Wearable device senses blood glucose and administers insulin accordingly
spectrum.ieee.org
spectrum.ieee.org
They do, however, use a sensor that measures blood glucose (fairly accurately) and a controller that estimates how much insulin to give (very crudely).
Maybe this seems unintuitive, but the bottleneck of these systems is not the hardware or the control algorithm, but rather the time it takes for the insulin to start "working" once injected.
In a normal situation, the pancreas secretes insulin directly into the hepatic portal vein and reaches peak concentration within minutes. When you inject insulin, depending on how good ($$$) it is, it can take anywhere from 2-4 hours to peak [1]. In technical terms, you have a closed loop controller with a very slow actuator, which severely limits how tight the control can be.
There is ongoing work to make faster insulins. But personally, I think the "cure" for type 1 diabetes will come from restoring the insulin-producing capabilities of pancreatic islets, or synthetically adding such capabilities to other highly vascularized tissues.
Also worth checking out is Tidepool, an organization trying to take an open source closed loop controller through FDA submission. https://www.tidepool.org/automated-insulin-dosing
[1] https://images.squarespace-cdn.com/content/v1/5b7c64af620b85...
My understanding is that it has a predictive algorithm based on what it's learned about the response of the user. Also we work out carbs and put them into the device 20 minutes before meals so that it can pre-dose.
Another advantage is the minute to minute monitoring providing data that allows us to understand how well she is doing and what we need to do to improve her levels.
If you don't eat your meal then the bolus is active and I guess that there are problems, this hasn't happened for us yet.
In the background the system delivers basal insulin - this is a low dose of insulin that is used by the app to manage glucose levels. The sensor shows the level, the app makes a prediction and then delivers insulin (or decides not to) in order to keep within limits. Sometimes she can go low or high (below 4 or over 7) when she goes low sometimes she has to eat sweets, when high sometimes engage "boost" mode. There is also an "ease off" feature that she engages to tell the system to expect a low insulin period; for example before playing sport.
The drawback of having a sensor is that you get alarms. The benefit is that you know what is happening and can act. The Camaps system with the always on pump means that you get relatively few alarms or overshoots. She is in bounds for a much improved amount of time (I think 95+% vs <70% before), also the over/undershoots are smaller and more quickly managed (I have no stat, but I have seen the data and believe this to be so).
My daughter is nearly 12 and is able to manage this independently. It's certainly much easier for her than the injections were. I think that as a parent or school this would be much easier to manage for a smaller child - you would have to not give them the password for the phone of course! I'm no expert but I expect that there would be a lower limit for when the app will work because little kids are so different - but I don't know.
So what this article is talking about is, basically, taking the thing DIYers are already doing, and making a more foolproof, FDA-approval-grade version. And I'm sure they're massively better in terms of how fiddly they are to set up and get working, but the end result is actually worse; they wind up not including features like autotune (which auto-adjusts basals and carb ratio based on historical data), because there's less risk of liability if they make doctors do the tuning, even if the doctor is only looking at the patient's data once every six months and is doing a poor by-eye approximation of what the algorithm does.
It offers guidance on insulin injection amounts. It doesn’t make insulin and doesn’t do any of the many other critical functions of the pancreas.