Components of an Artificial Pancreas (AP)
I'm going to discuss an AP built out of the technology that we have now, so an "off the shelf" pump connected to an "off the shelf" CGM, with some extra software on the pump to handle the control functions. A successful artificial pancreas is going to involve several different components that can be developed and tested separately. Understanding these components will help help track progress toward the final goal. The components are:
- A pump that can deliver small enough doses of insulin.
- A continuous glucose monitoring system (CGM) that is accurate enough.
- Insulin which acts quickly enough.
- Software that can take the data from the CGM and do a good enough job of automatically calculating the commands to send to the pump.
- Government approval of all of the above.
- Higher density insulin, so that a longer supply can be stored inside a person.
- Enough miniaturization so that the pump, CGM, software, and insulin can fit inside a person.
- And government approval of all these components.
Part of the uncertainty in the status I describe below, is that these components work together. So imperfections in one might or might not be overcome by advances in the others. For example, if CGM were perfectly accurate, that would make the control software much easier to create. Similarly, the faster acting the insulin that can be used, the easier the control software is to write, because it does not need to take into account the delay in the insulin or build up of insulin in the body. So the question is not do we have good enough CGM. It is more complex than that: do we have a good enough pump, CGM, software and insulin, so that together then can form an AP. And the same is true for all the major components.
For pumps and fast acting insulin, we are in good shape. Today's mass produced pumps are likely precise enough to support the first generations of AP, and currently available insulin is likely fast enough.
For the computer control of an AP, we are very close. There are several computer algorithms which have been tested in different clinical trials, and which seems to work well. The biggest hurtle remaining is to get the software good enough to work with the current generation of CGM.
For CGM, we are not there yet. There are two problem areas for CGMs, both important. First there is accuracy. BG meters in general are only accurate within 20%, but CGM are not even that accurate. No CGM has been approved as a replacement for regular BG meters for exactly that reason. One of the problems with CGMs is that the interstitial liquid that they measure sometimes has higher glucose levels than the blood which standard meters measure (and it is blood glucose levels that really matter). This means that the current generation of CGM devices can sometimes measure higher glucose levels than really exist (which would cause an AP to put in more insulin than it should).
But there are also non-accuracy related issues, which may be a larger problem. Many CGMs are not comfortable to wear, often fall off, have calibration problems, need to be replaced often, etc. My memory of comments in brave buddies, from people who tested CGMs, and from those who now use them, is that there are occasional worries about accuracy, but much more complaining about replacement, falling off, skin space, pain, recalibration work, etc. The bottom line is that no AP will be successful if the CGM component of it is constantly falling off, etc.
As for an implanted artificial pancreas:
There have been European tests of U-1000 insulin (which lasts ten times as long as the U-100 insulin commonly used). And this insulin was enough to operate an implanted pump for 3 months between refills. So I think we have the insulin needed, although it is not yet FDA approved in the US. There are currently clinical trials in the US on U-400 insulin, also.
Also, there have been clinical trials of an implanted pump in the US, including a pump produced by Minimed. So that too, is within sight, although not yet US FDA approved.
For a technical review of where we are, especially in terms of CGM systems, here is a whole journal issue focused on that question:
Six Stages to an Artificial Pancreas
The JDRF framework for a pump is based on six stages as described below. Note that stages 1 and 2 are more of emergency shut off functionality than artificial pancreas functionality, but you need to start somewhere! Indeed, I think it is fair to say that only at stage 4 is there a limited AP, and stage 5 is a full AP:
Stage 1. An emergency shut off when the patient experiences too-low BGs.
Many of us live in fear of extreme lows in the middle of the night which lead to seizures. In the worst case, type-1 diabetics occasionally die in the middle of the night from too-low BG. So a straightforward safety improvement which could be made to a pump, if it was attached to a CGM system would be to shut down the pump if the patients BG levels were too low for too long. Research has shown that most (although not all!) night time seizures are preceded by several hours of low BG numbers. So by simply telling the pump to stop putting in more insulin into the patient, many of these seizures could be prevented. Also, this plays into the strengths of the current CGM technology (the accuracy problems usually effect high BG measurements, not low ones). No any complex software is needed, and if the AP is wrong, BG may go up more than it should, but that is a relatively minor problem. The problem it solves ("dead in bed") is real (if rare) and terrifying. Finally, the FDA can view it as a simple safety interlock, for approval purposes.
Stage 2. A shut off when the patient is heading to a situation where they will have too-low BGs.
This is similar to stage 1, except that it uses software to predict when a low BG situation is likely to happen in the future, and causes the pump to dose less before the low BG situation occurs, hopefully avoiding it. This is much like stage 1, except that it requires more complex, predictive software.
Stage 3. Improving BG levels at all times (avoiding both too-low and too-high BG levels).
The goal of this stage is limit the time the patient spends with either too-high or too-low BG levels. This stage is not trying to keep the patient at ideal BG levels, rather it is just trying to avoid unhealthy BG levels.
Stage 4. Limited control with meal announcements ("overnight control")
At this stage the AP can control BG, aiming at optimum levels in the relatively undemanding time while the patient is asleep. And if it is notified ahead of time about eating. This stage does not need to worry about exercise, or emotional changes, or unexpected eating.
Stage 5. Fully reactive control of insulin.
This is what most of us consider to be an artificial pancreas. It keeps BG levels in the healthy range without having to do separate blood checks or counting carbs.
Stage 6. Fully reactive control of insulin and other hormones.
Beta cells produce other hormones besides insulin, and a better AP will provide those hormones as well as insulin, so this stage is icing on the cake.
Basically, steps 1 and 2 are designed to prevent serious low BG events, especially at night.
Steps 1, 2, and 3 are really focused on avoiding bad BG numbers, rather than keeping the patient with good BG numbers. They are fundamentally trying to avoid mistakes, rather than create a better system.
Steps 4, 5 and 6 are trying to improve what we've got.
Step 4 is sort of a night time AP, while 5 is an all the time AP.
Step 6 is a complete solution for multiple hormones.
How I Plan to Cover Artificial Pancreas Research
Except for this posting, I don't plan to cover each component separately, because that would be a lot of work, and I don't think it would be very useful. Instead, I will cover human trials of complete APs as they are tested, noting which components are in use for each test. I don't think it is useful to report of every trial of U-400 insulin, and U-500, and U-1000, etc. Instead, I will just note if an advanced insulin is used in a specific test of an AP.
Also, I'm probably going to limit my coverage to clinical trials where each patient uses the AP for at least 24 hours (not 24 consecutive hours, but 24 hours total), at least to start. There are a large number of clinical trials aimed at gathering data on a very specific algorithmic questions, and I don't plan to report on all of those. They are very important to the end goal, of course, but there are to many of them, and each one is individually such a small step, that I don't plan to cover them. These advances get incorporated into the longer studies which I will cover.
One problem that I have already hit, is how to name these studies. I don't want to name them after the company or institution running them, because one of those could be running many AP trials at the same time (Minimed, is running more than one right now.) Many of them have one researcher who developed the software being tested, and another who is running the clinical trials. Which one should I use? And what if several people worked together on the software, or in running the trials? I could name them all based on their Clinical Trial number, but that seems so sterile. I'll figure something out. (Maybe they will all get code names or something. :-)
In the next 2-3 weeks I will post a summary of clinical trials posting, which will give a very short status of a large number of current clinical trials involving APs. After that, I'll make follow up postings covering trials, much like I do with non-AP clinical trials.
Why I have not posted about the Artificial Pancreas before; Why I will Now
I have not previously posted about an Artificial Pancreas, for three reasons:
- Many don't consider it a cure at all.
- There was not that much news on it.
- I didn't have a framework to measure progress, in the same way I had a framework for drugs.
First, I was particularly worried that a posting referring to an AP as a cure would trigger a long a pointless argument about weather an AP was a cure or a crutch. If you gave someone a new heart valve, have you cured their heart condition? If you gave someone a brace, had you cured their withered leg? It is a philosophical question and different people will have different answers. Even worse (for this discussion) I doubt one side could convince the other they were right, because it really is a philosophical difference, so the email discussion could last forever.
Personally, I had always considered an IAP as a cure, but I was not sure if an AP was a cure or not. However, in the past I also thought that this was a minority opinion, and that most people felt that APs were not cures, not even IAPs. However recent discussions with other parents of children with diabetes has convinced me that many people consider APs to be cures, or at least important enough to be very valuable even short of a cure.
The one thing that I don't want to happen, is a long email argument about if an AP is a cure or not. There is no doubt in my mind that some people will consider it a cure, while others will not. Therefore, I'm posting on it's progress for those who consider it a cure, and people who don't should ignore those posts.
Second, in the last four months or so, there has been several reports of progress toward an AP. I think that Weinzimer's work at Yale has been generating much of this press coverage, but others are moving forward in this area as well.
Third, the JDRF has published a framework for evaluating APs. These are the stages that I've discussed above. This framework is based on APs working in progressively more and more realistic (and difficult) situations. It is not based on progress in individual components of an AP, and so it is perfect for me to use as I report on progress.
In the past I have thought about covering APs by covering progress in each major component of an AP. However, this tended to scare me off because of the work involved in covering several different components all of which are making many small, incremental improvements all the time. By limiting my coverage to complete AP testing, I'm hopeful that it will be a reasonable amount of work, and our kids will end up using a complete AP anyway.
Thanks to Dr. Mary and Dr. Steve (both at Bearskin Meadows) for getting me to think about these issues, and providing some of the information used here. All mistakes are my own.