11th Annual Diabetes Technology Meeting

October 27-29; Burlingame, CA Days # 2-3 Commentary - Draft

Executive Highlights

Days #2 and #3 of the 11th Annual Diabetes Technology Meeting were filled with absorbing views on the current and future status of diabetes technology, most notably the artificial pancreas. We heard from some of the field’s most highly regarded researchers (Drs. Roman Hovorka, Bruce Buckingham, Howard Zisser, Boris Kovatchev, and Lutz Heinemann, among others) on where the field is, what’s holding it back (faster insulin, CGM accuracy, and fidelity of data transfer between devices), and where it needs to move (greater mobility, remote monitoring, and integrated communication). Notably, we saw demonstrations of two mobile-phone-based artificial pancreas systems: one from researchers at the University of Virginia and the other from Medtronic (for images of both systems, please visit http://bit.ly/uyxD2z and http://bit.ly/t6NUHT). It’s refreshing to see so much progress and, in particular, such intense development of mobile outpatient systems – clearly major questions are coming to the fore – not so much whether outpatient trials can be done on a technical level, but how well the systems work, how easy they are to use (or will be), how regulators will assess the risks, and how payors will perceive the benefits.

On the regulatory note, day #2 marked the announcement of the investigational device exemption (IDE) for the ASPIRE study, an in-home trial of the Medtronic Veo, using the more accurate Enlite sensor. Lots of credit was given to FDA’s Dr. Chip Zimliki, chair of the Artificial Pancreas Critical Path Initiative and winner of this year’s Diabetes Technology Society Artificial Pancreas Award. We appreciated hearing his and Dr. Courtney Lias’ FDA perspectives, especially the stimulating panel discussions that followed their presentations. They made it clear that the FDA has a tough job (~50% of the recalls in Dr. Lias’ division in CDRH are due to software errors) and will continue to set high demands for researchers and companies. However, both speakers also emphasized that the FDA needs help and input from experts in the blood glucose meter, CGM, and artificial pancreas fields. Luckily, many such specialists seemed to be in the crowd, as an audience response question reported that 50% of respondents were an “Engineer or Computer Scientist.”

Finally, we relished listening to (and hearing people’s thoughts on) the latest updates on other technologies, including C8 MediSensors’ non-invasive CGM (some skepticism regarding accuracy and real-world usability), implantable CGM (more activity that we’ve seen in the past, while still a lot to prove), Bayer’s next-generation blood glucose meter (over 98% within ±10%/10 mg/dl), and an intriguing cellular-based technology from Wireless Medical (headed by the inventor of the USB Flash Drive, Dov Moran). Overall, we’re glad to see research moving in the direction of greater device integration, better accuracy, wireless/cellular communication, data collection, and miniaturization. All of these advances seem sure to benefit patients down the road – as usual, the big questions are how long will it take, how easy will it be to prescribe, teach, and use, and who will pay for it.


Table of Contents 


Detailed Discussion and Commentary

Live Demonstrations


Patrick Keith-Hynes, PhD (University of Virginia, Charlottesville, VA)

In a very exciting live demonstration, Dr. Patrick Keith-Hynes from the University of Virginia showcased the team’s newest closed-loop development: a mobile-phone-based artificial pancreas system. The control algorithm runs on an Android cell phone and can communicate wirelessly with an insulin pump and CGM sensor (the recently-started study in Europe uses a Dexcom Seven Plus and an Insulet OmniPod pump, with other manufacturers planned for the future). Much like the previous laptop-based Artificial Pancreas System (APS), this platform is designed to simplify the process of conducting a closed-loop trial. However, this system has the added benefit of portability, remote monitoring, and a user-friendly design, all crucial aspects for the outpatient trials where this system will be deployed. The team has also stripped the browser and telephone application out of the phone, only allowing a one-way outbound connection to a remote server. As we understand it, this makes things much easier from a regulatory perspective. We went up after the presentation and were able to view and play with the device – it’s very cool, well- designed, and we were happy to see Dr. Chip Zimliki, the head of the Artificial Pancreas Critical Path Initiative at FDA, eagerly demoing the system. To view screenshots of the system, please visit: http://bit.ly/uyxD2z. Thus far, two patients have used the Diabetes Assistant in outpatient closed-loop trials, one in Montpellier (Eric Renard) and one in Padova (Claudio Cobelli). For more information on this study specifically, see our coverage of Dr. Boris Kovatchev’s presentation.

  • The diabetes assistant is a portable artificial pancreas system that runs on an android-capable cellular phone. The system can wirelessly communicate with an insulin pump and CGM (at this time, only with Insulet’s OmniPod and Dexcom’s Seven Plus, although in the future other manufacturers will be added), and can operate multiple control algorithms (e.g., bi-hormonal system). It has a touch screen and streams de-identified patient data for real-time remote monitoring. In short, it is a convenient platform for outpatient artificial pancreas trials, running the entire artificial pancreas system on a cell phone.
  • The system’s home screen features hypoglycemia and hyperglycemia “traffic lights,” CGM information, system information, and various buttons (various screenshots are available at: http://bit.ly/uyxD2z). The idea of the hypoglycemia and hyperglycemia traffic lights is that “a user can pull the phone out of their pocket, and if the lights are green, things are okay.” If the lights aren’t green, “they may need to think a bit more.” The rest of the buttons were designed to simplify outpatient closed-loop experiments. The fork and knife button takes users to the meal bolus screen. The “Not Exercising” button allows users to inform the system that they are about to start exercising. The graph button takes users to CGM and insulin infusion graphs. The “Hypo Rx” button informs the system that the user just treated hypoglycemia. Users can also started closed-loop delivery from the home screen and inform the system of a recent CGM calibration.
  • Based on the desired system requirements, the diabetes assistant was designed with the following features:
System Requirements

Diabetes Assistant Feature


Runs on any Android cell phone

Approvable (Regulatory)

The Android platform was modified for use as a medical platform. The system has no phone or browser applications, no third party apps, and the 3G data connection is outbound only.

Software Portability

Can run on consumer devices or as an embedded code on custom designed hardware.

Inherently Modular Architecture

Ten applications were created for this system. Software modules are implemented as Android applications. Researchers can use their own control algorithms.

Wireless Communication 3G, Wi-Fi, Bluetooth, ANT+, BLE.
Scalable Naturally supports new and extended functionality (e.g., integration with heart rate monitors).
  • Android was specifically chosen as the mobile operating system of choice. Dr. Keith-Hynes mentioned this was because: it’s widely deployed (over 100 phones), it’s open source, it’s actively supported by Google, it’s designed for multitasking and interprocess communication, it’s inherently modular, and easy to develop code. The Diabetes Assistant running on the Android phone has several applications removed (e.g., phone, browser, games) – as we understand it, this helps with regulatory requirements.
  • The system is built with a robust remote monitoring system Data is anonymously streamed from the mobile device to a secure server at UVA Medical School. Once on the server, researchers can access the data through a web-based interface. Dr. Keith-Hynes showed a picture of the main screen in the online interface, which includes tabs for CGM and insulin delivery. Certain screens are designed for clinicians while others are for engineers.

Questions and Answers

Dr. John Pickup (King’s College London School of Medicine, London, England): So as I understand it, this is a generic device that will receive information from a number of sensors and pumps?

A: Yes. The device is also designed with a control algorithm, but it’s modular. It is intended as a research platform and people can drop in their own algorithms.

Dr. John Pickup: But it doesn’t control the pump?

A: Oh no it does. It executes closed-loop control. It sends basal and bolus commands to the pump.

Dr. Dorian Liepmann (University of California, Berkeley, CA): I realized I started getting advertisements on Gmail that are based off the emails I receive. I think this is the problem with a cloud-based system unless it is insulated. There is a tendency in the cloud to collect data. And you’re using Android…

A: The signal is only outbound and it’s strictly going to the server. I don’t think by using Android you’re going to start getting advertising on this device. It’s a research platform and a tool for us. It has to be a medical platform and we’ve cut out the browser and telephone applications.

Dr. Pickup: What happens when the cell signal cuts out?

A: It is no different than in any other closed loop scenario. The device is wireless to the sensor and the pump on the body. The data is stored on the device, and then transmitted to the online servers when the cell signal returns.

Dr. Bruce Buckingham (Stanford University, Palo Alto, CA): This is wonderful. What is the communication going to be between the pumps and CGMs?

A: We’ve used Bluetooth for the first round. Particular cell phones use ANT+. Bluetooth low energy is the next thing coming. The advantage of using a cell phone is you can get the latest wireless technologies as they come along.



Michael Kremliovsky, PhD (Medtronic Diabetes, Northridge, CA)

In an effort to “significantly raise the stakes” of the demonstration session, Dr. Kremliovsky presented Medtronic’s smartphone-based closed-loop system and showed (via a live Skype call and remote, real-time CGM/pump data feeds) how it is being used in Australian inpatient trials. The commercialized product, which Dr. Kremliovsky predicted would be similar in cost to today’s sensor-augmented pumps, will run the software through the pump rather than third-party phones or software; this is in keeping with Medtronic’s preference for using only its own proprietary systems. It was exciting to see real-time results from the system as it was being used in Medtronic’s inpatient Australian trial – definitely well worth the wait for some technical difficulties to be resolved with the communications equipment. (Dr. Kremliovsky quipped that on retirement, he would invent conference audio/visual equipment that operates “the same way that closed-loop control is working in Australia…flawlessly.”) He predicted that several years of clinical study would be required to bring the system to market; we hope that the FDA’s upcoming draft guidance sets out a feasible pathway in this regard. For a picture of the system, which we demoed after the presentation, please visit: http://bit.ly/t6NUHT.

  • The system consists of a Veo pump (though Dr. Kremliovsky noted that the study does not involve low glucose suspend), two separate Enlite sensors with modified software (the next-generation system will use all off-the-shelf products), a proprietary translator that converts signals from Medtronic’s frequency to Bluetooth, and a smartphone that integrates all the data and runs the PID control algorithm. During Q&A, Dr. Kremliovsky explained that Medtronic is using the smartphone as the crux of the system because phone-based software is relatively easy to modify. However, once the system is finalized, the algorithm and user interface will likely be run from the pump itself. We were not surprised by Medtronic’s decision not to open up its system to third-party developers and hardware (though this seems to be the trend elsewhere in the field as well), as the company has a strong historic preference for using its own proprietary technology. On the plus side, we expect that this will simplify the regulatory pathway. Though we think that open-ended collaboration with the commercial and academic community is generally beneficial, clearly Medtronic has made great strides toward a clinically useful closed-loop system thus far; we hope that the company continues to prioritize artificial pancreas development and push the field forward.
  • Dr. Kremliovsky displayed real-time results from a remote monitor of the portable system in use during a closed-loop study in Australia. (He also used Skype to call one of the study’s researchers, a member of Dr. Tim Jones’ group at the Princess Margaret Hospital for Children in Perth, Australia.) The two-night study has so far enrolled six patients, with two more planned before completion. Dr. Kremliovsky noted that the goal is not necessarily to achieve a certain level of glycemic control, but rather to see how the system operates. He said that so far the performance has been excellent, despite some overnight compression artifacts when patients sleep on the sensors (Dr. Kremliovsky did not give details on how the system accounts for these events, and we assume that Medtronic is still determining the best way to handle simultaneous input from two sensors).

Questions and Answers

Dr. Pickup: I know this is an investigational device, but presumably you are looking toward the stage of entering clinical practice. Can you say something about cost-effectiveness and affordability?

Dr. Kremliovsky: I don’t see a particular price difference between today’s sensor-augmented pumps and this technology. Of course, we are going to be on the pump – a smart phone is probably not Medtronic’s choice for running the algorithm. Everything is designed to move from the phone to the pump. We understand the regulatory challenges and that the FDA and other agencies will look to this. We know we will probably need to run a few years of clinical research. We wanted something that would be easy to modify – a pump is not, since you have to re-test and resubmit with every modification. Software-based approaches are much easier to get through the development cycle.



Dov Moran, BS (Technion Israel Institute of Technology, Haifa, Israel)

Emphasizing the need for connected, wireless, and easy-to-use devices, Mr. Moran (the inventor of the USB flash drive) provided an overview of a cellular-connected blood glucose meter that automatically uploads data to a web site, providing a patient and their healthcare provider with a personal journal of blood glucose measurements. This data, which was quite easy to read, allows for more powerful trend analysis and includes average blood glucose values for the past week, daily measurements over the past week, percentage of time in range over the past two weeks, last measured blood glucose value, last out-of-range value, and last missed measurement. Customized alerts can also be programmed for various events, such as three consecutive out-of-range measurements, measurements above a certain value, and no measurements recorded in the past two hours. The alerts can be sent to the patient, a caregiver, or a healthcare provider. During the question and answer session, Mr. Moran stated that both the wireless connection and server were very secure – noting that he even keeps his bank account information on the server. From our understanding, while other systems and devices exist (such as Telcare’s cellular enabled meter) that transmit and store data online, what differentiates Mr. Moran’s system is its unique hardware. Although the company will develop its own meter, the small chip that transmits the data could theoretically be placed in any meter or CGM product. We certainly find this approach intriguing, and we are eager to hear more about this system in the future (submission plans not provided) as well as track any partnerships that may emerge with other device companies.

Questions and Answers

Dr. John Pickup (King’s College London School of Medicine, London, England): Are there any security issues?

Mr. Moran: First of all, we use a common encryption for the data. It is not the most advanced encryption, but it is good enough. Then there is a very secure server. I even keep my bank data there. The privacy of the data will be maintained. There will be a password for the patient and a password for the physician and caregiver. The system is very secure.

Dr. Pickup: Can you upload CGM data instead?

A: We can do it with regular meter data or CGM data. It will also be a great research tool. We won’t have to continually check glucose meter results or look at look books anymore.

Q: Security is a very important issue in Europe. We are more sensitive about privacy then in the US. There is high risk for keeping all of this information in one central database as well as sharing this information on Facebook.

A: With regards to Facebook, the patient can make the decision on whether or not to share his or her data. It is not required. I have found that many people with diabetes would love to share their data with other people with diabetes and receive advice and tips. We also want to extend this capability beyond Facebook to include other social groups. Regarding Europe, I think this system could be a huge benefit for patients. I also believe that it represents a good business opportunity, but of course, this consideration is far less important.

Dr. David Klonoff (University of California, San Francisco, San Francisco, CA): You described the web page. But there are other websites that store data in a similar way. There are products that can send data to the web as well. What makes your product so unique is the hardware. From my understanding, you have developed unique hardware that is similar to flash drives in terms of data transmission. It can be put into a blood glucose meter and it will wireless transmit the data to a secure server. It doesn’t need to pass through an iPhone or Android phone. In effect, you are turning a blood glucose meter in a phone itself to get information out. Once it gets to the server, it can send messages back to your phone or Facebook. The smallness of the chip makes it unique. Dr. Pickup asked about CGM. So you can transmit data from any device. Your software is nice, but it’s not the only software out there. Your hardware is absolutely unique. So I wish you the best of luck with it.



Cynthia Marling, PhD (Ohio University, Athens, OH), Frank Schwartz, MD (Ohio University, Athens, OH)

Excessive glycemic variability is believed to be associated with increased risk for long-term complications. Dr. Marling and Dr. Schwartz discussed the development of a new system, called the 4 Diabetes Support System, that aims to detect glucose control patterns and characteristics (including glycemic variability) in people with diabetes. Currently, the system is designed to import CGM and insulin pump information from the Medtronic CareLink Network. The system analyzes the data, attempts to detect patterns, and provides physicians with guidance on how to solve the problem(s). In addition to glycemic variability, the system is trained to automatically detect 18 distinct problems with glucose control. According to Dr. Marling and Dr. Schwartz, as the system continues to be used, its detection capabilities continue to improve and expand. Additionally, the system remembers the specific glucose control issues detected for each patient. Glycemic variability presents a distinct challenge because of a lack of consensus over which measure(s) best captures glycemic variability (over 5o different measures exist). However, Dr. Marling and Dr. Schwartz highlighted that healthcare providers are usually able to easily recognize glycemic variability based on CGM data. A unique solution for detecting glycemic variability, therefore, could be to use machine-learning classifiers, which are capable of capturing physician perception and using this information to automate the detection of glycemic variability from CGM data. Thus far, the best automated classifier created by Dr. Marling and Dr. Schwartz agrees with physician ratings for glycemic variability 93.8% of the time. Stressing that the incorporation of additional physician ratings will improve its accuracy, they asked diabetes specialists in the audience to provide glycemic variability ratings on the project’s website.


Artificial Pancreas: Clinical


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

Dr. Kovatchev gave an exciting presentation describing the use of a recently developed mobile artificial pancreas system in an outpatient closed-loop trial. Yes! The system includes a Sony Ericsson cell phone running a hypoglycemia and hyperglycemia minimizer (the H2MS Virginia), an OmniPod insulin pump, and a Dexcom Seven Plus CGM. (The mobile system itself is described in much more detail in Dr. Patrick Keith-Hynes’ presentation; to view screenshots of the system, visit: http://bit.ly/uyxD2z) Two patients in Europe (Montpellier: Dr. Eric Renard; Padova: Dr. Claudio Cobelli) have participated in the study, which includes an impressive 18 hours of outpatient closed-loop therapy. Thus far, the system has had no major problems and managed to keep both patients in target most of the time. Notably, Dr. Kovatchev and the development team are “really excited” – we are also very glad to see outpatient closed-loop trials finally getting underway (albeit in Europe), and can’t wait to see specific data on patients’ glycemic control.

  • Dr. Kovatchev described some recent major milestones in artificial pancreas research:
    • 2008: FDA approval of an in silico simulator. This allowed algorithms to be tested quickly and easily without the need for time-consuming animal trials. The system uses a huge model of human metabolism, allowing various scenarios to be run on 300 simulated patients with a simulated sensor, simulated pump, and any control algorithm. In 2008-2009, the first study in humans used a control algorithm designed entirely in silico.
    • 2009: The UCSB/Sansum Artificial Pancreas System (APS). According to Dr. Kovatchev, this “opened up the field for pilot studies.” As a reminder, this system is a research platform that allows for seamless communication between different manufacturers’ insulin pumps, CGM sensors, and the researcher’s control algorithm from a single laptop. The JDRF multicenter trial of modular control-to-range is using the APS.
    • Now (2011): A portable artificial pancreas platform running on a cellular phone. This newly developed research platform runs on a cell phone, allowing communication between the control algorithm on the phone, the CGM, and the pump. The system’s CGM, pump, and insulin delivery information displays on the phone and allows user input for meals, exercise, etc. For more specific information on the portable system’s capabilities, please see the presentation given by Dr. Patrick Keith-Hynes (screenshots are located at: https://closeconcerns.box.net/shared/x9urg6tntge5t3ua9iod).
  • Dr. Kovatchev described an outpatient, control-to-range, closed-loop study that recently used the portable system. As of October 26, 2011, the study has been run in two patients, one at Montpellier with Dr. Eric Renard and one at Padova with Dr. Claudio Cobelli. The study is using a Dexcom Seven Plus, an OmniPod insulin pump, the aforementioned portable system running on a Sony Ericsson cell phone, and the H2MS Virginia control algorithm (hypoglycemia and hyperglycemia mitigation system). The researchers booked three hotel rooms near the clinical research center: one room for the participant, one for a medical doctor, and one for the engineers. The doctors and engineers were on call during the closed-loop portion of the study and could be called by the patient if anything went wrong. The two study participants had an evening meal and slept overnight with open-loop therapy. Then, they underwent ten hours of inpatient closed-loop therapy. They subsequently used the system for 18 hours of outpatient closed-loop therapy. This included a meal at a restaurant, sleeping overnight, and breakfast the following morning. Dr. Kovatchev noted that the “control was pretty good” – closed-loop therapy had no hypoglycemia and the “patient was virtually in target all of the time.” The system has worked fine from a technical point-of-view and there have been no major problems.



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

Dr. Zisser’s discussion of the artificial pancreas stressed that, “An artificial pancreas is not one thing.” He compared it to “getting to the moon” with many iterations and technologies along the way (like Kevlar and space blankets!). Similar to other speakers at DTM, he called the major limiting factors the delay in insulin action and inaccuracies in CGM – this seems to be the consensus in the field these days, and we hope that innovations in the coming years truly enable more ambitious and efficacious designs. However, in his view, “If we have a perfect sensor, we have amazing results.” Dr. Zisser noted that there is no clear winner among the three competing algorithms – MPC, PID, and MD-Logic. He also discussed the plethora of different insulin delivery strategies and new methods, including intraperitoneal delivery (5-7 minute onset; closed-loop study with Eric Renard beginning soon), inhaled insulin (closed-loop trial using MannKind’s Technosphere insulin to hopefully begin in early 2012), multihormone (glucagon and pramlintide) and others (Medtronic’s implantable pump, VIAject, Halozyme, InsuLine’s warming device, BD’s intradermal). His discussion of a tablet or cell phone platform to run the AP was particularly intriguing – these technologies are either under development or already being used in outpatient studies around the world (e.g., Montpellier/Pavia study discussed in Dr. Kovatchev’s talk in this session; Moshe Phillip’s overnight study discussed on page five of our ISPAD Day #3 coverage at http://bit.ly/rpmC3y). As these studies continue to happen outside the clinic, Dr. Zisser believes remote monitoring will become ever more important – fortunately, “We’re well on the way.”

  • A hybrid closed-loop study hopefully starting “in early 2012” will make use of MannKind’s Technosphere inhaled insulin. The rationale is that the inhaled insulin’s rapid onset will mimic the physiological first phase and a part of the second phase of insulin secretion seen in non-diabetic individuals. Subcutaneous insulin will be used for most of the trial, with inhaled insulin administered at the lowest dose to give a priming shot of insulin. In a closed-loop computer simulation using model predictive control and Technosphere insulin for a 75 g carbohydrate meal in 100 subjects, adding in the lowest dose of inhaled insulin blunted the post prandial peak without causing hypoglycemia later. The study will be submitted to the FDA shortly. Dr. Zisser is “very excited about the study,” which if all goes well, should hopefully be underway in early 2012. Assuming the study can get through the FDA, we think this is a very interesting and promising strategy to overcome one of the major limitations of current iterations of the artificial pancreas.



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

In a talk similar to his recent presentation at ISPAD 2011 (see our ISPAD Day #3 coverage at bit.ly/rpmC3y), Dr. Buckingham discussed clinical results on three partially closed-loop systems – predictive low-glucose suspend, control-to-range, and hybrid closed-loop (HCL). His group hopes to hear back from the FDA about the design of an outpatient study of predictive low-glucose suspend in the next few weeks, a longer time scale than was anticipated at ISPAD. They also look forward to participating in JDRF’s Control-to-Range study, which has begun internationally and will start soon at Stanford. Dr. Buckingham flashed through in silico CGM data on dozens of simulated adult and pediatric ‘patients,’ which predict that the system’s PID algorithm with insulin feedback will perform well in actual people. With regard to HCL, he happily announced that his group’s study recently completed enrollment (n=44 to intensive control, n=22 to conventional control), and he shared data on the first 34 patients to complete the hybrid closed loop portion of the study (the same results presented earlier this year in an ADA poster – see page 28 of our ADA 2011 Artificial Pancreas report at bit.ly/sJyX6Z). He said that the trial included only one notable failure, which involved overnight drift of the glucose sensor; nonetheless, no severe hypoglycemia or hyperglycemia occurred. After declaring that the evening’s celebratory drinks would be on him, Dr. Buckingham closed by sharing his patients’ ideal diabetes device: a single wearable patch (for both CGM and pump) that links with the same smartphone they use for calls, texting, and games.

  • Dr. Buckingham presented data on the first 34 patients to complete the three-day hybrid closed loop study, in which no severe hypo- or hyperglycemic events occurred. (Although all 44 patients have completed the study, the researchers have received institutional review board clearance to share information only from the first 34). Dr. Buckingham displayed a condensed version of the data table that he had presented in a poster at ADA 2011, which we reprint from our coverage of that meeting (see page 28 of the ADA 2011 Artificial Pancreas report at bit.ly/sJyX6Z for more details from that poster presentation).

First 6 hrs

By 24 hours

Day (7 am – 12 am)

Night (12 am – 7 am)






















Glucose Readings, hrs











Mean, mg/dl











SD, mg/dl











% 71-180











% ≤ 70











% ≤ 60











% ≤ 50











% > 180











% > 200













Moderators: J. Geoffrey Chase, PhD (University of Canterbury, Christchurch, New Zealand); Jeffrey Joseph, DO (Thomas Jefferson University, Philadelphia, PA)
Panelists: Bruce Buckingham, MD (Stanford University, Stanford, California); Howard Zisser, MD (Sansum Diabetes Research Institute, Santa Barbara, California); Boris Kovatchev, PhD (University of Virginia, Charlottesville, Virginia)

Q: Dr. Buckingham, you said you had one failure in this study. Did you look back at that failure to see what was different about him or her, or how the sensor was placed?

Dr. Buckingham: Sometimes the sensors just don’t calibrate as you’d like them to. To simulate real-world conditions, we calibrated once at bedtime, and we didn’t adjust at night. It was simply a drift in the glucose value, up to a 50 mg/dl difference. Things like that will happen. I don’t think we’ll ever be perfect, but that shouldn’t be the enemy of the good.

Q: What happens if the sensor delay causes the system to dose insulin when a glucose dip has not yet appeared on the sensor?

Dr. Buckingham: The algorithm is meant to predict roughly 35 minutes into the future. We know there is a delay; that’s why we have predictions. And once you’re past the nadir, it takes just a few minutes to turn insulin on again.

Q: Why do you all allow such a large breakfast? And for Bruce, what was the weight gain over those three-day periods of closed loop after diagnosis?

Dr. Zisser: As far as meal sizes, it depends on the type of controller. With a fully automated controller, it’s hard to give a really large meal and do really well. In our control-to-range trial, we’re trying to really challenge the system and so we give really large meals. Half the subjects will be adolescents, and when we talk to pediatric endocrinologists, they say these kids eat a lot. We want to make this as real world as we can.

Dr. Buckingham: At diagnosis, these kids come in with polyuria, polydipsia, and weight loss over a period of weeks or months. Some of that is liquid and some is muscle and tissue mass. They are in a rebuild state for 7-14 days. None of these kids were overweight or obese. There was weight gain but I think it was appropriate.

Q: Don’t you think breakfast is so hard because there are so many factors: large meals, cortisol, and decreased insulin sensitivity? I wonder if evolutionarily we’re really supposed to eat a large meal for breakfast?

Dr. Buckingham: I agree 100%; breakfast is the hardest meal of the day. You also have no insulin on board coming in to breakfast – a point that Dr. Tamborlane has brought up. We try to get some kids to eat some fat with breakfast. It really does help with gastric emptying. It’s the one meal of the day where you have to give a pre-meal bolus.

Dr. Joseph: It’s totally amazing how much progress has been made in the last few years. Is the existing technology – with sensors and their delay and inaccuracy and slow and variable insulin absorption – is this good enough to become a product for patients some day in the near future, or do we still need some breakthrough?

Dr. Kovatchev: Current existing technology is sufficiently good for control-to-range. This means that this type of therapy is in addition to standard basal-bolus therapy. This would allow for prevention of hypoglycemia and correction boluses when needed to attenuate hyperglycemia. It’s not going to replace manual insulin delivery. More precise sensors and faster acting insulins will make control to target possible – a full replacement claim. In other words, control-to-range is adjunctive, whereas control to target is a replacement for manual delivery.

Dr. Buckingham: It depends on what you’re comparing to. If you have a compulsive person that takes insulin before meals, checks their blood sugar 15 times per day, and does everything right, that person will do better without closed loop. But my typical adolescent forgets a lunch bolus and has hypoglycemia overnight after exercise. Parents stay up all night worried and do overnight tests. This is a huge improvement for them.

Dr. Claudio Cobelli (University of Padova, Italy): For a while, we have to live with some limitations of the technology. We are improving and interesting work is going towards improving sensors. Not even the technology itself but equipping the sensor with some software that makes it more intelligent. The major breakthrough will get rid of the wiring and make things wireless. This is a major step and it’s not far from what I can tell.

Dr. Zisser: I think even with current technology, we could get stuff like pump suspension, meal detection, and remote monitoring up and going.

Dr. Joseph: In your opinion, do you think insulin alone as single-hormone therapy is adequate for safety, or would a bi-hormonal system with glucagon be required?

Dr. Kovatchev: With the appropriate safety system, hypoglycemia can be reduced several-fold. It is a tradeoff whether to add the complexity of glucagon dosing.

Dr. Joseph: If it were yourself or a family member, would you want that safety feature?

Dr. Kovatchev: I think bi-hormonal is way better than single-hormone, and way better than normal control. Predictive pump shutoff or dosage attenuation and remote monitoring are already possible. These will reduce hypoglycemia very substantially. As further reductions are needed in the future, bihormonal systems may come about.

Dr. Buckingham: Dr. Ed Damiano’s group has shown that if there is too much insulin on board, insulin trumps glucagon. You have to control insulin levels also.

Dr. Chase: The title of the session was about clinical experiences. Two years ago David Klonoff let me moderate a similar session, and then the trials were smaller and had a more handcrafted feel. Now they are sort of done in bulk. To each of you – I know you are academics and so may augment your response, but what is the main lesson learned in your experiences over the last two years?

Dr. Zisser: I would say our main lesson is that we are spending a lot less time on regulatory work and actually doing clinical work. Two years ago we were figuring out how to get our ducks in a row for the FDA. From our standpoint, when we have new protocols to implement, we know who to talk to.

Dr. Buckingham: I would agree. I hope to see that as we get more studies under our belt, we can look at requiring less supervision and making less supervision required.

Dr. Kovatchev: Control-to-range is entirely possible with the current technology. The weakest link is the connection between pumps, sensors and the central processing unit. Improving communication is the focus of our efforts – to make that very solid and working all the time. This goes along with what Claudio said – making miniature devices that communicate wirelessly.


Artificial Pancreas: Scientific


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

Dr. David Klonoff presented the Diabetes Technology Society Artificial Pancreas Award to the FDA’s Chair of the Artificial Pancreas Critical Path Initiative, Dr. Chip Zimliki. Dr. Klonoff emphasized Dr. Zimliki’s hard work with Medtronic to get agreement on the design of the ASPIRE study of the low glucose suspend system (the IDE approval was announced this morning). Dr. Klonoff remarked, “Dr. Zimliki has taken a lot of abuse, and he takes it in good spirit.” We thought this was a notable choice from Dr. Klonoff and the Diabetes Technology Society and we look forward to hearing more about the FDA’s AP efforts as the deadline for the draft guidance approaches later this year.



Charles Zimliki, PhD (Chair, Artificial Pancreas Critical Path Initiative, FDA, Silver Spring, MD)

To start off his presentation, Dr. Zimliki responded to the award from DTS: “I have one comment about this award. As a type 1 diabetic thinking about the ASPIRE study, it’s about time. Lets get this study going and let’s get this device approved” – we’re glad to hear this enthusiasm from Dr. Zimliki considering the Veo has had European approval since 2008 and considering that, in our view, there should be some embarrassment around FDA for how long this has taken. Turning to his presentation, Dr. Zimliki explained the iterative path to an artificial pancreas and noted many of the challenges that will need to be addressed, most related to sensors and insulin. He was particular enthusiastic about the ASPIRE study and the cooperation of Medtronic in developing a protocol that meets the needs of all parties involved (from our view, this seems odd that FDA is thanking Medtronic for being cooperative and presumably moving things faster since to date, it seems FDA has been the bottleneck). Additionally, Dr. Zimliki repeatedly emphasized the Agency’s flexibility and willingness to speak with experts who may have answers to the tough questions. Nevertheless, we have heard disenchantment from some companies that this is not at all the case at all levels of the Agency, so we hope Dr. Zimliki’s sentiments are apt on some level and that they are becoming a greater part of the culture at FDA. It’s clear that his job is challenging and the Agency always has much to lose by approving an inadequate product, but we question the celebration associated with the ASPIRE trial since it is clearly a long trial and clearly a limited trial, and even a best-case scenario approval is likely at least two years away – putting at least five years between EU and US approval. .

  • In addition to the four devices in an artificial pancreas system, the FDA has added the “patient effect.” A blood glucose meter calibrating the CGM, the CGM itself, the insulin pump, and the algorithm are all commonly thought of as components of the artificial pancreas. However, the FDA also considers the effect patients have on the system – Dr. Zimliki specifically remarked that insulin resistance changes over time and we need to take into account these physiological effects. This makes sense from our view since technology influences different patients differently.
  • In June, the FDA released draft guidance on low glucose suspend (LGS) systems, with notable input from Medtronic. Dr. Zimliki stated, “I’d like to applaud Medtronic. They have come in and worked with us for the last four months to resolve outstanding issues. What you’ll find in the ASPIRE study and this draft guidance is that they’re not exactly alike. I believe we have developed a study that meets the needs of the company, patients, and FDA.” In Dr. Zimliki’s opinion, one of the more important aspects of the LGS draft guidance includes the use of CGM as a valid research tool to determine clinical effectiveness. Studies will be three months in length, must show non-inferiority in A1c, and superiority for CGM-based events for hypoglycemia (developed by Roy Beck). With a proposed suspend set at 70 mg/dl (which we believe will be characterized as too high for many patients), this will include CGM values <60 mg/dl, CGM values of at least ten continuous minutes at <60 mg/dl, no patient intervention for 30 minutes, and events filtered to avoid erroneous signals (e.g., exclude CGM rates >5 mg/dl/min). Dr. Zimliki highlighted that an incremental approach must be taken to the artificial pancreas, and anyone who looks at the ASPIRE study will see that there was flexibility in the design. He also encouraged the development of alternate endpoints (“the more the better”) – we note that later on, he backtracked from this point by explaining the difficulty of regulating a trial with many endpoints.
  • Control-to-range is the next step following LGS, although there are key areas for improvement in sensors and fail-safe system design. Dr. Zimliki highlighted sensor inaccuracy in particular, showing evidence of sensor attenuation from Dr. Ed Damiano’s studies at Boston University. While this might disappear with calibration, how to deal with these inaccuracies in the home setting without monitoring is an important question. Additionally, Dr. Boris Kovatchev and others have demonstrated sensor compression artifact (i.e., while sleeping, a user rolls on to the sensor and compresses the tissue, causing inaccuracy), which must be taken into account. In the overnight use period, sensor dropout from lack of communication is also a “big issue. It occurs, it’s a fact.” On the pump side of things, failures can occur as well – Dr. Zimliki showed data that it can often take over a day for a pump to recognize it has failed. This is especially concerning in light of controllers that ask pumps to operate in low dosing realms. Finally, comparing a picture of a hospital bed in a clinical research center to a child on a playground swing set, Dr. Zimliki noted there are serious real world challenges that need to be considered.
  • There are two ways to mitigate risk in an artificial pancreas system: 1) system reliability and 2) clinical mitigation. On the first, Dr. Zimliki noted developers could change the system or the device so that it “fails safely.” For clinical mitigation, manufacturers must recognize that no system is perfect and mitigation strategies must be in place to deal with CGM inaccuracies, dropouts, compression artifacts, and pump failures. Going to outpatient studies, FDA also believes that human factors is a significant issue – while PDAs and PDMs are made in the lab, companies must address whether patients can actually use these devices safely. One example he gave was a recent infusion pump with a “Go” button in red and a “Stop” button in green. He emphasized that we need to start thinking about patients, and children in particular, who may not be as savvy or familiar with the tools.
  • The high number of algorithm parameters makes clinical studies difficult to regulate. Dr. Zimliki gave one example of a study with hypothetical prediction thresholds at 70, 80, and 90 mg/dl at 35, 45, and 55 minutes. Assuming five other parameters, this leads to 45 different combinations that could be evaluated. For the FDA, it’s a “tough question” and “terribly challenging” to evaluate this parameter space. Dr. Zimliki emphasized, “We don’t have the answer. We ask anyone that does to please come talk to us.” We appreciate this openness and willingness to engage in dialogue and hope the FDA is making good use of the best and brightest in diabetes.



Roman Hovorka, PhD (University of Cambridge, UK)

Dr. Roman Hovorka gave an evidence-based, passionate endorsement of model predictive control (MPC) algorithms for use in the closed loop, calling this “the right approach to take.” After a thorough introduction to MPC and how it works, he discussed the major important benefits of MPC: 1) It deals effectively with insulin and sensor delays; 2) It can handle feed-forward information (e.g., manual boluses, food, exercise); 3) It can perform fault detection. He also noted that MPC has substantial clinical experience and success with numerous research groups. Finally, Dr. Hovorka pre-emptively criticized the two purported benefits of PID algorithms, arguing that these algorithms are 1) not physiological and 2) not better in cases where there is low prediction accuracy. Overall though, Dr. Hovorka believes that it is not important which algorithm is used as long as it works.

  • What is MPC and how does it work? Dr. Hovorka explained that MPC calculates insulin delivery using a mathematical model of insulin action. Over 100 models exist, including a compartmentalized physiological model (absorption and distribution of glucose, delay in CGM, absorption of insulin; used by Dr. Hovorka’s Cambridge team), a linearized physiological model (Magni et al., Journal of Diabetes Science and Technology 2007; used by the University of Pavia group), autoregressive MPC with exogenous input (i.e., the next CGM value is calculated by a weighted combination of the previous CGM value, insulin, previous meals, and previous errors; used by Dr. Ed Damiano’s group in Boston). Dr. Hovorka also described how MPC uses historical values to make predictions about the future. It uses moving time horizons (every five, ten or 15 minutes) with a prediction window of one to three hours (up to eight hours has been studied, but “this is challenging”). According to Dr. Hovorka, prediction accuracy in fasting is about 15 mg/dl per hour, while postprandial has higher error. Certain systems are adaptive in real time (Hovorka and Damiano), while others are adapted offline (Doyle).
  • MPC has a number of important advantages that make it ideal for closed-loop control: dealing with delays, handling feed-forward information, and fault detection. Considering the dramatic delays associated with subcutaneous insulin (70-80 minutes until peak insulin action is reached and another ten minutes for delay from interstitial glucose), closed-loop delivery of insulin can take over two hours to observe the maximum reduction in glucose. MPC is particularly well suited for this situation, as the model can accommodate this delay. MPC can also handle feed-forward information such as manual boluses, meals, and exercise. Finally, MPC can also do fault detection, such as situations when glucose excursions are not conforming to the expected pattern. Examples include insulin set failure, sensor fault, or pump failure (two DTM abstracts explore precisely these issues, demonstrating that such detection is possible: A20 Cameron et al., A42 Facchinetti et al.).
  • Designing an MPC algorithm requires important assumptions that can drastically affect the performance of the system. Knowledge of insulin PK/PD and meal absorption (only partially known in most cases) must be put into model, which gives insulin delivery advice under the supervision of a safety module. Dr. Hovorka summarized Dr. Ed Damiano’s fully closed-loop study, which showed how incorrect PK estimates could dramatically affect the system’s performance. While the model had a time to peak of 33 minutes, in reality it was over 100 minutes in some patients. In these cases of poor model fit, over-delivery in insulin resulted in the need for subsequent glucagon rescue by the system.
  • The Cambridge team’s MPC algorithms have achieved excellent results to date in over 100 people and more than 2,000 hours of closed loop. In previously published day and night studies, the system has reduced variability, increased time in target, halved hypoglycemia (<70 mg/dl), and reduced exposure to hyperglycemia. It has also been used successfully to adapt to meals, snacks, walks, and more strenuous exercise.
  • In a “pre-emptive PID rebuttal,” Dr. Hovorka argued that delivery using PID is 1) not better when there is low prediction accuracy and 2) PID algorithms are not physiological. Regarding the first, Dr. Hovorka asserted that bad predictive ability of a model is an engineer’s fault, which is equally the case with either a poorly designed PID controller or MPC controller. Additionally, if the glucose response to MPC control is unpredictable, it will also be unpredictable for PID control, rendering both systems unable to control glucose very well. Regarding physiology, Dr. Hovorka made a strong case that PID is nothing like physiological insulin secretion in non-diabetic individuals:


Subcutaneous Insulin Delivery

Pulsatile insulin concentration

Pulse every 4-6 minutes

No pulsatility

Portal-systemic gradient

100 fold

0 fold

Cephalic secretion


Pre-meal insulin dosing

Incretin effect

Incretin effect

Meal announcement/detection

First-pass insulin clearance



Time to peak in plasma

0 minutes

40-90 minutes

  • Audience Response Question: What parts of the closed-loop system need further improvements?
    • Glucose sensor: 45%
    • Control algorithm: 27%
    • Insulin analog: 23%
    • Insulin pump: 6%
    • Audience Response
  • Question: The time to peak action of rapid acting insulin is:
    • 40 minutes or less: 20%
    • 60 minutes: 36%
    • 80 minutes: 31% (correct answer)
    • 100 minutes or more: 13%
  • Audience Response Question: What is the most promising control approach for closed-loop?
    • Don’t know: 32%
    • MPC: 30%
    • Other (fuzzy logic, etc.): 23%
    • PID: 15%
  • Audience Response Question: What are the most important benefits of MPC?
    • Can handle insulin absorption delays: 35%
    • Can represent well PK and PD knowledge: 28%
    • Can help in fault detection: 20%
    • Can handle meal data: 17%
  • Audience Response Question: What are the most important aspects of control algorithms?
    • Should work well (be safe and efficacious): 93%
    • Should be easily initialized: 3%
    • Should be simple: 2%
    • Should be understandable: 2%
  • Audience Response Question: I would like to know about control algorithms:
    • Much more: 52%
    • Just a bit more: 13%
    • Know enough: 18%
    • Don’t care: 17%



Gary Steil, PhD (Children’s Hospital Boston, Boston, MA)

Dr. Steil made a case for PID over MPC algorithms, focusing on the clinical research that has been published for the two approaches (three papers each). Based on averaging all the participants’ CGM data for each trial, PID seemed to perform as well or better than MPC, with generally better post-breakfast stability and smaller post-dinner glucose excursions. Dr. Steil acknowledged that his comparison did not account for differences in study design. For example, glucose control was roughly similar between a 2001 PID study that he led and a 2010 MPC study from Dr. Ed Damiano’s group; the MPC study had significantly larger meal size, but it also was conducted nearly a decade later (and had the advantage of bihormonal control). He noted that even a head-to-head trial would not resolve the debate, since both MPC and PID algorithms can be endlessly modified. Instead, he argued for an environment where both approaches continue to be explored. He closed by observing that although MPC is favored in modeling studies, PID “dominates” in terms of use across medical technology and other industries.



Lutz Heinemann, PhD (Profil Institute for Metabolic Research, Neuss, Germany)

Dr. Heinemann gave an update on the progress of the AP@home Project, an EU-funded initiative designed to make artificial pancreas systems commercially available. The collaborators – 12 companies and academic institutions from seven European countries – are developing general classes of artificial pancreas prototypes: “two-port” systems that use commercially available pumps and CGM, and “single-port” systems that integrate insulin delivery and glucose sensing at a single site. In the single-port category, the Medical University of Graz and the Austrian company 4a Engineering are developing modified insulin catheters with commercially available CGM sensors built-in, so that the sensors are either inside or just outside the body; Dr. Heinemann noted that these devices have entered trials in rats and in non-diabetic humans. Taking a less conventional approach, Ecole Polytechnique Federale de Lausanne and Swiss-based Sensile Medical are in the early stages of developing their own single-port device: a porous catheter surrounded by a glucose-responsive hydrogel, which expands at low glucose and shrinks at high glucose. Thus, each time the pulsatile delivery mechanism sends insulin through the catheter, the pressure signal generated is modified depending on the external glucose levels. This signal can be analyzed and provides information about the glucose levels surrounding the catheter. As for two-port systems, Dr. Heinemann said that the CAT study – a 24-hour crossover-design comparison of closed-loop model predictive control algorithms from Cambridge and the University of Padova (iAP) with open-loop control – is winding down and will hopefully complete by the end of 2011, enabling submission of an ADA 2012 abstract. He said that two more studies (SPACE and CIPHER) are slated to begin in 4Q11; he did not discuss the designs in detail. The talk closed with a preview of more AP@home presentations to come – specifically, at a single-day, 150-participant symposium on February 7, 2012 in Barcelona, Spain (the day before the start of ATTD…we can’t wait!).



Moderator: Jeffrey Joseph, DO (Thomas Jefferson University, Philadelphia, PA)
Panelists: Charles Zimliki, PhD (Chair, Artificial Pancreas Critical Path Initiative, FDA, Silver Spring, MD); Roman Hovorka, PhD (University of Cambridge, UK); Lutz Heinemann, PhD (Profil Institute, Neuss, Germany); Garry Steil, PhD (Children’s Hospital Boston, MA)

Dr. Joseph: Chip, you brought up the concept that all devices eventually fail. What mechanisms do we have to identify failures and how do we build in fail-safe mechanisms?

Dr. Hovorka: It’s a property of the system, not the algorithm. An example is communication failure. The safe fall back in many situations is open loop. This is the easiest we can do. We’re lucky in type 1 diabetes that the safe failure model is open loop therapy.

Dr. Steil: Something that was in Roman’s talk was the idea of having a model that can supervise the system. If you put safety supervision over the controller, it can make the model safer. Models can do a good job at protecting against any kind of failure – all of it can be detected.

Dr. Gerold Grodsky (University of California, San Francisco, San Francisco, CA): I suspect I was the first to demonstrate the kinetics of insulin secretion. I was no engineer, just a physiologist. We made a model for the first- and second-phase insulin response, and it came out to be PID – I had never heard of this before. I have since spent lots of time studying both MPC and PID. One problem is that in both cases engineers are doing the work, and the algorithms have built-in rigidities that are not there physiologically. Sometimes the same constant is used for both of two parameters, when actually neither is a constant physiologically. I am glad to have both of you speaking today. I think we have the benefit of two terrific people. I don’t think we should look for who won – this was not a contest. I think that you both have demonstrated that people who know what they are doing can make either work. I suggest that these algorithms are always limited until they’re tweaked, and everyone is always tweaking one or the other. I think in both cases the engineering math is a little too restrictive relative to physiology. I think whoever wants to set up a program should pick the algorithm that you understand the best and that you can work with the best; don’t feel you have to go with one or the other. How’s that for being right in the middle? That’s not usually my position. (Applause.)

Dr. Hovorka: I agree we should focus on how it works, that tuning is always needed, and the proof is in the pudding clinically. I think it is very difficult to discover whether PID or MPC is better, and whichever you choose, you constrain your environment in some way. MPC in principle provides greater flexibility in terms of inputs, but I am not against PID.

Dr. Steil: I don’t really think this is the next question we have to answer. I think we need an environment where multiple approaches continue to be investigated.

Dr. Robert Engler (University of California, San Diego, San Diego, CA): I study physiology in a different area. What does the system do if the patient says: I just ate a meal, and the glycemic index is high, medium, or low. What is wrong with the patient input? Physiologically, the brain tells the CV system and digestive system what to expect from meals. What if you put this into the model?

Dr. Hovorka: It will aim to predict what will happen to glucose. This is how MPC algorithms work. Similarly, it will predict how insulin will affect glucose disposal.

Dr. Engler: So that is in the model?

Dr. Hovorka: I failed in my presentation if that was not clear. Yes. The models take patient input that includes carb content and manual boluses as well as blood glucose

Dr. Steil: Carb counting is very difficult, and the information that people input is suspect. The issue becomes further complicated by adding something like glycemic variability, or – as Dr. Howard Wolpert and I recently studied – fat content. It is commonly believed that high-fat meals simply delay gastric emptying, but we believe that it can also effect acute changes in insulin sensitivity. MPC has to track changes in these parameters, and it has to make control actions as soon as it identifies the changes.

Dr. Frank Doyle (UCSB, Santa Barbara, CA): Roman, I really enjoyed your discussion of MPC. There are three other attributes that make it ideal and explain why it became a driver in the refining industry in the 70s. One is handling multivariable systems. This is helpful for multihormone systems, something PID cannot handle. The second is constraints: thinking about safety constraints is natural in the algorithm of MPC. Finally, the major argument that drove its use in the refining industry: flexibility in the use of a cost function. In other words, what is your objective and what are you trying to minimize. MPC offers the flexibility to do zones and ranges. This favors MPC over a simple algorithm.

Dr. Hovorka: I agree, it offers great flexibility. Thank you, Frank.

Dr. Kenneth Ward (Oregon Health and Science University, Portland, OR): It almost seems like we’re discussing a binary choice. In Oregon, a PID-like system works well in some cases, but when we create insulin resistance with steroids, we needed a model to use concurrently with the PID system. Sensitivity adjustments showed us how to change the gain factors. In this case, patients underwent PID alone vs. PID with a supervising model.

Dr. Steil: On the previous idea of a cost function, these work perfectly with real dollars. But Chip was already talking about the high number of parameters. If the cost function becomes complicated there’s that. As soon we put PID with supervision, we’re encroaching on MPC-like control. None of the controllers I discussed had an adaptive PID controller. On the idea of fuzzy logic, these were a resounding failure in industry. It is incredibly difficult to show stability in an algorithm that constantly changes model parameters. These are all ideas that we should be evaluating in studies.

Dr. Stuart Weinzimer (Yale University, New Haven, CT): This was terrific – I enjoyed it very much. I wanted to make a plug – tonight Gary and Roman will be performing a duet from Annie Get Your Gun: “Anything You Can Do, I Can Do Better.” [some laughter] I think that given what Frank mentioned about building in constraints and safety, I see a place for adding on some model component to PID. Speaking as a non-engineer, there are so many fewer challenges that you are asking the system to do at night. For PID, we see wonderful overnight control. But at mealtimes, maybe we could have more modeling just for that. Have people thought about a day-night disconnect – sort of simpler or more complicated as we need it?

Dr. Hovorka: I think the combination of algorithms has been suggested. We want to take good things, not bad things, and there is an issue of complexity. Once someone has done closed-loop research with any algorithm, they will find things that could be improved. This could be done through a combination. I think that combining them in the way you describe, we need to be careful because one would feed into the other. Intellectual property and other issues come into play as well. In principle I am not against it, and I think it is good to think about.

Dr. Dorian Liepmann (University of California, Berkeley, Berkeley, CA): I am amazed that the controllers don’t use other data than glucose. This is in some ways a plug for my research area. MEMS (microelectromechanical systems) stuff is getting so easy to use. We can measure movement, activity, sweat… More data is always better for control systems. Why aren’t these other sensors included? I remember we talked about this 11 years ago at the first Diabetes Technology Meeting.

Dr. Heinemann: Some companies have tried to improve the signal of their CGM, such as Solianis from Switzerland. At the end of the day, I’m not sure how much they have gained. [Dr. Hovorka nods.] To have more data alone is not of help. I can clearly see that certain accelerometer data can be of help – for example, as a signal that exercise is starting. But to take it into the models…I have difficulty seeing it.

Dr. Steil: This is a very popular line of thought. As a side note, I would claim that exercise changes metabolic parameters, changing the model in an MPC design. Just as PID is really a one-line, incredibly simple algorithm, we should always be comparing the more complex systems to simpler ones. Something with new signals really has to perform better to justify the added complexity. There are a million places it can go wrong. It’s hard to go wrong with a one-line algorithm.

Q: What is the minimum requirement for frequency of glucose measurement?

Dr. Hovorka: A group in San Diego found that if you measure glucose every five-to-ten minutes, you can capture the blood glucose behavior. This is the minimum. Also, the measurements are smoothed, correlated because of data processing. I think five or ten minutes is enough.

Dr. Steil: All the PID data I showed used CGM measurements taken every minute. I have been an advocate of frequent monitoring to correct time lag, which I think is around three-to-five minutes. With control theory in general, increasing the sample interval almost universally decreases the stability of control. It is very difficult to hurt yourself by sampling more frequently, and it almost always helps. I’m not sure I would even attempt using PID at five-minute intervals.

Comment: When patients have different metabolism from one day to the next, making a static model that is useful in all these circumstances will require increasing the number of factors. An adaptive controller would eliminate this complexity. There was a question about adding other input variables. By using these, we can complement blood glucose information and at times give information farther ahead than what would be displayed through blood glucose. A multivariable approach is much easier to do with MPC than with PID, but also adaptive control will be the next stage. We have been using this at the Illinois Institute of Technology, and the results definitely look better.

Dr. Hovorka: Yes, a next step is to look at how adaptive control helps. As for the multiparametric approach – I would like to have a sensor for insulin concentration. And if we are going to a fully closed loop, a swallowing monitor would also be great. That’s my wish list on that.

Dr. Steil: I wouldn’t disagree with any of those statements.


Decision Support Software


Courtney H. Lias, PhD (Director, Division of Chemistry and Toxicology Devices, CDRH, FDA, Silver Spring, MD)

Dr. Lias opened the session on decision support software with a review of the FDA’s perspective, noting the Agency’s regulation of software is “an area of great confusion.” She highlighted the FDA’s rationale for regulating software: ensuring quality and safety in the systems. Surprisingly, approximately 50% of the recalls in Dr. Lias’ division are due to software problems. We found her one-off example of a misreported lab result quite alarming, truly reminding us how much the FDA has to deal with and why the Agency often seems so cautious.

  • The FDA regulates software in medical devices as well as stand-alone software when it has a medical purpose. The important question is the purpose of the software: whether it is intended to treat, diagnose, mitigate, or prevent a disease. Put simply, software that makes a medical claim is regulated as a medical device. Dr. Lias gave examples of decision support software, which include: systems that flag drug-drug interactions, hospital and lab information systems (high/low checks on data, delta checks), automated clinical calculators (eGFR), and risk predictions of disease. She emphasized that software with “library functions” (e.g., eBooks) and electronics like laptops and printers are not generally considered medical devices.
  • The FDA’s regulation of software is meant to ensure quality. The Agency is concerned with the development of that software, such as making sure the program was made deliberately, that the inputs and outputs are verified and validated, and it works in the way it was designed to work. Dr. Lias asserted that standard risk analysis and mitigation (e.g., FMEA) is key, although the Agency recognizes that you can’t always anticipate what might go wrong. She called database integrity “very important,” and with increased submission of Bluetooth devices, cyber security is becoming more and more important. These potential risks could be due to misuse or hacking of insulin pumps and glucose meters, even for nefarious purposes. For the Agency, “Building this security in is extremely important.” Finally, the FDA recognizes that bugs are inherent in any software development – as a result, the Agency has provided a process to identify and correct these problems. She encouraged industry to view the FDA’s software guidance document for greater clarity.
  • Even simple software functions can have a large impact. Dr. Lias mentioned that approximately 50% of recalls in her division are due to software problems that usually have some impact on patients. Commonly, these are the result of software bugs leading to a patient mismatch in the database of lab results. Ultimately, these can lead to the wrong lab result reported. Dr. Lias gave an example of an email she received yesterday, which described an instrument malfunction due to software. A patient with a lab result of 148.3 had a result displayed of 48.3. Of course, some of these results are ludicrous and doctors can catch the problems, but alarmingly, this is not always the case. Dr. Lias gave these examples “not as doomsday predictions but as common things we see.” For this very reason, FDA wants to see quality built in to the software development process.
  • The FDA considers “simple clinical decision support systems” as low risk devices. ​These include alarms (CGM at home, bedside monitors in the ICU), simple calculators (carb counters), and data manager software (uploading blood glucose monitoring data for trend analysis/alerts and pattern recognition). These are treated as low risk, and some are considered accessories to blood glucose meters. FDA simply wants to ensure that the data transfer integrity is there. Finally, online health tools are “an interesting area” with a “spectrum of gray” in terms of what is a device and what’s not. Dr. Lias encouraged everyone developing a software tool to adopt practices of design control and have a good quality system in place. For higher risk software, she encouraged dialogue with the FDA.
  • Model assumptions and misunderstandings of absolute vs. relative risk are “common” in FDA submissions. In particular, Dr. Lias mentioned risk tools that try to predict the future risk of type 2 diabetes. She noted that the model assumptions and inputs are very important in determining the reported result (“garbage in = garbage out”). In her view, the consternation in this area is in part due to environmental and family history factors that outweigh genes. Overall, she believes the incremental information given by these tools has not been great (we wonder if she is talking about Tethys’ PreDx test in particular). She also expressed criticism over pooling data from incomparable studies, the “biggest thing FDA sees” – a misunderstanding of absolute and relative risk.
  • Research is needed for more complex clinical calculators. She asserted that “no one is against the idea” of tight glycemic protocols in the hospital, but the question is how to ensure you have a tool that is well-developed. With this in mind, she believes, “It’s unclear where that field is going to go.” For treatment prediction for type 2 diabetes and insulin dosing, FDA would like to see more standardized testing and software tools.



Bruce Bode, MD, FACE (Atlanta Diabetes Associates, Atlanta, Georgia)

In a rich review of bolus calculators that will be published in an upcoming issue of JDST, Dr. Bode compared the technical specifications and clinical implications of all the major pump companies’ systems. Although the systems are broadly similar under many circumstances, they also differ in several ways – especially how they calculate (and account for) duration of insulin action. Dr. Bode said that several of these differences arise from non-clinical concerns, such as intellectual property issues (e.g., whether insulin action is modeled curvilinearly or linearly) and company marketing (e.g., differing recommendations for how long patients should set their duration of insulin action). Other areas of discrepancy (e.g., how to calculate food boluses when insulin is on board) seem as though they could have a single right answer, albeit one that is not clearly established or clinically agreed upon. Dr. Bode suggested that a single, FDA-mandated system for calculation might be better than the current range of approaches. He closed by polling the audience (who, as seen in a poll the day before, was predominantly industry); 54% of the respondents said that all bolus calculators should use the same rules and methodology, but only 38% thought that the FDA should regulate this standardization.

  • Dr. Bode briefly reviewed the history of bolus calculators. The first pump to include a bolus calculator was the Deltec Cozmo from Smiths Medical, which came to market in 2003. Subsequently Medtronic, which held a great deal of patents in the area, successfully sued Smiths and drove the Deltec Cozmo off the market. Today, all pumps have a form of bolus calculator, though Dr. Bode noted that some companies keep this software on a handheld device rather than the pump itself due to intellectual property concerns.
  • Dr. Bode questioned the universal practice of letting patients choose their own duration of insulin action (DIA), as opposed to using DIA values based on clinical data. In pharmacodynamic studies with insulin aspart (Novolog), Dr. Robert Henry’s group found that roughly half of the insulin dose works during the first two hours after insulin injection, with the other half acting between hours two and six (Mudaliar et al., Diabetes Care 1999). Dr. Bode lamented that although these data were submitted to the FDA, marketed bolus calculators allow patients to choose their own DIA from a range of 1.5 to eight hours. (During Q&A, Dr. Chip Zimliki of the FDA said that this accommodates the wide range of inter-patient DIA variability that has been reported by Dr. Edward Damiano’s group and other researchers.) Dr. Bode noted that pumps differ with regard to how they model the action of insulin that has already been dosed (insulin on board, or IOB). Some calculators use a curvilinear model for IOB, while others use a linear model (due to intellectual property concerns, he said). Also, companies differ in their recommended settings of DIA; Dr. Bode said that these recommendations tend to come from marketing departments rather than scientific studies.
  • Clinicians and companies disagree about how best to deal with insulin stacking – the common and potentially dangerous occurrence when insulin from two or more boluses is active at the same time. For correction boluses, essentially everyone agrees that IOB should be subtracted directly from the dose that would otherwise be given. For food boluses, however, two main schools of thought exist: 1) subtract IOB from a food bolus, just as for a correction bolus, either a) all the time or b) only if BG is below some threshold value; 2) never subtract IOB from a food bolus, but do reduce food boluses with a “reverse correction” if blood glucose is below some threshold – either a) the pump’s target BG value or b) 70 mg/dl. For example, if the target blood glucose is 100 mg/dl, the actual BG is 70 mg/dl, and insulin sensitivity factor is 30 units per mg/dl, one calculator would reduce the food bolus by one unit ([100 – 70]/30 = 1), while a different calculator might not adjust the food bolus unless the actual BG were below 70 mg/dl. In a poll of the audience (see immediately below), Dr. Bode found a range of opinions similar to that in the field as a whole.
  • All bolus calculators will subtract insulin on board (IOB) from a correction bolus but not from a food bolus. What should happen with extra IOB in bolus calculators for a food bolus?
    • Subtract from a food bolus if there is IOB remaining: 22%
    • Always cover a food bolus but if below glycemic target, use negative or reverse correction: 48%
    • Subtract from a food bolus only if blood glucose is below 70 mg/dl: 18%
    • I don’t believe in bolus calculators: 12%
  • Dr. Bode compared five different bolus calculators, using several brief clinical examples to make sense of the complexity. A few of the main differences are listed below.
    • Medtronic’s Paradigm pumps never subtract IOB from a food bolus, and they use a target range rather than a single glucose value, such that hypoglycemic values will be corrected to the lower end of the range, and hyperglycemic values will be corrected to the upper end of the range.
    • J&J (Animas)’s OneTouch Ping does not subtract IOB from a food bolus unless the blood glucose value is below target (which can be either a single value or a range).
    • Insulet’s OmniPod counts only insulin from correction boluses toward IOB, but not from food boluses. Also of note, the OmniPod will not calculate any boluses when the BG value is below a user-set minimum value (chosen from the range of 50-70 mg/dl). It uses a single user-settable target as well as a user-settable “correct-above” hyperglycemia threshold.
    • Under a complicated system that Dr. Bode said took him 10 hours to figure out, Roche’s Accu-Chek Spirit adjusts its recommendations for correction boluses after meals to account for an anticipated postprandial excursion. When initializing the Spirit, the patient inputs what they consider to be an acceptable post-meal glucose by entering both an acceptable meal excursion rise (e.g., 80 mg/dl) and an offset time where BG should not fall due to food still being absorbed (e.g., 120 minutes). Based on this value and the duration of insulin action, the calculator determines which blood glucose values should be corrected with insulin and which should not. For example, if the normal BG target is 100 mg/dl but the highest “acceptable” postprandial level is 180 mg/dl, the bolus calculator will not recommend correcting a BG of 160 mg/dl until a couple hours after the meal (i.e., when the food bolus should have kicked in). The Spirit calculates IOB from meal and correction boluses, but not snack boluses (e.g., up to 24 g carbohydrates), and it recommends carbohydrate consumption when BG is below target.
    • Tandem’s t:slim, pending FDA approval, subtracts IOB from (and performs reverse corrections to) food boluses only if BG is below 70 mg/dl. The pump has a single user settable glucose target, and Dr. Bode said that it shifts away from multiple user-settable options in an effort to simplify pump therapy.

Is IOB calculated using this type of bolus?

Is IOB subtracted from this type of bolus?

Reverse Correction







Animas OneTouch Ping



< Target BG



Insulet OmniPod†





User choice

Medtronic Paradigm






Roche Spirit






Tandem t:slim



< 70 mg/dl


< 70 mg/dl

*Uses predictive dose curve, †Will not calculate any bolus below set minimum BG

  • Dr. Bode closed with three more audience response questions about the different methodologies and whether they should be somehow standardized.
  • Which pump appears to have the most logical bolus calculator?
    • Animas OneTouch Ping: 17%
    • Insulet OmniPod: 6%
    • Medtronic Paradigm: 23%
    • Roche Accu-Chek Spirit: 8%
    • Tandem t:slim: 6%
    • None: 40%
  • Should all bolus calculators use the same rules and methodology for calculation?
    • Yes: 54%
    • No: 36%
    • Don’t care: 10%
  • Should the FDA mandate that industry all adopt a standardized set of rules and methodology for bolus calculators?
    • Yes: 38%
    • No: 59%
    • Don’t care: 3%



Charles Zimliki, PhD (FDA, Sliver Spring, MD)

After explaining the FDA’s rationale for classifying insulin dose calculators as medical devices, Dr. Zimliki detailed the agency’s primary review areas for such products and outlined outstanding challenges facing their development and approval. With regards to the FDA’s rationale, Dr. Zimliki stressed the significant risk posed to patients if the software malfunctions or contains errors. Insulin dose calculators are currently classified as Class 2 devices, and an investigational Device Exemption (IDE) is required to initiate clinical trials. The FDA’s primary review areas include: 1) the intended use (what therapy it recommends, how the inputs are measured, who is the end user); 2) the algorithm (what the parameters are, can parameters be adjusted); 3) the clinical evidence (how often the calculators are overturned, why are they overturned, clinical outcomes); and 4) human factors. Finally, for outstanding challenges, Dr. Zimliki underscored the need for industry to find ways to reliably measure inulin on board (particularly for home use devices), to minimize human use errors in the target population, and to obtain more comprehensive clinical data for these devices, such as how often the dosing recommendation is followed, the frequency of hypo- and hyperglycemia, and patient demographics.



Moderators: Carol Herman, RN (FDA, Silver Spring, MD), Elena Hernando Perez, PhD (Universidad Politecnica de Madrid, Madrid, Spain)
Panelists: Courtney H. Lias, PhD (FDA, Silver Spring, MD), Charles Zimliki, PhD (FDA, Silver Spring, MD), Bruce Bode, MD (Atlanta Diabetes Associates, Atlanta, GA), Ulrike Pielmeier, PhD (Aalborg University, Aalborg, Denmark)

Mr. John Walsh (Advanced Metabolic Care + Research, Escondido, CA): I have a comment on Bruce’s talk. The biggest factor in calculating accurate bolus is not having hidden insulin stacking. The duration of insulin action (DIA) sets the amount of insulin calculated. When I did a survey of 400 pumps in 2007, the average DIA was set at 3.1 hours. This is well short of research that shows DIA of 5.5 hours. With a DIA of three hours, the calculated insulin on board for a 10 unit bolus after three hours would be zero. But more likely, it would be 2.4-4 units if a more accurate DIA was set.

Dr. Bode: To FDA, why do you let patients change the true PD of insulin? You let them set this anywhere from 1.5 to 8 hours.

Dr. Zimliki: Because that’s what industry wants. That’s a tough question. There was a recent publication from the Boston University group showing the PK profiles of insulin and how to address this in the artificial pancreas. There is a wide range of profiles with regards to the diabetic population. We wanted to give opportunities for different algorithms to address this wide range.

Mr. Walsh: I talked with Lutz Heinemann from Profil on the way that PD is measured. It does not reflect the true DIA because of basal suppression during the test. We don’t have an accurate pharmacodynamic measure for pumped insulin. There are few studies.

Dr. Lutz Heinemann (Profil Institute, Neuss, Germany): Data on insulin infusion at different sites is relatively scarce. We need additional studies to not only obtain PK data, but PD data, ideally in different patient groups.

Q: To the FDA, you discussed your guidance on insulin calculators. For academic clinical trials, where would the guidelines fall for paper-based algorithms in converting to computerized spreadsheets? This is not a product in development, just for in an academic setting.

Dr. Zimliki: Just a reminder, this is FDA’s perspective, not guidelines. When you deal with paper-based systems, it doesn’t meet the definition of a medical device. I caution you on paper-based systems – paper alone following a flow chart is not a medical device. You can conduct studies as is. If you want to evaluate these algorithms without the need for an IDE that may be the way. But once you go into software, you cross that line – be careful.

Q: You showed how the insulin on board calculators can be quite complicated. With the new methods for accelerating insulin absorption with a heat pad and other methods, do you think this will make things more complicated? How do we handle this?

Dr. Bode: When you look at the different absorption of rapid acting insulin, dermal insulin, and heating an area up, you change the duration of action. Given that duration, the curvilinear relationship may not be appropriate and the action time might still be long. Until it comes to market, there is no way for industry to tell you. There’s also some recent research that suggests calorie counting is more accurate than carb counting for food bolusing in type 1 diabetes. There is no pump on the market doing that. To FDA, what would the process be like for calculators based on calorie counting?

Dr. Zimliki: Why do I feel like this is the inquisition? It depends on how complicated the algorithm is. We encourage flexibility, diversity, and want to make sure the clinical evidence supports the algorithm being proposed. We do not want to restrict the development process. We encourage you to develop algorithms and provide the appropriate data.

Mr. Hal Joseph (Clearwater Valley Hospital, Orofino, ID): I have a question about insulin pump algorithms. From my experience, over the whole population of pump wearers (or their parents), people never know their insulin sensitivity factor or insulin-to-carb ratio. The algorithms tend to dumb it down so much that people don’t think about it. My computer password at work has to be reset monthly, mainly as a measure against my forgetfulness. Maybe bolus calculators could similarly require regular user input. Also, why can’t your pump tell you a projected A1c based on the blood glucose information it has, compare it to your target value, and if you’re off, say that you have to make an appointment with your doctor or your pump will default?

Dr. Bode: I know offhand that Medtronic’s data management system can convert information to a projected A1c. For the pump to tell you to contact a physician: I think that’s appropriate but it has to go through the FDA. I’m not sure about the other part of your question. I think it’s important that when people use smart devices, they have to have a piece of paper with their insulin sensitivity factor, insulin-to-carb ratio, and target, and basal rate settings. You should always be able to go back to a manual method. You should print out the manufacturer’s readout of these numbers or create your own sheet, in case people aren’t able to use their pump system anymore. This happens a lot – you can ask the FDA.

Dr. Lias: We don’t have all the answers, but generally we are in favor of new devices that help patients.

Dr. Michael Marvin (Pronia Medical Systems LLC, Louisville, Kentucky): We have FDA approval for the Yale Protocol and have developed multiple protocols for the system that we have not yet used. In order to start clinical trials comparing these protocols, do we have to obtain an IDE?

Dr. Zimliki: I’m not sure how much we want to talk about that right now. However, yes, you will need an IDE to conduct that study.

Dr. Barry Ginsberg (Diabetes Technology Consultants, Wyckoff, NJ): I’d like to take you back 27 years to when I was a researcher in the DCCT in Iowa. All my patients were using “bolus calculators.” Ten were on paper, a few used simple calculators, and four were on HP-programmable calculators. I gather that if this had been done today, I would have to put in an IDE? Also, all four of these patients crossed state lines.

Dr. Zimliki: A calculator is a general-purpose article. If you are talking about entering numbers to get results, you are straddling a fine line.

Dr. Ginsberg: What if it is programmable?

Dr. Zimliki: We would have to discuss it. Thanks for the history lesson – I think 27 years ago was around the time I was diagnosed with type 1 diabetes.


Technologies for Metabolic Monitoring


Rudy Hofmeister, PhD (C8 MediSensors, San Jose, CA)

Dr. Hofmeister presented early clinical data on C8 MediSensors’ noninvasive CGM device, the HG1c. He reviewed the device’s Raman spectroscopy technology and its relative advantages to current glucose monitoring systems, and he then described calibration studies conducted in the spring and summer of 2011 (for background on C8 and the HG1c, see coverage of the EASD 2011 exhibit hall at bit.ly/nbXJDZ). The proof-of-concept study in the spring suggested early promise, while the larger summer study showed better results that reflected an improved skin interface (thanks to a biocompatible gel). Dr. Hofmeister said that no data were thrown out by the algorithm, though the commercialized version could reject obvious outliers (i.e., those that result from loss of optical coupling between the sensor and the skin). Notably, only 60 measurements were shown from the eight-patient calibration study – a small number from which to extrapolate. The company has since conducted a variety of human factors and usability studies across a range of conditions. Dr. Hofmeister said that the results have been positive and, importantly, show that the measurements are insensitive to abdominal movement (data not shown). The company is highly optimistic for future clinical studies and planned commercialization in Europe next year, though several key opinion leaders expressed skepticism during Q&A. We look forward to seeing larger datasets from patients under more challenging circumstances, as we have not yet been able to get a clear idea of the system’s accuracy in real-world settings.

  • Dr. Hofmeister showed select data from a calibration study of C8’s HG1c device, which was conducted during summer 2011 at the Diablo Clinical Research Center. The study included 48 subjects who came to the study for up to three days each, wearing two sensors for 90 minutes per visit (150 minutes during glucose excursions), for a total of 88 subject-days. Reference measurements were taken every 15 minutes with YSI and a blood glucose meter (J&J’s OneTouch Ultra). By our calculations, this adds up to over 1,000 data points (88 visits x 1.5-2.5 hours per visit x four reference measurements per hour x two sensors per patient), though results were shown from only a single device, worn by “many” patients at multiple skin sites. For these 73 data points, the Parkes Consensus Error Grid A score was 84.9%, and the MARD was 7.2% over a range of roughly 55-305 mg/dl. The study overall included adults with type 1 or type 2 diabetes (48%/52%), across a wide range of ethnic groups (white, black, Asian, Hispanic, Native American), skin types (Fitzpatrick skin types II-VI, which includes the entire range except for people most sensitive to sun exposure) and ages (19-70, median 49). Demographics were not given for the smaller number of people in the dataset presented. Encouragingly, Dr. Hofmeister said that minor abdominal motion does not impair measurement (and even helps somewhat); hopefully this trend continues under more extreme physical activity.
  • A prospective measurement study was also conducted in the summer of 2011, using the same device and calibration shown from the calibration study. In this small study of patients who had not been part of the calibration study (n=8), the Parkes Consensus Error Grid A score was 68.3%, with MARD of 13.3%, based on 60 data points across a 200-mg/dl range (roughly 60-260 mg/dl). Dr. Hofmeister noted that this dataset included outliers that could have been automatically identified and rejected in real-time, but were presented to show the dataset without any information discarded. He said that the good performance from a small calibration set (n=6) suggested that a universal calibration could be developed based on a fairly small dataset; he did not discuss the similarities or differences between the people in the prospective study and the calibration set.



Richard Stadterman, MA, MBA (VP, Global R&D, Bayer HealthCare LLC, Tarrytown, New York)

Mr. Richard Stadterman, Vice President of Bayer’s Global R&D, discussed the company’s newest blood glucose technology, the Contour XT, which he hopes will “set new standards for accuracy and precision.” This claim about both accuracy (meter value compared to YSI) and precision (variance around YSI) was emphasized throughout the presentation. Impressively, the company’s data shows 100% of points within ±15%/15 mg/dl (the new proposed ISO standard) and 98%+ of points within ±10%/10 mg/dl. In developing this new technology, Bayer produced a new mediator (“Without the mediator, we would not have gotten the performance”) that has a stable, high signal-to-noise ratio. Additionally, the company’s new proprietary algorithm helps compensate for errors from multiple sources. The result is excellent accuracy that we hope holds up in further clinical and regulatory studies.

  • Bayer’s next generation system reduces the sources of error inherent in current blood glucose meter systems.

Sources of Inaccuracy

Current BGMs

15% BGMs

Bayer’s Next Generation

Blood Sample Variation

(Hematocrit, Interferences)

5.1% SD

3.0% SD

2.4% SD

User and Environment Variation

(User Techniques, Temperature)


5.7% SD

5.0% SD

3.0% SD

Sensor and meter variation

(Lot calibration, aging over shelf life, electrode area, reagent deposition)

3.1% SD

2.8% SD

2.5% SD

True Performance

95% ± 16.5%

95% ± 13.0%

95% ± 9.2%

Clinical Trial

(Sample differences, reference method error)

4.4% SD

4.4% SD

4.4% SD

Clinical Trial Performance

95% ± 18.7%

95% ± 15.7%

95% ± 12.7%


  • Bayer’s next generation system has achieved excellent accuracy, with 100% of points within ±15%/15 mg/dl, and 98%+ of points with ±10%/10 mg/dl.


Percentage of Readings within specific error limits*

<75 mg/dl (n=90)

±5 mg/dl

±10 mg/dl

±15 mg/dl





>75 mg/dl (n=510)









*Fingertip capillary blood samples from 100 subjects were tested in duplicate by a trained user using three test strip lots. Reference samples used YSI.



Jeffrey LaBelle, PhD (Arizona State University, Tempe, AZ)

Dr. LaBelle discussed his lab’s efforts to develop a monitor for tear glucose (TG) as an alternative to fingerstick monitoring. He noted that TG has been studied for decades and that at least two academic groups today are developing contact-lens-like sensors (Parvis et al., IEEE Spectrum 2009; Baca et al., Clin Chem 2007). Dr. LaBelle said that by comparison his own group’s plan is a “stopgap” measure – a system in which patients would capture fluid from the eye (<0.5 ul) and transfer it to a BGM-like device, which includes microfluidics and an electrochemical sensor. The measurement process involves three sub-elements: the sensor assay itself (so far GDH-FAD seems preferable to glucose oxidase; Lan et al., JDST 2011]), the fluidics of the meter (he and his students are currently using self-manufactured version-1.0 systems, but they have planned out a version 4.1 that is sleeker and easier to mass-produce), and capture of the tears (in animal studies the researchers are swabbing the eye with a polyurethane foam plug; they have done preliminary research into vapors and other more advanced methods for humans). Dr. LaBelle’s group has received funding to conduct a yearlong series of rabbit studies. These began with a simple confirmation that touching the rabbits’ eyes does not cause damage, and they will include tests of the correlation between blood glucose and TG measurements. The researchers and their business partner are trying to bring the product to market as soon as possible; Dr. LaBelle’s slides referred to a tentative launch target of 2015, which seems to us extremely ambitious for this early-stage technology.



Pietro Galassetti, MD, PhD (University of California, Irvine, Irvine, CA)

Dr. Galassetti discussed the results from several of his own studies that had examined the use of exhaled gas analysis to monitor glucose, insulin, and lipid levels. For glucose levels, the integrated analysis of exhaled acetone, ethanol, methyl nitrate, and ethyl benzene were found to correlate very well with plasma glucose levels in healthy participants and participants with diabetes (r=0.88 to 0.97). Five exhaled gases (not specified) and four exhaled gases (toluene, PdONO2, CH3ONO2, and CO2) were also found to correlate well with plasma insulin levels (r=0.98) and plasma triglyceride levels (r=0.97), respectively. While impressive, Dr. Galassetti noted that the major pitfall with this technology was the large size of the machinery (it currently fills half a room) and the high cost. Moving forward, he expressed hope that the introduction of alternative gas-sensing technologies and the further refinement of the methodologies used will lead to smaller and cheaper devices, possibly paving the way for commercial development within a few years time. Given the greater simplicity, affordability, evidence base, and ongoing research associated with blood- and interstitial-fluid-based glucose monitoring, we think it unlikely that exhaled gas will become widely used any time in the near term.



Moderators: Dorian Liepmann, PhD (University of California, Berkeley, Berkeley, CA); Athanassios Sambanis, PhD (Georgia Institute of Technology, Atlanta, GA)
Panelists: Pietro Galassetti, MD, PhD (University of California, Irvine, Irvine, CA); Rudy Hofmeister, PhD (C8 MediSensors, San Jose, CA); Richard Jeffrey LaBelle, PhD (Arizona State University, Tempe, AZ); Richard Stadterman, MA, MBA (Bayer HealthCare LLC, Tarrytown, NY)

Dr. Liepmann: How did you pick the compounds you looked at?

Dr. Galassetti: There were a number of possible candidates based on existing literature. Acetone was one example. We looked at compounds that were potentially linked to glucose metabolism. Then we did one-on-one correlation analyses with over 200 compounds. Same gases were correlated with glucose or insulin, some not at all. Then we did a step-wise analysis of different iterations of three or four gases. So, it was combination of expert knowledge and a fishing expedition.

Dr. Barry Ginsberg (Diabetes Technology Consultants, Wyckoff, NJ): On the calibration for C8, is this going to be a universal calibration? A single device calibration? If a single device, how long does the calibration last?

Dr. Hofmeister: The calibrations are universal. They are set in the factory and applied to all devices universally.

Mr. Steve Perlman (OnLive, Palo Alto, CA): About C8 – I really applaud you guys for sticking with it. I love to see it and it’s getting better and better. To what extent is the FDA going to allow you to do the calculation on the device and then display it on a smart phone? Is that an acceptable structure? Is that your plan and something the FDA will approve?

Dr. Hofmeister: Our initial goal is to release the device in Europe through a CE Mark. Following that, we hope to initiate the FDA approval next year. You’re right that the smart phone introduces complexities. The FDA recently released guidelines on the use of mobile devices with medical devices. This was eagerly anticipated and it was encouraging. We believe it will be possible to have a medical device working on the smart phone, where the medical device is defined as our unit and the application that runs on the smart phone. This does not include the phone’s hardware and operating system itself.

Mr. Steve Perelman: I hope you succeed. Thank you.

Dr. Lutz Heinemann (Profil Institute, Neuss, Germany): A comment about C8. I’ve seen many non-invasive monitors through the years. We’ve been enthusiastic under control conditions for many of these technologies. Some of the figures you showed look very familiar. But in daily life, the reliability and accuracy was not as good with these devices Each and every time, people tend to be very positive about it, but in reality it fails. And then the people are excited and then very disappointed.

Dr. Bruce Klitzman (Duke University, Durham, NC): With the Raman spectroscopy, it seems there may be several artifacts. In the examples you showed, as optical coupling decreases, the measurements seem to go up. Do you see that as a problem? The last thing you want to do is measure glucose way too high. In normal use I imagine there would often be a decrease in optical coupling, though you said that algorithms would probably be able to catch these outliers.

Dr. Hofmeister: Not probably – the device is reading the spectrum of what it sees. If you identify something that is obviously out of family, you can reject it as the device not being installed correctly. I can’t tell you exactly what happened that time – whether the device slipped off the skin or clothing slipped under it – but the system will recognize when it is not in contact with the skin.

Dr. Liepmann: They left this in to show what the raw data would be.

Dr. Klitzman: What about hydration state, local hematocrit, or other physiological variability that could interfere?

Dr. Hofmeister: We have tested over a wide range of hematocrit range, from mid-30s to 60.

Dr. Klitzman: But it is probably 10-60 in micro-capillaries – the blood test doesn’t reflect the site where you’re measuring.

Dr. Hofmeister: I understand. We’ve seen no issues of interference across the hematocrit range, nor with medications, drugs, or other sugars.

Dr. Klitzman: What about hydration state?

Dr. Hofmeister: We can measure water directly with the system – this is another potential application, of course, and potentially a very lucrative one. We are measuring that as precursor step to measuring glucose.

Q: For the breath gas analysis, you are perturbing glucose levels and insulin levels. I am following the development of analytical methods closely. A colleague of mine has found that there is quite a lot of variation in breath composition in different conditions, such as during a lung infection. Have you considered this?

Dr. Galassetti: That is a main concern for us. Breath is a melting pot. There are so many different things coming out of the same avenue. You have to account for all of these possible confounding factors. We need to find gases that are not perturbed depending on other situations. We actually have much more data than I had time to show. We stared with people that were healthy or had type 1 or type 2 diabetes. We are extending to these analyses to other types of patients as well. We are beginning to pick up gases that are impacted by confounders. We want to have four or five gases at the most, otherwise we have a situation that is not applicable. As I showed, the gases we have found give a correlation coefficient of 0.9. We have alternative gases though that only lower the correlation by a few percentage points. These would still be valid in case any of the four I showed you are not usable. We are looking at the performance of these gases over time. It is a complicated issue, however.

Q: There are multivariate calibration methods and pattern recognition that may be useful at diagnosis.

Dr. Galassetti: For glucose monitoring, diagnosis is one issue. But we don’t use breath in diagnosis. In the lifespan of a diabetic, we maybe have three or four tests for diagnosis, vs. thousands of tests one they have been diagnosed. We could use a subset of gases in type 1 or type 2. The PNAS paper that came out in 2007 found that surprisingly, methylnitrate is high only in hyperglycemia in type 1 diabetes. Not in healthy subjects and not in type 2 diabetes. It correlates with hyperglycemia AND low insulin, which is a hallmark of type 1 diabetes but not the case in type 2 diabetes or healthy individuals. We could perhaps use an equation with methylnitrate for type 1 diabetes.

Dr. David Klonoff (Mills-Peninsula Health Services, San Mateo, CA): Do you think that we’ll see contact lenses used for measuring tear glucose?

Dr. LaBelle: It’s possible down the road. We did play with putting sensors in the lachrymal gland. But we had wires hanging out of the eye. It’s not possible today, but maybe down the road.

Dr. Klonoff: Rick, the data you showed was so impressive – it was in the range for hospital glucose monitoring requirements. Do you have plans to distribute in hospitals?

Mr. Stadterman: I think at this point we want to effectively launch that for personal use. And once we have launched around the world, that’s more of a commercial decision. Clearly we open up a lot of options with this technology.

Dr. Klonoff: Service dogs can detect hypoglycemia. What do you think they are detecting?

Dr. Galassetti: This issue comes up a lot; I was talking to Fran Kaufman earlier today about it. A trainer came to the lab and we tried to come up with a joint experiment. I haven’t done a lot of hypoglycemia studies – mostly we have studied hyperglycemia. But I haven’t seen any specific gas increases in hypoglycemia, and I would be surprised if the dogs are able to perceive a drop in gas concentrations. I’m not excluding the possibility that the concentrations changed, but possibly it involved skin emission – Mark Evans in the UK published a paper on this a few years ago – or other gas emissions. Also – possibly more importantly – the dogs may be picking up some whole-body signaling that they are trained to perceive: heart rate, sweating, and other more-subtle signals. I think it is a complex signal. It may include gases but I am unclear on the contribution.

Dr. Robert Vigersky (Walter Reed Army Medical Center, Washington, DC): I was wondering if you could tell us what happens to the selected gases during exercise. Does this limit their applicability?

Dr. Galassetti: This is a complicated issue. We did some exercise studies. There are two factors here. There is the change in metabolism that occurs during exercise. Glucose metabolism accelerates depending on duration, timing, and rigorousness of exercise. In addition, you have very acute changes in gas exchange dynamics. A lot of these gases can be traced and integrated correctly, but this will require enormous amounts of effort. Isoprene, for instance, exists in stores in the muscle. These stores are depleted in the first three to five minutes of exercise. It appears in the breath and then it disappears. We are developing a system right now that will allow us to also to detect and measure the concentration of gases in the blood stream. The results for isoprene have been very different from what we expected based on breath analysis. It is very complicated. We are trying to get some insight into it. If we come up with a breath measure for glucose, it will be a while before it becomes applicable to the post-exercise state.

Dr. Robert Vigersky: To Dr. Hofmeister: you can measure a wide range of glucose, but I didn’t see anything in the hypoglycemia range. In fact, I don’t think I saw anything under 100 mg/dl.

Dr. Hofmeister: The data I showed did have points below 100 mg/dl, some were into the 60s on those charts. We have also measured the effect of low glucose sensitivity in bench testing. We can go arbitrarily too high or too low, even to zero.

Dr. Ted Zhang (Dexcom, San Diego, CA): Dr. Stadterman, you touched on a very important point – that CGM accuracy is mostly determined by SMBG calibration. When we report CGM accuracy data, we base MARD on 100% of the data. For strip data, I saw it was based on reporting 95% of the data. Does data including 100% of measurements exist? For the new-generation sensor strip, what is the MARD based on 100% of the data? This would be very helpful for us – it is a seeding point for us.

Dr. Stadterman: Above 75 mg/dl, 100% of the data was within ±10%. You are getting a very narrow distribution; I think it will meet your requirements. I think it will depend on the pivotal trial data, but we are very confident in getting the data within ±10%.

Dr. Zhang: Currently?

Dr. Stadterman: I think all four of the manufacturers ought to be embarrassed. We were saying 95% plus or minus 18%. You are asking about the other 5% – it could be 20%-25%. I think that is a question the FDA has also asked: where are those outliers? It is a very critical question you ask. That is one reason we are so excited about this data – no outliers to deal with.

Dr. Zhang: The improvement in interference you showed – was this because of a lower device bias, or the mediator?

Dr. Stadterman: It was driven almost entirely by the mediator.


Hypoglycemia Detection


Francine Kaufman, MD (Chief Medical Officer, Medtronic Diabetes, Northridge, California)

Dr. Kaufman began her presentation on the Veo with an acknowledgement: “We’ve just had good news of the Veo IDE. I would like to recognize Chip Zimliki from the FDA and Mark Fallaice from Medtronic for working so arduously to make that happen.” She discussed the three published studies of the Medtronic Veo, which have shown reduction in the duration and severity of hypoglycemia with no evidence of increased hyperglycemia. Dr. Kaufman also showed new interim data from in-clinic study of the Veo, showing good results thus far. This data will be presented to the FDA, in addition to the data collected from the ASPIRE in-home trial of the Veo pump. Q&A was dominated by Veo discussion – we learned that the ASPIRE study will use the Enlite sensor and the LGS shut-off will be set at 70 mg/dl (to have a variable setting, the FDA would make Medtronic test it at multiple levels!).

  • The basis of the Veo is the stepwise, physiologic response to the lowering of glucose levels. As glucose levels fall, endogenous insulin release stops, followed by the release of epinephrine and glucagon, and symptoms of neuroglycopenia (shortage of glucose in the brain). Dr. Kaufman mentioned that the point of the Veo is to mimic the first step in this physiologic response – the decrease in insulin production. Additionally, she noted that hypoglycemia is a rate-limiting step in diabetes, making the Veo an important addition to diabetes management.
  • There are three published studies demonstrating the effectiveness of the Veo: a user evaluation from Choudhary et al. appearing in Diabetes Care 2011, a German study from Dr. Thomas Danne et al. presented at ADA (Diabetes Technology and Therapeutics 2011; for more information, see page 65 of our ADA 2011 full report at https://closeconcerns.box.net/shared/dz9hr6m94mu0ehsteybc), and finally, a data mining study from Agrawal et al. appearing in the Journal of Diabetes Science and Technology in September 2011; for more information see page 140 of our ADA 2011 full report [data for all ages] or our Day #3 coverage of ISPAD 2011 [data in pediatrics] at https://closeconcerns.box.net/shared/gtmf7g7hv84pmei808kh).
  • Studies show that the Veo mimics physiology and can help decrease hypoglycemia with no increase in hyperglycemia. Dr. Kaufman noted that the Veo is not intended to prevent hypoglycemia, as the shutoff occurs when a hypoglycemia threshold is reached rather than when hypoglycemia is predicted. However, we look forward to Dr. Bruce Buckingham’s work on this front – as we understand, his team is waiting for clarity from the FDA to begin an outpatient study of a predictive LGS system.
    • Common Findings with the LGS Across Studies
      • LGS set between 50 and 60 mg/dl
      • LGS event everyday or every other day
      • Over 50% turn on insulin in less than five minutes
      • Two-thirds of LGS events occur during the day
      • LGS events lasting two hours mainly occur at night and represent 10% of all LGS events.
    • Decrease hypoglycemia
      • Fewer sensor glucose values in hypoglycemia range
      • Increase of approximately 20 mg/dl/hour with suspend and two hours post
    • No increase in hyperglycemia
  • Dr. Kaufman presented new interim data from an in-clinic study of the Medtronic Veo. The data presented was in 30 out of 50 planned patients. Medtronic will present this data to FDA once the study is completed and analyzed. Participants entered the clinic after an overnight fast and subsequently exercised until the sensor glucose reached 85 mg/dl, at which time they stopped and were observed. Once the sensor glucose hit 70 mg/dl, the low glucose suspend feature was activated and insulin delivery was suspended for two hours. This protocol was repeated another time in the same subjects, but without the LGS feature turned on. Dr. Kaufman showed the baseline characteristics of the study participants thus far (n=30, mean age 35.8 years, 40% female, mean BMI 27.3 kg/m2, mean A1c 7.8%), specifically noting that BMI is continuing to go up in the type 1 diabetes cohort. In terms of LGS duration and the lowest glucose value observed thus far:

LGS On (n=30)

LGS Off (n=30)

Mean (SD) duration (in mins)

139.6 (77.96)

192.9 (72.35)

Median duration



Min, Max duration

5, 240

17, 240


59.3 mg/dl

56.9 mg/dl

  • In the in-clinic study, LGS reduced the duration and severity of hypoglycemia. Dr. Kaufman did not give specific numbers, but showed a graph illustrating the glucose profiles of study participants. After an overnight fast, subjects entered the clinic with an average blood sugar (YSI) between 100 and 140 mg/dl. Following exercise, average blood glucose in both the LGS on and LGS off groups dropped similarly to ~60 to 70 mg/dl. In the subsequent four-hour observation period, blood sugar in the LGS-off group remained hypoglycemic (<70 mg/dl). With LGS turned on, blood glucose rose to around 70 mg/dl at 90 minutes and between 90 and 110 mg/dl by the end of the four-hour observation period, indicating that the LGS was able to bring patients out of hypoglycemia safely and without a significant risk of hyperglycemia.



Henning Beck-Nielsen, MD (Odense University Hospital, Odense, Denmark)

Dr. Beck-Nielsen discussed the development of a hypoglycemia alarm based on continuous EEG monitoring and real-time data processing by Denmark-based HypoSafe. He noted that specific changes in EEG occur in people during the state of hypoglycemia (characterized by delta waves with low frequency and high amplitude), which he argued was unsurprising given the reliance of the brain on glucose for energy. While other activities, such a sleeping, can also produce changes in EEG, an algorithm created by the company has demonstrated the ability to specifically separate the hypoglycemia signal from other signals with high sensitivity in initial studies. The device itself consists of an EEG sensor (about the size of a quarter) implanted under the skin near the ear. The procedure takes approximately 15 minutes. An external battery powered device placed behind the year powers the sensor, receives and processes the EEG data, and produces audible alarms during hypoglycemia. Approximately 100 individuals with type 1 diabetes have been implanted with the device for up to six weeks thus far. According to Dr. Beck-Nielsen, no participants have complained of discomfort or experienced other adverse events related to the device. In these individuals, the device was found to reliably detect both induced and spontaneous hypoglycemia with the occurrence of only a “few” false alarms. Alarms typically occur about 20 minutes before cognitive impairment begins. The company is currently preparing to initiate a six-month randomized-controlled phase 3 study in 120 people with type 1 diabetes (timeline not given). Dr. Beck-Nielsen concluded by noting that the device could be particularly beneficial for those with hypoglycemic unawareness and could be used in a closed-loop system to improve its safety.



Xian Huang (Columbia University, New York, NY)

The winner of this year’s JDRF Student Research Award, Mr. Huang discussed two prototype microelectromechanical systems (MEMS)-based continuous glucose sensors with potential applications for subcutaneously implantable CGM. The sensors are small (<4 cm long), and they have demonstrated good glucose sensitivity across a range of 0-500 mg/dl, responsiveness to changes from high-to-low glucose, and excellent in-vitro stability for five hours (longer-term data not shown). Mr. Huang also presented data from mouse studies in which sensors were subcutaneously inserted and connected to a computer by wires. The Clarke Error Grid A zone scores were favorable (95.3% and 83.6% in two studies, as compared to blood glucose measurements in the tail), though the sensor/BGM discrepancies looked rather wide on the glucose traces he showed. We have heard great enthusiasm for MEMS technology and think that it will eventually become widely used in diabetes devices; we look forward to increasing academic and commercial research in the years to come. (We note that Debiotech uses MEMS in its JewelPUMP, which is reported to dose insulin with extremely high accuracy; we have not seen an update from this Switzerland-based company for some time.)

  • Mr. Huang presented on two prototypes, both of which use a glucose-sensing polymer (PHEEA-ran-PAAPBA) sealed inside a semipermeable membrane. Depending on the glucose concentration, the polymer changes in both viscosity and dielectric constant (capacitance). One prototype sensor measures the polymer’s viscosity by detecting changes to the vibration of membrane inside the system; the other sensor measures the polymer’s capacitance (this approach has the advantage of not requiring a vibrating membrane). Each sensor includes both the glucose-sensing polymer and a reference polymer to enable differential analysis (i.e., in order to cancel out interference from temperature, light exposure, motion, etc.).



Moderators: B. Wayne Bequette, PhD (Rensselaer Polytechnic Institute, Troy, NY); Lutz Heinemann, PhD (Profil Institute for Metabolic Research, Neuss, Germany)
Panelists: Henning Beck-Nielsen, MD, DMSc (Odense University Hospital, Odense, Denmark); Xian Huang (Columbia University, New York, NY); Francine Kaufman, MD (Medtronic Diabetes, Northridge, CA)

Dr. Heinemann: Dr. Kaufman, listening to your presentation and acknowledging the weakness of the CE Mark, can you elaborate on the concerns the FDA has about the Veo?

Dr. Kaufman: I don’t want to presume speaking on the behalf of the FDA, but there has been a lot of conversation back and forth. They are asking us to show safety and efficacy. The in-clinic study will hopefully show the efficacy of continuing or not continuing to infuse insulin during hypoglycemia. The in- home study will show safety and efficacy. I didn’t want to get into what sensor we were using during my presentation. With the new sensors available outside of the US, some patients are beginning to use that sensor. We will be going into the in-home trial with the new Enlite sensors. So, the FDA is asking us to show safety and efficacy. We are appreciative that we have now come to an agreement of what to do.

Dr. Galassetti (University of California, Irvine, CA): I’m wondering if you factored in the issue of hypoglycemia-associated autonomic failure with prior hypoglycemia. Subsequent responses to hypoglycemia are blunted if there was a prior hypoglycemic episode. Isn’t there an effect on the EEG signaling?

Dr. Beck-Nielsen: This is a very important question of course. I cannot give you a clear answer, but it’s an obvious problem. In some cases, a patient got a hypoglycemia, ignored it, and they got a new alarm. It is possible to get an alarm within a couple hours. To study that, we’ve set up a whole PhD project. We think it is important to study and we are working on it.

Dr. Douglas Muchmore (Halozyme, San Diego, CA): Thanks Fran. I’m wondering about the patients in the European studies with the full two hour suspend mode. How many of the two hour suspends at night were responded to? How many let the suspend continue?

Dr. Kaufman: Through the CareLink database, we can only see if they have they done a glucose value or done something on the device. Otherwise, the presumption is they slept through the alarm. Sometimes we see that people shut off the alarm and keep the pump suspended. Of course, stopping insulin infusion is would not be my first choice for treating hypoglycemia.

Dr. Muchmore: Is the alarm sufficient?

Dr. Kaufman: This alarm is louder and the pump flashes “Emergency, call 911.” It is significantly louder than past alarms. People still sleep through it. This is also a problem when people get older and can’t hear as well. We know that there are significant issues with sleeping through alarms.

Q: Congrats to you and Medtronic on the ASPIRE study. With the technology you have, can you differentiate the causes of hypoglycemia? Was it excess basal or excess bolus? Patterns with physicians? Most hypoglycemia is caused by user error and we often see the coaster pattern: hyperglycemia, then over treatment, then hypoglycemia, then overeating, etc. I’m wondering if you’d see a small improvement in A1c with this device over a couple of months.

Dr. Kaufman: We did have a poster here that one of our medical students worked it. It used one of our algorithms in CareLink and look at the events preceding pump shutoff. A lot of it is bolus insulin – a large manual bolus, a bolus for a meal, and bolusing without the wizard. We have not seen an increase in hyperglycemia in these studies. The hope is we won’t see a deteriorated A1c. The data also suggests that we can decrease the hypoglycemia events and perhaps the hyperglycemic events as well.

Dr. Robert Vigersky (Walter Reed Army Medical Center, Washington DC): Why did you choose 60 mg/dl as the level to suspend? Why is that the default? It this what will be used in the US studies?

Dr. Kaufman: The Veo pump in Europe has a broad range – it can be set from 40 to 110 mg/dl. In the US, the lower limit for an alarm will be 70 mg/dl and the higher limit will be 90 mg/dl.

Dr. Lutz Heinemann (Profil Institute, Neuss, Germany): Once the glucose suspend is on, it’s not only switching off the basal, but any ongoing bolus? I was also puzzled by the overnight alarms with short duration? My suspicion is that these are compression alarms where patients are sleeping and rolling on the sensor – not a real hypoglycemia.

Dr. Kaufman: Thomas Danne presented a study at ADA with the LGS set at 70 mg/dl. This is a relatively high value for alarming and shutting off. The low alarm was set at 75 mg/dl. For pediatric endocrinologists, you would expect a fair number of these alarms. The lower you set it, the fewer alerts and alarms you get. When we data mined for paired values of fingerstick blood sugars at LGS shutoff, 82% of the time the blood sugar was at least under 100. So there may be some false alarms. To your first question, there is no ability to deliver insulin while the LGS is turned on.

Dr. Bruce Buckingham: I am surprised that the shutoff level will be 70 mg/dl. I think you would get a large number of false alarms and you would want a lower number?

Dr. Kaufman: Our pivotal trial will be at 70 mg/dl. We are hoping to think about changing the device at some point. In the pump, the LGS is either on or off. The goal is reduction in hypoglycemia. We would like to change it so that LGS could be used at night, and you could pre-program the pump so that the LGS is turned off during day. We have had some early negotiations with the FDA. They said we could move this level of 70 mg/dl wherever we want, but we would have to study it at each level. So we’re starting at 70 mg/dl.

Dr. Sanjoy Dutta (Juvenile Diabetes Research Foundation, New York, NY): With the Veo, there also seems to be a reduction in hyperglycemia. Can you comment on that?

Dr. Kaufman: In people with diabetes, there is a lot of cycling between going low, overreacting, going high, overreacting, etc. Maybe some of the potential benefit is related to patient behavior. By mitigating hypoglycemia or making them feel safer or providing an earlier alarm, patients change their behavior and possibly achieve better control. In the study, the time period on was much longer than the time period off. There is not much suggestion that we are going to disrupt glycemia.


New Insulins and Insulin Delivery Methods


William Tamborlane, MD (Yale University Medical School, New Haven, Connecticut)

Dr. William Tamborlane gave an engaging presentation on the early results of the InsuPatch (InsuLine’s insulin pump infusion site warming device) in thirteen individuals with type 1 diabetes. As a reminder, the warming device is integrated into the infusion site and increases blood flow, thereby speeding insulin action. This clamp study showed a very beneficial impact of the InsuPatch in terms of T max GIR (reduced 36 minutes), T early 50% GIR (reduced 19 minutes), and AUC GIR0-90 min (increased 40%) vs. no InsuPatch. The device is now being tested at a higher temperature (40 degrees Celsius or 104 degrees Fahrenheit) and eventually as part of a closed-loop study. Assuming this device makes it out of the clinic, we’ll be interested to see how patients like it in the everyday setting, as well as data on the rate of adherence. Compared with newer formulations of insulin, we’d have to guess the InsuPatch is certainly a more favorable regulatory approach to speeding the action of insulin.

  • InsuLine’s InsuPatch infusion site warming device was tested in thirteen teenagers with type 1 diabetes. Participants had a mean age of 13 years, mean A1c of 7.3%, BMI <95% for age and sex, diagnosis of diabetes more than one year prior, and on insulin pump therapy for at least three months. Patients came to the research center for two separate overnight admissions within eight weeks. Upon entry, participants received a new infusion site for their insulin pump. Then, they randomly underwent a euglycemic clamp with or without the InsuPatch. The device’s temperature was set at 38.5 degrees Celsius (101.3 degree Fahrenheit) – Dr. Tamborlane emphasized that this did not cause discomfort in patients, although we have to admit, it sounds quite warm for a patch on the skin. A bolus dose of 0.2 units/kg was given via an insulin pump and the basal infusion rate was suspended. Dr. Tamborlane mentioned that the InsuPatch was turned on 15 minutes before the bolus and for up to 75 minutes after the bolus.
  • The InsuPatch significantly accelerated the onset of insulin action as measured by T max GIR (reduced 36 minutes), T early 50% GIR (reduced 19 minutes), and AUC GIR0-90 min (increased 40%).

Without InsuPatch

With InsuPatch

GIR max

7.2 ± 3* 8.2 ± 4*
T GIR max 126 ± 28 minutes 90 ± 21 minutes
T early 50% GIR 58 ± 20 minutes 39 ± 13 minutes
AUC of GIR0-90 min 262 367

*Not statistically significantly different; Dr. Tamborlane believes the small sample size may be to blame.

  • The InsuPatch is currently being tested at a higher temperature (40 degrees Celsius [104 degrees Fahrenheit]) and will eventually be used in a closed-loop study. In the first three patients at the higher temperature, T GIR max was 70 minutes, compared to 90 minutes in the previous study (at 38.5 degrees Celsius) and 126 minutes without the InsuPatch. Dr. Tamborlane did not share further details on the proposed closed-loop study, but he emphasized that the Yale group’s previous study in 2008 (Weinzimer et al., Diabetes Care) underscores the need for faster insulin action. As a reminder, this study showed that insulin on board lasted until subsequent meals hours later, increasing the risk of delayed hypoglycemia.



Robert Baughman, PharmD, PhD (VP, Clinical Pharmacology, MannKind Corp., Valencia, California)

Dr. Baughman gave a detailed presentation on MannKind’s Technosphere insulin (Afrezza), highlighting its rapid-acting profile (peak effect in 14 minutes and basal concentration by 180 minutes vs. 45-60 minutes and four to five hours for rapid acting analogs), the most recent clinical data presented at ADA (see page 164 of our ADA 2011 Full Report at http://bit.ly/ousuyX), the improvements of the Gen2 (“Dreamboat”) inhaler over the MedTone, and most notably, the agreements with the FDA for the new type 1, type 2, and PK/PD studies. The studies are currently listed on Clinicaltrials.gov and have an estimated completion date of September 2012 for the currently recruiting type 1 study and February 2013 for the type 2 study (not yet recruiting). We’ll certainly be following these studies closely and listening carefully to MannKind’s upcoming earnings calls for further details.

  • The type 1 diabetes study design agreement with the FDA (ClinicalTrials.gov Identifier: NCT01445951): a multicenter, multi-national, open-label, randomized trial over 24 weeks. It will compare the efficacy and safety of Technosphere insulin (delivered via the Gen 2c inhaler) plus a basal insulin vs. insulin aspart plus a basal insulin (glargine, detemir, and NPH). The study will also employ forced-titration algorithms for both basal and prandial insulins. A third treatment arm will also be included in which Technosphere insulin is delivered using the older MedTone C inhaler. This will allow for a head-to-head comparison of the pulmonary safety of Technosphere insulin using the Gen 2c inhaler vs. the MedTone inhaler. The primary outcome measure is change in A1c at 24 weeks compared to baseline. According to Clinicaltrials.gov, the estimated enrollment is 471 patients with a study start date of September 2011 and an estimated study completion date of September 2012 (July 2012 is the estimated primary completion date). The study is currently recruiting participants.
  • The type 2 diabetes study design agreement with the FDA (ClinicalTrials.gov Identifier: NCT01451398): a multicenter, multi-national, open-label, randomized trial over 24 weeks. It will compare the efficacy and safety of Technosphere insulin (delivered via the Gen 2c inhaler) to a placebo powder in insulin naïve type 2 patients inadequately controlled on metformin with or without one or more oral antidiabetic therapies. The primary outcome measure is change in A1c at 24 weeks compared to baseline. Courtesy of Clinicaltrials.gov, the estimated enrollment is 328 patients with a study start date of November 2011 and an estimated study completion date of February 2013 (December 2012 is the estimated primary completion date). The study is listed as “Not yet recruiting.”
  • The dose proportionality and dose effect (PK/PD) study design based on FDA feedback (MKC-TI-176): Single center, randomized, four way crossover design. Four Technosphere insulin doses will be administered: 10, 30, 60, and 80 units. A final 15-unit dose of regular human insulin will also be administered. The study will use a euglycemic clamp with a two-hour run in period. Blood sampling will occur for four hours for Afrezza and eight hours for regular human insulin. The primary objective will be proportionality of insulin PK as measured by C max and AUC0-180 min). Secondary objectives will include AUC GIR and Technosphere insulin bioavailability relative to regular human insulin. At this time, we have not been able to find this study on Clinicaltrials.gov.
  • Regarding the old MedTone inhaler, Dr. Baughman remarked, “The engineers came up with a simple way to improve it…replace it!” He described the benefits of the Gen2 “Dreamboat” inhaler, which offers a number of improvements over the older MedTone: less powder per dose, easier to use with no maintenance, and no reverse discharge (i.e., no recoating of the inside of the inhaler). A single use cartridge is placed in the Gen2 device, which will last 15 days without cleaning. With the previous MedTone inhaler, 15 units of Afrezza provided similar absorption to four units of injected insulin. Now, only ten units of Afrezza are required in the Gen 2 device.



Alan Smith, PhD (Altea Therapeutics, Atlanta, GA)

Dr. Smith described Altea’s PassPort transcutaneous therapeutic delivery system, which is being developed for use with insulin and Amylin/Lilly’s exenatide. The PassPort makes use of a thermal ablation device that create micropores on the skin’s surface; these pores enable delivery of proteins from a patch placed on the skin. In a study presented as a DTM 2010 poster, PassPort-mediated insulin delivery was shown to raise blood insulin levels similarly to basal insulin injections. However, insulin levels decreased rapidly around 12 hours due to closing of the pores. The company has since developed a proprietary technique for keeping the pores open over 24 hours, and they are now targeting 24-hour delivery of an insulin formulation that has bioavailability of 10%-20% (as opposed to 5% in the earlier study). As for exenatide, phase 1 pharmacokinetic/pharmacodynamic studies have shown sustained 24-hour delivery, with peak levels above those seen with twice-daily exenatide injections. Altea is working on a new exenatide formulation with reduced lag time (down from ~12 hours to reach target levels in the PK/PD studies) and improved bioavailability (up from 3%). The company’s goal is to commercialize daily patches that produce physiological exenatide levels equivalent to those of Bydureon (Amylin/Lilly/Alkermes’ exenatide once weekly).

  • Dr. Smith gave an overview of efforts to develop simpler and less painful alternatives to subcutaneous delivery of proteins (e.g., insulin and exenatide). He said that one of the most promising approaches for basal delivery is skin ablation, which can be achieved by laser (e.g., Pantec’s PLEASE system), radio-frequency energy (e.g., TransPharma’s ViaDor), or thermal means (as used by Altea). The ablation creates micropores through the stratum corneum, the hard top layer of the epidermis. This enables delivery of proteins into the skin for as long as the pores stay open. Dr. Smith said that this period is limited to 24 hours or less unless something is done to keep the pores open; he also noted that Altea has developed a proprietary method for doing exactly this.



Patrick Scannon, MD, PhD (XOMA, Berkeley, CA)

Dr. Scannon discussed preclinical results for XOMA’s human monoclonal antibody XMetA, which is an allosteric activator and partial agonist of the insulin receptor. Notably, Dr. Scannon indicated that the XOMA’s goal with XMetA was to replace the use of basal insulins. Using in vitro data, Dr. Scannon showed that: 1) XMetA binds to the insulin receptor at an allosteric site; 2) XMetA binding does not disrupt insulin binding at the insulin receptor; 3) XMetA specifically induces the autophosphorylation (activation) of the insulin receptor but not the IGF-1 receptor (which has been implicated in carcinogenicity); 4) XMetA activates the insulin receptor to a lesser degree than insulin; 5) XMetA and insulin together do not increase insulin receptor activation beyond insulin alone; 6) in contrast to insulin, XMetA only activates the Akt insulin signaling pathway but not the Erk1/2 signaling pathway; and 7) XMetA selectively activates glucose transport but not proliferation (insulin-activated proliferation). In a mouse model of diabetes (streptozotocin/high fat diet), XMetA lowered fasting blood glucose to levels just above those observed in control non-diabetic animals. This effect was dose-dependent as well as self-limiting, reaching a plateau at 3.0 mg/kg. Importantly, no hypoglycemia or weight gain was observed. In conclusion, Dr. Scannon stressed the potential of XMetA to improve glycemic control in people with type 1 and type 2 diabetes while avoiding the weight gain, increased risk for hypoglycemia, and potential mitogenicity associated with insulin therapy. During the subsequent panel discussion, he noted that XOMA was currently attempting to confirm these results in other species (including non-human primates) and designing toxicity studies. No timeline was given for the initiation of clinical studies.



Andreas Pfutzner, MD, PhD (IKFE GmbH, Mainz, Germany)

Dr. Pfutzner highlighted the many benefits provided by insulin pens over vial and syringe systems in the administration of insulin. These benefits include improved accuracy, safety, ease of use, convenience, and cost. In the future, Dr. Pfutzner argued that pens will increasingly include features that help improve insulin therapy adherence (currently at 77.4% according to a recent study). In particular, he underscored the potential of telemedicine (the transmission of pen data to electronic records) to eliminate under-/overdosing. Arguing that insulin pen needs differ for each patient, he proposed the development of a modular pen system that a patient could customize him- or herself. He concluded by stating his belief that insulin pens would remain the preferred method for delivering insulin for the next eight to ten years.



Moderators: Michael Weiss, MD, PhD (Case Western Reserve University School of Medicine, Cleveland, OH); Howard Zisser, MD (Sansum Diabetes Research Institute, Santa Barbara, CA)
Panelists: Robert Baughman, PharmD, PhD (MannKind, Valencia, CA); Andreas Pfutzner, MD, PhD (IKFE, Mainz, Germany); Patrick Scannon, MD, PhD (XOMA, Berkeley, CA); Alan Smith, PhD Altea Therapeutics, Atlanta, GA); William Tamborlane, MD (Yale University Medical School, New Haven, CT)

Dr. Zisser: Patrick, what happens when you give insulin in addition to the antibody?

Dr. Scannon: The allosteric mechanism predicts that at low doses of insulin, you would see glycemic effect. But as you approach maximal doses of insulin, there would be no greater effect.

Q: Do you have data on the weight status of the patients and fasting blood sugar in the post meal supplemental study?

A: Yes we do, but I cannot recall it.

Dr. Dorian Liepmann (University of California, Berkeley, Berkeley, CA): Pat, the data you showed is impressive. What is XOMA’s plan going forward?

Dr. Scannon: We are now extending into other species, especially non-human primates. The goal is to move to humans as quickly as possible, after our second species and our toxicity studies, which we are designing now.

Q: You showed area under the curve data for the onset of insulin with or without warming. During the fall, there didn’t seem to be any difference. Does that have to do with the warming scheme? Does the positive effect only take place during onset?

Dr. Tamborlane: We recently looked at insulin levels from a PK analysis. Early on, we had higher inulin levels with warming, but the decay after five hours was very similar. By increasing the rate of absorption, we may be increasing bioavailability as well. More insulin may be getting into the peripheral circulation.

Dr. Barry Ginsburg (Diabetes Technology Consultants, Wyckoff, New Jersey): Your binding data looks almost identical to insulin. All of the other data looks like you cannot activate all the receptors.

Dr. Scannon: In some pathways we see more activation and in some we see partial activation. We don’t know if that’s the assay variability or a real effect. We do know that we don’t affect the uric pathway and we do affect AKT pathway. This shows we can affect pathways differently. Once we get into glycemic control, we’re still investigating the various components and there are many steps in between each part of the pathway. But we’re getting a much greater definition. Some of it will be assay variability and some will be properties of allosteric.

Dr. Tamborlane: In our studies in adolescence, the rise in growth hormone in puberty causes selective insulin resistance. Higher glucose is then a major signal for insulin secretion. In the non-diabetic, this insulin resistance increases insulin levels and promotes protein anabolism and generation of IGF-1. So it may have a plus or minus effect. Having an antibody that stimulates glucose uptake may have either positive or negative consequences as well.

Dr. Scannon: We’re trying to understand the full implications of this allosteric antibody. Pediatrics are important. We’ve designed this for patients who are insulinopenic. In at least one model of type 1 diabetes, the antibody also significantly improves glycemic control. We need to start studying and asking these questions in non-human primates.

Dr. Tamborlane: Wouldn’t lowering insulin levels in obese type 2 diabetes cause weight loss?

Dr. Scannon: In a type 2 diabetes model, the animals do not gain weight. The animals that receive insulin gain weight – very, very large amounts of weight. The antibodies look like the normal controls. This could result in lower secretion of endogenous insulin.

Dr. Robert Vigersky (Walter Reed Army Medical Center, Washington, DC): Have you done studies to look at antibodies to XMetA?

Dr. Scannon: We are dealing with human antibodies in mice, so they are neutralized in 6-8 weeks. We can do only so much in mice. We will look in non-human primates where antigenicity is less of an issue.

Dr. Bruce Buckingham (Stanford University, Palo Alto, CA): I have a wise-guy question for Bill. We can’t get pediatric patients to take their insulin 15 minutes ahead of time. Now we’re asking them to turn on a heater in advance of meals. If we just gave insulin ahead of time, it would have almost the same effect.

Dr. Tamborlane: I have known Bruce for a long time and have tried to teach him about clinical trial design. For a proof-of-concept study, you want to give the maximum chance to show effect, and then in follow-up studies you push the envelope. The inpatient-use study will involve starting the warming element at the time of bolus; this could be keyed in right on the pump.

Dr. Tamborlane: I don’t want to leave Andreas out…the greatest advantage of pumps over pens in pediatric patients is the history function, particularly for boluses. Industry should move to pens with a history function – this could be on the iPhone – and dose calculator function.

Dr. Pfutzner: As I said, I think the immediate future is telemedicine, not only for pens but for all therapeutic approaches. I like your preference for pumps, but I would suggest pumps are more inconvenient for type 2 patients, for example. There is also resistance from insurance carriers for any new technologies, as we are all aware. I think the combination of these economic and convenience issues means that pens will remain more used in the near term.

Q: How close with temperature alone can we get to the ideal PK/PD profile for a closed-loop system? Will we have to combine it with other strategies to achieve this profile?

Dr. Tamborlane: We got T GIR max down from 125 to 90 to now 70 minutes. There is no reason you can’t combine it with other approaches. One attractive feature is that you are not affecting the insulin molecule itself. You can add Halozyme, monomeric insulin, etc. and get an additive effect.

Dr. Barry Ginsberg: I don’t think you can be so hard on industry. It turns out that there is a patent that does not expire until 2016 that is necessary for putting microprocessors into pens. Without putting a microprocessor into a pen, you can’t develop a pen with telecommunication abilities.

Dr. Pfutzner: Why don’t we put the microprocessor in the needle then?

Dr. Baughman: Does the warming pad increase the bioavailability of insulin? The bioavailability of insulin is generally low. For regular human insulin, it’s 55%. For analogs, it’s about 60%.

Dr. Tamborlane: I didn’t show any PK data, but we just received this data recently. It looks like the area under the curve for insulin is greater over the first 90 minutes.

Steve Perlman (OnLive, Palo Alto, CA): Good to see you Bill. You talk about warming the area around the point of injection. Have you tried warming the insulin? Then you could build a pen with heater.

Dr. Tamborlane: The concern from the FDA’s perspective when we obtained the IDE for the InsuPatch was the idea that warming the skin would inactivate the insulin. This is not a worrisome condition. Why do you get increased blood flow? Because the blood is taking the heat away from the source of warming. It’s a very pleasant warming sensation. I don’t know if putting a heater in a pen has been thought of.

Mr. Perlman: I have a comment on the patenting of pens brought up earlier. I personally believe, as a group of people working in diabetes, that it’s very critical that we come up with standardized methods for having data collected on a large scale. The tools exist. In my lab, we’ve done this with low energy Bluetooth. We have them working up to 30 feet. If patents are a problem, we’ll look at them, find ways around them, or chat with the patent owners. We have to stop having this individual data and we must aggregate and anonymize it. An example: my son’s nurse noticed she was getting irregular results when testing kids. She drew a correlation between the hand sanitizers and the problem. Someone actually mentioned that yesterday. How many people know about that? Very, very few. If we had meters reporting that “suddenly around this school there was an installation of hand sanitizers and the blood glucose average is markedly different,” this would be beneficial. If we could get these devices built around these things and get standards in place, we could use the data.

Dr. Pfutzner: Within the FDA, there are really attempts and efforts to standardize tools. There are gazillions of preference questionnaires, but no standards. This gives the regulatory frame for developers – they can know when they meet certain standards and can get that approval. Right now it’s a discussion and it’s a best guess if the data is sufficient or not. I entirely agree with you. So many things interfere with laboratory studies’ accuracy and we need standardized protocols.


Information Technology


Zach Little, BS (Program Manager, Health Vault, Microsoft, Seattle, WA)

Zach Little, a program manager for Microsoft’s Health Vault and type 1 patient himself, gave a great overview of where data management in diabetes needs to go. We valued his vision of a future where data is shared and easily integrated: “We need to break down those siloes. All of that data working together is much more valuable than working separately for various entities.” Mr. Little asserted that data should be stored in the cloud, available to anyone, easy to upload and access, and the hardware should use modern technologies like USB and Bluetooth. We appreciated this consumer electronics perspective and think that data management has not been given the attention it deserves from many companies in diabetes.

  • What do users want? According to Mr. Little, users overwhelmingly want to be and feel healthier. They want to share information with care providers and own their own data. For most, privacy is a concern and they don’t want insurance companies to have access to their data. Mr. Little noted the difficult balance this creates: the more secure something is, the less usable it is. He also argued that online health data imposes a higher security bar than online banking – while your bank will refund you money that is stolen, if someone steals your health data, you can never get it back.
  • What do users not want? Mr. Little made it clear that complex graphs, charts, and lists are clearly unwanted. He also noted that users don’t want to do manual input, even if it’s something as simple as stepping on the scale and putting a weight in a phone. This seems to be a consensus we’ve heard from many patients. Finally, Mr. Little emphasized that anything that’s complex to install won’t gain massive adoption. Instead, we have all been trained to plug things in and they immediately work.
  • The current state of diabetes technology does not generally support good data management.
    • “Almost no glucometers are connectable out of the box.” Mr. Little noted that a special cable is typically required, which is not easy to find and usually costs $15-20. He also argued that the software available to connect meters is highly variable and leaves users in a bad place. In his view, “We have to have better collaboration.”
    • Despite the benefits of CGMs, “the data is siloed.” Mr. Little expressed disenchantment that CGM data cannot be exported to an Microsoft Excel file. We agree with his view that the real value in data comes from combining it with other data sources, rather than just viewing it in isolation.
    • Insulin delivery data is rarely connected to blood glucose data.
    • Sharing is hard.
    • Current device connectivity is almost exclusively device to PC.
    • The utility of the 1000s of mobile apps for diabetes ranges from “highly variable to not very useful.” Most of these programs require manual data entry, which in Mr. Little’s view is not a big upgrade from writing in a logbook to begin with.
  • Microsoft HealthVault partnered with Kaiser and Cleveland Clinic, producing very mixed results in people with type 2 diabetes. People were sent home with blood pressure monitors, blood glucose meters, and a full set of instructions. Mr. Little noted that the results were extremely good for people who got things to work – readmissions were down over 50% (“huge statistical differences”). However, he revealed that “getting it to work” was really the key. As an example, out of the first five patients who used the program, none were able to set up their digital device and download HealthVault without having a technician visit their home! In our view, this speaks to the challenges of implementing cutting-edge technology in this population.
  • Bayer’s Contour USB, Sanofi’s iBGStar, and Telcare’s blood glucose meter demonstrate that things are moving in the right direction. Mr. Little noted that Bluetooth glucometers would open up the mobile world, and encouragingly, the latest Android release from Google includes a health device profile. He believes that there is some “pretty cool stuff that’s coming,” but it’s moving more slowly than he would have expected. We would guess that part of this is due to the FDA, although Telcare’s rapid approval shows that it is possible to get approval quickly.
  • What needs to happen to move the needle? Mr. Little argued that devices must comply with modern standards such as USB and the newest version of Bluetooth – in his view, “the great technology from Microsoft and Apple goes out the window when the old version of Bluetooth is used.” He also believes diabetes technology needs to work right out of the box, patients must be empowered to use the data, and it should be fun (e.g., FarmVille is a fun game, but the ultimate purpose is to encourage people to expand their social networks in a fun way. This is especially important in the teenage population, where individuals have more motivation and touch with social networking technologies.) Finally, he believes electronic medical record systems need to be able to take in large amounts of patients’ data and display it in a meaningful way for physicians.
  • Why do we need data in the cloud? Mr. Little asserted that once data is in the cloud, it’s in a generic format and can then be transformed for anyone that needs it. This includes state agencies, pharmacies, administrators, researchers, and healthcare providers.



John Hatem, MSN, RN (Oracle, Redwood City, CA)

Mr. Hatem discussed how clinical data repositories in healthcare information exchange solutions can support improvements in diabetes management and care. He began his presentation by highlighting the many challenges facing data management in the healthcare setting, including: 1) the huge volume of data; 2) increasingly complex treatments in a number of diverse settings (especially for people with diabetes); 3) the existence of data silos with diverse terminologies, forms, etc.; and 4) an absence of essential data acquisition, transformation, and analytical capabilities. Mr. Hatem noted that a number of comprehensive data models and repositories exist today that can help address these challenges. Such systems can integrate financial (billing, charges, payers, etc.), administrative (provider relationships, demographic data, organization hierarchies, etc.), and clinical data (lab results, consent, prescriptions, diagnoses, etc.). These repositories can also help standardize medical terminology, making the data much easier for providers, payers, and patients to analyze, transform, share, and use for both clinical and research purposes.



Marcus Grindstaff, BS (Intel-GE Care Innovations, Roseville, California)

Highlighting the importance of developing people-centered technology solutions that fit seamlessly into normal life and work flows, Mr. Grindstaff provided an overview of the Intel-GE Care Innovations Guide, a next-generation remote health management product. In the home of the patient, the product consists of a color touchscreen device (much like a tablet) as well as peripheral medical devices (such as a blood glucose monitor, blood pressure monitor, and scale) that automatically upload data into a program that is continually monitored by a clinical team (typically a nurse care manager). The patient can also report psychosocial data. Other features include: interactive health sessions, video-conferencing between the patient and healthcare providers, audio and visual notifications, and multimedia education. The system is intended to allow patients, such as those with diabetes, to take a more active role in their own care without needing to significantly disrupt their lives. Meanwhile, it enables healthcare providers to provide personalized and more continuous care at a remote location (we are interested in whether this program is currently reimbursed). Studies have shown that the Guide leads to improved medical adherence, quality of life, self-management, and awareness of personal health issues. Notably, Mr. Grindstaff noted that Care Innovations just entered into a collaboration with the American Diabetes Association (ADA) to develop personalized programs for each patient using the Guide that would address health issues extending beyond diabetes (such as smoking and hypertension). Patients would also have access to a multi-media package from the ADA. Mr. Grindstaff concluded by noting that improving patient outcomes is not all about the technology, which he said is necessary but not sufficient for optimal care. Rather, we need to engage patients and providers in the process by delivering solutions that interact with and support each individual’s lives, motivations, and goals.



Moderator: Charles Peterson, MD (US Army, Telemedicine and Advanced Technology Research Center, Fort Detrick, MD), Jan Wojcicki, PhD (Institute of Biocybernetics, Warsaw, Poland)
Panelists: David Klonoff, MD (University of California San Francisco, San Francisco, CA), John Hatem, MSN, RN (Oracle, Redwood City, CA), Zach Little, BS (Microsoft, Seattle, WA), Marcus Grindstaff, BS (Intel-GE Care Innovations, Roseville, California),

Dr. Peterson: While people are coming up to the microphones, I’d like to touch on the themes this morning: whatever we do should be useful and fun. It should be plug and play. It should be interoperable. It should have standards and solve the ontology problems and privacy problems. We’d like it to be personalized, predictive, and preventative. We need to know who our community and network is. Diabetes is a perfect microcosm to work in to solve these problems – we have a tremendous opportunity. This is a time of real creative problem solving where we can really make a difference.

Mr. Hal Joseph (Clearwater Valley Hospital, Orofino, ID): I’m wondering not about connectivity, but why doesn’t each blood glucose meter give actionable information? Things like averages for seven days, 14 days, and 30 days? Or, you test your blood glucose at 7 am and it gives an average and a notification that you are always high at this time. This is more actionable material. These are simple things that would help people change their behavior. I don’t think people would always want to give information to cloud.

Mr. Hatem: I think that’s an admirable goal. One of first questions: is the technology and device capable of handling those decisions? The technology you’re talking about is not sophisticated. I still see room for cloud or information exchange and movement of data. The challenge is to make devices smarter and useful.

Mr. Grindstaff: I completely agree with that. In many cases, glucose measurement is a key piece of information. But there are many other aspects. The more of that information we can bring into algorithms that are making recommendations, the better we can tailor the message to individual patients and make it more personal.

Mr. Little: I think that’s right. I think for blood glucose values that’s perfect. The value of the cloud is that we can combine multiple types of data. I can add in Nike+ data from runs. We can also combine weight.

Dr. Eda Cengiz (Yale University, New Haven, CT): Many hospitals and clinics are implementing EMRs. None are designed specifically for patients with diabetes. When you try to extract diabetes-specific information from an EMR, or interface with another database, the common response from IT personnel is that it cannot be done. Is it possible to extract diabetes-relevant data from the EMR, and specifically for patients with diabetes? Is there an opportunity to collaborate with big EMR companies?

Mr. Hatem: With health information exchange, we are agnostic on which EMR system is used. We have been successful in extracting data from some of these companies. There are mechanisms that make it difficult. But I don’t know any EMR where it’s not possible, especially data that’s related to diabetes.

Mr. Little: I was recently at the Connected Health Conference in Chicago, and providers would say, “You just don’t know how much data we have.” Well, I went to a large hospital in Boston, and they went through the last 25 years and looked at every single data on every single patient. It ended up being the amount of video uploaded to YouTube every five minutes, or the amount of photos to Facebook every hour. We have the technology there. It is large data in health, but these are solvable problems. Computers are good at this, this is what they do. Physicians also need to help us solve these problems.

Mr. Grindstaff: The barrier is business process and approach rather than technical.

Q: While I agree that you have to get the products out there, that they have to be fun, that they have to work, etc., I disagree with Marcus. It does matter what technology underlies the product. Can you discuss the technological challenges that remain and highlight areas that we should be working on to help in this effort? What is the next big thing?

Mr. Grindstaff: You are right. Maybe a better statement would have been that the technology needs to support work flow and the needs of patients. Maybe my statement was pushing the needle too far to make a point. But it really needs to start with the people that are involved, what they do, and what they need. We need to develop technology that will support those functions and activities. The other panelists were right in saying that we needed technology that is interoperable. We need to bring down the walls separating different systems. We need further collaboration between different companies. No single system alone will tackle all the challenges in the industry. We have to bring together the best ideas and the best systems.

Mr. Hatem: I’m a big advocate of standards and developing technology that will support processes used by clinicians and patients as well as businesses at the same time.

Q: Yesterday, we had a great session on communication between devices and data receivers, trackers, and analyzers. One of the hard parts that emerged is that companies do not want data from their devices intercepted and used to make treatment recommendations. How are we going to bridge this gap in diabetes? Other data can also help when analyzing blood glucose data. It would be great to be able to integrate all this data, but we have to negotiate with the companies to make this possible.

Mr. Hatem: I hate to sound like a broken record, but I think this is again a case of standards. We need to build standards that make use of current technology for device data capture. I don’t want to disrespect device manufacturers, but it doesn’t matter what device is collecting data to comply with standards in a data exchange infrastructure.

Q: The issue is not with obtaining data. It is with the legal and liability issues surrounding the use of data obtained from CGMs, for example, and using that data to give medical advice that can lead to a bad outcome.

Mr. Little: We need institutions that have the knowledge to take that data and give reliable and safe decisions. I do not see an application coming out in the next six months that will guide insulin-dosing decisions based on data from a device. It will likely remain a one-on-one, patient to physician process. I don’t think I’ve ever heard a physician say that there is too much information for me to make a clinical decision. Our job is to empower the physician to make decisions using all of this information. Unless there are changes in tort laws, etc., I foresee insulin dosing remaining a decision made between the physician and each individual patient.

Q: I’m working on a project that integrates hospital physicians, multi-group practices, and home health agencies. Is anyone thinking about one-click systems such that every physician, nurse, and doctor has their own personalized desires as to how they would like to look at the data? As you standardize and pull these things together, I suggest that for every individual that accesses the data, the system ought to be able to establish a one-click relationship with the data. When I walk into a room and want to access summary data from the last time I saw a patient, I don’t want to negotiate the system. It knows how I like to look at the data. That capability should be in the hands of patients and providers.

Mr. Little: I don’t think anyone would disagree with that. The technologies exist to do that. It’s more just enabling it and setting it up. When you’re talking about Oracle, Microsoft, and GE, these are three hypercompetitive companies. All of our products do work together. It’s a matter of getting the data to a form where it can be used and where nurses and doctors are able to use it in a meaningful way. However, I bet if you took a cross section of the people in this room, I bet you would come up with some similarities in terms of how people like to look at the data.

Q: A comment was made about physicians never complaining about having too much data. They do complain though. They don’t have time to look at all of it. One-click solutions would be wonderful if we can get there. Some of these solutions, however, may be slowing down the healthcare system and may be having unexpected consequences. What kind of studies and testing should be conducted in clinical setting before they are widely adopted?

Mr. Hatem: Can you elaborate on what you mean by slow down?

Q: Electronic medical records have slowed down the productivity of healthcare providers. My healthcare facility has a particularly inefficient system, which has slowed the ability of physicians to see patients by 30%. And it affects every healthcare professional, not just physicians. We need better systems that do not create other unintended consequences.

Mr. Hatem: Right now, we are quite successful at taking all this data and putting it into a unified infrastructure and making it available to patients, physicians, payers, etc. You are discussing more the user interface. If the user interface is not efficient, where you have five clicks where only one is needed or there is a page no one can understand, that’s a separate issue that can be addressed. The end users that have this perspective need to be part of this process.

Mr. Grindstaff: The challenge is presenting the right information to the right person at the right time. As more electronic medical records come online and more remote monitoring occurs, we are learning together how to do this more efficiently. But you are right, we have to have the end users at the table to get the right design.

Mr. Little: When we first came into the healthcare system, we came in a little arrogantly. We thought we could just understand how it worked and come up with solution quickly. We experienced a lot of tough lessons from our first product. It really blew up in our faces. What we learned is that we have to look at who is going to use the product and what they need it for. The initial pilot studies we did for the Cleveland Clinic and Kaiser were good. They involved 20-40 users. We need to do that on a much larger scale, though. It’s also going to be monetary based. Are we saving payors money? If we can’t prove that we are saving money, nothing will happen based on just intrinsic good.

Dr. David Klonoff (Mills Peninsula Health Services, San Mateo, CA): I’m hearing there are many needs to improve software and hardware. To me, the easiest system to develop is standards for transmission of blood glucose meter data to computers. I will tell you an anecdote from a meeting a few years ago. We at DTS thought that this was a good idea and industry needed encouragement. So we brought in a panel of people from the four major blood glucose companies along with an engineer from an electronics company in Silicon Valley. The engineer was the first to speak. He said, “I want everyone to know that I don’t know the first thing about diabetes. But I do know about flat screen TVs, and sales went through the roof once we had interoperable standards.” The other speakers gave their presentations, and by the end, there was unanimous agreement that they didn’t want to work with each other. I hope things have changed since then.

Mr. Little: There are at least five standards off the top of my head. There are two Bluetooth standards, ANT has a blood glucose profile, USB has personal health device class. All of these cover blood glucose meters. This is where part of problem lies. If I’m a large diagnostic company, I don’t know what standard to pick. They make this investment and go through FDA and it’s a hard call as to what you do. We’ve talked to them over and over and there is a lot of waffling back and forth. I think if you look across the board, standardization in other settings has been good for all involved. There will be pain and consolidation to start. And companies are worried that they will become a commodity and not a propriety system, making it harder to make money.

Mr. Grindstaff: There is a gentleman in the back who has a remote mouse, and he can unplug it from his computer and plug it in to another computer and it will work. We’ve each probably worked with dozens of interoperable systems this morning alone. That’s the model. A lack of interoperability puts enormous impediments on progress.

Mr. Hatem: Vendors would find it helpful if there were standards that apply to different use cases. Vendors could select one out of five standards. Certain countries have organizations that deal with this, such as Canada and the UK’s NHS. These entities will help us drive that.

Comment: I think there is a hierarchy of problems. Data and technology can alert us to a problem, but we cannot expect this to create a resolution plan. I’m thinking of book by John Pickup, The Brittle Diabetic. Basically, there is no such thing as brittle diabetes, but there are many brittle people with diabetes. It’s a complex problem. I think if we can have an alert/alarm system, which I think is very doable, that would be state of the art.

Q: What efforts do you envision on the problem of drug adherence? We have devices with dose memory and technology being applied to tracking pill consumption. They are kind of clunky and adherence is at 40-50%. I was amazed yesterday to hear that there is an application out there (FarmVille) where you can use real money to buy seeds to plant on a virtual farm. If we can achieve behavior change through this program, we should be able to get patients to improve adherence, leading to cost saving and improved health outcomes.

Mr. Grindstaff: You’ve really hit on a valuable point. It is not just about a box that dispenses pills. It is above all about behavior. Patients have to want to comply. To improve adherence, we have to drive awareness in the right way. We have to understand how patients behave and how we can deliver the right rewards in the most effective manner. These can be pictures of pictures of a grandson or a sailboat, etc. Every patient is unique, and we need to address that.

Mr. Little: Unless someone is intrinsically motivated to treat himself or herself, it is not going to matter. You can have someone test their blood sugar often, but if they continually eat doughnuts, it doesn’t matter. If you implement a system in which someone is watching on the backend, like a family member of physician, and reinforcing that particular behaviors are bad for a patient and that they want them to be around in the next 10 years to see their grandkids, etc., then we will be effective. We need to identify that intrinsic motivation. It is something we tend to overlook. We have to figure this out for different cohorts of patients. This will help out improve outcomes.

Mr. Grindstaff: The concept of delayed consequences is the major challenge here. The consequences may be years away. One of the things we try to do is to shorten the time between the action and the perception of the consequence. When we shorten it to a 72- or 24-hour cycle, they can see the effects of their actions more clearly. The shorter the interval, the more likely they will adapt.

Mr. Hatem: Let’s start collecting data on what is working now to increase patient adherence and what is not working. I think technology can be very helpful on this side of the equation.

Dr. Klonoff: I like the idea of shortening the interval between an action and the consequence. This is especially important for those with type 2 diabetes and going on insulin.

Mr. Little: There’s a book called Not Dead Yet by Phil Southerland. He’s a professional cyclist and one of the youngest people diagnosed with type 1 in the US. He has an A1c of 5.1%. His motivation since he was four years old was his doctor telling him he would go blind if he didn’t control his diabetes. That’s his single motivation to keep it under control. In his case, that’s what it was. For me personally, I’ve always been a runner. I was running 17-minute 5K’s. I was in the low 14 minutes in college, and this was really frustrating. My doctor said, “You might improve if you controlled your blood glucose better.” I controlled it better for two weeks and knocked a minute and a half off my time. I think it’s about finding this motivation.


Poster Presentations


M Matson, L Georgopoulos, S Arnold, W Kramer, L Shi, P Strange

This double-blind, placebo-controlled phase 1/2a study examined the safety, tolerability, pharmacokinetic, and pharmacodynamic profile of PhaseBio’s once weekly GLP-1 agonist Glymera in people with type 2 diabetes (n=28). In Part A of this two-part study (the data from which was presented in this poster), participants were randomized 1:3 to receive a singe dose of placebo or Glymera (0.10, 0.30, 0.90, 1.35, or 2.00 mg/kg). Over a 28-day period, Glymera was found to be well tolerated and safe. No dose-related trends in type or severity of adverse events were reported. Two participants treated with Glymera reported mild nausea, although at exposure levels two-to-three times higher than necessary to elicit a significant pharmacodynamic effect. The pharmacokinetic analysis demonstrated a slow and flat absorption, long half-life (not specified), proportionality between area under the cure and maximum concentration, and no relationship between clearance and body weight or BMI. To the investigators, this data suggested that Glymera was amenable to a fixed and weekly dosing regimen. Glymera also demonstrated a dose-dependent lowering effect on both fasting plasma glucose (-0.53 mg/dl [placebo corrected] for the 0.90 mg/dl dose at close to maximal effect) and liquid mixed meal area under the curve glucose. Emax modeling determined the effective dose (ED50) to be 0.35 mg/kg. Results from Part B of this study (a four-week ascending dose study) are expected to report in 4Q11.

  • Glymera is a 636 amino acid polypeptide of GLP-1 genetically fused to a physiologically inert repeating elastin like peptide (ELP) biopolymer. The individual subunit of the ELP polymer is a five amino acid motif (VPGXG), where X can be any amino acid. ELP biopolymers reportedly improve the solubility and bioavailability of attached peptides while allowing the peptides to maintain a similar activity profile as the native peptide. Modifying the amino acid sequence of the individual biopolymer subunits as well as the biopolymer’s length allows for the optimization of the physical and chemical properties of the ELP-fusion protein. In this way, Glymera was engineered to form a compact and highly ordered hydrogen bonded structure through the exclusion of its water shell as it moves from room temperature to body temperature upon subcutaneous injection. Studies conducted by the company suggest that the formation of this highly ordered structure results in the controlled and slow absorption of the drug as well its long half-life (not specified) as it dissipates from the site of administration and the process of hydrogen bonding reverses.
  • The phase 1/2a study was conducted in two parts. Part A was a single ascending dose study, and Part B was a four-week ascending dose study using once-weekly dosing. Results from only Part A were presented in this poster. Participants in Part A (n=24) treated with one or two oral antidiabetic drugs discontinued their use during a two-week minimum run in period. Participants were randomized 1:3 to receive placebo or Glymera (0.10 mg/kg [n=3], 0.30 mg/kg [n=3], 0.90 mg/kg [p=6], 1.35 mg/kg [n=3], and 2.00 mg/kg [n=3]). Following a baseline mixed meal tolerance test, participants were dosed. Pharmacokinetic, pharmacodynamic, tolerability, and safety data were collected on participants for 28 days post-dosing.

-- by Adam Brown, Ben Kozak, Joseph Shivers, and Kelly Close