Closed loop coming closer?

[Books1]

This looks interesting https://diatribe.org/jdrf-aims-make-closed-loop-more-customizable-interoperable

 

[ringi]

"closed loop" will never work based only on current BG monitoring, as by the time BG has increased, it is too late to add the insulin. There will have to a way for the system to know what food someone is eating. Or a BM sensor embedded on the blood that leaves the digestive system.

Also, current systems add the insulin too far away from the liver to get a very quick response.

(We may get system where someone can take a photo of the food they are about to eat, and the AI works out the carb content etc.)

 

[DCUKMod]

I wonder if Tim might have time to have a look at this one. I know he is very knowledgeable of this subject.

 

[tim2000s]

@Books1 - yes - it's been interesting talking with the JDRF on this one. The idea is what we've been campaigning for in the We Are Not Waiting community for some time. It's now down to the manufacturers to step up to the plate.

"closed loop" will never work based only on current BG monitoring, as by the time BG has increased, it is too late to add the insulin. There will have to a way for the system to know what food someone is eating. Or a BM sensor embedded on the blood that leaves the digestive system.

Also, current systems add the insulin too far away from the liver to get a very quick response.

(We may get system where someone can take a photo of the food they are about to eat, and the AI works out the carb content etc.)

Sorry @ringi, I'm afraid (and I feel qualified to say this for reasons that will become obvious) that you have absolutely no idea what is possible.

As a closed loop user of around 14 months, and for the past 6 of those, using functionality that combines Fiasp with some smart algorithmic work, I can eat limited carb meals (less than ~40g carbs) without bolusing or meal announcing and stay in a range of 3.9-10mmol/l (70-180 mg/dl - the range used in clinical trials), and if I announce something like Fish and Chips, or eating out at a restaurant, then I don't have to bolus. At all. The speed at which current CGMs change in relation to blood levels is good enough for this, and with insulins like HDV, we're starting to see possibilities of not announcing for larger meals.

Using a month of very little manual intervention, as described above, I'm getting average glucose levels of about 6.6, my median is 6.1, the standard deviation is 30% and predicted Hba1C is 5.8%. My time in range is 87% (again using the clinical trial numbers) which is vastly ahead of normal MDI use. As you might be able to tell it's something I'm passionate about.

What we can also then do, again with smart algorithm use, predict when food might be eaten and also how long it is before absorption starts when food is eaten. By doing this we can safely prepare the body to eat. We're currently working on a load of things.

On your point about insulin delivery, the majority of exogenous insulin doesn't get to the liver anyway (hence the work on HDV-insulin), so it's more about how you can get existing insulin types absorbed and in use by the musculature mostly. Again, we currently have Fiasp and Lilly are working on "faster-Humalog". That's anticipated later this year.

Even now, with these capabilities enabled, I almost never exceed glucose levels of 12 in normal operation. But we shouldn't just think of artificial pancreases as "IT-like" technology. The Smart-Insulin initiative is another interesting approach to "Artificial Pancreas" technology that is even better placed to manage glucose levels more effectively.

All I'm saying is never say never!

 

[ringi]

I can eat limited carb meals (less than ~40g carbs) without bolusing or meal announcing and stay in a range of 3.9-10mmol/l (70-180 mg/dl - the range used in clinical trials), and if I announce something like Fish and Chips, or eating out at a restaurant, then I don't have to bolus.

Low carb clearly makes the problem easier due to the "rule of small numbers" and slower changes in BG. I expect the key will be the design of a UI that allows people to announce meals that is easy to use, yet gives the system enough data to learn from. (One day it may be possible to measure the hormones that are released when a meal starts to be digested.....)

However I see most of practical benefits being from preventing hypos, as it is not hard for someone to press a button when they start eating, indicating if the meal is "low carb", "mid carb", or "high carb". I expect that sharing data between users (along with GPS) will help for larger restaurants.

(I assume a link to something like a fitbit and GPS is used to get input on exercise level.)

However I thought healthcare providers worldwide have refused to pay much more for "better" insulins, hence how will these new insulins be funded?

I just don’t like the terminology "Artificial Pancreas", as it makes people think they will be able to forget about diabetes and hand over all responsible for their health to the system. Even worse when they start to be used with Type2 leading to an even fasting cycle of increasing insulin resistance.

(It is a few years since I have written control software (mostly open loop), but I have worked on some large scale industrial control systems. Telling when a sensor is giving invalid data is always an issue.)

 

[ringi]

Thinking more about this, I think the issue I have is the concept of "closed loop", I have no problem seeing how mostly automated systems that are not complely "closed loop" can work.

 

[tim2000s]

Low carb clearly makes the problem easier due to the "rule of small numbers" and slower changes in BG. I expect the key will be the design of a UI that allows people to announce meals that is easy to use, yet gives the system enough data to learn from. (One day it may be possible to measure the hormones that are released when a meal starts to be digested.....)

However I see most of practical benefits being from preventing hypos, as it is not hard for someone to press a button when they start eating, indicating if the meal is "low carb", "mid carb", or "high carb". I expect that sharing data between users (along with GPS) will help for larger restaurants.

(I assume a link to something like a fitbit and GPS is used to get input on exercise level.)

However I thought healthcare providers worldwide have refused to pay much more for "better" insulins, hence how will these new insulins be funded?

I just don’t like the terminology "Artificial Pancreas", as it makes people think they will be able to forget about diabetes and hand over all responsible for their health to the system. Even worse when they start to be used with Type2 leading to an even fasting cycle of increasing insulin resistance.

(It is a few years since I have written control software (mostly open loop), but I have worked on some large scale industrial control systems. Telling when a sensor is giving invalid data is always an issue.)

Again, an interface already exists, and we use it. It's simply IFTTT buttons configured for certain actions. They are on my smartwatch and I can tap them with food. I don't have to be massively accurate either.

The low prevention has been around for a few years in both commercial products and was the first iteration of OpenAPS when it was created. There's not a lot of benefit in sharing "Low", "Medium" and "High" data across restaurants as at that level it's an observation thing. A plate of chips will always be "High" for example, regardless of the restaurant. What would be more beneficial would be the range of what each of those means based on the geolocation that the entry was made at, as you could then dose appropriately for the bottom end of the range and increase dose as glucose readings dictated.

Exercise is proving to be one of the biggest hurdles in all of this. It's not the time to peak insulin action that causes the biggest issues with exercise, it's the time to decay, and the majority of current fast insulins have a half life of about 57 minutes, compared to endogenous insulin which is approximately ten minutes. This means they all last six plus hours in the body, and are therefore still quite concentrated a few hours post meal when exercise is likely. The real breakthrough here will be "faster out" insulin.

Some are using the term "Automated insulin delivery system", but I think most of us with broken pancreases prefer the term Artificial Pancreas. There are other names given as well, such as, currently, Hybrid Closed Loop.

I'd agree that the sensor data is the key to this, and there are plenty of studies to demonstrate how poor sensor data affects outcomes. The existing systems all incorporate some level of "noise" detection to deal with this, which does seem to work effectively. As it's also (in real time system terms) a fairly slow moving system, it's possible to correct from an error with little impact.

Thinking more about this, I think the issue I have is the concept of "closed loop", I have no problem seeing how mostly automated systems that are not complely "closed loop" can work.

I think it's fair to say that if you operate fully closed loop, the results are not as good as the almost closed loop approach, but they are still good enough to achieve a surprisingly good hba1c for example!