AI and T1D: a quick overview

How will AI affect type 1 diabetes?

Most T1Ds think that AI will improve automated pump algorithms to achieve even better glycemic control. But, in a new article I recently posted, I believe it’s a bit more complicated than that, where I posit that where AI really can improve things is by new generations of AI-driven communication tools that can both distill highly complex and meaningful medical knowledge into formats that normal people can understand, and make such material more widely disseminated, especially by (and through) healthcare providers, so patients can apply more personalized self-management techniques.

It may not be easy, but there is precedent. I hope my article can stimulate ideas in this area.

1 Like

For me, I won’t use any automated control because I find the CGM is too unreliable. For me, sometimes it reads exactly the same as the finger stick and other times as much as 50 mg/dl higher. The problem is that CGM is measuring interstitial glucose, not blood glucose. For emergency alerts, I have the G7. But, for insulin dosing, I use finger stick

John

While this often works, it can swing the other way too. That is, sometimes the CGM (a dexcom G6) can be more reliable, not because its readings are more accurate, but that it’s a better representation of what systemic glucose levels are–and especially its trajectory.

Read that carefully: I did not say the CGM is “more accurate,” I said “more reliable.”

Dosing should never be done on glucose values alone, but the rate of change leading up to that value (and the actions that were taken that might account for that rate of change).

If you see 200, whether it’s a BGM or CGM, what really matters is where it was just prior to that. If it was 250 coming down at 5mg/dL per minute, that’s entirely different than if was 100 rising at the rate.

You might say, “yeah, well, of course…” Fine, but now let’s get into that:

Now that you take into account the rate of change—which a BGM can’t show you, but a CGM can, accuracy notwithstanding—you then need to consider what you did beforehand that might have caused that rate of change. Did you eat beforehand? Or dose beforehand? Or exercise? Or even more? Chances are that there isn’t just one thing, but multiple things.

This is where things get murky, particularly for an algorithm. I already know I don’t need to convince you, Dr. John, about that. You already know the CGM’s value (or even rate of change) is insufficient for an algorithm to make “good” dosing decisions, but your description of using a CGM for dosing decisions alone is similarly error-prone. There are a lot of factors that have to be taken into consideration that only you the human can know, and the weighting to give each of those factors is what most T1Ds eventually intuit from years of experience: when I do this, that happens. But when I do that, this happens.

Obviously, that doesn’t mean it’s easy to fully take into account all the factors, or how much weighting to give to each of them. This all accounts for why glucose is so hard to manage. Duh. But relying on an automated system to know this–or to see patterns that will reliably repeat themselves–is a pipe dream.

But I always have to emphasize, AID systems can do better than a T1D that never looks at a CGM at all, and/or who makes really poor dosing decisions. And sadly, that’s a huge number of people (mostly children and adolescents).

But let’s get back to your point, Dr. John: Dosing based on BGM values alone can sometimes work, but it can also fail, not just for the reasons I described above, but because of the mistaken belief that either meter represents the body’s “systemic glucose levels.” That is, the total amount of glucose in the body. That’s not the case, and can lead to just as many bad decisions as good ones.

When glucose values are volatile–which is more often that not if you look at your daily chart—a BGM test may be useful to do a finger-in-the-wind sanity check on whether your CGM is in the ballpark, but it should never be the true guiding light.

A full discussion of this, with citations to medical literature and graphic illustrations of glucose levels and CGM vs. BGM readings, can be seen in my article, “Continuous Glucose Monitors: Does Better Accuracy Mean Better Glycemic Control?

I read the complete article. It was interesting.

John

Of course it can. You do a test, you wait a few minutes, and then you do another test.

You just need to make sure you are using a good meter, like the Contour Next.

The shorter your time window, the less reliable your “rate of change” estimate is for effective dosing decisions. Sure, you could use a few minutes, but you have no idea if those few minutes show a short-term leveling off from an earlier spike, or a spike that may just now be starting, or a continuation, etc…

I would argue you need at least 30 minutes—and optimally, 90 minutes—of glucose movement patterns to make the best-informed dosing decision, which a CGM can provide that a BGM cannot. (Again, I’m not arguing against a BGM by any means. I’m only saying it’s not the heavily weighted variable in a more complex equation. Useful, but not nearly as much as a 90-minute window.)

Again, all of this is discussed in detail in the aforementioned article, which shows charts that illustrate these points.

You are just trying to get people to go to your site! If you have something beneficial to impart to diabetics, why don’t you post here instead of using this site to herd people to your own site?

1 Like

because you can’t easily post long articles here with embedded graphics and other elements that make something a more substantive “article” than what this platform provides (which is more designed for discussion). There is no incentive to “bring people to my site” – there’s no advertising, subscription fees, nothing. It doesn’t even have discussion forums. It’s merely a better platform for posting content than discussion groups.

If anything, I choose this platform to have all my discussions–I don’t post to other forums (for some reason–I guess just habit).

It’s the discussion that matters. From what I can tell, people are engaging and either asking questions or offering input, which benefits everyone.

If the moderators of this platform don’t like that I post links to my articles, I’, more than happy to hear from them and I’ll stop.