Does Glucose Variability Influence the
Relationship BetweenMean Plasma
Glucose and HbA1c Levels in Type 1
and Type 2 Diabetic Patients?
JUDITH C. KUENEN, MD1
RIKKE BORG, MD2
DIRK J. KUIK, MSC3
HUI ZHENG, PHD4
DAVID SCHOENFELD, PHD4
MICHAELA DIAMANT, MD1
DAVID M. NATHAN, MD5
ROBERT J. HEINE, MD6
ON BEHALF OF THE ADAG STUDY GROUP*
OBJECTIVE—The A1C-Derived Average Glucose (ADAG) study demonstrated a linear re-
lationship between HbA1c and mean plasma glucose (MPG). As glucose variability (GV) may
contribute to glycation, we examined the association of several glucose variability indices and the
MPG-HbA1c relationship.
RESEARCH DESIGN AND METHODS—Analyses included 268 patients with type 1
diabetes and 159 with type 2 diabetes. MPG during 3 months was calculated from 7-point self-
monitored plasma glucose and continuous glucose monitoring. We calculated three different
measures of GV and used a multiple-step regression model to determine the contribution of the
respective GV measures to the MPG-HbA1c relationship.
RESULTS—GV, as reflected by SD and continuous overlapping net glycemic action, had a
significant effect on the MPG-HbA1c relationship in type 1 diabetic patients so that high GV led
to a higher HbA1c level for the same MPG. In type 1 diabetes, the impact of confounding and
effect modification of a low versus high SD at an MPG level of 160 mg/dL on the HbA1c level is
7.02 vs. 7.43 and 6.96 vs. 7.41. All GV measures showed the same tendency.
CONCLUSIONS—In only type 1 diabetic patients, GV shows a significant interaction with
MPG in the association with HbA1c. This effect is more pronounced at higher HbA1c levels.
However, the impact of GV on the HbA1c level in type 1 diabetes is modest, particularly when
HbA1c is close to the treatment target of 7%.
Diabetes Care 34:1843–1847, 2011
S ince the Diabetes Control and Com-plications Trial (DCCT) and theUK Prospective Diabetes Study
(UKPDS) (1,2) established the relation-
ship betweenHbA1c and the development
of long-term diabetes complications,
HbA1c has become the key monitoring
tool in diabetes management.
During the lifetime of the erythrocyte,
hemoglobin (Hb) is gradually glycated. The
proportion of the glycated sites, HbA1c,
within the erythrocyte increases through-
out its life span and reflect the exposure to
mean blood glucose (MBG) levels during
the preceding 2–3 months (3). This non-
enzymatic posttranslational modification is
relatively slow. In vivo and in vitro studies
have shown that HbA1c levels are directly
proportional to the time-averaged con-
centration of glucose during the erythro-
cyte’s life span (3–6). Given the kinetics
of glycation, brief periods of hyperglyce-
mia should not have a major impact on
HbA1c levels (7–9).
However, increased glycated protein
levels are documented in some nondia-
betic pathological states. So, hyperglyce-
mia is not the complete answer to the
etiology of increased early glycated prod-
ucts in nondiabetic conditions. A common
denominator is oxidative stress. It has
been hypothesized that oxidative stress
either via increasing reactive oxygen spe-
cies or by depleting the antioxidants may
modulate the genesis of early glycated
proteins in vivo (10,11). Hyperglycemia
stimulates oxidative stress (12) and glu-
cose variability; in particular, postprandial
glucose excursions have been regarded
as potentially deleterious as a result of,
among other factors, their association
with the increase of oxidative stress (13).
Therefore, glucose variability (GV) could
influence the glycation of HbA1c.
Previous studies have examined
whether the relationship between mean
plasma glucose (MPG) levels and HbA1c is
influenced by glucose variability and
found no or minimal influence (10,14,15).
However, these studies used limited self-
monitoring of blood glucose (SMBG) data
to assess mean glucose levels and variabil-
ity in relatively small numbers of measure-
ments. Thesemethods could underestimate
glycemic excursions. Continuous glucose
monitoring (CGM) provides a more com-
plete view of glycemic excursions, includ-
ing the duration and frequency of the
excursions, and allows calculation of
features of GV. Our aim was to examine
the influence of GV on the MPG-HbA1c
relationship in the A1C-Derived Average
Glucose (ADAG) study.
RESEARCH DESIGN AND
METHODS—The ADAG study was
conducted at 10 centers in the U.S.,
c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c c
From the 1Diabetes Center, VU University Medical Center, Amsterdam, the Netherlands; the 2Steno Diabetes
Center, Copenhagen, Denmark; the 3Department of Epidemiology and Biostatistics, VU University Medical
Center, Amsterdam, the Netherlands; the 4Biostatistics Center, Massachusetts General Hospital, Harvard
Medical School, Boston, Massachusetts; the 5Diabetes Center, Massachusetts General Hospital, Harvard
Medical School, Boston, Massachusetts; and 6Lilly, Minneapolis, Minnesota.
Corresponding author: Judith C. Kuenen, jc.kuenen@me.com.
Received 2 December 2010 and accepted 12 May 2011.
DOI: 10.2337/dc10-2217
This article contains Supplementary Data online at http://care.diabetesjournals.org/lookup/suppl/doi:10.
2337/dc10-2217/-/DC1.
*A complete list of the members of the ADAG Study Group can be found in the Supplementary Data.
© 2011 by the American Diabetes Association. Readers may use this article as long as the work is properly
cited, the use is educational and not for profit, and thework is not altered. See http://creativecommons.org/
licenses/by-nc-nd/3.0/ for details.
care.diabetesjournals.org DIABETES CARE, VOLUME 34, AUGUST 2011 1843
P a t h o p h y s i o l o g y / C o m p l i c a t i o n s
O R I G I N A L A R T I C L E
Europe, and Africa from 2006 to 2008 to
define the relationship between HbA1c
and average glucose levels. Because a
full description of this observational
study has been published (14), we de-
scribe it here only briefly. A total of
268 individuals with type 1 diabetes
and 159 individuals with type 2 diabetes
(age 18–70 years) completed the study.
Participants were selected based on sta-
ble glycemic control as evidenced by two
HbA1c values within one percentage
point of each other in the 6 months prior
to recruitment. Individuals with a wide
range of HbA1c levels were included.
Participants with conditions leading
to major changes in glycemia (infectious
disease, steroid therapy, and pregnancy)
or conditions thatmight interfere with the
measurement of HbA1c or the relation-
ship between HbA1c and MPG (hemoglo-
binopathies [16], anemia, increased
erythrocyte turnover, blood loss and/or
transfusions, or chronic renal or liver dis-
ease) were excluded (14). The study was
approved by the human studies commit-
tees at the participating institutions, and
informed consent was obtained from all
participants.
Measurements of glycemia
During the study period, CGM (Med-
tronic Minimed, Northridge, CA) was
performed at home four times with 4-week
intervals during the 16-week study pe-
riod.Monitoring period lasted at least 48h,
during which time glucose levels were
assessed every 5 min. CGM data were
accepted for analysis if there were no gaps
longer than 120 min and if the mean
absolute difference with the Hemocue
calibration results was ,18%, as recom-
mended by the manufacturer. For calibra-
tion purposes, participants performed
SMBGwith the Hemocuemeter (Hemocue
Glucose 201 plus; Hemocue, Ängelholm,
Sweden) during the days of CGM.
For adequate calculation of MPG,
subjects additionally performed a seven-
point SMBG (OneTouch Ultra; Lifescan,
Milipitas, CA) for at least 3 days per week
during the weeks when CGM was not
performed. All blood glucose values
stated are plasma equivalents.
HbA1c samples were analyzed with
four highly intercorrelated DCCT-aligned
assays: a high-performance liquid chroma-
tography assay, two immunoassays, and
an affinity assay (all approved by the Na-
tional Glycohemoglobin Study Program).
The mean HbA1c value at the end of the
12 week study period was used (14).
Calculating glucose variability
Three indices of intraday glucose variabil-
ity were calculated based on CGM: the SD
of mean glucose concentrations, the mean
amplitude of glycemic excursions (MAGE),
and the continuous overlapping net
glycemic action (CONGA). High SD,
MAGE, and CONGA values indicate high
intraday glucose variability. MAGE is the
mean of the differences between consec-
utive peaks and nadirs, only including
changes of.1 SD of glycemic values and
thus capturing only major fluctuations
(17). For the calculation of CONGAn,
the difference of the current value com-
pared with the value n hours previously
was calculated for each observation after
the first n hours. The CONGAn is the SD
of these differences (18). In the analyses,
we used CONGA at 4 h (CONGA4). Cal-
culations based on CGM data were calcu-
lated after exclusion of the initial 2 h of
monitoring, which is considered to be an
unstable calibration period.
Statistical analysis
First, we explored the correlations be-
tween MPG and HbA1c and measures of
glycemic variability as SD, MAGE, and/or
CONGA4 for the total diabetic population
and the two diabetes types. Multiple lin-
ear regression was used to investigate
confounding and effect-modifying influ-
ence of clinical parameters (glycemic var-
iability) on the relation between the
determinant (MPG) and outcome (HbA1c)
of interest. We then assessed which of the
variability measures (SD, MAGE, or
CONGA4) had the strongest impact on
the MPG-HbA1c relationship by con-
founding or effect modification.
Effect modification was concluded
when the slope of the interaction term of
glycemic variability and determinant was
significant. If no effect modificationmight
be concluded, a parameter DB was com-
puted as the relative difference of the
slope of the determinant in the model
without and with the clinical parameter.
Confounding was concluded when the
absolute value of DB exceeded the gener-
ally accepted threshold of 10%.
Multivariate confounding was inves-
tigated with a variant of stepwise regres-
sion, in which the stepping criterion was
not a P value but the DB as long as it ex-
ceeded the threshold. For significance, a
threshold of a = 0.05 was used.
Analyses were done for the total pop-
ulation and stratified for the type of diabetes.
Finally, we illustrated the magnitude of
the effect caused by the variability indices,
by confounding or effect modification, on
the MPG-HbA1c relationship.
RESULTS—Of the 507 patients en-
rolled, 427 completed the study and had
adequate glucose monitoring and HbA1c
samples for inclusion in the analyses. Two
hundred and sixty-eight participants had
type 1 diabetes, and 159 had type 2 diabe-
tes. The CGM and the SMBG data during
the 3-month period included approxi-
mately 2,400 and 300 measurements per
subject, respectively. The relationship be-
tween the HbA1c level at the end of the
3-month study period and MPG calcu-
lated over the preceding 3 months was
expressed as the simple linear regressions.
The formula for the total diabetic popula-
tion was as follows: HbA1c (%) = 0.028 3
MPG (mg/dL) + 2.66 (R2 = 0.80). The for-
mula for type 1 diabetes was as follows:
HbA1c (%) = 0.028 3 MPG (mg/dL) +
2.77 (R2 = 0.77). The formula for type 2
diabetes was as follows: HbA1c (%) =
0.0283MPG (mg/dL) + 2.62 (R2 = 0.82).
The clinical and glycemic character-
istics are shown in Table 1. Mean HbA1c
(SD) for type 1 diabetic patients was 7.3%
(1.1) and for type 2 diabetic patients was
6.8% (1.1).
All GV measures had significant in-
fluence on the MPG-HbA1c relationship
for the total population. The variability
index SD showed the strongest influence
on the MPG-HbA1c relationship. None of
theGVmeasures showed confounding for
all diabetic patients pooled or for the type
1 or type 2 diabetic patients separately
(Table 2).
In the type 1 diabetic patients, the
effect modification of SD and CONGA4
was significant (P , 0.01 and P = 0.02),
and for the MAGE it was just not signifi-
cant (P = 0.06) (Table 2). TheMPG/HbA1c
linear regression formula with confound-
ing for type 2 diabetes was as follows:
HbA1c (%) = 2.64 + 2.63 3 MPG/100 +
0.58 3 SD/100. The MPG-HbA1c linear
regression formula with effect modifica-
tion for type 1 diabetes was as follows:
HbA1c (%) = 3.91 + 1.79 3 MPG/100 2
1.37 3 SD/100 + 1.25 3 MPG/100 3
SD/100. The impact of effect modification
of low GV (SD = 30 mg/dL) versus high
GV (SD = 100 mg/dL) for an MPG level of
160 mg/dL in type 1 diabetes on the
HbA1c level was 6.96 vs. 7.41%, respec-
tively, as shown in Table 3. At an MPG
level of 220 mg/dL (HbA1c following the
regression formula of 8.89%), a decline in
the SD parameter from 100 to 30 mg/dL
reduced HbA1c from 9.23 to 8.26%.
1844 DIABETES CARE, VOLUME 34, AUGUST 2011 care.diabetesjournals.org
Glucose variability, plasma glucose, and HbA1c
For all patients pooled, there was no
effect modification of the respective GV
measures on the MPG-HbA1c relation-
ship. For type 2 diabetic patients, the im-
pact of effect modification from the
respective GV measures was far from sig-
nificant (Table 2). The number of patients
with a predefined SD is shown in Table 1
for all patients pooled and for the type 1
and type 2 diabetic patients separately.
CONCLUSIONS—This study demon-
strated a significant effect of GV, as
reflected by SD, on the MPG-HbA1c rela-
tionship. High GV (SD) is associated with
higher HbA1c levels for a given MPG, and
this effect wasmore pronounced at higher
HbA1c and MPG values. However, the
magnitude of this effect of GV was small
and only demonstrable in type 1 diabetic
patients. Possibly, the type 2 diabetic pa-
tient group was too small (n = 159) and
the variability in this group too low to find
this interaction.
The ADAG study showed a tight
correlation between HbA1c and MPG, al-
lowing the translation of HbA1c into esti-
mated average glucose (14,19). It has
been suggested that GV could affect the
MPG-HbA1c relationship, but this has not
previously been demonstrated (20–22).
To our knowledge, the current study is
the largest study reporting an influence
of GV—as expressed by SD, MAGE, and
CONGA4 calculated from CGM—on the
MPG-HbA1c relationship. The discrepan-
cies in the MPG-HbA1c relationship are
less likely caused by technical errors be-
cause this study included accurate and
centralized measurements of HbA1c val-
ues and intensively measured plasma glu-
cose concentrations (;2,700 values) in a
large and diverse population. Also, indi-
viduals with conditions or treatment that
might result in major changes in glycemia
or interference with the HbA1c assay, or
the MPG-HbA1c relationship, were ex-
cluded. These precautions allowed us to
search for factors other than MPG that
may contribute to HbA1c.
In general, GV is higher in patients
with poor glycemic control and in type 1
diabetic patients compared with type 2
diabetic patients, which can be attributed
to insulin therapy and higher insulin
sensitivity. High GV may affect glycation
because of periodic exposure of the eryth-
rocyte to high glucose levels and therefore
to faster irreversible glycation.
Other factors like hyperglycemia-
induced oxidative stress may affect the
glycation process. In recent literature, it
has been speculated that oxygen free rad-
icals per se or with an associated decrease
in antioxidants may modulate the forma-
tion of early glycated protein (10,11).
Brownlee (12) demonstrated that hy-
perglycemia stimulates oxidative stress.
High GV and especially postprandial glu-
cose excursions were also previously
Table 1—Baseline clinical and glycemic characteristics
All Type 1 diabetes Type 2 diabetes
n 427 268 159
Age (years) 47.6 6 13.6 44.1 6 12.9 56.6 6 9.4
Sex (% female) 54 52 51
Ethnicity (% non-Hispanic whites) 83 93 73
Current smokers 11 12 9
Insulin treatment 76 100 38
Glycemic measures
HbA1c (%) 6.8 6 1.3 7.3 6 1.1 6.8 6 1.1
MPG (mg/dL) 149.4 6 39.6 162 6 36 149.4 6 36
Measures of GV
CGM SD (mg/dL) 48.6 6 25.2 64.8 6 16.2 39.6 6 16.2
MAGE (mg/dL) 86.4 6 43.2 115.2 6 32.4 68.4 6 27
CONGA4 (mg/dL) 66.6 6 28.8 88.2 6 23.4 52.2 6 21.6
SD (mg/dL)
#30 61 (14.3) 9 (3.4) 52 (32.7)
,30–60 173 (40.5) 84 (31.3) 89 (56)
,60–90 173 (40.5) 155 (57.8) 18 (11.3)
.90 20 (4.7) 20 (7.5) 0 (0)
Data are means 6 SD, %, or n (%).
Table 2—The P values of the influence of the respective GV measures themselves, as well as effect modification and the D of
confounding, calculated from the respective slopes (B and B9) from the regression equations, on the HbA1c-MPG relationship for all
patients pooled and for type 1 and type 2 diabetic patients separately
Influence of the GV
measure (P)
Slope of MPG (B) in the
main regression formula
Slope of MPG (B9) in the regression
formula with the GV measure
DConfounding
(%)*
Effect
modification (P)
SD
All ,0.01 2.818 2.624 6.9 0.06
Type 1 diabetic 0.01 2.781 2.631 5.4 ,0.01
Type 2 diabetic 0.06 2.782 2.637 5.2 0.74
MAGE
All ,0.01 2.818 2.700 4.2 0.37
Type 2 diabetic 0.19 2.781 2.721 2.2 0.06
Type 2 diabetic 0.19 2.782 2.698 3.0 0.19
CONGA4
All ,0.01 2.818 2.667 5.4 0.15
Type 1 diabetic 0.06 2.781 2.687 3.4 0.02
Type 2 diabetic 0.07 2.782 2.661 4.3 0.46
*DConfounding in % = 100 3 absolute (B9 – B)/B.
care.diabetesjournals.org DIABETES CARE, VOLUME 34, AUGUST 2011 1845
Kuenen and Associates
associated with oxidative stress in type 2
diabetes (13). The activation of oxidative
stress, estimated from urinary excretion
rates of isoprostanes, was highly corre-
lated with MAGE calculated from CGM
(13). However, Wentholt et al. (23) could
not replicate these results in type 1 diabe-
tes. Recently, Ceriello et al. (15) demon-
strated that high intraday GV was more
damaging to endothelial function than
stable hyperglycemia and that oxidative
stress plays a key role. Whether oxidative
stress influences glycation needs to be
determined.
On the other hand, it has been dem-
onstrated that erythrocyte survival is
shorter at chronic high glucose concen-
trations levels, which might falsely lower
HbA1c levels. Peterson et al. (24) showed
that the life span of 51Cr-labeled eryth-
rocytes increased in all seven subjects
when their poorly controlled diabetes
was adequately treated. Virtue et al. (25)
concluded that there is a hyperglycemia-
related decrease in erythrocyte survival as
measured by carbon monoxide in the
expired air, which results in an exponen-
tial underestimation of the severity of
hyperglycemia at higher HbA1c levels.
Similarly, hyperglycemia-related osmotic
stress may influence erythrocyte perme-
ability and could cause damage to the
erythrocyte and shorten its life span.
These findings could lead to underesti-
mation of HbA1c at higher MPG levels,
concealing a glycemic control worse
than indicated by HbA1c measurements.
However, we found that type 1 diabetic
patients with high GV display higher
HbA1c levels than suspected based on
MPG. This effect was more pronounced
at higher HbA1c levels, indicating that fo-
cus on reducing GV, especially in patients
with poor glycemic control, could help
reduce HbA1c levels.
Limitations of our study are that CGM
has a limited range of reliable measure-
ments between 2.2 mmol/L and 22.2
mmol/L. Therefore, theoretically, CGM
performance could be less precise in
patients with high glycemic variability.
Furthermore, CGM has a lag time in
glucose values compared with the venous
measured values (the physiological gap),
and this can result in underestimation of
the influence of GV on the glycation of
HbA1c, and no measures of erythrocyte
survival, oxidative stress, or clinical fol-
low-up are available in this population.
In conclusion, at higher levels of GV
the relationship between HbA1c and MPG
in patients with type 1 diabetes is altered,
leading to a higher HbA1c level for a given
MPG.However, the impact (near theHbA1c
treatment target of 7%) is only modest.
Acknowledgments—The ADAG study is sup-
ported by research grants from the American
Diabetes Association an