Challenges Facing Automated Insulin Delivery Systems

Along with the increased popularity of closed-loop systems, I’ve been following the clinical studies and technology reviews published in peer-reviewed literature, and though they can be rather long and technically complex, I’ve summarized them in this article.

To sum it up, the primary barrier facing fully automated systems is the fact that both insulin pumps and continuous glucose monitors (CGMs) attach to the body via interstitial tissue, which creates two highly problematic conditions: time-lag and pharmacokinetic inefficiencies, both of which make any kind of AI algorithm too chaotic and error-prone to yield a reliable dosing protocol. The article includes citations, of course.

I know very few people who are unhappy with their closed-loop systems, but the literature seems to suggest this is a case of selection bias as well as healthy user bias. My previous article on that cites assertions that this is a case of reverse causality.

Happy to discuss here, of course!


I think this summary is on point. For me, the tricky part has been the lack of control as a lot of closed-loop systems don’t allow the patient to adjust things as needed.

I had an Omnipod 5 and simply hated it because it didn’t allow me to adjust my ranges. I know I need to be careful when my glucose level gets around 160 mg/dl, and the system at the time I used it wouldn’t let me adjust that.* Eventually, I just went back to MDI. I intended to start Looping, which I just started doing yesterday. I’m very excited about it. I chose it because it simply gives me more control. Even setting the Loop interface up once it was on my phone was pretty straightforward for me. Probably because I’ve been pumping in one way or another for years, so even after months of MDI, I was able to pull my last pump settings and start.

Before the O5, I had a Tandem that was on their Basal-IQ system. I opted not to do their closed loop system at the time. My reasoning was that I would be unable to roll the software back if their closed-loop system didn’t work for me. The Basal-IQ system worked great for me. I have a good feeling about the Loop system so far and know a few people who also Loop. I hope that gives me a good network of people to touch base with as well as my endocrinology team should I have any questions.

Also, from my observation of discussions in the diabetes community in general, the loudest people are the ones who are upset and are having problems. Those who are doing well overall pop up to share a good a1c or time in range number, and that’s about it. I’d count myself in that group, as I had an 87% in range last week. Let’s see what I get for next week now that I’m Looping.

*maybe that’s changed, but I’ve already given that system up and won’t go back.


How old is the data that was relied upon? I think the new automated pumps are much improved from old models. And while they do require a lot of investment from the user, they are a major improvement in treatment options. If your cgm is not providing BG numbers that closely reflect actual BG, that can hinder proper dosing, but my cgm is almost always 1-2 points from a finger-stick. I still finger-stick twice a day to ensure numbers.

The new world of automated pumps offers a great opportunity for a1c improvement, imo. And, eventually I suspect that not having manual override won’t seem so terrible. I’m one of those T1s who is learning better control from my Medtronic 780g, After an adjustment period, my projected reduction in A1c is incredible! I won’t get ahead of myself, but expect to see awesome results in July. Will come here to post for sure! I’m so excited!

So, I feel the new systems are groundbreaking (I can’t address other brands, as I haven’t used them lately) and the numbers will support it. I agree that those who are having great experiences may not be as vocal as those who are struggling.

both of the comments above reflect a common experience among those who are successful with their systems, and it boils down to this subtle phrase that’s easy to ignore:

while they do require a lot of investment from the user, they are a major improvement in treatment options

This validates what my original article stated, success with pumps and looping systems are mostly attributed to those who invest their time and attention into their self-management. But that is the real cause of improved health, and the studies show this to be true for MDI users as well.

As it pertains to individuals, a person who is given similar levels of T1D education and is also willing to invest time and attention (mind-share) to their self-management, typically yield equivalent A1c levels to themselves… that is, the same person typically does the same with a pump or MDI, provided that they attempt to invest that time into either equally.

Outside of clinical trials–that is, the real world–the perception that pump users have better A1c levels than non-pump users is mostly due to a “healthy user bias,” where only those pump users who are doing well are the ones who stick with it.

There’s also the Christmas Tree effect (which I also mention in my article), which is a phenomenon that happens with some people, where they are more willing to invest that time in pumps because they are enamored with the technology itself and “want it to succeed.”

The primary barrier to widespread (societal) improvement of glycemic control is the resources (people, money, infrastructure) to provide education and training. Without systemic institutional support, only those in higher economic and educational tiers get that support, which also happens to be those who can get pumps (due to better insurance plans, etc.).

1 Like

I admire you’re very thoughtful article, but it leaves me a little confused. I get that CGMs rely on interstitial fluid and are not completely accurate and do not reflect current blood sugar exactly. Why do you think not being completely accurate is a fatal flaw? I think we’re aiming for time in range, not time at an exact blood sugar so it seems to me that a slightly fuzzy number is okay for the slightly fuzzy outcome that is the goal.
I 100% agree that the unreliability of subcutaneous infusion is a big problem. Don’t get why you think that delivering insulin more directly is technically infeasible, though. My understanding is that it’s already been done with implantable insulin pumps decades ago.

I get that CGMs rely on interstitial fluid and are not completely accurate and do not reflect current blood sugar exactly. Why do you think not being completely accurate is a fatal flaw?

Keep in mind, I’m either summarizing and/or quoting excerpts from the reports I cited in my article, so my interpretation is from those articles. And those articles are saying that the delicacy of glycemic control is more than merely pairing insulin with glucose levels, because the need for insulin (and other hormones) can be triggered by other factors, such as exercise, inflammation and other circumstances that would not necessarily be known by glucose levels only.

If you were tracking only glucose, then you have two other problems. First, glucose is not evenly distributed throughout the body–different organs and even the bloodstream–does not contain equal levels, making it hard to infer what the systemic levels are, and proper insulin dosing needs that level of granularity. (Without it, such as what we have with CGMs, we can make macro decisions, such as big doses over hours, but not the kind of nuanced detail that is needed to keep glycemic levels “normal.”)

That leads to the second problem: measurement. Glucose is highly volatile in fluid (even small amounts), so any given sample (no matter how small) is going to yield one reading, whereas the next sample can be different. (This problem is exacerbated when glucose levels are higher, and/or when levels are changing rapidly.) This is why CGMs need to collect a lot of readings continually over a five-minute period before giving a single reading. And even then, it’s still an estimate. It could wait longer to get a better estimate, but CGM manufacturers thereby face the trade-off between the need to know something right away, versus waiting longer to get greater precision. For detail on this, see my article, "Continuous Glucose Monitors: Does Better Accuracy Mean Better Glycemic Control?

To improve upon today’s pump technology, one needs to infer systemic glucose levels, which is currently not possible from interstitial tissue. The researchers who wrote the articles I cited say that this limitation imposes an upper ceiling on how well pumps can perform unless that barrier is crossed, and that ceiling is pretty much where we are now.

As for insulin delivery, you asked,

Implantable anything is problematic: infections, faulty devices, cost, bulk, psychological factors. The implantable pump never took off, and development pretty much tapered off in 2015. If you’re going to implant something, why not make it actual islets, which is where current R&D is headed now.

Implantable or not, delivering anything directly into the bloodstream is problematic, largely due to risk of infections and other problems. Besides, the bloodstream is not the portal vein, where insulin really needs to go, so it can get to the liver first and perform the many functions within that ecosystem that is part of the nuts and bolts of the delicate process of glycemic control. To get to the portal vein, that brings us once again to islet transplants, and now that they can be very inexpensively grown using stem cells, availability is no longer a problem. It’s still the immune response.