Clinical Chemistry 53:5

897–901 (2007) Endocrinology and

Metabolism

Variability in the Relationship between Mean

Plasma Glucose and HbA1c: Implications for the
Assessment of Glycemic Control
Eric S. Kilpatrick,1* Alan S. Rigby,2 and Stephen L. Atkin3

Background: Previous studies have shown a single sically linked to both increased hypoglycemia and de-
linear relationship between mean plasma glucose creased microvascular complications compared with
(MPG) and hemoglobin A1c (HbA1c). We examined the conventional treatment. These findings may also have
relationship in different treatment groups of patients implications for expressing HbA1c as mean blood glu-
with type 1 diabetes participating in the Diabetes Con- cose equivalent.
trol and Complications Trial (DCCT). © 2007 American Association for Clinical Chemistry
Methods: Seven-point glucose profiles (premeal, post-
meal, and bedtime) and HbA1c were measured quarterly The role of hemoglobin A1c (HbA1c)4 in the assessment of
during the DCCT. We studied measurements from (a) glycemic control in patients with diabetes has been ce-
intensively treated patients at study commencement, (b) mented by the results of the Diabetes Control and Com-
intensively treated patients after stabilization of their plications Trial (DCCT) and the United Kingdom Prospec-
glycemia (from 6 months onward), and (c) convention- tive Diabetes Study (1, 2 ). These studies showed that
ally treated patients from 6 months onward. Only com- HbA1c is an important marker in assessing a patient’s risk
plete glucose profile and HbA1c pairings were consid- of microvascular complications and hypoglycemia. As a
ered (n ⴝ 589, 11 483, and 11 855, respectively). result, both HbA1c and blood glucose targets are now
Results: From 6 months into the trial, conventionally used in the routine management of patients with type 1
treated patients had consistently higher MPG concen- and type 2 diabetes (3 ).
trations than intensively treated patients at any given A linear relationship between HbA1c and mean plasma
HbA1c value (mean difference, 1.6 mmol/L at 7% HbA1c, glucose (MPG) was first established in 1984 (4 ) and
increasing to 2.8 mmol/L at 11% HbA1c). Similarly, at the corroborated by the feasibility study of the DCCT (5 ).
same HbA1c, the MPG of intensively treated patients at More recently, data from the full DCCT trial were used to
baseline was higher than in the same individuals after 6 compare the average of every 7-point glucose day profile
months of intensive treatment (1.2 mmol/L difference at measured with the mean HbA1c of each participant (6 ).
7% HbA1c, increasing to 4.6 mmol/L at 11% HbA1c). From this, a linear equation between the 2 measures was
Conclusions: The relationship between MPG and derived, allowing patients and healthcare workers to set
HbA1c is not constant but differs depending on the plasma glucose targets based on their HbA1c goals. In-
glycemic control of the population being studied. Hav- deed, it has been suggested that this relationship be used
ing lower mean glucose at the same HbA1c may help to express HbA1c as mean blood glucose equivalent so
explain why intensive DCCT treatment appeared intrin- that patients can equate the test to their own self-moni-
tored blood glucose records (7–9 ).
In the DCCT itself, further analysis of the data seemed
1
Department of Clinical Biochemistry, Hull Royal Infirmary, Hull, United to suggest that HbA1c predicted the risk of hypoglycemia
Kingdom. and microvascular complications differently between the
2
Academic Department of Cardiology, University of Hull, Hull, United intensively and conventionally treated patients, such that
Kingdom.
3
Department of Diabetes, Hull York Medical School, Hull, United King-
intensive treatment was associated with a greater risk of
dom.
* Address correspondence to this author at: Department of Clinical Bio-
chemistry, Hull Royal Infirmary, Anlaby Rd., Hull HU3 2JZ, United Kingdom.
4
Fax 441482-607752; e-mail [Link]@[Link]. Nonstandard abbreviations: HbA1c, hemoglobin A1c; DCCT, Diabetes
Received September 7, 2006; accepted February 26, 2007. Control and Complications Trial; MPG, mean plasma glucose; AUC, area
Previously published online at DOI: 10.1373/clinchem.2006.079756 under the curve.

897
898 Kilpatrick et al.: Variability Relating Mean Glucose to HbA1c

hypoglycemia (10 ) but a decreased risk of small vessel sively treated group reached a nadir 6 months into the
disease (11 ) for any given HbA1c value. The reasons for trial (1 ), so only data from 6 months onward were
this observation remain speculative but are thought to considered for both groups. We compared mean HbA1c in
relate to the treatment regimens used by the 2 treatment each patient with their average within-day MPG. We also
groups. studied the relationship between MPG and HbA1c in the
We reanalyzed the DCCT dataset to establish the intensively and conventionally treated groups at their first
relationship between MPG and HbA1c for each of the visit before randomization (baseline).
treatment groups to see whether this may help explain We analyzed the data using least-squares regression
why the rate of complications differed between them. analysis weighted for the number of MPG/HbA1c pair-
ings in each individual. Residuals were checked for gaus-
Patients and Methods sian distribution by plotting a histogram. Unpaired t-tests
datasets were used to compare the MPG of patient groups at the
We used the publicly accessible datasets stored in SAS same band of HbA1c, and the same test was used to
format collected by the DCCT ([Link]). The compare the slopes of regression lines. Paired t-tests were
DCCT was a 9-year follow-up study of 1441 participants used to establish differences between 7-point and 8-point
with type 1 diabetes comparing the effect of intensive vs glucose profiles.
conventional blood glucose management on the develop-
ment of microvascular complications. Study participants Results
were randomized into intensive (n ⫽ 711) and conven- Seven hundred seven intensively treated and 726 conven-
tional (n ⫽ 730) treatment groups. tionally treated patients had 11 483 and 11 855 complete
(7-point) MPG and HbA1c pairings collected from 6
glycemic variables and statistical methods months into the study. At baseline, complete glucose
Capillary blood glucose samples were shipped on dry ice profiles and HbA1c measurements were present in 589
to the DCCT Central Biochemistry Laboratory and ana- participants who were subsequently randomized to inten-
lyzed by use of a hexokinase method (12 ). HbA1c samples sive treatment and 585 individuals who were assigned to
were express-transported at 4 °C to the Central HbA1c conventional treatment. There were no significant differ-
Laboratory, where samples were originally analyzed on a ences in age (P ⫽ 0.46), duration of diabetes (P ⫽ 0.41), sex
HPLC column (Biorex 70; Bio-Rad); within- and between- (P ⫽ 0.30), or phase of treatment (P ⫽ 0.90) between the
assay CVs for this HbA1c method were ⬍6%. From 589 intensively treated patients at baseline and the 122
November 1986 onward, a Diamat HPLC instrument who did not have a complete glucose profile at baseline.
(Bio-Rad) was used, with which both CVs were consis- The MPG values of these 4 groups categorized accord-
tently ⬍3% (13 ). ing to their HbA1c are shown in Table 1. After stabilization
In the DCCT, a blood glucose profile was taken at of glycemia, the intensively treated group had lower MPG
3-month intervals. Blood glucose was assessed at 7 points values than both the conventionally treated group after
throughout the day: prebreakfast (we assumed a time of stabilization and the intensively treated group at baseline
7:00 AM), postbreakfast (8:30 AM), prelunch (12:00 AM), (P ⬍0.05 from 6% to 10.4% HbA1c, P ⬍0.0001 from 6.5% to
postlunch (1:30 PM), presupper (6:00 PM), postsupper 8.4% comparing stabilized intensive patients with both
(7:30 PM), and bedtime (10:00 PM). An additional data groups). Only 1 of 15 comparisons between the conven-
point was collected at 3:00 AM on ⬍1% of occasions and tionally treated group at baseline and the same group
was used here only to validate that an absence of over- after stabilization showed any significant difference in
night measurement was not influencing our findings. In MPG (P ⫽ 0.043 at 9% to 9.4% HbA1c).
addition, not everybody had a complete 7-point blood Fig. 1 shows the relationship between MPG and HbA1c
glucose profile during every quarter of their participation, using regression in each of the treatment groups and for
so we restricted the data for analysis to the time points both groups combined after stabilization (6 months on-
where there was a complete glucose profile and HbA1c ward). The slope of the conventionally treated group [1.54
pairing. Mean blood glucose was calculated by the area mmol/L (95% CI, 1.42–1.66) increase in MPG for every 1%
under the curve (AUC) method using the trapezoidal rule increase in HbA1c] was significantly greater than in the
(14 ). This was performed as by Rohlfing et al. (6 ), except intensively treated patients [1.23 mmol/L (1.12–1.33), P
that rather than assuming a constant blood glucose con- ⬍0.001]. The regression relationship of the combined
centration between bedtime and the following morning, group was skewed by the larger number of intensively
we assumed the following morning glucose value to be treated patients with low HbA1c values and by the larger
the same as that of the previous morning. The same 11% number of conventionally treated patients with higher
increase in mean blood glucose values was made to values, so the slope in this case was greater again [1.87
express results as an MPG equivalent (15 ). mmol/L (1.81–1.94), P ⬍0.001 compared with both indi-
The relationship between MPG and HbA1c was estab- vidual groups]. Weighting the regression by instead using
lished for each of the treatment groups (conventional and the reciprocal of each patient’s HbA1c variance provided
intensive) separately and combined. HbA1c in the inten- very similar results.
 Clinical Chemistry 53, No. 5, 2007 899

Table 1. MPG values at different mean HbA1c values.a

Conventional from 6
Intensive from 6 months months Intensive at baseline Conventional at baseline

MPG, MPG, MPG, MPG,

HbA1c, % mmol/L n mmol/L n mmol/L n mmol/L n
5–5.4 7.9 (1.4) 3 0 0 12.0 1
5.5–5.9 8.2 (0.8) 29 8.9 (1.0) 2 6.5 1 10.3 (3.5) 3
6–6.4 8.0 (1.1) 110 9.2 (1.4) 5 10.0 (2.0) 12 8.5 (2.9) 19
6.5–6.9 8.8 (1.0) 183 9.8 (1.8) 26 9.8 (2.7) 41 10.5 (2.7) 41
7–7.4 9.4 (1.4) 173 11.3 (1.9) 38 11.5 (3.3) 56 11.3 (2.9) 37
7.5–7.9 9.9 (1.4) 101 12.0 (1.8) 72 12.0 (4.1) 68 11.5 (3.0) 67
8–8.4 11.0 (1.6) 55 12.6 (1.9) 103 12.8 (3.6) 93 13.1 (3.4) 95
8.5–8.9 11.0 (2.3) 24 13.3 (2.1) 119 13.5 (3.6) 73 13.6 (3.8) 70
9–9.4 11.6 (1.4) 11 13.7 (2.0) 96 14.3 (3.8) 71 14.6 (3.6) 67
9.5–9.9 12.2 (2.4) 8 15.1 (2.6) 86 15.5 (4.1) 44 15.6 (3.4) 48
10–10.4 12.8 (2.8) 5 16.4 (2.1) 81 17.2 (3.3) 54 15.8 (4.1) 41
10.5–10.9 0 16.9 (3.0) 43 16.9 (4.5) 16 17.4 (4.1) 26
11–11.4 14.2 (1.7) 5 16.7 (3.2) 21 18.5 (5.2) 24 17.4 (4.1) 29
11.5–11.9 18.3 (3.8) 17 20.6 (4.2) 11 20.0 (5.2) 16
12–12.4 17.6 (3.6) 8 21.9 (2.6) 10 19.6 (5.1) 14
12.5–12.9 17.3 (2.7) 8 24.1 (5.4) 5 17.3 (5.0) 2
13–13.4 0 24.5 (2.5) 3 24.7 (4.4) 4
13.5–13.9 24.0 1 28.1 (3.5) 4 0
14–14.4 24.4 (7.9) 3 26.8 (1.7) 2
14.5–14.9 24.0 (11.6) 3
a
Data are mean (SD).

Nearly identical findings were also obtained when ing as per Rohlfing et al. (6 ) [slope of conventionally
MPG was calculated using a constant blood glucose treated group, 1.69 (1.56 –1.83); intensively treated group,
concentration between bedtime and the following morn- 1.21 (1.09 –1.33); combined, 1.97 (1.89 –2.04); P ⬍0.001
between all 3 groups].
From 6 months into the study, there were 1361 glucose
points collected at 3:00 AM, 1068 of which were at the
6-month visit itself (528 intensive and 540 conventional
treatment). Using these 6-month data, we calculated each
patient’s 8-point AUC and compared this with what their
7-point profile would have shown had the 3:00 AM
sample not been collected. The 8-point AUC was, on
average, 0.06 mmol/L greater than when just 7 points
were used (P ⫽ 0.01). In the intensive group alone, the
8-point AUC was 0.02 mmol/L lower than the 7-point,
and in the conventional group it was 0.13 mmol/L greater
(P ⬍0.001, intensive vs conventional).

Discussion
This study shows that the relationship between MPG and
HbA1c in the DCCT was different between intensively and
conventionally treated patients such that the former
group had clinically significantly lower MPG concentra-
tions at the same HbA1c values. The difference became
greater as the HbA1c increased, being 1.6 mmol/L at 7%
HbA1c and 2.8 mmol/L at 11% HbA1c. This observation
Fig. 1. Relationship between MPG and HbA1c in conventionally treated
does not seem to be a consequence of any differences
(MPG ⫽ 1.54 ⫻ HbA1c ⫺ 0.10; r ⫽ 0.69; - - - - -), intensively treated
(MPG ⫽ 1.23 ⫻ HbA1c ⫹ 0.47; r ⫽ 0.67; 䡠 䡠 䡠 䡠 䡠), and both patient between treatment groups, as a very similar finding was
groups combined (MPG ⫽ 1.87 ⫻ HbA1c ⫺ 3.61; r ⫽ 0.82; ——) after identified among just the intensively treated patients
stabilization. All P ⬍0.0001. when their glycemic control was improved.
900 Kilpatrick et al.: Variability Relating Mean Glucose to HbA1c

A plausible explanation can be found for this differ- These findings have several clinical implications, the
ence if the concept of high and low glycators is assumed first of which relates to our current interpretation of the
(16 ). High glycators have consistently higher HbA1c than DCCT itself. In that trial, the major problem associated
expected for their MPG, whereas low glycators have with intensive treatment was the greater risk of hypogly-
lower HbA1c than their MPG would suggest. The pro- cemia compared with conventionally treated patients
posed reasons for this between-individual variability in with the same HbA1c (10 ). In showing that intensively
hemoglobin glycation rate include differences in erythro- treated patients had consistently lower MPG than conven-
cyte survival and other genetic elements (17, 18 ). In tionally treated ones at the same HbA1c (equivalent to an
relation to the DCCT, the majority of patients at study HbA1c difference of 0.75% to 1.5%), this study is likely to
baseline had HbA1c values between 8% and 10%. A have at least partly explained the discrepancy in risk
proportion of those at 10% probably found it difficult to between the patient groups.
achieve a lower HbA1c because they were high glycators Intensive treatment was also associated with an unex-
and so had HbA1c values that were more pessimistic than plained decrease in the rate of retinopathy progression at
their MPG values. By comparison, more of those starting the same HbA1c value (11 ). For example, an intensively
at 8% were probably able to achieve the lower HbA1c treated patient with 9% HbA1c had an event rate similar to
because they were low glycators. Conventionally treated that of a conventionally treated patient at 8% HbA1c.
patients continued at these same HbA1c levels throughout Increased glucose variability in the latter group has been
the study but if the HbA1c was to improve by 2%, as postulated as the cause (20 ), although a recent analysis of
happened with intensive treatment, then the larger pro- DCCT data seems to discount this as a possibility (21 ).
portion of high glycators who started at 10% would Again, however, the lower MPG in intensively treated
remain being high glycators but would now have the patients at the same HbA1c provides a simple alternative
same HbA1c (8%) as the conventionally treated popula- explanation.
tion, with more low glycators at this HbA1c. This means Our data analysis could also be of relevance in the
that high glycators who improve their glycemic control current debate on how to report standardized IFCC
end up being compared with more low glycators, so at HbA1c values. Some advocate continuing to report IFCC
every chosen HbA1c level they would, on average, have HbA1c using existing DCCT numbers; others advocate
the lower MPG that this study has shown. expressing HbA1c as an average blood glucose equivalent
An alternative explanation posits that a component of (7, 8 ). The latter is seen as a preferred option (9 ), but our
DCCT-measured HbA1c exists that is not only unrelated study suggests doubt about what constitutes the average
to glycation but also variable in concentration between blood glucose for any given HbA1c value, given the fact
individuals. The term high glycator would then be a that this can change substantially between (and even
misnomer, because these individuals would simply have within) groups of patients. Researchers in the ongoing
more of this component than a low glycator. Certainly, it prospective study organized by the American Diabetes
is accepted that the original DCCT assay is not entirely Association/European Association for the Study of Dia-
specific for HbA1c because HbA1c values are 1.5% to 2% betes/International Diabetes Federation Working Group
lower using the standardized IFCC method (19 ). What is for the HbA1c Assay to further clarify the relationship
not so clear is whether this difference in HbA1c values between MPG and HbA1c (Mean Blood Glucose Study)
between the 2 assays is constant between individuals, (9 ) need to be aware that the overall glycemia of patients
despite there being a good overall correlation between the chosen for the study could have a major influence on the
2 (19 ). If this non-HbA1c component was indeed the cause outcome.
for our findings, it might be expected that future studies Last, our findings may help physicians understand
will show that MPG relates more closely to the IFCC why many intensively managed patients seem to have
standardized HbA1c method but not necessarily to any broadly acceptable self-monitored glucose records but
routine (possibly less specific) HbA1c method calibrated comparatively high HbA1c results. Rather than question-
from it. ing the monitoring technique (or honesty) of their pa-
A previous analysis of the DCCT data to establish the tients, the MPG values in Table 1 show that in intensively
relationship between MPG and HbA1c combined data treated patients the increase in mean glucose (1.23
from both treatment groups into 1 (6 ). In so doing, as mmol/L) per percentage HbA1c is not nearly as large as
shown in Fig. 1, this may have overestimated the slope of has been previously assumed. Thus, many may be close to
the regression equation, i.e., the increase in MPG for a their glucose target, although they appear some way from
given increase in HbA1c. In our analysis, the slope showed their HbA1c goal.
a 1.23 mmol/L increase in plasma glucose for every 1% The DCCT dataset is not without its limitations. Al-
increase in HbA1c in the intensively treated population, though more than 163 000 laboratory-measured glucose
which was significantly different from the 1.54 mmol/L samples were used in this analysis, newer techniques such
increase in the conventionally treated group, which in as a continuous glucose monitoring system could give a
turn was significantly lower than the 1.87 mmol/L in- clearer indication of glycemia, especially overnight. How-
crease when the 2 groups were combined. ever, it is reassuring that when glucose samples were
 Clinical Chemistry 53, No. 5, 2007 901

collected at 3:00 AM they showed that the 7-point MPG 6. Rohlfing CL, Wiedmeyer H-M, Little RR, England JD, Tennill A,
was just 0.06 mmol/L different from the 8-point MPG. In Goldstein DE. Defining the relationship between plasma glucose
fact, the differences between 7- and 8-point measurement and HbA1c: analysis of glucose profiles and HbA1c in the Diabetes
Control and Complications Trial. Diabetes Care 2002;25:275– 8.
in the 2 DCCT treatment groups indicated that our
7. Home P, Mbanya J-C, Horton E. Standardisation of glycated
findings would be further reinforced if an overnight haemoglobin. BMJ 2004;329:1196 –7.
sample had been collected on every occasion. It is also 8. Report of the ADA/EASD/IDF. Working Group of the HbA1c Assay,
reassuring that preliminary data from the Mean Blood London, UK, January 2004. Diabetologia 2004;47:R53– 4.
Glucose Study have demonstrated a remarkably good 9. Sacks DB, for the ADA/EASD/IDF. Working Group of the HbA1c
correlation (r ⫽ 0.94) between continuous glucose moni- Assay: Global harmonization of hemoglobin A1c. Clin Chem 2005;
toring system and 8-point AUCs (22 ). 51:681–3.
10. The Diabetes Control and Complications Trial Research Group.
Hypoglycemia in the Diabetes Control and Complications Trial.
In summary, this study has shown that the relationship Diabetes 1997;46:271– 86.
between MPG and HbA1c can differ substantially depend- 11. The Diabetes Control and Complications Trial Research Group.
ing on the glycemic control of the population studied. The The relationship of glycemic exposure (HbA1c) to the risk of
lower mean glucose at any given HbA1c value in inten- development and progression of retinopathy in the diabetes
sively treated DCCT patients may go some way toward control and complications trial. Diabetes 1995;44:968 – 83.
12. The Diabetes Control and Complications Trial Research Group.
explaining why they appeared more prone to hypoglyce-
The Diabetes Control and Complications Trial (DCCT). Design and
mia and less likely to develop microvascular complica- methodologic considerations for the feasibility phase. Diabetes
tions when HbA1c was used as the sole marker of 1986;35:530 – 45.
glycemia. Researchers investigating the feasibility of ex- 13. Steffes M, Cleary P, Goldstein D, Little R, Wiedmeyer H-M,
pressing HbA1c as an MPG equivalent need to be aware of Rohlfing C, et al. Hemoglobin A1c measurements over nearly two
this finding, as do healthcare staff who are currently decades: sustaining comparable values throughout the Diabetes
relating MPG to HbA1c in their clinic patients. Control and Complications Trial and the Epidemiology of Diabetes
Interventions and Complications Study. Clin Chem 2005;51:
753– 8.
14. Tai MM. A mathematical model for the determination of total area
Grant/funding support: None declared. under glucose tolerance and other metabolic curves. Diabetes
Financial disclosures: None declared. Care 1994;17:152– 4.
Acknowledgements: We thank the DCCT investigators for 15. Fogh-Andersen N, D’Orazio P. Proposal for standardizing direct-
making their trial dataset public and therefore allowing reading biosensors for blood glucose. Clin Chem 1998;44:655–9.
independent investigators to analyze their work for the 16. Yudkin JS, Forrest RD, Jackson CA, Ryle AJ, Davie S, Gould BJ.
Unexplained variability of glycated haemoglobin in non-diabetic
benefit of patients with type 1 diabetes.
subjects not related to glycaemia. Diabetologia 1990;33:208 –
15.
References 17. Virtue MA, Furne JK, Nuttall FQ, Levitt MD. Relationship between
1. The Diabetes Control and Complications Trial Research Group. GHb concentration and erythrocyte survival determined from
The effect of intensive treatment of diabetes on the development breath carbon monoxide concentration. Diabetes Care 2004;27:
and progression of long-term complications in insulin-dependent 931–5.
diabetes mellitus. N Engl J Med 1993;329:977– 86. 18. Snieder H, Sawtell PA, Ross L, Walker J, Spector TD, Leslie RDG.
2. UK. Prospective Diabetes Study (UKPDS) Group. Intensive blood- HbA1c levels are genetically determined even in type 1 diabetes:
glucose control with sulphonylureas or insulin compared with evidence from healthy and diabetic twins. Diabetes 2001;50:
conventional treatment and risk of complications in patients with 2858 – 63.
type 2 diabetes (UKPDS 33). Lancet 1998;352:837–53. 19. Miedema K. Towards worldwide standardisation of HbA1c deter-
3. American Diabetes Association: Standards of medical care in mination. Diabetologia 2004;47:1143– 8.
diabetes—2006. Diabetes Care 2006;29:S4 – 42. 20. Ceriello A. Postprandial hyperglycemia and diabetes complica-
4. Nathan DM, Singer DE, Hurxthal K, Goodson JD. The clinical tions: is it time to treat. Diabetes 2005;54:1–7.
information value of the glycosylated hemoglobin assay. N Engl 21. Kilpatrick ES, Rigby AS, Atkin SL. The effect of glucose variability
J Med 1984;310:341– 6. on the risk of microvascular complications in type 1 diabetes.
5. The Diabetes Control and Complications Trial Research Group. Diabetes Care 2006;29:1486 –90.
Diabetes Control and Complications Trial (DCCT). Results of 22. Heine RJ. HbA1c-standardization and relationship to BG levels and
feasibility study. Diabetes Care 1987;10:1–19. complications. Diabetic Med 2006;23(Suppl 4):S17.