I’m skeptical too. CGM accuracy has to become MUCH more accurate and reliable before closed loop would be viable. A closed loop would be like cooking on a gas or electric stove instead of an open fire… much easier to control the heat, but it still needs attention. I don’t think, in the beginning at least, that it would require any less attention than a pump/cgm. I certainly wouldn’t want a system that did anything without my approval, so it would be something different to pay attention to, a different set of variables, and still require intelligent thinking and interaction.
I’d love it if one of those good medical engineering brains out there could figure out a safe way to make a CGM that measures actual blood glucose instead of interstitial fluid. That, for me, would be a huge leap forward. I believe the technology is there. I had a port for chemo that went directly into a blood vessel.
There is a basic problem in closing the loop, insulin only lowers blood sugar. A responsive control loop requires an ability to both lower and raise blood sugar. Initial attempts using insulin and glucagon have made some starting progress, but not enough to be safe. Another alternative would be to augment with glucose. But even when these obstacles are surmounted, any closed loop system would need to have multiple layers of defense so that any single failure in the system enables it to gracefully fail. If your sensor fails, what happens? In the end, we are unlikely to see an approved closed loop system that doesn’t involve multiple redundant sensors and elements or a sensor that never fails. This is also a huge engineering challenge.
Closed loop control has many problems. Inaccuracy of the BG sensors, Lag in the sensor reading. (Dead time compensation algorithm can cope some what with lags but control quality suffers.) Lag in the insulin action. (Injecting directly into a vein does not sound safe) I/C ratios that change. ( some adaptive control may cope with this somewhat) Illness. You can always add feedforward control to cope with meals and illness. And of course dual action control systems with insulin and glucogon is the way the body works.
And of course there is always the problem of some sort of system failure.
I was also reading some implantable pump stuff that sees promising and at risk for extermination. This is really interesting as the insulin is secreted into the peritoneal cavity which is more psysiologic. Better control according to study participants.
Totally agree bsc. We need that simplified and 99% accurate sensor. Not to mention the feedback mediator hormone like Glucagon. Lots of kinks to iron out! Interesting nevertheless…
I read a blog on the JDRF website from a guy who had the same reservations as you. He did not feel a “machine” could do the job he does. He still signed up for a clinical trial. During this trial they gave a meal with wine and after put them on a treadmill (can’t remember the exact protocol but something like this- know to frequently cause hypos). They gave the human a couple of chances and the “machine” a couple of chances. The guy ended up going hypo consistently where the machine only hit a very mild low once. He came away with the experience with a whole new mindset about whether he could give a “machine” the control.
At least initially you will still have to bolus so you will have the oppurtunity to give pre boluses and I imagine as they advance they will give options to allow the user as much control of the system as they want.
Steve Edelman (of TCOYD fame) told me that he had a peritoneal insulin pump as part of a study and that it offered unbelievably good control…until he got a nasty infection and had to have it removed. Still would probably be a great tool if we could make it work.
No doubt. I know with Peritoneal Dialysis infection rates are high, and nasty.
Very cool. And the glucagon micro dosing again. Thanks for the info.
I have a background in chemical engineering and am very familiar with doing this type of control in industrial applications, which have many of the same characteristics and issues as BG control. As mentioned in some earlier posts the main challenges are
- CGM accuracy and lag time
- Slow acting insulin
- Lack of automated failsafe (glucagon)
I see the CGM issue as a minor problem, the technology is advancing at an incredibly fast pace and within a few years I believe that CGM will be accurate and reliable enough to dose insulin from (maybe Dexcom G5 or implantable sensor?), as for the CGM lag, this is not a major issue and can be overcome very easily with fairly simple predictive algorithms, which are currently available.
The slow acting insulin is the major obstacle, since it stays in your system for up to 4 hours, the controler (computer) needs to sense a change in BG, then give the appropriate amount of insulin, unfortunatly the glucose is hitting your system much faster than the controller can keep up with, since it takes 4 hours for the insulin to react completely, this means that you will have large post meal spikes(much bigger than you currently have by dosing insulin manually). Even with the next gen ultra rapid insulins, it is not possible to get a SC injection to difuse through your fatty tissue into the bloodstream anywhere near fast enough to keep up with automated insulin dosing for meals. Closed loop will never happened using subcutaeous insulin delivery. The only solution to this is to have insulin delivered directly into the bloodstream, the same way that a non diabetic gets it, that way it takes effect in a couple minutes and is used up competely in 10 - 15 minutes allowing the controller to re-analyse and re-dose insulin every 10 minutes or so. Implantable pump or diaport are the only options to achieve this.
I am not up to speed on the latest research into protable glucagon delivery, so I can’t comment on how soon this will be available. Having the glucagon avaiable is not competely neccessary, but it would be a very nice feature, as it would allow the controller to be much more aggressive with insulin dosing and then counteract the insulin when a low BG is predicted, which would be a major help in reducing the time spent with BG above target and would give more confidence that severe lows would not be an issue.
I believe that the AP will be done in 3 phases
- The only thing that the AP could do effectively with today’s CGM and external insulin pumps is make minor adjestments to fasting BG values, which would still be a great way to manage overnight BG control and would be a blessing for people who have issues with poor overnight control.
- This is likely in the 2015-2020 timeframe, when CGMs are truly accurate enough to dose insulin from, ultra rapid SC insulin is approved and portable glucagon is available for pumps. At this point a controller could manage BG from a meal with the user giving a carb estimate and bolus type (normal, square, super, etc) before the meal. The controller would read BG values after the meal and either give insulin or glucagon as needed. This will not work perfect, but it might be possible and with proper user understanding and input, it could be done. This system would also handle fasting BG mcuh better than in phase 1.
- Completely automated closed loop, this will happened when GCMs are accurate enough to dose insulin from and the insulin is delivered directly into the bloodstream, glucagon would not be needed, as cutting off the insulin delivery would quickly bring up BG very quickly since the insulin on board would be gone within 10 minutes.