A couple of entries ago, I mentioned an upcoming publication. It has now appeared in the J Diab Sci Techol ahead of print (subscription required).
Basically, each glucose meter result in the A zone of an error grid has its difference from reference squared and that value scaled ranging from 0 (result=reference) to 1 (result is on the A zone boundary). If the value is outside of the A zone no A zone Taguchi loss value is calculated. These “Taguchi loss” values are then averaged to give the average Taguchi loss (ATL). One can calculate the ATL for all zones, although I did not do this for the article.
The ATL represents a way to distinguish performance among different glucose meters with similar performance (most values in the A zone and no values beyond the B zone). I believe it is an improvement over the MARD statistic (mean absolute relative deviation).