Advanced Technologies & Treatments for Diabetes (ATTD) – 5th Annual Conference

February 8-11, 2012; Barcelona, Spain Day #1 Commentary

Executive Highlights

Hola from a very frigid Brrrrrrrrrr-celona and the first day of the 5th Annual Advanced Technologies and Treatments for Diabetes (ATTD) conference. This year’s meeting has 1,656 participants from 78 countries, up quite substantially from 1,400 in 2011 and 915 in 2010. An all-day AP@home pre- conference meeting was the major focus of the day today, with Roche and Medtronic workshops to kick off the official conference and an evening keynote session from Dr. John Pickup to close out the day.

The AP@home meeting included roughly 110 academic and industry attendees that were engaged and excited to take part in the day’s closed-loop discourse. An update from Dr. Boris Kovatchev was quite noteworthy – new data from now six patients participating in outpatient closed-loop suggest no difference in time in range vs. open loop and no episodes of hypoglycemia. We believe the positive data will continue to mount as the group works to demonstrate the feasibility and safety of the system. Also valuable was an update from Dr. Steven Russell on the bi-hormonal work at the renowned Mass General – the team’s ambitious five-day outpatient closed-loop study will begin this summer using an iPhone 4S and two Tandem insulin pumps.

A full session on single-port AP approaches (combining insulin infusion and CGM sensing into a single site – a boon for patients) was a unique topic in the AP@home meeting that we had not previously seen or heard much about. While the data is still very early stage, it’s a sign of what we might expect down the road. The meeting closed with an excellent panel discussion, headlined by an outspoken Dr. Aaron Kowalski (JDRF, New York, NY). We left with no doubt that the AP has made immense progress in the past few years, but there are also important R&D hurdles and regulatory/reimbursement questions left to answer.

The ATTD opening session showcased the witty Dr. John Pickup, who discussed diabetes technology over the next ten years. His wish list included cost-effectiveness, optical glucose sensing, mobile healthcare, patch pumps, closed-loop control-to-range glucose control, and nanotechnology. The authentic Spanish flamenco dancing and guitar performance that followed was a major highlight – we always appreciate the local flavor at ATTD opening ceremonies! (There was an opera singer last year in London.)

We also took an early tour through the exhibit hall and picked up a few product updates:

C8 MediSensors: The company anticipates CE Mark in 1H12, with launch to occur immediately thereafter; data will be presented at EASD.

Cellnovo: Launch in the UK is expected in April/May timeframe; CGM integration is expected by the end of the year; the handheld was available for demo in the exhibition hall and is using OneTouch Vita test strips (not Verio strips as we previously reported at EASD).

Sanofi: The iBGStar will launch in the US “very soon.”

Ypsomed: Had planned to launch the second-generation OmniPod in April 2012, but due to manufacturing backup it will now be later in the spring or in the summer.



Detailed Discussion and Commentary

AP@home: Two-Port Approach


Hans de Vries, MD, PhD (Academic Medical Center, University of Amsterdam, Netherlands)

Dr. de Vries opened the session with an overview of the AP@home project. AP@home is an EU-funded collaboration of 12 companies and research institutions from around Europe aimed at making the artificial pancreas commercially available. The project hopes to benefit both patients and European information and communication technology/device industries and is currently pursuing two approaches: “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. Dr. de Vries indicated that as the two-port systems are moving toward the commercial phase, single-port systems are entering clinical research – all to be discussed in detail during the day’s sessions.


Andrea Facchinetti, PhD (University of Padova, Padova, Italy)

Dr. Facchinetti is using sophisticated algorithms to improve the output of a CGM sensor in three main areas: precision, accuracy, and the prediction of hypo/hyper events. He treats the CGM as a ‘black box’ (meaning he starts with the output from commercially available sensors). Having collected data from 12 patients over a week’s wear, he has tested his algorithms and shown that noise can be reduced by 50%, accuracy improved 23% and he can gain 15 minutes of prediction time before an event (with less than 25% of false alerts). (Of course, prediction is less useful in an MPC artificial pancreas since the algorithm is already performing the prediction.) Dr. Facchinetti disclosed that the Padova team is working with Dexcom – he believes that by getting inside the black box, he will be further able to improve the quality of the CGM output.

  • The output of conventional CGM sensors could be improved in terms of precision, accuracy, and prevention of hypoglycemic/hyperglycemic events. Noise in the sensor outputs affects precision. Since CGM measures glucose in the interstitial fluid, diffusion of glucose from blood to the interstitial fluid affects accuracy, as does sub-optimal calibration. Current CGM alarms often don’t give enough notice to avoid time spent in hypoglycemia or hyperglycemia. A better CGM sensor would forecast an upcoming hypoglycemia event in enough time to avoid it.
  • This work created a set of algorithms to process the CGM output to create a ‘smart’ CGM. The algorithm consists of three modules in a cascade. These are denoising (precision), enhancement (accuracy), and prediction (avoidance of events). The denoising algorithm uses a Bayesian framework (a priori knowledge that ‘true’ CGM is a regular biological signal, and that the noise component is uncorrelated). It doesn’t require external parameters and copes with variations of noise during monitoring. The enhancement algorithm selects a portion of CGM data between two fingerstick calibrations and uses a two compartment model (of blood and interstitial fluid) to reconstruct the blood glucose. The prediction algorithm is an auto-regressive model, with a thirty-minute prediction horizon.
  • The results from the algorithms represented a noticeable improvement over the output of the current CGM output. To develop and test the smart sensor algorithms, datawas collected from 12 patients with type 1 diabetes at four sites, over seven days, wearing two sensors. The denoising algorithm reduced noise by 50%. The enhancement algorithm showed a 23% improvement in accuracy. The prediction algorithm had a gain in detection time of 14 minutes for hypoglycemia and 17 minutes for hyperglycemia, with 93-96% of events predicted and ~25% false alerts. (Note that the prediction algorithm is less useful for an artificial pancreas, since an MPC algorithm already performs this prediction.) In the future, he believes that adding meal information will enable prediction of 100% of events with false alerts <7%. CGM and insulin data can also be used in combination to detect potential failures of either the CGM or the pump and generate an alert.

Questions and Answers

Q: Do you really use a meal announcement or a specification of number of grams of carbohydrate?

Dr. Facchinetti: It’s actually an estimate of the number of carbs. But we did a sensitivity analysis and we can show that the key factor is the announcement of a meal, regardless of the exact number of carbs.

Dr. Aaron Kowalski (JDRF, New York, NY): You treat the sensor as a black box – could you do better if you didn’t?

Dr. Facchinetti: We are collaborating with Dexcom to improve the Dexcom sensor. If we can work inside the sensor, we can further improve the performance.



Eric Renard, MD, PhD (University of Montpellier, Montpellier, France)

Dr. Renard presented preliminary data from the CAT Trial, which directly compared the Cambridge team’s MPC algorithm to the Padova/Pavia/UVA (iAP) MPC algorithm in the same patients with the same devices (Insulet OmniPod and Dexcom Seven Plus). Eight patients at six centers took part in the study. Each patient underwent 24 hours of closed loop with the Cambridge algorithm, 24 hours of closed loop with the iAP algorithm, and then 24 hours of open-loop therapy. Both algorithms performed similarly and were not statistically significantly different from open loop therapy for time in range (around 60% in all arms for 70-145 mg/dl [basal periods] and 70-180 mg/dl [postprandial]). However, there was a threefold reduction in hypoglycemia to around 2% for both the Cambridge and iAP algorithms. Dr. Renard emphasized that the algorithm was tuned quite conservatively, but the effective prevention of hypoglycemia bodes well as the AP moves to the home environment. We were surprised that both MPC algorithms did not offer time-in-range improvements over open loop therapy; Dr. Kowalski and Dr. Tom Peyser’s comments during Q&A pointed out that potential patient confounds that may have obscured the benefits of closed loop.

  • The CAT trial compared closed-loop control using MPC algorithms from Cambridge and Padova/Pavia/Montpellier (iAP). Patients used the Insulet OmniPod, a Dexcom Seven Plus, and either the UCSB artificial pancreas system or manual validation from a nurse. Each patients underwent three life conditions (rest, meal, exercise) over three 24-hour periods (open- loop, closed-loop with Cambridge algorithm, closed-loop with iAP algorithm). Six centers took part in the study, with eight patients per center participating. Dr. Renard noted that the Insulet/Dexcom system was chosen because Abbott’s commitment to artificial pancreas development was uncertain.
  • The study included meal, rest, and exercise periods with time in range as a primary endpoint. Participants entered in the evening, ate dinner, spent the night, had breakfast and lunch the following day, and concluded the final afternoon with an individualized exercise session. Target range was defined as 3.9-8 mmol/l (70-145 mg/dl) in the basal periods and more than three hours following a meal and 3.9-10 mmol/l (70-180 mg/dl) in the first three hours after a meal. Only preliminary results are available now; all intent-to-treat data should be available by ADA 2012, with per-protocol data (e.g., excluding when the CGM was defective) available by EASD 2012.
  • Both the Cambridge and iAP algorithms were not significantly different from open loop therapy for time in range, but were significantly better on hypoglycemia. Time in range was around 60% in all three arms of the study. However, closed-loop control with both algorithms had a significant benefit for hypoglycemia, offering a three-fold reduction over open- loop therapy. Dr. Renard noted that the algorithm was quite conservative and prioritized avoidance of hypoglycemia. Of course, this is the side to err on as the AP moves to the home environment.
  • The study revealed several strengths and weaknesses of the AP model.
Strengths Weaknesses
The two AP algorithms performed similarly.

The algorithms did not increase time in range.


There were no major deviations (e.g., severe hypoglycemia or ketosis, even when the AP system crashed).

There was too conservative tuning of the closed- loop algorithms.

There was effective prevention of hypoglycemia.                   The stability of sensor accuracy is improvable.
Automation works. Insulin infusion may cause issues in some cases.

Questions and Answers

Q: I’m wondering about your selection of time in range as a primary outcome measure. We’ve talked to third party payers in the US and they have more of a predisposition for a reduction in A1c. How do you bridge that?

Dr. Renard: When you enter the world of closed-loop control, you cannot choose the same tools. A1c is important, but you cannot evaluate such studies on A1c because they’re only a few days long. The time scale of treatment adaptation is not the same in closed loop and overall diabetes treatment. I agree this is a new view of diabetes. Until now, we’ve taken a long view of diabetes, A1c, which foresees ten years down the road. When you use closed loop control, it’s minute by minute. Clinicians and patients know what good diabetes control means having a good mean and staying in range. This is even more important when an automated system is managing diabetes at home. You need to avoid the wide deviations and hopefully, the small deviations too. We are entering a new era in evaluating diabetes – A1c is the big scope, and time in range is the minute-by-minute view.

Dr. Aaron Kowalski (JDRF, New York, NY): I’m not surprised that time in hyperglycemia was not changed with closed loop. In the artificial setting of the CRC, you have motivated open-loop patients. They will control hyperglycemia as much as possible. But they also have a tendency to overdose insulin. My question: are we hitting a limit on the algorithm side with hypoglycemia? Maybe we need to look more at insulin kinetics?

Dr. Lalo Magni (University of Pavia, Pavia, Italy): When we have more data and more days of closed loop control, we will be able to do more. The main difficulty in closed-loop control, is that we start with very little information about the patient. We only start with some marker of clinical information. That is very important to go at home. Safety is the goal on hypoglycemia. Then, when we have data, we can improve by using good models and adding constraints.

Q: Can we do better on hypoglycemia?

A (University of Cambridge, UK): There are quite a lot of hypos that are related to the meal bolus. It’s not necessarily the closed loop delivery.

Dr. Renard: The algorithm is a safety belt. They have less hypoglycemia.

Dr. Tom Peyser (Dexcom, San Diego, CA): Was there a difference in baseline A1c between patients who had an improvement with open loop therapy vs. closed loop therapy?

Dr. Renard: I don’t know yet. The general scope of A1c was good and the entering patients were quite good people. They were well controlled and it was tough to improve. The open loop control was around 60% time in range. The average patient is around 40%. It’s hard to do better than that.

Dr. Peyser: Maybe that’s an unrepresentative result.

Q: You have a large data set compared to other studies. Given the size of the data set, do you plan to publish information about defaulting to a backup sensor or when you had data loss?

Dr. Renard: All the data will be detailed. The study was only 24 hours. The sensor was set with little time before the start of closed-loop induction. As you know, the time needed to stabilize the signal is important. Most of us with experience in two-day AP studies know that the second day is always better than the first day. That’s in part related to the sensor. There’s also a carryover effect of the first period of closed loop. It may influence it that it was only 24 hours.



Martin Ellmerer, PhD (Medical University Graz, Graz, Austria)

Dr. Ellmerer outlined the AP@home Project’s plan to move from clinical research center (CRC) to home- use studies. He explained that researchers must analyze CRC data on individual “critical situations” in order to extrapolate the consequences if the same events had occurred in the home environment. He illustrated this process by presenting individual patient data from the University of Graz site of the CAT study. (The data on sensor/pump errors generally looked encouraging, though of course it’s impossible to generalize from this presentation to CAT as a whole.) Dr. Ellmerer emphasized the importance of further CRC studies that push the limits of current systems, especially to assess how algorithms deal with errors in the sensor and/or pump. Meanwhile, researchers have already begun single-center studies in which patients transition from the CRC to outpatient setting, and a multicenter home-use trial is slated to begin in 2013. We hope that AP researchers worldwide adopt a consistent system for assessing home-setting risks based on CRC data. Of course we are even more hopeful for the time when home-use AP studies become the norm, so that this process of risk extrapolation becomes obsolete!

  • Dr. Ellmerer believes that previous research experiences should be considered in designing and evaluating new trials; fittingly, he began his talk with a brief look at the previous artificial pancreas devices. In 1990, a company called AVL Graz collaborated with the University of Technology Graz to develop a miniaturized version of the Biostator (18 x 9 x 6 cm; 300 g [7 x 3.5 x 2.4 in; 10.5 oz). The system’s subcutaneous glucose monitoring was described in Diabetes Care (Trajanoski et al., 1997), though it was never actually studied inclosed-loop experiments. Dr. Ellmerer also highlighted efforts from the Advanced Insulin Infusion Using a Control Loop (ADICOL) project, conducted from 2000 to 2002. The personal digital assistant (PDA)-based system was one of the first to combine subcutaneous insulin delivery and subcutaneous glucose sensing for semi-closed-loop control (Hovorka et al., Diabetes Technol & Ther 2004).
  • To overcome regulatory hurdles, researchers must analyze critical situations from clinical research center (CRC) data, projecting what would have happened to the same individual patients had they been in the home environment. To illustrate this sort of analysis, Dr. Ellmerer presented select data from patients enrolled at the University of Graz site of the CAT study. None of the four examples he showed would have led to adverse events at home, though of course it’s impossible to generalize from this – Dr. Ellmerer was simply outlining the process that will ultimately be applied to all the CAT data. We are curious to hear more on how EU and US regulators will formally consider this sort of analysis when evaluating risk-benefit ratio. We also hope to hear more on specific study-design changes that researchers might take to address any problem areas (in addition, of course, to continuing to develop better sensors, smarter algorithms, and faster insulin).
    • Few sensor- and pump-related critical situations arose at the Graz site of the CAT study, and most of these did not require intervention. Five falsely high readings were detected and addressed by carbohydrate administration, while the few false low readings did not require outside intervention. One patient experienced a loss of sensor data for less than two hours (no intervention required), while loss of sensor data for greater than two hours occurred twice (requiring sensor replacement). As for pump- related critical situations, Dr. Ellmerer said that the study included one undetected occlusion and one leaking infusion site; both cases were addressed by replacing the OmniPod.
    • In each of the four specific critical situations that Dr. Ellmerer described from the Graz site of CAT, no adverse events would have occurred even if the patient had been in the home environment. In the first event, the CGM sensor read falsely high and then, after insulin was delivered, switched to reading falsely low. However, the patient took a confirmatory fingerstick and thus did not attempt to correct the false hypoglycemia. In the second event, the CGM read falsely high (leading to overcorrection) but then returned to reading accurately, so that the ensuing hypoglycemia was detected and addressed. In the third event, a patient in the open-loop portion of the study experienced a pump occlusion – the sort of situation that, unfortunately, already occurs in the real world (and can be readily addressed by patients). Finally, Dr. Ellmerer described a temporary loss of sensor data during which the patient had to use SMBG data to steer the algorithm. The researchers placed a second sensor, but while it was warming up the original sensor came back online; during this interval glucose control remained fairly stable.
  • Dr. Ellmerer closed by outlining other future steps for the AP@home Project. One goal is development of an AP platform that is optimally compatible with a variety of diabetes management hardware (given the wide adoption of the UVA Diabetes Assistant, for example, we think these efforts are fairly well on their way). Another big aim is to conduct feasibility studies ofsensor performance and of closed-loop control in difficult CRC conditions, especially during and after meals. And, of course, research must transition to the home environment: first with single- center feasibility studies in which patients transition from the CRC to home settings (already begun), then with a multicenter clinical study conducted exclusively in the home environment (slated to begin in 2013).
  • In a preview of the next session, Dr. Ellmerer said that even when the AP does come to market, one final goal will still remain: development of a “single-port” system that combines glucose sensing and insulin delivery in the same insertion site. He noted that such approaches will increase patient compliance – a somewhat negative way of framing the single- port advantage, in our view. (Although compliance will doubtless improve with systems that are easier to use, we think that patient quality-of-life is a worthwhile goal in itself – especially if it means a larger fraction of the population would be interested in using an AP).

Questions and Answers

Dr. Aaron Kowalski (JDRF, New York, New York): Risk vs. benefit is a big topic in the US with the FDA. I would urge folks to attend a presentation at this meeting from the Helmsley Foundation’s T1D Exchange: data on standard of care in the US shows that severe hypoglycemia is much more common, and A1c values much higher, than we are reporting in the patients in these studies. The clinical community, industry, and patient representatives need to provide better examples of the current risk-vs.-benefit environment, so that we don’t have an environment where one adverse event will hold back development of the AP.

Dr. Ellmerer: Yes. As was mentioned before, most patients in these studies are very well educated and compliant. If they are the control group, it is hard to show additional benefit. But if we move into the home environment, of course in early studies we can enroll only patients who know what to do in critical situations. It is thus difficult to get the full picture of what is happening at the moment with open-loop control.

Q: The uncertainty of sensor data falls into two big buckets: completely random errors and exaggerated trends. You showed examples of exaggerated high or low signals. Those types of errors can be compensated for as Dr. Facchinetti laid out, whereas truly random errors are more difficult to address. Can you talk a bit more about how uncertain is the measurement you’re getting, as to whether it’s random or exaggerated?

Dr. Ellmerer: That is difficult. Who can help me?

Dr. Heinemann: I can help you by saying we will have coffee in a second.

Dr. Ellmerer: Thank you very much, Lutz.


AP@home: Single-Port Approach


Werner Regittnig, MD (Medical University of Graz, Graz, Austria)

Dr. Regittnig provided an overview of the single-port approach, presenting two proof-of-concept studies in support of the integration of insulin delivery and glucose sensing into a single site. Given the high concentrations of insulin at the infusion site, Dr. Regittnig hypothesized that a maximal and constant arterial-to-interstitial fluid glucose gradient would be established, allowing for accurate sensing. Both proof-of-concept studies – the first in healthy subjects examining basal infusion rates and the second in type 1 diabetes patients mimicking an insulin bolus – confirmed this constant gradient with strong accuracy, justifying the single-port approach. Dr. Regittnig suggested the approach could both reduce the number of needle sticks necessary and the size of the treatment system, increasing patient convenience.

  • The single-port approach attempts to integrate insulin infusion and glucose sensing into a single site. Given the high concentrations of insulin at the infusion site, Dr. Regittnig indicated that insulin action would be saturated, providing maximal and constant glucose uptake. He hypothesized that this would establish a maximal and constant arterial-to-interstitial fluid gradient as well, allowing for accurate glucose sensing at the infusion site.
  • The first proof-of-concept study for the single-port approach applied two catheters for basal insulin delivery and interstitial fluid extraction at the same site in five healthy subjects. Insulin was delivered for six hours at three different infusion rates, with plasma glucose also monitored using blood sampling. Results indicated a decline in tissue glucose during the first 60 minutes of infusion, after which levels stabilized; as hypothesized, glucose concentrations in the tissue paralleled those seen in the blood at a constant gradient despite various infusion rates, potentially allowing for estimation of the blood glucose concentrations from measurement at the site of delivery.
  • The second proof-of-concept study (first presented at EASD 2009 – see the October 19, 2009 Closer Look) performed tissue glucose sampling and insulin delivery at a single site in 10 type 1 diabetes patients. Insulin was infused at rate to maintain euglycemic levels through the night and following a glucose bolus. Similar to the first proof-of-concept study, results confirmed that the tissue glucose concentration closely reflected the blood glucose concentration at a constant gradient – Dr. Regittnig indicated that 99.6% of data points were in Zones A & B, with only one data point falling outside the clinically acceptable range.

Questions and Answers

Q: How far apart were the sensing and the infusion sites? Can you generalize results to them being at the same site?

Dr. Regittnig: In our experiments, we didn’t use continuous glucose sensors. We extracted the interstitial fluid from the infusion site directly, and we analyzed it with laboratory instruments. So it was exactly at the site of insulin infusion.

Dr. Hans DeVries (Academic Medical Center, University of Amsterdam, Netherlands): The slide you showed with interstitial fluid glucose concentrations – I understand those are not actual but calculated.

Dr. Regittnig: It’s a little bit more complicated with these measurements, because there’s also mixing with insulin solution and interstitial fluid. We determined the mixing degree by using the ionic reference method.

Q: What is the time lag between how long it takes to get the interstitial fluid into the analyzer? Did you correct for this?

Dr. Regittnig: We considered this delay. We were sampling so that the time it takes for the fluid to get from the catheter to the sampling vial was considered.

Q: Can you comment on your expectations for delivery rates greater than 1U per hour?

Dr. Regittnig: In the first study we used basal infusion rates. In the second study we also used an insulin bolus, and we were also able to show that during this bolus time and shortly after the tissue glucose concentration was still reflective of the blood glucose concentration.

Q: Even with fairly large infusion rates?

Dr. Regittnig: The mean bolus size was 6U.

Dr. Howard Zisser (Sansum Diabetes Research Institute, Santa Barbara, CA): Did you measure the insulin in the dialysate?

Dr. Regittnig: The concentration when you consider this mixing is similar to the infusate concentration.

Q: Did you measure the glucose and what was going on with the gradient when the site was in for more than 48 hours?

Dr. Regittnig: We did these studies over 24 hours, which was the longest time we performed this. We did not do experiments over 24 hours.



Franck Robin, PhD (Sensile Medical AG, Hägendorf , Switzerland)

Most artificial pancreas systems are ‘two port’ – they have two skin perforations for insulin infusion and glucose measurement. The ‘single port’ approach is to combine insulin and glucose at the same site with only one skin perforation and one device on the body. Sensile Medical AG is developing a glucose- responsive cannula that delivers insulin more or less quickly depending on interstitial glucose concentration. By measuring the delivery time for an insulin micro-bolus, glucose levels can be inferred every 3-10 minutes. This elegant approach has the safety benefit that insulin won’t flow at very low glucose levels and that a failure in the fiber can be immediately detected. No clinical results were presented at this stage, but a lively Q&A session demonstrated considerable interest in this intriguing development project.

  • Research work has shown that it is scientifically feasible to have a ‘single port’ approach to the artificial pancreas. ‘Single port’ means only one skin perforation for insulin infusion and glucose measurement.
  • Sensile Medical is developing a single port sensor in which a five-millimeter porous cannula simultaneously delivers insulin and measures glucose, by changing the size of its pores in response to glucose concentration. A pressure transducer measures the time it takes for a small pulse of insulin to flow through the pores, which is a measure of the glucose concentration. This time is normally less than a minute and is not physiologically relevant.
  • The cannula is made of a nano-porous polypropylene fiber coated with a hydrogel based on phenylboronic acid chemistry. The pore size is 100-200 nm, and the coating thickness is a few tenths of a nanometer thick. The cannula is five millimeters in length and placed subcutaneously. It is connected to the pump and the end is closed. When a pulse of insulin is applied, the insulin is forced to flow sideways through the pores of the cannula. The hydrogelswells at low glucose, narrowing the pore size and delaying the insulin delivery. A pressure transducer measures the delivery time, which can then be used to infer glucose levels. A 250 nl micro-bolus takes roughly five seconds to deliver at low glucose and 40-50 seconds at high glucose. For a typical patient, this bolus is delivered every 3-10 minutes, which sets the frequency of the glucose measurement. Calibration procedures that are independent of air bubbles and changes in the size of the bolus have been developed. Tests with insulin lispro have shown no clogging of the nano-pores over time.

Questions and Answers

Dr. Howard Zisser (Sansum Diabetes Institute, Santa Barbara, CA): At low glucose do the pores close completely?

Dr. Robin: Yes

Q: How reversible is the system?

Dr. Robin: The system is highly reversible. The coating is only a few tenths of a nanometer in thickness. So we get a very fast response. It’s not a bulk situation where changes in the gel could take minutes or hours – here, the response is a few seconds or a few tens of seconds.

Q: We are thinking about the regulatory aspects of this technology. Can you discuss leaking or breaking of the cannula?

Dr. Robin: If the cannula leaks or breaks, you get a very fast infusion of insulin and you can detect it immediately.

Q: Are you thinking of using the cannula as a control mechanism for insulin delivery as well?

Dr. Robin: Yes.

Q: How specific is the chemistry to glucose vs. maltose?

Dr. Robin: I am not a chemist, but I understand that there has been some modification to make it very selective to glucose.

Q: How long will the cannula last?

Dr. Robin: In the end, this will be limited by FDA requirements, which are 3 days

Q: What work did you do on the dimensions of the pores?

Dr. Robin: The uncoated core diameter is a tuning parameter that we can use to adjust delivery time from between five seconds (fully open) to 40-50 seconds (fully closed).



Werner Doll (Medical University of Graz, Graz, Austria)

Mr. Doll discussed the Graz approach to a combined CGM sensor/insulin infusion site, which integrates commercially available CGMs (Dexcom, Medtronic, Abbott) into an insulin cannula. He focused on an in vitro experiment that tested the function of the three companies’ sensors when directly exposed to lispro and aspart insulin. Encouragingly, the linearity and accuracy of all three sensors was unaffected by direct exposure to insulin. Mr. Doll concluded that all three commercially available sensors could be used to develop a single port AP. We note that the work is pretty early stage, although animal studies are ongoing and prototypes have been developed for all three sensors. Mr. Doll mentioned that the prototypes showing the best performance will be selected for further development in the AP@home project. No timeline was given, although we look forward to seeing this work hopefully move into humans in the future.

  • The Graz approach to a combined CGM sensor/insulin infusion site includes both ex vivo and in vivo sensor setups using commercially available CGMs. The in vivo sensor model integrates the sensor into the insulin cannula wall, meaning the sensor is in direct contact with both tissue and insulin. For ex vivo sensor placement, the sensor is built into the cannula holder but situated outside the skin. The pump is periodically reversed and the interstitial fluid is sampled and exposed to the sensor.
  • An in vitro experiment tested whether the function of currently available sensors is impaired by exposure to aspart and lispro insulin. The output signals of the Abbott, Dexcom, and Medtronic sensors were tested using insulin containing various glucose and additive concentrations. Three conditions were tested for each sensor: (1) long term exposure over 12 hours to test the sensor’s stability; (2) sequential exposure with different glucose concentrations to test the sensor’s linearity; (3) sequential exposure with different concentrations of the insulin additives phenol and metacresol (both preservatives have chemical structures similar to acetaminophen, which has known interferences with sensors). The sensors were put into a thermo-regulated box to test under stable temperature conditions. A potentiostat was used to measure the raw signal of each sensor.
  • Direct contact of commercially available CGM sensors to rapid acting insulin solutions did not impair accuracy and linearity in vitro. Mr. Doll only showed the results from one sensor for each of the three conditions, though he noted that all three sensors performed similarly in the three tested conditions. In the long-term exposure case over 12 hours, the Abbott sensor had a stable glucose sensor signal using insulin with a glucose concentration of 200 mg/dl. In the second case, the Medtronic sensor signal exhibited good linearity and stability when exposed to insulin aspart containing 100, 50, 200, and 0 mg/dl of glucose (45 minute sequences). Finally, Mr. Doll showed the results from testing the Dexcom sensor with various concentrations of insulin additives (insulin lispro alone, + 3 mg/ml phenol, +6 mg/ml of metacresol) – there was no variation according to the higher concentration of the additives.

Questions and Answers

Dr. Hans de Vries (Academic Medical Center, Amsterdam, Netherlands): Looking ahead, are you thinking an in vivo or ex vivo sensor placement is most promising?

Mr. Doll: It’s too early to go into detail. My personal view is that an in vivo sensor placement is preferable.

Q: What about inflammatory reactions for the in vivo system – fibrinogen or any other reaction?

Mr. Doll: The results presented were in vitro testing. This context did not do in vivo tests. I cannot speak to coating effects or nano fibers.



Giordano Lanzola, PhD (University of Pavia, Pavia, Italy)

Dr. Lanzola discussed the role of telemedicine in closed-loop control – a relatively early-stage but especially fast-moving aspect of AP development. University of Pavia researchers have developed a system that integrates closed-loop control with a web-based server for remote, real-time patient monitoring – useful both for research and, potentially someday, in regular clinical practice. The system has performed with perfect fidelity in early clinical studies, and the next step is to connect to the web- based server through a smartphone rather than a personal computer. (With foresight, Dr. Lanzola and his colleagues chose a PC/web synchronization protocol that is readily compatible with mobile phones – thus the transition to mobile should be straightforward.) Dr. Lanzola closed by discussing several variations of smart-phone-based telemedicine, including the system developed by the University of Virginia and discussed in detail by Drs. Boris Kovatchev and Patrick Keith-Hynes.

  • The Pavia telemedicine system uses a modified version of the UCSB artificial pancreas system (APS) control platform run from a PC, which in turn is synchronizes data with a web-based server. Dr. Lanzola gave a quick tour of this custom web application, which gives researchers an in-depth, real-time look at all the data recorded by the artificial pancreas system (e.g., sensor values, insulin delivery, calibration fingersticks, reference blood glucose, announced meals, announced exercise). The Pavia team deliberately chose a PC/server synchronization protocol that is already used by mobile phones. Thus, adapting the system from PC to smartphone should be straightforward.
  • Dr. Lanzola described favorable results from the early clinical tests of his team’s remote application, which has been implemented at Pavia, Cambridge, and Amsterdam. So far the AP@home researchers have remotely monitored 11 patients in 21 trials, each of which lasted roughly 24 hours (some patients were in multiple trials). Notably, Dr. Lanzola said that so far no data have been lost through the remote monitoring infrastructure.
  • Dr. Lanzola outlined a general model for a telemedicine-integrated closed-loop system, discussing potential modifications to simplify inter-device connectivity. In the basic model, the glucose sensor and insulin pump both communicate wirelessly to a dedicated piece of hardware (embedded device) that runs the AP software. The embedded device sends Bluetooth signals to an Android smartphone (a protocol that the researchers have studied preliminarily in partnership with STMicroelectronics). The smartphone, in turn, syncs with the web-based server using the same protocol already used for the PC-based AP system. Dr. Lanzola noted that many variations of this arrangement are possible. One option is to use a pump and sensor that are already integrated, so that the embedded device needs to connect with only one piece of hardware (we think that pump-CGM integration is becoming standard, as seen in Medtronic’s and Dexcom’s new and upcoming offerings). Another twist would be to integrate the controller algorithm with the pump, which in turn would send Bluetooth signals directly to the server-synched-smartphone. Finally, Dr. Lanzola discussed integration of the control algorithm with the smartphone (the approach being used by University of Virginia researchers, with whom the Pavia team is collaborating).

Questions and Answers

Dr. Eda Cengiz (Yale University, New Haven, CT): There were times our computer crashed mid-study. Is it possible to have two laptops running simultaneously so that one can take over if the other fails? I know it’s not practical, but could it be done for clinical purposes?

Dr. Lanzola: I guess you are envisioning a couple of processors in the platform, so that the backup can take over if the primary processor fails. This is technically a problem on the front end, not the back end.

Steve Lane (Triteq Innovations, Hungerford, United Kingdom): For our artificial pancreas work, we are moving to an embedded platform with dedicated software – no other operating systems like Windows or Linux. The product is being designed as a medical device, so we can take it through regulatory approvals.

Dr. Cengiz: Does it allow reverse signaling? Generally companies don’t want to share their software.

Mr. Lane: We are working with both Roche and Dexcom to be able to look at their communications, but we can see only what they want to give us. We don’t get direct access to their software.


AP: Current Status


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

The stellar duo of Drs. Kovatchev and Keith-Hynes provided an update on the Diabetes Assistant, a mobile phone-based artificial pancreas system. First presented at the Diabetes Technology Meeting in October 2011 (see the November 1, 2011 Closer Look at for a complete breakdown of the system’s features), the Diabetes Assistant runs on an Android-capable cellular phone and is compatible with a variety of insulin pumps, sensors, and control algorithms. Dr. Keith-Hynes also provided an overview of additional elements newly available or in development for the Diabetes Assistant, including a remote monitoring package, an Android-capable wristwatch that communicates with the Assistant via Bluetooth (dubbed the “AP Companion”), a system completely running on Bluetooth, and the incorporation of Insulet/Dexcom’s integrated pump/CGM iDex and Dexcom’s new G5 sensor platform. Early data from the JDRF outpatient closed-loop trial, which employs the Assistant, suggest positive results: summary figures from the first six patients indicate no statistical difference between the percentage of time spent within range (70-180 mg/dl) across the outpatient open-loop (~80%), inpatient closed-loop (~70%), and outpatient closed-loop (~60%) periods of the trial, with no episodes of hypoglycemia (<70 mg/dl) observed during the outpatient closed-loop period. Fresh data (from last week!) from the first patient in the AP@home outpatient closed-loop trial, which also uses the Assistant, showed no episodes of hypoglycemia/hyperglycemia, and 76.2% of readings within the 70-180 mg/dl target range. We look forward to seeing if such results can be replicated further and are eager to see how the Assistant will be incorporated into future research.

  • Following a brief history of artificial pancreas platforms from Dr. Kovatchev, Dr. Keith-Hynes reviewed the components of the Diabetes Assistant. First presented in a hands-on demonstration by Dr. Keith-Hynes at the Diabetes Technology Meeting in October 2011 (see the November 1, 2011 Closer Look at for a complete breakdown of the system’s features), the Diabetes Assistant is a portable artificial pancreas system that runs on an Android- capable cellular phone. The system provides for wireless communication with an insulin pump and CGM as well as the operation of multiple control algorithms. It is intended as a research platform for outpatient artificial pancreas trials, allowing researchers to incorporate their own modules into the interface independently, similar to “apps” on a mobile phone. Given the rapidly evolving nature of consumer electronics, Dr. Keith-Hynes suggested the phone-based system would also allow researchers to take advantage of the latest in communications technology. Indeed, the Assistant is already being used in several major trials, including the JDRF multicenter artificial pancreas trial, the European AP@home trial, the NIH/NIDDK RO1 and DP3 Diabetes Impact projects, and the JDRF multicenter trial of the ambulatory closed-loop.
  • Dr. Keith-Hynes provided an overview of additional elements newly available or in development for the Diabetes Assistant. These include: 1) A remote monitoring package, allowing researchers to collect and monitor data remotely and in real-time; 2) the AP Companion (introduced in 2012), an Android-capable wristwatch that communicates with the Assistant via Bluetooth; this would allow users to monitor the system’s hyperglycemia and hypoglycemia “traffic lights” without needing to take the phone out; and 3) a system completely running on Bluetooth. The incorporation of Insulet/Dexcom’s integrated pump/CGM iDex is also possible, as will be the inclusion of Dexcom’s new G5 sensor platform.
  • Dr. Kovatchev ended the presentation with early data from the first patients of the JDRF and AP@home trials of the outpatient closed-loop. As a reminder, both trials begin with a 12-hour outpatient open-loop period (dinner, overnight), followed by a 10-hour inpatient session when closed-loop is initiated (breakfast, lunch), and ending with an 18-hour fully outpatient period (restaurant dinner, overnight). Both trials utilize the Diabetes Assistant system; the JDRF trial uses a control-to-range algorithm developed at UVA, a communication layer powered by APS, and connection to a Dexcom receiver and an OmniPod pump, while the AP@home trial uses the iDex and the AP@home control algorithm. Eventually, the iDex – which combines the fourth-generation Dexcom sensor with Insulet’s second-generation OmniPod – will be incorporated into the JDRF trial as well.
    • Sessions with the first two patients in the JDRF outpatient closed-loop trial were performed at the end October 2011. Tracings from the 40-hour sessions presented by Dr. Kovatchev suggested reasonably good control, with only two isolated readings over 250 mg/dl in the outpatient closed-loop period in the second patient. Additionally, summary results from the first six patients in the trial indicated no statistical difference between the percent time spent within a range of 70-180 mg/dl across the various periods of the trial (~80% in the open-loop period, ~70% in the inpatient closed-loop period, and ~60% in the outpatient closed-loop period); percent time within range increased to near-complete when the range was widened to 70-250 mg/dl (~85%, ~85%, ~95%, respectively). Notably, no episodes of hypoglycemia (<70 mg/dl) were observed in any patients during the outpatient closed-loop period. Dr. Kovatchev suggested that the average glucose values during the periods (~135, ~160, and ~165 mg/dl), while not great, demonstrated the functionality of the system.
    • The first patient in the AP@home outpatient closed-loop trial initiated early in February 2012 – just days before the meeting. Data from the first patient showed no episodes of hypoglycemia and no readings above 240 mg/dl, with 76.2% of readings within the 70-180 mg/dl target range. Dr. Kovatchev concluded with a short film showing a Diabetes Assistant user strolling a market and riding a bike through the streets of Padova – a nice reminder of the progress made thus far.

Questions and Answers

Q: How is the size of meal boluses determined?

Dr. Kovatchev: You’d have to ask the attending physicians or Claudio Cobelli, because I was not even there. As far as I know the algorithm does adjust the meal bolus.

A: In the last trial the meal bolus is automatically computed based on the information that is given on the knowledge of the standard therapy. The standard therapy is used as a baseline, but the MSPC changes according to the information and the state of the system.

Dr. Aaron Kowalski (JDRF, New York, NY): I’m interested in the Italian study you just showed because it illustrated an overestimation of glucose by the sensor. But there was no hypoglycemia – can you explain that?

Dr. Kovatchev: So it was overestimated by the algorithm, but this was control-to-range. This can deal with errors, as it is not a target but wants to bring the patient back in range. The main difference between range control and target control is the sensor accuracy; we’ll get one day to a sensor that can power target control.



Simone Del Favero, PhD (University of Padova, Italy)

Models are extremely valuable in the development of the artificial pancreas. They are used for simulation (in silico testing), control (MPC), and fault detection (unexpected behaviors). Individualization in all these areas is crucial to cope with intersubject variability. A ‘Holy Grail’ is the development of a personalized control algorithm for the artificial pancreas. Dr. Del Favero described advances in modeling, including a way to upgrade the FDA approved simulator for artificial pancreas testing to better handle the hypoglycemic range. He also described an approach to help artificial pancreas control algorithms to ‘learn’ so that they can more closely model an individual patient and make better predictions.

  • In the artificial pancreas, models are used for simulation (in silico testing), control (MPC), and fault detection (unexpected behaviors). Individualization in all these areas is crucial to cope with intersubject variability.
  • Models for simulation can replace animal testing and can save years. At Padua they created the original ‘maximal model’ simulator which carefully models the GI tract, liver, the glucose system, muscle and adipose tissues, beta cell, and the insulin system. A complex triple tracer experiment in humans identified fluxes for all the sub-systems, which was used to model synthetic subjects. A model for the sub-cutaneous space replaces the beta cell to simulate people with type 1 diabetes. The FDA has approved this in silico model, although it is less effective in describing glucose kinetics in hypoglycemia. A new model presented by Dr. Del Favero has managed to solve this problem. In future, the ambitious goal would be to create a virtual clone of the patient (without having to do triple tracers). The approach is to use Bayesian techniques and to integrate and exploit patient diary information.
  • Models for control don’t have to be as clinically accurate as the models for simulation. “The object is good control, and that’s it”. One goal is to have individualized models, in which the controller learns from the patient. A particular learning approach is a ‘pure math’ black box approach, which doesn’t use any physiological information, just tries to predict glucose, based on the mathematical form of the inputs (meals, insulin) for a particular patient. Dr. Del Favero presented a ‘stable spline’ method, which seemed to have promise in this area.
  • Models can be used to detect the failures of pumps and glucose sensors. Once we have a model and an estimate of its accuracy we can detect failures of devices. Using the model wecreate a future predicted path and a ‘confidence window’ (which gets bigger over time). If the actual data doesn’t fall into the confidence window we can infer that a failure has occurred.

Questions and Answers

Dr. Aaron Kowalski, JDRF New York, USA: – can refinements of the triple tracer help us get to the individualized model?

Dr. Del Favero: Running a triple tracer for an individual patient would be infeasible. So adding any other physiological data would be good, but there is a lot of intra-day variability, which makes things more difficult for us.

Q: How do you start to a brand new patient?

Dr. Del Favero: You can’t start with a brand new patient. You have to tune open loop with a patient during the early stages of diagnosis. Once they have stabilized their open loop treatment, then it will be time to move to closed loop.

Q: How does the model adapt to the changing metabolism of a patient?

Dr. Del Favero: This is a new method – we don’t have an adaptive algorithm right now. But we can take a window.

Dr. Howard Zisser, Sansum Diabetes Institute, Santa Barbara, USA: How many models will we need? An infinite number – one for each patient - or just a handful?

Dr. Del Favero: This is a really interesting question. My hope is that having a truly personalized model could lead to the best improvement, but maybe we don’t need different models for each subject, we could have a kind of quantization.



Moshe Philip, MD (Tel Aviv University, Petah Tikva, Israel)

Dr. Philip reviewed the efforts of the DREAM Consortium, a team of artificial pancreas researchers studying fuzzy logic algorithms. (As a reminder, fuzzy logic AP algorithms are designed to simulate the way that clinicians currently manage basal-bolus therapy – hence the consortium’s name for their software, MD Logic.) Dr. Philip took issue with Dr. Kovatchev’s description of the UVA/Padova study as the first outpatient artificial pancreas study, saying that the title instead belongs to DREAM3 (a multicenter study of closed-loop control at diabetes camps, which began in September 2011, and for which Dr. Philip showed favorable data on average glucose – under 140 mg/dl throughout 24 hours, with a narrow interquartile range – a dream, indeed, for many patients). Keeping up his mock-feud with the MPC-using AP community, Dr. Philip promised that the official presentation of DREAM3 results later in the week would include its own movie!

  • Fuzzy logic is the preferred approach of the DREAM Consortium, which includes researchers from Dr. Philip’s team in Petah Tikva, Dr. Tadej Battelino’s group in Ljubljana, Slovenia, and Dr. Thomas Danne’s group in Hannover, Germany. Dr. Philip noted that the DREAM group’s software (MD Logic) is compatible with multiple pumps (including Medtronic, Roche, and Cellnovo) and CGM systems (including Medtronic and Dexcom) – “for most,” he joked, “we even have permission.” The software is individualized and programmed to learn as it goes, and it is integrated with real-time remote monitoring. (Though the system cannot yet be run from a smart phone, Dr. Philip said that this transition is “very close.”) The consortium achieved good results in its first clinical research center studies, DREAM1 (n=7 young patients with type 1 diabetes, n=12 overnight sessions) and DREAM2 (n=12 young patients with type 1 diabetes, n=15 nights).
  • Final results from the consortium’s study of youth and adolescents at diabetes camps (DREAM3) will be presented later at this year’s ATTD, so Dr. Philip was able to give only a preview. The study enrolled campers in Israel (n=18), Slovenia (n=20), and Germany (n=18), who were studied for a total of 75 patient-nights. One could argue that diabetes camps are somewhat controlled environments by nature and thus not fully outpatient. However, Dr. Philip said that recurrent hypoglycemia tends to be quite common at camps, making for a challenging test of closed-loop control. In an enticing sneak peak at results, Dr. Philip showed that the average glucose value was below 140 mg/dl (with a narrow interquartile range) throughout 24 hours.
  • Dr. Philip briefly mentioned DREAM4, an overnight, home-use study that will enroll youth, adolescents, and young adults (n~150). We hope that the presentation of DREAM3 results also includes a more in-depth preview of DREAM4.



Steven Russell, MD, PhD (Massachusetts General Hospital, Boston, MA)

Dr. Russell gave an overview of his team’s work on bi-hormonal closed-loop control (including Ed Damiano, Firas El-Khatib, and others). He reviewed the team’s two major published trials of bi- hormonal control, emphasizing the system’s effective prevention of hypoglycemia, robustness to sensor and pump failure, the use of only weight for algorithm initialization, and the wide variability observed in insulin pharmacokinetics (a mean lispro Tmax of 70 minutes and a range from 24-166 minutes). He called the latter a “critical determinant of success.” Most notable was Dr. Russell’s discussion of the team’s upcoming portable AP experiments (starting in summer 2012). We last heard about this trial at ADA last summer, so this was a welcome update (see page 59 of our ADA full report at The newest version of the team’s portable bi-hormonal AP will use an iPhone 4S communicating with two Tandem insulin pumps. The team is still evaluating which sensor to use (Dexcom G4, Medtronic Enlite, or Abbott Navigator), though has settled on the Abbott for the time being. Regarding the supply problems with the Navigator, Dr. Russell stated, “We think we found a solution to get around it.” We thought this was quite interesting and wish he had provided details – as a reminder, Abbott has an ongoing clinical of the next-gen Navigator II ( identifier NCT01455064).

  • The team is moving to a portable, wearable system to be used in a trial on the Mass General Hospital campus starting this summer. The five-day hybrid closed-loop study will use an iPhone 4S and two Tandem t:slim insulin pumps (glucagon and insulin). The team is bolusing on the Tandem pumps using an app on the iPhone. We note that these hardware choices are a deviation from Dr. Damiano’s statement at ADA: an iPod touch and two Insulet OmniPods. We think the change to an iPhone is smart given the potential for remote monitoring. The team seems to have settled on the Abbott Navigator CGM for now (see accuracy data below), although they will evaluate the Dexcom G4 and the Enlite to “have more options going forward.” As a reminder from what we reported at ADA, this study will involve five-days of hybrid closed-loop control with free roaming around the Massachusetts General Hospital Campus. Participants will have no set schedule or diet, free access to the gym, point of care blood glucose checks, and a nurse chaperone for safety during the day. At night, patients will sleep in the hospital with reference blood glucose sampling using a GlucoScout. The study will use meal-priming boluses that are adapted based on the pharmacokinetics from previous boluses. Adults will be studied inthis initial study, with further studies planned for 12-17 year olds at a diabetes camp (2013) and newly diagnosed patients (2013-14).
  • The team is also planning a 12-day study of Massachusetts General Hospital staff with type 1 diabetes (2013). There will be no set schedule or diet, normal work, sleeping at home, point of care blood glucose checks, and remote telemetry for safety. This is certainly ambitious and we hope the team can get it through regulatory and IRB approval.
  • “We felt we had no choice but to continue with the Navigator.” A head-to-head CGM comparison convinced the Boston team that the Abbott Navigator is the best of the currently available sensors. The Dexcom Seven Plus recorded a MARD of 16%, compared to 11.8% with the Abbott Navigator, and 20.2% for the Guardian. The team will be testing the fourth generation Dexcom sensor and the Medtronic Enlite to have “more options going forward.” Dr. Russell emphasized that the Dexcom has large periods of signal dropout – while the Navigator recorded 99.8% of possible readings and the Guardian logged 98.5%, the Dexcom came in at 75.9%. We did not catch the specifics of this CGM comparison study, but are following up with Dr. Russell.
  • The team is also planning studies at a diabetes camp (2013) and in newly diagnosed patients (2013-14). The planned study for the diabetes camp will include patients 12-17 years old, two weeks of control (!), a full range of activities, a normal diet, point of care blood glucose checks, a nurse during the day, and remote telemetry (via the iPhone 4S). They are hoping to study up to 15 kids at once. For newly diagnosed patients (over seven years old), the goal is to enroll them in the first three months after diagnosis, and ideally at the time of diagnosis. The current plan is three days of control, no set schedule or diet, a normal routine, point of care blood glucose checks, and remote telemetry.

Questions and Answers

Dr. Hans de Vries (Academic Medical Center, Amsterdam, Netherlands): What about the availability of glucagon?

Dr. Russell: We have an IND exemption from FDA to use Lilly glucagon for up to 24 hours. We then change out the OmniPod at the 24-hour mark. Going forward, we will do the same thing for future studies until we have access to stabilized glucagon. At least four groups have stabilized glucagon formulations. We are hoping to test some of these. We think the most promising approach is using a stabilized formulation of synthetic human glucagon – the regulatory path is shorter. The data looks promising. We were worried about that being a limiting factor, but with the support of JDRF, we don’t think it is. We’re very optimistic.

Q: How early do you give the meal-time priming bolus?

Dr. Russell: At the time the meal is placed in front of the person. We give a fraction of insulin for the meal to get around some of the disadvantages of subcutaneous administration. In our new studies, the pre meal priming bolus is set to adapt over time. By the second meal, the algorithm will start adapting. We are targeting 50% of the insulin for a meal. The ultimate plan is no carb counting and patients would choose a small, medium, or large meal. The system would learn what you mean if you are consistent. That’s a less demanding level of granularity. We think if we’re only targeting 50% of insulin for a meal, it’s okay.

Q: Couldn’t you take a meal bolus before the meal is presented?

Dr. Russell: Yes, patients have that option. The algorithm won’t know any better if they choose to do that. If you pre-dose too much, you will get hypoglycemia before the food is absorbed. This is a problem for those with slower gastric emptying, etc. For achieving mean levels of blood glucose to prevent complications, I think a level of 150 mg/dl is good. I’d rather be at that than lower. If you look at the DCCT data, it’s pretty flat for complications for those under 6.5% or 7%.

Dr. Aaron Kowalski (JDRF, New York, NY): What is prediction horizon for the glucagon PD dosing?

Dr. Russell: For that earlier study, we didn’t have glucagon dosing before blood glucose hit 120 mg/dl. In our newer algorithm, we can get glucagon dosing if the blood glucose is as high as 180 mg/dl, if the derivative is high enough. In the current experiment, we’re seeing substantially less hypoglycemia.

Dr. Boris Kovatchev (University of Virginia, Charlottesville, VA): Can you talk about the pre-meal priming bolus for the individual that was insulin sensitive?

Dr. Russell: In this experiment, we gave 0.05 units per kilogram. For this guy, it was more insulin than he needed and the meal priming bolus was too big. The control algorithm was giving almost no additional insulin. We needed glucagon to prevent hypoglycemia, but we were only giving 470 micrograms of glucagon per day. We weren’t getting glucagon levels above the normal range. In our current experiments, we have adapted the meal priming bolus so that it changes over time.

Dr. Larry Hirsch (Becton Dickenson, Franklin Lakes, NJ): Steve, these are elegant studies. You’ve reinforced something that all of us as clinicians appreciate: the incredible variability of insulin uptake and action. You’re in a controlled research environment, and Tmax varied from 20 odd minutes to almost 3 hours. That was in the same people with the same device in a controlled setting. What do you think is happening in the real world? It must be far worse. Maybe we’re struggling with very sophisticated attempts to derive an AP when a lot of what we could do is to improve the consistency of insulin delivery.

Dr. Russell: We’ve done some thought experiments. We think we would get substantially better control with more rapid insulin absorption. But you cannot take advantage of this unless it is consistent and you tell the algorithm that there is a faster PK. If we do tell the algorithm and we’re wrong, we stack insulin and get hypoglycemia. This happened in our first experiment. If we could make insulin twice as fast and shift the entire curve leftward, that would make a huge difference. The problem would be if it made 90% of people faster and 10% had no benefit. That would be hard to deal with.

Q: For glucagon to work optimally, there must be glycogen in the liver. What about prolonged exercise depleting glycogen?

Dr. Russell: We looked at this. There was no difference in glucagon efficacy before or after exercise. We think it was not enough exercise to deplete the liver of glycogen.


AP: Will We Bring It Home? Promises and Hurdles


Aaron Kowalski, PhD (JDRF, New York, NY)

Dr. Kowalski gave a very useful overview of how JDRF is thinking about artificial pancreas development in the years to come. We liked the smart reorganization of the six-phase development timeline put forth a few years back – Dr. Kowalski broke down AP systems into first generation (low glucose suspend, hypoglycemia minimizer, control-to-range), second generation (control-to-target), and third generation (multi-hormone control) systems. He tactfully showed how JDRF is attacking the gaps in each phase of development, including investment in better sensors, research into insulin and other hormones, and regulatory pressure. JDRF is certainly taking a diversified, multi-pronged approach to help bring an AP to market, but as Dr. Kowalski put it, being careful to “under-promise and over-deliver.”

  • “We’ve vigorously lobbied FDA to accelerate progress on the Medtronic Veo.” Dr. Kowalski emphasized that the FDA’s concerns with the Veo – sensor errors and risk for ketoacidosis – were “completely unfounded.” We look forward to hearing an update on the Veo’s pivotal trial, ASPIRE, on Day #2 of ATTD. Dr. Kowalski noted that since the technology already exists, JDRF is devoting little funding to this part of the AP timeline. The one exception is a bit of funding for research on hypoglycemia unawareness.
  • A hypoglycemia minimizer is “possible today, has very little risk to the patient, and significant potential upside.” Dr. Kowalski noted that from a technological perspective, the field is very, very close to having this system developed. Notably, Dr. Bruce Buckingham and Dr. Peter Chase have just launched outpatient trials of a hypoglycemia minimizer in collaboration with JDRF. JDRF’s ultimate hope is that companies combining sensors and pumps could commercialize such a system.
  • “The technology exists for treat to range,” but “we need to understand how to drive to a commercial embodiment of this system.” Dr. Kowalski mentioned JDRF’s artificial pancreas consortium and work with companies to tackle the development of a treat-to-range system. Specifically, the ongoing partnership with Animas is working on this problem, and JDRF has also funded some research with Medtronic (this is the first we’ve heard of JDRF funding Medtronic and we will be looking into this). Dr. Kowalski emphasized that the treat-to-range approach makes a lot of sense from a safety perspective and should offer clinically meaningful benefits for patients.
  • Looking forward, the AP can be broken down into first generation systems (low glucose suspend, hypoglycemia minimizer, treat-to-range), second generation systems (treat-to-target), and third generation systems (multi-hormone systems). The gaps, opportunities, and research priorities differ for each generation.
    • First generation systems: “We are very, very close.” Dr. Kowalski noted that research is progressing well on this front, but the key will be the clinical and regulatory pathway. In the real world, patients must interact with these devices and the companies must be incentivized to make an investment – obviously easier said then done given the current regulatory environment. Dr. Kowalski emphasized that “until patients can purchase these devices, we’re not there yet.”
    • Second generation systems: “We see gaps and opportunities.” As systems move towards treat-to-target, sensors will need to improve in accuracy, reliability, redundancy (beyond glucose oxidase), and become calibration free. JDRF is actively targeting all of these facets through its sensor initiative in partnership with Helmsley (see below). Algorithm development and improved insulin infusion kinetics are also a focus of the foundation. Regarding the latter, Dr. Kowalski reminded the audience of JDRF’s ongoing partnership with BD. We would be very interested to see clinical data and more information on where this partnership stands though we assume the company is staying quiet on this front for a reason.. As a reminder, BD is working on both infusion sets and intradermal needles.
    • Third generation systems: “A huge, huge opportunity,” but we have “pragmatic issues.” Although groups are working on using glucagon and amylin (Ken Ward and colleagues in Oregon, Steven Russell and colleagues in Boston, the team atUVA), dual chambered pumps will need to be developed. Additionally, a single point of injection and sensing will be an important development (see talks from earlier in the day for the status on this front). For glucagon specifically, the major limitation is the need for a soluble formulation. Dr. Kowalski noted that JDRF has approached Sanofi, Lilly, and Novo Nordisk, but the lack of a current market for soluble glucagon has prevented interest from these large companies. He called it a “hard sell” from a business perspective– this is disappointing but perhaps not all that surprising.
  • JDRF is investing heavily in insulin and other hormones. Dr. Kowalski called insulin a “pressing problem” and noted that current insulins are unable to mimic the quick cephalic response in people without diabetes, leading to postprandial hyperglycemia; they also have a long tail of action, leading to delayed hypoglycemia. JDRF is also “very, very interested” in amylin, which is “almost never talked about and lightly used in the community.” As a reminder, JDRF has an ongoing partnership with Amylin aimed at developing a coformulation of amylin and insulin.
  • The Sensor Development Initiative is a partnership between JDRF and Helmsley (“The most important new entrant into type 1 diabetes.”) The partnership is up to $20 million and will work with researchers and industry to improve all aspects of sensors: reliability, sensitivity, specificity, and error detection, with the constraint of maintaining patient quality of life. Thus far, JDRF has seen a “very good response from the community” and they hope to have funding in place by the end of June.
  • “Gaining regulatory approval will be challenging.” JDRF is investing significantly into the regulatory side of AP development, including frequent meetings with FDA staff. Broadly speaking, Dr. Kowalski believes the FDA’s conservatism has led to an imbalance in the prioritization of risk over efficacy.



Ian Fowler, BSc (Hons) PMP (Triteq Innovations, Hungerford, United Kingdom)

Mr. Fowler reviewed the regulatory requirements for medical devices in the EU and US, broadly laying out the probable regulatory pathway for the artificial pancreas. His company, Triteq Innovations, has established a core compliance plan for the AP@home Project so that AP developers will be well prepared to file the device in Europe. (He noted that although the AP would probably be defined as a Class IIb device in Europe, the plan is designed to meet all Class III requirements to ensure against “gotcha!” moments down the road – a good precaution in our view, even though CE marking tends to be much more readily attainable than FDA approval.) Overall we would have been interested to hear more on artificial-pancreas-specific considerations and challenges. As a side note, as mentioned during Q&A, reimbursement requirements should also be considered from the outset of device development.

Questions and Answers

Comment: To receive the CE mark is one step, but of course reimbursement is important as well, and this varies widely among different countries.

Dr. DeVries: Reimbursement, specifically with regard to CGM, will be discussed in a symposium on Saturday. This is an intriguing issue, and I would say a real difference between Europe and the US.


The Future of AP Development in Europe


Moderators: J.H. DeVries, MD, PhD (Academic Medical Center, University of Amsterdam, Netherlands), Lutz Heinemann, PhD (Profil Institute for Metabolic Research, Neuss, Germany),

Panelists: Henry Anhalt, DO (J&J Animas, West Chester, PA), Claudio Cobelli, PhD (University of Padova, Padova, Italy), Aaron Kowalski, PhD (JDRF, New York, New York)

Questions and Answers

Dr. de Vries: We have significant problems with CGMs and insulin pumps and we’re hoping for improvements. But what about algorithms – can we get better?

Dr. Cobelli: In my opinion, algorithms can really improve the sensor. Dr. Facchinetti’s presentation this morning was convincing – a battery of algorithms can improve precision, accuracy, prediction, and fault detection. I have to give credit to the open minds of people at Dexcom. They have started talking seriously with people in academia. We can publish nice papers in prestigious journals, but now we can have these implemented in real-world devices. The ultimate goal is for diabetic patients to have more accurate technology. On algorithms, I’m MPC biased. Steve is using MPC, we are using MPC, Roman uses MPC, Frank uses MPC. But each of the various MPC strategies is different and each has pros and cons. MPC uses a model to predict the future and select the best infusion rate for the next step. This model is forcing the quality of the prediction. Something we are trying to work on is to improve the model inside the MPC algorithm – to individually tailor the MPC algorithm. We can certainly improve in this area. We will certainly learn a lot with longer clinical trials. Now we have one-day, two-day trials, but it’s not where the advantage of MPC strategies can be seen. If we are able to build a tailored algorithm for two or three weeks, the results will improve dramatically.

Dr. Eric Renard (University of Montpellier, Montpellier, France): We have come to think that remote monitoring is a necessity for clinical trials, at least at the beginning. How do each of you consider remote monitoring – optional? Mandatory? Do you think it will remain a part of the system even when thousands – millions – of patients are using it? Or just be a temporary piece during the regulatory process?

Dr. Kowalski: The proof will be in the pudding. Everyone agrees there will be transitional studies that use it, as have been performed already. I don’t think it’s mandatory, though. Diabetes management is already risky – today’s pumps dose insulin when people are hypoglycemic. To me, it’s about identifying and neutralizing potentially catastrophic risks of artificial pancreas systems. I think remote monitoring could play a role, but I do not think that it is the most important piece. The level of safety that we are layering onto these systems is significantly more than the average person with diabetes is dealing with, even without remote monitoring. Remote monitoring also raises issues about infrastructure, as well as the question of how many patients actually want to have their diabetes and their location constantly tracked. I can’t speak for industry, but I don’t think that it is a must-have from a patient perspective.

Dr. Anhalt: I look at remote monitoring in two senses. First there’s the safety perspective in transitional studies, on the way to fully outpatient model. But then as medicine evolves, we will think about how telemetry applies in regular practice. Will it be passive uploads of blood glucose values? Warnings when the values are out of range? I don’t think we can answer that question today – for now we can only speculate.

Dr. Heinemann: 15 years ago no one had a mobile phone. Probably 10 years from now we will have significantly more computer power in our hands than now; monitoring a million people will not be an issue anymore.

Dr. Kowalski: I have talked to Lane Desborough of Medtronic about this. Access to this type of data could be very, very powerful. Having an Android phone controller that minimizes the burden will be critical, I think. How to convince regulators to treat a consumer product as a medical device will be a tough nut to crack. But I think that massive data collection will start a virtuous circle for research and future algorithm development.

Dr. Heinemann: Of course, development of new systems and approval of that system are very different things.

Dr. DeVries: I think this topic has been well covered.

Dr. Cobelli: Speaking to Lutz’s comment, we probably already have more processing power than we need. To build the system that Boris [Kovatchev] presented earlier was a relatively easy three weeks of work porting the AP algorithm onto a smartphone. The next step is telemedicine. Phone is much more powerful than needed for either of these – we are already there.

Dr. Martin Ellmerer (Medical University Graz, Graz, Austria): We heard three talks about the single port approach – one port for both CGM and insulin. And then I saw it in your talk Aaron, which was positively surprising. But the insulin infusion cannula can only be applied for three days, while sensors are for seven plus days. Is there a misfit in that approach?

Dr. Kowalski: That’s a great question. The perfect future as I see it is a patch pump with a single point of infusion on a phone like Boris presented. This would be minimal burden. The number one thing when I talk to lay people is they hate two sites. But they’re willing to do it for better control. On the mismatch of time, the CGM manufacturers need to work on the warm-up and the instability of the signal early in the sensor life. The data gets better as time goes on. Getting back to the first question, I think one of the weakest links is the pump site. For CGM, there are errors that we know of; everyone knows the different error distribution of the systems. But we have no idea of the variation in insulin infusion. We see a huge variety in inulin kinetics. I think pumps are incredible accurate. But JDRF is investing in this through a major initiative with BD. The point of infusion from a patient perspective is top of mind. One site is a blockbuster home run.

Dr. DeVries: I’d like to come back to the regulatory climate in the US vs. EU. I’ve got some outline of the differences in the approval requirements, but I don’t understand how that translates to having more difficulty conducting clinical trials in the US. Why does one do early-stage research in Europe instead of the US?

Dr. Marc Breton (University of Virginia, Charlottesville, VA): We start applying for studies at the same time, but the EU process tends to go faster. In particular, EU regulators accepted use of the iDex device as long as it was supervised, whereas the FDA had more involved requirements. I think we submitted our latest study protocol to the FDA two weeks ago, so we will probably have our first refusal in about two weeks.

Dr. Kowalski: In the US, we have an investigational device exemption (IDE) process. The FDA has taken a very rigid stance on validation and verification of systems and software. I could agree with these requirements for outpatient studies, but the agency requires such data even for CRC studies. We have been co-funding development of the UVA device. A huge problem with the FDA has been their requirement for full validation of every device used in study. Thus we have been co-funding development of the UVA device in the hopes that using the same master device every time will speed up the process. By contrast, we’ve had no institutional review board (IRB) issues – the FDA has stepped beyond IRB. We also don’t have these problems in the EU right now. JDRF has been lobbying the FDA to provide better guidance for AP development, and the draft document was put out on December 1. We brought together all the key clinical organizations – the American Diabetes Association, the American Association of Clinical Endocrinologists, the Endocrine Society, the American Association of Diabetes Educators. This made it difficult for conservative FDA reviewers to cry “safety,” which is one problem we’ve been having with the agency. The FDA doesn’t like devices that they haven’t fully verified and validated, and when we’re doing research, we rely on these devices.

Dr. Heinemann: I have to address the comment you made about the process being easier in Europe. Things are relatively difficult in Germany, Austria, and the UK. It took us nearly a year to get CAT approved in Germany, with much difficulty arising because they regard software as a class IIb device. EU is not homogeneous in this direction, unfortunately.

Dr. Kowalski: Software is a huge issue in the US as well.

Comment: Dr. Steven Russell (Massachusetts General Hospital, Boston, MA): I think it’s really interesting to explore whether an array infusion site would be more reliable. You see this tremendous intra-subject variability. This might be due to the local infusion site. An array of microneedles would take multiple sites. It might be that the average of multiple sites is more reliable, in the same way that more sensors are more reliable. Ken Ward has shown that when you average three sensors, you get more reliable data. Using a single site, with multiple infusion needles in a small area, might be a solution.

Dr. Kowalski: We need to understand what don’t we know: the infusion site. We know absolutely that people do better on pumps. We’ve all seen the meta-analyses from John Pickup. But this is a chink in the armor that if we can fix, will have benefits for patients.

Dr. Heinemann: You need to think about this from a patient psychology point of view. Keep in mind, even in the JDRF CGM study, many kids weren’t willing to wear the system more than three times per week. Even if we develop an AP, will they wear it every day? The single port, with all of its complexity, is a key component.

Dr. de Vries: Research has shown that absorption of insulin pumps is changing over time.

Dr. Cobelli: When Aaron was saying the bottleneck is insulin infusion, I was thinking it was because of the delay. Nobody talks about the delay and the variability in insulin itself. For designing control algorithm, having delays is one of the most difficult problems. The strategy on the table is to improve the insulin molecule so it works more rapidly. But we still must cope with this incredible delay. It’s 60-70 minutes delay with insulin compared to five or ten minutes with the sensor. One strategy is to use pramlintide. It leads to a delay in the appearance of glucose. This makes it less peaky and can bring us back to a situation that is more normal – glucose and insulin time courses that are more in the same window.

Dr. Bruce Buckingham (Stanford University, Palo Alto, CA): I’ll have a talk on infusion site failure during the meeting and I’ll have data to present. Another thing is a high-fat meal versus a normal meal. It makes a huge difference. The high fat meal is like giving amylin – it’s completely different absorption.

Dr. Russell: You mentioned conservatism from a regulatory perspective, but we’ve seen it from a company standpoint too. We’ve had to change tack several times when devices that were promised weren’t ready or priorities changed. And that may be in part because of the regulatory climate, and in part because the pathway for clinical trials is several years in advance. That does not fit into a quarterly profits. We’ve found that the larger companies have been stodgy about moving forward. It’s the smaller companies, the Dexcoms, the Insulets, that have been more cooperative.

Dr. Anhalt: That’s a very challenging question. It might appear from the outside as evasiveness or backing away. But on the inside, there’s a lot that goes into where resources get allocated. I would say, and I’m not speaking on behalf of J&J, that there’s a commitment to this. We wouldn’t be in this room if there weren’t. One of the greatest things about this meeting is the opportunity for open dialogue. The commitments are there. But things change in terms of prioritization of specific products and the product pipeline. When there is a risk taken for product in a clinical trial that has implications for safety on the company, there are different considerations. And different companies operate differently. It may seem that bigger companies have a more entrenched regulatory approach. It may be because bigger organizations have more experience. But suffice it to say everyone is trying to find the best path forward in any way that they can.

Significant dollars have also gone into investigator-initiated studies. There are paths for investigators to come to companies. Tens of thousands of dollars have been expended for support of AP work. Obviously, we all want to do this safely and effectively. Before it gets into the marketplace, we must do our due diligence.

Dr. Kowalski: We’ve learned to under-promise and over-deliver. We’ve done the opposite on cure-related things. The companies deserve credit here. We’ve had issues with some companies getting out of type 1, but the investments from Insulet and all the studies they supported – they have a business to run. The same goes for Animas, Dexcom, and Medtronic. I’m going to turn this a bit and say a lot is incumbent on clinicians. Look at CGM – it’s an absolutely transformative tool right now. But frankly, we have quite pitiful adoption and pitiful reimbursement. You can have an AP, but if it’s not prescribed and not reimbursed for a company, there is no market. If I’m a company starting an AP program, it’s not going to be prioritized right now. There are lots of early CGM adopters, especially in this room, but when you go back to your countries and organizations, talk to your colleagues. CGM uptake is not as good as it should be from a patient perspective. I hear all the time, “Why won’t my doctor prescribe CGM?” I’ve been on CGM for five and a half years. But there are tons of patients in the US that don’t have reimbursement or access to CGM. The community needs to be pushing and we need to keep the standard of care moving forward. There are still lots or people doing three shots and two fingersticks a day.

Dr. Heinemann: I’d like to close by briefly looking toward the future of AP development in Europe. We are at half time of the AP@home Project. What happens when it ends? Will it continue? As I mentioned, we hope that it does continue, and that we have a product in perhaps 10 years time. On the one hand Europe has a more favorable – depending on your view – regulatory environment. But the economic climate is not that favorable – for example, compared to how much money the NIH gives out.

Dr. Anhalt: The NIH does have a lot of money, but it’s not growing. It’s a fight every year for the budget. Dr. Heinemann: Dr. de Vries, any final remarks?

Dr. de Vries: I thought this was a great day. I really enjoyed hearing all the conversations, not only during

the sessions but also during lunch and coffee breaks. I hope we can convince Moshe Philip that we should have a similar day every year at ATTD.


Opening Ceremony


John Pickup, BM, DPhil (King’s College London School of Medicine, London, UK)

With his characteristic dry wit and gravitas, Dr. Pickup opined on the coming decade of diabetes technology trends. Quoting the American computer scientist Alan Kay, he said, “The best way to predict the future is to invent it.” Thus, he said, his talk should not be viewed as a series of predictions, but rather as a list of inventions to be developed in the next 10 years (“as I go away and retire”). In brief, the “buzzwords” that Dr. Pickup chose to discuss included: cost-effectiveness, optical glucose sensing, mobile healthcare, patch pumps, closed-loop control-to-range glucose control, and nanotechnology.

  • Cost-effectiveness: Dr. Pickup explained that the current system of medical device development involves long delays in establishing clinical efficacy, cost-effectiveness, and finding the most appropriate patient population. For example, he said that pump therapy began in 1976 and took roughly 30 years to reach “clinical maturity.” He added that the current status of continuous glucose monitoring is “not really clinical maturity” despite its having debuted over 10 years ago. To accelerate future development and more quickly find ‘responders’ to therapy, Dr. Pickup recommended earlier use of randomized controlled trial evidence, analyzed by patient data meta-regression. To illustrate this technique, he described his group’s recent meta-analysis of six CGM vs. SMBG trials, which indicated that CGM success is predicted by baseline A1c and sensor wear-time (Pickup et al., BMJ 2011).
  • Optical glucose sensing: Dr. Pickup noted that current continuous glucose monitoring technologies are not accurate or reliable enough. Fluorescence-lifetime-based sensing, however, is less susceptible to electrochemical interference. Dr. Pickup’s group has published encouraging early-stage research on the topic (Saxl et al., Analyst 2011) – for more on this emerging research area, see our coverage of Dr. Pickup’s talk from the 2011 Diabetes Technology Meeting at
  • Mobile healthcare: Dr. Pickup proposed that mobile health could be a solution to the current system of diabetes management, which he said involves “too much information, too many people, and too many devices.” Patch Pumps: Dr. Pickup briefly reviewed some potential benefits of new/upcoming patch pumps and micropumps, each of which features one or more of: small size, integrated cannula, touchscreen remote, and greater suitability for type 2 diabetes patients.
  • Closed-loop insulin delivery: Dr. Pickup forecasted that predictive low-glucose suspend would “soon” enter clinical practice, with predictive low- and high-glucose control-to-range systems to follow. He believes that these sorts of control-to-range systems will become an “important part of clinical practice in the next few years.”
  • Nanomedicine: Dr. Pickup proposed a variety of applications for nanomedicine. One potential use is “smart tattoos” for glucose monitoring – a technique that involves subcutaneous or intradermal injection of fluorescent nano-/microsensors, illumination to excite the sensors’ fluorescence, and non-invasive measurement of the resulting signal. Another application could be nanofilm encapsulation of islet cells, which would protect the cells from autoimmune attack while still allowing flow of oxygen, nutrients, glucose, and insulin. Dr. Pickup also mentioned the use of nanoparticles for drug delivery (e.g., to target weight-loss areas in the brains of obese patients) or for in vivo imaging of complications (e.g., to detect foot ulcers in their early stages).


Company Workshop: Sensing and Delivering in a Closed Loop Environment (Sponsored by Roche Diagnostics)


Sanjoy Dutta, PhD (JDRF, New York, NY)

 Dr.  Dutta provided an overview of the JDRF’s proposed path to the artificial pancreas, highlighting current challenges. While he believed first-generation systems could be delivered with currently available technologies, he suggested second-generation to-target systems would require further scientific and technical developments. Of these developments, he focused particularly on improving sensor accuracy (noting JDRF and the Helmsley Charitable Trust’s recent up to $20 million initiative toward improved sensors) and accelerating insulin action (highlighting numerous products in development, including Halozyme’s analog-PH20, Diasome’s HDV-insulin, InsuLine’s InsuPatch, and Roche’s DiaPort). Moving from the scientific realm, he also underscored JDRF’s work to overcome regulatory hurdles, recapping the three major guidance documents released by the FDA on the development of the artificial pancreas in the past year – while only drafts, he suggested these could be tremendously useful for research and eventual commercialization.

  • While Dr. Dutta believes first-generation systems can be developed and delivered with currently available technologies, second-generation systems will require improved sensors and insulins. Dr. Dutta highlighted the JDRF and the Helmsley Charitable Trust’s recent initiative putting forth up to $20 million for the development of improved sensors able to drive control toward a euglycemic target. He also made note of other priorities, such as ultra-fast insulins, improved delivery methods (e.g., intradermal, inhaled, intraperitoneal), glucose-responsive insulins, hormones beyond insulin (he focused particularly on glucagon, suggesting JDRF was working to accelerate development of a formulation that could be used in the artificial pancreas), and repurposed drugs.
  • Of the challenges facing the closed-loop, Dr. Dutta spent considerable time on improving insulin action. Ongoing projects he reviewed included Halozyme’s analog-PH20 (he highlighted results suggesting 15-20 min earlier onset of action and 45 min shorter duration of action compared to analog alone), Diasome’s HDV-insulin (liver-directed insulin to better mimic physiological delivery – results indicated a 38% in AUC during OGTT), InsuLine’s InsuPatch (a thermal patch that applies controlled heat around the infusion site to speed action; as in last year’s presentation, he showed data demonstrating a 36 min advantage over subcutaneous insulin in the time to maximum concentration), and Roche’s DiaPort. He suggested such improvements and perhaps their combination could provide the insulin action necessary for tighter control with an artificial pancreas system.
  • Moving from the scientific realm, Dr. Dutta also highlighted JDRF’s work to overcome regulatory hurdles. With the artificial pancreas serving as mostly uncharted territory, he emphasized the need for guidelines from FDA and noted the JDRF’s continued pressure on the agency. He also recapped that three major guidance documents were released by the FDA on the development of the artificial pancreas – while only drafts, he suggested these could be tremendously useful for research and eventual commercialization.



Bernhard Gehr, MD (Fachklinik Bad Heilbrunn, Bad Heilbrunn, Germany)

Dr. Gehr provided first impressions of safety and tolerability with the second-generation DiaPort, Roche’s implantable transcutaneous port that enables intraperitoneal insulin delivery. The second- generation system features a polyester felt layer for more stable in-growth and a single, more flexible and wider catheter to prevent adhesions, overgrowth, and occlusions. In the ongoing open-label trial of 12 type 1 diabetes patients with the device initiated in late 2011, early results suggest some improvements in adverse events, with three cases of superficial wound infections, two cases of abdominal pain, no intraperitoneal infections, and seven cases of subtotal catheter occlusions (Dr. Gehr indicated these were potentially due to reactions with contaminations from the manufacturing process and to be addressed by Roche further). As only an early impression, we look forward to seeing if low adverse event rates can be maintained with prolonged use to gauge the DiaPort’s utility for patients outside of its currently limited indication.

  • Dr. Gehr opened with an overview of the DiaPort system, Roche’s implantable transcutaneous port that enables intraperitoneal insulin delivery. (For a complete overview of the features of the DiaPort, see the ATTD 2011 full report in the October 14, 2011 Closer Look.) While intraperitoneal insulin delivery has been shown to potentially better mimic physiological insulin delivery with reduced hypoglycemia and weight gain, the first-generation DiaPort was prone to infections and occlusions. The second-generation system features a polyester felt layer for more stable in-growth and a single, more flexible and wider catheter to prevent adhesions, overgrowth, and occlusions.
  • Dr. Gehr provided first impressions of the second-generation DiaPort as observed in an ongoing clinical trial. The open-label, single-center, single-arm trial initiated 12 patients with type 1 diabetes on the second-generation DiaPort in late 2011 and is expected to be finalized in August 2012. Dr. Gehr reviewed the implantation procedure, which is outpatient and primarily utilizes general anesthesia to minimize pain. In the trial, patients are then transferred under the care of physicians and a nurse for two days loose bed rest and light physical activity; they are discharged on day 5 on thromboprophylactic therapy and education on daily management of wound healing (we also appreciated that a dummy system is available for patients to practice use of the system). Dr. Gehr suggested some recalculation of basal rate management is necessary, though bolus and correction factors remain unchanged (patients must take care with bolus calculators, however).
  • Early responses suggest improvements in adverse events versus the first-generation DiaPort. Though only short-term data is available, Dr. Gehr indicated three cases of superficial wound infections (treated successfully with oral antibiotics), two cases of abdominal pain, no intraperitoneal infections, and seven cases of subtotal catheter occlusions (all treated by flushing with saline). Interestingly, Dr. Gehr suggested occlusions were likely occurring due to insulin crystals formed by reactions of insulin lispro with silicone contaminants in the catheter – when patients were switched to regular insulin (which has no effect on insulin action with intraperitoneal delivery – insulin lispro was chosen for its wide availability), occlusions no longer occurred (Dr. Gehr suggested Roche was looking into the catheter manufacturing process to address this issue further).

Questions and Answers

Q: Could the lispro be reacting with the titanium?

Dr. Gehr: We think lispro may be interacting with silicone. In EM microscopy of the catheter you can see silicone in patches, which could be contamination in the manufacturing process; special care will be there in the future so there’s no silicone in the catheter.

Q: When did you start insulin intraperitoneally?

Dr. Gehr: We started on the first postoperative day, but it can be right away no problem.



Lesley Jordan (Brentwood, United Kingdom)

Type 1 diabetes patient Lesley Jordan provided a patient’s perspective on the first-generation DiaPort. Ms. Jordan initially began on the DiaPort after she was having trouble finding suitable injection sites and control on her pump began to waver. While she suggested her experience was overall beneficial (she liked the improved insulin profile, convenience of the single site, and lack of needles), she did note a number of challenges in her eight-year history with the device, including seven surgeries (two for DiaPort implantations, two for overgrowth, two for salvage, and one for the removal of an insulin crystal – the size of a baseball!), the need to revert to subcutaneous treatment during device difficulties, severe hypoglycemia during infections due to overtreatment, and the need for regular appointments every three months for catheter replacement in a sterile environment. We greatly appreciated Ms. Jordan’s honest perspective – with indicated patients who do stand to benefit from treatment, we hope the second generation will improve upon these limitations.

Questions and Answers

Q: Can you comment on your A1c levels before and after?

Ms. Jordan: I saw some benefit, but I’ve had a lot of trouble over the years. In the recent history there always seems to be something wrong that doesn’t help. When it was agreed I could have the DiaPort, the consultant said my control was exemplary. My best A1c was 6.3% with the DiaPort; at the moment I’m around the near seven’s with difficulties again. Last year I went up to the eight’s because of difficulty and the salvages; but every time I need to revert to subcutaneous the A1c climbs.

Comment: One reason we get unexpected highs is occlusion or kinking. Really the main issue is the material and the insulins we’re using. I cannot help but wonder if that’s the problem here. It makes me think the type we’re using is the wrong one.

Ms. Jordan: That’s way beyond my understanding, so I can’t comment. But I have to use what’s available now.

Comment: I can explain some of the occlusions – most of the occlusions that occur are caused by overgrowth. Many think it’s crystals, but when we look that is not so. When you get it, you get it back – and the time frame is prone to be smaller. I hope the new system will help you.

Q: Regarding hypoglycemia, did you experience any differences?

Ms. Jordan: I can’t say I have. I have had two severe reactions with the DiaPort – when I had an infection I was chasing my high blood glucose levels, so when I sat down I crashed completely. I only realized I was hypo right before I started convulsing.



Dirk Völkel, PhD (Roche Diabetes Care, Mannheim, Germany)

The renowned Dr. Völkel previewed ongoing studies at Roche exploring high performance continuous glucose sensors, highlighting current limitations in CGM technology. In all, he recounted a number of current limitations in CGM technology, including the time lag between measurements and actual values, signal artifacts due to noise and sensor/tissue compression effects, difficulties with prospective calibration, and compromises in signal filtering. While he (regrettably!) did not provide details of the sensor used, he previewed early results from ongoing internal trials performed in clinic in type 1 diabetes patients over the course of eight days – he suggested main improvements included a stronger mean relative deviation and an uninterrupted readings. With data points compared from two identical sensors, results did suggest some inter-sensor variation though decreasing greatly with time. He suggested goals for future CGM systems include reliable and robust readings under real life conditions, accurate reflections of interstitial glucose concentrations, and fewer false alarms – while clearly details are limited now, we look forward to learning more about Roche’s work in this area as it progresses.


Ohad Cohen, MD (Sheba Medical Center, Tel Hashomer, Israel)

The indomitable Professor Cohen spoke once again on the topic of Medtronic’s CareLink therapy management software for healthcare professionals. Since we last heard him give his workshop at ATTD, the software has been improved and upgraded to CareLink Pro 3. You can read our detailed report on CareLink Pro 3 from December 2010 at One of the most exciting aspects of the upgrade is the ‘therapeutic considerations’ for physicians, which seek to provide actionable advice on events leading up to hyperglycemic or hypoglycemic episodes. The software analyzes the circumstances leading up to these episodes and makes suggestions such as “consider reviewing the correct way to change an infusion site with your patients.” The workshop offered a hands-on practical experience with the software, organized as cases. The participants each worked with a computer with access to CareLink Pro. The workshop made it clear that the level of insight that can be obtained from CareLink is really impressive, but it requires familiarity with the program, careful study of the data, and good clinical judgment.

  • CareLink data management software can be used with the latest technology, but also is fine for patients who just want to download a simple meter. Patients can download their data at home via CareLink personal, or at the clinic. (CareLink supports most meters, but only a Medtronic pump and CGM sensor).
  • Incorrect times and dates affect the accuracy of the reports; Dr. Cohen asked physicians to help patients synchronize all the devices to within one minute of each other. (This is yet another burden for busy physicians – wouldn’t it be great if the devices could set the correct time themselves, like a cellphone?)
  • Medtronic explained that they had improved CareLink Pro based on customer feedback. Customers indicated that the prior version of the software could be overwhelming, with too many reports and a difficult learning curve. Accordingly, Medtronic added two reports, which are meant to make information easier to find, to substitute for many of the other reports, and to proactively provide analysis of the data.
  • The first new report is the Therapy Management dashboard report, which is an information-dense snapshot of key patient information all on one page. It includes a 24-hour graphical analysis of glucose and insulin (shaded to highlight glucose peaks and troughs), mealtime analysis, statistics, pump use, sensor use, and detection of hypo and hyper patterns.
  • Another useful report that was added is Episode Summary, which provides a summary of the hypoglycemia and hyperglycemia episodes and the events preceding the episodes. The system also gives ‘therapy considerations’ to help explain the root causes and correct them. This report is one of the most interesting and exciting aspects of CareLink Pro, since it attempts to simplify the interpretation of the data and to provide actionable therapy solutions – a big time saver for clinicians.
  • The workshop used a series of clinical cases to demonstrate how the software can be used. Dr. Cohen managed to make the cases unfold like a detective story. There were many and varied subtle points that arose in the discussion. It became very clear that making appropriate therapy decisions requires careful study and interpretation of the reports and quite a lot of clinical skill and judgment. But with the appropriate investment, the level of insight is very impressive and would otherwise be impossible. In the right hands, and providing the data is there, this is a really powerful resource.

-- by Adam Brown, Eric Chang, John Close, Joseph Shivers, and Kelly Close


Note: We updated this report after publishing it February 9 in order to better characterize a talk about the artificial pancreas.  We regret any mischaracterizations with the original report.