As you’ll be aware by now, I’m wearing a number of different CGMs with a view to comparing them.
But as you’ll also be aware, CGM comparison is notoriously difficult, so this is a brief article to describe how I plan to do this.
The story so far…
As you’ll have seen, I’ve applied and started the sensors, and the video of that serves to show the process. I leave readers to look at that and decide which they think is most appropriate for them.
I’ve also tried to briefly describe waste and insertion tools, and again, that’s really down to end user choice as to which you prefer.
I’ll also produce brief videos of interacting with the apps, trying to do similar things so that anyone can see what they look like and if anyone has preferences.
Many are very similar, and once again, individual preferences tend to play a major part in choice on this level, and it’s hard to make a recommendation, but it’s easier to highlight where there are differences.
Ultimately though, it’s again, a user choice if the app drives sensor choice.
I’m also gathering reams of data. And that’s where I will make comparisons on myself.
As with other comparisons, I will generate a Mean Absolute Relative Difference, or MARD vs fingersticks, using the Ascensia Contour Next system as my reference. MARD is usually used versus venous blood and is commonly used as a way of referencing accuracy and comparing systems. As:
a) I don’t have access to a way of measuring glucose in venous blood; and
b) fingerpricks are how we’d dose without CGM;
it seems like a sensible alternative.
MARD doesn’t really give a good picture of accuracy across the range of blood test data though, and one of the things we’re particularly interested in is how these devices do when we’re either high or low.
With this in mind, I’ll also produce some datasets that show percentage of readings within the 20/20 rule, and potentially some of the others (5/5, 25/25, etc).
This shows percentage of readings within 20% of the reference fingerprick test above 80mg/dl or 4.4mmol/l, and within 20mg/dl below that level.
It’s then worth breaking this out further into the percentages at different glucose ranges to gauge at what levels the readings are most aligned with the fingersticks, or for some people, “most accurate”.
Finally, the data will be added to a consensus error grid to give a visual representation of the dispersion of the data points.
What are you hoping to achieve by this?
It’s very straightforward. Given the differences between studies used to generate MARD numbers, I’m trying to demonstrate in an n=1 way that a useful dataset can be created to compare sensors, and I’m also trying to highlight any places where sensors may not be entirely trustworthy, to ensure that people using and issuing them give due consideration to the choices they are making.
I’d also encourage anyone out there who was willing, to fund powering up a study like this to provide a proper dataset across systems that would help people make better choices.
Not that big an ask!