Taking the Artificial Pancreas Home, 24 hours per day!

By Adam Brown and Kelly Close

Want more articles like these, right in your inbox?

Sign Up Now!

Twitter Summary: Wearing UVA’s DiAs #artificialpancreas in a 3-month study testing overnight and 24-hr closed-loop at home; takeaways, surprises, next steps

“Control” is a diabetes word that comes with a lot of baggage. It implies that perfect blood sugars are possible. It implies that patients can manage all of the factors that affect blood sugar. And on a deeper level, “control” begs an interesting question: can a real-world computer algorithm ‘control’ blood sugars and insulin delivery better than a human?

Over the past few months, Kelly and Adam have had an opportunity to explore this question as part of an awesome trial testing the University of Virginia’s DiAs artificial pancreas system at home (also called the “closed-loop” or “automated insulin delivery”). This research platform consists of an Android smartphone running a control algorithm, a Roche insulin pump, and a Dexcom G4 Platinum CGM with a special Bluetooth box. Though we have been remotely monitored by the UVA and Stanford teams while on closed-loop at home (the phone sends the CGM and insulin data to a server in real-time), we are generally on our own with the system – we sleep in our own beds, eat our normal diets, go to work, exercise, drive, and have to troubleshoot problems. Because it’s still research, both of our significant others have been trained on the system and equipped with glucagon – just in case. It’s amazing that this type of system, which was tested in tightly controlled hospital settings just a few years ago, is now out in the real world! The artificial pancreas has come very far in a short amount of time.

This article shares our positive and learning-filled experience wearing the DiAs system overnight and 24 hours per day at home over the past few months. We have decided to separate overnight from daytime closed-loop wear, since the experience is very different. Overnight wear has been amazing, while daytime wear has been positive overall but a learning experience for both of us on many fronts. Below, we also discuss some of the biggest surprises, our biggest worries, what other groups in the field are doing, and some of the key questions on our minds. Please share your questions or thoughts on Twitter!

How Does DiAs Work?

1. An Android smartphone with a control algorithm wirelessly receives the Dexcom CGM value every five minutes.

2. An algorithm running on the phone calculates how much insulin to deliver (based on the current and predicted blood glucose, and insulin on board). If glucose is predicted to go too high, additional insulin is given. Insulin is reduced or suspended if glucose is predicted to go too low. The glucose target varies by time of day (more aggressive at night, more conservative during the day).

3. A command is sent wirelessly to the pump to give the calculated amount of insulin.

4. This repeats every five minutes while closed-loop is running during the day or at night. During the day, we still enter meal information into the system (number of carbs), which makes this a “hybrid closed-loop” (“treat-to-range”) system.

At Night: Dear DiAs, You Are Way Better At This Than Me

Bottom Line: We’d take the overnight version of the DiAs research system today! Waking up at 120 mg/dl nearly every morning is powerful.

Overnight, the DiAs control algorithm aims to gradually bring blood sugars down to 120 mg/dl by 7 AM. When it works, it’s incredible from both a diabetes and quality of life perspective. Reliably waking up at 120 in the morning feels like a one-mile head start on the daily marathon of blood sugars: no waking up low and eating the entire fridge, no waking up high in a state of insulin resistance and needing to take lots of insulin. Staying in range for most of the night – without significant hypoglycemia or hyperglycemia – also improved our sleep quality and energy on the following day. Last, we both found that insulin needs could vary substantially from night to night (e.g., very active days vs. sedentary days) – DiAs can cope with these changes reasonably well, since it is reacting to glucose every five minutes.

Both of us would take this overnight system in its current form, as it feels so much safer than how we manage diabetes on our own insulin pumps right now (“open-loop”). The system is especially good at avoiding hypoglycemia overnight, the scariest time for us as patients. On his own pump therapy, Adam spent an average of 4.4% of the time at night at values less than 70; during the two closed-loop phases of the study, he spent just 2.6% and 0.6% of the time low. Kelly’s numbers at night also changed meaningfully for the better – not only did she settle at an excellent number just above 100 mg/dl a couple of hours into each night, she also really loved all the “soft landings.” Kelly experienced virtually no hypoglycemia on nighttime DiAs, whereas on open-loop therapy, she had a lot of unstable numbers.

Both of us were a bit nervous in hitting “Closed-loop On” for the first time overnight, but the remote monitoring and our partners by our sides made us feel safe. Soon, it became second nature to run closed-loop overnight, and before long, times when we couldn’t run overnight closed-loop (e.g., traveling, CGM sensor died) actually felt scarier. It’s amazing how quickly we both built trust in the overnight system! Both of us really support “night only” systems coming to market first.

We did find a few areas where DiAs sometimes struggled or was frustrating overnight:

1. Alarms. Since this is a research study with remote monitoring, the alarms are loud, frequent, and annoying – of course, the goal is to demonstrate safety, not to optimize the product profile for a commercial product. We look forward to future systems that will operate in the background at night – the goal is not to wake up at all, even when something is going wrong. For instance, if the CGM sensor drops out and closed-loop can’t continue, the system could default back to the pre-programmed basal rate without alarming. Hopefully, alarms will also be more customizable, allowing patients to set their own preferences.

2. Eating high-fat meals close to bedtime. The system often didn’t increase insulin enough to bring Adam down into range, which was understandable for safety reasons. Perhaps future versions of DiAs will allow patients to set their own targets or tell the system to be slightly more aggressive at certain times. 

Daytime: Human + DiAs = Who Is In Charge?

Bottom Line: Good at avoiding hypoglycemia, though the daytime algorithm is more conservative. Users must find a comfortable balance between automation and manual input.

During the day, DiAs is more conservative, since there are more disturbances that can affect the system (e.g., meals and exercise). The system targets a conservative blood sugar of 160 mg/dl during the day, and seeks to avoid hypoglycemia (<70 mg/dl) and hyperglycemia (180 mg/dl). If blood sugar is predicted to go out of the range of 70-180 mg/dl, DiAs will give a small correction bolus or reduce/suspend insulin. Meals are entered manually on DiAs (number of carbs), alongside the current meter glucose value (for a correction). The system also permits manual boluses of insulin and takes these into account.

During waking hours, Kelly has seen that eating food that isn’t so diabetes friendly still means negative numbers – that hasn’t changed! But, the high numbers tend to go to a “soft landing” that is really reassuring, and Kelly has had very little hypoglycemia during daytime DiAs. Far less hypoglycemia also translates to fewer unexpected highs. Overall, Kelly found that the DiAs algorithm was more conservative during the day than the bionic pancreas – of course, the addition of glucagon allows Drs. Damiano and Russell’s system to be a bit more aggressive than the insulin-only DiAs system.

The “hassle factor” involved in connectivity is much more apparent during the day, but if that were not a problem, Kelly would want DiAs 24/7. Even so, she very much appreciated the chance to help doing research!

Adam’s daytime glucose values have been very similar on DiAs vs. his normal open-loop days, though he did not expect much of a change – he rarely forgets boluses, tends to eat few carbs, and catches out-of-range blood sugars quickly (he wears CGM). As a result, he didn’t expect a significant improvement with DiAs during the daytime.

For Adam, a couple weeks of using daytime DiAs brought an important realization: his ingrained diabetes management habits were more aggressive in dosing insulin than how DiAs approaches things. He finally realized that DiAs performed best during the day when he left it alone and didn’t try to: (i) manually correct high blood sugars; (ii) ignore the bolus screen warning messages (e.g., “if you take 2 units of insulin, you are predicted to go low”); and (iii) eat too much to correct lows (the incessant hypoglycemia alarms make it easy to do this). Indeed, when Adam was too hands on, he often saw consecutive rises and falls in glucose during the day (going from 80 to 180 back to 80). Essentially, DiAs was (appropriately) responding to his aggressive insulin dosing by seeking to avoid hypoglycemia. The alarms would then cause Adam to eat too much, resulting in a subsequent high and the process repeating. Lessons learned:

• There is more of a learning curve for daytime closed-loop;

• Closed-loop systems must tradeoff automation with human intervention, and the balance might be different for the wide variety of patients out there;

• Letting the system correct highs and avoid lows generally resulted in better outcomes vs. manually tinkering with the dosing. However, this took some patience and getting used to.

In the upcoming extension phase of the study, Adam plans to take a more hands-off approach to using DiAs during the day. He hopes it will reduce the burden of managing his diabetes, while keeping his time-in-range roughly the same.

Biggest Study Surprises

1. Connectivity. Since the DiAs team is using off-the-shelf devices for research purposes (Roche pump, Android phone, Dexcom CGM receiver with special Bluetooth box), getting everything to connect can be a hassle. This required a lot of troubleshooting and repairing/rebooting/replacing devices. We would caution that this isn’t a major worry, as final commercialized products will benefit from a new wave of sensors and pumps designed for integration with consumer electronic devices. Some systems will also integrate the algorithm and CGM receiver right into the pump. In the extension phase, Adam has seen better connectivity with the Dexcom Share Receiver, which eliminates the need for the Bluetooth box.

2. The need for adaptability. The DiAs algorithm uses the normal, open-loop basal rates as a background guide. In other words, our normal basal rates serve as a guardrail around the insulin dosing to cap how much additional insulin the system can give. At one point during the study, Adam got sick and needed additional insulin overnight – ideally, the system would have realized this and increased insulin dosing on its own. But since it uses the open-loop basal rates as a baseline, Adam needed to manually increase the pre-programmed basal rates to give the system “permission” to give him more basal insulin. Future versions of DiAs will hopefully be more adaptable and able to cope with such changes in insulin sensitivity. 

3. How hard researchers work! The team at Stanford, led by Drs. Bruce Buckingham and Trang Ly, has been absolutely incredible. They must be on call in the wee hours of the morning every night, and both of us have certainly received texts at all hours of the day! We both gained a major, major appreciation for how hard researchers are working, how committed they are, and how much it takes to move this field forward.

What’s Next for DiAs?

DiAs has been tested by over 300 people with type 1 diabetes around the world to date. The system will soon complete its last round of pilot studies, and two large-scale trials are planned to begin later in 2015 and in 2016:

• Project Nightlight is a pre-pivotal trial of overnight closed loop. The study will take place at UVA, and with additional support from JDRF, will expand to Mount Sinai (New York City) and the Mayo Clinic (Rochester, MN).

• The international Diabetes Closed-Loop (iDCL) Trial will be a large-scale, long-term study of 24/7 closed-loop control. Very notably, the iDCL Trial has been designed in collaboration with FDA and with industry as a “pivotal trial” of closed-loop technology, meaning it could be used to support regulatory approval. The study will take place at UVA, UC Santa Barbara, the William Sansum Diabetes Center, Stanford University, the Barbara Davis Center (Colorado), Mayo Clinic (Rochester, MN), Mount Sinai (New York City), the Academic Medical Center of Amsterdam, and the Universities of Montpellier (France) and Padova (Italy).

In addition to these studies, the creators of DiAs have formed a company, TypeZero Technologies (Charlottesville, VA), and licensed the DiAs software and algorithms to develop a commercially available closed-loop system. TypeZero will pursue two routes to commercialization: (i) embedding artificial pancreas software in insulin pumps in collaboration with others; and (ii) a standalone, smartphone-based system that will communicate wirelessly with existing devices. The idea is a system that can adapt to the preferences of each person (e.g., “I want overnight only”) and to signal availability (a “pump mode” when no CGM is available). The team’s ultimate goal is to establish the artificial pancreas as an adaptable wearable network surrounding each person in a digital treatment ecosystem – not a single-function device. 

Concluding Thoughts

Participating in this trial has been a fantastic learning experience for both of us. It’s amazing to think closed-loop systems are now in very real-world, home trials, and on the cusp of being launched as real products! Our research experience reminded us of how much potential these devices have to improve blood glucose and to reduce the burden of managing diabetes. Early versions will have their challenges, but we have no doubt that these systems are coming, that they will really help many patients with type 1 diabetes, and they will get so much better over time.

JDRF generously extended our study for six months, so we will gain even more experience wearing and using the system!

Our sincere thanks go to Stanford, UVA, JDRF, the FDA, and all those working day and night to close the loop! Adam and Kelly both feel incredibly grateful to have been given the chance to be part of this research.

Who Else is Closing the Loop and How Fast Are They Moving?

Key Questions for the Artificial Pancreas

Are patient expectations too high? If we expect too much out of first-generation artificial pancreas systems – e.g., “I don’t have to do anything to get a 6.5% A1c with no hypoglycemia” – we might be disappointed. Like any new product, early versions of the artificial pancreas are going to have their glitches and shortcomings. Undoubtedly, things will improve markedly over time as algorithms advance, devices get more accurate and smaller, insulin gets faster, infusion sets improve, and we all get more experience with automated insulin delivery. But it takes patience and persistence to weather the early generations to get to the truly breakthrough products. We would not have today’s small insulin pumps without the first backpack-sized insulin pump; we would not have today’s CGM without the Dexcom STS, Medtronic Gold, and GlucoWatch; we would not be walking around with smartphones were it not for the first brick-sized cellphones. This trial recalibrated our expectations a bit – these systems are going to be an absolutely terrific advance for many patients, but they will not replace everything out of the gate. Let’s all remember that devices need to walk first, then run, and it’s okay if the first systems are more conservative from a safety perspective.  

Overnight-only or 24-hour? There is some debate in the field on this topic, since some believe a night-only system brings additional challenges – e.g., What if someone wears the night-only system during the day? What about people who work the night-shift? Such challenges would not be present during the day, and as this study showed, 24-hour systems do work! However, daytime closed-loop brings higher patient expectations (we can see what the system is doing!) and other unique challenges, such as meals and exercise. The idea of an overnight-only system first is somewhat attractive, especially to build early experience and get an easy win – closed-loop systems at night pretty much always beat patients in the real world. For the most part, everyone in the field is developing 24-hour systems, though we imagine some patients will prefer to wear them at night-only.

Insulin-only or insulin+glucagon? Ultimately, we believe that the question is partially one of patient preferences. There will be some patients who may want the extra glycemic control offered by the dual hormone approach and will be willing to accept a bit more risk or a more aggressive algorithm. We believe the Bionic Pancreas could be especially helpful for those with hypoglycemia unawareness, a sizeable percentage of patients. If the Bionic Pancreas makes it to the market by 2018 with insulin and glucagon, some patients will want to give it a try – we believe a range of options is a good thing for people with diabetes, since all systems and products have pros and cons. Ultimately, cost considerations may present the largest factor in adoption. The Bionic Pancreas certainly brings multiple cost elements to consider – glucagon, a dual-chambered pump, custom infusion sets, potentially higher training, etc. It’s hard to know at this point how the relative costs/benefits will exactly compare.

What’s the right balance between automation and human manual input? The Holy Grail is a fully-automated, reactive artificial pancreas that requires no meal or exercise input. But insulin ideally needs to get faster to make that a reality. For now, daytime systems need to deal with balancing human input with automation, and as discussed above, there’s an associated patient learning curve. How much should artificial pancreas systems ask patients to do? How do we ensure patients do not forget how to manage their diabetes (“de-skilling”) as systems grow in their automation abilities?

How will academic groups commercialize their devices? This continues to be an open question for groups like the Bionic Pancreas and Cambridge – who will build and supply the devices, secure FDA approval, manufacture, support training and customer service, etc.? As noted above, TypeZero Technologies licensed DiAs to commercialize a closed-loop system. In mid-April, Medtronic also licensed DreaMed’s artificial pancreas algorithm, which was developed by an academic group (DREAM) that later formed the startup. Of course, Medtronic has an advantage, since it has both a pump and CGM in-house; other insulin pump companies will not only need an algorithm, but need to find a CGM supplier, such as Dexcom.

What fraction of patients will be willing to wear some type of artificial pancreas system? Right now, many estimate that ~30% of US type 1s wear a pump, and about 10%-15% wear CGM. There are a lot of reasons why that may be the case, including: cost; hassle factor; no perceived benefit; no desire to switch from current therapy; wearing a device on the body; alarm fatigue; etc. Will an artificial pancreas address enough of these challenges to expand the market?

Will healthcare provider embrace closed-loop systems? Today, healthcare providers lose money when they prescribe pumps and CGM. We need to make sure that closed-loop systems make providers’ lives easier, not more complicated.

Will there be a thriving commercial environment and reimbursement? It’s extremely expensive to develop and test closed-loop systems, and companies will only develop them if there is a commercial environment that supports a reasonable business. Reimbursement is a major part of that, and it’s hard to know if insurance companies will pay for closed-loop systems – we likely won’t know until there is longer-term data on these devices, especially changes in A1c or severe hypoglycemia. We are optimistic that reimbursement will be there, especially if systems can simultaneously lower A1c and reduce hypoglycemia.