Thanks, Adam. It’s always interesting to find data backing up anecdotes.
A1c is indeed an imperfect and, at this point, loaded metric. For better or for worse, it is also the standard by which efficacy of long-term management is most frequently evaluated. Your point on dispersion is well taken. From that 2008 study (Nathan, et al, available here), the 95% confidence interval around an A1c of 7.0% is 123–185 mg/dL, and the corresponding 95% CI around a mean glucose of 154 mg/dL is 5.9% to 8.0%. To your second bullet point, that statistical analysis does not take into account the error bars of each individual data point, which are significant along both axes (A1c and AG). I’ve also read elsewhere that, for a given mean glucose, the intra-patient variation is small, meaning that, if you fall above or below that regression line, you will tend to do so with some regularity.
Both the science and statistics are quite interesting, but to me, the pressing points (as expressed in the follow-up article) are actually more rudimentary: why do meters made by the same company show such glaring and persistent biases, and, given the magnitude of the issue, why is it neither acknowledged not addressed?