April 14, 2019; Hayward, CA; Highlights - Draft

Carb DM’s annual Bay Area Diabetes Summit day kicked off with a stirring ode to changing lancets (to the tune of Frozen’s “Let it Go”), courtesy of the day’s MC and self-proclaimed Weird Al Yankovic of diabetes: Bigfoot’s Ms. Melissa Lee. After some digging, we later discovered that she has 13 such parodies up on YouTube! Today's meeting had 560 registrants, up from 483 last year – we love to see this event growing, and hope even more people can attend in the future.

Read on for our top six highlights from a very packed day!

Top Six Highlights

1. Tandem Control-IQ 6-Month Pivotal (Ages 14-75) Complete, Now Recruiting for Peds Pivotal; “Lot of Hope for the Adolescents” -> One Had A1c Drop 10.7% to 7.4%; Recipe for Full Closed Loop = Meal detection + Pramlintide?

Stanford’s Dr. Bruce Buckingham confirmed that, in line with expectations, the pivotal study of Tandem’s six-month Control-IQ/G6 hybrid closed loop pivotal in ages 14-75 has wrapped up: “We just finished the 168^th subject last Tuesday – data from all the sites is now being compiled, and going to be sent to FDA.” Per Tandem’s 4Q18 call, these results will be presented on the Sunday of ADA 2019, and the system is presumably still on track for a launch between “summer and end of Q3” (September). Dr. Buckingham added that Stanford is now recruiting for a study to investigate the system in 7-14-year-olds (note: Tandem said 6+ on the call), which Tandem has said could be wrapped up and approved by FDA in time for a pediatric launch right out of the gate. If not, the younger age indication would roll out after 14+. Citing a specific participant in the pivotal, Dr. Buckingham said Contol-IQ gives him a “lot of hope for the adolescent age group”: “A kid came in with an A1c of 10.7%, and that’s with bolusing about twice a day. Bolusing now same frequency, six months in to the study, his A1c was 7.4% because of the automatic corrections every hour that help to bring blood glucose down. That’s a huge improvement.” This system has proven itself in this age group (specifically 13-18 year-olds) before, boosting time-in-range in a >48-hour ski study (n=25) from 50% on SAP to a remarkable 73% on closed loop, translating to 6 more hours per day in range (p=0.01). Needless to say, we are eager to see the full study results. As a reminder, this system gives a hourly automated correction boluses during the day, targeting 112.6-160 mg/dl, and at night, the target is gradually tightened to 112.5-120 mg/dl, but no automatic corrections are administered.

• Forlenza et al. recently published data in DT&T from a 3-day, n=24 home use study comparing Control-IQ with SAP in 24 children ages 6-12. Compared to SAP, Control-IQ increased time-in-range by 4.4 hours per day (71% vs. 53%), decreased mean glucose by ~26 mg/dl (154 vs. 180 mg/dl), and only came with ~25 minutes of time <70 mg/dl per day (vs. 13 minutes in SAP; difference not significant). This system was active for 94% of the study. These results bode very well for the upcoming pivotal.

• Dr. Buckingham stated that there are “over 200,000” people using 670G now. The last company update in February touted 157,000 trained, active users; it’s possible that Dr. Buckingham may have updated information, or that he was rounding up/extrapolating (especially in light of recent international launches). He described the system as useful, “but you really need to know how to use it. It works well overnight, but there’s still a lot of work during the day. Stanford is currently recruiting for a six-month RCT comparing the 670G against “standard control” – they are specifically looking for people on MDI.

• Dr. Buckingham’s group is also in the middle of an adolescent Airbnb study with Insulet’s Omnipod Horizon system. Why can adolescents be such a difficult group for closed loop algorithms? “You don’t know what it’s like to test adolescents in a closed loop system, because they do things…last night a guy had a bowl of fruit loops, it was huge, with milk that filled it to the brim, it must’ve been a cup and a half of milk. And he had it at 1 o’clock in the morning. XCheck the overnight control. Truly free range.”

Selected Questions and Answers

Mr. Lane Desborough (Bigfoot): Any comments on hypoglycemia with pramlintide co-infusion?

Dr. Buckingham: You can get hypos, particularly if you start out low. If you give a pretty good bolus for food, then the pramlintide will stop the stomach from emptying, so you’ll end up getting a low. The way to get around that is to give a small bolus up front, then much more of an extended bolus. Pramlintide can also cause nausea, probably dependent on how much you give in the bolus – you could probably have a limit on how much is given at the beginning and then give more later. These are all things that need to be played with, and I’d really like to begin playing with these parameters. The Klue watch reminder has worked well. Maybe you can tell when someone’s eating, increase basal, give a bolus, then see the glucose go up – that’s the second confirmation that someone’s eating – then have pramlintide on board, so maybe you don’t need to bolus. Maybe you can be an adolescent for the rest of your life.

Q: How will diabetes education need to change to keep up with all of this new technology?

Dr. Buckingham: It is a whole new world in technology. By the end of the year, we’ll have two hybrid closed loop systems on the market, and 3-4 by the end of 2020. Each has different ways the algorithm works, different ways to adjust it, different tuning parameter. Educators have to learn this. It’ll be a real challenge.

Q: You’ve worked with all of these closed loop systems – which is your favorite one so far?

Dr. Buckingham: It’s all a matter of personal choice.

2. Stanford Convatec Coated Lantern Infusion Set 10-Day-Wear Study Update: 18 of 24 Subjects Complete, Average Duration of 8.6 Days; RCT up Next

Dr. Bruce Buckingham shared that 18 of 24 subjects have now completed the Stanford clinical study of 10-day wear of Convatec’s Coated Lantern infusion set: 90% of subjects have worn the set for seven days, with an average duration of 8.6 days. This data is an update from the 10 subjects who had been completed the study in February. Dr. Buckingham noted that the next study of this set will be an RCT, and that his group will also study Capillary Biomedical’s SteadiSet “in the near future.” Capillary is eyeing a 2020 launch of a 7-day wear set.

• The table below makes the case for new approaches to infusion sets to extend wear. Dr. Buckingham’s summary of 353 weeks (~6.8 years) of 7-day infusion set wear shows that only 40% make it the full 7 days, with a mean wear duration of 5 days. The most common failure mode is unexplained hyperglycemia (26% of cases). We didn’t catch whether this slide was summarizing data from studies of a single type of infusion set or all types, though we assume the latter.

3. Excitement for Klue Apple Watch Auto Meal Recognition App from Stanford Docs; Three-Month Stanford Study of Klue Bolus Reminder Feature Recruiting; Roadmap to “Meal-Aware Closed Loop”

Stanford’s Dr. Rayhan Lal and Bruce Buckingham both expressed considerable excitement over the potential of Klue, an Apple Watch app that automatically detects meal start and eating speed without user intervention. Klue uses the motion sensors on the Apple Watch and artificial intelligence to detect when someone starts eating and how fast (via gesture recognition). Dr. Lal noted that he “would really like to eventually see this feature integrated in” to closed loop systems, and even showed a slide with a Kowalski-esque roadmap to from open loop (bolus reminders; currently beta testing), to hybrid closed loop (prompted meal announcements), and finally to “meal-aware closed loop” (fully automated insulin delivery). Dr. Buckingham outlined later in the session how this might look: Klue detects meal-like gestures, commencing the delivery of a small bolus (or ramped basal), and then rising glucose would be the second meal confirmation, triggering delivery of more insulin. At this stage, however, Dr. Lal and colleagues are still studying step one: Stanford is currently recruiting for a three-month study, where participants will use the app with bolus reminders for six weeks, followed by six weeks off. This is certainly an intriguing area of study.

• Dr. Lal spent a few minutes discussing DIY; he demoed the Loop simulation feature, acknowledged the Jaeb-coordinated Loop observational study, and noted that “there are test builds of OpenOmni Loop and AndroidAPS in the wild.” Tidepool CEO Mr. Howard Look was in the house, and during Q&A, shared that his organization will soon have another pre-submission meeting with FDA on June 17 where, after review by the Tidepool Loop Medical advisory committee that includes Drs. Buckingham and Lal, will approach the FDA with the outcomes that they believe should be analyzed based on the data from the observational study. This came in response to an audience member question to Stanford’s Dr. David Maahs about whether FDA would recognize real-world evidence from this study as sufficient demonstration of safety. 

4. Dr. Simon Friedman Light-Induced Insulin Delivery Keynote; Next Step Large Animal Studies, Potential Hurdle Thickness of Skin; Simplicity vs. Smart Insulin; Illuminating Audience Q&A

UMKC’s Dr. Simon Friedman delivered the morning keynote on his team’s efforts to deliver a light-induced insulin delivery system. We covered this work most recently at DTM 2018, and though there were no material updates today, we enjoyed a transparent audience Q&A. The concept of the system is simple: An insoluble polymer that releases human insulin when exposed to 356-nm light is injected below the skin and stimulated non-invasively through an LED light source controlled by CGM. Dr. Friedman did share that larger animal (swine) studies could begin in a “couple” or “few” years, a single injection of the hyper-concentrated polymer could contain a week’s-worth of insulin, and that next steps will be to look at longer-wavelength photolysis (to enable greater penetration into the skin), multiple day animal studies (“tow, three, four days of use”), and glucagon (“the really crazy idea is to have a single material with two different wavelengths, one that stimulates glucagon and one that stimulates insulin”). He also noted that true “smart insulin” is “exciting but demanding” – requiring a single material to measure glucose, make a decision, and meter out insulin – while “we’re just asking our material to do one thing and do it well: Meter insulin. The role of decision-making and blood glucose measuring we defer to existing technologies.” See our prior coverage for more details, and below for more details.

Questions and Answers

Mr. Howard Look (CEO, Tidepool): It’s easy to get excited about these cool technologies, but how feasible is it really? Give us your honest assessment – is this 50/50 odds it works in five years?

A: That is a very difficult question to answer. I can tell you, from my perspective, I’m trying to maximize impact. I want to work on things If I feel like they have a chance of working. You don’t know unless you keep going. The universe so far has given us a lot of green lights on this. That first material, there are so many ways we could’ve gotten the red light, but we’re getting green lights. I’ll tell you, my grant runs out July 31^st of this year, so that’s one part of the puzzle. In terms of actual work, we have enough positive feedback to keep pushing forward on it. There are these divergent paths – right now we’re on this one path, and at one point we’ll do an experiment that changes the probabilities. The next big test is to do large animal model like a swine model. Swine skin is thicker, so that’s a barrier. If that works, it could be a couple years before getting into human being. Maybe we’re a couple years out from doing a swine test, I don’t know.

Q: What part of the light spectrum are you using?

A: We’re in the visible or sub-visible range. In vivo was at 365 nm, which was ok, not great. We’ve got second-gen materials getting into the visible range. Getting things to photo-cleave with infrared would be great because it penetrates so deeply into tissues but the downside is its very hard to get things to cleave with infrared light – low energy photons.

Q: On the rats, the bulge looked fairly large – could the depot be inserted using a regular subcutaneous syringe?

A: With the first results we have, we went into dermal layer, which is shallower than subcutaneous – we wanted to have best chance of seeing something. Also, the dermal layer is nicely vascularized, so there’s a great chance of seeing response. The downside to the dermal layer, there’s not as much volume as there is in the subcutaneous space. There are always tradeoffs. We haven’t given up on subcutaneous delivery, but there are pros and cons. Dermal delivery is more painful than subcutaneous, surely.

Q: What do you think the shelf life could be for this kind of molecule?

A: On our minds, but a couple steps away. I can speculate, but as we make these modifications, they can have unexpected effects on fibrillations that would take place in insulin solutions. I don’t have an answer to that question.

Q: How does this relate to 50 units of Humalog insulin, for a day? Is there a bulge of 50 units of insulin just sitting there?

A: We’re releasing human insulin. 20 microliters of our polymer contains one day’s-worth of insulin. That might cause issues with storage, I don’t know, but the idea is that the material is maybe five-times as dense as U500. It’s much more concentrated, potentially. 140 microliters of this would contain a week’s-worth at that concentration.

Q: How dangerous could this be – ambient light?

A: One way of dealing with that – Senseonics has an adhesive that sticks on, and it just doesn’t come off. That’s a simple, direct solution. In terms of a more scientific solution, we do a lot of work measuring absolute irradiance of sun, how intense it is in specific wavelengths, and we’ve found that our light source can deliver wavelength of much greater intensity than the sun in that narrow range. So even if it did come off, you might get a low level basal release, not a big insulin spike.

Q:  When could this be commercially available?

A: Depends what happens. A big hurdle in the next few years is, when we get into a larger animal model, how effective light will get across the thicker skin. That’s in the next few years I think, and if that looks good, then we can do human trials.

5. Frank Conversation on T1D Cure Research with Drs. Hebrok, Meyer, and Adi Delves into Cure Science, but also Non-Scientific Barriers (Reimbursement, Funding, Attracting Talent)

JDRF International Board Member Ms. Karen Jordan led a fantastic panel entitled “Where’s the Cure?” featuring two of the Bay Area’s top type 1 cure researchers (Stanford’s Dr. Everett Meyer and UCSF’s Dr. Matthias Hebrok) and one highly-regarded clinician (UCSF’s Dr. Saleh Adi). Below, we’ve outlined our favorite quotes from the discussion, which we were delighted to hear touch on both the science of the cure, and also other practical factors related to the cure (reimbursement, funding, etc.).

On Scientific Progress in Cure Research

• “Our cells are 85%-90% identical to the beta cells you have in the body – they’re close. But the cure initially will be like the Model T was 100 years ago, not a Beamer or Porsche. You have to understand that we’ll go through a number of iterations to get to the final cure – the injection of a cell that’ll last many, many years. Possibly, if everything goes perfectly right, it could last the rest of your life. We are born with all of the beta cells we’ll ever have, so theoretically, it’ll be possible one day. Baby steps. First, maybe we’ll start with a device, encapsulated cells that’ll last a year or two at first; the cells won’t be as efficient as those in a functioning body. But over time I think stem cells or transplantation of functioning tissue will come close to what I envision as the cure.” – Dr. Hebrok

• “In order to cure t1d, we need to find stem cells (check), pull them out (check), teach them how to be beta cells (check), make them beta cells (check), put them back in (that’s easy; we haven’t yet found the optimal place, but we’ll do that), and protect them from immune system. Can we do that last one 100%? Probably yes. Are we doing it in most efficient, safest way? Probably not yet.” – Dr. Adi

• “CRISPR has been an absolute godsend – we can take a single base pair out of the 3 billion base pair genome. By modulating this, we can modulate what is presented on the beta cell, what is recognized by T cells – we learn a tremendous amount about how disease starts and accelerates.” – Dr. Hebrok

• “You can manipulate bone marrow so the patient can accept organ transplants – if you give bone marrow along with the organ, you can eliminate the need for immunosuppression. For 70 years, it’s been too toxic. Now we’re doing a less toxic approach, and with no immunosuppressant, in kidney grafts. We have a lot of learning to do about overcoming genetic barriers, and logistics of making it less toxic, but hopefully within the next 3-5 years we could be testing this in patients who are getting islet transplants.” – Dr. Meyer

• “Some approaches are not necessarily the best of the companies to put forward. If your embryonic stem cell is here [extends left hand] and your mature beta cell is here [extends right hand], right now, some approaches stops somewhere in the middle. They’re not beta cells – and now those cells take 120-150 days in the body to turn into beta cells. In early studies, when these cells are put in an encapsulation device, most of the cells die.” – Dr. Hebrok

• “People are trying to put the beta cells into the omentum now, which might be a better place for them. Those put in the portal vein live in liver – in there, they are constantly exposed to a much higher level of glucose than in the rest of the body. They are stimulated constantly to make certain level of insulin, which can cause hypoglycemia and/or stress the beta cells, reducing survival. It’s quite possible that when you put them in the abdomen, they survive a lot better.” – Dr. Adi

• “We are really, really close to a cure. How close? In science, when we say really close, we mean 5-10 years.” – Dr. Adi

On Non-Scientific Barriers to Cure

• “70,000 people in the US could benefit from islet transplantation, and they can’t get it because it’s not reimbursed. Go to Canada, Britain, Australia, and they can get reimbursed.” – Dr. Meyer

• “Talking about medical infrastructure and reimbursement. If you have something that’s safe, exogenous insulin, and few clinical trial endpoints, then there’s no way for insurance to reimburse a transplant. A lot of attention is placed on endpoints in clinical trials – we should be very demanding on those endpoints as a community. For those eligible for implantation, hypoglycemia and self-reported quality of life has been accepted in other countries as evidentiary for financing. Also, 1.4 million people will go through procedures. UCSF can’t handle that many people. Even if we had the cure today, who would get it is a huge question?” – Dr. Meyer

• “We scientists are under-funded. I’ve trained 10s of people. It is getting harder because they are observing that people like me … I’ve sent in seven different grants to NIH in the first eight weeks of year. One was 343 pages. Our success rate is a little higher than 10%, but not where it should be, at 25%-30%. If this grant comes back and we get as many stupid comments as we did last time, then PhDs and postdocs will come back and say, ‘this is ridiculous.’ We worked through Christmas and New Years on this. We are not being supported at the level we need, and the science is going ridiculously well. People are working on these cells every day of the week. The devotion and commitment to what we’re doing is fantastic, and I’m hoping they have the career they want – some of them will go into other fields.

When I give talks at schools, I always ask the PhD students how many of them will go into academia vs. industry. It used to be 50:50. Now, it’s down to 9:1. The one person who wants to go into academia is the MD/PhD in the room, because they have security – if something goes wrong, they can go take care of patients. We are depleting, not completely, our next generation. Industry is not supporting basic science in their organizations anymore. If we don’t train the next generation of people, this will have an effect.” – Dr. Hebrok

• “Endocrinology is not the top desired field for medical students. If we had clinical trials, procedures reimbursed by insurance, and someone could look at that and say they could make a life and do pretty well, that’d help. But that’s what we’re up against. Part of it is stability. The field is not as stable as it should be right now for many people, and it limits their options. It’s important to say what can we get in the field reimbursed to start moving toward the cure.” – Dr. Meyer

6. Bigfoot’s Lane Desborough: 28 Assertions on Automation, Feedback Control, #WeAreNotWaiting

Similar to his November DiabetesMine talk, Bigfoot Chief Engineer Mr. Lane Desborough shared 28 assertions about automation, patient-led innovation, interoperability, and challenges in closed-loop engineering. You can see the whole 30-minute talk here. Slides will be posted shortly.

1. Bad things can happen during mode transitions, when the state of the system is changing. For example, computer to human control, plane takeoff to cruise or cruise to landing. This has clear implications for shifts between AID auto and manual modes, or perhaps even for shifts between various auto modes, such as day and night.

2. Software is a harmless mental abstraction until it is instantiated in the physical world. Mr. Desborough told of how his friends at an ethylene plant were doing a routine software upgrade on a computer, but when they went to apply a patch, they shut every valve in the plant, wreaking havoc. “We have to be very careful about how we instantiate software in the physical world.”

3. Disturbances are rejected using hierarchical, temporally decoupled control algorithms. Certain controllers are designed to reject disturbances that are happening quickly, like cooling water temperatures, to slower disturbances, like a new tanker arriving every couple of weeks.”

4. Remote monitoring and diagnostics create new value; new feedback loops engage additional stakeholders. Mr. Desborough used to travel the world as an engineer, looking at data, but then someone had the idea to bring the data to the engineer. Thus, remote monitoring was born – at Honeywell, in 1998, the remote monitoring software was dubbed “Loop Scout” (sound familiar?).

5. Feedback is amazing, allowing excellent control to be achieved, although the sensor needs to be accurate. Sensor noise is fed into the system.

6. Automation transfers variability from a place where it hurts (sensor) to a place where it doesn’t hurts as much (actuator) so that we don’t have to do as much work. E.g., A thermostat shifts variability from room temperature to use of fuel/gas/electricity.

7. Feedback automates tasks; sensing, deciding, and acting can be automated to various degrees. Not all of it needs to be fully automated or fully manual – aspects can be automated. E.g., a bolus calculator automates the decision part, while sensing and actuating are still partly manual.

8. Automation adds new tasks such as supervision, troubleshooting, and system maintenance. “Don’t forget the human!”

9. Humans and computers are good at different things; improper task allocation creates many problems. Humans good at recognition, “blink,” troubleshooting, new situations; computers are good at cognition tasks, vigilance/repetitive tasks, automated procedures. It’s important to properly allocate these tasks.

10. Automation not a panacea. It introduces new challenges for supervising, troubleshooting, and maintaining the system. On the one hand, there are risks of deskilling, miscalibrated trust, complacency, and addiction (think Homer Simpson asleep at the nuclear power plant controls); on the other hand, there are risks of task saturation, brittleness, mode confusion, brittleness, and loss of situational awareness (think Homer freaking out at the nuclear power plant controls). Mr. Desborough noted that the grounding of Boeing 737 Maxes is a result of automation confusion – a new mode was introduced and there was insufficient training, leading to tragedy.

11. There are well-proven, established engineering methods for developing safety critical complex systems. E.g., modeling and simulation.

12. Computer modeling and simulation enable the design, automation, and operation of complex, hazardous, sociotechnical systems. Bigfoot is a big believer in simulation. Modeling is advantageous because it doesn’t put anyone at risk, is faster than real-time, and lower cost.

13. Interoperability: with great power comes great responsibility; benefits may be offset by challenges and new considerations. “Automated insulin delivery is not the first railroad for interoperability.” Interoperability identifies skill gaps, cybersecurity issues, a whole host of new considerations.

14. Interoperability shifts responsibilities and creates complex organizational dynamics with potentially conflicting goals. Issues have arisen around governance – different suppliers contributing to a single system may be incentivized to work together on one project while competing in others, so they may try to keep information proprietary or engage in other destructive behaviors.   

15. If automation works in other safety-critical systems, why shouldn’t it work for insulin delivery? This is what crossed Mr. Desborough’s mind when his son was diagnosed with type 1 diabetes in 2009.

16. Software in industrial automation systems is held to a very different standard than software in academic research. Mr. Desborough was invited to be an independent reviewer on an academic algorithm submission to FDA, and he was surprised by the quality and complexity of the work.

17. Insulin therapies have progressively increased the amount of variability transferred, with an ultimate goal of reducing burden by mimicking natural insulin delivery. With a single injection of insulin per day, you have one chance a day to impact glucose variation; with MDI, you have multiple chances per day; with automated insulin delivery, you have even more.

18. Subcutaneous delivery of “rapid acting” insulin limits performance of automated insulin delivery systems. The 25-40-minute deadtime (delay) for rapid-acting insulin is the limiting factor right now in how much variability can be transferred to algorithm. “Imagine if your car had the same thing from the time when you press on the gas pedal to the time you see the change on the speedometer, it’d be very difficult to design a feedback control system for that situation.”

19. Dozens of factors contribute to variation in blood glucose – most of them unmeasured – making automation challenging. Mr. Desborough showed Adam’s ~42 factors that contribute to glucose variation.

20. An automated insulin delivery system needs to work across a wide range of users and use conditions. Insulin needs change a lot between and within people.

21. Experiments with humans are difficult, dangerous, and expensive, often producing poor results. See this plot (assertion #5) of messy glucose traces, documenting Mr. Desborough’s son’s highly-varied blood glucose responses (n=304!) to the same juice box (16-grams of carbs).

22. Beyond the algorithm, human factors, use conditions, and economics must be considered in the development of automated insulin delivery systems. Mr. Desborough’s son participated in a Yale closed loop trial with Hylenex, an adjuvant drug that can increase the absorption of other injected drugs. In short, conditions were not right for the molecule, and it kept denaturing rapidly, not to mention its extreme cost. Mr. Desborough mentioned that these additional factors are critical when thinking about commercialization.

23. Something which appears simple to the user belies the complexity of the effort that went into creating it. When Mr. Desborough arrived at Medtronic to lead the team working on the 640G PLGS algorithm, what he inherited was overly-complex and demanded too much of the user. His team spent six person-years building a simulator to identify optimal parameters, eventually excised 6 of 7 tuning knobs, leaving just one. One of our favorite quotes of the day followed: “If it’s a complex interface, that should be a flag that that effort probably hasn’t taken place.”

24. NightScout was born from the efforts of techies who paid it forward so others could share the lifechanging benefits they were enjoying.

25. #WeAreNotWaiting catalyzed the engagement of many passionate innovators who remain active in development to this day. He showed a picture including himself, Beta Bionics’ Mr. John Costik, Bigfoot’s Mr. Bryan Mazlish, and Tidepool’s Mr. Howard Look.

26. #WeAreNotWaiting is a powerful expression and demonstration of the power of consumer-led innovation. “Bigfoot is not waiting either … The best way to reduce burden for people with diabetes, doctors, and payers, is to get together and deliver services and technologies largely surrounding the notion of automation of insulin delivery.” Bigfoot now has 130 employees!

27. Computer modeling and simulation can complement traditional methods of characterizing the expected behavior of automated insulin delivery systems.

28. Development of safe and effective automated insulin delivery is enabled by comprehensive simulation of behaviors, physiology, algorithms, and hardware. The system also includes human behavior, how one responds to an alarm, when they eat, how much they eat, how they count carbs, when and how much they exercise. “I really believe that a comprehensive simulation capability is existential to automating insulin delivery.”

-- by Brian Levine, Adam Brown, and Kelly Close