One of the most striking results I observed (other than the fact that real life accuracy wasn't as good as advertised) was the existence of a significant bias in the Dexcom measures when they were compared to SMBG values.
Here is that chart again
What it shows is that the Dexcom G4 non AP had a tendency to over-estimate low values (outside of incidents such as compressions) and under-estimate high values.
At the time, I did my best to validate that interpretation, clean up my data of any eventual biases, double check. But of course, there's always room for mistakes. That's why I read with great interest that large scale study by Lori Laffel MD, MPH, of the Joslin Diabetes Center where the very same bias is clearly visible. The scale, presentation and units are different but one can clearly see that in the low range the average errors tend to be on the up side, from 70 to 200, the errors are balanced and that above 200-250, errors are on the low side)
These results are mostly irrelevant today as the standard G4 is living its last days and the G4 with 505 update is now available outside the USA.
However, this result comparison shows that, with the kind collaboration of normal users, it is possible to extract valuable - and valid - insight.
I'll get back to the paper in more details on this blog if I can find the time.