First "at home" trial for an artificial pancreas

[SamJB]

A JDRF trial for the first "at home", unsupervised, overnight use of an artificial pancreas:

https://www.jdrf.org.uk/news/latest...with-type-1-diabetes-average-ten-hypos-a-week

Very promising. I'm confident we'll hand over our daily management to one if these things soon. Almost as good as a cure!!

 

[ElyDave]

very encouraging.

Get it sensitive enought to deal with rapid changes due to exercise as well and we're there.

 

[iHs]

My advice to peeps at the moment using bolus basal is to save yr tummy for the artificial pancreas with infusion sets etc and to use the arms or legs fot the bolus and backside for the basal. You need yr tummy in good condition

Sent from the Diabetes Forum App

 

[noblehead]

All looks encouraging and not before time too. The 10 hypo's a week seems an awful lot

 

[AndyS]

With some of the new MEMS based pumps coming down the pipeline soon the accuracy of the insulin delivery will be a trivial thing I think.
I still feel that the real key to getting them sensitive enough to cover rapid changes due to exercise is getting a better, more accurate and real-time, blood glucose monitor.
The software is a challenge, yes, but it's something that can be more easily fine tuned. You cant have software making good decisions based on data that is out of date by the time it hits the system, which is how pretty much all CGMs are these days.

I still agree, certainly promising and agree on saving your stomach injections sites, though I do inject pretty much everywhere

Keep up the good fight folks

/A

 

[ElyDave]

All looks encouraging and not before time too. The 10 hypo's a week seems an awful lot

I could easily rack up 5 a week if you include those during or post exercise, so far this week I'm up to three sub-4.0 readings since Saturday

 

[noblehead]

I could easily rack up 5 a week if you include those during or post exercise, so far this week I'm up to three sub-4.0 readings since Saturday

Yeah, I suppose with you been a runner Dave it will be hard not to have a few hypo's a week.

 

[ElyDave]

exactly, same with kids I would think with random spontaneous activity levels regardless of IOB

 

[noblehead]

exactly, same with kids I would think with random spontaneous activity levels regardless of IOB

and people with Brittle Diabetes.

When I was on twice daily injections I would hypo daily (if not every-other-day),

 

[AlexMBrennan]

I can't help but feel that it's a significant step back to go back to chasing highs - e.g. If you eat something, the system won't respond to it unitil after your BG has already gone up noticeably... Or maybe I'm missing something there.

Also, excellent celebrity soundbites: "When you do the maths, 10 per week means I may have had 20,000 hypos in my lifetime. That’s more than many people have had hot dinners" - that's because there are 7 days in a week, and thus no more than seven dinners per week.

 

[SamJB]

Maybe someone else could chip in here, but my understanding of the pancreas, is that this is what it does - detect an increase in BGs and release some insulin. Because it's got better direct access to the bloodstream than an injection, the effect of pancreatic insulin is faster than injected. This is pure intuition, BTW.

I guess one of the difficult thing in all this is responding to an increase in BGs, knowing that the insulin won't have any effect for a given (or unknown, unless calibration has be done) delay. Changes to insulin sensitivity due to exercise, time of day are going to be big challenges too. It's going to involve some complex analytics, but that's my line of work and I'm confident that it can be done.

 

[AlexMBrennan]

Maybe someone else could chip in here, but my understanding of the pancreas, is that this is what it does - detect an increase in BGs and release some insulin. Because it's got better direct access to the bloodstream than an injection, the effect of pancreatic insulin is faster than injected

Even if that's true, the artificial pancreas will be using the usual subcutaneous infusion set so it can't take advantage of that effect.

Changes to insulin sensitivity due to exercise, time of day are going to be big challenges too. It's going to involve some complex analytics

The goal of the new system is to completely disregard all information (e.g. If you are going to eat a burger, or go for a run), and the result is necessarily going to be worse than a system that takes this information into account - that's an inevitable mathematical fact.

In practise the difference might not be significant, and this artificial pancreas might be a significant improvement over MDI, in particular for patients who have trouble with hypos but the point remains that I feel they are intentionally crippling the system by discarding information.

 

[Spiker]

the point remains that I feel they are intentionally crippling the system by discarding information.

I agree. It's frustrating.

 

[Spiker]

Bah and humbug. I am annoyed about this whole situation. Closed loop systems have existed for 40 years. Why are they not available for patients yet? It's not a technology issue. It's just like the 1970s and 80s when the medical profession refused to give blood glucose meters to Type 1 patients as this would "confuse them".

I am very ****** off that I am walking round with a pump and a very accurate CGM, paired, such that the pump continually knows my blood glucose, yet it does nothing active with this information beyond alarms. It goes out of its way to pretend it doesn't know my BG. When I use the BG wizard it defaults to 4.4 even though it knows my actual BG and could trivially set the actual value as a proposed value. So if doesn't even support a person in the loop mode, let alone a closed loop. How many more years are going to be required before we get our hands on this kit? If infuriates me because I could write a safe and effective algorithm for a closed loop in under a minute. If I could patch the pump software that is exactly what I would do.

Even apart from the lack of a closed loop or even person in the loop support, why doesn't my pump/CGM pair:

- calculate my insulin:carb ratio on the fly from actual empirical data and notify me when that value varies much from what I have set?

- same with my correction ratio?

- same with my basal dose? Why isn't the pump measuring, every time IOB is zero and no carbs are active, and advising me when BG is not flat?

All of this stuff is TRIVIAL to implement. It can all have a person in the loop for safety. And what's the safety risk? Hypos? Remind me what we do when someone is in the worst risk category for hypos again? How do we address this issue? Oh yeah, we GIVE THEM A CGM! So anyone using a closed loop or semi open loop is already in the safest situation we can devise.

It makes me want to start my own company like the Omnipod guy, was it? My blood boils at the delay.

Sent from the Diabetes Forum App

 

[Engineer88]

Back in your box Spiker (joke)

 

[SamJB]

Even if that's true, the artificial pancreas will be using the usual subcutaneous infusion set so it can't take advantage of that effect.

The goal of the new system is to completely disregard all information (e.g. If you are going to eat a burger, or go for a run), and the result is necessarily going to be worse than a system that takes this information into account - that's an inevitable mathematical fact.

In practise the difference might not be significant, and this artificial pancreas might be a significant improvement over MDI, in particular for patients who have trouble with hypos but the point remains that I feel they are intentionally crippling the system by discarding information.

I agree entirely, Alex.

I am very ****** off that I am walking round with a pump and a very accurate CGM, paired, such that the pump continually knows my blood glucose, yet it does nothing active with this information beyond alarms. It goes out of its way to pretend it doesn't know my BG. When I use the BG wizard it defaults to 4.4 even though it knows my actual BG and could trivially set the actual value as a proposed value. So if doesn't even support a person in the loop mode, let alone a closed loop. How many more years are going to be required before we get our hands on this kit? If infuriates me because I could write a safe and effective algorithm for a closed loop in under a minute. If I could patch the pump software that is exactly what I would do.

One thing that I've noticed working in the world of software, is that any logical decision, or calculation, made by a person, can be made by a computer. A closed loop system is entirely possible. Difficult to do? Yes, but achievable nonetheless.

 

[ElyDave]

Coming back into this later on, I think the issue, as alluded to is the delays that are built into any artificial system.

Thinking logically what a pancreas/the "normal" endocrien system does is
- Detect change in blood sugar - decide on action based on rate and direction of movement - initiate change in insulin and/or other hormone secretion - repeat.

Even when detecting dlood glucose directly and changing rate of secretion directly into the bloodstream there are still tiem lags resulting in peaks/dips of blood sugar aroudn meals and exercise.

Teh artificial pancreas then is something like

Set basal rate OK - CGM detects change in blood sugar, rate and direction - CGM informs pump - pump algorithm checks rule set and chooses action - pump initiates action - CGM detects resulting change in BG, hopefully meets target, informs pump - repeat loop

The issues for me are
1) the additional lags built into the system by using proxies for blood sugar, which have an inherant lag
2) the lag in pump action
3) the type of control algorithm
4) wide range of pump delivery rates compormising accuracy at top and bottom end

I used to study these kinds of problems from a control pespective as a young chemical engineer and one of the hardest things to do was to tune a control loop, which is essentially what this problem is, needing elements of proportional control (feed back) and derivative and integral control (feed forward).

Yes, as Sam JB says it can be done, but the further you get from a directlly measured instantaneous value, the harder it gets.

 

[Spiker]

I see where you are coming from but the problem of a feedback system with laggy inputs is ubiquitous in engineering and a well solved problem both in theory and in practice. The solution is damping. You under react to the inputs, slowly and progressively. Remember we don't need to replicate the performance of an actual pancreas in order to get improvements that will render MDI and manually controlled pumps obsolete. I'm not interested in the problem of completely replicating organic pancreas levels of control. Not yet anyway. There will be probably three or four generations of operational closed loop devices before we are tackling that problem, which is admittedly a hard one. I just want to get to the first operational generation. I'm frustrated it's not already here because the current technology could easily do the job.

Sent from the Diabetes Forum App

 

[ElyDave]

I see where you are coming from but the problem of a feedback system with laggy inputs is ubiquitous in engineering and a well solved problem both in theory and in practice. The solution is damping. You under react to the inputs, slowly and progressively. Remember we don't need to replicate the performance of an actual pancreas in order to get improvements that will render MDI and manually controlled pumps obsolete. I'm not interested in the problem of completely replicating organic pancreas levels of control. Not yet anyway. There will be probably three or four generations of operational closed loop devices before we are tackling that problem, which is admittedly a hard one. I just want to get to the first operational generation. I'm frustrated it's not already here because the current technology could easily do the job.

Sent from the Diabetes Forum App

I agree, and that's really my point.

We could get a first generation fairly easily in all probabaility, and try to fine tune the PID loops (the lead and lag bits, and damping) when in use.

To be honest, the lead and lag bits should really be specifically tailored to each user, which would mean the propietary software being open to limited fiddling by the user. That may be as much of a stumbling block as the engineering. Meanwhile we can improve both pumps and CGM to try adn remove the lagginess

If I was looking forward to now from 20 years ago, perhaps my career woudl have taken a different direction into biomedical engineering instead of oil and gas safety and environmental management

 

[Spiker]

My grumpy frustration has driven me to research pump hacking. I think from available information put out by hackers, I could hack a pump and hack a CGM so that a 3rd device could sit in the middle, read the CGM (by hacking) and control the pump (by hacking). The whole system then would form an artificial pancreas. My message to the pump and CGM companies would then be, as the great Eric Cartman once said, "Screw you guys!" ;-)