Defining the Relationship Between Plasma Glucose and HbA_1c: Analysis of glucose profiles and HbA_1c in the Diabetes Control and Complications Trial

The DCCT data set was provided by the National Institutes of Diabetes, Digestive, and Kidney Diseases of the National Institutes of Health and was prepared by the Data Coordinating Center at George Washington University. The DCCT was a multicenter, randomized clinical trial designed to compare intensive and conventional therapies and their relative effects on the development and progression of diabetic complications in patients with type 1 diabetes (1). The study population consisted of 1,441 patients with type 1 diabetes recruited by 29 centers located throughout the U.S. and Canada. Patients were between 13 and 39 years of age and did not show evidence of severe diabetic complications at the time of admission into the study. Intensive therapy consisted of three or more insulin injections daily or use of an insulin pump with the intent of achieving BG values as close to the normal range as possible. Conventional therapy consisted of one or two insulin injections per day. Mean duration of participation was 6.5 years (range 3–9 years).

Quarterly HbA_1c measurements (n = 37,058) and corresponding BG profiles were obtained from 1,441 subjects. After exclusions due to incomplete profiles, there were 26,056 HbA_1c values with corresponding seven-point profiles from 1,439 subjects (an average of 18 HbA_1c values and corresponding profiles per patient).

For the seven-point BG profiles, capillary blood hemolysates were collected before meals, 90 min after meals, and at bedtime by patients in the home (6). BG was measured in a central laboratory using a hexokinase enzymatic method (7). Blood for HbA_1c analysis was collected by venipuncture. HbA_1c was measured in a central laboratory using an ion-exchange high-performance liquid chromatography method (8,9).

Statistical analysis was performed using SAS and SPSS (Chicago, IL). Mean BG was determined using area-under-the-curve analysis (10). For each profile, the seven time points were connected by straight lines over time for a 24-h period, and then the trapezoidal areas under each curve were determined, added together, and divided by time. A constant BG level between bedtime and the following morning was assumed. Mean plasma glucose (MPG) was estimated by adding 11% to mean BG estimates (11). Mean MPG and HbA_1c were calculated for each subject and used to perform least-squares linear regression analysis. Due to variation in the number of observations per subject, the regression analysis was weighted to account for this. The relationships between individual PG time points and HbA_1c were also examined.

The results of linear regression analysis are summarized in Fig. 1. The Pearson correlation coefficient (r) was 0.82; change in MPG per increase of 1% HbA_1c was 1.98 mmol/l (35.6 mg/dl). The 95% prediction interval for a subject with 18 observations (the average number of profiles per patient in this study) was ±3.81 mmol/l (69 mg/dl) at levels of 6–9% HbA_1c. Within-subject (intraindividual) variation in HbA_1c was much lower than for seven-point PG (mean intraindividual coefficient of variation = 9.7 vs. 29.8%, respectively).

MPG at increasing levels of HbA_1c is shown in Table 1. Along with regression-estimated MPG, the table shows approximate MPG based on increments of 2 mmol/l or 35 mg/dl per 1% change in HbA_1c to facilitate clinical interpretation and use of these data.

Results of regression analyses correlating HbA_1c with individual premeal and postmeal PG are summarized in Figs. 2 and 3. All individual time points showed lower correlations than the seven-point profiles. Prelunch and earlier PG time points showed lower correlations with HbA_1c than postlunch and later PG time points.

The increasing use of HbA_1c to monitor long-term glycemic control in diabetic patients is largely the result of data from the DCCT and the U.K. Prospective Diabetes Study showing that HbA_1c is strongly correlated with adverse outcome risks. For patients and health care providers, a clear understanding of the relationship between PG and HbA_1c is necessary for setting appropriate day-to-day PG testing goals with the expectation of achieving specific HbA_1c targets.

The relationship between HbA_1c and PG is complex. Many studies have shown that HbA_1c is an index of MPG over the preceding weeks to months. Erythrocyte life span averages ∼120 days. The level of HbA_1c at any point in time is contributed to by all circulating erythrocytes, from the oldest (120 days old) to the youngest. However, recent PG levels (i.e., 3–4 weeks earlier) contribute considerably more to the level of HbA_1c than do long- past PG levels (i.e., 3–4 months earlier). Therefore, HbA_1c is a “weighted” average of BG levels during the preceding 120 days; PG levels in the preceding 30 days contribute ∼50% to the final result, and PG levels from 90–120 days earlier contribute only ∼10% (12,13). This explains why the level of HbA_1c can increase or decrease relatively quickly with large changes in PG; it does not take 120 days to detect a clinically meaningful change in HbA_1c after a change in MPG.

Another factor that complicates efforts to describe an accurate and precise relationship between PG and HbA_1c is that, for practical reasons, previous studies and our present study have attempted to define this relationship using a limited number of PG levels measured over a limited time period (in this case, 1 day every 3 months) to estimate HbA_1c. Short-term PG levels can fluctuate markedly, particularly in patients with type 1 diabetes; this can result in significant discrepancies when attempting to estimate HbA_1c based on a single PG measurement or even a series of measurements on a single day. In this study, the time between sampling also contributes to intraindividual variation, especially for PG. However, we have achieved greater certainty in our estimates of the relationship between PG and HbA_1c than was possible in previous studies by using a considerably larger number of patients and observations obtained over a longer period of time. The resulting strong correlation suggests that, although a single PG measurement or a single daily profile may not reliably predict HbA_1c, PG levels measured over time can provide a reasonably accurate estimation of HbA_1c.

Several studies have suggested that, although intraindividual variation in HbA_1c is minimal, there is evidence of wide fluctuations in HbA_1c between individuals that are unrelated to glycemic status, suggesting that there are “low glycators” and “high glycators” (14–16). However, a recent study showed that when multiple observations per patient are used to minimize the effects of assay variation, the interindividual range of HbA_1c results in nondiabetic individuals is actually quite narrow, <1% HbA_1c (17). Therefore, for any individual patient, a consistent discrepancy between patient-monitored PG determinations and estimated HbA_1c should be investigated; there may be other factors causing this discrepancy, such as improper meter use, laboratory error, a physical condition that alters red cell life span, or a variant hemoglobin interfering with the HbA_1c assay method. With the advent of new technologies that are capable of monitoring PG on a 24-h basis (18), it will be interesting to see how our estimate of the relationship between PG and HbA_1c compares with estimates obtained using these technologies.

Our data indicate that fasting PG alone should be used with caution as a measure of long-term glycemia. Fasting PG tended to progressively underestimate HbA_1c (and seven-point MPG) at increasing PG levels. The data also suggest that postmeal PG contributes appreciably to HbA_1c; however, all postmeal times are not equal in their contribution. We found that compared with the seven-point profiles, postbreakfast levels markedly overestimate HbA_1c, whereas postlunch levels show a relationship to HbA_1c that is very similar to that of MPG. A previous study of patients with type 2 diabetes also found that postlunch PG is a better indicator of glycemic control than fasting PG (19). However, that study did not examine bedtime PG, which we found also shows a relationship to HbA_1c that is very similar to that of MPG.

The ADA currently recommends that patients with diabetes attempt to achieve average preprandial PG levels of 5.0–7.2 mmol/l (90–130 mg/dl) and average bedtime PG levels of 6.1–8.3 mmol/l (110–150 mg/dl) as well as HbA_1c <7% (2). Our results show estimated average preprandial PG and bedtime PG levels of 8.7 and 9.2 mmol/l (157 and 166 mg/dl), respectively, at 7% HbA_1c. These data suggest that patients who consistently achieve ADA-recommended BG and PG targets will also achieve an HbA_1c level <7%.

In summary, there is a predictable relationship between PG and HbA_1c. Understanding this relationship will allow patients with diabetes and their healthcare providers set appropriate day-to-day PG targets based on HbA_1c goals. It is important to note that the relationship between PG and HbA_1c defined in this study only applies when HbA_1c is measured using assay methods that are certified by the National Glycohemoglobin Standardization Program as traceable to the DCCT reference method, as recommended by the ADA (20). Fasting PG should be used with caution as a surrogate measure of MPG because it may significantly underestimate HbA_1c and, therefore, risks for complications at increasing HbA_1c levels.

We thank the DCCT study group and the Data Coordinating Center at George Washington University for providing the data set as well as the patient volunteers who participated in the DCCT.