Our analysis (1) of the Diabetes Control and Complications Trial (DCCT) dataset showed no relationship between pre- and postprandial glucose variability (SD) and the risk of microvascular complications. Wilson, on behalf of the Diabetes Research in Children Network (DirecNet) Study Group (2), makes the point that continuous glucose monitoring can detect larger postprandial glucose excursions than the single-point measurements taken in the DCCT. We agree that it would seem appropriate that future studies should employ this technique.
As part of their analysis of glucose measurement in the DCCT, the article by Service and O’Brien (3) did indeed contain an analysis that found glucose variability to have the same lack of influence on retinopathy risk as ourselves, and for that they deserve credit. By specifically investigating the question of glucose variability, our article (1) was also able to assess the effect of variability both within and between each glucose profile, to apply this to nephropathy as well as retinopathy, and to adjust the data for possible confounders such as treatment group, age, sex, and duration of disease.
Our approach to missing data was different from that of Service and O’Brien (3). In their analyses, they excluded those with missing blood glucose values, accepting a bias that this might create. In contrast, we tried to take account of some of this data. We included all profiles with five or more observations during the 24-h period, assuming that a missing value lay on a straight line between the two surrounding data points. The compliance with glucose profiling using these criteria was extremely good, with a median of 91% of patients (range 84–97%) having such a profile during each quarter of our study period and a minimum of three- quarters of these being full seven point. We are aware that many methods can be used to extrapolate missing values for longitudinal data (4), with each method having its advantages and disadvantages depending on the setting (5).
Both Monnier et al. (6) and Service and O’Brien (7) are curious as to whether our findings would have been the same had we used the mean amplitude of glycemic excursions (MAGEs) assessment of variability rather than SD. We chose SD after undertaking preliminary work (not shown in our article) using a variety of methods for assessing variability. They all pointed to one thing: that variability in blood glucose was not related to microvascular risk after adjusting for mean blood glucose. We therefore only presented our SD data, citing a reference to a study in type 1 diabetes, which showed that SD was highly correlated with other measures of glucose variability (8).
From a statisticians viewpoint, we must admit to having reservations about the use of MAGE. The 1 SD difference used in the calculation seems somewhat arbitrary (9), and, crucially, a review of the archive JSTOR failed (unlike SD) to find one statistical critique of the method. However, for the record, with regard to retinopathy progression, the adjusted hazard ratio for MAGE was 1.06 (95% CI 0.94–1.19), P = 0.33.
We entirely agree with Monnier et al. (6) that to definitively establish any possible role of glucose variability in microvascular complications requires a prospective interventional study. We never thought that our article would, or should, be the last word on the subject. However, one of the attractions of analyzing the DCCT dataset is that the study was conducted before possible confounders, such as antihypertensive and lipid-lowering agents, came into common use. Consequently, it seems likely that the feasibility of any new study powered to take into account any of these and other factors could prove to be challenging to any future investigators.
As things stand, the world’s most complete dataset relating glycemia to microvascular complications has found glycemic instability to play no additional role in complication risk. As such, it means that the burden of proof for any future study is no longer to confirm an association but to disprove the lack of one.
