Relationship Between A1C and Fasting Plasma Glucose in Dysglycemia or Type 2 Diabetes: An analysis of baseline data from the ORIGIN trial

The design, methods, and baseline characteristics of the ORIGIN trial have previously been published (13). Briefly, ORIGIN included participants with established type 2 diabetes (T2DM) (82%) or with newly diagnosed T2DM, impaired fasting glucose, or impaired glucose tolerance based on either FPG or an oral glucose tolerance test (18%). This analysis is restricted to 12,527 participants with both a baseline FPG and concurrent A1C measurement who were recruited in North America, South America, Europe, and Pacific Rim countries (Australia, Asia, and India) and who comprised white, Latin American, Asian, black, and other ethnicities. FPG (glucose dehydrogenase method) and A1C (high-performance liquid chromatography method) levels were determined in local laboratories around the world. Minor differences in A1C assays and their normal ranges across sites were accounted for by dividing the obtained result by the upper limit of the normal for the local assay and multiplying by 6%.

All statistical analyses were performed at the ORIGIN project office in the Population Health Research Institute using S-Plus version 8.1.1 (Tibco Software, Palo Alto, CA). Baseline characteristics were summarized as either counts (%) or mean (SD). Cubic B spline curves with 95% CIs were generated to show the relationship between the A1C (dependent variable on the ordinate) and FPG in all participants, participants not on an OAD, participants on any OAD, and participants on metformin, a sulfonylurea, or any other OAD. Similar spline curves were also generated for persons using and not using an OAD within each geographic region and each ethnicity.

FPG knots of 5.6 and 9.0 mmol/L that were selected by the statistical program for all participants were used for all of the analyses. These two knots identified three FPG intervals: <5.6, 5.6–9.0, and >9.0 mmol/L. Linear regression models were generated to characterize the FPG-A1C relationship within each of these three intervals for 1) all participants, 2) participants on an OAD, and 3) participants not on an OAD. The FPG-A1C relationship in each of these groups was therefore described by three regression models (i.e., one regression model per FPG interval). In each model, the dependent variable was A1C and the independent variables were FPG, the subgroups of interest, and terms for the interaction of FPG with the subgroups. Thus, the independent variables for each of the three equations generated for the entire group were FPG, OAD, and FPG × OAD, and the independent variables for those participants on an OAD included 1) each OAD category and interactions of FPG with each category, 2) each geographic region and the relevant interactions with FPG, and 3) each ethnicity and the relevant interactions with FPG. Similar models for each geographic region and ethnicity were generated for those participants not on an OAD.

The following approach was used to determine whether the FPG-A1C relationship was significantly affected by use versus nonuse of an OAD, ethnicity, and/or geographic region. First, the interaction terms for each model were tested for statistical significance. As each FPG-A1C relationship was described by three models, the cut point for significance was set at P = 0.017 (i.e., one-third of 0.05). If the interaction term in any of the three models described a relationship that was significant at P < 0.017, it was inferred that the FPG-A1C relationship was significantly affected by the subgroup being tested and that different estimates of the FPG-A1C relationship are appropriate for persons in that subgroup versus those not in that subgroup. When that occurred, the difference in the estimated A1C in persons in and not in that subgroup was estimated from the respective spline curves and plotted.

Participants’ mean (SD) age was 63.6 (7.8) years. As noted in Table 1, 8,146 (65%) were men, 7,390 (59%) were on an OAD (comprising 27% on metformin and 30% on a sulfonylurea), 6,067 (48%) were from Europe, and 7,395 (59%) were white. The mean BMI was 29.8 (5.2) kg/m²; the mean FPG in people not on an OAD and on an OAD was 6.9 (1.6) and 7.6 (2.2) mmol/L, respectively (P < 0.0001); and the mean A1C in these two groups was 6.3% (45 mmol/mol) (0.9) and 6.7% (50 mmol/mol) (0.9), respectively (P < 0.0001).

Inspection of the spline curves illustrating the relationship between FPG and A1C for the entire group (Fig. 1A) showed that the shape of the relationship is consistent with the two knots of 5.6 and 9.0 mmol/L identified by the statistics program. It also suggested that the relationship differed depending on whether the participant was taking an OAD (Fig. 1B) or was not taking an OAD (Fig. 1C). This possibility was tested by inspecting the interaction terms for the three linear regression models that described each of the three segments of the FPG-A1C relationship (FPG <5.6, 5.6–9.0, and >9 mmol/L) in all participants (Table 2). Indeed, as the interaction term was significant (at P < 0.017) for two of these models (P < 0.0001 and P = 0.004 for FPG 5.6–9 and >9 mmol/L, respectively), OAD use was clearly a significant determinant of the FPG-A1C relationship in this population (due to the differences in the relationship for FPG levels ≥5.6 mmol/L).

At FPG levels of 5.6–9.0 mmol/L and without OAD use, for every 1.0 mmol/L higher FPG, A1C increased by an average of 0.43%. (β coefficient 0.430 [95% CI 0.402–0.458], P < 0.0001). With OAD use, the corresponding average increase in A1C was 0.26% (0.255 [0.230–0.280], P < 0.0001). At FPG levels of >9.0 mmol/L without OAD use, the average increment in A1C was 0.21% (0.211 [0.135–0.286], P < 0.0001), and with OAD use, the average increment was 0.10% for every millimole per liter rise in FPG (0.097 [0.063–0.132], P < 0.0001). The impact of this interaction for FPG levels >5.6 mmol/L is illustrated in Fig. 2, which shows that the difference in the estimated A1C levels for people on an OAD versus not on an OAD was greatest for FPG levels >9 and <8 mmol/L. This can be appreciated by noting that the predicted FPG for participants with an A1C of 6.5% (48 mmol/mol) would be 6.95 mmol/L (95% CI 4.0–12.0) with OAD use and 7.5 mmol/L (6.6–8.6) without OAD use.

In all subsequent analyses, performed separately for people on and not on an OAD, no significant interactions were seen with respect to type of OAD used (in people on an OAD [Fig. 1D–F]), geographic location (Fig. 3), or ethnicity (Supplementary Data Fig. 4).

These analyses of cross-sectional data from >12,500 people with impaired fasting glucose, impaired glucose tolerance, or early diabetes who volunteered for the international ORIGIN trial demonstrate a clear graded relationship between FPG ≥5.6 mmol/L and A1C levels. They also show that the nature of the relationship differs depending on whether an individual is taking an OAD but is unaffected by either ethnic origin or geographical location. Scrutiny of the FPG-A1C relationship suggests several insights from these observations.

Both inspection and statistical analysis of the curve for FPG versus A1C levels in the whole population (Fig. 1A) revealed a complex relationship, with a steeper, relatively linear part of the curve between FPG 5.6 and 9.0 mmol/L. At levels <5.6 mmol/L, no statistically significant relationships between FPG and A1C were found. At levels >9.0 mmol/L, the slope of the curve was less steep, demonstrating a lesser increase of A1C with higher FPG levels. While speculative, potential explanations for alteration of the relationship between FPG and A1C when FPG is >9.0 mmol/L can be proposed. One may be blunting of postprandial peaks by glycosuria when they are above the renal “threshold” for reabsorption, which is typically ≥9 mmol/L (14). Also, erythrocyte survival may be shortened during marked hyperglycemia (15,16).

The relationship between FPG and A1C is particularly relevant for FPG levels between 5.6 and 9.0 mmol/L, as patients with diabetes typically have FPG in this range. For persons not taking any OAD, the slope of the relationship indicated a 0.43% higher A1C for each 1 mmol/L of FPG concentration. For individuals taking an OAD, this relationship was significantly blunted, with a 0.25% higher A1C for every 1 mmol/L higher FPG within this range. The linear relationship between FPG and A1C in both these subgroups (albeit with different slopes) and the fact that the A1C measurement is known to reflect both fasting and nonfasting hyperglycemia suggest that fasting and nonfasting glucose levels generally change in parallel within this range.

Why the FPG-A1C relationship was altered when OADs were used is unknown. One possible explanation is that both metformin and sulfonylureas (the most commonly used OADs) lower FPG more than nonfasting glucose levels (17–19). Another is that drug-naïve individuals are less inclined to attempt lifestyle interventions or less likely to be checking their glucose levels because they either do not have a diagnosis of diabetes or are not taking a drug that can affect glycemic patterns. Whatever the mechanisms for the effect of OAD therapy, both clinical management and interpretation of epidemiologic data may be illuminated by this alteration of the FPG-A1C relationship. In contrast, the absence of differences in the FPG-A1C relationship across different populations suggests that FPG levels are much more strongly related to A1C levels than to any variables associated with either geography or ethnicity, whether genetic or behavioral in origin.

The strength of the study was the large number of subjects available for the overall and subgroup analyses. The numbers were high for subgroup analysis in relation to the use of OAD. The large sample size with a wide range of A1C and FPG levels facilitated analysis of data at clinically relevant A1C levels. Important limitations include the fact that participants were all volunteers and were all deemed to be at high risk for cardiovascular disease. Whether similar relationships would be apparent in individuals from the general population at lower risk for cardiovascular disease is unknown, although there is no compelling reason to believe that they would be different. These observations are also limited by the fact that glucose and A1C values were locally measured using a variety of assays worldwide and that a mathematical approach was used to adjust for differences in each laboratory’s normal range for A1C. Thus, laboratory-related variability may have influenced the results. Also, information on the presence of variant hemoglobins or other physiologic factors that might affect A1C values was not available, and whether such factors may have altered the FPG-A1C relationships observed is unknown. However, the fact that clear FPG-A1C relationships were observed despite these potential sources of variability strengthens the conclusions and suggests that a stronger relationship would be apparent if the measurements were all made in the same laboratory.

In summary, our cross-sectional analyses in a large population of persons with dysglycemia or early T2DM confirm a strong relationship between FPG in the range of 5.6–9.0 mmol/L and concurrent A1C measurements across different geographic regions and ethnic groups. They also show a clear effect of oral therapies on this relationship, with a smaller increase in A1C per unit increase in FPG for persons taking an OAD. These data highlight the relevance and appropriateness of the A1C test as a useful measure of glycemic status throughout the world and also suggest that OAD use or nonuse should be considered when interpreting A1C level.

The ORIGIN trial is funded by sanofi-aventis. No other potential conflicts of interest relevant to this article were reported.

A.R. and M.C.R. contributed to, researched, analyzed, discussed, and prepared the manuscript. C.K. contributed to, researched, analyzed, and performed the statistical analyses. H.C.G. contributed to, researched, analyzed, discussed, and prepared the manuscript. A.R. and H.C.G. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

A.R. acknowledges the help provided by Dr. C. Snehalatha, Director, Research in India Diabetes Research Foundation, in the preparation of the manuscript.