While talking with Dave about his Libre3 I was searching through the last 500 MAUDE reports Abbott submitted to the FDA. Completely unrelated to Dave’s issue I saw Abbott reports accuracy issues using the Parkes Error Grid. I know from looking at MAUDE reports previously Dexcom also uses the Parkes Error Grid in submissions. I started wondering if being able to check CGM and BGM readings that seem very different might be helpful to some people so I did some reading. The Parkes Error Grid is based on research and was created to categorize the risk inaccurate BGM readings pose to patients.
I was wondering if it would be useful for us so I recreated the Parkes Error Grid for T1D in something that was easy. https://student.desmos.com/join/xbmqu6
Is it useful? I was pondering creating my own error grid with just three zones, “fine”, “it might get better” and “replace”. How does it compare to https://www.diabetestechnology.org/seg/ ? Comments and criticisms are welcome.
the scatter plot illustrates a BS (finger prick) device reading compared to reference (or real and accurate actual blood sugar)
if you have a BS device, and compare it to CGM, how do you reconcile the scatter error of each device without a reference? my question is, without knowing accurate blood sugar at the moment (reference) then each device reading exists somewhere on it’s own scatter plot and the potential error of each device could add, negate or subtract and (again without reference) you will not be able to tell.
please please this is not a criticism… my experience is with technical calibration of scientific instruments, its one of the things I do at work. I have been tooting the “we need more accurate instruments” horn for 25 years. I am just uncomfortable with calling my finger stick meter “reference”.
Thanks for taking a look @joe. I’ve made a note to remove the references to “reference” next time I update the tool. I also tried making the points into lines the length of +/-15% of the BGM reading. The zones are wide enough that most of the time 15% doesn’t matter. Reducing the number of points where the accuracy of the BGM spans 2 or 3 zones can be accomplished by reducing the number of zones. The other thing to keep in mind is don’t let perfect become the enemy of good.
I’m not sure this will be helpful, but I think I’m echoing @joe’s comments on using one somewhat inaccurate device to judge another somewhat inaccurate device. The old, “A man with one watch knows what time it is, a man with two watches is never sure.” In reality, I’ve had both my finger stick meter (Contour Next One) be way off once in a great while (I chock it up to the strips), but I trust it much more than my G7 CGM, which has been way off much more frequently. I trust the finger stick reading much more and it has a good reputation among other T1Ds; but the margin for error is still highly suspect. When things “just don’t seem right”, I’ll violate the saying and test twice with a finger stick, if they’re pretty close, I’ll “stick” with it and move on…but everytime I do it, I think of the above quote.
@spdif on the plus side and I think it is very important, the tool emphasizes that on a high bs (actual or by finger stick) that literally a huge CGM variability is “normal” I see a lot of people misinterpret a big variation and mathematically it is irrelevant. I might consider that as long as the CGM is not also tracking rapidly changing blood sugar, the zoning seems about right.
@Tlholz I’m a fan of that saying. Also “don’t let perfect be the enemy of good”.
Thanks @joe. I was thinking about letting the user input more CGM readings to calculate the rate of change and displaying a warning when over +/-2 mg/dL/min.
I walked a mile to get my lab work done last week and the CGM said 160-ish and falling and the labs came back at 110-ish. 50mg/dL difference and nothing wrong, I was just past the limitations of the technology. The CGM caught up eventually.
