12th Annual Diabetes Technology Meeting

November 8-10, 2012: Bethesda, MD Day #2 Highlights – Draft

Executive Highlights

Greetings from North Bethesda and Day #2 of the 12th Annual Diabetes Technology Meeting, where more than 400 attendees have gathered from 24 countries. The morning kicked off with a keynote speech by FDA’s CDRH Director Dr. Jeffrey Shuren. It was notable to see such a prominent FDA figure at the conference, and indeed, Agency participation seems to increasingly be a theme at every Diabetes Technology Meeting we attend. Dr. Shuren’s speech touched on a broad array of topics, including the challenges FDA is facing (funding is a big one – a measly $15 million budget for regulatory science for medical devices!), the creation of a public-private partnership to help industry collaborate, and the Agency’s efforts in CGMs, pumps, and the artificial pancreas. His comments were often unfortunately general in scope, though we did appreciate some very encouraging remarks about how FDA can get devices to market faster and better work with industry.

In an afternoon dedicated to the artificial pancreas, we heard from some of the biggest names in the field. Dr. Edward Damiano (Boston University, Boston, MA) shared the latest update on his five-day, outpatient bi-hormonal closed-loop study: an IDE for the controller device (an iPhone 4S that wirelessly connects to two Tandem t:slim pumps and a Dexcom G4 sensor) was submitted to FDA last week and the hope is to start the study next month! We were also keen on Dr. Boris Kovatchev’s (University of Virginia Health System, Charlottesville, VA) presentation of the first results of a closed-loop trial (which started just one week ago!) using a new control-to-range algorithm and employing the Diabetes Assistant smart phone platform, Dexcom G4 Platinum CGM, and Tandem t:slim insulin pump. Impressively, the first patient had no hypoglycemic readings, and spent 83% of the 40 hour closed loop control in range (70-180 mg/dl) – we hope this initial patient will prove representative of the group.

New data also came from Dr. Bruce Buckingham (Stanford University, Stanford, CA), who presented results from a pilot study of a predictive low glucose suspend algorithm. The data are strong and the team is now enrolling for a larger, multi-center trial. Dr. Frank Doyle III (University of California, Santa Barbara, Santa Barbara, CA) mapped out the next year-and-a-half of the NIH-funded consortium that also includes Sansum, UVa, and the Mayo Clinic: it starts with a new controller that adapts its target zone based on time of day (studied in-clinic for the first time last week) and leads to an eight-week outpatient trial in mid-2014. Dr. Claudio Cobelli (University of Padova, Padova, Italy) updated us on a variety of recent and ongoing improvements to closed-loop simulations (an update to the UVa/Padova metabolic simulator was filed with FDA in October) and smarter sensors (i.e., a next- generation Dexcom sensor with enhanced onboard processing algorithms). He also reported the very encouraging results of the first outpatient study of the AP@home European project which is conducted in Padova, Montpellier and Amsterdam.

For the first time since the meeting’s inception, DTM Chair Dr. David Klonoff presented the Diabetes Technology Society Leadership award to an organization as opposed to an individual. The Center for Devices and Radiologic Health (CDRH) at FDA was this year’s recipient. Of note, upon accepting the award on behalf of CDRH, Dr. Jeffrey Shuren (Director, CDRH), announced that just 10 minutes prior the center had posted the final guidance on the artificial pancreas, having combined the draft guidances in development for the low glucose suspend and the closed-loop system. Also deserving of recognition, the esteemed Dr. Lutz Heinemann (Science & Co, Dusseldorf, Germany) and Dr. J. Hans DeVries (Academic Medical Center, Amsterdam, The Netherlands) received the Artificial Pancreas Award for their dedicated work on the AP@Home project.

 

Detailed Discussion and Commentary

Keynote Address

THE IMPACT OF REGULATORY SCIENCE ON DIABETES TECHNOLOGY: AN FDA PERSPECTIVE

Jeffrey Shuren, MD, JD (Director, CDRH, FDA, Silver Spring, Maryland)

We were glad to see the FDA’s CDRH Director Dr. Jeffrey Shuren front and center at DTM in the Friday morning keynote speech. Dr. Shuren emphasized CDRH’s vision: “to give (diabetic) patients access to high quality, safe, and effective devices of public health importance first in the world” – admittedly, we are pretty far from this today. He explained that the division is taking this vision “very seriously” through a number of key steps: collaborating with companies (CGMs, pumps, the artificial pancreas, data management), developing a public-private partnership to focus on regulatory science for medical devices (this sounds encouraging though few details were shared), developing better tools and software, and use of the new innovation pathway and entrepreneur in residence program. Overall, we found Dr. Shuren’s words encouraging to hear, though they were fairly general in scope and light on details (e.g., “we are working with companies”). However, he also had some very promising and frank comments about FDA’s challenges and how it can better encourage innovation and help companies get devices to market faster – this perspective was really great to hear. In terms of specific diabetes technology, he only mentioned the Medtronic Veo and was very non-committal on its status: “we’ll see where that goes and whether we’ll have that technology for US patients in the near future.”

  • Dr. Shuren emphasized the major challenges facing regulators and regulatory science: communication, infrastructure, and funding. He explained that regulatory science (the tools, standards, and approaches needed to evaluate safe and effective medical devices) is not well understood or appreciated in the medical device ecosystem. Additionally, most of the evaluation work is scattered throughout country – it is inefficient (one expert here, one project there) and there is very little investment by federal government. For comparison, NIH’s FY12 budget for research was $30.7 billion, including $575 million for the NIH’s new Center for Advancing Translational Sciences (NCATS). By contrast, FDA has $15 million for medical device regulatory science (excluding staff). “The disparity is huge,” he noted.
  • Dr. Shuren highlighted classes of diabetes technology as examples of how CDRH is working to get better devices approved sooner. His discussion was fairly general for the most part, more about broad approaches than specific companies or devices.
    • 1) Data management: Dr. Shuren explained that we have meters, CGMs, and pumps collecting data, but clinicians have limited time to look at it and must deal with lots of cables and downloading hassle. For patients, this makes it challenging to manage diabetes. FDA is currently working with companies to bring data management technologies into a single stream of information. Additionally, the Agency is working to develop analytical tools that make interpretation easier. Dr. Shuren’s slide highlighted that “medical cell phone apps [are] coming soon.” We hope this means the finalized mobile medical applications guidance is on the horizon.
    • 2) CGM: Dr. Shuren emphasized that “CGMs are not as accurate as we need them to be” for the closed loop. FDA is working with companies to develop better, more reliable, and more accurate sensors. He stated that companies “are taking on that challenge” and it’s “encouraging to see some of the advances that are hitting the market” – this may have been an indirect reference to Dexcom’s recently approved G4 Platinum.
      • How do we better assess CGM? FDA is working on better in vitro screening of substances that can interfere with CGM readings. Work is also ongoing to better understand the effects of biofouling and how to manipulate a sensor’s surface. The idea is to understand if there are important physiological differences that lead to differences in long-term sensor function. FDA would then provide this feedback to companies to make better technology.
      • Emerging technologies in hospital glucose sensors. FDA is working to understand how changes in physiological pH affect sensor accuracy in the hospital setting. The Agency is also focusing on optical glucose biosensing and minimally invasive sensing.
    • 3) Insulin pumps. Dr. Shuren reminded the audience of the FDA’s 2010 initiative on infusion pumps. Previously, there were thousands of adverse events being reported for infusion pumps, many related to insulin pumps. Interestingly, Dr. Shuren believes the new initiative has resulted in higher quality regulatory submissions. In the two-year period prior to initiative, the Agency cleared 51% of pumps. Now, the FDA is clearing about 70% of them. The FDA has also developed better tools for manufacturers to use and development of a generic insulin infusion pump safety model is ongoing.
    • 4) Artificial pancreas. Dr. Shuren noted, “We’re not there yet because we need better components, but we’re well on our way to getting there.” He emphasized that the Agency is committed to this technology and would “love to see it come to the US first.” The FDA has consolidated the review team in CDRH to help improve oversight over the AP. In the past year, Dr. Shuren highlighted that the FDA has approved four or five clinical trials every single month devoted just to the AP. The Agency has also approved the first outpatient closed loop AP study in the US.
      • On the FDA status of the Medtronic Veo, Dr. Shuren was disappointingly non-committal and fairly vague: “We have an in-house application and we’ll see where that goes and whether we’ll have that technology for US patients in the near future.” He mentioned that it has LGS technology and “is already CE Marked in Europe” – we would of course add that it’s been a three- plus year delay…
      • Dr. Shuren explained that the AP draft guidance documents (subsequently finalized and posted a few hours after his talk) were somewhat unique – usually, the regulatory pathway must catch up with the science. For the AP, it was the other way around: the regulatory pathway needed to get ahead of the science (we would note that this was really not the case with low glucose suspend).
    • 5) Bioartificial pancreas. FDA is working to better understand combination products and is focusing on different ways to have successful encapsulation of pancreatic islet cells.
  • To overcome the challenges of funding and inefficiency, FDA is setting up a public- private partnership with LifeScience Alley. The Agency is working to set up a 501(c)(3) organization that will be separate and only focused on advancing the regulatory science for medical devices. One of the areas will be diabetes. The hope is to get this off the ground in the “near future.” The partnership will allow sharing of resources, dollars, expertise, data, and allow for companies to come together and not run into legal challenges. We hope this could allow for independent testing of devices, especially blood glucose meters, which would jointly be supported by money from all companies.
  • Dr. Shuren closed with a review of the FDA’s innovation pathway, a new route to market for breakthrough technologies. It serves as an “incubator cell” for new approaches and tools to reduce the time and cost of development, assessment, and review of breakthrough (and other) devices. It also transforms how the FDA and innovators work together. Part of the program includes the entrepreneurs in residence program, which invites experts from the medical device industry (VCs, patients, experts, and companies) to the FDA for some of the Agency’s day- to-day work. The pathway includes an application process, a collaboration phase, a clinical trials phase, and market approval. A new version of the pathway was recently launched and Dr. Shuren specifically mentioned that there are three products for end-stage renal disease. He did not specifically address diabetes, though we certainly hope industry experts are taking part in the process.

Questions and Answers

Dr. Yogish Kudva (Mayo Clinic, Rochester, MN): It’s very important to all of us that the US be ahead of the curve and have the first device approvals. What are the lessons we can all learn from how the Medtronic Veo process has gone? Why are we lagging behind?

A: From an FDA perspective, in the past few years, we’ve had challenges on a variety of fronts. Our programs have not been sufficiently predictable, transparent, or timely. We’ve been very public about those challenges. We recognized those problems when I came and we put out two reports on it. We were very frank about the challenges. In early 2011, we developed a plan of action and steps to start fixing the program. Not only has there been lots of progress, but we are now for the first time seeing changes in our performance that we have not seen in some cases for a decade. We will put this data out in the coming weeks on what it’s been like before and where we’re headed now. We’re focused on making our programs better.

We need to tackle this issue on science. Most other countries are not safety and effectiveness. The bar is different – other countries don’t need to be effective. We believe that’s important for patients. If that bar becomes irrationally too high, it becomes a barrier. You need to find a sweet spot to have safety and effectiveness, but do it in a way that is rational, timely, reasonable, and not costly. Regulatory science is the linchpin for getting there. Through things like computer models, we can test drive technology without animal studies – that is a game changer. Those are the advances we’re talking about. Better pathways, putting out guidance, and advancing the very science itself.

Dr. Robert Vigersky (Walter Reed National Military Medical Center, Washington, DC): One of the barriers in getting devices into hands of patients is the IRBs at the various institutions. What is the FDA doing to work with IRBs on an individual basis or to put out guidance to get protocols to IRBs?

A: We don’t have authority over IRBs. This is one of the issues on the table for the entrepreneur in residence program. How can we streamline clinical trials? The scope isn’t about what’s solely within the jurisdiction of FDA. It’s anything that we can influence in any way, shape, or form. IRBs are something we’ll look at. Do you move to a central IRB model? It could be a time saver. How much time does it take contracting if you’re doing a 70-site study? Why not have set templates with contracts? That can lead to big efficiencies and not big costs to do.

Peter Rule (OptiScan, Hayward, CA). If we could envision the ideal collaboration between industry and FDA, with certain endpoints and certain outcomes, and jointly meet them, there would be a high probability of approval. But the current risk benefit standard makes that difficult. It’s hard to power trials as a manufacturer. Do you see the day where there is more refinement around the notion that risk-benefit is a subjective state?

A: At the end of the day, there will always be a little bit of subjectivity. Science is often gray. I think we make a lot of smart decisions, but not as consistently as we could. What is the right kind of assessment – can you get out ahead and work with industry and academia. That’s what we try to do with the AP. Just this past April, we released a final framework on benefit-risk for those seeking PMA or de novo decisions. It’s very patient centric. When technologies come on the market, they’re not coming to be used on you. They are used on patients. We need to be focused on patients’ perception of risk-benefit. With a new technology, you cannot expect it to be a home run on the first iteration. You must take that into account. If you don’t, you will never let it on the market and it will never get better. My staff is applying that to every single PMA and de novo submission. For some technologies, there is a risk that some people would not take. But some would. In that circumstance, if you’re explicit about that risk, let’s let patients and practitioners make that call.

Q: For technologies like glucose sensing, what about an implantable vs. a non-invasive sensor – is there a useful domain for both?

A: You go where the technology takes you. If the technology is good enough that you didn’t need implantable, you wouldn’t use it. If the answer is no, you would still have implantable. It would be a wonderful world if we didn’t have to use as many invasive technologies on patients – we’re a long way from that, but it’s a terrific goal that we should be shooting for.

Dr. David Klonoff (Mills-Peninsula Health Services, San Mateo, CA): Thank you for a really good presentation. It’s nice for us to hear that you really get it and understand the issues that are going on. On behalf of everyone, we want to thank you for the hard work.

Artificial Pancreas: Engineering Aspects

ROBUSTLY ADAPTIVE CLOSED-LOOP BLOOD GLUCOSE CONTROL IN CHILDREN AND ADULTS WITH TYPE 1 DIABETES USING A BIHORMONAL BIONIC PANCREAS

Edward Damiano, PhD (Boston University, Boston, MA)

Dr. Damiano provided the latest update on his team’s quickly moving bi-hormonal work, a very valuable update from what we last heard at Children with Diabetes in July. Most notably, Dr. Damiano discussed the status of the five-day outpatient study that we’ve been looking forward to for over a year. The IDE for the controller device (an iPhone 4S communicating with a Dexcom G4 CGM and two Tandem t:slim pumps) was submitted to the FDA last week and the hope is the study can start in December. It will take place in Boston’s Beacon Hill neighborhood, a three square mile area downtown where subjects will able to roam freely while wearing the device with a nurse chaperone during the day. Dr. Damiano also presented interim results from the third inpatient study that is finishing up 51-hour experiments in 12 adolescents and 12 adults. Half of the study participants are getting adaptive pre- meal priming boluses while the other half are using a fully reactive closed-loop system (i.e., no meal boluses from the user). Notably, the closed-loop algorithm adapts over time, starting quite conservatively and then fine-tuning insulin dosing. The interim results look very solid in nine adolescents and 11 adults by day two of the study – mean blood sugars [projected A1c] of 143 mg/dl [6.6%] in adults and 171 mg/dl [7.6%] in adolescents receiving no meal boluses, improving to 138 mg/dl [6.4%] and 157 mg/dl [7.1%] when adaptive pre-meal boluses were used. We hope Dr. Damiano’s team can continue the momentum pending positive feedback from the FDA and IRB for the outpatient study.

  • Dr. Damiano’s much awaited five-day, transitional, outpatient closed-loop study will hopefully begin in December. The IDE for the controller device (an off-the-shelf iPhone 4S that communicates with a Dexcom G4 CGM and two Tandem pumps) was submitted to the FDA last week. He is awaiting feedback from the Agency and pending IRB approval (fingers crossed!), the study will start next month. It will take place in Beacon Hill, a three square mile neighborhood in downtown Boston. Patients (20 adults) will have the ability to roam freely (with a chaperone) during the day with unrestricted eating and exercise and point of care blood glucose testing. At night, they will sleep with a GlucoScout for reference blood glucose checks.
    • The iPhone 4S will run the control algorithm and communicate with two low-energy Bluetooth Tandem t:slim pumps (insulin and glucagon) and a Dexcom G4 CGM. The G4 will wirelessly stream data into the iPhone through a new custom hardware attachment connected through the 30-pin connector. This was an update over the system we saw at ADA in June and Children with Diabetes in July, which was hardwired to Abbott’s FreeStyle Navigator CGM receiver. Dr. Damiano was wearing the system during the presentation and showed the audience his real-time streamed blood glucose value from the G4 along with the Tandem pumps dosing saline.
    • The Beacon Hill study will test both fully reactive (no meal boluses) closed- loop control and closed-loop control with adaptive pre-meal priming boluses. For the latter, patients will select whether a meal is small, medium, or large, and pre-meal doses will be adapted over time by the algorithm. More broadly, the control algorithm itself will also adapt over time and fine tune dosing based on its performance and changing insulin requirements.
  • Dr. Damiano reviewed the design and interim results from his group’s ongoing third clinical feasibility study in 12 adults and 12 adolescents. The trial involves 51-hour experiments using the Abbott Navigator CGM as the input to laptop-driven insulin and glucagon control. The laptop directs dosing on two Insulet OmniPods. Participants ate six high carbs meals (the level of control achieved given the carb content is quite impressive) and had 30-40 minutes of structured exercise (4,000 heart beats). The algorithm initializes with only the subject’s weight and adapts over time – notable robustness considering both adults and adolescents are taking part in the study. Half of the adolescents and half of the adults receive adaptive priming boluses at meal presentation (i.e., the algorithm automatically changes the size of the pre-meal priming bolus over the course of the study), while the other half are on fully reactive control with no priming bolus.
    • Similar to previous trials, Dr. Damiano’s group tested CGMs head to head: Dexcom’s G4 Platinum and Abbott’s FreeStyle Navigator (first gen) – accuracy was very comparable. Dexcom’s G4 had a MARD of 12.3%, very comparable to the Abbott FreeStyle Navigator’s MARD of 12.6%. The CGMs were compared to blood sampling every 15 minutes. Data was used from eight to 48 hours of closed loop experiments. The CGMs were inserted 24 hours before the first calibration. The system to be used in the new outpatient study (see above) will use the Dexcom G4.
    • Dr. Damiano displayed interim study results, demonstrating good average control and a low prevalence of hypoglycemia. He urged the audience to pay more attention to the slightly better day two numbers since the algorithm takes six to 12 hours to adapt to the patient and establish optimal control. In his view, these numbers are more predictive of how the system would perform for several months. Dr. Damiano also emphasized that the A1c’s achieved in adults and adolescents in both experimental conditions were much better than standard of care. Additionally, hypoglycemia was infrequent, though it remains to be seen if there will be an increase once the outpatient study gets going and patients are not so sedentary.
  CGM Average 
(mg/dl)
BG Average (mg/dl) 
[Projected A1c]
% BG Values
< 70 mg/dl
Carbs 
(g/kg/day)
  Day 1 Day 2 Day 1 Day 2 Day 1 Day 2  
Adults No Meal Bolus (n=5)

141

136

145

[6.7%]

143

[6.6%]

2.0

5.7

3.6

Adolescents No Meal Bolus (n=6)

165 156

175

[7.7%]

171

[7.6%]

1.6 0.5 4.5

Adults Auto Meal Bolus (n=5)

125

126

130

[6.1%]

138

[6.4%]

5.5

0.0

4.2

Adolescents Auto Meal Bolus (n=3)

162 147

166
[7.4%]

157
[7.1%]

0.0 0.7 5.2
  • Dr. Damiano briefly touched on his team’s second clinical feasibility study (just published in Diabetes Care), highlighting that children are much different from adults. Initially, the closed-loop algorithm performed well in six adults: an overall mean blood glucose of 158 mg/dl (68% in the range of 70-180 mg/dl, 0.7% <70 mg/dl) and a mean of 123 mg/dl overnight (93% in the range of 70-180 mg/dl, 0.5% <70 mg/dl). However, when the same system was brought into children, it could not get them in range – average BGs were 180-190 mg/dl with the same controller. The team iterated the algorithm and eventually generated a more adaptive system, which has since been used in the third feasibility study (see above) and will be part of the Beacon Hill study (also described above). It is initiated with only the subject’s weight and comes online with conservative dosing that adapts over time.

MOVING THE CLOSED-LOOP ARTIFICIAL PANCREAS FROM THE CLINIC TO THE HOME: ALGORITHMIC DEVELOPMENTS

Frank Doyle III, PhD (University of California, Santa Barbara, Santa Barbara, CA)

Laying out the near-term goals of the NIH DP3-grant funded Ambulatory Control project, Dr. Frank Doyle III described the control algorithm developed for the project’s first outpatient studies. The algorithm, called periodic zone model predictive control (PZMPC), is similar to UCSB’s initial zone MPC algorithm in that insulin delivery reverts back to a fixed basal rate unless CGM values start to trend outside of a specified glycemic zone. However, with PZMPC the boundaries on this zone gradually shift at night, from the daytime zone of 80-140 mg/dl to an overnight zone of 110-220 mg/dl. To further mitigate the risk of nocturnal hypoglycemia, overnight insulin delivery is constrained to be no more than 150% of the basal rate. In silico modeling suggests that such an algorithm should enable excellent overnight safety, and on November 2 PZMPC was successfully tested for the first time in a human patient. Dr. Doyle said that next steps include additional feasibility evaluations at UCSB/Sansum in the coming weeks and months, larger in-clinic studies in the spring and summer of 2013, supplemental in silico testing in late 2013, and eight-week outpatient experiments starting in mid-2014.

  • Starting with the basic on-off controller described by Kadish in 1963, Dr. Doyle reviewed the history of algorithm development in closed-loop glucose control research. Later the precursors to today’s proportional-integrative-derivative controllers were developed by Albisser et al. (1974), Clemens (1979), and Fischer et al. (1980). In 1996 Dr. Doyle worked on the first application of model predictive control (MPC) for glucose control, and in 2001 Dr. Roman Hovorka’s group introduced non-linear MPC to artificial pancreas research. Promising work has been done with other algorithmic approaches such as pole-placement, H- infinity, adaptive, and fuzzy logic, Dr. Doyle noted. To supplement this overview he presented a slide with all but the most recent published clinical trials of artificial pancreas studies: these included four studies of PID algorithms, one with a PD/PI algorithm, one with a PD-based controller that also dosed glucagon, one hybrid MPC-/PD-driven insulin/glucagon system, 11 using MPC-based algorithms, and one with zone MPC.
  • One of the latest developments in MPC-based glycemic management has been UCSB’s zone MPC algorithm (Grosman et al., J Diabetes Sci Tech 2012), whereby insulin delivery changes from a pre-set basal rate only if the patient’s CGM values leave the target zone or are predicted to leave the target zone (80-140 mg/dl, in initial applications). Dr. Doyle noted that zone MPC has been demonstrated feasible in both a 12-patient UCSB/Sansum study of fully closed-loop control (Zisser et al., ADA 2012) and in larger industry trials of an artificial pancreas precursor product (Mackowiak et al., ADA 2012).
  • Dr. Doyle described the rationale for and design of a periodic zone model predictive control (ZMPC) closed-loop algorithm. Wide agreement exists that closed-loop control should be relaxed overnight in order to minimize risk of hypoglycemia, he explained; the question is just what approach to use. One approach would be to turn off the controller altogether or “de- tune” it so dramatically the controller never takes action (e.g., by widening the target zone to go all the way up to 1,000 mg/dl). As an alternative, UCSB researchers have designed the PZMPC algorithm that smoothly adjusts the boundaries of the zone from their daytime values (80-140 mg/dl) to establish a wider zone overnight (110-220 mg/dl). The PZMPC algorithm also puts a hard constraint on how much insulin can be delivered, even in hyperglycemia: no more than 50% of basal rate. The UVa/Padova simulator was used to compare PZMPC, traditional zone MPC, and a regime that switches to a fixed basal rate at night; PZMPC led to more overnight hyperglycemia but less overnight hypoglycemia (Gondhalekar, Dassau, Doyle III Eur Control Conf 2013).
  • The initial clinical evaluation of PZMPC closed-loop control will enroll 5-12 of the 12 patients who participated in the first study of UCSB’s original zone MPC algorithm (Zisser et al., ADA 2012). Except for the difference in control algorithms, the experimental design is identical to that of the zone MPC study (day-and-night study with unannounced meals, unannounced exercise, skipped lunch). The first patient tested PZMPC on November 2, 2012 and experienced favorable glycemic control (including less overnight insulin delivery – possibly safer if a patient were not frequently testing, Dr. Doyle noted).
  • Dr. Doyle closed with a look to the future of the NIH DP3-grant-funded Ambulatory Control project, a collaboration of artificial pancreas researchers at UCSB, the Sansum Diabetes Research Institute, the University of Virginia, and the Mayo Clinic. He reminded the audience that the five-year, $4.5-million initiative is designed to develop closed-loop systems that respond to glucose on a scale of minutes, adapt to day-to-day and week-to-week glycemic changes, and “monitor and supervise” glucose control over months and months.
    • Communication between the hardware and algorithms will occur through the UCSB/Sansum Artificial Pancreas System (APS) platform, which Dr. Doyle said is being ported to a new iDevice framework so that it can run on mobile phones (as opposed to laptops or tablets as previously used). In addition to the PZMPC controller, the system will incorporate the UCSB/Sansum Health Monitoring System (HMS) safety algorithm, which can send text messages and graphical alerts to physicians for remote monitoring. Dr. Doyle also noted that the FDA has approved the inclusion of fingersticks with Bayer’s Contour Next BG meter in an upcoming UCSB/Sansum closed-loop study – hopefully the first step toward outpatient studies that fingerstick tests as the sole reference values.
    • Dr. Doyle looked forward to upcoming clinical studies in the DP3 project, which will begin inpatient closed-loop studies at the University of Virginia, Sansum Diabetes Research Institute, and Mayo Clinic during the spring and summer of 2013. Each study will include two closed-loop sessions: one for behavioral initialization of the individual patients, and another with “behavioral adaptation.” Additional in silico are slated for late 2013 as a prelude to the main event: eight-week, case-controlled outpatient comparisons of closed-loop and open-loop control.

 

OUTPATIENT ARTIFICIAL PANCREAS: TECHNOLOGY AND CLINICAL TRIALS

Boris Kovatchev, PhD (University of Virginia Health System, Charlottesville, VA)

In an engaging, data-driven presentation, Dr. Boris Kovatchev presented the first-patient results from a closed-loop efficacy trial using a new modular algorithm and the latest device technologies. The algorithm used an enhanced control-to-range system comprised of: 1) an insulin on board (IOB) tracking module; 2) a range control module; and 3) a safety supervision module. The algorithm ran on a DiAs smart phone (a portable AP system developed at the University of Virginia by Dr. Patrick Keith- Hynes that has a modified medical-grade Android OS platform designed for AP application), which connected to a Dexcom G4 Platinum receiver via USB and wirelessly to the Tandem t:slim pump via low energy Bluetooth. (The DiAs also could be monitored remotely over 3G or Wi-Fi connection.) The study, which just commenced a week ago, is being conducted at four sites, with five patients at each site. In a crossover design, patients were randomized to either open- or closed-loop control (run on the DiAs) for a 40-hour session. Dr. Kovatchev presented the very first patient experience. Over 40 hours with closed- loop control, the participant was in range (70-180 mg/dl) 83.4% of the time, never below 60 mg/dl, in range overnight (80-140 mg/dl from 11:00 pm to 7:00 am) a notable 100% of the time, and >180 mg/dl 14.4% of the time. While Dr. Kovatchev said he knew he shouldn’t present data from just one person, we were sure glad he did! After seeing the promising first tracing, we can’t help but look forward to when full results emerge.

  • Dr. Kovatchev began by detailing the specifics of the Diabetes Assistant (DiAs) portable AP platform. Notably, the system has a medical grade Android operating system designed for AP applications. (The operating system and graphical user interface are deposited in the FDA master file MAF 2109, “AP Mobile Medical Platform.”) DiAs was developed at the University of Virginia by Dr. Patrick Keith-Hynes. The system can wirelessly communicate with an insulin pump and CGM and can operate multiple control algorithms. It’s color touch screen features a home screen with hypoglycemia and hyperglycemia “traffic lights” to inform the patient whether intervention is needed, and various system statuses (e.g., battery time, whether there is connection to the pump or sensor). DiAs can be used for closed-loop or open-loop control, and can enable remote monitoring (even simultaneous real-time remote monitoring of several patients, as Dr. Howard Zisser [Sansum Diabetes Research Institute, Santa Barbara, CA] demonstrated – he controlled several patients from a single iPad.) For a deeper delve into the user interface, please see page three of our DTM 2011 Day #2-3 report at https://closeconcerns.box.com/s/vitjvebk02cb0mx4gl1k.
  • Early closed-loop feasibility studies with DiAs demonstrated the ability to maintain inter-device communication. The system consisted of DiAs, a communication box (Google Galaxy Nexus phone), and iDex (an Insulet OmniPod PDM integrated with the Dexcom Seven Plus CGM). Across four centers (UVA, Padova, Montpellier, and Sansum Diabetes Research Institute), patients received both open- and closed-loop control using DiAs. Inter-device communication was maintained 98.9% of the time in open-loop control (out of 277 patient hours) and 97.1% of the time in closed-loop control (out of 550 patient hours).
    • After 13 hours of open-loop control, participants had closed-loop control for 29 hours. The closed-loop control was two-fold: during the day the system implemented control-to-range and overnight the system was in safety mode (i.e., more relaxed control to reduce hypoglycemia risk).
  • Remote monitoring using DiAs connected to Dexcom’s G4 sensor reduced nocturnal hypoglycemia in a trial in young children at three summer camps sessions (n=20/camp [n=10 G4 + DiAs; n=10 G4 only]). Total study time was 1360 hours, of which remote monitoring was operational for 1314 hours (97%). For a deeper delve into the study, please see our coverage of Dr. Bruce Buckingham’s (Stanford University, Stanford, CA) dedicated presentation on the trial on page 12 of our EASD Day #2 Highlights report at https://closeconcerns.box.com/s/phnv2z6hpe8x4r81v1kw.
  • Just last week, a multi-center efficacy trial of closed-loop control using a control-to- range algorithm and the newest generation devices commenced. Participating centers include UVA Center for Diabetes Technology, Sansum Diabetes Research Institute (UC Santa Barbara), Padova (Italy), and Montpellier (France). This randomized crossover study consists of one 40-hour session each of open- and closed-loop control (DiAs runs both). Five patients are enrolled per site; patients are responsible for system communications.
    • The modular control-to-range algorithm is comprised of three modules: 1) an IOB tracking module (UCSB); 2) a range control module (Pavia); and 3) a safety supervision module (UVA). Importantly, the algorithm allows for enhanced control-to- range during the day for intensive treatment, but relaxes control overnight.
    • The closed-loop system consists of DiAs smart phone, which connects by USB to the Dexcom G4 receiver and by low power Bluetooth to the Tandem t:slim. The G4 Receiver, of course, wirelessly communicates to the G4 sensor. Dr. Kovatchev said that to the best of his knowledge, this was the first time a closed-loop used the G4 and t:slim.
  • Results from the first patient tracing were encouraging, with 83.4% of time in target range (70-180 mg/dl) and 100% of time in target range (80-140 mg/dl) overnight. Dr. Kovatchev drew attention to the accuracy of the G4 – the 12 fingersticks shown seemed to fall closely in line with the G4 tracer. Further, Dr. Kovatchev highlighted the “traffic light” system of DiAs, with a color charting beneath the tracer showing hypoglycemia lights. Dr. Kovatchev noted two examples: 1) when the blood sugar was rapidly declining the safety system picked up the event at 140 mg/dl, the yellow hypoglycemic light came on, and insulin delivery was cut; 2) when the blood sugar reached ~90 mg/dl, the red late came on indicating that carbohydrates were needed and hypoglycemia was avoided.

Closed Loop Control

Time in range of 70-180 mg/dl

83.4%

Time above 180 mg/dl

14.4%

Time in range of 80-140 mg/dl overnight (11:00 pm to 7:00 am)

100%

Number of hypoglycemic episodes below 60 mg/dl

0

OUTPATIENT ARTIFICIAL PANCREAS: SMART SENSORS AND ALGORITHMS

Claudio Cobelli, PhD (University of Padova, Padova, Italy)

Dr. Claudio Cobelli described three ongoing projects to improve closed-loop control: the latest AP@home trials, an improved Dexcom sensor, and an updated simulator. The AP@home consortium is in midway through a set of overnight, partially outpatient experiments (n=12) that use UVa’s latest DiAs controller, a specialized remote monitoring system, an algorithm that incorporates recent insulin delivery information in its control decisions, and a simplified protocol for open-loop insulin dosage at meals. Meanwhile Padova engineers are working with Dexcom to develop a “smart” CGM transmitter with new onboard processing algorithms to detect noise and enhance calibration. Dr. Cobelli also described several recently submitted modifications to the UVa/Padova metabolic simulator, which include a more-physiological nonlinear response to hypoglycemia, a model of glucagon counterregulation, and revised definitions of insulin-to-carbohydrate ratio and correction factor. Ongoing research on the simulator will introduce a new model of sensor error, improve the model of subcutaneous rapid-acting insulin, and attempt to “clone” the results of the initial AP@home studies.

  • Dr. Cobelli explained that several improvements have been introduced in the latest AP@home clinical trial, which is targeted to complete by the end of the year. The crossover-design study uses outpatient closed-loop during the day and inpatient closed-loop control at night; the design also includes exercise and video games. The first four patients have completed the study at Padova, and the trial is planned to conclude with four patients in Montpellier and four in Amsterdam. The DiAs system used in the new trial has been improved over that in the first outpatient European studies, and the MPC “observer” module has been modified to monitor the pump for information on insulin delivery (enabling more accurate glycemic predictions). Also, integration of open-loop meal control in the closed-loop scheme has been simplified. Pre-clinical simulations were run on a recently modified simulator (see below), and improvements have been made to the “worst-case analysis” CVGA grid used to tune the controller’s aggressiveness. (Basically, the previous grid scored a particular simulated patient’s performance based only on whichever was worse in a given experiment, the highest hyperglycemic excursion or the lowest hypoglycemic excursion. By contrast the new curvilinear grid would rate an algorithm’s performance differently if a patient’s respective maximum and minimum values were 300 mg/dl and 110 mg/dl, instead of 300 mg/dl and 70 mg/dl, for example.)
    • Dr. Cobelli presented data from one of the patients in the study (“as you can imagine,” he smiled, “I chose the best.”) This patient’s time in target range (70-180 mg/dl) was improved dramatically with closed-loop control (99.9%) compared to open- loop control (72.7%); the mean time in target for closed-loop control in all four patients was in the mid-80% range.
  • In collaboration with Dexcom, Padova’s bioengineering team is exploring improved CGM algorithms for better noise detection (Facchinetti et al., IEEE Trans Biomed Eng 2011) and enhanced calibration (Guerra et al., IEEE Trans Biomed Eng 2012). The published work on these algorithms has involved post-processing sensor data that had already been converted to a glucose signal. However, by building algorithms directly into a future Dexcom “smart” transmitter, Dr. Cobelli hopes to further improve sensor performance and simplify wireless communication in closed-loop systems. 
  • Modifications to the Virginia/Padova metabolic simulator were submitted to the FDA on October 17, 2012, and subsequent improvements are already underway. The changes currently under FDA review include a non-linear response to hypoglycemia, a counterregulation model that includes glucagon secretion, kinetics, and action), a new way to define insulin-carbohydrate ratio and correction factor (to mimic the way that real patients would determine these values), and an altered model of absorption parameters. In a CE-EGA of how well actual patient data (n=96) agreed with the old and new simulations, the new simulation performed significantly better in hypoglycemia (and agreed more closely with real data on interquartile range and high and low blood glucose indices, as well).
    • Ongoing work on the simulator includes a new model of CGM error, which is based on data that the Oregon researchers shared from a recent clinical trial (Castle et al., Diabetes Care 2012). This work includes individualized models of blood to interstitial glucose kinetics as well as models of the calibration function, sensor variability, and measurement noise. These components can be analyzed individually to see how errors in each might affect the sensor result. The researchers can also assign various probabilities to the likelihood transient artifacts, error codes due to noise, and disconnection of the sensor from the body, to see what these problems would mean for glycemic control.
    • Padova engineers are also changing their module of subcutaneous insulin kinetics to incorporate data from a clamp study of insulin lispro in 41 patients with type 1 diabetes. Dr. Cobelli thanked Biodel’s Alan Krasner for donating these data, and he expressed hopes that the updated module would be completed by the end of 2012.
    • To conclude, Dr. Cobelli explained that the simulation is being adjusted so that it can “clone” the clinical data from the AP@home CAT Trial. (As a reminder, this dataset includes a total of 141 traces from eight patients.) For example, these modifications would allow the simulation to incorporate intraday variability in glucose absorption and insulin sensitivity, as occurs physiologically.

PANEL DISCUSSION

Jeffrey Joseph, DO (Thomas Jefferson University, Philadelphia, PA), Eyal Dassau, PhD (University of California, Santa Barbara, Santa Barbara, CA), Frank Doyle III, PhD (University of California, Santa Barbara, Santa Barbara, CA), Edward Damiano, PhD (Boston University, Boston, MA), Boris Kovatchev, PhD (University of Virginia Health System, Charlottesville, VA), Claudio Cobelli, PhD (University of Padova, Padova, Italy)

Dr. Joseph: It seems feasibility has been demonstrated out of the clinic and in the home. What has to happen to make this a tool that patients can use in the real world setting? Is it two years off? Five years off? Is it just refinement that is needed or a major breakthrough?

Dr. Kovatchev: There will be advancements needed to bring this to seamless outpatient use, and there will be stages. We have a clear line of sight. The first step is to have reliable, robust wireless connection between devices in the closed-loop system and I believe that is coming sometime next year. Then, there will be some miniaturization of the devices, but the hardware will be to a large extent set. And we are beginning with relaxed safety modules taking care of hypoglycemia over night that will progress to intensive therapy with enhanced control. As soon as we get wireless connections, we can do larger studies.

Dr. Damiano: We’ve staged out our approach to include experiments at three levels. First, are feasibility studies in carefully monitored inpatients, which we have been conducting over the past four months and which will conclude for us next month. Then, we have transitional studies where you loosen the reigns a little bit. You might have nursing attention of one-on-four or one-on-eight. Experiments are longer in duration – five days, then two weeks. Beginning next month, we expect to do these over the course of the next 18 months in our group. These will include the Beacon Hill study, camp studies in children, and studies with adults wearing the device for a couple weeks and bringing it home and taking it to work. We’ll have close reigns on the participants, but not to the extent of feasibility trials. The last major step would be to move to a pivotal study. That will require a custom medical grade hardware/software platform that is dedicated to the task of controlling glucose. We want to build this over the next 18 months. The pivotal trial would have hundreds of subjects, be a parallel study and take about six months. I envision something fully closed-loop perhaps available in the next four to five years. In our case we are relying on the importance of glucagon, but we still need stable pumpable glucagon. We need reformulation of glucagon, not a glucagon analog, so that there can be a faster regulatory path, and then we need a dual-chamber pump that can deliver both insulin and glucagon.

Dr. J. Hans DeVries (Academic Medical Center, Amsterdam, The Netherlands): When you presented your third experiment results, you divided them between day one and day two. I like that. But perhaps it needs some justification from a statistical point of view. Why is better in the second 24 hours?

Dr. Damiano: When we talk about day one and day two, we’re talking about the system adapting in real time over the course of the experiment. To begin, the system does not know anything about the subject other than the subject’s weight; it is not informed about carb-to-insulin ratios, correction factors, basal rates, total daily dose, etc. The system begins dosing insulin conservatively. It then adapts to the higher insulin needs in adolescents, if that is required, or it might stay close to where it started, which is more typical of the insulin needs of adults. While it’s always adapting basal infusion rates every few minutes, the degree it may dial up may be greater in one subject than another. The point is we have a time scale of adaptation that looks like six to 12 hours to converge upon the subject’s insulin requirement. On the first day, control is not as good as what we expect on day two. We don’t know how much better it will get on days three, four, and five. We do see that control on day two is better and anticipate that it might continue to improve somewhat more over time.

Q: Are you planning to compare dual-hormone to single-hormone control in clinical trials? Maybe starting from a simulation mode?

Dr. Damiano: The studies we have done have been limited to dual hormone control. In order to assess the true potential of glucagon, we have to design a study that uses our own bihormonal and insulin-only control systems or work collaboratively with another group that follows the same clinical protocol and does insulin-only control. It’s hard to make comparisons between closed-loop systems when protocols are not handled in the same way.

Dr. Gary Steil (Children’s Hospital Boston, MA): Dr. Damiano, I think you told me in your last diabetes paper that the variability between subjects on any two days was as large as between any two subjects. In other words, each patient is a new patient on a different day. When you find these new parameters over the first six hours, what is the stability of that new identification? And for Dr. Cobelli, in the past, it’s been very difficult to identify new virtual patients. If the patient is changing very rapidly, how identifiable is the UVa/Padova model for all of the parameters? Are the patients really that variable? Is the same patient tomorrow different from the one today?

Dr. Damiano: When we described the variability from day one to day two in our previous study, we were referring to the variability in insulin dosing by the controller, not in glycemia. We analyzed this variability in insulin dosing between day one and day two in the same subject and compared that to the variability in insulin dosing between day one in that subject and day two in each of the other subjects. That paper was restricted to our adult cohort. What we observed from this analysis in our adult cohort was that intra- subject variability was not less than inter-subject variability. In the adult study, both are comparable and not very large – there is relatively low variability in both. There was not nearly as much inter-subject variability in that adult cohort as there is in the cohort now that includes both adolescents and adults. The variability between adolescents and adults is profound. For adolescents, we’re seeing insulin requirements that are two-to-three fold higher than for adults of the same weight.

Nevertheless, if you take the same adult subjects and run them in a study for three months, they might develop an inter-current illness along the way. They may require significantly more insulin during the illness and can start to look like an adolescent, in terms of their insulin requirement. The system must be able to automatically adapt up to meet those insulin-dosing demands. Alternatively, they may also have a vomiting illness, where the system must be able to automatically dial down the insulin. That is what I think our system must be able to demonstrate, and I think our data suggest that we can achieve this. Variability might not be that significant from day to day, but it very well can be from one week to another.

Dr. Cobelli: Thanks to the Bayesian identification strategy it is possible to identify the UVa/Padova model on the 24-hour glucose traces of the AP@home project. Important changes on glucose absorption parameters and insulin sensitivity were observed in each individual. The ability to well describe the traces starting from the a priori information of the UVa/Padova simulator provides an indirect validation of the simulator. We are midway in the analysis of the 141 traces.

Q: As to comparing insulin alone to insulin plus glucagon, we’ve published that study. It was a crossover design so people started on insulin plus placebo then switched to insulin plus glucagon. There was less hypoglycemia with the glucagon. So that study has been done, but as Ed pointed out, there is no pumpable glucagon.

You each mentioned exercise. Have any of you considered measuring something like heart rate and how it would input into the system? Maybe it could turn on glucagon earlier based on heart rate?

Dr. Cobelli: Heart rate is an important signal but may not be the signal. Sometimes it’s good in detecting the immediate change. But as exercise progresses, I don’t think heart rate catches up with physical activity. There is a variety of physical activity – low, medium, high. It is not just the exercise you are doing on a bike. We are doing experiments with the Mayo group. They have a sophisticated system to capture physical activity, and we have presented the first results but are still working to better understand the relationship between physical activity and the glucose signal (CGM and plasma). I’m not sure if heart rate is the signal. It is important, but not the totality.

Dr. Damiano: In particular, while you did the dual-hormone versus insulin-only comparison study, it has to be taken in the context of how that study was done. That study was conducted in sedentary subjects and will likely understate the importance of glucagon. What I think we’re going to see is greater importance of glucagon with physical activity. On the question of using input signals to the controller other than glucose, I think heart rate is very ambiguous and I don’t think that is something that should drive control. Various levels of exertion impact the way glucose is cleared. Extreme exercise can actually require enhanced insulin secretion while moderate exercise results in tremendous glucose clearance. I think the verdict is born out in glucose.

Dr. Steil: Can we identify all the model parameters without a glucose tracer?

Dr. Cobelli: Yes, we have been able to reproduce the traces of the AP@home patients by using the simulator. Obviously, we are using the a priori information of the simulator. The distribution of the parameters of the simulator probably resembles reality since we were able to describe a variety of traces in a variety of situations.

I certainly agree with all the comments that Boris made in terms of the timeline. Keep in mind, that there are ongoing studies in parallel in terms of understanding the physiology – the intra-day and inter-day variability of insulin sensitivity. And what is the model for stress? What is the model for exercise? These studies are being done in parallel with the technological advancements.

Q: Do any algorithms have the ability to discern hyperglycemia from eating versus hyperglycemia by pump failure? It warrants very different responses from algorithms and patients.

Dr. Doyle III: To distinguish between this, you need a fault diagnosis module – something to detect for abnormal events. It is impossible to do blindly from glucose alone. By marrying this with fault diagnosis tool you could manage it, but not with a straight MPC algorithm.

Dr. Cobelli: It is a different module; it has nothing to do with MPC. A paper is in press of our group in IEEE Transaction of BME describing this module.

Dr. Yogish Kudva (Mayo Clinic, Rochester, MN): Earlier this year in August we had a paper in Diabetes Care in low grade physical activity using accelerometers and it has a different effect as far as glucose lowering is concerned after a meal. We also compared heart rate and accelerometers in the same patients. We have a poster, number 39. And we have data being analyzed now on V02 max.

Dr. Howard Wolpert (Joslin Diabetes Center, Boston, MA): Dr. Cobelli, you said that you were integrating the glucagon counterregulatory response in the hypoglycemia module. But in type 1 diabetes, patients lose the glucagon response to hypoglycemia.

Dr. Cobelli: The glucagon model is based on the tracer data that has been done. It is what is the state of the art in terms of glucagon secretion, kinetics, and action. Incorporation of glucagon into the model allows the simulator to have glucagon in it.

Dr. Wolpert: So it’s only relevant to dual hormone closed loop systems?

Dr. Cobelli: Yes, exactly.

Dr. Hovorka: On the issue of prandial insulin dosing – there seems to be a question about whether to give priming boluses or have fully reactive systems. It’s quite an important decision about which way to go. Without prandial dosing, I’m not sure it’s possible to get as good of control as with open loop therapy.

Dr. Damiano: We have not included prandial dosing to the exclusion of testing our fully reactive system. Our first study tested a fully reactive system using venous blood glucose as the input to the control system. In our second study, we added a weight-based prandial dose at the presentation of the meal because of the 10- to 15-minute delay in the CGM signal. In the study we’re doing now, which will finish in a couple weeks, there are two arms. One arm includes adults and adolescents receiving an adaptive prandial bolus at the beginning of each meal. The other parallel arm includes adults and adolescents receiving entirely reactive control with no prandial insulin boluses. I entirely disagree that you cannot beat open loop control using a fully reactive system. In adults, we were able to achieve average blood glucose levels of 143 mg/dl with entirely reactive control, versus 138 mg/dl with a prandial insulin bolus. So we didn’t see much difference with or without a prandial insulin bolus in adults; in either case, this would correspond to an A1c of about 6.5%. In adolescents, we achieved average glucose levels of 171 mg/dl with hypoglycemia less than 1% of time. This is in a sedentary condition. You see that even moderate exercise causes greater glucose clearance, so we should see these numbers drop further in the outpatient setting. That corresponds to an A1c of 7.6%. That’s way better than the standard of care. It turns out it’s not a hard bar to beat open-loop therapy in adolescents.

Dr. Doyle: We need to inform the decision of the best strategy. We’re pushing hard. Data from ADA showed that we can get to within 70% time in range with a purely reactive algorithm. I think that competes favorably with prandial dosing.

Dr. Kovatchev: I have a slightly different take on prandial dosing. It can also can be beneficial from a design and regulatory point of view. If you have prandial dosing and the basal rate controlled by the person, then a range controller on top of this basal-bolus therapy, that controller has an adjunct claim – that’s opposed to a replacement claim in regulatory terms. That’s important for closing the loop and taking this out there.

Comment: Overnight glucose control is the time to get better control. It’s prime time for the closed loop to improve control over the open loop. I am concerned that by relaxing control, you are losing the greatest benefit.

Dr. Doyle III: I don’t dispute that at all. We relax measurements over night as we begin to worry more about the robustness of the system. My personal comment is that the greatest challenge is getting away from clean in-clinic environments and having truly free living. That will be a real tough test. In anticipating this, we need to minimize the risk of hypoglycemia and we offer this tool to relax control. I completely agree with you, but this is an alternative to turning it off.

Q: On zone MPC, you mentioned that between the range of 80-140 mg/dl, the algorithm would keep a constant basal. Does that mean when the subject eats a meal, the controller won’t give anything but basal until the glucose goes over 140 mg/dl? What’s the rationale for using a zone rather than single set point?

Dr. Doyle: It starts delivering when the prediction of glucose leaves the zone. If it’s showing a trajectory that will leave the zone, dosing will commence. In our years of working with doctors at Sansum, we’ve learned that there is no single magic value. Glucose is always in a range.

Artificial Pancreas: Clinical Aspects

IN-HOME USE OF A PREDICTIVE NOCTURNAL PUMP SHUT-OFF ALGORITHM

Bruce Buckingham, MD (Stanford University, Stanford, CA)

The esteemed Dr. Bruce Buckingham presented pilot-study results assessing a predictive low glucose suspend (PLGS) system, one that Dr. Buckingham characterized as “home-grown.” The system was designed with the intention of eliminating prolonged episodes of nocturnal hypoglycemia, while working “in the background” with a reduced number of alarms. The system only alarms if blood glucose is less than 60 mg/dl or greater than 250 mg/dl and does not alarm for system failure (instead, it reverts back to usual basal). “Glucose is important for you brain,” he said, “but so is sleep.” These home trials included 19 patients (375 nights total), and tested three iterations of the algorithm. Nights were randomized such that 2/3 of nights were controlled by active PLGS. After finding that PLGS reduced nocturnal hypoglycemia by ~50% and nights with prolonged lows by ~70%, Dr. Buckingham said that they are moving forward with the third version of the algorithm (which conferred the best overall nighttime glucose management) and are enrolling 44 subjects in a multicenter study with a goal of obtaining 1,600 nights of system use.

  • Nineteen subjects completed 375 nights, with 2/3 of nights randomized to predictive low glucose suspend (PLGS); the system is run from a bedside computer. Studies were conducted in homes, with no nurse present or remote monitoring. Participants ranged in age from 18-56 years old and in A1c from 6.0% to 7.7%.
  • The algorithm went through three stages of fine-tuning, such that three versions of the algorithm were tested. The first had a prediction horizon of 70 minutes with basal insulin being resumed when glucose was predicted to rise above 100 mg/dl. The second used a prediction horizon of 50 minutes with basal insulin resumed with any positive rate of change; there was no insulin suspension if glucose was above 230 mg/dl. The third iteration was the same as the second, but used a prediction horizon of 30 minutes.
  • After the third fine-tuning of the algorithm, the system had the lowest mean glucose at first shutoff and the lowest median peak glucose following fist shutoff. As Dr. Buckingham explained, one of the concerns with PLGS is that insulin suspension leads to rebound hyperglycemia, which did occur in the study he presented. However, the third iteration of the algorithm seemed to mitigate this risk somewhat.
 

Intervention

 

Algorithm 1 (n=5)

Algorithm 2 (n=12)

Algorithm 3 (n=77)

# Intervention nights

67

108

77

# Nights with Pump Suspension

52 (78%)

84 (78%)

41 (53%)

# Pump Suspensions Per Night

 

0

22%

22%

47%

1-2

43%

47%

31%

3-4

31%

19%

12%

5-8

3%

11%

10%

Mean Glucose at First Shutoff (mg/dl)

139

121

113

Median Peak Glucose Following First Shutoff (mg/dl)

185

146

135

  • Dr. Buckingham presented safety data showing that the number of nights with nocturnal hypoglycemia, and the duration of that hypoglycemia, was reduced with PLGS compared to control nights. However, the reduction in hypoglycemia seemed to some extent to come at the cost of higher overnight glucose. Dr. Buckingham polled the audience (see below) to see how much of a rise in overnight glucose people would be willing to tolerate in order to reduce hypoglycemia. He would be willing to accept about ~12 mg/dl. Of course, as this was a pilot study, we await the larger, better-powered study Dr. Buckingham is currently enrolling to determine whether these findings will be replicated, statistically significant, and clinically significant.
 

Algorithm 1

Algorithm 2

Algorithm 3

 

Control (n=5)

Active (n=5)

Control (n=10)

Active (n=12)

Control (n=8)

Active (n=9)

Number of nights

38

67

48

108

37

77

Mean Bedtime Glucose (mg/dl)

166

155

139

148

152

157

Mean Overnight Sensor Glucose (mg/dl)

145

158

123

137

133

148

Nights with CGM values 71-180 mg/dl

76%

71%

91%

90%

94%

89%

Nights with CGM values > 180 mg/dl

63%

78%

29%

56%

49%

60%

Nights with CGM values >250 (mg/dl)

29%

24%

6%

15%

8%

21%

Nights with CGM values < 60 mg/dl

29%

11%

35%

17%

24%

10%

> 60 min duration

11%

3%

15%

5%

3%

3%

> 120 min duration

3%

0%

4%

3%

0%

0%

  • Dr. Buckingham asked the audience: to avoid severe nocturnal hypoglycemia, what is the highest increase in mean overnight glucose values that would be acceptable to you? Surprised by the majority response, Dr. Buckingham noted that a 45 mg/dl increase for 24 hours would raise your A1c 2%. 
    • 5 mg/dl: 8%
    • 12 mg/dl: 18%
    • 20 mg/dl: 33%
    • 25 mg/dl: 41%
  • He closed his presentation with a second audience response question: for nocturnal closed-loop control, what is the maximum number of times you would be willing to wake up to do a calibration on a nightly basis?
    • 0: 82% (Me too, said Dr. Buckingham.)
    • o1: 17%
    • o2: 1%
    • o3: 1%

RESULTS FROM THE AP@HOME PROJECT

J. Hans DeVries, MD (Academic Medical Center, Amsterdam, The Netherlands)

Dr. J. Hans DeVries presented an overview of the AP@home Project’s CAT trial, including new secondary analyses as well as both intent-to-treat and per-protocol results (previously presented at ADA 2012 and EASD 2012, respectively). As a reminder, CAT was a 23-hour, crossover comparison of open-loop control to two different MPC algorithms (one from Cambridge and another from the iAP consortium, both “de-tuned” to prioritize hypoglycemia reduction). Compared to open-loop control, both algorithms significantly reduced time spent in hypoglycemia but significantly increased time spent in hyperglycemia. The Cambridge algorithm delivered significantly less insulin than that from iAP, and both algorithms used less insulin than open-loop control. Results were equivalent regardless of whether CGM was calibrated with YSI or glucose meter measurements. Control was generally similar whether centers used manual input of the algorithm’s commands (control decisions made every 15 minutes) or automated control with the Artificial Pancreas System (decisions every five minutes), though time in hyperglycemia was significantly lower with manual control.

  • In addition to reviewing the main findings of the CAT Trial, Dr. DeVries unveiled several secondary analyses. For the primary intent-to-treat and per-protocol analyses, see page 64 of our ADA 2012 coverage at http://www.closeconcerns.com/knowledgebase/r/c6afb200 and page 6 of EASD 2012 coverage at http://www.closeconcerns.com/knowledgebase/r/9f88794c.
    • Compared to YSI reference values, the Dexcom Seven Plus sensors used in the CAT Trial had a mean absolute relative difference (MARD) of 15.1% in the intent-to-treat analysis. The MARD fell only slightly to 14.1% in the per-protocol analysis, though the most common reason for data exclusion from the per-protocol analysis was CGM failure (defined as a MARD above 50% for 45 minutes or more). Such instances of CGM failure accounted for 0.4% of sensor values in the open-loop setting but 13% and 17% of time under control by the iAP and Cambridge algorithms, respectively. He alluded to an ADA 2012 oral presentation from the Cambridge group in which the Dexcom Seven Plus was found to have more long lag times than Abbott’s FreeStyle Navigator, but he added that these problems seem to have been largely addressed in the Dexcom G4 Platinum.
    • Overall control did not differ when CGM was calibrated with self-monitoring of blood glucose (SMBG) as opposed to YSI. Dr. DeVries indicated that this bodes well for the transition to outpatient studies. The next challenge will be to replace YSI reference values with a glucose-sensing technology that can be used by patients at home.
    • Dr. DeVries also compared centers using “automated” vs. “manual” closed- loop control. In three of the six CAT centers, sensors and pumps were linked to the control algorithms via a laptop running the Artificial Pancreas System (APS), which enabled control every five minutes. The other three centers were located in countries where regulatory clearance of APS was thought unlikely. In these centers, communication from the sensor to the control algorithm to the pump was carried out by clinical trial staff every 15 minutes. The results in these sub-studies were similar to the main results, except that a lower percentage of time was spent in hyperglycemia in the manual vs. automated centers (31.6% vs. 44.4%). We speculate that automated closed-loop control may have placed a strain on the devices involved, limiting the systems’ functionality, whereas the15-minute vs. five-minute difference was relatively inconsequential given that thealgorithms had been designed not to be aggressive. Dr. DeVries noted that in a sub-sub- analysis of manual-control centers, the Cambridge algorithm outperformed open-loop for time in range (64.7% vs. 51.7%); he did not hypothesize as to why this result was seen.

MOVING TO ARTIFICIAL PANCREAS AS A THERAPY FOR TYPE 1 DIABETES: HOW TO SELECT AND QUALIFY THE CANDIDATES

Eric Renard, MD, PhD (Hospital Lapeyronie, Montpellier, France),

Dr. Renard gave a unique presentation focused on two major questions: 1) how much can we trust the artificial pancreas today? and 2) who are the ideal candidates for an artificial pancreas? On the first, Dr. Renard outlined what has been proven thus far with the AP: feasibility (Steil et al., Diabetes 2006); reduction of nocturnal hypoglycemia (Hovorka et al., Lancet 2010); improved mean blood glucose (Breton et al., Diabetes 2012); and the possibility of outpatient use (Cobelli et al., Diabetes Care 2012). He also outlined the biggest current weak points of the AP: CGM sensors (the one he most emphasized), delays in insulin absorption, and mealtime control. Dr. Renard outlined two groups of patients that will be ideal candidates for the AP: those with recurrent hypoglycemia and those that are afraid of hypoglycemia (“hypophobic”) and purposely keep their blood glucose high. Although patients with high glycemic variability might be good candidates for the AP, Dr. Renard believes the subcutaneous delivery route may be challenging –intraperitoneal may be a better route in these cases. He concluded with a review of how to prepare candidates for use of the AP.

  • Dr. Renard believes the major benefit of the artificial pancreas is reduction in hypoglycemia. He explained that this has been demonstrated in nearly all closed-loop studies Dr. Renard specifically referenced the results of the CAT trial, which found equivalent overall mean glucose control between open loop and closed-loop control, but a significant reduction in hypoglycemia. Encouragingly, the CAT trial also took place at several centers without any closed- loop experience, demonstrating the potential feasibility of rolling the AP out to a broader population.
  • In Dr. Renard’s opinion, CGM remains one of the weakest points in the artificial pancreas. He explained that CGM was responsible for the highest number of problems in the CAT trial. Pump errors occurred less than 1% of the time, while software problems occurred 1.6% of the time. By contrast, 7% of time was spent without a sensor signal and almost 5% of time was spent without a sensor reading due to sensor failure. We note that the Dexcom Seven Plus was used in this trial. We expect that once Dexcom’s G4 Platinum is used in AP studies, these rates will drop precipitously (as a reminder, one of the benefits of the G4 Platinum is improved signal transmission).
  • The AP is most ideal for those with recurrent hypoglycemia (i.e., overzealous insulin delivery) and those who are afraid of hypoglycemia (“hypophobic”). Dr. Renard showed CGM trace examples of both patients, explaining that automated insulin delivery stands to benefit both problem areas. However, he feels that different experiments are needed for these two populations and they should not be mixed. Dr. Renard did agree that patients with high glycemic variability might be good candidates for the AP, though he thinks “the subcutaneous route is challenging.” The big question is whether the subcutaneous route – fraught with its own challenges – will be able to manage variability. In his view, intraperitoneal delivery is probably the best route for this population.
  • How to prepare candidates for AP use? 1) Move the patient to CSII; 2) experiment with sensor use; 3) train the patient to carbohydrate count (patients will still need to input carbs at mealtime); 4) train patients on the AP platform; 5) start closed loop at the hospital (or a similar controlled environment) and check remote monitoring before leaving for home use.

PANEL DISCUSSION

Roman Hovorka, PhD, MSc, BSc (University of Cambridge, Cambridge, United Kingdom), Arleen Pinkos, BS (U.S. Food and Drug Administration, Silver Spring, MD), Bruce Buckingham, MD (Stanford University, Stanford, CA), J. Hans DeVries, MD (Academic Medical Center, Amsterdam, The Netherlands), Eric Renard, MD, PhD (Hospital Lapeyronie, Montpellier, France)

Dr. Hovorka: If you are moving patients with the AP to the home, in your experience how much more knowledge and training would you say the closed loop requires compared to what they already know?

Dr. Renard: For patients who have used a pump, a sensor, and know how to carbohydrate count, within two days they can go home. We need to select patients who want to improve their diabetes management and if they are dedicated, I don’t expect it to be a long time in the hospital before they can go home.

Dr. DeVries: That being said, considering pump penetration around the world, for many countries it would be a huge step for a lot of patients. There are countries with pump penetration below 5%, so from that perspective it’s not for everyone.

Dr. Jeffrey Joseph (Thomas Jefferson University, Philadelphia, PA): Is the improved control with intraperitoneal insulin due to suppressed hepatic glucose production?

Dr. Renard: Animal studies suggest that this occurs. In these studies, the main difference compared to subcutaneous insulin is with the speed that hepatic glucose production is suppressed.

Comment: It is evident that for any algorithm, whether PID or MPC, one size does not fit all. We need to do more physiology studies so we can individualize the algorithm to the patients. From a sensor standpoint, that is one of the weakest links in closed-loop systems. We need to understand the physiology of the delay in glucose sensing. That will hopefully improve sensor algorithms.

Dr. Ken Ward (Oregon Health & Science University, Portland, OR): Dr. Renard, I agree with the examples you gave of likely candidates. However, I would note that our closed- loop studies have enrolled patients who are keep themselves in good glycemic control but with great personal cost – 12-to-13 fingersticks per day and constant vigilance. These patients say that they greatly appreciate being in closed-loop studies and having a vacation from diabetes management, even for a day-and-a-half. They might be good candidates for an artificial pancreas as well.

Dr. Buckingham: I would add that the typical adolescent, who is on a “vacation” most days, is also a good candidate.

Dr. DeVries: I would agree that a one-day break can be of huge value to patients.

Dr. Renard: At the same time, in these patients it would probably be hardest to show the benefit of AP. The patient with whom we did outpatient research said that the only benefit was that he didn’t have to care about diabetes. But this is difficult to explain scientifically.

Dr. DeVries: Let alone from a reimbursement point of view.

Q: To Hans, in the study you just did, was the meal bolus a full calculated meal bolus? And then Eric, you showed a sensor profile where it overestimated by a 1/3. We are concerned about a sensor that is 1/3 high, but we routinely acknowledge that a patient can’t count carbohydrates. How big is meal bolus and how important is it to get it right?

Dr. DeVries: The bolus was determined by the algorithm and the input differed by the two algorithms. I’ll defer to Roman.

Dr. Hovorka: The Cambridge algorithm gave 80% of the calculated meal bolus.

Comment: We routinely acknowledge that in open-loop parameters patients will have the wrong carbohydrates. It strikes me that the closed loop should be more robust to these errors.

Dr. Renard: If you look at current methods, you just look at what’s on the plate and compute carbohydrates in your brain. We need some tools for patients that will be easy to manage. We have developed better tools now, but in this we are still very conservative. There is room for technology to improve this. We can use components of the closed loop to help patients count carbohydrates.

Comment: Carbohydrates are difficult to determine if patients go to a restaurant and have something they didn’t prepare. Dieticians in California were asked to rate meals on the amount of carbohydrates and they pretty much all got it wrong. Counting carbohydrates is difficult.

Dr. Renard: You cannot control everything in life.

Dr. DeVries: If I go to US restaurants, the number of carbohydrates is often on the menu.

Dr. David Klonoff (Mills-Peninsula Health Services, San Mateo, CA; University of California, San Francisco, San Francisco, CA): Every one of these systems uses a specific combination of hardware. How do you choose a combination? Also, when you consider all of these possibilities, not to mention the variety of control algorithms and ways of handling meals, what do you think this means for establishing performance standards?

Dr. DeVries: In the CAT Trail we used the products that were made available to us. Companies like Dexcom invest a huge amount of time, and also money, making devices for closed-loop research. The same goes for Insulet – these companies should be commended.

Dr. Renard: We should have the companies working on platforms that work with any pump and any sensor. As a patient you should not be constrained in your choice of pump because you want to have an artificial pancreas. This will be a new world of diabetes management; I think that the environment should be open. The platform will be validated on pumps X, Y, and Z, not limited to a single manufacturer.

Dr. Buckingham: I think that we all use what is made available to us and appreciate the companies that make their systems available. I think that if you take a car and it gets you from point A to point B – i.e., a good glycemic result – it is a good automobile. However much it costs, I think that it’s worth it.

Dr. Hovorka: I think that instrumentation has been the limiting factor thus far.

Q: For the predictive system, did you impose any constraints on the suspend time?

Dr. Buckingham: I didn’t state that the threshold was 80 mg/dl and a single episode couldn’t be longer than two hours. There was a maximum of three hours overnight for total suspension time. That was protection against a sensor failing overnight.

Q: In the future we may have ultrafast acting analog insulin. What will be the impact of having that insulin?

Dr. Hovorka: I think it will have a tremendous affect on efficacy and safety. If you have something in your sleeve, as soon as possible would be good.

Dr. DeVries: It is always difficult to predict future and if we would have these faster do we need to go the intraperitoneal route?

Dr. Renard: It depends on whether it is variable in absorption.

Dr. Stuart Weinzimer (Yale University, New Haven, CT): To address Dr. Klonoff’s question: as primarily a clinician, I could envision picking out a closed-loop system that I think is best for individual patients, just as I now do when choosing among diabetes products now. There won’t be one system ideal for every scenario.

Dr. Hovorka: Agreed.

Q: To repeat a question asked to the panel of engineers, what do you think has to happen to get an artificial pancreas on the market? I was surprised that none of the engineers mentioned the sensor as being a problem, but I noticed that two of you did.

Dr. DeVries: I think the engineers did a very good job answering the question. One thing that they perhaps did not mention is the legal ramifications of closed-loop control. We will probably see severe hypoglycemia with a closed-loop device at some point in the future. Who is responsible? What are the consequences? My view is that if the trials are decisive and the labeling is clear, we will just have to accept these risks.

Dr. Buckingham: The Helmsley Charitable Trust data indicates that 7.5% of patients are having seizures or loss of consciousness each year; people are dying now from the dead-in-bed syndrome. I think that if you can reduce these risks with an artificial pancreas, then it is good to use one. The sensors are getting better, and this will lead to better system performance.

Dr. Renard: When you use closed-loop algorithms, you reduce time in hypoglycemia and thereby in a way improve sensor performance, since accuracy is worst when glucose is low.

Dr. Renard: Hypoglycemia reduction isn’t seen in every case when mean glucose goes up, though; using a closed-loop algorithm is preferable to simply relaxing control.

Dr. Hovorka: How are we going to evaluate home studies? With CGM? Is CGM good enough?

Dr. Buckingham: I think it is. You don’t have another choice other than someone living in the house with a YSI machine. We’re going to have to rely on continuous glucose.

Dr. DeVries: We see now in translation trials that we completely rely on CGM, but groups apply telemonitoring too. Your group, Roman, is going out of hospital without telemonitoring. I’ll turn the question back to you, is it safe?

Dr. Hovorka: My question was different. Once you collect CGM, we found in our analysis that there was a bias introduced in using a CGM.

Dr. Renard: Safety authorities make it so difficult for outpatient trials. People with diabetes have nothing when they inject insulin before going to bed. There is no CGM, no measurement...what should be forbidden is to use insulin or a pump without CGM or frequent measurements. It’s curious to see how many hurdles there are to outpatient trials, when it is a safer way than how we use insulin now.

 

-- by Adam Brown, Kira Maker, Joseph Shivers, and Kelly Close