McCarter et al. (1) report that the propensity to glycate hemoglobin, as evidenced by a higher-than-expected HbA1c level for the concomitantly measured mean blood glucose (MBG) level, is itself a risk factor for diabetic retinopathy and nephropathy. They use publicly available data from the Diabetes Control and Complications Trial (DCCT) to calculate a hemoglobin glycation index (HGI) to support their hypothesis. Their results appear to offer a possible explanation for anecdotal reports that some patients with chronically excellent control are nonetheless severely affected by diabetic retinopathy and/or nephropathy, while others with chronically poor glycemic control nonetheless escape these complications.
The data from the National Institute of Diabetes and Digestive and Kidney Diseases–funded DCCT were placed in the public domain to allow other investigators to pursue additional analyses, and we applaud their efforts. However, we the principal investigators of the DCCT cannot accept their conclusion based on the results presented.
The HGI used in these analyses is heavily dependent on the observed HbA1c. The authors regress HbA1c on MBG and other factors and then define the HGI as the residual: HGI equals observed HbA1c minus predicted HbA1c from the regression equation. However, the residual is not independent of the HbA1c. When HbA1c is regressed on the residual, the slope is 1 (a statistical fact whenever the dependent variable is regressed on the residual). Thus, the higher the residual or HGI, the higher the HbA1c. HGI is then simply a surrogate for HbA1c. This is a statistical tautology. Their analysis (their Fig. 2) showing that the risk of diabetic retinopathy progression is directly related to HGI is therefore certainly a function of HbA1c.
The question that is not addressed is whether HGI is a risk factor for diabetic retinopathy and nephropathy independent of the observed HbA1c. In other words, is the difference between the HGI groups in diabetic retinopathy or nephropathy risk attributable in whole or in part to the differences in HbA1c between these groups? This would be addressed by examining the differences between HGI groups after also adjusting for the levels of HbA1c. This critical analysis is not presented.
Related questions are whether HGI explains more of the variation in diabetic retinopathy and nephropathy risk than the observed HbA1c and whether HGI is a better and more useful predictor of risk than HbA1c alone.
In the authors’ methods report (2) they developed the HGI from numerous self-monitored blood glucose values over the preceding 30 days. However, the only blood glucose values available during the DCCT were the values obtained immediately before the time of the HbA1c collection. HbA1c is an index of MBG over the preceding 4–12 weeks and not necessarily of the MBG determined on the preceding day. For example, a patient with chronically elevated blood glucose levels and a high HbA1c might try hard to have lower blood glucose levels on the test day. This would explain Fig. 3A, where subjects with a low calculated MBG but high HGI (meaning high HbA1c) are at greater risk for retinopathy. Thus, the regression relationship between the concurrent MBG and HBA1c would be expected to be less than that between the HbA1c and the preceding 4–12 weeks of blood glucose values, data which were not obtained in the DCCT. Accordingly, a better regression estimate and a more accurate HGI index would be obtained from a regression line generated from each individual DCCT patient’s mean DCCT HbA1c versus the mean of each patient’s quarterly MBGs over the entire DCCT follow-up.
It is conceivable that a greater tendency to glycate hemoglobin and other proteins adds to the risk of complications above that created by the level of hyperglycemia per se. However, this hypothesis has not been addressed in this report. Moreover, the potential for sampling errors in measuring MBG and, consequently, the poor appreciation of the relationship between MBG and HbA1c remains a major impediment in understanding whether “biologic” variations in glycation truly exist. Until these issues are resolved, the observed HbA1c level still remains the easiest obtained and best proven clinical predictor of the risk of microvascular and neuropathic complications (3).