As many can have seen, the Dexcom G7 by no means had, and not too long ago the G6, has eliminated back-smoothing from the apps on the telephone, for varied causes. Does that matter? To most individuals, in all probability not, however let’s take a look at comparisons of the information with the 2 units plugged into me.
In all the pictures of knowledge aspect by aspect, the model with the inexperienced dots is a Dexcom ONE visualised through Drip Companion App setting, and Blue dots are Decom G7 visualised utilizing the companion app setting in an xDrip Variant model, which permits a number of xDrips to be run on the identical telephone. This permits us to:
Visualise the 2 aspect by aspect;
Work across the limitation that makes screenshots within the Dexcom G7 app not possible on Android units.
What we’ve heard from varied person teams on the Social Media is that they really feel that the G7 doesn’t produce as easy knowledge because the G6 (or the ONE), so listed below are a number of photos of the ONE and G7 aspect by aspect to attempt to see if that’s true in an n=1 sense, and what the influence could be.
It’s price mentioning that each these sensors had been utilized on the similar time and are subsequently the identical age.
Aspect by Aspect
In each these photos, screenshotted from widgets on a Samsung telephone, there seems to be extra Jitter or noise within the Blue dots than within the inexperienced dots. If we then dig into the underlying knowledge inside xDrip, you see the next:
Among the knowledge right here (pre-5.30am) could be put right down to compression lows inflicting jumpy readings, nevertheless, after 5am, there’s a notable lack of smoothness within the output, in comparison with the inexperienced dots of the ONE. At this level, the sensors ought to each have overcome their first couple of days of instability, as each had been a little bit over two days outdated.
However even now, after we’re seven days into the sensors, there’s positively a distinction within the readings, and nonetheless some clear jitter, despite the fact that it appears to have calmed down a bit on the G7.
So how a lot does this “noise” matter? .From an on a regular basis perspective, maybe not an excessive amount of, though, it may intervene along with your studying at a cut-off date whenever you had been maybe about to dose insulin for lunch. The shortage of backsmoothing, additionally doesn’t actually matter vastly on this use case.
However what about with an AID system? That’s a really completely different query.
Jitter and AID techniques
The quantity of noise that may be seen in these photographs of the G7 might be an excessive amount of for a non-filtered system to work with. The place a system is delivering correction boluses, a sudden sharp bounce would possibly encourage over-correction leading to supply of an excessive amount of insulin. However most AID techniques are smarter than that, and use one thing to handle that jumpiness, for instance, one thing like a Kalman filter, which is, I perceive, how CamAPS FX offers with it.
Equally, within the Open Supply world, a wise particular person has written a smoothing algorithm to scale back the jumpiness of the information, that’s, in essence, not a good distance from a Kalman filter. That is out there in Dev in AndroidAPS proper now to check out. I’d advocate that Loop would possibly maybe need to introduce comparable performance with the microbolusing code that’s now out there, to reinforce the security of the system. This additionally all helps with the variations exterior of this degree of jitter, from compression low rebounds, for instance.
What’s inflicting this jumpy, noisy knowledge?
To this point, I’ve tried three G7 sensors and seen the identical throughout all of them. Others have additionally used a number of sensors and had comparable difficulties. My preliminary hypothesis was the distinction in how the sensor sits in tissue. As well as, it could recommend that possibly the onboard algorithms within the two completely different sensors have variations.
Within the G6, and former Dexcom units, the sensor is at an angle into the physique tissue. Within the G7, it’s vertical. I assume that this someway makes it much less secure and extra inclined to maneuver throughout the scar sheath that it makes on entry. However that’s solely hypothesis. Since then I’ve heard that others have mentioned this challenge instantly with Dexcom, and had an analogous opinion in response. It’s considerably irritating to see this taking place, and one wonders what the tip end result will likely be. It additionally makes me surprise simply how a lot “magic” is occurring within the Abbott sensor algorithm, given that every one these sensors are vertical somewhat than angled.
Any influence on industrial AID techniques?
It will rely fully on how they use CGM knowledge. If the don’t rely solely on the Dexcom knowledge and apply some type of filtering of their very own, then I think that the important thing level will likely be testing. If nevertheless, they rely solely on the information, then I think there will likely be extra of a delay in releasing G7 enabled units as they work out how one can handle this. Sadly, it’s not clear from any of the documentation who does what.
So the place does it depart us as finish customers? Within the open supply world we’re making progress. Within the industrial world? When you’re already managing filtering, in all probability not too frightened. when you’re not? Then you definately’e an entire heap of labor to do.