We read with interest the article by Dyck et al. (1), in which the authors described a chronic glycemic exposure variable (GEi) in the Rochester Study. They examined GEi and its individual components (A1C, duration, and age at onset) in terms of prediction/correlation with complications and concluded that GEi is generally predicted better than its individual components (see Table 3 of ref. 1).
Dyck et al. compared their results with our previously published analyses (2) using a different chronic glycemic exposure variable, A1 months, noting that (as also reported by the Diabetes Control and Complications Trial [3]) this combination variable did not predict better than its components (A1 and duration). Our analytic approach, however, was different; we compared the fit of models, including the components to a model, with the composite alone. The differences in fit were small but favored the separate components. It would thus be interesting to compare the total R2 of alternate models, one with GEi and another with its components, in the current study. We suspect that, as in our case, differences would be small.
Another interesting issue is the use of “age at onset ” and “duration” (1) together effectively defining age itself. Could any enhanced prediction be related to age itself? Inclusion of the partial R2 for age in Table 3 (see ref. 1) would be useful.
Dyck et al. further suggested that differences between these studies may be explained by the “choice of patients” and differences in outcome assessment. As the Epidemiology of Diabetes Complications study (4) is comprised of community-treated type 1 diabetic individuals from a childhood-onset cohort shown to be epidemiologically representative of type 1 diabetes, selection bias was unlikely. However, the inclusion of type 2 diabetic subjects in the Rochester Study may have influenced results. Nevertheless, we agree that a continuous neuropathy outcome measure may be preferable and that this difference also may have contributed to the differences reported. Consequently, a comparison of A1 months and GEi would be more informative if performed for the outcome common to both studies (Diabetes Control and Complications Trial protocol neuropathy).
Finally, one motivation behind developing the A1 month measure was to address whether a glycemic threshold exists above which complications develop. Were the authors able to examine this issue using GEi? While unable to determine a clear threshold, we found that ∼1,000 A1 months were experienced before the advent of advanced complications. This translates to 42 years of A1C 2% above normal or 18 years at 5% above normal, which reflects another motivation for our chronic glycemic exposure variable—a clinically useful concept of risk.
