FDA/JDRF/NIH Workshop on Innovation Towards an Artificial Pancreas, April 9-10, 2013; Bethesda, MD Full Commentary – Draft

Executive Highlights

The FDA/JDRF/NIH Workshop on Innovation Towards an Artificial Pancreas (AP) was held at the NIH campus in Bethesda, Maryland from April 9-10. The meeting provided an excellent forum for discussion, bringing academia, industry, government, and patient voices together in one room. We heard rich discussion on how artificial pancreas development has progressed and is expected to progress in the coming years. Although the conversation had a narrow focus at times as speakers aired their concerns about some of the very minute details concerning AP development, or considered AP development in an isolated context, many attendees encouraged the community to adopt a “big picture” view and remember that the artificial pancreas should be compared to how type 1 patients are treated at present. We thought this was an especially important message for the FDA to hear.

One of the biggest takeaways from the meeting was the importance of continued, frequent open dialogue among invested parties: regulators, researchers, funders, industry, patients, etc. We look forward to watching how the meeting’s conversation informs near- and long-term AP development, especially as companies figure out where to direct their efforts and how to invest in this area. Certainly, bringing an artificial pancreas to market has lots of challenging elements from a business model perspective; that said, we loved hearing JDRF’s Dr. Aaron Kowalski urge companies to seize the opportunity. Big picture, there seemed broad agreement that, this technology has the potential to improve clinical outcomes significantly, even with a product as simple as predictive low glucose suspend.

Below we include our full coverage of the FDA/JDRF/NIH Workshop on Innovation Towards an Artificial Pancreas. Following the conference theme section, we include coverage of individual presentations divided by day. We particularly encourage close reading of the panel discussions, which we thought contained some of the best insights of the meeting.

  • We were reminded of the challenges industry faces along the path to bringing an artificial pancreas to market: Pivotal study design seems to pose a particular challenge; the FDA’s final AP guidance, while valuable, is not as specific as similar guidance on the drug side. Dr. Courtney Lias (FDA, Silver Spring, MD) explained that the document has “a lot of ‘if this, then that.’” Though the conditional formatting allows the guidance to adapt to the various types of systems it will apply to, some seemed frustrated by the lack of clarity. In addition to regulatory concerns, legal concerns were also a topic of discussion. “Would one death stop [artificial pancreas development] in its tracks?” asked one attendee. Speakers seemed acutely aware of the litigious environment that the artificial pancreas would enter, but many urged that risk be considered in the context of the current standard of care. Said Dr. Aaron Kowalski: “Look at what we’re doing today – drawing up insulin in a syringe…open loop therapy is terrible right now. As these studies ramp up, it’s going to be very clear that these systems are much safer than what we’re doing right now. We need to capitalize on that. We need to get companies comfortable with a pathway to market.” Dr. Kowalski charged, “I want to throw down the gauntlet to industry. We’re working well with industry, but I would like to see within the next 24 months every pump that is speaking to a sensor have low glucose suspend.” We hope that the delay in Medtronic’s MiniMed 530G (threshold suspend pump and Enlite sensor) approval will not discourageindustry investment on this front. Certainly, there is a case to be made the predictive low glucose suspend is safer, as CGM trend accuracy is better than point accuracy.
  • Many speakers described inter-device communication as the “weakest link” in the closed loop, to use the words of Dr. Boris Kovatchev (University of Virginia, Charlottesville, VA). A decade ago, algorithms, sensor accuracy, and speed of insulin were identified as the biggest barriers to artificial pancreas development – the focus on communication standards is a clear sign of how far the field has come. It’s also a stark reminder of how much rests on industry to making the closed loop happen – without pumps and sensors and algorithms that seamlessly talk to each other, research moves much more slowly and the path to commercialization will be much longer. Speakers uniformly agreed that communication failures need to be addressed, though discussion was more nuanced on what a potential solution could look like. Some proposed uniform communication standards, while others were fans of the physical integration of system components. Dr. Joseph Cafazzo (Toronto General Hospital and the University of Toronto, Canada) gave a convincing presentation describing the efforts of industry members of the Continua consortium to adopt universal communication standards (IEEE, Bluetooth, and USB) for pumps, sensors, and BGMs.
  • Speakers discussed whether artificial pancreas design should lend towards ambition or iteration. Said Dr. Ed Damiano (Boston University, Boston, MA) of his team’s path towards the artificial pancreas, “I do not want perfection to be the enemy of the good, but I don’t want incrementalism to be the enemy of the good either. I think we can do this in four years.” [While we believe this timeline is quite ambitious, even for Dr. Damiano, the quest for a homerun first product vs. an average one strike us as a fascinating discussion.] At GTC Bio in late April, Dr. Damiano discussed how today’s CGMs were colored by perceptions of early devices that “weren’t ready for primetime.” He commented that for his artificial pancreas, “It’s how ambitious we’re being that makes the difference…Hopefully the tremendous improvement in glycemic control is what will get people to wear these devices.” (For his full discussion, see page six of our GTC Bio 2013 report: Conversely, Dr. Moshe Phillip (Schneider Children’s Medical Center of Israel, Petah Tikvah, Israel) gave a more incremental perspective through the lens of the DREAM consortium: “The enemy of the good is the best. Closing the loop for 24 hours and having the best solution with a mean blood glucose of 110 mg/dl jeopardizes the progress.” He believes that “even if we pay the price of remote monitoring at the beginning until we have enough safety data, I think we can get there faster [than four years].” As studies become more ambitious and research progresses towards a commercialized product, we expect this topic of discussion to heat up. In our view, it’s a particularly critical question for industry, who must decide what basket(s) to put their eggs in – predictive suspension? Overnight closed-loop? Full 24-hour closed loop? Each has its pros and cons from an R&D, regulatory, and reimbursement perspective.
  • “Perfect should not be the enemy of the good.” If we had to pick the one message we heard most over the course of the two-day workshop that appeared to resonate with most meeting attendees, this would be it. Said FDA’s Arleen Pinkos, the Agency recognizes that the first artificial pancreas out of the gate will not be perfect for everyone. Ms. Pinkos believes that smart labeling is the best way to communicate system limitations to patients. To quote Closer Concerns’ own Ms. Kelly Close (Close Concerns/diaTribe, San Francisco, CA): “Without the first backpack insulin pump, we wouldn’t have small pocket sized pumps now. Without the GlucoWatch, we wouldn’t have today’s amazing CGM. On the artificial pancreas, let’s also keep this perspective – the first product won’t be perfect and has no need to be. What we’re doing today as patients is SO far from perfect.”
  • There was a clear divide between Medtronic and Dexcom on the topic of CGM redundancy: Medtronic has made redundancy a critical piece to its artificial pancreas development. Mr. Rajiv Shah (Medtronic, Northridge, CA) focused on Medtronic’s use of data from multiple/orthogonal sensors and algorithms to ensure CGM reliability, accuracy, and safety. Dr. Tom Peyser (Dexcom, San Diego, CA) provided a counterpoint to Medtronic’s approach, arguing that highly accurate systems drastically reduce the value of or need for redundant sensing technology. We think the debate on the merits of redundancy is only just beginning to emerge. Certainly, comfort factor for patients will play an important role in this regard, as will cost.
Table of Contents 


Day One



Griffin Rodgers, MD (Director, NIDDK, Washington, DC); Alberto Gutierrez, PhD (Director, Office of In Vitro Diagnostic Device Evaluation and Safety, FDA, Silver Spring, MD); Aaron Kowalski, PhD (Vice President, Treat Therapies, JDRF, New York, NY)

Representatives from each meeting sponsor – FDA, NIH, and JDRF – provided opening remarks to begin the conference. All were optimistic and encouraging (“we’ve become like a family”), though there was a clear emphasis on past and future challenges as well.

  • Dr. Alberto Gutierrez provided a historical perspective on the FDA/NIH/JDRF meetings, which began in 2005, had a rocky bump in 2010, and are now on an uptrend trend. The three groups began working on CGM and “had a good working relationship back then.” Dr. Gutierrez called the 2008 workshop “a very good meeting” with “ a lot of excitement.” Unfortunately, communication between the FDA and investigators subsequently “became more difficult” at the 2010 meeting. According to Dr. Gutierrez, that workshop’s disappointing tone led to the Agency and investigators “at each other’s heads.” This rocky relationship culminated in November 2011 when JDRF took out a full-page ad in the New York Times and Washington Post. Dr. Gutierrez was frank in saying that the Agency “had some issues to work out,” and now “we have turned the corner in many ways.” The main change has beenputting the AP group into one office, a move that has allowed the Agency to work hard at improving overall response time and interaction with researchers. He also closed optimistically, echoing CDRH Director Dr. Jeffrey Shuren’s goal to make the US first in the world in terms of medical devices coming to market first. In our view, this is a highly ambitious goal given the Agency’s recent history.
  • Dr. Aaron Kowalski emphasized the “incredible progress” that’s been made since the first AP innovation workshop in 2005. He reminisced that at that initial meeting, there were few AP studies even going on and the transition was just beginning to the age of portable CGM. Fast forward to today, where there is “tremendous data” from outpatient studies and many devices that are approved and in use. Yet, Dr. Kowalski also underscored the “huge unmet medical need today,” referencing the high A1cs seen in the T1D Exchange and the significant burden posed by hypoglycemia. He thanked the FDA for working tremendously hard on the final AP guidance, and also acknowledged JDRF’s partnership with NIH. Promisingly, the latter allows the organizations to pursue co-funding initiatives and leverage each other’s strengths to accelerate research.
  • Dr. Griffin Rodgers discussed the efforts at several government agencies to prioritize development of an artificial pancreas. The FDA has established the critical path initiative, Health and Human Services has developed the Interagency AP Working Group (helped in the development of the final FDA AP guidance), while NIDDK and its partners have supported a variety of cutting edge research. Dr. Rodgers highlighted that it “takes a village,” giving a particular shout-out to JDRF’s important efforts to get the Special Diabetes Funding renewed. He also highlighted the Helmsley Charitable Trust’s important efforts to coordinate research. Dr. Rodgers closed with optimism that the meeting’s focus on innovation would result in a valuable exchange of ideas and strategies for future developments.

State of the Art: Design, Results, and Challenges from the Latest Clinical Studies


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

Dr. Bruce Buckingham reviewed pilot study results assessing a predictive low glucose suspend (PLGS) system and presented initial safety data from the team’s ongoing randomized outpatient study. 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 rate). For the ongoing randomized trial, Dr. Buckingham is targeting 42 subjects each completing 42 nights. Each night is randomized, such that 21 nights use active PLGS and 21 nights serve as the control. The study will assess the primary outcome of nights with sensor glucose <60 mg/dl; secondary outcomes include mean overnight glucose, percentage of time in range (71-180 mg/dl), area under the curve >180 mg/dl, percentage of nights with sensor glucose >250 mg/dl, and number of mornings with ketones >0.6 mmol/l. To date, 49 subjects have enrolled (five of whom did not complete the study run-in period) and 21 individuals have completed the study for a total of 1,354 randomized nights and 1,266 valid nights (i.e., at least four hours of CGM data was available for the night). Given that the study is blinded, Dr. Buckingham could not share efficacy data; however, pooled safety data from the 1,354 nights showed no episodes of diabetic ketoacidosis and no seizures. Morning ketone levels above 0.6 mmol/l occurred in less than 1% of nights and there were no morning ketone levels above 1.5 mmol/l. The study is slated to complete in May and we look forward to learning whether its findings will corroborate the promising reductions in hypoglycemia seen in the initial pilot study.

  • The pilot study included 19 patients (375 nights total) and tested three iterations of the algorithm; data was first presented at the Diabetes Technology Meeting (see our discussion on page 19 of our DTM full report at The first algorithm had a prediction horizon of 70 minutes, the second used a prediction horizon of 50 minutes, and the third iteration used a prediction horizon of 30 minutes. The third algorithm conferred the best overall nighttime management; as such, a 30-minute horizon algorithm was used for the larger outpatient study.
  • Notably, final visit data from the 19 patients indicated that 100% of patients would opt to use a similar system if it was commercially available, despite 47% of patients (n=9) saying the system “occasionally” disrupted their sleep and 11% of patients (n=2) saying the system “often” disrupted their sleep. Fifty-eight percent of patients felt that the hypoglycemia alarms were not appropriate, which they mainly attributed to sensor inaccuracies. We think the 100% positive response towards continued use of such a system is an impressive testament to the system’s benefits and exemplifies the sentiment that “perfect should not be the enemy to good.”

Subjects that felt system disruptive of sleep

n (%)


8 (42%)


9 (47%)


2 (11%)

Subjects that felt hypoglycemia alarms were not appropriate


11 (58%)

Because wanted to change thresholds


Because of sensor inaccuracies


Questions and Answers:

Dr. William Tamborlane (Yale University, New Haven, CT): What percent of patients wore the sensor six to seven days per week of the 19 that finished?

A: Most were using it almost continuously.

Q: The Agency’s concerns with low glucose suspend is that it would increase A1c over time and our contention is if you are using it for the prevention of nighttime hypoglycemia, patients will wear the sensor during the day too and they will actually lower A1c.

A: That is a good point. Our trial cannot address A1c because we randomized each night.

Q: How did you deal with exercise?

A: Our algorithm does not incorporate exercise. It was an information item. We are also doing work using an accelerometer with the system to tell when the person is sleeping, which is a good way to tell if they are sleeping on the sensor.

Dr. Robert Vigersky (Walter Reed National Military Medical Center, Washington, DC): What is the relationship you’ve seen between ketones and snacks? Could there be starvation ketosis?

A: That’s a good question – final data cuts will come later.


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

Dr. Ed Damiano – interrupted mid-slide three by an NIH fire alarm test – shared exciting results from his ambitious five-day, ongoing outpatient Beacon Hill closed-loop study. In the five patients completed thus far, overall average blood glucose (CGM) has been a stellar 124 mg/dl on days two and three and 121 mg/dl on days four and five, with less than 1% of blood glucose values <60 mg/dl. Notably, the FDA approved the IDE for Dr. Damiano’s 32-patient, five-day-long camp study today, which will take place this summer at Joslin. He noted that the team’s new mobile device “has increased our productivity by eight or nine fold,” enabling the team to perform more closed-loop experiments in the first eight months of 2013 than in the previous four years combined! In Q&A, Dr. Damiano also shared his two key wish list items: calibration free CGM and stabilized glucagon.

  • The ongoing Beacon Hill outpatient closed-loop study involves five-day experiments in 20 adults with type 1 diabetes (>21 years). The randomized crossover design compares five days of closed-loop therapy to five days of usual care. Patients have free run of a three- square-mile area of downtown Boston. Patients have point of care capillary blood glucose tests during the day (Hemocue) under 1:1 nursing conditions. Patients are allowed to eat and exercise as they please, with a maximum of two drinks per day. At night, participants sleep in a hotel with venous blood glucose monitoring (GlucoScout) under 1:2 nursing. The study began in February 2013 and Dr. Damiano expects to collect over 3,000 hours of data.
  • The Beacon Hill Study is using an iPhone 4 controller, a Dexcom G4 Platinum CGM/receiver, and two Tandem t:slim pumps (insulin and glucagon). With help from SweetSpot, the G4 Platinum CGM receiver is attached to the iPhone via hardwire through the 30- pin connector – we had only previously seen the Navigator version of the iPhone controller, which was far more brick-like than the sleeker Dexcom version. The Tandem pumps are controlled wireless (Bluetooth) from the iPhone controller. Glucagon reservoirs are changed every day, while insulin is changed every other day. The researchers elected to go with the Dexcom G4 Platinum based on the results of their head-to-head-to-head comparison studies: Dexcom G4 Platinum had a MARD of 10.8%, compared to 12.3% for the Navigator and 17.9% for the Enlite.
  • The study is essentially fully closed-loop control, with the control algorithm initialized only with a participant’s weight. The controller does have a manual pre-meal priming bolus component, which allows patients to select if this is more than, less than, or about equal to the number of carbs they typically eat (i.e., no carb counting). The algorithm learns over time from its pre-meal priming bolus performance and improves its dosing. In the study, patients are told about the pre-meal priming bolus, but are not required to use it. In fact, if a patient forgets to take the priming bolus before the meal, he or she is supposed to allow the controller to take care of the meal by itself. To date, few patients have been taking the pre-meal priming boluses, which makes the results below that much more real world. We also like the no carb- counting and learning algorithm approach to meals, since highly accurate carb counting is quite difficult in practice.
  • Dr. Damiano showed preliminary, very impressive results from five Beacon Hill patients tested to date (“we’re nearly normalizing blood glucose”). The average is certainly one of the best we’ve seen in a daytime closed-loop study, especially with the low prevalence of hypoglycemia. Dr. Damiano emphasized that these averages predict an A1c around6.1%, handily beating the standard of care. He also noted that patients in the study are probably more active than they otherwise might be, perhaps because the presence of glucagon gives patients more confidence in avoiding lows.
  • The FDA just approved the IDE for a Beacon-Hill-like 2013 summer camp closed- loop study. Thirty-two patients will be studied over four weeks in a randomized, crossover design comparing five days of closed loop therapy to five days of open loop therapy. The study will occur in an integrated camp experience at Camp Joslin (boys) and Clara Barton (girls). There will be point-of-care blood glucose checks and study staff providing 24-hour coverage via telemetry. Like Beacon Hill, the camp study will be transitional and real world in nature, with no restrictions on eating, activity, etc.

Questions and Answers

Dr. Ken Ward (Oregon Health and Science University, Portland, OR): You initialize the system with just weight. But wouldn’t total daily dose as an index of insulin sensitivity allow you to come in with tighter control sooner?

A: What happens during inter-current illness? Why would I do this if my system works this well with just weight? Why pour all that information into the device. And what about someone with an A1c of 9% vs. 12%?

Dr. Ward: You can adjust A1c for total daily dose.

A: Let’s let the device find that out. I do not need that information. I’ve got two subjects at 70 kilos. One is a 15 year old with raging hormones vs. a 70-year old. We’re controlling these people and beating standard of care. And there’s still the issue of inter-current illness. Any system that requires that kind of insulin information is going to have a hard time.

Dr. Ward: I meant you could use it just to start out.

A: It’s transient. People go through changes over months. You can see a two or three fold change in insulin sensitivity. Or worse, say you have a vomiting illness. You need to be able to dial that down. I’m getting good results on day one. We’re testing a time course that is relevant to inter-current illness. That is key. You’ve got to be able to adapt to that in a six to 12-hour window. The system initializes at the same weight, but could end up dosing at 1.5 units per kilos vs. 0.3 units per kilo. That’s what we want to be able to show. I don’t want a system that relies on that total daily dose.

Q: Have any subjects had inter-current illness during the study?

A: We do not allow subjects with inter-current illness in the trials. We have studied a system that knows nothing more than weight. You could say adolescence is a chronic state of inter-current illness [laughter]. What we haven’t tested is adolescents with inter-current illness. The highest total daily dose we have seen is 1.5 units per kilo per day, and the lowest is 0.3 units per kilo. That’s a five-fold difference.

Q: How does the glucagon release in this outpatient study compare to your previous study?

A: We’re using more. The last study was 0.5-0.7 mg per day. Now we’re dosing 0.8-0.9 mg per day.

Q: In terms of Hemocue fingersticks every two hours, are those blinded?

A: Yes, they are blinded. One of the highest things on my propriety list besides stable glucagon is calibration free CGM systems. We do the fingersticks calibrations before dinner and before breakfast in this study. We do have rules – if you are rising or falling too rapidly, we will delay the calibration. We generally do two calibrations per day with the Dexcom G4 Platinum.

Dr. Robert Vigersky: What’s the inclusion/exclusion criteria for subjects? Are we studying the cream of the crop and the best-controlled and most reliable patients? What about those with hypoglycemia unawareness?

A: We used to only allow strictly C-peptide negative participants, but we are relaxing that in the camp study going forward. We used to only allow A1cs of 9% or below, now we’re opening the doors to that. The averages in these studies are lower than 8% or 8.5% in the usual care group. Maybe it’s a bit of a self- selecting group.


Roman Hovorka, PhD (University of Cambridge, United Kingdom)

Complimenting his recent ATTD presentation, Dr. Roman Hovorka gave an overview of the Cambridge team’s three ongoing home studies using the FlorenceD closed-loop system. The ongoing studies consist of an adolescent overnight trial (APCam06; n=16), adult overnight trial (Angela03; n=24), and adult day-and-night trial (AP@home02; n=18). To date, 14 patients have completed APCam06 (for over 250 closed-loop nights), three have completed Angela03 (for over 80 closed-loop nights), and three have completed AP@home02 (for 21 closed-loop days). Notably, the studies do not employ remote monitoring. Individual glucose profiles from the outpatient studies suggest positive benefits of FlorenceD and that overnight and day-and-night closed-loop control is feasible. Further, a snap shot of controller effort data over 21 days for one adolescent in APCam06 demonstrated how difficult maintaining glucose control overnight is without such a system given intra-individual variability in insulin needs (for background, controller effort is measured by the percentage of insulin used relative to programmed basal). Dr. Hovorka reserved the last minute of his presentation to engage in broader discussion about closed-loop studies. He raised awareness to the potential for CGM to overestimate treatment effects when it is used to simultaneously control the system and assess the outcome. We don’t hear much discussion on this front and wish time had allowed for Dr. Hovorka to delve into how best to address this bias. For greater detail on FlorenceD and study design, see our discussion of Dr. Hovorka’s excellent ATTD presentation on page 12 of our full report at

Questions and Answers

Arleen Pinkos (FDA, Silver Spring, MD): Can you comment on frequency of blood glucose monitoring and how much diversity there was between that frequency?

A: The minimum compliance was at least four blood glucose readings per day. We asked patients to use to use the FreeStyle meter, but some used their own meter. Our requirements were at least four readings and the range was up to seven.

Dr. Aaron Kowalski: With respect to the insulin load variation day to day, my interpretation is that it could relate to the infusion set. How much of that variation was due to food consumption versus bad infusion sets?

A: That is an important topic. We are capturing the data, but we haven’t really analyzed it yet. My view is that infusion sets cause a major failure, so those are the extremes. My other view is that sometimes people overdose for evening meals and this would mean that not as much insulin was needed for the rest of night. The factors are difficult to assess and there is a lot of underlying variability. We are looking into that.

Q: Can you comment on day-and-night closed-loop control and the use of meal boluses? Where are you with those today?

A: Meal boluses are used based on the bolus wizard.

Q: So patients still have to do what they do in open-loop control?

A: I’d love to be able to run, but I think we need to walk first.

Q: Giving meal boluses in advance creates a huge risk factor in children under seven years old or people who don’t count their carbohydrates correctly.

A: We did a study in the CRC where we didn’t bolus in meals. Our system can cope, but we don’t recommend it to be used as a standard feature.


Linda Gonder-Frederick, PhD (University of Virginia Health System, Charlottesville, VA)

Dr. Linda Gonder-Frederick gave an excellent behavioral perspective on the artificial pancreas and use of diabetes technology. She presented a number of psychological models and studies that characterize adoption and continued use of diabetes technology. We particularly valued her discussion of a UVa patient focus group on concerns over wearing an artificial pancreas – the two common themes that emerged were 1) patients’ loss of perceived control over their own diabetes management (“humans are control freaks”) and 2) systems that lack personalization. Dr. Gonder-Frederick concluded that behavioral research “lags far, far behind” technology development research and we need more research to improve the cost-benefit ratio of diabetes technology. We couldn’t agree more and hope to see far more emphasis in this area – with only ~30% of US patients on insulin pumps and ~5% on CGM, there is much work to be done on patients’ perceived barriers to tech and how to overcome them.

  • Dr. Gonder-Frederick emphasized the importance of perceived ease of use and usefulness. She called these perceived factors the primary determinants of technology acceptance. This ultimately leads to intention to use and behavior. We thought this was a very interesting point, since perceived ease of use is much broader than just designing a simple-to-use product – it also has to do with marketing, education, training, etc.
  • UVa focus groups have revealed two main patient concerns with using an artificial pancreas. The first is a loss of perceived control over diabetes management – said Dr. Gonder- Frederick, “For humans, loss of control is anxiety provoking. Humans are control freaks.” She believes we need to incorporate perceived control into technology design (i.e., making us think we are in control). This reminded us of the encouraging potential of hybrid systems that could augment open-loop therapy, though not completely take it over. The second main patient concern was lack of system personalization; in other words, how could an AP take into account all the information a patient takes into account? Dr. Gonder-Frederick encouraged a greater look into personalized technology – ultimately, she believes it will allow systems to maximize individual risk tolerance for high and low blood glucose.
  • Dr. Gonder-Frederick presented a temporal model of patients’ adoption of diabetes technology. First, patients decide whether or not to adopt a device – this is termed “initial use” and is influenced by things like perceptions, distress factors, hassle factors, and ease of use. Then, patients decide whether to continue using a device or not – does it improve glycemic control, quality of life, and reduce diabetes burden? The last step is incorporating technology into the daily diabetes management routine (“long-term adherence”).
  • Research has revealed several predictors of discontinued or inadequate CGM use: high hassle factor/low satisfaction, a mismatch between patient expectations and patient experience, physicians’ failure to review CGM data, and patient difficulty interpreting CGM data. Dr. Gonder-Frederick believes there are things providers can do up front to give patients more realistic expectations about CGM. She also covered a study from Dr. William Polonsky of 2,500 Dexcom CGM users. The two major predictors of discontinuing CGM use were physicians’ failure to review CGM data with patients and patient difficulties interpreting CGM data.
    • We hope that better software in the coming years can directly address both of these issues. In that vein, UVa has developed a Glycemic Risk Feedback output, which reports data in an intuitive red-yellow-green light manner. It provides feedback about glycemic variability, overall glycemic risk, and hourly risk. It is being tested right now with patients.
  • While consistent CGM use is difficult for many patients, some data supports less frequent use: 40-50% use (Scaramauzza et al., 2011), intermittent use (Vigersky et al., 2012), and a stepped care approach (Riveline et al., 2012). Dr. Gonder-Frederick believes it is important to think in terms of flexibility, tailored therapy, and “different strokes for different folks.” We completely agree – certainly, while 24/7 use is optimal, intermittent CGM use is better than none at all.
  • Out of 93 CGM studies listed on, only two are focused on behavioral interventions (Dr. Lori Laffel on family teamwork and Dr. Tim Wysocki on family behavioral therapy).

Questions and Answers

Dr. Yogish Kudva (Mayo Clinic, Rochester, MN): In terms of closed-loop subjects, one of our concerns is that the same patients keep getting enrolled in short duration studies.

A: This is something we need to do something about. We cannot keep studying the same people. We need to expand out subject population and study the less ideal populations. We also need some standardization of how we ask people things. There are opportunities to ask anonymous questionnaires.

Q: What about next gen device studies – it seems like the CGMs used in many of these studies were older generations.

A: The technology is constantly changing and we are running to keep up with it. But there are some fundamental issues that won’t change. We should be looking for what they are. One is people’s need for perceived control. That’s not going to change. It’s definitely a challenge. We already have so many different kinds of technologies. For something like overnight closed loop control, I think parents will adhere. That’s not an issue.

Dr. Moshe Phillip (Schneider Children’s Medical Center of Israel, Petah Tikvah, Israel): Is there any consensus about which tools to use?

A: Yes, I think I’m going to write a paper on that. We need a paper on that and we need to think about this. I’m part of the fear of hypoglycemia research effort. I think that’s a critical factor. We need diabetes specific measures of quality of life, as well as more general measures.


Lutz Heinemann, PhD (Profil Institut für Stoffwechselforschung, Germany)

Dr. Lutz Heinemann reviewed the AP@home consortiums’ clinical studies as well as gave background on a single-port artificial pancreas (being developed by GRZ/4A). The single port fuses the sensor and infusion cannula to enable a single insertion site: both ex-vivo (where the sensor rests outside the body with glucose drawn up to contact the sensor) and in-vivo (where the sensor is attached to the outside of the infusion cannula) approaches have been considered. Prototypes of the system have been tested in murine models and recently, single-port experiments have been done in healthy humans. As we understand it, the in-vivo approach is being more actively pursued; data on the single-port system is expected to emerge at ADA.

  • The AP@home’s CAT Trial showed that closed-loop control gave similar time in target compared to open-loop control with diminished hypoglycemia. For background and details on the CAT trial, see Dr. J Hans DeVries (Academic Medical Center, Amsterdam, The Netherlands) overview on page 21 of our Diabetes Technology Meeting full report at
  • The consortium’s CIPHER study suggested improved glucose control on day three of pump use vs. day one of pump use, with no significant inter-pump differences. The aim of CIPHER was to access the effect of pump time on glucose control and whether the effect was mediated by pump tubing length (Medtronic Paradigm was compared to Insulet OmniPod).

Average Postprandial Glucose (mg/dl)




Day 1



Day 3






  • Dr. Heinemann reviewed data from a head-to-head comparison of the Abbott FreeStyle Navigator I, the Medtronic Enlite, and the Dexcom G4 Version A (i.e., the version used with the Animas Vibe in Europe, not the more advanced G4 Platinum available in the US). In addition to assessing accuracy, sensor longevity was assessed. For results details, see our discussion of Dr. Yoeri Luijf’s (Academic Medical Center, Amsterdam, The Netherlands) presentation on page 44 of our ATTD Full report at


Ali Cinar, PhD (Illinois Institute of Technology, Chicago, IL)

Patients “don’t want to do the work of the beta cells,” said Dr. Ali Cinar. As such, his team’s closed-loop system does not use meal and activity announcements. The system uses a generalized predictive control algorithm that incorporates energy expenditure and galvanic skin response information gleaned from BodyMedia’s Sensewear armband; his presentation was tailored to the engineer in the room as he described algorithm’s design and rationale in detail. Dr. Cinar also reviewed results from a small fully closed-loop study in three patients (seven experiments) using the system and spoke to the promise of integrating a hypoglycemia early warning system. See our discussion of the closed-loop study on page 29 of our ATTD full report at


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

Dr. Boris Kovatchev reviewed UVa’s mobile artificial pancreas platform, Diabetes Assistant (DiAs), emphasizing the system’s flexibility through a modular design, unified safety system, user interface, and upcoming studies. Three studies have been completed thus far on the system, with data from two of them to be presented at ADA 2013. Notably, the four studies planned for 2013-2014 will use a Dexcom Gen 5 sensor wirelessly sending data directly to the DiAs smartphone (i.e., no translator), which will ultimately dose a Tandem or Roche insulin pump. Dr. Kovatchev called communication between devices the biggest roadblock in AP development and something that “desperately” needs to be changed.

  • Two studies of DiAs will be presented at ADA 2013 this summer in Chicago: The first was a November 2011-May 2012 42-hour feasibility study of the DiAs platform with a USB connection to the Insulet OmniPod iDex PDM. The second study is a multi-center November 2012-May 2013 trial that included five subjects per site and two 40-hour outpatient sessions. That study had no meal restrictions, mandatory restaurant dinners (alcohol permitted), and 45 minutes of light exercise. The DiAs system was connected via USB to the Dexcom G4 Platinum receiver. DiAs is connected to the Tandem t:slim via low energy Bluetooth.
  • The planned 2013-2014 studies of DiAs will use the Dexcom Gen 5 sensor directly connected wirelessly to DiAs. The investigators will use either Roche or Tandem insulin pumps. The four slated studies include 1) a multisite trial of advisory control and telemedicine (NIH); 2) a multi-center outpatient trial of closed control (AP@ home); 3) diabetes camp studies of closed loop control (Helmsley Charitable Trust); and 4) a multi-center home trial of closed control (Jaeb Center and JDRF).
  • The Diabetes Assistant is a modular, portable AP platform based on an Android smartphone. Dr. Linda Gonder-Frederick following human factors studies in 2011 developed DiAs’ user interface. The system incorporates hypoglycemia and hyperglycemia traffic lights: green means things are okay, yellow means the system is doing something to mitigate hypoglycemia or hyperglycemia, and red means human intervention is needed. DiAs also has a unified safety system (USS Virginia) that gauges the safety of insulin delivery. If the USS Virginia determines delivery is not safe, it will alert the user or attenuate delivery.
  • Dr. Kovatchev emphasized the impressive flexibility of the DiAs platform. It can run with a CGM alone; a pump alone; open loop control with SMBG, CGM, and a pump; advisory mode with SMBG, CGM, and a pump; and closed-loop control. These modes can run interchangeably and can be used differently as studies require (e.g., Dr. Bruce Buckingham’s summer camp study used DiAs in a remote monitoring capacity, while other studies have used itfor closed-loop control). The modular design allows researchers to drop their own algorithms into the system. Importantly, DiAs also allows inputs above and beyond glucose readings and insulin delivery – for example, Dr. Marc Breton has studied the addition of heart rate and accelerometry data.

Questions and Answers

Dr. Robert Vigersky: What types of parameters go into the safety system (USS)?

A: It’s a one-sided MPC algorithm. It can give less insulin but not more. From the insulin request, it’s working down to attenuate it. It has all the characteristics of MPC – looking at insulin delivery, glucose traces, and predicting where glucose is going. It only attenuates and discontinues insulin. It has been used in probably 200 people so far.

Dr. Vigersky: In those 200, how often does it activate?

A: Small attenuation of insulin delivery is common. It happens almost every 24 hours. That’s the yellow light of the safety system – it is working and doesn’t bother the patient. The red light means human intervention is needed. That’s very rare and happens once a day at most. It signifies that with the current insulin on board, even if the pump shut off right now, hypoglycemia would result.

Q: What is the impact of not having a common communication standard? Is that something that needs to be addressed moving forward?

A: I have repeatedly stated that the communication between devices is the weakest link in the closed loop. We desperately need this.


Moshe Phillip, MD (Schneider Children’s Medical Center of Israel, Petah Tikvah, Israel)

Dr. Phillip reviewed data on the DREAM studies presented at ATTD 2013. See pages 13-17 in our ATTD 2013 full report at

Questions and Answers

Dr. Steven Russell (Harvard University, Boston, MA): Do you have thoughts on overnight- only closed-loop control – is that a viable product? What about from a regulatory standpoint? Or do you see it as a step towards 24-hour closed-loop control.

A: I don’t know even one parent who sleeps a full night at our clinic. They get up in the middle of the night and check if the child is breathing, sweating, cold… I think that nocturnal hypoglycemia is an unmet need and should be a specific indication for closed-loop control during the night before we reach 24 hours.

Q: How would you keep people from using a nocturnal system during the day? Once they have it, wouldn’t they use it all the time?

A: That’s a wonderful question. I’m not afraid of them using it during the day. That’s why we started doing daytime studies – just to make sure it is safe. We’re not trying to show efficacy; we’re focusing on safety.

Dr. Ron Pettis (Becton Dickinson, Franklin Lakes, NJ): Can you speculate on the decrease in variability?

A: It was the algorithm and the way we developed it. We take into consideration changes in insulin sensitivity and other parameters.

Dr. Gary Steil (Children’s Hospital, Boston, MA): I’m wondering about the switch to pre- programmed basal rates from overnight fuzzy logic control. What if the fuzzy logic basal rate was turned down, the patient was hypoglycemic, and then a higher open-loop basal rate was turned back on?

Dr. Eran Atlas (Schneider Children’s Medical Center, Petah Tikvah, Israel): When changing the basal, we’re changing it as a temporary basal. It will revert back to the default program. Usually, that’s in the morning, so the patient is awake and can control it.

Dr. Atlas: So during the night there is not a pre-programmed basal rate?

A: Right – it’s totally on the logic.


Arleen Pinkos (FDA, Silver Spring, MD)

Dr. Arleen Pinkos gave the FDA perspective, with particular emphasis on “truth in labeling.” She opened her presentation by building on the positive energy from investigators’ presentations that morning – results look promising, she said. The Agency recognizes that “one size does not fit all” and that the first artificial pancreas out of the gate will not be perfect for everyone; ultimately, patients and their HCPs will determine whether the AP is right for them, she said. However, when it comes to weighing benefits and risks during regulatory review, Dr. Pinkos explained that the Agency places great weight on whether users will thoroughly understand the limitations of the system and when they may need to take additional action. Thus, Dr. Pinkos challenged the audience to determine how best to communicate clinical performance (and potential limitations) through labeling. Said Dr. Pinkos, we need to relay the complexity of study design in an easy to understand way that is concise yet comprehensive, and suitable for all audiences – this is no small task. While Dr. Pinkos finished her presentation with time to spare (one of the rare few), disappointingly, no Q&A session was held after her presentation in order to allow adequate time for the round table discussion that followed.


Roy Beck, MD, PhD (Jaeb Center for Health Research, Tampa, FL); Kelly Close (Close Concerns, San Francisco, CA); Barry Ginsberg, MD, PhD (Diabetes Technology Consultants, Wyckoff, NJ); Aaron Kowalski, PhD (JDRF, New York, NY); Lori Laffel, MD (Joslin Diabetes Center, Boston, MA); Ed Damiano (Boston University, Boston, MA); Bill Tamborlane, MD (Yale School Of Medicine, New Haven, CT); Moshe Phillip, MD (Schneider Children’s Medical Center of Israel, Petah Tikvah, Israel); and Chip Zimliki, PhD (Medtronic Diabetes, Minneapolis, MN)

Dr. Richard Bergenstal: Just to start, is there really a need for this exciting technology?

Dr. Beck: There certainly is a need. You can assess that in different ways. If you look at the percentage of patients with type 1 diabetes that actually meet target and minimize hypoglycemia, it’s a minority of people. There’s a need for better ways to manage type 1 diabetes. We need to reduce burden and improve quality of life. Take that together, there’s a tremendous need.

Ms. Close: I don’t think that there is any other therapeutic area when patients are commended for being so far from normal. Patients are eager to get back to normal, and if we had something like the AP, we would all be much closer to normal. There is so much anxiety and work and stress that comes on the families and partners of those with diabetes, and all of that is so silent. Something like this would have a profound impact.

Dr. Laffel: One of the challenges Roy articulated is we aren’t achieving goals by A1c levels. We haven’t really reduced the occurrence of hypoglycemia nor the presence of long-term complications. Despite wonderful tools, we need to talk about the need for adherence. Until we have seamless approaches to care that don’t require a lot of input, we’re going to be burdened with having behaviors as a focus of any technology. There will be that behavioral component with any technological advancement.

Dr. Tamborlane: In pediatrics, the mean A1c in adolescents is almost 9%. Still. That’s as bad as the control group in the DCCT. We have inadequate ways to assess the quality of life of technologies. That might be a problem down the road, especially as we become more cost conscious. We’ve seen this in the JDRF CGM trial. If you use standard tools, the quality of life perceived by patients with diabetes is pretty good. In a DCCT manuscript in prep, we’re using fairly sophisticated methods to measure quality of life. One is a perfect score, and the difference between perfectly healthy diabetes patients and those on dialysis was a tenth of a point. We need better ways to assess quality of life.

Dr. Kowalski: I have a very provocative question. Are we crazy right now? What I’m hearing is that we have this pressing unmet medical need. We want data, perspective from patients, and perspective from clinicians. And what we’re seeing from all the investigators who have spoken is that these systems are much safer, giving much better glycemic control, and folks don’t want to give these devices up. I worry we are paralyzed by over analysis to some degree. We have a tool that could transform diabetes and we are sitting on it to some degree. What is the hang up in our minds that is preventing us from moving faster?

Dr. Damiano: This perception that we’re not moving quickly, I don’t understand. We are flying. We have a device that needs to be tested. We just entered into an outpatient study; we just got approval to test the device in children. In eight months, we will have done more closed-loop hours than in 4.5 years combined. I don’t feel like we are delaying things, I can’t imagine moving faster. I think the technology we have will be transformative, but we need to test it in a sufficient number of people and circumstances. I do not want perfection to be the enemy of the good but I don’t want incrementalism to be the enemy of the good either. I think we can do this in four years. I think we are moving at a reasonable pace. I don’t want to do it so quickly that we compromise what’s out there. If something goes wrong, it will be the device that will be at stake; the entire initiative will be at stake. We have to do the transitional and pivotal studies.

Ms. Close: I think we are flying too. I totally agree with Arleen, there has been so much progress, but for those of us who look at drug side, sometimes things get really stalled as well. We are flying and it’s good we are talking about what regulators want to see and, with respect to the commercial landscape, what does liability look like. There are so many patients that are going to want this.

Dr. Phillip: I don’t think we need four years. We have to have a safe product. Even if we pay the price of remote monitoring at the beginning until we have enough safety data, I think we can get there faster.

Dr. Ginsberg: One of the problems is that people fail to look at opportunity cost. Patients with type 1 diabetes on intensive therapy have a high rate of severe hypoglycemia – maybe as high as two events per year. When we talk about safety, we have to take that cost into account. We’ll never get a device that absolutely prevents all hypoglycemia from happening.

Dr. Robert Vigersky: FDA has said that they’re not going to tell you what those safety parameters should be. What does the pivotal trial look like in terms of safety parameters? Can we inform FDA about safety?

Dr. Phillip: It’s easy. In terms of safety, it’s reduction in events of hypoglycemia and time spent in hypoglycemia, without increasing severe events, without DKA, and without increasing A1c. That’s enough. We don’t have to show the patient is happy and we don’t have to show that we improved A1c. The enemy of the good is the best. Closing the loop for 24 hours and having the best solution with a mean blood glucose of 110 mg/dl jeopardizes the progress.

Dr. Tamborlane: There are two aspects of safety. One is the benefit of the closed loop: preventing or reducing the risk of hypos. Despite what we’ve heard, we still don’t have the Veo approved. We should also have a predictive low suspend pump. Then there’s what Ed was talking about – the potential injury due to system failure. That’s a lot more challenging. There are unusual cases and there should be sufficient mitigation to prevent horrendous events from happening. This is why we don’t have everyone climbing on board to start on closed loop right now.

Dr. Sanjoy Dutta (JDRF, New York, NY): Can we hear about the design of pivotal trials with respect to end points? What are we looking for? It can’t be A1c. What do we agree the end point should be?

Dr. Bergenstal: Is time in range still a good marker?

Dr. Beck: I think it depends on the patient population. There won’t be one-size-fit-all for outcomes. If we take patients who have high A1c, we want to reduce A1c. Whereas if we take a group with A1c less than 7%, I think there we are looking to reduce hypoglycemia, which would be biochemical. It depends on whether you are doing a night study too. I think overnight we can look at time in range and reducing hypoglycemia.

Dr. Tamborlane: I have a controversial point. Look at all the data - the control patients are doing too well. If you have a treat to target vs. open loop and you are looking at A1c differences, it should be non- inferiority. If you do just as well with closed loop and reduce the burden of therapy by 90%, you have a big winner.

Dr. Roman Hovorka: To be approved in the UK, the system has to be cost effective. We should aim for those patients who will benefit most: those with high A1c and those with hypoglycemia. From my viewpoint we need these patients in our trials. If it were an overnight study, implementing A1c would be challenging, so maybe time in target is appropriate.

Dr. Damiano: With respect to the timing of pivotal trials, people are saying four years is a long time. In 2008, I remember people talking about closed loop being available now. In a system that just suspends glucose at night, you have a smaller effect size, but we are going to absolutely crush the control group. Our effect size will be so big that our pivotal trial can be smaller and faster – six months – compared to what can be done with an inferior system that’s not as ambitious. I think the pivotal study does drive a lot of the timeline. We are going with a system that can really substantially improve quality of glycemia.

Dr. David Klonoff: In addressing the targets for trials, I was on a JDRF panel that looked at this. We looked at three groups of people. First, are those with too much hypoglycemia. The goal is less hypoglycemia with no sacrifice in A1c. Second is people in poor control – we want a better A1c with no increase in hypoglycemia. The third group of people are doing fine – we don’t want to see A1c or hypoglycemia get worse; we just want to improve quality of life.

Todd Zion (cofounder, SmartCells): Aside from the stability of glucagon, why not use it? Dr. Damiano, you are the only one really adopting it in a clinical manner.

Dr. Damiano: I am not the only one. Ken Ward is using it. There are a lot of groups. I think we will see popularity over time. I think it will shine in outpatient settings. When people are physically active,

glucagon is critical. It is dysfunctional in type 1 diabetes. So we have a dual hormone problem in type 1 that we are addressing. The biggest obstacle has been that it is not a stable, pumpable drug. But the fact of the matter is, there has been a lot of attention as a result of what we’ve done and what Ken has done. I’m seeing small pharmaceutical companies coming up with stable glucagon on a fast regulatory pathway. Tandem is developing a dual chambered pump. We are getting into more sophisticated real-life studies and these other things are coming in parallel. I think there will be fewer barriers.

Dr. Tessa Lebinger (FDA, Silver Spring, MD): I have a scientific question for the panel. Is anybody looking at correlation of either glycemia or insulin requirements dependent on how much insulin has been infused one, two, or three days earlier? Dr. Damiano presented some data previously that patients with antibodies had longer duration of action of insulin. You get patients where no matter what they do they are running low. Other days they are running high without explanation except possibly, low affinity antibodies. Is blood glucose on a given day a function of a couple days ago? Is any one looking at these longer-term associations?

Dr. Boris Kovatchev: Claudio Cobelli and colleagues from the Mayo Clinic looking at inter-day variability will answer that question. The answer is yes, there are people looking at it.

Q: We have Apple systems, Android system, two Linux systems…Is the fact that we are choosing different systems making sharing results more difficult? Can we move towards one system?

Dr. Kovatchev: That’s why we’re introducing this – for people to use it and move faster. For choice of operating system, there are some reasons to believe that at the first stages, you must have an operating system that is medical grade. The only way to do so is with Android. We would love to do that on an iPhone, but it would require some participation from Apple.

Dr. Hovorka: There are limitations in Europe when it comes to manufacturing responsibility. Someone must take legal responsibility. There are also cost challenges. What is more limiting is connectivity to the pumps and CGMs. That has been holding up the progress.

Dr. Damiano: I have no desire to run this on an iPhone. I don’t envision this running on an iPhone. If there is a medical grade system that can run on Android, we should consider that. But I think of it as just a miniature laptop that gets us out of the CRC and into the real-world setting. Android or iPhone is a portable computer. The algorithm is not going to live on this. It’s just one step towards the final embodiment.

Q: Will any AP be restricted to a small number of users? Would one death stop this in its tracks?

Dr. Ali Cinar: I’ve had this question in my mind for a while. At the end of the day, most of the studies are in the proof of concept phase. I don’t think that any of the universities will be selling APs to the general public. We haven’t heard from industry about additional elements that are necessary to produce a mass scale AP system. It integrates many components, but it also has legal problems. A university can sell 1,000 of these and make a small profit, but if there is one litigation, they would pay 10x more and go bankrupt. It would be good to hear from industry on the things that would motivate them.

Dr. Zimliki: I will say that our president has devoted a lot of time in moving the AP initiative forward. I do think there are concerns related to the risks. I love the progress we have made, but we have to be concerned about what happens when the sensor is wrong. I think there are really interesting answers to that in the next session.

Anna McCollister-Slipp (Co-founder, Galileo Analytics): The biggest issues we’ve had as we’re trying to think of how to put together a steering committee and a public-private- partnership relates to data silos. As we assess the potential of the AP and where we are going to go in the next three years, whether regulators or patients, what we’re going to need is – and we have power to – get better data on device stability, pump data, and device failures. The FDA only has the capability of getting reported device failures. We don’t have any context for really judging these devices. The data I’ve seen is much safer than what I see on the day to day. We need to break down data silos and get data on pump safety and the number of hypoglycemia events, whether severe or non-severe. We need this data to properly judge the safety of the artificial pancreas.

Dr. Tamborlane: I think there are fundamental differences between open-loop and closed-loop CSII. For 30 years or more with CSII, there has been remarkably few if any events where the pump was responsible for the over delivery of insulin that caused injury. Usually it related to patients’ interactions with the pump and usually it is under delivery.

Dr. Jeffrey Joseph: I have experience in the operating room trying to document and do evaluation of rare events. An adverse event happens rarely, and there is usually a series of events that led up to that. What do we have to do make this device in the hands of the end user safe enough, especially so that it is taking the patient out of the loop as much as possible. If the patient does too much, it actually may do harm and increase the risk of an adverse event. When you take a meter and put it in the hands of a lab researcher, the results are better than when you put it in the hands of an end user.

Dr. Kowalski: I think we’re stuck on the trees and not looking at the forest. We need to focus on the comparator. These trials are pretty straightforward. Look at what we’re doing today – drawing up insulin in a syringe. There are kids who don’t test ever. They are dosing insulin on no information. These systems are going to blow away open loop therapy. Open loop therapy is terrible right now. As these studies ramp up, it’s going to be very clear that these systems are much safer than what we’re doing right now. We need to capitalize on that. We need to get companies comfortable with a pathway to market. The systems are not going to dump 300 units of insulin into you.

Dr. Alberto Gutierrez: The liability issue is a strange one because there is enough law out there. When things go through a PMA and are approved by FDA, a certain liability gets taken away. FDA is certainly very aware of the problems. We are clearly trying to take a measured approach, but things happen that you make you wonder if we are moving too fast. An example just happened recently on a system that one would have thought was well vetted by now and we wouldn’t make stupid mistakes: glucose meters. A company didn’t think about values over 1,000 mg/dl. They coded the system so that when it a patient’s blood glucose value was 1024 mg/dl or higher, it would give an erroneously low value. What about if you do this in the closed loop? We need to approach this in a way that makes sense. I’m not sure we’re quite there yet. I think the timeline of four years to an artificial pancreas is a doable one, though companies will need to vet their systems and do lots of safety studies. [Editor’s Note: Dr. Gutierrez was referring to LifeScan’s recent recall of the OneTouch Verio IQ, Verio Pro, and Verio Pro+ meters; see our report at for more information].

New Developments in Modeling, Algorithms, and Technology


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

Dr. Frank Doyle described his team’s work to bring personalization into algorithm development. To set the stage for his discussion, Dr. Doyle explored the fundamental tradeoff between controller robustness and performance. Dr. Doyle simplified the demonstrative mathematical formula with a single tenant: the ability of a controller to handle variability (e.g., noise, varying insulin action, uncertainties) is diametrically opposed to performance. As such, he argued that the key to balancing robustness and performance is mitigating uncertainty through better patient descriptions. Algorithms have traditionally been personalized either by 1) adaptive control, which is used when patient characteristics are poorly understood or change unpredictably and the system learns day to day; or 2) initialization, which is used when sufficient a priori information about the patient is available to adjust available settings. However, the adaptive approach generally takes six to seven days to reach optimal performance and the initialization approach requires a sufficiently rich data set, which may often be unavailable. Dr. Doyle and his team are exploring a third approach whereby the algorithm’s patient model is personalized a priori with total daily insulin. “Patient customization is crucial for closed-loop AP algorithms,” concluded Dr. Doyle. His statement seemed to reflect the permeation of personalized medicine into every level of diabetes care – from the doctor’s office to the devices to the algorithms that run the devices.


Wayne Bequette, PhD (Rensselaer Polytechnic Institute, Troy, NY)

Dr. Wayne Bequette discussed fault detection theory relevant to the artificial pancreas, with particular focus on insulin set failure and sensor attenuation. Drawing from lessons learned in aircraft and chemical refinery fault detection, Dr. Bequette explained that closed-loop control can potentially mask problems. As such, he underscored that measured output (e.g., glucose value) doesn’t tell the whole story - you really need to understand what’s going on with input (e.g., insulin delivery, insulin on board), hesaid. To date, Dr. Bequette has focused on two fault detection models: infusion set failure and pressure- induced sensor attention (i.e., when a patient rolls onto the sensor during the night causing sensor attenuation; PISA). Dr. Bequette and his team are exploring real-time PISA detection based on CGM rate of change, CGM rate of change increase rate, and the attenuation time window. Further, the PISA model incorporates activity monitoring to account for daytime exercise that could be misconstrued as PISA. Preliminary data was presented at ATTD – see page 45 of our ATTD full report at


David Klonoff, MD (Mills Peninsula Health Services, San Mateo, CA)

Dr. David Klonoff explored the mHealth landscape, with particular focus on current barriers to uptake. Chief among them is that patients, clinicians, regulators, payers and industry all have different concerns and priorities muddying mHealth’s path forward. For example, clinicians want mHealth to save them time, save them or the patients money, or improve outcomes. Meanwhile, patients want improved human factors, regulators are laser-focused on safety, payers want to save money, and industry wants to make money. However despite these hurdles, Dr. Klonoff envisions a future in which the artificial pancreas will provide remote cloud-based mHealth interventions (coined “KlOnStar” to capture the utility of the familiar OnStar car navigation system). The system would funnel sensor data through an enhanced 911 device to deliver alerts to physicians, designated “next of care,” or call centers (Dassau, Journal of DST 2009). In addition to improving safety, Dr. Klonoff believes that mHealth research can improve AP efficacy, potentially through new wearable sensors that can provide additional novel input to AP systems and improved human factors (which would increase adherence and therefore the benefit of the AP). To meet this end, Dr. Klonoff outlined four buckets of research that he considered “worthy” of NIH funding: 1) incorporation of data from multiple wearable sensors in addition to CGM to determine insulin dosing; 2) creation of remote safety systems to notify HCPs; 3) validated human factors tools to simplify necessary manual patient overrides; and 4) security for wireless data transmission and pump control. We thought Dr. Klonoff’s presentation was one of the more clear and crisp talks we’ve heard with respect to mHealth in the diabetes arena. We will be curious to learn whether his research buckets will receive the necessary investigator interest and research funding in the short term; certainly, if ever there was an audience to pitch these ideas to, it was the collection of expert thinkers in Lister Hill Auditorium.

Improving Insulin and Non-insulin Hormone Replacement


Howard Zisser, MD (Sansum Diabetes Research Institute, Santa Barbara, CA)

Dr. Howard Zisser discussed two approaches to improving closed loop control: inhalable Technosphere insulin (MannKind’s Afrezza) as an adjunct to closed-loop control and intraperitoneal insulin delivery. As to the former, Dr. Zisser explained that without precise insulin dosing, large carbohydrate meals (i.e., >50 g) can lead to hyperglycemic excursions and/or rebound hypoglycemia. As such, his team is investigating the use of Technosphere, which has an ultra-rapid on/off profile, to blunt postprandial glucose excursions without elevating the risk of hypoglycemia. In his team’s ongoing clinical trial, ~four of 12 patients have completed the study; the first patient’s postprandial glucose profile looked promising. Turning to intraperitoneal delivery, Dr. Zisser is collaborating with Dr. Eric Renard to compare closed loop control using Roche’s DiaPort 2 vs. a subcutaneous pump. Dr. Zisser expects to show data from the study’s 10-patient cohort later this year. For greater detail, see our discussion of Dr. Zisser’s Technosphere and DiaPort 2 presentations on page 23 and 25, respectively, of our ATTD full report:


Ron Pettis, PhD (Becton Dickinson, Franklin Lakes, NJ)

Dr. Ron Pettis reviewed the design, published studies, and upcoming research on BD’s microneedles for intradermal insulin delivery (1.5 mm long x 34 gauge). We look forward to seeing results from a 24- hour biomechanical device functionality study, which will be presented at EASD 2013. Meanwhile, the company recently completed a 72-hour basal-bolus insulin infusion study to examine device function and extended PK/PD variability/stability. The company has ongoing, unspecified work to optimize microneedle device design. Important next steps include defining the regulatory pathway (current analogs are not approved for intradermal delivery) and incorporation of microneedles in closed-loop trials.


Jeffrey Joseph, DO (Thomas Jefferson University Hospital, Philadelphia, PA)

Dr. Jeffrey Joseph explored how tissue trauma might contribute to CSII failure and variable subcutaneous insulin delivery. To set the stage for his discussion, Dr. Joseph explained that insertion causes initial trauma to the tissue as well as ongoing trauma due to micro- and macro- movements at the interface between the catheter and tissue. Catheter issues start to become clinically significant on day three (after which longer catheter use tends to correlate with worsening glucose control), necessitating insulin set change every two to three days. He proposed two histologically based hypotheses for the altered performance: 1) insulin could be degrading in the wound site around the catheter; and 2) insulin may be leaking out through the entrance wound. In order to optimize CSII catheter insulin performance, Dr. Joseph proposed myriad approaches: standardize catheter depth and location through automated insertion; minimize initial insertion trauma through a non-cutting pencil point needle or lubricious catheter/needle surface; minimize ongoing tissue trauma by using a soft, flexible, rounded catheter; increase capillary blood and lymphatic flow through heat or medication; inhibit blood coagulation and inflammation; inhibit insulin degradation with hyluronidase; and increase fibrinolysis (the break down of the fibrin that builds as a result of bleeding upon needle insertion). Dr. Joseph’s presentation and the round table discussion that followed drew attention both to the need for greater focus in this area and the potential to improve insulin absorption through catheter design.


Steve Prestrelski, PhD (CEO, Xeris Pharmaceutics, Austin, TX)

  • Dr. Steve Prestrelski reviewed Xeris’ efforts to develop a stabilized liquid glucagon. The company has three products in in the pipeline: a glucagon rescue pen for the treatment of severe hypoglycemia (G- Pen), a glucagon mini-dosing pen for the treatment of moderate hypoglycemia, and a formulation for use with a bi-hormonal pump. As to the latter, Xeris is targeting a product that is pump agnostic, has a two year expiration date at room temperature, stability at 37C (98.6F) for up to four weeks, and is marketed in 1 ml vials (5 mg/ml in vial). According to Xeris’ Brett Newswanger’s ATTD presentation, the company was expecting IDE approval of the pump solution by 2017-2018. The “key issue” to a bi- hormonal pump agnostic formulation, explained Dr. Prestrelski, was ensuring compatibility between DMSO (the solvent Xeris uses) and pump parts/infusion sets. For example, most infusion sets used in the US use polycarbonate, which is incompatible with DMSO; however, Xeris believes incompatibility is a solvable problem and noted its collaboration with multiple pump companies to make sure all pump parts could be used with a DMSO-based solution. Turning to Xeris’ lead G-Pen product timeline, Xeris expects to begin its phase 2 trial in 3Q13 (a slight delay from the company’s previous 1Q13 target). Dr. Prestrelski’s talk nicely complimented Mr. Newswanger’s presentation on the topic at ATTD – for greater detail on Xeris and its ongoing development efforts see our ATTD discussion on page 109 of our full report:


Claudio Cobelli, PhD (University of Padova, Italy); Sanjoy Dutta, PhD (JDRF, New York, NY); Bruce Frank, PhD (Thermalin, Cleveland, OH); David Klonoff, MD (Mills Peninsula

Health Services, San Mateo, CA); Boris Kovatchev, PhD (University of Virginia, Charlottesville, VA); Allan Krasner, Biodel, Danbury, CT); Sean Saint (Tandem Diabetes Care, San Diego, CA); Bill Tamborlane, MD (Yale University, New Haven, CT)

Dr. Kovatchev: If you had to pick up a second hormone for dual hormone closed loop, would you pick up glucagon or pramlintide?

Dr. Stuart Weinzimer: It’s easier to pick a hormone where you can have a co-formulation or co-delivery potential. That’s a benefit and part of the attractiveness of pramlintide. You are also mimicking natural physiology a little better. The idea is that it’s just like the native beta cell. However, it’s a bit reductionist to pick a hormone like it's a game show.

Dr. Tamborlane: I agree with Stu. I’m not sure it’s one or the other. We can achieve the same effects of pramlintide with a GLP-1. Once a day liraglutide is something we’re very interested in. We’re surprised Novo Nordisk up until now has not been looking into that, but they are now. Or a Bydureon once a month injection, along with glucagon as an adjunctive therapy. We’re approaching things from two ways: glucagon to rescue from a low, GLP-1 to prevent highs that may contribute to lows.

Dr. Aaron Kowalski: I have been obsessed with infusion sets over the last few years. It stems from a transition I made from MDI to pumps – I did that on CGM. I posit that the reason we don’t see a bigger separation between MDI and CSII is because of pump set failure. I commend what you’re doing because there is such a huge problem. JDRF is funding BD on this front. But I would challenge the community that we’ve got to get insulin into people, and we’ve got to do it more consistently.

Dr. Jeffrey Joseph: Thank you Aaron. One thing I got out of a thorough review of the literature is that there’s never been a systematic study to understand what happens to an infusion set after days three, four, and five. We need systematic studies.

Dr. Bruce Buckingham: We’ve studied 120 weeks of seven days of insulin infusion with one set – 40% of people can go seven days with no deterioration in glucose control. Some people can only go two or three days. There is a huge biologic factor that is not well understood, whether it’s fat cells, cytokines, histology, etc.

Day Two

Future Devices and Other Innovative Approaches to Optimize a Closed-loop System


Tom Peyser, PhD (Dexcom, San Diego, CA)

Dr. Tom Peyser discussed Dexcom’s broad pipeline of CGM innovation: 1) a review of previously shown G4 Platinum and G4 AP version data; 2) a brief discussion of the G5 smart transmitter (available this fall to closed-loop researchers under an IDE); and 3) pump partnerships with Tandem and Animas. Particularly valuable was his detailed discussion of redundancy; Dr. Peyser provided a competitive counterpoint to Medtronic’s redundant sensing approaches. We also appreciated his review of key CGM performance characteristics for the artificial pancreas. Now that the technology has come so far, we think it will be valuable for researchers, industry, and patients to hash out what level of system accuracy and reliability is good enough.

  • Dr. Peyser discussed Dexcom’s view of key CGM performance characteristics for the artificial pancreas: 1) accurate and reliable; 2) easy to use; 3) connectivity to AP controllers; 4) cost-effective and large-scale manufacturability; and 5) self-monitoring to assess signal fidelity. Dr. Peyser believes that the required level of CGM performance for the artificial pancreas should be seriously discussed within the closed-loop community. He proposed a MARD <10%, Clarke A- Zone points >85%, an occurrence of egregious errors <1% of the time, no pharmacological interferences, and an ability to prevent or identify fault conditions.
  • Closed-loop researchers have particularly appreciated the G4 Platinum’s improvements in signal transmission and reliability. Dr. Peyser specifically mentioned the system’s improved transmitter-receiver range (20 foot minimum) and no loss of transmission when the receiver is connected to a laptop or smartphone. He candidly told a story from the Seven Plus days, where AP researchers would call him in the middle of the night with signal transmission problems. Often, they would have to hold the receiver within six inches of the transmitter to get data.
  • Notably, the JDRF-funded version of the Gen 5 smart transmitter (ANT+ to a smartphone) will be available to closed-loop researchers this fall under an IDE. Dr. Boris Kovatchev specifically mentioned on day one of the workshop that the G5 will be used in two UVa Diabetes Assistant studies in 2013-2014: a multi-center outpatient trial of closed control (AP@home) and a multi-center home trial of closed-loop control (Jaeb Center and JDRF). As of Dexcom’s JP Morgan 2013 presentation, commercial launch of the Gen 5 was slated for 2014- 2015.
  • Dr. Peyser extensively addressed CGM redundancy, noting that it is “not a panacea” and must be put into a larger system context. He showed a humorous picture of many kids on a field aimlessly surrounding a soccer ball: high redundancy (many kids playing soccer) though low accuracy and reliability. He comically compared this to an Italian soccer player, who has low redundancy (just one person playing) but high reliability and accuracy. The same applies in a basketball analogy – the quest for redundancy does not mean it’s worth replacing a 60% shooter with two 30% shooters.
    • Applying this to the G4 Platinum, Dr. Peyser’s emphasized that the benefits of redundancy are significantly reduced with a more accurate sensor. He mentioned the oft-cited study from Drs. Jessica Castle and Ken Ward, which found accuracy benefits to using multiple redundant Dexcom Seven Plus sensors. Since the Seven Plus had lots of random error, such redundancy was an effective risk mitigation strategy. However, Drs. Castle and Ward repeated the study with the G4 Platinum and found a much lower benefit to using multiple sensors.
    • Many of the current sensor failure mechanisms are common to all sensors. These include improper SMBG calibration (one of biggest sources of error in Dexcom studies), wound response post-insertion, pressure induced sensor attenuation, data transmission problems, unexpected pharma interactions, and age-related bio-fouling.
    • Dr. Peyser argued that a system level redundancy strategy is a more effective approach to safety mitigation. First, intelligent CGM redundancy works to prevent severe adverse events by taking into account insulin dose history; the goal is to avoid hyperinsulinemia. System level redundancy also includes a physiological component – since CGM can reverse hypoglycemia unawareness, it could allow patients to detect an errant AP system on their own.
    • Sensor performance can be enhanced by self-monitoring of the sensor signal. This has been an integral part of the design of all Dexcom sensors. Algorithms can verify signal integrity, identify anomalies, and check for non-physiological rates of change. Ultimately, they can decide whether data is displayed or not (appearing as “???” on the Dexcom receiver). Dr. Peyser believes a sensor reliability metric would be key to provide to AP controller algorithms, as it would allow controllers to dose insulin more or less aggressively.
    • Dr. Peyser’s critical discussion of redundancy was a counterpoint to Medtronic’s enthusiasm. Dexcom is taking the view – perhaps rightly so – that since the G4 Platinum is more accurate and reliable, redundancy through multiple sensors is not needed. On the other hand, Medtronic has taken the view that using more than one sensor provides additional accuracy and reliability benefits, especially when the sensors are orthogonally redundant (glucose oxidase and optical). This is the first time we can recall seeing this dichotomy at a conference, and as CGMs begin driving insulin dosing, we think this issue is sure to come up again in the future.


John Mastrototaro, PhD (Medtronic, Northridge, CA)

Dr. John Mastrototaro provided a look at Medtronic’s stepwise plan to bring an artificial pancreas to market: 1) the Paradigm Veo/MiniMed 530G (no updates on the FDA review timeline); 2) the MiniMed 640G predictive low glucose management pump (expected to launch this year in Europe); 3) overnight closed loop and treat to range; and 4) fully closed loop. Dr. Mastrototaro emphasized that Medtronic is taking a cautious and risk mitigation approach to stages three and four, since these have higher risk – indeed, given Medtronic’s scale in the US alone (~300,000 patients on pumps = 100 million nights per year), even a small risk of an adverse event is still large in an absolute sense. The company is currently testing several fail-safe approaches and new algorithms in studies in Australia (and coming up in the UK).

  • Dr. Mastrototaro emphasized that closed-loop components must be treated as a system, especially in the context of commercialization. Although many researchers talk about seamlessly interchanging components, Dr. Mastrototaro argued that it’s not quite that easy. He explained that different pumps have different occlusion detection alarms, meaning that algorithms need to individually understand the unique parameters for each pump. On the sensor side, he used Dexcom’s interference with Tylenol as an example of limitations unique to certain CGMs (we note that data showed on the first day of the Workshop suggested this interference is not an issue with the G4 Platinum like it was for the Seven Plus). Dr. Mastrototaro’s system level discussion even mentioned the specific blood glucose meter used to calibrate a closed-loop system.
  • Low glucose suspend “is an improvement, but it’s not perfect. It won’t remove all hypoglycemia.” Dr. Mastrototaro described three basic failures of LGS: 1) times when the sensor is reading high and a low alarm is never triggered in the first place; 2) the sensor reads too low and erroneously suspends insulin, even though the blood glucose is actually higher (according to data from Dr. Bill Tamborlane, DKA is not a risk in these cases); and 3) suspension occurs properly, but the patient still goes low (e.g., a patient takes far more bolus insulin than is needed, and suspending the pump will not stop hypoglycemia).
  • Medtronic’s predictive low glucose management pump, the MiniMed 640G, will “likely be launched this year in Europe.” This was less specific timing than remarks in the company’s F3Q13 call, where EU launch was expected this summer. We’ll be interested to hear in the company’s May 21 F4Q13 call whether a summer launch is still on track.
    • Dr. Mastrototaro showed data from 22 experiments of the MiniMed 640G, where 14 hypoglycemia events were prevented. The mean glucose nadir was 79 mg/dl and the mean suspension time was 92 minutes. He did not give further study design details or any sort of comparison to Veo or open-loop therapy. The last data we heard on the MiniMed 640G came at EASD 2012 – a computer simulation of the new pump algorithm reduced the number of hypoglycemia events (<70 mg/dl) by 18% and the average duration of hypoglycemia by 50%, a significant improvement over the Veo’s corresponding reductions of 1% and 28% (see page three of our EASD Diabetes Technology report at In our view, predictive low glucose management would be very beneficial for patients and represents a very achievable step near-term.
  • Dr. Mastrototaro believes that the biggest AP safety concern is a large CGM error sustained for a long period of time. By contrast, he does not view transient sensor compression artifacts as a big worry, since the pump would suspend insulin and then resume quickly (i.e., assuming that the sensor artifact results in a transient dip in blood glucose). This represents a ~20-30 mg/dl rise in blood glucose, which would subsequently be corrected by the system.
  • Dr. Mastrototaro believes overnight closed loop is the next logical step after low glucose suspend and predictive suspension systems. Since patients are not dealing with meals, overnight insulin delivery is only provided by the basal rate. This means much less aggressive insulin delivery (“one daytime meal equates with the total insulin you would give in the whole overnight period”) and a greater opportunity to mitigate the risk of sensor inaccuracy. Dr. Mastrototaro was also positive on the potential of treat to range to limit postprandial peaks.
  • Commercializing an overnight closed-loop system is “a bit tricky,” with “lots of concerns from a legal perspective and risks of liability.” He explained that assuming Medtronic has approximately 300,000 people on insulin pumps (as an aside, based on 450,000 US pumpers per Halozyme’s most recent estimates, this translates to ~67% market share by volume for Medtronic, in line with the company’s market share by sales in our most recent model). That many patients on pumps translates to over 100 million closed-loop nights per year. Even if sensors have serious issues on one in one thousand nights, that’s still over 100,000 nights per year that Medtronic pumpers could be in trouble on an improperly designed overnight system. Put differently, based on STAR-3 data of 13 severe hypoglycemia events per 100 patient years, which is the lowest rate generated thus far, that’s still 40,000 severe hypoglycemia events for US Medtronic customers each year.
    • Said Dr. Mastrototaro, “When we’re looking at commercializing [an overnight] system, we’re adding all sorts of fail safe mechanisms”. These will include sensor checking algorithms and mitigations against calibration errors. Medtronic’s clinical trials will assess fake sensor and meter values to make sure the system is safe. The company has started one trial in Australia using the new system with fail-safes, and another is starting in the UK within a month. The basic idea is to have five different algorithms running in the background.


Rajiv Shah, MS (Medtronic, Northridge, CA)

Mr. Rajiv Shah’s presentation focused on Medtronic’s use of sensor redundancy and electrochemical impedance to improve accuracy. Data from a flat ribbon probe with electrodes on both sides showed that sensor readings can differ from one side of the sensor to the next, suggesting that localized cellular environments affect accuracy. As such, Medtronic is investigating the use of glucose-independent electrochemical impedance spectroscopy to determine when a sensor starts to fail. If impedance values fall outside of an “ideal” or predicted impedance band, information from the sensor can be de- emphasized. Thus, by fusing data from redundant sensors and using electrochemical impedance information to de-emphasize readings from failing sensors, Medtronic hopes to avoid sensor inaccuracy caused by localized effects. In other pipeline news, Mr. Shah noted that Medtronic was in the midst of finishing its clinical evaluation of its integrated sensor/infusion set in support of regulatory submission. Mr. Shah’s ATTD presentation gives additional detail on Medtronic’s pipeline products – see page 57 of our ATTD full report for detail:


Kerstin Rebrin, MD, PhD (BD Medical, Franklin Lakes, NJ)

Dr. Kerstin Rebrin provided updates on BD’s proprietary glucose binding protein (GBP)- based optical sensor in development. The system is currently in its second generation and Dr. Rebrin noted that the most immediate challenge going forward was to further miniaturize the optical device. Diving into the technical detail, Dr. Rebrin explained that the protein folds when glucose binds to it, resulting in the quenching of fluorescence. BD measures two wavelengths of light – blue and green – and uses the ratio of these wavelengths to reduce signal to noise ratio. Next, BD applies a non-linear calibration curve and time delay correction. (The system has a time delay of ~five minutes.) Dr. Rebrin was encouraged by initial simulation data and the sensors overall profile – she noted a good in vivo interference profile, reduced insertion trauma, consistent placement, signal stability over time, fast warm up time (<20 min), and favorable wear comfort. As of ATTD, the system was being tested in two clinical studies: a 24- hour study through UVA/JDRF in 12 patients and a 29-patient, three-day study in Canada. For greater detail, see page 36 of our full report


Kenneth Ward, MD (Legacy Research Institute, Portland, OR)

Before discussing his ongoing work to integrate sensing technology into a catheter, Dr. Kenneth Ward explored whether insulin could indeed be delivered in the same microenvironment in which glucose is sensed. Turning to the literature, a microdialysis/microperfusion study showed that when there was no physical separation between glucose sensing and insulin delivery, the degree of local glucose reduction was fixed (~15-20%) and irrespective of insulin rate (Lindpointer et al., Diabetes Care 2010). A study of subcutaneous insulin delivery in a swine model that examined sensor data near and far from the site of delivery led the authors to conclude that the test to test variation between sensors substantially exceeded the effect of the distance between glucose sensing and insulin infusion (Rodriguez et al., Diabetes Technol Ther 2011). Said Dr. Ward, subcutaneous insulin seems to have less of a localized effect than one would predict. He explained that one of two competing fates befall injected insulin: 1) the insulin acts (and is degraded) at the fat cell; or 2) insulin is absorbed into the blood stream. Due to rich adipose blood flow rate, negative cooperativity (whereby very high local concentrations of insulin inhibit insulin action), and the comparatively inefficient insulin uptake by adipose tissue (vs. muscle uptake) Dr. Ward argued that absorption is the primary fate of subcutaneously injected insulin. As such, Dr. Ward believes that an integrated glucose sensor and catheter is feasible. He is working to develop such a device through collaboration between his company Pacific Diabetes Technologies and Oregon State University. The idea is to use a flex circuit to dispose glucose sensors on a flat surface that can be wrapped into a catheter (i.e., each catheter is encased by a glucose sensor). While the system is early stage (pre-animal testing), we can’t help but be excited by its potential to reduce patients’ care burden by decreasing the number of devices worn.


Natalie Wisniewski, PhD (Chief Scientific Officer, Profusa; Medical Device Consultancy, San Francisco, CA)

Dr. Natalie Wisniewski gave an overview of the pros and cons of non-glucose oxidase-based glucose sensing technologies. Her presentation focused on fluorescence and had a clear takeaway: there is a bright future, in her view, for fluorescence-based sensing chemistries alone or in combination with glucose oxidase. She highlighted two significant benefits of fluorescence: a lack of glucose and a lack of oxygen consumption. Dr. Wisniewski was positive on Senseonics’ recent implantable data (MARD of 12% over several months), and wondered what the outlier values were. She also highlighted the “really encouraging” chemistry of BD’s glucose binding protein CGM, as its accuracy may be better in the hypoglycemic range. She did not give any sort of timeline or prediction on who would come to market first. She covered a few other non-glucose-oxidase technologies (direct electro-oxidation, non-invasive, holographic crystal, osmotic pressure sensing, MEMs), highlighting that all are early stage and “not ready for prime time.”


Joe Lucisano, PhD (President and CEO, GlySens, Inc)

Dr. Joe Lucisano discussed GlySens’ efforts to develop an implantable continuous glucose sensor. The GlySens sensor uses a dual detection method, whereby the system monitors the enzymatic reactions of both glucose oxidase and catalase (which breaks down hydrogen peroxide, a byproduct of the glucose oxidase reaction). Dr. Lucisano reviewed results from the human feasibility trial of the first-generation device; data was first presented at the Diabetes Technology Meeting (see our discussion on page 29 at Data were presented: (1) showing operation of the fully implanted sensor for month-long periods without need for recalibration and (2) affirming patient acceptance of the device. The feasibility study also revealed that the sensor’s data transmission to the external receiver did not perform at the maximum desired distances (data transmission was often disrupted at night when the receiver was on a bedside table) and that, in some cases, the tissue pO2 would drift outside the operational range and preclude glucose readings (background oxygen levels are used to calculate glucose readings); however, Dr. Lucisano commented that these two issues have been corrected for in GlySens’ second-generation design. The company expects to begin clinical testing for its second-gen system this year.


Dan Burnett, MD (Theranova, San Francisco, CA)

Dr. Dan Burnett believes that sensor failures are a byproduct of the space in which they rest as opposed to the sensor technology itself. As such, Dr. Burnett is exploring the viability of the peritoneal cavity as a sensing site. Unlike the subcutaneous tissue, he explained, the peritoneal cavity has a constant core temperature and is physically protected from external mechanical stress. Further, it has minimal tissue response to biocompatible materials, a reliable blood supply, and proven history of safe use for other medical devices. Dr. Burnett described his team’s initial work in juvenile non-diabetic pig models comparing sensor lag time during subcutaneous vs. peritoneal glucose sensing. Indeed, intraperitoneal latency was shorter than subcutaneous latency with respect to time to half maximal values (p <0.001). Dr. Burnett also recently presented the data at the 13th Annual Rachmiel Levine Diabetes and Obesity Symposium – see page 49 of our report for detail:


Tom Peyser, PhD (Dexcom, San Diego, CA); John Mastrototaro, PhD (Medtronic, Northridge, CA); Arleen Pinkos (FDA, Silverspring, MD); Barry Ginsberg, MD, PhD (Diabetes Technology Consultants, Wyckoff, NJ); Aaron Kowalski, PhD (JDRF, New York, NY)

Dr. Ginsberg: Some sensors are more accurate than others. Is there a way to pick them out post-manufacturing?

Dr. Peyser: We actually would like to pick out only the most accurate sensors and sell them. We check performance of every sensor we manufacture in vitro. But we haven’t found a good way to predict which of those will be good in the body. Our approach is to try to reduce variance in manufacturing. There was a lot of variance on the Seven Plus. That’s been a big focus of ours with the scale up. The concept of the MARD distribution is to highlight that at a MARD of 10-12%, or even at 14-15%, there are some sensors on the left of the distribution – these may be more accurate than BGM. There are also sensors on the right side, which can cause problems for patients. We are exploring ways to identify good performing sensors, but also addressing manufacturing.

Dr. Ginsberg: There was a large study in 1993 that examined the accuracy of BGM in the hands of patients. The average error was 13%, and those were meters.

Dr. Kowalski: Tom’s presentation highlighted that G4 AP is a software improvement. According to the data, it conveys significant improvements in terms of accuracy. Is there a way at FDA to incorporate software-based improvements into industry systems faster? Isn’t there a faster way than filing PMA supplements over and over again?

Ms. Pinkos: FDA has accepted software modifications with very little clinical data, but it depends on the nature. In the case of Tom, I thought it was a membrane change?

Dr. Peyser: The membrane change was from the Seven Plus to G4 Platinum. G4AP is just a software improvement. We are looking forward to discussions with you and others for ways of expediting that.

From my perspective, Aaron, the changes are in the calibration algorithm. We would expect to have to do a clinical study. Maybe not as much of a clinical study as we did for the G4 Platinum originally. That would be up to discussion with the Agency.

Ms. Pinkos: We often hear someone tweaking their algorithm for improvements. You might need a bridging study to show what kind of effect it had on performance.

Dr. Mastrototaro: If you are going to modify the labeling with improved accuracy, you need clinical data and a PMA supplement. It’s the same thing – you’re improving sensor performance and want to have data to back it up and modify labeling.

Ms. Pinkos: Another option includes taking old data and running it through the algorithm retrospectively.

Q: You mentioned hypo events at a rate of 13 events per 100 patient years. What is the bare minimum target for sensor accuracy and serious adverse events? A single event can be catastrophic.

Dr. Mastrototaro: 13 events per 100 patient years came from our STAR-3 trial using sensor augmented pump therapy. The DCCT and other trials are many times higher than that. In sensor performance for the closed loop, we’re looking at it differently than MARDs. You could do closed-loop with 20% error if you consistently had that error all the time. The issue is big mg/dl errors for sustained periods of time. What we focus on, and we have tens of millions of days of sensor data, is periods when the sensor is grossly off for extended periods of time. That can be a catastrophic failure. We try to figure out what aspects of those events we can mitigate. So we’re mitigating calibration errors, for instance. If you can identify when the CGM is inaccurate, then you just won’t be in closed loop at that time. The big issue for us, however, is how many times you are going to have the big errors sustained for a long period of time.

Dr. Gary Steil (Children’s Hospital, Boston, MA): We’re starting a closed-loop study where we take YSI glucose values and randomly throw in the types of errors John just talked about. The target is 120 mg/dl. The YSI error will be 30 mg/dl, which pushes it down to 90 or 150 mg/dl. The intention is to do closed loop through nighttime and breakfast. If YSI thinks you’re at 90 and you’re really at 120, that’s a sustained 24-hour error. All glucose errors are not created equally. Most of the errors that people are incredibly concerned about need to be taken into account by robust algorithms.

Dr. Mastrototaro: Another example is the sensor gets pulled out of the body or the sensor has died or failed. That’s different from the sensor just reading low.

Elements for a New Generation of Devices/Technologies


Joseph Cafazzo, PhD (Toronto General Hospital and the University of Toronto, Canada)

Dr. Joseph Cafazzo gave a nice update on the work towards standardizing BGM, CGM, and insulin pump communication, a desperate need cited throughout the Workshop. After making a strong case for the value of standards, Dr. Cafazzo discussed the work of Continua, a consortium of 200 member companies that includes a handful of diabetes players: Bayer, Dexcom, Medtronic, Sanofi, JDRF, and Roche. Continua calls for adoption of the IEEE standards, Bluetooth and Bluetooth low energy, and USB. The blood glucose meter standard has already been published and is in its second revision. Meanwhile, the insulin pump standard has been drafted and is in balloting, while the CGM standard is in draft form. These standards are released to manufacturers on an annual basis. Looking forward, Continua hopes to develop a standard for dual-chambered insulin pumps. We wonder whether companies could be incentivized to join Continua, perhaps by either FDA or JDRF. Overall, we love the focus on device standardization and believe it has massive potential to help the entire field, especially AP research.

  • Dr. Cafazzo made a strong case for the benefits of standardization for industry, patients, and regulators. Arbitrary data formats and radio transport systems result in redundant, wasteful efforts in building individual interfaces for individual products (e.g., artificial pancreas systems). Dr. Cafazzo believes standards can encourage new entrants (researchers and companies), help create an innovation ecosystem, and build a competitive advantage. From a regulatory perspective, standards mean less variability in products considered by FDA. This also means more transparency, industry consensus, and an ability to address security issues. Dr. Cafazzo also appealed to common sense, everyday reasons for standards – there are clear standards for planes, trains, automobiles, all forms of construction, Internet, and data communication. Without these, widespread innovation and use would not be possible.
  • The proposed standards are not intended to be performance requirements for the artificial pancreas. Dr. Cafazzo emphasized that they “are not like accuracy requirements for SMBG,” and they are not about post-market mixing and matching of components. The intent is for the standards to be used in pre-market research when companies are developing products.


Margaret Joan Taylor, PhD (De Montfort University, United Kingdom)

Dr. Margaret Taylor described her team’s effort to develop a biomaterial device for closed-loop insulin delivery. In contrast to conventional designs, the device utilizes no sensor or algorithm. Rather, glucose sensitive gel that is implanted into the peritoneal cavity governs insulin release (the insulin rests in a depot within the gel). The gel contains polymerized ConA and dextran, which are competitively displaced in the presence of glucose. This causes the gel to form a sol, thereby increasing the material’s permeability and releasing insulin directly to the liver. The gel is housed in a pebble-shaped device with ports that allow for ex-vivo refilling. Dr. Taylor described her teams’ investigation of the device in pigs; the study was first presented at ATTD (see page 30 at While the study started with 10 pigs, only one pig was used in the analysis (pigs were lost for myriad reasons, including anesthesia-related complications, under dosing of streptozotocin [used to induce diabetes], and peritonitis). However, Dr. Taylor was optimistic about the pig’s glucose profile with the implanted device and she encouraged the audience to keep their minds open to solutions that were neither biological nor electronic in nature.


Cheryl Stabler, PhD (University of Miami, Miami, FL)

When it comes to islet cell transplants, said Dr. Cheryl Stabler, it’s all about “location, location, location.” Dr. Stabler explained that approximately 60% of islet cells die within the first week of transplant due to initial inflammatory responses and that the remaining islets often experience metabolic burnout and a smoldering immune response overtime. The high inflammatory response of the subcutaneous tissue has motivated Dr. Stabler and her team to investigate alternative sites. She believes that the omental pouch is a promising site because it has portal drainage, is richly vascularized, can be accessed minimally invasively, and has immune suppressive qualities. Next, she proposed that silicon scaffolds could further optimize the islet cell microenvironment by providing physical protection (i.e., the islets would be housed within the scaffold). Potentially, scaffolds could also enable localized drug delivery, in situ oxygen generation, and co-delivery of helper cells, all of which could improve an islets’ cell survival chance. Microencapsulation of the cells could also help improve survival by camouflaging the islets from the immune system – her team is currently exploring a microfluidic technique to create a thin conformal coating. Certainly, there are myriad challenges to overcome for long-term successful islet cell transplantation; however, we are encouraged to see Dr. Stabler systematically addressing the issues that have come to light with current techniques.


Robert Farra (MicroCHIPS, Waltham, MA)

MicroCHIPS is manufacturing an implantable microchip that houses multiple reservoirs for drug or sensor storage; the device is implanted just below the subcutaneous tissue. The device provides long- term protection for the reservoirs’ contents. It can release the drug or expose the sensor on demand via an electronic signal such that medicine can be delivered remotely. Dr. Robert Farra envisions that the technology could enable longer-term glucose sensing or redundant glucose sensing (i.e., multiple sensors could be exposed and when those sensors start to fail, electronic signals could be used to open new reservoirs and thus, expose new sensors). Additionally, Dr. Farra believes that the device could be a good delivery technique for glucagon (or could even house both glucagon and glucose sensors). The company’s next-generation device will have 200 reservoirs (vs. 20 in the previous gen) that are 1.0 mg in size (vs. 0.5 mg). We look forward to following the company’s pursuit of potential diabetes applications.


Eric Schreiter, PhD (Janelia Farm Research Campus, Howard Hughes Medical Institute)

Dr. Eric Schreiter gave a basic-science-focused talk on engineering a glucose responsive fluorescent sensor protein. The basic idea is to express this glucose-responsive fluorescent protein in a group of cells by safely delivering it through a gene transfer or autologous cell transfer (e.g., take a skin biopsy, get the sensor in, and then re-implant the cells). Ultimately, the researchers plan to build a wearable spectroscopic device to measure the fluorescence. Dr. Schreiter and colleagues recently built a thermus thermophilus glucose/galactose binding protein (TtGBP) sensor and optimized it for the physiologic glucose range. It undergoes a ~300% fluorescence increase from hypoglycemia to hyperglycemia levels of glucose. Dr. Schreiter’s rushed concluding slide indicated that the sensor still needs improvement on a few fronts. Still, despite the very early stage of this approach, it was quite interesting and we hope to hear more.

Roundtable Discussion: Perspectives for the Future


Kenneth Ward, MD (Legacy Research Institute, Portland, OR); Barry Ginsberg, MD, PhD (Diabetes Technology Consultants, Wyckoff, NJ); Todd Zion, PhD (Cofounder, SmartCells); Natalie Wisniewski, PhD (Medical Device Consultancy, San Francisco, CA); Howard Zisser, MD (Sansum Diabetes Research Institute, Santa Barbara, CA); Aaron Kowalski, PhD (JDRF, New York, NY); Courtney Lias, PhD (FDA, Silver Spring, MD); Guillermo Arreaza, MD (NIH, Bethesda, MD); Rich Bergenstal, MD (Park Nicollet Health Services, Minneapolis, MN); and Robert Vigersky, MD (Walter Reed National Military Medical Center, Washington, DC)

Dr. Bergenstal: What are the hopes, expectations, and needs in the short term to keep this momentum going? What are we going to expect and hold each other accountable to?

Dr. Kowalski: I am going to throw down the gauntlet. I started out the meeting talking about the need for delivery to patients. I want to throw down the gauntlet to industry. We’re working well with industry, but I would like to see within the next 24 months every pump that is speaking to a sensor have low glucose suspend.

Dr. Bergenstal: What will that take?

Dr. Kowalski: We heard a lot of stretch goals here. As a simple safe thing to do, predictively turning off pumps could be transformative. Medtronic is on that path and hopefully the FDA is working with them in the States. Other companies need to make a commitment to this. This will be a major funding piece for JDRF this year and we could put it out to the industry. I hope that any company that sells a pump will make a firm commitment to at least PLGS systems in the very near term.

Dr. Lias: We’d like to see that happen too. What we get out of this type of interaction is everybody in the same room. We get a lot of information, can look at bigger goals, and everyone else can kind of hear what’s going on. I share your point of view that that would be a great thing. We will do what we can to continue that.

Dr. Zion: Anyone can cut down on hypoglycemia. As you get more aggressive, you get more and more hypoglycemia. It’s a laudable goal to turn off pumps, but we should look at whether you’re losing control. The bar needs to be set quite high. I’d like to see more talks look at other variable like MAGE and incidence of hypoglycemia.

Dr. Kowalski: I agree with you and I disagree. I had this talk at FDA. I grew up in a house where my brother had a lot of severe hypoglycemia. I saw a lot of ambulance trips and I have given a lot of glucagon injections. I personally think that the hypoglycemia/A1c tradeoff should be a decision that clinicians and patients make. My brother’s A1c is 6.8%. Would he be willing to sacrifice and go to a 7.3% and not have two hospital trips this year? That should not be an FDA decision. That’s a clinician decision. That tradeoff is in the guidance right now. If you set the bar too high, you’re tying the hands of the medical practitioner and patient.

Dr. Lias: I think that’s the point Arleen was making yesterday. We care about safety, but the real issue is whether there is enough information about the device so people can make informed decisions. Maybe they will be able to label the device to say that hypoglycemia is reduced, but that there is an A1c trade off.

Dr. Arreaza: That’s a kind of massive community work. We, as a government agency, cannot be doing any advocating work. We rely on other groups to lobby congress. Constituents of Congress members have tools to make the community aware of problems and to put pressure on their Congressman. The NIDDK remains committed to supporting basic and applied research and we are going to continue doing that independent of the resources we have available. Even if the Type 1 diabetes special program is not available, we will find a way to continue supporting research. We hope some of those questions on the board today will be answered by the end of this forum. For example, what are the biggest gaps in technologies? We are able to, in some way, identify gaps through internal processes, but we need help from the scientific community to put together more efficient programs and perspectives to lay a plan for the upcoming years. For example, the second question [“How much focus in future research is needed on understanding the biological interactions (e.g. device-tissue interface, immune response) vs. engineering elements (e.g. algorithms, better polymers)?”] – we are already funding, but if you can help us identify in a more precise fashion those key aspects to getting to this goal for a working wearable artificial pancreas, we want to continue supporting those efforts.

Dr. Wisniewski: I love Aaron that you threw down the gauntlet for an immediate artificial pancreas product. We need short-term practical goals. But remember members of the research community are also working on more advanced technologies that lay the groundwork for five to ten years from now. We need to make sure that we are appropriately funding the advance research that will get us to a “wear-and- forget” device that has wide patient and physician adoption. We need to support a strong research foundation that moves us beyond percutaneous needles that have to be replaced every week and that patients have to calibrate several times a day. From a Silicon Valley viewpoint, funding models of the past to support R&D to develop truly advanced, innovative products are dead. We (medical device start-up industry) very much rely on DARPA, NIH and foundations. We need to ensure that pathway is laid for long-term fully implantable AP. We should have the race in the short-term, but let’s not lose sight for the long-term.

Dr. Bergenstal: Do you have any advice for us on how to renew the Special Diabetes Program funding? It got funding for one year – how can we get that going for three years?

Dr. Ward: I was going to say that system integration is completely key to all of this. I heard several people say that the problem in the AP is not the algorithm or science or sensor or pump. It’s communication failures. We talk about going home with the AP. The key question is how many times does that doctor, tech, or engineer walk home and fix the system? How often do they have to call up the patient on the phone? We’re talking about reliability and simplicity. We must also reduce and combine elements. The dual chambered pump is a perfect example, as is integrating sensors into catheters.

Dr. Zisser: I think that will come. Currently, researchers are using off the shelf products. We are trying to build what we think in our minds will work. Anything we can do to get standards is good, though it’s a long process. Eventually, phones, computers, and the systems will be on the pump itself.

Dr. Ward: Do we wait for that time to release a product? Can’t we make it now?

Dr. Zisser: Making it and getting it released are two different things. A rate-limiting thing is power consumption. You’re communicating all the time. These systems aren’t designed with that capability. When I look back on where we were a couple years ago, we’ve done a lot of what we hoped we would do. One of my takeaways from this meeting was the variability of insulin over days and sensors over days. Anything we can do to decrease that is beneficial.

Dr. Vigersky: I recall someone mentioning the bell-shaped curve of IQ and we have that for physicians in their ability to understand systems and prescribe them appropriately. Do we need a full proof system or should we allow only certain people to subscribe them who have met certain standards?

Dr. Kowalski: Right now there is a bell-shaped curve and right now every one with type 1 diabetes has to manage their diabetes or they die, so they do it. What’s clear is that glycemic control is not great. It gets back to what Courtney was saying – we need to test these devices and define clearly what they can do, then provide tools to well-trained physicians, just like pumps and sensors are today. I come back to the comparator of where we are right now. We have a point of time in diabetes where we made a lot of progress, but there is still a ways to go.

Dr. Bergenstal: We’re talking about patients with diabetes and human factors. What’s simple and what can we use? There are some people with diabetes in the audience. I don’t want to put them on the spot, but perhaps I could ask Kelly if she might have a comment?

Ms. Kelly Close, MBA (Close Concerns/diaTribe, San Francisco, CA): We hope the field will keep in mind that incremental changes are immensely valuable as patients. Without the first backpack insulin pump (you all know that picture!), we wouldn’t have small pocket sized pumps now. Without the GlucoWatch, we wouldn’t have today’s amazing CGM. On the artificial pancreas, let’s also keep this perspective – the first product won’t be perfect and has no need to be. What we’re doing today as patients is SO far from perfect. It would be amazing to see excellent progress on technologies that make open loop therapy much better and less of a hassle. That could be smarter CGM, software that helps adjust insulin dosing, or technology that better incorporates into our daily lives.

Along with that technological progress, we absolutely hope we will have seen far more commercial progress in the coming years. We hope that companies and regulators both become less worried about liability. We all know that there are so many accidents that happen EVERY DAY with this crazy drug called insulin. There are so many accidents that happen with cars and driving, or even just people going to bed in their own homes every night. They aren't accidents that are attributed to FDA or to a company, but they are due to the difficulty of dosing insulin. Let's agree that every single accident won't disappear, but let's leave open the possibility that many of them will.

We hope we'll see more creative thinking about how to prove cost-effectiveness. The environment has never seen so much cost pressure. So many people want to see a modern day DCCT. We now have sensors that would make that happen. There are so many brilliant people here who know how to design trials and how to show benefit – we hope to see more collaboration like with the JDRF CGM trial that changed so many of our lives.

Imagine a world where the AP is available to everyone who needs it. A world where we wake up in the morning, our BG is reset. A world where we don't have to stop and think all the time. A world where we get a guaranteed A1c in target that will protect us from complications. Imagine the 'health dividend' that the AP will create – that would have an enormous long-term impact on healthcare costs. Imagine equally much change as we have seen since our last meeting. We look forward to watching our world, following it, writing about it, and to preparing to live a life that is more normal, healthy, and productive and predictable. [Applause]

Dr. Boris Kovatchev (University of Virginia, Charlottesville, VA): I want to take the next 45 minutes to say… just kidding [laughter]. The system approach should be the most important thing we focus on now. There were only three talks that mentioned it – John Mastrototaro’s talk, Joe Cafazzo’s talk, and probably mine. At the end of the day, the AP is a system. If we want to get there quickly, we have to do some things. I agree with Howard that the system is important, but I disagree that everything needs to be built into the pump. That’s only looking at one dimension of the pump – basic computing in a single device. It’s the horizontal version. The vertical version is distributed computing between these devices, an intermediate device (e.g., a smartphone), and the cloud. To get there, we need a base. One question is do we need a medical grade operating system. I think the answer is yes. I think we should get there ASAP. Do we need core testing facilities? I think the answer is yes as well. We are capable of establishing system testing facilities that may be independent of anybody and would use in silico validation of systems. This allows fast and rapid development with real components that are linked to a simulator. So there are two things that we need: the base and the testing facility for rapid evaluation of system functionality.

Dr. Arreaza: To establish the core testing facility, will it will be necessary to develop a consortium of scientists working together with industry to standardize this process?

Dr. Kovatchev: Correct yes, it’s cooperative work. That’s why we need to decide this here in this room.

Dr. Kowalski: I do think clinicians are a huge barrier. If you look at CGM adoption, CGMs are tremendously powerful tools. Most of people in this room are the few people that are actively prescribing them. We need tools to facilitate clinician adoption. I talk to the JDRF team about our goals – improve glycemic control and make patients lives easier by reducing burden. What will help clinicians and help facilitate adoption? We need data and tools such as the Ambulatory Glucose Profile. That’s an output to interpret information in a short office visit and give good recommendations. We aren’t going to have a fully automated, fully closed loop tomorrow. For now, there’s going to be this interaction and you are going to have to go to your doctor. We need to have the types of tools Rich is working on. We need to reduce the burden on the health care providers as well as the patients. Many doctors don't prescribe CGM to many patients (who could benefit) because they find it tedious, don't get reimbursed for their time, and feel they don't get much out of it.

Dr.: If I can provide an analogy – for osteoporosis, you must take an exam and get certified to read DEXA scans to get reimbursed. Perhaps JDRF could set up some sort of course or certifying exam.

Dr. Zisser: This is also done in fellowship programs.

Dr. Tessa Lebinger: What elements of artificial pancreas testing of artificial pancreas component testing could and should be standardized? We see a lot of data today comparing various sensors to blood glucose levels. What should the comparison be? Some people use arterialized venous blood, regular venous blood…venous blood is used to diagnose diabetes but fingersticks/capillary blood testing is used to manage diabetes. There are postprandial differences between these different matrices. What should be the standard comparison?

Dr. Zisser: The difference you are referring to in the postprandial phase is in this direction (gestures up)? You could have postprandial in this direction (gestures down)?

Dr. Lebinger: Yes, up.

Dr. Zisser: If the controller and sensor says 180 mg/dl vs. 210 mg/dl, I don’t really care so much. It’s when you get down around euglycemia, that difference is very important. With CGM we calibrate with fingersticks and we use venous YSI for reference. There’s always going to be noise in each one of those compartments.

Dr. Lebinger: Some are using arterialized venous blood for YSI and some are using venous blood for YSI.

Dr. Zisser: What is that difference?

Dr. Lebinger: I have papers with me I can show you after. Dr. Ward: Arterialized is a little higher.

Dr. Zisser: A little. There is no true glucose. If one says 105 mg/dl and one says 95 mg/dl, is that wrong?

Dr. Lebinger: It can be up to about 11% difference postprandially.

Dr. Kowalski: I always get caught up in this argument. I give Dexcom a lot of credit. Dexcom focused on hyperglycemia and hypoglycemia. I call this a clinical chemistry argument. Let’s not miss the forest from the trees. Sometimes we get so hung up on this accuracy question and individual system components. To make an analogy, if I put my cruise control on at 55 mph, am I going to be worried about the performance characteristics of my fuel injector? At this meeting, my concern is that people are getting hung up in things that may not be absolutely critical or that important. Errors may be completely washed out. In a clinical trial, we’re not asking how much insulin is coming out of the cannula. I would urge the community not to get so focused on the little pieces and miss the big picture.

Dr. Zisser: I think the discrepancy is important enough in the hypoglycemia region. But there is inconsistency in how CGM is compared to reference.

Dr. Kowalski: Don’t get me wrong – I totally agree that we do need standardized outputs for these trials.

Dr. Wisniewski: I appreciate your viewpoint Tessa. I think that’s a very focused question. I don’t have a definitive answer and it seems the science doesn’t…

Dr. Lebinger: I was saying that when you are evaluating your sensors, we should have a standard for reference.

Dr. Wisniewski: There is a big need for that. As we are developing new sensors, we want to be able to compare the next one to the next one. You want to make a fair comparison, but we need a level playing field – it really is a hindrance to development.

Dr. Moshe Phillip: At my clinic, I don’t care that the sensor is accurate. What I care about is patients.

Dr. Lebinger: It becomes a question when you’re sensing different areas, especially intraperitoneal sensing vs. interstitial fluid.

Dr. Ginsberg: If you have sensors measuring different areas of the body, they are different sensors. If you want to do head to head comparisons, you need to do third party testing.

Dr. Kovatchev: Testing and standardizing the outcome of testing of components is mostly done already. It’s done by many people in the room including, Barry and myself. We spent several months five years ago producing a document that is now the continuous sensing standard. It has seven different tests for how you test continuous glucose sensors and that is out there. Lets focus on what Dr. Moshe Phillip said.

Dr. Lias: I want to go back to the forest. We really do try to look at the point of whatever is in front of us. We will use a clinical endpoint if we think that’s appropriate. If it’s about accuracy, we would look at that. We are very interested in hearing if there are suggestions on standardization. We take an approach of fit for purpose.

Dr. Sanjoy Dutta: I want to add one more big tree into that forest. Can we discuss pivotal or definitive clinical trials for the artificial pancreas? How will it help the Agency make a decision? How would you design such a trial? What will you be looking for?

Dr. Lias: Our thoughts on AP trials are most eloquently stated in the guidance document. It has a lot of “if this, then that.” There are a lot of different devices. If you’re using CGM as an endpoint, maybe CGM accuracy is more important. There’s a lot of flexibility for intended use. That’s why this discussion is helpful to us – it helps define what is the device for and what is the appropriate trial designed for.

-- by Adam Brown, Kira Maker, and Kelly L. Close