Society (JDS) developed a set of na- Working Group on HbA1c Standardization, and was
tional calibrators with the recommendation to adjust the thereafter thoroughly evaluated and optimized by the
calibration of all routine HbA1c methods to these calibra- IFCC network of reference laboratories. In 2001, the final
tors (13 ). In Sweden, the Mono S method, a high-resolu- method was unanimously accepted in a ballot by the
tion ion-exchange HPLC method, has been chosen as the national member societies of the IFCC as the “Approved
DCM for the harmonization of HbA1c results (14 ). All of IFCC Reference Method for the Measurement of HbA1c in
these national initiatives were important steps toward Human Blood”. The method was published in 2002 in the
improvement of the comparability of HbA1c test results, IFCC section of the journal Clinical Chemistry and Labora-
but national standardization programs based on different tory Medicine (20 ). A PDF version of the method can be
DCMs cannot replace uniform worldwide standardiza- downloaded free of charge from the IFCC section of Clinical
tion anchored on a metrologically sound international Chemistry and Laboratory Medicine (www.degruyter.de/
reference measurement system (15, 16 ) comprising (a) a journals/cclm/pdf/401_78.pdf).
clear definition of the analyte based on its molecular The IFCC Reference Method has three steps. In the first
structure, (b) a primary reference material containing the step, Hb from washed and lysed erythrocytes is cleaved
analyte in a pure form, (c) a validated reference method into peptides by the proteolytic enzyme endoproteinase
that specifically measures the analyte in human samples, Glu-C. The resulting glycated and nonglycated N-termi-
and (d) a global network of reference laboratories that nal hexapeptides of the ␤-chain are then separated from
guarantees that the reference method is performed with the crude peptide mixture by reversed-phase HPLC. In
the necessary analytical quality and is capable of assign- the third and final step, the glycated and nonglycated
ing reliable values to matrix-based secondary reference hexapeptides are quantified by mass spectrometry or by
materials and calibrators. capillary electrophoresis with ultraviolet detection. The
The IFCC Working Group on HbA1c Standardization percentage of HbA1c is determined by the ratio of gly-
has developed such a reference system for HbA1c (17 ). cated to nonglycated ␤-N-terminal hexapeptides of Hb.
HbA1c is defined as the stable adduct of glucose to the The measurements in this study were performed by the
reference laboratories of the IFCC Network of Reference
Laboratories (IFCC-NRL), which are listed in the Appen-
17
Nonstandard abbreviations: HbA1c, hemoglobin A1c; DCM, Designated
dix. The network comprises laboratories from Europe,
Comparison Method; NGSP, National Glycohemoglobin Standardization Pro- Japan, and the US that have successfully established the
gram; JDS, Japanese Diabetes Society; NRL, Network of Reference Measure- reference method. The network is coordinated by the
ment Laboratories; DCCT, Diabetes Control and Complications Trial; CPRL,
Network Coordinator, who is responsible for the organi-
Central Primary Reference Laboratory; PRL, Primary Reference Laboratory;
SRL, Secondary Reference Laboratory; and JSCC, Japanese Society of Clinical zation of meetings, updating the Standard Operating
Chemistry. Procedure of the reference method, and organizing regu-
168 Hoelzel et al.: HbA1c IFCC Reference Method and DCMs

lar quality-control surveys within the network. The net- swedish standardization scheme
work works on behalf of, and is supervised by, the IFCC The Swedish standardization scheme uses the Mono S
Working Group on HbA1c Standardization (21 ). method (strong methylsulfonate cation exchanger on
monobeads; Amersham Biosciences) as DCM for the
ngsp standardization scheme harmonization of HbA1c measurements (25 ). The mea-
The NGSP system was established after the Diabetes surements in this study were performed by the Swedish
Control and Complications Trial (DCCT) study showed IFCC reference laboratory at the Malmo University Hos-
the relationship between HbA1c and risks for develop- pital. The laboratory is a part of the EQAS organization,
ment and/or progression of diabetes complications. Im- External Quality Assurance in Laboratory Medicine in
plementation of treatment goals based on the results of Sweden (EQUALIS), located in Uppsala. Split samples of
the DCCT in clinical settings made it necessary to harmo- fresh EDTA blood are distributed once a month to 40
nize HbA1c results (22 ). The anchor for the NGSP labora- hospitals using different HPLC methods. Five of them are
tory network is a DCM, which is ion-exchange HPLC contracted to run the Mono S system in a national
using Bio-Rex 70 resin (Bio-Rad Laboratories) (23 ). The network. These laboratories are used for calibration of all
method is performed in the Central Primary Reference hospital and point-of-care instruments in Sweden every
Laboratory (CPRL) and backup Primary Reference Labo- second year.
ratories (PRLs) (12 ). Secondary Reference Laboratories
(SRLs) have been established to assist manufacturers with design and logistics
calibration to the DCCT as well as serving as the compar- The study was designed by the IFCC Working Group on
ison methods for NGSP certification. These laboratories HbA1c Standardization and organized logistically by the
use HbA1c assay methods of various method types (ion- IFCC reference laboratory in Winterswijk (The Nether-
exchange HPLC, affinity HPLC, and immunoassay) that lands), which currently holds the function of the IFCC
are convenient and robust, provide excellent analytical Network Coordinator, and the reference laboratory in
Zwolle (The Netherlands), which was responsible for the
performance, and are calibrated to the CPRL method. The
blood collection. The design was adjusted to the aim of
CPRL, PRLs, and SRLs work closely together in a network
the study, which was to generate equations that describe
of reference laboratories (NGSP-NRL). The network lab-
reliably the relationship between the various reference
oratories are monitored monthly by sample exchanges
systems. Therefore, low uncertainty of the resulting equa-
with the CPRL. The “NGSP-HbA1c values” in this study
tions as well as an acceptable workload for the reference
were measured by the SRLs located in the US and Europe.
laboratories performing the measurements had to be
considered. The uncertainty of the equations depends on
jds/jscc standardization scheme
many factors, such as the analytical performance of the
The basis of the JDS/JSCC standardization scheme are
measurements, the number of measurements per sample,
national calibrators. In 1995, the JDS developed a first set
systematic differences between the laboratories within the
of national calibrators, called JDS Calibrator Lot 1, which networks, the number of participating laboratories per
was recommended to be used for the calibration of all system, the number of samples analyzed, the biological
routine HbA1c assays in Japan. The calibrator values were variation of samples, and the number of measurement
assigned with the HPLC ion-exchange chromatography campaigns (covering shifts attributable to changes in
methods of TOSOH and Kyoto Daiichi. These two meth- calibration, reagents, and instruments). To evaluate all
ods were chosen because at the time when the calibrators factors that contribute to the uncertainty and to reduce the
were established, most of the Japanese medical laborato- overall uncertainty, blood pools were used instead of
ries used one of these HPLC methods. In recent years the single samples (explanation see below). The results
Japanese standardization scheme has evolved. The Japa- among networks and not among laboratories were com-
nese Society of Clinical Chemistry (JSCC) developed a pared, each specimen was measured four times by each
high-resolution ion-exchange HPLC method, named laboratory, and the experiments were repeated in four
KO500 (24 ), and a second set of national calibrators independent studies. The studies were performed 2001–
(deep-frozen blood), called JDS/JSCC Calibrator Lot 2. 2003 and were called Marrakech (2001), Chicago (2001),
The KO500 method was used to assign target values to the Kyoto I (2002), and Kyoto II (2003).
Lot 2 calibrators and to samples for national proficiency
testing. To keep consistency in the HbA1c values, the samples
calibration of the KO500 method was adjusted to the first When evaluating the quantitative relationship between
lot of JDS calibrators. For the measurements in this study, the reference systems, the biological variation in the
the KO500 HPLC method was calibrated with JDS Cali- samples used for the method comparison is a major
brator Lot 2. The measurements were performed by the confounding factor. The biological variation results from
three Japanese IFCC reference laboratories, which are also the fact that the composition of blood in each individual is
reference laboratories of the JDS/JSCC standardization slightly different. There are Hb derivatives such as car-
scheme. bamylated Hb and other adducts, and Hb forms that may
 Clinical Chemistry 50, No. 1, 2004 169

interfere with the NGSP-SRL methods in particular; these were obtained [regression equation: y ⫽ 1.006x ⫺ 0.035%
are commercial methods with a broad spectrum of HbA1c; r2 ⫽ 0.999; slope and the intercept did not deviate
method principles (ion-exchange chromatography, immu- significantly from 1 (P ⬎0.999) and 0 (P ⬎0.99), respec-
noassays, affinity chromatography) (26 ). Theoretically, tively]. However, whole blood is stable for ⬃1 week,
the influence of biological variation should be substan- whereas hemolysates stored at ⫺70 °C are stable for many
tially reduced if the samples used are mixtures of blood years (e.g., a hemolysate manufactured in 1999 and mea-
from various donors rather than samples derived from sured by the IFCC-NRL in 1999, 2001, and 2002 had HbA1c
single donors. This hypothesis was checked and con- values of 9.39%, 9.37%, and 9.32%, respectively). Because
firmed to be true in a separate study in which 36 dona- the IFCC Reference Method is time-consuming and the
tions were used to prepare 36 single specimens as well reference laboratories are spread globally, there was a
as 6 pools (each pool being a mixture of 6 individual need to have a time buffer for shipment and performing
donations), and HbA1c was measured by the network the measurements. Therefore, frozen hemolysates were
laboratories. The outcome was that the HbA1c values of used in this study. The hemolysates were prepared from
the pools were not significantly different from the mean of the blood pools according to the Standard Operating
the HbA1c values of the respective single donations (mean Procedure of the IFCC Reference Method (20 ) within 32 h
of singles, 6.84%; mean of mixtures, 6.85%; P ⬎0.999) and after blood drawing and stored at ⫺70 °C until shipment.
that the intralaboratory CV were the same for pools and The samples were shipped on dry ice (⫺79 °C; amount
singles donations (combined CV for singles 1.1%; for sufficient for 5 days) by courier to the IFCC-NRL labora-
mixtures 1.1%; P ⬎0.999), but the scatter of the HbA1c tories. The reference laboratories checked for the presence
values around the regression line characterized by Sy兩x of dry ice, which was still present when the samples
was significantly lower for pools than for single donations arrived at all laboratories during all studies, and stored
(e.g., for the relationship of the IFCC-NRL and the NGSP- the samples at ⫺70 °C until analysis. The measurements
NRL, Sy兩x was 0.17 for single donations and 0.08 for pools,
were performed within 9 weeks of hemolysate prepara-
respectively). This means that use of blood pools could
tion.
substantially reduce the uncertainty of the resulting re-
gression equations. Therefore, the blood pools used in this
specimens for the DCMs
study were as follows: in the Marrakech study, eight
All of the DCMs work with hemolysates, but the methods
samples in the range 3.04 –9.65%; in the Chicago study,
use different hemolyzing agents and different dilutions.
eight samples in the range 3.30 –9.00%; in the Kyoto I
Therefore, whole-blood specimens were sent to the DCM
study, five samples in the range 3.09 –11.25%; and in the
laboratories, which were prepared locally for the mea-
Kyoto II study, five samples in the range 3.48 – 8.65%
surement according to the specific protocol of the DCM
HbA1c (HbA1c percentages determined by the IFCC Ref-
used. Blood from the pools was dispensed in 1-mL
erence Method). Each sample was prepared from 10
donations. aliquots, packed in special isolating boxes on cool packs
Before mixing, each donation was checked for (a) (2– 8 °C; capacity 4 days) within 8 h after preparation of
hepatitis B surface antigen, anti-HIV, and anti-hepatitis C the pools, and immediately shipped by courier. On ar-
antibodies (Abbott Laboratories; criterion, samples must rival, the DCM laboratories checked the temperature and
be negative); (b) abnormal Hb variants such as S, C, E, and stored the specimens until the measurement. All measure-
F (Menarini 8140 method; A.Menarini; criterion, HbF ments were performed within 5 days after blood drawing.
⬍1% and other variants absent); and (c) abnormal
amounts of urea to exclude increased concentrations of quality control
carbamylated Hb. The HbA1c value was used to make A precondition for the inclusion of the results of the IFCC
pools with appropriate HbA1c concentrations. The blood network laboratories was that they had successfully
samples (60 –90 donations for each campaign) were drawn passed the regular quality-control surveys of the IFCC-
and checked within a time frame of 32 h. Donations were NRL. The IFCC-NRL control surveys were organized
stored in the refrigerator (2– 8 °C) until pools were pre- twice a year. Six hemolysates with HbA1c concentrations
pared. The pools were the starting material to supply both covering the clinically relevant concentration range (3–
IFCC network laboratories and DCMs with specimens 13% HbA1c) were distributed to the network laboratories,
(described below). and each sample was measured four times by the refer-
ence laboratories. The criteria for passing were that the
specimens for the ifcc reference method mean intralaboratory CV was ⱕ2.5% and the mean devi-
Both whole blood and hemolysates are suitable materials ation from the overall mean of all NRL laboratories was
for the IFCC Reference Method and provide the same ⱕ2.5%. The mean intralaboratory CV of the NRL labora-
numerical HbA1c results. This was demonstrated during tories in the surveys performed during this study was
the development of the reference method: a panel of 1.0 –1.2%, the intralaboratory CV varied from 0.50% to
whole-blood samples and hemolysates from the same 2.2%, and the interlaboratory CV were in the range of
blood were measured in parallel, and identical results 1.4 –1.9%. In two studies, one reference laboratory did not
170 Hoelzel et al.: HbA1c IFCC Reference Method and DCMs

meet the criteria; its results were therefore not included

Table 1. Comparison of IFCC HbA1c Reference Method with
when calculating the overall IFCC-NRL HbA1c values.
the national DCMs: individual results for the four method-
The laboratories of the NGSP-NRL participated in a
comparison studies.
monthly monitoring program with specific precision and
Studya
bias limits (12 ). All laboratories met the requirements of
the program during the time of the method-comparison Marrakech Chicago Kyoto I Kyoto II
Reference system (n ⴝ 8) (n ⴝ 8) (n ⴝ 5) (n ⴝ 5)
studies (mean difference between CPRL and SRL HbA1c
NGSP-NRL
values ⬍0.35%; SD of difference replicates ⬍0.23%). (9 laboratories)
The JDS/JSCC-NRL laboratories twice a year ex- Slope 0.926 0.926 0.906 0.912
changed a set of samples (deep-frozen blood) with five Intercept, % 2.14 2.05 2.21 2.17
HbA1c concentrations (4.05–12.63%) and participated in r2 0.999 0.999 0.999 0.999
the IFCC/DCM trials. The interlaboratory CV in these JDS/JSCC-NRL
studies were ⬍1.0%. (3 laboratories)
The Swedish reference laboratory participated in the Slope 0.934 0.926 0.920 0.943
national network of Mono S reference laboratories, which Intercept, % 1.76 1.67 1.78 1.68
performed regular intercomparison studies. Split samples r2 0.996 0.999 0.999 0.999
of fresh EDTA blood were distributed once a month. The Swedish-RL
(1 laboratory)
interlaboratory CV in this Mono S network intercompari-
Slope 1.008 0.941 1.002 0.968
son studies were ⬍1.5%.
Intercept, % 0.90 1.09 0.78 1.08
r2 0.999 0.999 0.999 0.999
statistical evaluation a
n, number of samples.
The mean values of the four measurements per sample
and the SDs and CV for these measurements were calcu-
lated for each individual laboratory. The mean values of the JDS/JSCC-NRL. The slopes and intercepts of the
the laboratories were used to calculate the “overall mean regression equations and the r2 for the relationship be-
values ” of the IFCC-NRL, the NGSP-NRL, and the tween the IFCC-HbA1c values and the DCM-HbA1c values
JDS/JSCC-NRL. The overall mean values were used as of the four studies are listed in Table 1. The Kruskal–
“IFCC-HbA1c values”, “NGSP-HbA1c values”, “JDS/ Wallis test demonstrated that there were no statistically
JSCC-HbA1c values”, and “Swedish-HbA1c values” for the significant differences in the outcomes of the four studies
calculation of the relationship between the IFCC reference [IFCC-NRL, ␹2 ⫽ 0.70 (P ⫽ 0.87); NGSP-NRL, ␹2 ⫽ 0.83
system and the DCM-based systems. Each of the four (P ⫽ 0.84); JDS/JSCC-NRL, ␹2 ⫽ 0.90 (P ⫽ 0.82); Swedish-
method-comparison studies was evaluated separately. RL, ␹2 ⫽ 0.88 (P ⫽ 0.83)] so that the results of the four
Whether the correlation between the IFCC Reference studies could be combined to calculate the overall regres-
Method and the DCMs fits a linear regression model was sion equations (Table 2). The slope and the intercept of the
checked by visual inspection. Because this could be con- equation for the IFCC vs NGSP methods deviated signif-
firmed for all comparisons, linear regression analysis was icantly from 1 and 0, respectively (P ⬍0.001). However,
used for calculating slopes, intercepts, and r2, with use of the relationship between the two systems was linear, and
SAS software, Ver. 8.2 (SAS Institute). The SAS software
was also used to apply the Kruskal–Wallis test to check
whether there were statistically significant differences in Table 2. Comparison of the IFCC HbA1c Reference Method
the outcomes of the four studies. Finally, the results of the with the national DCMs: combined results for the four
four studies were combined, and the overall regression method-comparison studies.
equations were calculated for the correlation between the Reference system
DCM-based systems and the IFCC reference system. The NGSP-NRL JDS/JSCC-NRL Swedish-RL
presence of statistically significant differences between
No. of samples 26 26 26
the system values was checked, as indicated by the slope Intercept, % 2.152a 1.724a 0.884a
deviating statistically significantly from 1 and/or the SE, % 0.050 0.065 0.080
intercept from 0, respectively. 95% confidence 2.049–2.256 1.590–1.859 0.718–1.049
interval, %
Results Slope 0.9148a 0.9274a 0.9890
The mean intralaboratory CV in the four studies was 1.1% SE 0.0075 0.0098 0.012
(range, 0.38 –2.1%) for the 11 laboratories of the IFCC- 95% confidence 0.899–0.930 0.907–0.948 0.964–1.014
NRL; 1.0% (range, 0.47–2.4%) for the 9 SRLs of the NGSP; interval
0.54% (0.27– 0.87%) for the three Japanese reference labo- RMSEb 0.084 0.109 0.134
ratories, and 0.57% (0.30 – 0.58%) for the Swedish refer- r2 0.998 0.997 0.997
ence laboratory. The mean interlaboratory CV were 1.9%
a
Significant deviation of slope from 1 or intercept from 0 (P ⬍0.001).
b
RMSE, root mean square error.
for the IFCC-NRL, 1.9% for the NGSP-NRL, and 0.66% for
 Clinical Chemistry 50, No. 1, 2004 171

Fig. 1. HbA1c values measured in 26 pooled blood samples by the Fig. 3. HbA1c values measured in 26 pooled blood samples by the
IFCC-NRL, applying the IFCC Reference Method (mean value of 11 IFCC-NRL, applying the IFCC Reference Method (mean value of 11
network laboratories), and by the NGSP-NRL, applying various methods network laboratories), and by the Swedish-RL, applying the DCM Mono
adjusted to the NGSP DCM Bio-Rex 70 HPLC (mean values of 9 S HPLC.
network laboratories). The lines are the regression line and the y ⫽ x line, respectively.
The lines are the regression line and the y ⫽ x line, respectively.

There were also statistically significant differences in

there was very little dispersion of the measured values the slopes as well in the intercepts between the three
around the regression line (r2 ⫽ 0.998; see Fig. 1). There DCM-based systems (Table 3).
were also statistically significant differences between the
IFCC-HbA1c values and the JDS/JSCC-HbA1c values. Discussion
However, as with the NGSP, there was a strong linear To ensure optimum quality of patient care and a valid
correlation between both systems (r2 ⫽ 0.997; see Fig. 2). interpretation of clinical trials, good interlaboratory
There was a slightly higher variation in the outcome of the agreement of results in laboratory medicine is essential.
four method-comparisons studies of the IFCC-NRL vs the The most effective approach for achieving universal
Swedish-RL, but the differences were not statistically comparability of test results is the ability to trace the
significant. The higher random variation was attributable calibration of the routine assays back to an international
to the HbA1c values from the Swedish standardization metrologically sound reference measurement system.
scheme not being the mean values of a network of Therefore, an essential requirement of the IVD Directive
reference laboratories but rather the results from one of the European Union is that manufacturers have to trace
laboratory. In contrast to the two other systems, the back the calibration of their tests to reference methods or
overall slope of 0.989 did not deviate significantly from 1 reference materials of higher metrologic order if available
(P ⬎0.95; see Fig. 3), whereas the intercept was signifi- (27 ).
cantly different from 0 (P ⬍0.001). With the development of the IFCC reference system, a

Table 3. Comparison of the national DCMs: combined

results for the four method-comparison studies.
DCM

NGSP-NRL JDS/JSCC-NRL Swedish-RL

NGSP-NRL (9 laboratories)
Slope 0.985a 0.923b
Intercept 0.459b 1.345b
r2 0.999 0.998
JDS/JSCC-NRL (3 laboratories)
Slope 1.014a 0.936b
Intercept ⫺0.459b 0.907b
2
r 0.999 0.997
Swedish-RL (1 laboratory)
Fig. 2. HbA1c values measured in 26 pooled blood samples by the Slope 1.081b 1.065b
IFCC-NRL, applying the IFCC Reference Method (mean value of 11 Intercept ⫺1.442b ⫺0.947b
network laboratories), and by the JDS/JSCC-NRL, applying the JDS/ 2
r 0.998 0.997
JSCC DCM KO 500 calibrated with JDS calibrators (mean values of 3
network laboratories).
a,b
Significant deviation of slope from 1 or intercept from 0: aP ⬍0.05;
b
P ⬍0.001.
The lines are the regression line and the y ⫽ x line, respectively.
172 Hoelzel et al.: HbA1c IFCC Reference Method and DCMs

reference method and reference materials of higher met-

Table 4. Examples of the numeric relationship between
rologic order are now available for the measurement of
IFCC HbA1c values and DCM-based values.
HbA1c. In the IFCC reference system, HbA1c is defined on
NGSP HbA1c JDS/JSCC HbA1c Swedish HbA1c
its molecular structure and specifically measured with a IFCC
reference method, whereas in the existing DCM-based HbA1c, % HbA1c, % utrans,a % HbA1c, % utrans, % HbA1c, % utrans, %
national standardization schemes HbA1c is defined and 5.30 7.00b 0.018 6.64 0.024 6.13 0.029
measured as the “HbA1c” peak of the chromatogram of 7.00 8.56 0.017 8.22 0.022 7.80 0.028
the chosen DCM. These HbA1c peaks contain not only a
utrans is the standard uncertainty, which is generated if the equations
HbA1c but also, depending on the resolution of the resin described in this report are used to transform IFCC HbA1c values into DCM-based
values.
used, to some extent substances that are not HbA1c b
HbA1c target value for NGSP.
(28 –32 ). To minimize the confounding effect of biological
variation, blood pools were used as samples in this study,
and blood with abnormal Hb variants such as S, C, E, and when the nonspecific glucose routine methods, based on
F and abnormal urea concentrations (carbamylated Hb) the measurement of the reduction caused by glucose and
were excluded. other substances in blood, were replaced by methods that
Considering the lack of specificity of the DCMs, it is specifically measured glucose (34, 35 ). Because of the
not surprising that all three DCMs generate significantly close correlation between the IFCC Reference Method and
higher results than the IFCC Reference Method and that the DCMs, it is possible to derive IFCC values from the
there are significant differences among the results of the existing scales of numbers by use of linear regression
three DCMs as well (see Table 3). The NGSP-SRL gener- equations. Because the Bio-Rex 70 method was used as
ates the highest HbA1c values because the HbA1c peak of reference for harmonization of the HbA1c test results in
the Bio-Rex 70 method, used as anchor of the NGSP, the DCCT study (1 ), it is also possible to “translate” the
contains a high proportion of non-HbA1c substances, such risk curves generated in this landmark study and the
as HbF, minor Hb forms, and carbamylated Hb, and the HbA1c data of the important UKPDS study (2 ), which
peak is not clearly separated from the neighboring non- were also adjusted to the Bio-Rex 70 calibration. When
HbA1c peaks (23 ). The KO500 used in the Japanese regression equations are used for transforming IFCC
scheme is a high-resolution HPLC method, but because HbA1c values into DCM HbA1c values and vice versa, the
this method is based on calibrators with values that were additional uncertainty introduced by the uncertainty of
assigned with the older HPLC methods from TOSOH and the regression equations must be considered. This uncer-
Kyoto Daiichi, the JDS/JSCC-NRL values reflect the low tainty must be added to the uncertainty of the analytical
specificities of these methods. The results of the JDS/ measurement. The uncertainty values given in Table 4
JSCC-NRL are slightly lower than those of the NGSP- were calculated according to the Eurochem/Citac Guide
NRL. The Swedish DCM generates the lowest DCM for Quantifying Uncertainty in Analytical Measurement
HbA1c values. The Mono S system, developed in 1983, (36 ), considering the IFCC value as independent variable.
shows an almost homogeneous HbA1c peak in the chro- The standard uncertainty is the standard error of the
matogram, but it contains carbamylated Hb as well as free predicted DCM value.
␣-globulin chains (25, 31, 32 ), so that the values are higher Changing medical decision criteria is not just a matter
than those measured with the IFCC Reference Method. In for laboratory professionals but also for the healthcare
contrast to the DCM methods, new dedicated HPLC providers and patients who use these criteria. Therefore,
systems are today eliminating many of these interfering the IFCC Working Group on HbA1c Standardization has
adducts such as carbamylated Hb by use of more modern contacted the international scientific societies of diabe-
chromatographic material and improved gradients (33 ). tologists to discuss appropriate ways of adopting the
There is, however, a strong correlation between the IFCC standardization for HbA1c routine tests in clinical
IFCC values and the DCM values, which can be described practice.
by a linear regression model. The relationships were
identical in all four studies, and the dispersion of the
values around the regression line was small. It is therefore This work was partially supported by R&D Project CT
possible to establish reliable numerical links between the 98-2248 within the framework of the S, M & T Program of
results of the IFCC Reference Method and the DCMs the European Commission.
described by linear regression equations. When calibra-
tors with IFCC values are used, the HbA1c values of the Appendix
routine methods will be lower than those generated with participating ifcc reference laboratories
the previous calibration. There are significant numerical Department of Clinical Chemistry, Malmo University
differences between HbA1c values based on the IFCC Hospital, Malmo, Sweden; Roche Diagnostics GmbH,
Reference Method calibration and HbA1c values based on Penzberg, Germany; Department of Clinical Chemistry,
the DCM calibrations (see Table 4). These changes are Isala Klinieken, Zwolle, The Netherlands; Department of
similar to the change in glucose values four decades ago Clinical Chemistry, Queen Beatrix Hospital, Winterswijk,
 Clinical Chemistry 50, No. 1, 2004 173

The Netherlands; Department of Science and Biomedical 9. Weykamp CW, Penders TJ, Muskiet FAJ, van der Slik W. Effect of
Technology, University of Milan, Milan, Italy; IRCCS calibration on dispersion of glycohemoglobin values determined by
Hospital San Raffaele, Milan, Italy; Institute of Biomedical 111 laboratories using 21 methods. Clin Chem 1994;40:138 –
44.
Technology, Consiglio Nazionale Delle Richerche, Milan,
10. Weykamp CW, Penders TJ, Miedema K, Muskiet FAJ, van der Slik
Italy; Centers for Disease Control and Prevention, Atlanta,
W. Standardization of glycohemoglobin results and reference
GA; Standardization Reference Center, Kawasaki, Japan; values in whole blood studied in 103 laboratories using 20
Institute of Biopathological Medicine, Ono, Japan; Labo- methods. Clin Chem 1995;41:82– 6.
ratory of Analytical Chemistry, Faculty of Pharmaceutical 11. Little RR, Wiedmeyer H-M, England JD, Wilke AL, Rohlfinf CL,
Sciences, University of Gent, Ghent, Belgium. Jacobson JM, et al. Interlaboratory standardization of measure-
ment of glycohemoglobins. Clin Chem 1992;38:2471– 8.
participating ngsp SRLs 12. Little RR, Rohlfing CL, Wiedmeyer HM, Myers GL, Sacks DB,
Methods are given in parentheses: Diabetes Diagnostic Goldstein DE. The National Glycohemoglobin Standardization Pro-
gram: a five-year progress report. Clin Chem 2001;47:1985–92.
Laboratory, University of Missouri School of Medicine,
13. Shima K, Endo J, Oimomi M, Oshima I, Omori Y, Katayama Y, et al.
Columbia, MO (Bio-Rad Diamat HPLC; Tosoh 2.2 Plus
Inter-laboratory difference in HbA1c measurement in Japan. A
HPLC; Primus CLC330 HPLC); Collaborative Studies report of the Committee on an Inter-laboratory Standardization of
Clinical Laboratory, Fairview University Medical Center, HbA1c Determination, the Japan Diabetes Society. J Jpn Diabetes
Minneapolis, MN (Bio-Rad Diamat HPLC; Tosoh 2.2 Plus Soc 1994;37:855– 64.
HPLC); Core Laboratory for Clinical Studies, Washington 14. Arnquist H, Wallensteen M, Jeppsson JO. Standardization of
University School of Medicine, St. Louis, MO (Roche longterm glucose measurements established. Lakartidningen
Tina-quant II on Hitachi 917); Queen Beatrix Hospital, 1997;50:4789 –90.
Winterwijk, The Netherlands (Beckman CE; Bio-Rad Dia- 15. Müller MM. Implementation of reference systems in laboratory
mat; Menarini HA8160 HPLC); Isala Klinieken, Zwolle, medicine. Clin Chem 2000;46:1907–9.
The Netherlands (Roche Unimate-Modular Analytics; Pri- 16. International Organization for Standardization. In vitro diagnostic
medical devices–measurement of quantities in samples of biolog-
mus CLC385 HPLC).
ical origin—metrological traceability of values assigned to calibra-
tors and control material. ISO 17511. Geneva, Switzerland: ISO,
participating jds reference laboratories 2003.
Institute of Biopathological Medicine, Ono, Japan; Stan- 17. Hoelzel W, Miedema, K. Development of a reference system for
dard Reference Center, Kawasaki, Japan; Department of the international standardisation of HbA1c/glycohemoglobin de-
Laboratory Medicine, School of Medicine, Keio Univer- terminations. J Int Fed Clin Chem 1996;9:62–7.
sity, Tokyo, Japan. 18. Finke A, Kobold U, Hoelzel W, Weycamp C, Jeppsson JO, Miedema
K. Preparation of a candidate primary reference material for the
international standardisation of HbA1c determinations. Clin Chem
References
Lab Med 1998;36:299 –308.
1. The Diabetes Control and Complications Trial Research Group.
The effect of intensive treatment of diabetes on the development 19. Kobold U, Jeppsson JO, Dülffer T, Finke A, Hoelzel W, Miedema K.
and progression of long term complications in insulin-dependent Candidate reference methods for HbA1c based on peptide map-
diabetes mellitus. N Engl J Med 1993;329:977– 86. ping. Clin Chem 1997;43:1944 –51.
2. Stratton IM, Adler AI, Neil HA, Matthews DR, Manley SE, Cull CA, 20. Jeppsson J-O, Kobold U, Barr J, Finke A, Hoelzel W, Hoshino T, et
et al. Association of glycaemia with macrovascular and microvas- al. Approved IFCC Reference Method for the measurement of
cular complications of type 2 diabetes (UKPDS 35): prospective HbA1c in human blood. Clin Chem Lab Med 2002;40:78 – 89.
observational study. Br Med J 2000;321:405–11. 21. Jeppsson JO, Hoelzel W, Kobold U, Finke A, Mauri P, Miedema K,
3. Krishnamurti U, Steffes MW. Glycohemoglobin: a primary predictor et al. International network of reference laboratories for the
of development or reversal of complications of diabetes mellitus. determination of HbA1c [Abstract]. Clin Chem 1998;44 (S6):A22.
Clin Chem 2001;47:1157– 65. 22. Little RR, Goldstein DE. Endocrine standardization of glycohemo-
4. ADA. Clinical practice guidelines 2001. Diabetes Care 2001; globin measurement. Anal Chem 1995;67:393R–7R
24(Suppl 1):S1–133. 23. Goldstein DE, Little RR, England JD, Wiedmeyer HM, McKenzie
5. Little RR, Wiedmeyer HM, England JD, Naito HK, Goldstein DE. EM. Methods for quantitating glycosylated hemoglobins: high
Interlaboratory comparison of glycohemoglobin results: College of performance liquid chromatography and thiobarbituric acid col-
American Pathologists survey data. Clin Chem 1991;37:1725–9. orometry: In: Clarke WL, Larner, Pohl S, eds. Methods in diabetic
6. Weykamp CW, Penders TJ, Muskiet FA, van der Slik W. Glycohe- research, Vol. 2. Clinical methods. New York: John Wiley, 1986:
moglobin: comparison of 12 analytical methods, applied to lyoph- 475–504.
ilized haemolysates by 101 laboratories in an external quality 24. Hoshino T, Nakayama T, Kuwa K, Nakanishi T, Okahashi M,
assurance programme. Ann Clin Biochem 1993;30:169 –74. Tominaga M, et al. Reference method for St-GHbA1c determina-
7. Hoshino T, Okahashi M, Arai H. Survey and assessment of the tion—standard operating procedure, Ver.1.4. Hamamatsu, Japan:
actual state of routine measurement of glycohaemoglobin/GHb by Japanese Society of Clinical Chemistry Working Group on SOP for
commercial methods: warning to the users and the providers. St-GHbA1c Determination, 2000.
J Pharm Biomed Anal 1997;15:1551– 62. 25. Jeppsson JO, Jerntorp P, Sundkvist G, Englund H, Nylund V.
8. Mosca A, Paleari R, Trapolino A, Capani F, Pagano G, Plebani M. A Measurement of hemoglobin A1c by a new liquid chromatographic
re-evaluation of glycohemoglobin standardization: the Italian expe- assay: methodology, clinical utility and relation to glucose toler-
rience with 119 laboratories and 12 methods. Eur J Clin Chem Clin ance evaluated. Clin Chem 1986;32:1867–72.
Biochem 1997;35:243– 8. 26. Weykamp CW, Penders TJ, Muskiet FAJ, van der Silk W. Hemoglo-
174 Hoelzel et al.: HbA1c IFCC Reference Method and DCMs

bins S and C: reference values for glycohemoglobin in heterozy- cation exchange liquid chromatography. Evaluation of a Mono S
gous, double-heterozygous subjects and homozygous subjects, as column method. Clin Chim Acta 1996;253:159 – 69:
established by 13 methods. Clin Chim Acta 1994;231:161–71. 32. Koskinen LK, Ala-Houhala IO, Lahtela JT, Laippala PJ, Koivula TA.
27. Directive 98/79/EC of the European Parliament and of the Does uremia interfere with HbA1c results in the FPLC method with
Council of 27 October 1998 on in vitro diagnostic medical Mono S cation exchanger? Clin Chim Acta 1998;273:69 –79.
devices. Off J L 1998;331:1–37. 33. Little RR, Tennhill AL, Rohlfing C, Wiedemeyer HM, Khanna R,
Goel S, et al. Can glycohemoglobin be used to assess glycemic
28. Fluckinger R, Woodtli T. Effect of temperature on quantifying
control in patients with chronic renal failure? Clin Chem 2002;48:
glycated (glycosylated) hemoglobin by cation exchange chroma-
784 –5.
tography. Clin Chem 1985;31:114 –7.
34. Marks V. An improved glucose-oxidase method for determining
29. Cohen MP, Witt J, Wu VY. Purified hemoglobin preparations in the blood, C. S. F. and urine glucose levels. Clin Chim Acta 1959;4:
evaluation of HbA1c determination by ion exchange chromatogra- 395– 400.
phy. Ann Clin Biochem 1993;30:265–71. 35. Washko ME, Rice EW. Determination of glucose by an improved
30. Bisse E, Huaman-Guillen P, Wieland H. Chromatographic evalua- enzymatic procedure. Clin Chem 1962;7:542–5.
tion of minor hemoglobins; clinical significance of hemoglobin A1d. 36. De Zorzi P, Belli M, Babizzi S, Menegon S, Deluisa A. EURACHEM-
Clin Chem 1995;41:658 – 63. CITAC guide (2000). Quantifying uncertainty in analytical measure-
31. Koskinen LK. Specificity of hemoglobin A1c measurement by ments, 2nd ed. Accred Qual Assur 2000:7:182– 8.