As mentioned in this article, I bought myself a GlucoRX Aidex to see how well it worked. As what has to be one of the cheapest systems on the market in the UK, and also on the prescription tariff, it needed a review.
So far, I’ve written about the experience of my first sensor, the issues with its lack of stickiness and the rather uneven and difficult to find instructions and advice. But this article covers the big one.
And if that’s not something you want to read all the way through for the detail, then these are the headlines.
For me, the Aidex was mostly very inaccurate when compared to blood tests. The MARD over the official sensor life was 15.1%, which compares to 9.2% for Dexcom over its official sensor life and 11.6% when used for 14 days, with the inaccuracies that always brings.
It missed a high proportion of hypos (although a small number were measured) and had a high level of variability. All of this, combined with very unclear calibration guidance leads me to the conclusion that this shouldn’t be on the prescription tariff, and is dangerous enough to cause a driver who used it to check levels pre-journey to be involved in a hypo related crash. It’s that bad.
So while it maybe the “cheapest” to buy, there are plenty of costs that could be associated with it, and most of those really don’t want thinking about.
Sure it’s the cheapest on the market, but it’s not worth the money you’re paying for it.
It’s important that a CGM isn’t too far from fingerpricks in terms of its accuracy, as in its place, that’s what we’d use for determining insulin dosing and deciding that we are safe to drive , regardless of what a trial compared to YSI analysis says.
The output from the clinical trial that GlucoRX, sorry, MicroTech, the actual maker of the system and the company that have issued the license to GlucoRX, have published, suggests that the accuracy versus veinous blood through a YSI analyser is a MARD of 9.08% and versus fingersticks, 10.1%. They also go on to say that it’s comparable to the Abbott Freestyle Libre, which are bold claims for an unknown CGM.
As discussed in the recent article looking at accuracy trials, those of the Aidex were a bit of an outlier compared to what other manufacturers do. Will that play a major part in what I see versus the study outcomes?
To try and make this a fair fight, I’ve used the same model that was used in the clinical trial (finger pricks at 0700, 0900, 1100, 1330, 1500, 1700, 1900) and added an additional one at 2100 due to the way my day works.
Eight fingersticks per day for 14 days. 112 in total. Additionally, I have worn a Dexcom G6 alongside it, to see how the two compare. As the G6 expired at 10 days, I’ve used the trick to extend the sensor and we’ll see how well it compares outside of the official lifespan.
The fingersticks were done with the Contour Next One system, regarded by the Diabetes Technology Society as the most accurate of those available.
Calibration of either sensor was done on an “as needed” basis, given the lack of guidance for the Aidex.
As an additional data point, my Time in Range (70-180) is around 82% over the period, with a Time Below Range of between 3% and 4% (hard to be clear due to variation caused by compression lows). The GVI score is 1.49.
Overall then, these are the results of this n=1 assessment of how accurate the Aidex sensor is compared to fingersticks and the Dexcom G6 (over 10 and 14 days).
As we can see, the MARD from fingersticks for the number of measurements in this assessment is 15.1% for the Aidex.
This compares to 9.2% for the Dexcom G6 when used within its official life and 11.6% when restarted and used over 14 days.
Breaking it out on a day by day basis, we see the following outcomes:
Day 11 for the Dexcom was particularly bad due to the trauma calibration taking effect.
Given the regular timings, we can also compare the values at different time points (which typically relate to different levels of glucose variation), although, as for the daily values, don’t have a large number of values so will have a greater potential error.
Consensus Error Grid
Next, we move on to the Consensus Error Grid and modified Bland Altmann charts comparison, using the tools supplied by the Diabetes Technology Society.
It’s clear in the first of these three images that the Aidex has substantially more dispersion than the Dexcom does over its official life, and the major deviations for the Dexcom come from the extended period.
ISO Standards data:
These are the results from the 15/15, 20/20, etc rules, which determine how many data points fall within what’s considered to be accurate.
The following image shows the variance from the fingerprick tests over time, and is a graphical representation of the data presented in the day by day tables above. It’s clear that the Aidex took a long time to settle to a consistent level.
What’s very clear here, given that we’re comparing accuracy across the same glycaemic variability in a single person, is that there are significant differences.
Accuracy, or lack thereof
At a MARD level over the official life of the sensor, the Aidex is more than 50% worse than the G6. Even when the Dexcom is extended, it still generally performs better on the point in time comparisons, even if it goes into sensor error at higher values when it encounters noise.
We are told that we shouldn’t just look at MARD though, so taking the Consensus Error Grid into account, we have more of the same. The image highlights the same thing as the descriptive statistics. There is a very wide dispersion of the Aidex values compared to the reference, and these trend towards a bias.
Additionally, if we look at the ISO scores, we can see that none of the Aidex results fall within what’s considered to be the most accurate (see Jon’s article here). In fact, they would suggest that it’s very inaccurate when compared with the Dexcom at both 10 and 14 days.
Dispersion and Variability
The Consensus Error Grids really tell quite an important story. The Aidex has a number of readings in zone C and a significant proportion in Zone B, when compared to the Dexcom, even when it has been extended. Additionally, it took a third of its life to get into low double digits.
At 9am and 9pm particularly, there seemed a lot of difference from bloods, and given the times I eat, I can only put it down to that being when there is greater glucose variability. That’s not a great performance in a CGM.
As a friend of mine described it, you’d be equally well off with a weathervane.
And therein lies the problem. With that much variability in the results, this isn’t a safe device to either dose insulin from or to use to determine whether you are safe to drive. Whether we like it or not,, people will use it for these things. You just can’t rely on it. This isn’t helped by the distinct changes over time that we mentioned earlier.
As I’ve mentioned a couple of times, the Aidex is supposedly factory calibrated, however, the huge variability when first started and the apparent lack of adaptation raise a level of concern. On top of this, the calibration model seems to be fairly basic, with glucose values jumping to the newly entered blood value when calibration is selected.
Unlike with the Dexcom G6, where some level of guidance is provided as to when calibration makes sense (e.g. how far from blood the finger pricks values are), you’re very much left to your own devices regarding choosing what’s appropriate for Aidex.
I’m not convinced that the “factory calibration” provided with Aidex is particularly good either. The variation over the first four days was fairly extreme and calibration didn’t seem to make a huge amount of difference.
While the Aidex is now the cheapest available offering on the UK market, that’s not necessarily a good thing. It’s also the cheapest available offering on NHS prescription lists, but in this n=1 experiment, it seemed to perform significantly worse than the Dexcom G6 does. A low price often comes with hidden costs, and I think this device has a few, not least of which is that during the testing the Aidex simply didn’t report 75% of the measured low glucose tests, compared to the Dexcom, which missed 25%.
I hope that there isn’t pressure from CCGs on surgeries to have them select this over the Dexcom ONE or Libre2 because of cost, because if there is, any incidents that occur as a result of use of the Aidex should be laid squarely at their door.
We now have three cheaper CGMs available in the UK, and so far, in my limited n=1 experience, none of them has performed as well as the big players. That may just be me, but it does feel as though these devices don’t just come with a low price, they come with a higher risk.
Whilst getting everyone who wants one onto a sensor system is the right thing to do, I’d rather it wasn’t with these.