I would like to extend my apologies to Gibson et al. (1) for having misunderstood their description of the feedback about blood glucose levels offered to patients in their trial of a real-time telemedicine system (2) and for incorrectly attributing a quotation from an article by Bode et al. (3) to their report. I, too, do not wish to see the interesting results of the Farmer et al. (2) trial data misrepresented. It was (and still is) not my intent to either blemish or deny the results of Farmer et al. or to assume an antagonistic posture toward the efforts of the authors.
However, as to their application of statistical methods to analyze A1C measurements, at the recent American Diabetes Association 2007 Scientific Sessions it was reported that using A1C to assess glycemic control will soon be replaced by mean blood glucose (MBG) (4). This will presumably “add clarity for diabetic patients looking to manage their disease” (5). Since this will require the replacement of one set of reference values (those used to define the standard MBG = f(A1C) relationship) with a second set (those from the patient’s own history), it will also call for the use of better statistical methods to deal with the temporal correlations between the repeated glucose values used in real-time monitoring. Averages of uncorrelated data are not interchangeable with averages of correlated data. Further, I concur with Gibson et al. (1) in my hope that this confusion has not distracted from discussion of the important issue of the statistical analysis of blood glucose time series.
