November, 11-13 2010; Bethesda, MD Full Notes, Draft

Executive Highlights

Dr. David Klonoff’s 10th Annual Diabetes Technology Meeting (DTM) gave attendees a great feel for the state of the field as we exit the first decade of the 21st century. Excitement among the over 500 doctors, engineers, and industry representatives was high, especially given that DTM came immediately after the FDA Artificial Pancreas Workshop on November 10th (included as an appendix in this report). The daylong special meeting featured presentations from clinicians, industry, and the FDA; appeals from patients and JDRF; and hours of lively discussion. The artificial pancreas was also featured prominently (and less contentiously) at DTM, and we left Washington feeling slightly more optimistic about the prospects of an AP in the coming years. Big-time researchers (Drs. Claudio Cobelli, Edward Damiano, Boris Kovatchev, Eric Renard, Stuart Weinzimer, and more) presented their impressive closed-loop work, their thoughts on algorithms (modular MPC seems to be the consensus), and potential for glucagon in bi-hormonal systems down the road (a long way off, but strong research prospects). While no new data were shown, seeing the volume of evidence supporting closed-loop control certainly put things into perspective, despite the FDA’s concerns about CGM accuracy.

On that note, it was refreshing to see that CGM in the inpatient and outpatient settings will continue to dramatically improve. Talks by Rajiv Shah of Medtronic (the Enlite Sensor), Peter Simpson of DexCom (fourth- and fifth-generation sensors), and Michael Higgins of Edwards Lifesciences (inpatient CGM) reaffirmed our belief that the future looks incredibly bright for CGM. More accurate, smaller, and less painful sensors should be the status quo in the coming years.

We also found some of the less buzzed-about presentations interesting, including those on telemedicine, biomarker tests, and the security of wireless devices. As technology evolves in new and unfamiliar directions, we’re glad that so many top minds are concentrating on how to manage the risks and harness the opportunities – and we’re glad we get the chance to watch.

 

 

Conference Sessions

Artificial Pancreas

ENGINEERING AND CLINICAL PERFORMANCE STANDARDS FOR CLOSED-LOOP CONTROL OF DIABETES

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

An important figure in the field of the artificial pancreas, Dr. Boris Kovatchev outlined the engineering and clinical performance standards needed for effective closed-loop research. Both the Artificial Pancreas System (APS) and in silico testing of algorithms have become standards in the field; these developments have accelerated the pace of research and the ease of collecting data. According to Dr. Kovatchev, clinical performance standards will also need to be adopted that allow cross-study comparisons. In this case, appropriate statistical metrics and outcomes will need to be agreed upon and utilized. Finally, the modular architecture for system design will allow for further standardization, safety, and effectiveness. According to Dr. Kovatchev, adopting such standards will accelerate the research progress and the commercialization of closed-loop control systems.

• Current diabetes devices for the artificial pancreas (AP) don’t communicate with each other; the Artificial Pancreas System (APS) solves this problem. The APS is a hardware interface that allows currently marketed pumps, CGMs, and algorithms to communicate with each other during a clinical trial. It is the result of work from UCSB and Sansum and represents a significant advance in closed-loop system development. The APS makes research with current devices far easier and more standardized.
• The FDA approval of in silico testing of AP systems signaled an important advance in the establishment of preclinical standards. Now, researchers can easily test their algorithms in a simulated population of patients, pumps, and continuous glucose monitors. This system helps replace costly and frustrating animal testing.
• Dr. Kovatchev said that clinical performance standards will need to be adopted to compare different studies. Such standards could establish appropriate glycemic averages and weight the risk associated with different glucose values (e.g., 10 points of risk at very high and very low glucose values). Additionally, a universal control variability grid analysis (CVGA) could be adopted for comparison-sake. The CVGA has an x-axis with the minimum BG observed in a study (2.5th percentile), while the y-axis has the maximum BG observed in a study (97.5th percentile). “A zone” readings contain a minimum reading between 90 and 100 mg/dl and a maximum reading between 110 and 180 mg/dl. Subjects in the A zone would have “optimal control,” while other zones might indicate over correction of hypo- or hyperglycemia or just poor control in general.
• Algorithm architecture standards must fit together, and a modular control system may offer the best standardized approach for a closed-loop system. Modular control algorithms can be decomposed into four natural components: a safety algorithm to prevent hypoglycemia, a pre-meal bolus calculator, a range controller, and a basal rate controller. On top of all this would be another safety system detecting sensor errors and deviations. The architecture of the system would use the APS hardware interface (APS). The modular system allows different algorithms to be developed and tested for each of the four components, allowing much more collaborative research and variability in development.
• Two landmark studies are underway to test the safety and effectiveness of closed- loop systems. The JDRF control-to-range trials will use the previously mentioned modularsystem architecture. These trials will be conducted at seven centers around the world and should yield promising results for the future of the artificial pancreas. Another important study, underway as of early 2010, is the European AP@home project, in which researchers will test currently available artificial pancreas algorithms with CGMs and insulin pumps currently on the market. During the final year of this four-year project, a multinational clinical trial will compare the performance of an artificial pancreas system with standard intensive insulin therapy.
• To close, Dr. Kovatchev briefly reviewed the results of a promising closed-loop test conducted at UVA. Subjects underwent a moderate exercise test, and the closed-loop system reduced hypoglycemia more than twofold, boosting time in the range of 80-140 mg/dl from 36% to 52% and time in the range of 70-180 mg/dl from 64% to 82% (UVA, Breton, Demartini, and Clarke).

THE DESIGN OF CONTROL ALGORITHMS FOR THE ARTIFICIAL PANCREAS THAT YIELD ROBUST PERFORMANCE AND ARE SAFE FOR THE PATIENT

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

In a more technical presentation, Dr. Francis Doyle III explained the ideal design characteristics of AP control algorithms that would maximize both performance and robustness. Such algorithms will have customizable performance criteria, constraints, and control parameters. This customization, accompanied by the use of model predictive control and insulin-on-board calculators, will boost the efficacy of these systems. According to Dr. Doyle, with better knowledge of patients and improved CGM accuracy, we can achieve reduced variability in these systems and obtain better glycemic outcomes.

• Closed-loop algorithms face a fundamental tradeoff between performance (i.e., time in range) and robustness (i.e., tolerating uncertainty in a safe manner). Although it would be great to have a fast algorithm that minimizes postprandial glucose and quickly returns blood glucose to normal, this may not be feasible given concerns about hypoglycemia. As an added problem, subcutaneous insulin and sensors require a high bandwidth control design to blunt an unanticipated change in glucose. Finally, uncertainty between patients adds another difficulty in the algorithm development process.
• When thinking of artificial pancreas algorithms, there are three main elements to design: the performance block (e.g., outcome measures), the constraint block (e.g., safety, overrides, insulin-on-board), and the controller block (e.g., the model and its aggressiveness). Each block can be customized to the individual patient.
• Model predictive control allows the closed-loop system to make forecasts into the future by taking into account recent glucose history and predicted insulin requirements. Both transfer function models and autoregressive exogenous input (ARX) models have been explored. ARX models take into account lagged values of blood sugar, recently delivered insulin, and terms that capture carbohydrate consumption. A typical subject model has around eight coefficients and is individualized to the patient. In an 11 patient study, the model was able to predict roughly 90 minutes into the future.
• Insulin-on-board controllers can provide greater safety by accounting for delivered insulin that has yet to act. It can be customized to the subject. In one trial studying closed- loop system in unannounced meals, a model predictive control system using insulin on board achieved over 70% of readings in zone, with all values in the green A and B zones of the CVGA diagram (compare to exclusively yellow, orange, and red zones with open-loop control).

• Depending on the objective (e.g., A1c, time in range, CVGA), the algorithm components can be tuned appropriately. Target zones can be tailored and customized, invoking different strategies depending on the current blood glucose. In the promising JDRF Control-to-Range Trials, the question of tuning will be explored with a model predictive control algorithm.

DECISION SUPPORT FOR GLYCEMIC CONTROL IN THE HOSPITAL

Martin Ellmerer, PhD (B. Braun Melsungen AG, Maria Enzersdorf, Austria)

Shifting the emphasis away from outpatient diabetes care, Dr. Ellmerer discussed B. Braun’s Space GlucoseControl (Space GC) system for facilitating glycemic control in critical care settings. He explained that the decision support device is an open-loop system that signals a patient’s pump to deliver insulin based on a control algorithm, which treats to one of two target ranges (80-110 mg/dl or 80-150 mg/dl); the algorithm is in turn based on manually measured blood glucose values. Dr. Ellmerer explained that the control algorithm is compatible with point-of-care meters and blood gas analysis as well as future forms of blood glucose measurement. Additionally, the algorithm can determine and recommend how often measurements are needed (typically once every two hours or so). Dr. Ellmerer explained that the system can maintain safety despite infrequent glucose input because the algorithm automatically accounts for insulin dosage (the main cause of hypoglycemia) and parenteral and enteral nutrient delivery (the main causes of hyperglycemia). Space GC-informed therapy has been shown superior to standard of care in several clinical trials, and we hope that the system (which is commercially available in Europe and IDE-cleared in the US) continues to perform strongly during its ongoing path through FDA regulation.

• Dr. Ellmerer proposed several ways in which inpatient glucose control is more challenging than outpatient self-management. A patient at home monitors their own glycemic patterns all day, every day, whereas nurses must treat a variety of patients – generally for too short a time to become familiar with any of them. At home, glucose control can be a patient’s top priority, but in the hospital, it generally falls below other priorities (e.g., CV monitoring, respiratory monitoring and support, volume replacement therapy). At home, measurements are typically taken four-to-six times per day with a glucose meter, while in the hospital it is more customary to take two measurements every 24 hours. Dr. Ellmerer noted that nurses also have some advantages compared to patients at home: they have access to different measurements (e.g., blood gas analysis, lab values, glucose meters) and they deliver insulin intravenously rather than subcutaneously, enabling more responsive control.
• B. Braun’s Space GlucoseControl (GC), a decision-support system for critical care glucose control, automates two of the three elements of glucose control: dose calculation and insulin delivery. The system requires blood to be manually tested by either blood gas analysis or glucose meter, and the data can be entered on a touchscreen display. The system’s model predictive control (MPC) algorithm, modified for critical care patients rather than people with diabetes, receives the blood glucose data and also automatically monitors enteral and/or parenteral nutrition rates. Based on this information, the algorithm calculates the appropriate insulin rate to treat to one of two target ranges, 80-110 mg/dl or 80-150 mg/dl. Upon user confirmation, the system automatically delivers insulin at the new rate. The algorithm also calculates the time until a subsequent glucose check should be performed, displays a countdown, and sounds an alarm when it is time to test.

• In several clinical studies, the Space GC system has increased the time patients spend in their target glycemic range (80-110 mg/dl). In a study comparing Space GC to routine management in 60 cardiac surgery patients, arterial glucose was measured hourly for up to 24 hours. Patients using the Space GC spent a median of 52% of their first 24 hours in target compared to 19% for patients receiving routine treatment (Plank et al., Diabetes Care 2006). In another 24-hour study of 60 cardiac surgery patients, an improved MPC algorithm with 1.5-hour mean sampling interval outperformed a routine glucose management procedure with 2-hour mean sampling intervals; time in target was 60.4% and 27.5%, respectively. In a 48-hour study of 120 cardiac surgery patients with mean blood testing intervals of roughly two hours, the Space GC gave higher mean time in target (46%) than the two comparator protocols, Mathias (38%) and Bath (40%) (Blaha et al., Diabetes Care 2009). Substantial improvements were not seen in the Leuven hospital, but standard care there led patients to spend over 60% of their time in target.
• Dr. Ellmerer argued that although the Space GC does not continuously monitor blood glucose, the system maintains safety by continuously monitoring the major sources of hypoglycemia (insulin) and hyperglycemia (nutrition). As an example, he said that several times during clinical trials, the nutrient supply ran out. With continuous glucose monitoring there would be a sensor delay before hypoglycemia was recognized, but the Space GC could adjust IV insulin dosage immediately to leave only the 20-minute delay of insulin action. The algorithm can also compare its estimate of glucose level to input values, alerting nurses if there is a major discrepancy between the values (e.g., if the insulin supply is diluted).

PERFORMANCE OF CLOSED-LOOP SYSTEMS: WHAT ARE THE GOALS AND HOW DO WE REACH THEM?

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

Dr. Renard described the evolution of closed-loop control over the past decade, reviewing the work of researchers including Dr. Boris Kovatchev, Dr. Claudio Cobelli, and himself (sometimes affectionately termed a group of “three musketeers” for their work on the artificial pancreas). He stated that the goal of the artificial pancreas is simple: to enable the level of glycemic control seen in people without diabetes, whose blood glucose stays between 70 and 140 mg/dl. He followed the changes that have occurred in artificial pancreas designs: subcutaneous CGM was found effectively equivalent to intravenous monitoring, the proportional integrative derivative (PID) algorithm has given way to the superior model predictive control (MPC) algorithm, and subcutaneous insulin infusion is now typically used in place of the less-widely available intraperitoneal administration. Dr. Renard expressed confidence in the progress made by closed-loop control so far, and he called for better sensors, self- teaching algorithms, and transatlantic collaboration on outpatient trials.

PANEL DISCUSSION

Moderators: Jeffrey Joseph, DO (Thomas Jefferson University, Philadelphia, PA) and Marc Torjman, PhD (University of Medicine and Dentistry of New Jersey, Camden, NJ)

Questions and Answers

Q: For Dr. Kovatchev – in the new design of the control algorithm, an important part is played by open loop control. I wanted to know if you will design an algorithm for this part of the regulator.

Dr. Kovatchev: Decomposing the control process into four pieces, safety and postprandial control have been implemented consecutively, leaving basal and preprandial out for now, for control by the clinician. The plan is to consider moving into these areas, too; at least one fully automated version has already been developed.

Dr. Joseph: The problem with the artificial pancreas comes with meals. The subcutaneous insulin works slowly, leading to hyperglycemia in the postprandial period and then potentially hypoglycemia. What do we need: better sensors, insulins, algorithms? What about something that senses a meal is being eaten?

A: There are many answers to that question. One that has been tested is the manual pre-meal bolus [Tamborlane and Weinzimer, Yale]. In fully automated control, faster-acting insulin will be very beneficial. There are new approaches that are being built along stochastic models of MPC that can anticipate meals and patterns. There is a lot of work being done in this area.

A: I think what we should really reach is to minimizing the human component; the system should be as automatic as possible. It should become clever by itself, taking experience from previous days. I could see a system with a run-in period where it learns by itself. I think that will be the final way to solve this question.

Dr. Joseph: It was interesting to see the advantages of peritoneal insulin: faster action and less variability. Are there other physiological advantages to peritoneal insulin?

Dr. Renard: Peritoneal insulin is certainly more physiological. The lower the peripheral insulin, the less risk of postprandial hypoglycemia. The subcutaneous space is the worst place to put insulin. If you can decrease variability, that helps results.

Dr. Joseph: Dr. Renard, you have experience with the implantable intraperitoneal pump, and I saw the DiaPort on your slide.

Dr. Renard: There’s a study being done with JDRF support that uses subcutaneous sensing, the MPC algorithm, and the DiaPort. The pharmacokinetic benefits of intraperitoneal insulin should improve the closed loop control and also inter-patient variability. This has not been examined a lot, but variability from one patient to another is quite high with subcutaneous delivery. The peritoneal way, between-patient variability is much reduced.

Q: Dr. Ellmerer, did you treat any patients with type 1 diabetes?

Dr. Ellmerer: Type 1 diabetes was not an exclusion criterion, though most of the patients with diabetes had type 2. We didn’t look at type 1 diabetes; most patients did not have diabetes or had type 2.

Q: In critically ill patients, 70 mg/dl can be potentially harmful. The Leuven studies reported hypoglycemia as less than 40 mg/dl; in my mind that is dangerously low. Hypoglycemia below 70 mg/dl or 60 mg/dl is not really reported; have you looked at this?

Dr. Ellmerer: I agree there’s controversy about hypoglycemia. We moderated the target range of 80-110 mg/dl used by Dr. van den Berghe. I know lots of ICUs in the US and Europe don’t go so tight anymore. 80-150 mg/dl is the range we used.

Dr. Stuart Weinzimer (Yale University, New Haven, CT): I’m wondering about the other patient factors that will be important. I wasn’t quite sure what you meant by an algorithm using other information from that day’s activities. Do you mean meal timing? This is analogous to meal announcement. I noticed that in your slides, insulin delivery started to increase before the meal; it seemed like you had meal timing built in to the algorithms.

Dr. Kovatchev: In terms of patterns, the stochastic controller scans the previous days and determines a pattern for previous days. Using this pattern, it would start delivering insulin earlier than meal. In such a scenario, the cost function becomes more important, especially if there is a skipped meal.

Q: Dr. Ellmerer, have you had a chance to look at what the clinicians in the Leuven study were doing manually and whether there was anything you could do to improve your algorithm?

Dr. Ellmerer: Dr. Grete van den Berghe performed a landmark study. She uses a very simple guideline, not even a protocol: if blood glucose is high, raise insulin; if it’s going low, lower insulin. It’s more sophisticated than this, but it’s very simple. She trained her nurses for a half to three quarters of a year; she and her nurses are experts. That was how she prepared for the trial. Unfortunately and understandably, the starting and preparation was far too short in the other studies of inpatient tight glucose control. This may be why so many hypoglycemic events occurred and why the studies finished with different results.

Q: Do you think industry could take the components and make an application for clinical practice, or are more trials needed?

Dr. Renard: I think for night control we are quite close. There is lot of safety with the closed loop during the night – more safety than with doing what people do now, going to bed without any measurement. So it’s really nonsense that we don’t move to this at home, providing industry products for night control. During the day it’s quite tough: meals are hard. The exercise problem is almost solved, however – closed- loop control is safer than what patients typically do now.

Dr. Kovatchev: I would support all that Dr. Renard says. I think the first step of implementing the safety system can be made now.

Dr. Robert Vigersky (Walter Reed Army Medical Center, Washington, DC): I have a question that’s related to what we’re seeing in type 1 diabetes. We’re starting to see obesity and insulin resistance. These patients are requiring very large doses of insulin. I’m wondering about the robustness of these algorithms at high doses of subcutaneous insulin?

Dr. Kovatchev: The individualization procedure takes into account several factors: bodyweight, correction factor, basal rate, etc… and it determines the aggressiveness of the control algorithm. If someone is on a high basal rate, this makes the controller more aggressive because it suggests the person is more insulin resistant. This has been tested in silico, where we have a wide variety of patients, and it seems to work.

Comment: On the modular design, ideally you want a plug-in system. But my major concern is that it’s not about the components but the overall system design. We need the right brakes for the right engine. It’s important to have the overall design right. It might not be easy to move components from one system to another.

Dr. Kovatchev: I fully agree, and that’s the exact purpose of the modular design.

Arleen Pinkos (FDA): Can you describe whether it’s the safety module or something else that corrects for problems with the CGM? Have you observed CGM problems in your clinical research studies? How often do you think you need to observe that before you think it’s safe to go to outpatient trials?

Dr. Kovatchev: Safety is achieved by a system state estimation. The system does not only rely on CGM, but takes into account other factors. We want to know what the state of the person is. If we get the state estimation right, we can make the pump safer and correct sensor errors using the additional information from the pump. Based on that state estimation, both devices can be improved. We did observe in our clinical studies multiple instances of sensor problems, particularly overnight when people would sleep on the sensor and data would drop out. In these cases, the controller estimates the system state and won’t be influenced by problems with the CGM. It knows that the appropriate amount of insulin was delivered and the system won’t fall apart because of the drop in CGM data.

Q (FDA): To follow up on Arleen Pinkos’ question, do you think the controller could anticipate if the patient exercised before sleep?

Dr. Kovatchev: Yes, it will adapt to changes in sensitivity.

Dr. Joseph: Is there any use for an accelerometer that would tell the duration and intensity of exercise?

Dr. Kovatchev: To the best of my knowledge this has not been evaluated, but the system should get better with more inputs.

Dr. Renard: The key component is the safety module. What you are afraid of with exercise is hypoglycemia. Since you already have a system for preventing this risk, the system will react and stop delivery. The risk of hypoglycemia with the closed loop was been minimized. This is why I support going with the closed loop at home. The main risk has been minimized; now it’s time to go.

Closed-Loop Control: What Is Holding Us Back and What Can Be Done?

BETTER SYSTEMS: BUILDING AN ARTIFICIAL PANCREAS SYSTEM

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

“The FDA has outlined the device limitations of the current technology. Even with these limitations, we should be able to develop an artificial pancreas today.” To close the first official day of the conference, Dr. Charles Zimliki, Chair of the FDA’s Artificial Pancreas Critical Path Initiative (and someone who has had type 1 for 27 years), gave an informative and very positive outlook on the FDA’s approach to approval of an artificial pancreas system. In his opinion, an artificial pancreas can be developed today with existing technology provided the correct system-level studies are done. He described the FDA’s main issues with approval of the artificial pancreas as sensor inaccuracy, sensor artifact, sensor dropout, and pump failures. If devices can be improved and the correct studies can be carried out, Dr. Zimliki said he feels very optimistic about the potential approval of an AP. We appreciated his patient- centered perspective and his enthusiasm for the artificial pancreas, a welcome contrast to Dr. Patricia Beaston’s quite cautionary tone at the FDA Artificial Pancreas Workshop this past Wednesday.

• The FDA considers current sensor accuracy an inherent limitation in the closed loop. Data from closed-loop studies show both phenomenal sensor accuracy and terrible sensor accuracy. At times, CGMs trace YSI values perfectly and could be used for insulin dosing. However, sometimes CGMs lose the accuracy and can underestimate hyperglycemia and overestimate hypoglycemia. This has been observed in numerous closed-loop studies, but in one example, Dr. Edward Damiano’s team recalibrated the CGM and the accuracy returned to acceptable levels. According to Dr. Zimliki, potential fixes might include using redundant systems or calibrating the system with different algorithms (e.g., the Medtronic presentation by Rajiv Shah at this meeting demonstrated great improvements in sensor accuracy with better algorithms). According to Dr. Zimliki, the agency is willing to accept all potential solutions.
• Sensor artifact and dropout are also problematic for the artificial pancreas. In the case of sensor artifact, placing pressure on the sensor while sleeping can lead to situations whereCGM values deviate from actual blood glucose values. In sensor dropout, data simply stop being transmitted. Both scenarios present insulin-dosing problems in the AP, and appropriate backup and detection systems need to be designed to deal with such scenarios. Dr. Zimliki suggested solutions for sensor artifact could be tested in pigs, as this might be easier. He was less worried about sensor dropout, indicating a simple fix would involve an emergency backup basal rate/system to continue delivering insulin at appropriate levels.
• Instances where the pump is occluded are another roadblock to AP development. This problem has been observed in closed loop studies, and it’s even been documented that certain pumps can take 24 hours to detect the occlusion. The agency doesn’t have this solved and would encourage industry to think hard about the issue.
• Dr. Zimliki said that existing technology can close the loop, but the FDA will ask researchers to take a system-level approach in their studies. Dr. Zimliki slightly criticized control algorithm studies, which are only one piece of the equation and don’t take into account broader system-level performance. The FDA would strongly encourage studies on the system level, especially those that evaluate CGM problems. By addressing these issues and designing better devices, studies can progress to the outpatient environment more quickly and the closed loop will become more of a reality for patients.

BETTER ALGORITHMS: MODULAR SYSTEM APPROACH AND IN SILICO TESTING

Claudio Cobelli, PhD (Dept. of Information Engineering, University of Padova, Italy)

Similar to many of the other closed-loop presentations given at this meeting, Dr. Claudio Cobelli recommended the use of modular MPC controller algorithms for closed-loop control. He noted many of the difficulties the AP brings to the table, such as safety considerations, inter-subject variability, meals, and sensor algorithms. Modular-based MPC can help deal with all these difficulties and such systems facilitate easy testing, incremental regulatory approval, faster deployment, and clinical acceptance. Dr. Cobelli has been impressed with the merger of clinical experience and engineering, and he emphasized that this must continue for the fast pace of research to continue.

• Closed-loop systems are plagued by delays in subcutaneous insulin absorption/action and lags in sensor readings. To cope with delay, Dr. Cobelli advocated Model Predictive Control (MPC). An MPC model must use currently available data and deal with disturbances (meals, exercise). In an ideal world where all information is known, this is easy. But this is obviously not the case, and the solution is the modular system approach.
• The safety module of the algorithm protects against hypoglycemia, a type of “intelligent brake system” based on CGM and pump data, IOB, and model prediction. In a poster by Hughes et al., an in silico trial of a safety supervision system (SSS) tested two scenarios for increased risk of hypoglycemia: 1) increased insulin sensitivity (e.g., resulting form exercise) and 2) an overinsulinized meal. In scenario 1, the SSS prevented 43% of hypoglycemia (<60 mg/dl) and issued warnings for 74% of imminent hypoglycemic episodes not prevented, with average warning time of 35 minutes. When treatment with carbohydrate was simulated at the time of warning, 98% of hypoglycemic episodes were prevented. In scenario 2, 100% of all pre-meal boluses were intercepted by the system prior to delivery and appropriate warning were issued. The study suggests that an independent SSS can dramatically reduce the incidence of hypoglycemia.

• Better algorithms must account for high variability between subjects. It’s clear from the data that there will not be a universal MPC model; rather, model predictive control will be tuned to a particular subject through an individualization model that lies on top of the MPC and safety modules. An aggressiveness factor, q, varying as a function of body weight, basal delivery, and carbohydrate ratio, has had solid success in clinical studies and will be used in the JDRF Control-to-Range trials.
• Meals add further complexity to closed-loop systems and require feed-forward MPC. A new strategy learns from the patient during open-loop therapy; subsequent days involve closed-loop therapy based on the information garnered. This use of open-loop-informed closed- loop control appears to be a practical and promising strategy for taking into account all the available information regarding each patient without requiring ad hoc experiments for model individualization.
• Although the focus on CGM in the closed loop is often directed at the sensors themselves, the algorithmic part of the sensor model has been understated. At a given moment, the accuracy of the sensor can be determined and should be incorporated into the AP algorithms. Sensors with better calibration algorithms, self-diagnosis of problems, and failure detection will definitely help in the development of the closed-loop. Additionally, seeing the sensor and pump as a system where each can improve the other will be an important step.
• One of the key benefits of in silico testing is the inter-subject variability built into the system (100 adults, 100 adolescents, 100 children, wide distribution of insulin absorption and insulin action). Going forward, efforts will be made to add glucagon and to incorporate inter- and intra-day variability in patients.
• The JDRF Multi Center Control-to-Range trials will use the Artificial Pancreas System (APS) for hardware interface and a modular MPC algorithm. Thus far, there have been 11 successful studies: six in Montpellier, France and five in Padova, Italy. The researchers are confident in the results that have been obtained and are looking forward to next year’s upcoming trials: four in the U.S. and three in the EU. Also of note, on October 22, 2010, UVA and the Mayo Clinic received approval to conduct a study using zone MPC closed-loop control with pramlintide (#G100266).

BETTER INSULIN KINETICS

Stuart Weinzimer, MD (Yale University, New Haven, CT)

In a speedy and incredibly comprehensive presentation, Dr. Stuart Weinzimer gave a broad overview of the possible solutions to the limitations of subcutaneous insulin in the closed loop. Defining the problem, Dr. Weinzimer reminded the audience of the slow absorption profile of subcutaneous insulin, with peak insulin levels achieved in one hour and peak insulin action around 90 minutes. In early closed-loop studies, this delay led to hyperglycemia immediately following meals, overdosing of insulin by the system, and subsequent hypoglycemia. To optimize insulin kinetics, one approach is accelerate insulin appearance. This can be achieved through a pre-meal manual priming bolus (a Yale Study using ePID dropped peak postprandial glucose levels from 219 mg/dl to 196 mg/dl), heating the infusion site (e.g., InsuPatch: a Cengiz, et al. 2010 DTM poster showed statistically significant increase in GIR [2.4 to 3.7 mg/kg/min] and a decreased Tmax GIR of 133 vs. 84 minutes with use of the InsuPatch), and use of hyaluronidase (lispro plus PH20 significantly lowers peak postprandial glucose). Additionally, one can accelerate the disappearance of insulin through the use of “insulin feedback” models (a Yale study found such a model provided a more rapid return to target glucose), using individualized pharmacokinetic models and glucagon to control hypoglycemia (an approach explored by Drs. Russell, Damiano and colleagues in Boston), and finally, using pramlintide to slow carbohydrate absorption. The latter is currently being studied at Yale, and early results show a slight but statistically significant reduction in postprandial glucose. According to Dr. Weinzimer, the first generations of the closed-loop will require hybrid open loop/closed loop control and/or faster insulins, while subsequent generations will use bi- hormonal systems.

BETTER HORMONES: THE ROLE OF GLUCAGON FOR FULLY AUTOMATED GLUCOSE CONTROL

Steven J. Russell MD, PhD (Harvard University, Boston, MA)

Extolling the benefits of glucagon for the artificial pancreas, Dr. Russell reviewed two clinical trials of bihormonal pump control that he conducted with Drs. Edward Damiano, Firas El-Khatib, and David Nathan. The main use of glucagon in these early trials has been to correct high prandial insulin doses and avoid hypoglycemia. However, in the first trial, this approach was not consistently successful – a problem that the researchers corrected by revising their model of insulin pharmacokinetics (see Dr. Damiano’s DTM 2010 talk on this subject for more details). In the second (still ongoing) trial, glucagon is used to prevent hypoglycemia from exercise and insulin boluses; the lowest blood glucose measurement in the first 192 hours of the trial has been 72 mg/dl. Looking forward, Dr. Russell suggested that because subcutaneous glucagon onset (~23 minutes) is much faster than subcutaneous insulin offset (~78 minutes), a system using glucagon can respond to hypoglycemia much faster. Thus, it may be that only a bihormonal system could give true closed-loop control able to respond to real-world challenges such as exercise (an element of his group’s second clinical trial). Although he acknowledged that glucagon does not yet exist in a stable liquid form, he seemed strongly in favor of the hormone: he framed the choice of closed-loop control with one- or two-hormone systems as being between “automated insulin delivery” and an “artificial endocrine pancreas”; during the question and answer session, he insisted that glucagon should be an integral part of closed-loop control rather than just a failsafe measure. We hope that competition between bihormonal and insulin-only systems continues to drive innovation on both fronts and that it slows progress on neither side: a single-hormone system, even one that doesn’t fully close the loop, would be a huge step forward for the field.

• To end his talk, Dr. Russell addressed one of the greatest challenges to a bihormonal pump – the fact that glucagon is unstable in solution. For example, in his group’s closed-loop trials, glucagon is reconstituted from a solid no more than 24 hours prior to use. Dr. Russell expressed his optimism that a long-term version of glucagon would be available soon, in light of developments such as the new glucagons described in the November 2010 Journal of Diabetes Science and Technology (a formulation from Steiner et al., an analog from Chabenne et al., and a pH-stabilized analog from Ward et al.).
• During the question and answer session, Dr. Russell argued that glucagon should be an integral part of closed-loop control, not just a failsafe. He disputed the notion that a bihormonal pump increases the demand for insulin; patients in his group’s trials typically eat twice as many carbs as they previously had, and (between pre-meal boluses and the algorithm- controlled basal rate) they use half the total daily insulin dose.

BETTER SENSORS: THE CLINICAL PERSPECTIVE

Jens Christiansen, MD, FRCPI (Aarhus University Hospital, Aarhus, Denmark)

From his perspective as a clinician, Dr. Christiansen discussed critical issues related to the quality and uptake of CGM. He emphasized that health care providers are interested in a robust, reliable solution, not necessarily a fancy one. He suggested that some people’s skepticism might come from memories of the first subcutaneous devices, which he said were promoted disproportionately to their benefits. To illustrate the difference between an optimal and practical solution, he amused the audience with the story of how the American space program spent $1 million to develop a pen that wouldn’t leak in zero gravity… while the Russians used a pencil. He emphasized the differences between randomized- controlled trials and a clinical setting where the time for instruction is short, and he said that ease of calibration and maintenance are important considerations as well. Inspiringly, Dr. Christiansen said that he thinks about “the market” in terms not of money but patients. He contended that a cheap, functional CGM benefiting patients worldwide would be much better than a device that produces perfect Clarke scores but is made of solid gold.

WHAT WENT WRONG? THE IMPORTANCE OF BETTER HUMAN FACTORS

J Geoffrey Chase, PhD (University of Canterbury, Christchurch, New Zealand)

Dr. Chase presented seven medical vignettes to illustrate principles such as good study design, interdisciplinary cooperation, and common sense. We agree that human factors are hugely important for the success of any clinical intervention, although the human factors of this talk – which repeated itself and ran well over time – could have been somewhat better designed.

PANEL DISCUSSION

Moderators: Jeffrey Joseph, DO (Thomas Jefferson University, Philadelphia, PA) and Marc Torjman, PhD (University of Medicine and Dentistry of New Jersey, Camden, NJ)

Questions and Answers

Dr. Joseph: I’m hearing two groups emerge here, one group saying that glucagon is necessary and another group saying that turning off insulin may be equally safe and effective. Any thoughts?

Dr. Russell: I just don’t think the timing works out in the absence of glucagon. Hypoglycemia can happen fast and turning off insulin, even to zero, won’t do anything right now for the insulin that is already in the subcutaneous space.

Dr. Weinzimer: I don’t take the opposite approach. I actually agree. We’ve required a lot treatment for hypoglycemia in our studies. Looking at our DirecNet studies: by suspending insulin delivery, we still had 15% of subjects develop hypoglycemia. In order to completely avoid hypoglycemia, you either have to use glucagon or pre-treat with carbohydrates.

Dr. Kovatchev: There are two different uses of glucagon. One is using it as a safety feature to prevent hypoglycemia. There is also controlled use of glucagon, which may be more questionable and more dangerous. This triggers the system to deliver more insulin and rely on the glucagon when it over delivers. This needs much more investigation.

Dr. Russell: You won’t need to use glucagon in the safety sense if you can effectively use it in a control setting. We keep glucagon within the physiologic range and these are very small amounts. It’s also not true that the use of glucagon causes more insulin to be administered. Before our subjects do the trial, they tell us their total daily dose. In the study, we give them twice the number of carbohydrates they usually eat and with pre-meal boluses, they use half as much insulin. We’re very efficient with the amount of insulin we use.

Dr. Roman Hovorka (University of Cambridge, Cambridge, UK): Why do you think you use half the insulin?

Dr. Russell: I’d like to reference a study by Drs. Damiano and El-Khatib in pigs. It looked at basal and bolus insulin in response to glucose excursions. Then, the experiment was performed again, with all the insulin given in the form of pre-meal boluses. This led to worse control. It seems that if you treat an excursion when it’s small, it’s much more efficient. Additionally, in the pig experiments we’ve done, the amount of insulin used over 24 hours drops in multi-day experiments, perhaps because insulin resistance is being reduced.

Dr. Hovorka: I agree, it’s very interesting.

Comment: I have a comment on complexity. I think there’s a semantic problem with our understanding of what complexity is. The old definition is that complexity is a bad thing, and it seems like this is what your talk addressed. But the new understanding is that complexity is unavoidable and in fact a good thing: the way we build intelligent system is to make it more complex. It’s really a matter of controlling complexity, not avoiding complexity.

Dr. Chase: I didn’t communicate effectively enough. I didn’t mean to avoid complexity in how a protocol operates, but in how it interacts with users. For example, it’s not a matter of avoiding complexity in trial design but of making sure it’s easy to use that design. I think we’re basically agreeing.

Q: I’m wondering about the approval process for these systems. Do you see them being approved in an incremental fashion, with simpler systems approved first? How is the FDA looking at this?

Dr. Zimliki: Yes, incremental fashion seems to be the most logical progression to an artificial pancreas. I think investigating in a parallel fashion will get things approved as quickly as possible. I think that is the way forward. As a fellow diabetic, I would really encourage this style of research. You can develop simple and advanced systems simultaneously; you don’t need one before the other.

Dr. Torjman: I have a question for Dr. Cobelli following Dr. Weinzimer’s talk. Does that device facilitate better control by improving subcutaneous absorption?

Dr. Cobelli: That’s an absolute yes.

Dr. Joseph: I have a follow-up question. What would be your ideal pharmacokinetics and pharmacodynamics?

Dr. Cobelli: IV.

Dr. Joseph: I give intravenous insulin a lot; there are still long action delays.

Dr. Cobelli: I would be happy with IV.

Technologies for Metabolic Monitoring

HIGH ACCURACY OF CONTINOUS GLUCOSE MONITORS IN CHILDREN SUSCEPTIBLE TO CRITICAL ILLNESS HYPERGLYCEMIA

Mark Rigby, MD, PhD, FAAP (Emory University, Atlanta, GA)

Dr. Rigby presented exciting data on the use of subcutaneous CGM in the ICU, showing that the technology seems to be no less accurate than in an outpatient setting despite conventional wisdom to the contrary. Setting out to see how subcutaneous CGM would really fare in critical care, his group conducted a study (n=50) of glycemic control in pediatric (under 16 years-old) patients in the medical or surgical ICU. (As Dr. Rigby pointed out, glycemic control in the pediatric ICU is itself an under-studied area). None of the patients had diabetes and all were receiving mechanical ventilation. Notably, many had comorbidities (e.g., edema) or therapeutic regimens (e.g., ionotropes, vasopressors, glucocorticosteroids) that are generally thought to interfere with subcutaneous sensors. Medtronic Guardian CGM systems were used to monitor patients; impressively, the measurements were comparably accurate to those from outpatient studies of subcutaneous CGM. Of 1,555 CGM measurements that occurred within five minutes of a reference blood glucose draw, 74.6% fell in the A zone of the Clarke Error Grid and 23.3% were in the B zone. Dr. Rigby noted that this study was one of the largest experiences with currently available CGM in any critical care population, and that it demonstrates CGM’s strong prospects as an adjuvant to glucose control in the ICU. Best of all, subcutaneous sensors are still improving dramatically – some notable researchers (not from either company) suggested to us after the session that if Medtronic’s Enlite or DexCom’s fifth generation sensor can also maintain their accuracy in the hospital, they will offer enormous benefits for blood glucose care in the ICU.

• Dr. Rigby discussed two topics that are understudied in the ICU: glycemic control of pediatric patients and the accuracy of subcutaneous CGM sensors. He reminded the audience that trials such as VISEP (NEJM 2008), Glucontrol (Int Care Med 2009), and NICE- Sugar (NEJM 2009) have made clinicians generally cautious about tight glycemic control in the ICU. This wariness is especially true of tight control in the pediatric ICU population, where fears of hypoglycemia have not been allayed by the single recent RCT on the subject (Vlasselaers et al, Lancet 2009). Although the 700 patients in that study had lower mortality with tight control compared to standard of care (2.6% vs. 5.7%), they also experienced far higher rates of hypoglycemia (25% vs. 1.4%). However, Dr. Rigby said that the high rates of hyperglycemia among pediatric ICU patients suggest big potential health gains from glycemic control that avoids hypoglycemia.
• Subcutaneous CGM sensors are often said to be inadequate for the ICU setting, but these claims have generally been based on theory rather than evidence. Dr. Rigby listed several of the reasons the ICU is thought to confound subcutaneous sensors. Many patients in the ICU have altered perfusion due to conditions such as edema and vascular leakage; many patients use vasodilators, vasoconstrictors, corticosteroids, or other therapeutics believed to interfere with sensors; patients are often at the extremes of temperature, pH, osmolarity, and physical size; and the time lag of glucose readings is often thought to be more problematic for critical care than in an outpatient setting.
• Dr. Rigby investigated the use of subcutaneous CGM as an adjuvant to glycemic control in pediatric ICU patients receiving mechanical ventilation (n=50). Patients ranged in age from 6 weeks to 16 years old (mean 4.3 years) and in weight from 2.4 to 87 kg (~5.3 to ~191 kg). Exclusion criteria were coagulopathy and diabetes. 30 patients were on vasoactive drips, six received continuous renal replacement therapy (CRRT), three required venovenous extracorporeal membrane oxygenation (ECMO), and 20 had blood glucose values persistently above 140 mg/dl treated with insulin. Ionotropic therapy was used by 64% of patients, and glucocorticosteroids were taken by 68%. Roughly two thirds of patients remained in the ICU for more than a week.

• Blood glucose was targeted between 80 and 140 mg/dl using a system for routine glycemic control in pediatric critical care settings (Preissig et al., PCCM 2008). Repeated values above 140 mg/dl were treated with insulin infusions starting at 0.1 U/kg/hr.
• Every patient was monitored with Medtronic Guardian CGM devices. The Guardians were “functionally blinded” – placed in baskets beside patients’ bedsides. Alarms were set at 70 and 200 mg/dl, and although all therapeutic decisions were based on approved blood glucose monitoring devices rather than CGM readings, nurses were instructed to perform confirmatory glucose checks whenever the alarms activated. The Guardians were placed in the thigh or abdomen, they took measurements every five minutes, and they were changed every five days (we note that Guardians are approved for only three days of use). 47 patients using 89 sensors had usable CGM data, yielding over 64,000 readings.
• The study’s full results were recently published as Bridges et al, Critical Care 2010.
• Of the 1,555 interstitial CGM measurements taken within five minutes of a reference blood glucose value, 97.9% fell into zone A or B of the Clarke Error Grid (A=74.6%, B=23.3%, C=0.4%, D=1.7%, E=0.0%). Mean absolute difference (MAD) was 19.2 mg/dl, and mean absolute relative difference (MARD) was 15.3%. Clarke A + B scores were comparable regardless of CGM site, pooled illness severity score, length of ICU stay, above 95% even in such challenging populations as those with altered perfusion, edema, insulin-treated hypoglycemia, or vasopressor therapy. During Q&A, Dr. Rigby noted that although the hypoglycemic alarms rang during the study, no actual instances of hypoglycemia occurred; he said he preferred technologies that err on the side of caution.
• In Dr. Rigby’s (and our) opinion, these results are a strong refutation of many of the theoretical arguments against existing CGM in the ICU. He pointed out that these accuracy scores are comparable to those seen in outpatient trials of subcutaneous CGM.
• Dr. Rigby believes there is a strong clinical need to integrate continuous glucose monitoring in the ICU. Moreover, he argued that subcutaneous CGM is superior to intravascular CGM due to concerns about access (a “huge” issue), infection, thrombosis, and mobility. He said that for the foreseeable future subcutaneous CGM should be used only as an adjuvant rather than to directly inform therapy, but he noted that surrogate devices are regularly used in the ICU and that CGM tracking and trending data would be very valuable to trigger routine blood glucose checks.
• Although this was not addressed during Dr. Rigby’s presentation or the Q&A, we note that Medtronic is developing a subcutaneous, electrochemical CGM device for use in the ICU. On October 9, 2010 at the First Annual International Hospital Diabetes Meeting in San Diego, Dr. Rebecca Gottlieb presented feasibility data showing that the sensor is predictive of glucose trends, accurate down to 50 mg/dl, and resistant to interference from acetaminophen and ascorbic acid.

KEYNOTE ADDRESS: GLUCOSE TESTING IN THE CLINICAL SETTING – RISKS AND RESPONSIBILITIES

Alberto Gutierrez, PhD (Office of in Vitro Diagnostic Devices, FDA, Silver Spring, MD)

In an era when the FDA has had an increasing influence on diabetes technology, it seemed appropriate that Dr. Alberto Gutierrez gave the keynote address to lead off the conference. After proclaiming, “I am a bureaucrat, but I am here to help,” Dr. Gutierrez gave a brief overview of the FDA’s mission and the approval process for medical devices. Turning to off-label use of medical devices, he emphasized that the risks and responsibilities are greater and that they shift to clinicians. In these cases, clinicians must really understand the performance of the device in the intended population; he added that the FDA’s and manufacturer’s roles are less clear with such non-approved use of medical devices. Instances of off-label use are especially difficult because it’s hard for manufacturers to determine risk, and the failure of off- label medical devices may not meet the standards of a recall. In Dr. Gutierrez’s opinion, common off- label use brings with it a corresponding duty of manufacturers to design and study devices that meet the off-label clinical need. (In the Q&A, he discussed the example of in-hospital SMBG use.) To close his keynote address, the speaker briefly addressed issues of patient protection with investigational medical devices, cases where the rules are also not clear. In these cases, IRB approval and informed consent are needed along with FDA oversight. Dr. Gutierrez argued that the FDA can play a role in riskier devices and made sure to reemphasize that any off-label use increases the risk and responsibilities of those using the devices.

PERFORMANCE OF WEARABLE RAMAN SPECTROMETER FOR NONINVASIVE GLUCOSE MEASUREMENT

Rudy Hofmeister, PhD (Executive Vice President, C8 MediSensors, Inc., Los Gatos, CA)

In an interesting examination of a new technology, Dr. Rudy Hofmeister gave an overview of the design, clinical studies, and current progress on the C8 MediSensors noninvasive continuous glucose monitor. The system has been under development since 2004 and was designed with high standards in mind: accuracy (i.e., on par with existing CGM – we were surprised by this comparison since although CGM is valuable in our view, accuracy isn’t yet an advantage associated with these systems), noninvasive (i.e., painless with no skin puncture), continuous (i.e., measurement at least every 15 minutes – this is a different view of continuous than some would give, but it is certainly more frequent than BGM makes possible), nonintrusive (i.e., “like a watch – you use when need it, otherwise you forget about it”), and affordable (i.e., at a cost of two fingersticks a day). The sensor will also feature linking with an iPhone or Android, offering the corresponding mobile capabilities of these devices. A benefit of the C8 MediSensors device is the universal calibration factor, potentially boosting ease of use for patients. Small human clinical trials of the device have achieved solid accuracy, with a MARD of 11.5% and 100% of points in the A and B zones of the Clarke Error Grid. If these data can be replicated in larger clinical studies (this has been a big ‘if’ for previous noninvasive efforts), this technology could represent a breakthrough in the field. There appeared to be much interest in the audience about the device, and as expected, important questions were posed over how the technology actually works, whether the data can be uphold in a larger clinical trial, if the technology could be produced at scale, and if it will be convenient for patients to use.

• The C8 noninvasive CGM has been in development since 2003, with progressive improvements and constant adherence to company-imposed “critical success factors.” In 2003, C8 MediSensors, Inc. was founded with the charter of developing a noninvasive CGM. From 2004-2008, the company looked at every technique available for noninvasive CGM and settled on Raman spectroscopy. According to the speaker, this allowed the team to best meet five critical success factors: accuracy, noninvasiveness, continuousness, nonintrusiveness, and affordability. During this time, C8 MediSensors built a proof of concept, conducted clinical tests, and adopted a universal calibration. In 2009, the firm attempted to miniaturize the device, which is now wearable and about the size of a business card. C8 MediSensors is now proceeding with FDA and CE Mark regulatory approvals.

• Because the design of this sensor combines the five critical success factors and a universal calibration, Dr. Hofmeister characterized it as ideal for patients. The design leveraged results of the technology bubble in the late 1990s and early 2000s. C8’s sensor is a novel application of existing, well-established consumer technologies, making the production cost relatively low. The monitor takes measurements every five or 15 minutes (or can spot check), can be worn on the abdomen or thigh, and uses Bluetooth to communicate with an iPhone or Android mobile phone. The mobile application will also feature recent measurements, trends, an overlay of user entered activities, emergency contact notification, and data uplink. From our view, user interface will be critical to see if it is established that the system works.
• Dr. Hofmeister explained that the universal calibration factor makes the C8 Sensor easier to use than traditional CGM. It requires no calibration to the user and works “out of the box.” Using the example of a digital thermometer, Dr. Hofmeister mentioned that such a device would be entirely different if we had to have a different thermometer for every member of household, or the thermometer had to be recalibrated. All results presented are from universal calibration.
• In human clinical trials performed at Dr. David Klonoff’s lab at Mills-Peninsula Health Services (San Mateo, CA), the C8 sensor achiever a MARD of 11.5% with 100% of points in the A and B zones of the Clarke error grid. Tests were 90 minutes long, and patients were tested with glucose values upon entering the clinic and confirmed with YSI and HemoCue measurements every 15 minutes. These results were in a small study, but have improved significantly in the last year (MARD 18%, 98% in A and B). Dr. Hofmeister suggested that data should continue to improve as more studies occur.

MEASURING HYPERGLYCEMIA AND HYPOGLYCEMIA WITH NONINVASIVE NEAR- INFRARED SPECTROSCOPY

Mark Arnold, PhD (University of Iowa, Iowa City, IO)

Dr. Arnold described promising pre-clinical data in noninvasive continuous glucose monitoring, a field that he noted has been set back by previous, premature announcements of success. He listed several strengths of optical monitoring: the process is painless, nondestructive, and biocompatible, it allows for hypoglycemic alarms, and it can incorporate the detection of multiple analytes (possibly important for more comprehensive future systems, Dr. Arnold noted). Unfortunately, early noninvasive technologies based on changes in the skin’s optical properties were not specific to glucose, which has a weak absorption spectrum relative to many other common amino acids and sugars. Dr. Arnold also explained that multivariable statistical analysis can produce false correlations, leading to a need for confirmatory analytical methods (which have often been neglected in the past). He then presented findings from his own group’s work with the transmission of near-infrared radiation to test blood sugar in live rats, a method that yields glucose-specific signals confirmable by three separate methods (partial least squares [PLS] statistical modeling, net analyte signaling [NAS], and hybrid linear analysis [HLA]). He concluded that noninvasive near-infrared spectroscopy can collect glucose-specific in vivo information at both hypo- and hyperglycemic concentrations, with much greater reliability than was seen with noninvasive efforts of years past. Dr. Arnold made an intriguing case for optical monitoring technology’s prospects, and we are interested in learning more about this “strong foundation for success” in an area that looks to still have much building ahead.

PANEL DISCUSSION

Bruce Buckingham, MD (Stanford University, Palo Alto, CA) and Darrell Wilson, MD (Stanford University, Palo Alto, CA)

Questions and Answers

Gary Stiles, MD (Children’s Hospital, Boston, MA): I’m part of the randomized controlled trial of CGM in the pediatric ICU that Dr. Rigby described. I concur quite strongly that there are a lot of unsubstantiated rumors about CGM being ineffective in the ICU. We’ve studied 600 babies, all less than three years old, and we’ve gotten similar experiences to Dr. Rigby’s in our study of tight glycemic control vs. standard of care. I have a few questions. You commented on having an alarm set at 70 mg/dl, but you didn’t go through the data on false and true positives. Could you comment on this?

Dr. Rigby: Alarms did go off. In fact, when alarms went off at 70 mg/dl or lower, if anything it was a false concern. It seemed that when blood glucose was dropping and patients were in hypoglycemia according to the monitor, they actually weren’t quite there yet. From a clinical standpoint, this is reassuring; I would rather have the alarm be biased in that direction than the other.

Dr. Stiles: Do you have any comment on the true detection rate? Did the alarm ever fail to detect hypoglycemia?

Dr. Rigby: Actually, in our experience with these patients, we had no patients hypoglycemic by standard glucose readings. We did have some patients who went mildly low, and this was always detected by the CGM.

Dr. Stiles: I also had a comment on your analysis. You didn’t, as most groups don’t, report the slope and offset of the linear regression. We defined a bias in our slope and offset, and this creates problems with where we set the alarm threshold. We find the slope and regression of that data is different in standard of care and tight glucose control; the mean calibration point is lower in one group than another.

Dr. Rigby: No, we haven’t done that type of regression, but I’d be interested in looking at the data and talking about it.

Q: Dr. Gutierrez, thank you for a nice overview of the FDA’s perspective and the requirements for medical devices. What is the guidance for software programs that support metabolic monitoring such as insulin calculators?

Dr. Gutierrez: The agency has come out to say that software is a medical device, but as with everything it depends on what the intended use is. For personal health records, the agency is putting together a view on that issue which will come out soon. We consider insulin dosing calculators class II devices.

Dr. Robert Vigersky (Walter Reed Army Medical Center, Washington, DC): Dr. Rigby, I want to congratulate you for providing the data that we’ve been asking for. However, the data that we’re missing in the inpatient setting is in the hypoglycemic range. We don’t know the accuracy of these devices in hypoglycemia in the ICU. You mentioned a randomized trial for control in the ICU with different levels of control. What were those? It’s been thought that maybe a mean glucose of 150 mg/dl or 140 mg/dl is sufficient.

Dr. Rigby: I totally support the study of these devices in the hospital in the hypoglycemic range. I think the most important thing these sensors can do is alarm if the blood glucose level has reached a threshold. As a clinician, I care if the glucose is lower than 60 mg/dl and then I’ll do a confirmatory test. On your second test on the RCT we just initiated, we’re comparing control of 80-140 mg/dl vs. 90-220 mg/dl.

Many of the current studies out there compare very similar ranges. But it’s very difficult to determine if there is a benefit.

Dr. Buckingham: I use the CGM as a watchdog in the hospital. It’s hard to monitor kids more than once an hour and the CGM picks up the stuff in between.

Q (LifeScan): Dr. Hofmeister, could you tell us about the mathematical algorithm used by your device? My expectation is that it faces the challenges of IR methods, but maybe you could talk about that.

Dr. Hofmeister: First, as to the mathematical basis of our calibration, it’s very similar in many ways to what Matthew Arnold presented. We use multivariate analysis, typically partial least squares (PLS). Sometimes we might use principal component analysis (PCA) or other regression techniques. My general finding is that in machine learning, if you have good data, it doesn’t matter which of these methods you use.

Q (LifeScan): Existing blood glucose monitors tend to be susceptible to interferents like hematocrit changes. I was wondering how this applies to your method.

Dr. Hofmeister: I don’t believe we’ll be subject to interference to the degree that other devices are: common interferents like acetaminophen and changes in hematocrit don’t really affect our calibration for two reasons. First, the RAMAN effect is extremely selective; we can pretty much tell the difference between glucose and anything else. Secondly, even if a therapy produced a signal identical to glucose, it would be present in a small concentration, giving a small signal relative to the true glucose signal.

Q: You mentioned that the use of SMBG for tight glycemic control is considered off-label. Could you comment on glycemic control versus tight glycemic control and what actions are considered off-label?

Dr. Gutierrez: If you look at the use of blood glucose meters in the hospitals, they are intended to monitor. The use of meters in the ICU is off-label: they are not intended for use there or studied there. We know that meters are used in certain ways. Companies get in trouble with the FDA if they make unapproved claims, but sometimes they come very close to promoting that use without presenting evidence. Why are we concerned? Lots of times the meters don’t work properly: they are sensitive to changes in hematocrit, the hydration of the patient, and other issues. When these meters are tested and cleared by the FDA, we are not looking at data in the intended-use population per se, and that carries some risks.

Q: How were your calibration glucose values incorporated into your analysis? You were using a continuous, real-time monitor and YSI to confirm. If you use a YSI calibration value and put it in the CGM, that affects the analysis.

Dr. Arnold: I guess what I can say is that the values from YSI were not on top of the CGM values.

Dr. Jeffrey Joseph (Thomas Jefferson University, Philadelphia, PA): I have a question on the noninvasive monitoring system. What about the practical side of attaching it to the skin, motion, stability?

Dr. Hofmeister: We have spent a lot of time with the ergonomics of this device. The major issue we have encountered is that they are uncomfortable. But the mounting scheme has evolved over time. It’s a very wide sash or strap that goes around the abdomen or leg. The wide strap makes the device comfortable to wear. Tape was extremely uncomfortable, and we stopped using that. Despite the mounting, the device can still move around a bit, and that’s fine for our applications as long as it’s not moving around continuously. If it shifts when you get up from your chair, that’s not a big deal. What you don’t want is the device moving continuously. That represents a challenge, and we have not solved that.

Dr. Arnold: We haven’t focused on mechanical engineering; we’ve been focusing on the selectivity of the measurement. I’d say our current version wouldn’t be practical in the ways you asked about, but that if it’s recognized that glucose data can be extracted from spectroscopy, market pressures can be brought to bear to improve the technology. We’ve been using human measurements from the back of the hand. This works fine from a measurement standpoint, but I’m not sure it would be practical in the field. However, we could apply what we’ve learned to other areas; we’ve looked at the pinna of the ear and the earlobe.

Q: I have a question for Dr. Gutierrez. You mentioned that the current standards apply mostly to type 1 patients for home use. Can you comment on your expectations for the FDA’s standards for the in-hospital use of these systems?

Dr. Gutierrez: This is not an easy question. The law that we have to follow says that we approve devices if we compare them to ones that are legally on the market and find them substantially equivalent. We clear devices for the monitoring of type 1 patients, even in-hospital monitoring. They may be used very differently, and even though we know this, it’s difficult for use to force companies to do studies and make the proper claims. Based on what we’ve cleared in the past, we will continue to clear devices. If we find particularly troublesome safety issues, we may be able to require companies to deal with those issues. For example, in the future, certain point-of-care meters will likely be labeled for use by multiple patients, and we will have the companies make sure the meters include instructions so that they can be cleaned appropriately and used with single-use lancets, so we don’t have the problems with pathogen transmission that Dr. Perz described yesterday. We don’t have the ability to tell manufacturers that if their devices are used in the ICU then they have to come to us with that claim. Would we like companies to file for those claims? Yes. Yesterday Dr. Joseph had a wonderful talk on in-hospital monitoring, and I agree with his points except to say that while NICE-Sugar was indeed flawed, it showed how glucose measurement is or was being currently done across the US, maybe not in Dr. Joseph’s clinic but in other places. NICE-Sugar showed that when you do measurements in the ICU with devices not designed for that, you might not get good results.

Comment: I work at a children’s hospital in Boston and I’d like to comment about my experience with the CGM. In the hypoglycemic range, we have 350 kids in our study and our incidence of hypoglycemia is 2.7%. What we’ve found is that to get the CGM to detect 50% of the excursions that go below 60 mg/dl, we must set a threshold at 70 mg/dl and use a predictive alarm. We care about episodes 60 mg/dl or lower.

Q: These point of care meters are under a lot of criticism. If you have a lab measurement that’s within 3%, it takes a long time to get that information back to the bedside. Which matters: the potentially inaccurate point of care value or the accurate lab value that doesn’t get to the bedside in time?

Dr. Gutierrez: There is definitely use in getting the blood glucose data as soon as possible. But we would like to have data on CGM in the inpatient population. I have seen errors in people with blood sugars of over 600 mg/dl and a meter reading normally. You must ask if the meter is being used in a way that makes sense; did you understand the risks in the populations being used?

Q: My question is for the developers of the noninvasive CGM system. Maltose interference has led to catastrophic events. When you look at the ability of your system to pick up glucose, you should also look at interferences.

Dr. Hofmeister: Certainly that is something we intend to look at. The other sugars, while chemically similar, have unique spectral features. This makes it easy to identify them and ignore them.

Dr. Arnold: We have not verified this in vivo.

Q: Can you comment on pediatric populations wearing multiple sensors at the same time? What makes the CGMs work well or poorly?

Dr. Rigby: I have no experience in the intensive care unit with multiple sensors. I was surprised how well the device worked across the board. We did an extensive analysis, but haven’t found a sub-population that CGM did not work well in. Patients in the ICU for less than three days were actually the group that had the lowest Clarke Error Grid Zone A correlations. This might be expected; CGMs tend to get more accurate over time and these were the patients wearing the devices for the shortest amount of time.

Dr. Bruce Buckingham: The longer they wear them, the better they do. In my experience, the one case where a sensor didn’t work better over time was with a platelet count less than 20,000. Sensors work well in newborns with hypoglycemia. We have not used two sensors at one time, however. But the whole concept here is the watchdog phenomenon. You’re getting many more frequent measurements and can figure out when to test. Blood sugar checks come every few hours, but you need something available all the time. We also changed our alarm so the nurses could hear it.

Novel Formulations of Insulin and Metabolic Peptides – Rapid Acting

MECHANISM OF ACTION, CLINICAL SAFETY, AND EFFICACY OF RECOMBINANT HUMAN HYALURONIDASE (RHUPH20) CO-ADMINISTRATION WITH CURRENT PRANDIAL INSULIN FORMULATIONS

Douglas Muchmore, MD (Halozyme Therapeutics, San Diego, CA)

Dr. Muchmore gave an overview of Halozyme’s PH20, a promising solution to the slow speed of subcutaneously administered insulin. PH20 is based on recombinant human hyaluronidase, a compound that Dr. Muchmore emphasized has been studied for over 50 years and is already approved for speeding the absorption and delivery of certain drugs. Coinjecting PH20 with analog insulin can significantly drop the amount time required for subcutaneous insulin absorption and action in both type and type 2 populations. PH20 can also reduce postprandial hyperglycemia, reduce the occurrence of hypoglycemia, and increase area under the curve. Halozyme Therapeutics is currently pursuing further studies examining the superiority of insulin lispro coinjected with PH20 versus lispro alone. Hyaluronidase is a well-understood compound, with a long and safe clinical history. Hyaluronan forms a barrier to bulk fluid flow in the interstitial space, and the enzyme hyaluronidase converts the gel barrier to a liquid substance. Dr. Muchmore said that toxicity is not a concern, as the enzyme is transitory and turned over by the body at a rate of 5 g per day. The enzyme has a 50-plus year clinical history and has been tested and approved for increased dispersion and faster absorption in a broad range of drugs and fluids.

• Halozyme’s enzyme, recombinant human hyaluronidase (rHuPH20) rapidly increases the dispersion of coinjected molecules and disappears quickly. The enzyme has a rapid half-life of less than five minutes and interstitial permeability is fully and rapidly restored. Compounds coinjected with hyaluronidase spread out over a much broader band of subcutaneous space within 5-10 minutes, resulting in enhanced absorbability.
• “Ultrafast insulin is to analog insulin what analog was to regular insulin.” A plethora of published clinical data has demonstrated faster pharmacokinetic and pharmacodynamic profiles of analog insulins coinjected with rHuPh20. Studies have indicated twice as much insulin exposure in the first hour, greater and earlier peak exposure, and significant differentiation from standard rapid-acting analogs. There is no differential effect of PH20 based on which rapid-actinganalog is used. Additionally, adding PH20 to regular insulin effectively turns it into a rapid-acting analog; a study found almost identical PK/PD profiles between lispro alone and regular insulin + PH20.
• In a review of studies comparing postprandial glucose control with lispro vs. coinjection of lispro and PH20, Halozyme’s formulation led to clinically meaningful declines in peak postprandial glucose in both type 1 and type 2 populations (type 1: ~165 mg/dl to ~130 mg/dl; type 2: ~165 mg/dl to ~140 mg/dl). The studies also showed large increases in the number of patients meeting glycemic goals after meals (excursions above 140 mg/dl reduced by 80% in type 1 study and by 44% in type 2 study). A double-blinded study is ongoing to compare analog insulin alone with analog insulin plus PH20 in the take-home treatment setting.

NOVEL GLUCAGON FORMULATIONS

• Kenneth Ward, MD (Oregon Health and Science University, Portland, OR)In a well-attended afternoon session, Dr. Kenneth Ward gave the audience a true sense of the state of research into glucagon and delved into some promising strategies to stabilize the compound. He started with a biochemical review of glucagon and quickly transitioned to a physiologic explanation of the impaired glucagon response in type 1 diabetes. Turning to some of his bihormonal closed-loop research, Dr. Ward explained that glucagon was successful in preventing hypoglycemia two out of three times. After giving some potential reasons for the failure of glucagon to prevent low blood sugar in his research, he closed with a review of ongoing efforts to stabilize the compound for use in the closed loop. Although fully closed-loop control seems like a long way off, it’s great to see efforts being put towards an artificial pancreas that can deliver more than just insulin.
  • ​The nature of glucagon in type 1 diabetes is an “extremely confusing area” in the field, but an October 1984 study (Lorenzi et al., West J Med) examining diabetes duration and glucagon response helps explain some of the mystery. The study involved two groups of individuals, those with type 1 diabetes for less than ten years and those with a disease duration exceeding ten years. The researchers performed a hypoglycemic clamp down to 30 mg/dl and measured glucagon secretion. They found that those with shorter diabetes duration had significantly higher secretions of glucagon than those with longer duration diabetes. One hypothesis for this decreased response of glucagon may concern the physiology of the pancreas. Alpha cells of the pancreas are surrounded by beta cells, which may suggest that the alpha cells are regulated by the neighboring beta cells. Once these beta cells are destroyed in type 1 diabetes, the alpha cells may function incorrectly.
  • Approximately 20-25% patients with type 1 diabetes have hypoglycemia unawareness; thus, the use of glucagon to prevent hypoglycemia may offer substantial benefits to patients. Patients often keep their blood sugars high to prevent hypoglycemia. As a result, glucagon administration in a bihormonal pump may allow tighter glycemic control.
  • In bihormonal artificial pancreas research at OHSU, glucagon was successful roughly two-thirds of the time in preventing hypoglycemia. The research involved the use of two subcutaneous CGM sensors, subcutaneous glucagon administered via an insulin pump, and subcutaneous insulin delivered via pump. The controller gave small pulses of glucagon as the blood sugar fell. This system typically arrested the decline in glucose. The administration of glucagon via closed-loop control reduced the minutes of hypoglycemia (< 70 mg/dl) per day from40 minutes to just 14 minutes. The system also achieved a median of zero minutes of hypoglycemia at night, which we found quite impressive.
  • In the one-third of cases where the system failed to prevent hypoglycemia, potential reasons were high insulin levels, sensor delay, deficient liver glycogen stores, and pharmacologic instability of glucagon. Often, high insulin on board drove blood sugar down at a rate glucagon could not overcome. Additionally, lag between sensor values and blood glucose values often meant a person was hypoglycemic before glucagon was triggered to be delivered. Dr. Ward also hypothesized that insufficient liver glycogen stores would have rendered glucagon ineffective in raising blood glucose. Finally, the pharmacologic instability of glucagon over time could also have degraded the compound’s ability to raise blood sugar.
  • “The better the sensor, the less need for rescue glucagon.” Imperfect sensor data has many causes, including the puzzling topic of sensor drift. Sensor accuracy can also be affected by background current, and correcting for this factor improves sensor accuracy. Finally, studies also show that calibrating sensors more frequently also helps with sensor accuracy. Nevertheless, when current sensor error rates are extrapolated over a month, the error can be significant.
  • Glucagon has chemical stability problems; one strategy for dealing with this problem is keeping it at a high pH. Dr. Ward’s team found that at a pH of 10, glucagon is relatively stable and does not form amyloid gels, nor is it cytotoxic to cells in culture. Even after 30 days, this effect remained. Dr. Ward said soluble glucagon at a neutral pH may also have potential to deal with glucagon instability.
  • Several companies are developing new forms of glucagon. Biodel is using surfactants in its formulation; this strategy has achieved stability for 50 days and, as a variation of an existing product, Dr. Ward pointed out that this may have a regulatory advantage over Marcadia/Lilly’s novel glucagon analog (in late 2010, Marcadia was purchased by Roche – see Closer Look, December 28, 2010). Many are also finding exciting new delivery mechanisms, such as that of Enject’s – this is considerably easier than the old fashioned glucagon, which requires reconstitution.

THE IMPORTANCE OF INSULIN PHARMACOKINETICS IN CLOSED-LOOP CONTROL: IMPLICATIONS FOR ULTRA-RAPID-ACTING INSULIN FORMULATIONS

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

Dr. Damiano discussed his group’s inpatient trials of bihormonal closed loop control, with an emphasis on the importance of understanding and improving insulin pharmacokinetics (PK). He explained that when the system’s control algorithm was programmed with an estimate of 33 minutes for time to peak insulin concentration, glucagon failed to prevent hypoglycemia for subjects with slower-than-average insulin PK. However, when this estimated time-to-peak was doubled, hypoglycemia could be successfully prevented in all subjects. Although avoiding hypoglycemia with current rapid-acting analogs requires the trade-off of higher average blood glucose, Dr. Damiano said that ultra-rapid- acting analogs with lower PK variability could enable much tighter control than is now possible.

• In a 27-hour study of bihormonal closed-loop control in 11 subjects, a customized MPC algorithm delivered insulin lispro (Eli Lilly’s Humalog) in response to post- meal glucose rises and glucagon in response to subsequent glucose lowering. Dr. Damiano’s group found that glucagon was not consistently successful in averting hypoglycemia, in large part due to the high inter- and intrapatient variability of insulin pharmacokinetics (PK). He explained that the algorithm was initially programmed to assume a 33-minute time to peak insulin concentration, but that retrospectively analyzed measurements of blood insulin levels invalidated this estimate. Although the 33-minute estimate worked well for some patients, hypoglycemia tended to occur whenever a subject’s actual PK was longer than 71 minutes.
• To account for the slow insulin PK in many patients, Dr. Damiano and his colleagues roughly doubled their algorithm’s original 33-minute estimate of insulin time-to-peak. They considered developing individualized assumptions for each patient, but found that intrapatient variability was too high to make this worthwhile: “A person with diabetes is ten people with diabetes.” In subsequent experiments using the longer PK assumption, hypoglycemia was consistently prevented in all patients, although average blood glucose increased in those whose actual PK was faster than the new assumption. Notably, with the new version of the algorithm, hypoglycemia could be prevented with physiologic levels of glucagon, suggesting that a bihormonal closed-loop system would not deplete patients’ glycogen stores.
• To illustrate a best-case scenario for closed-loop glucose control, Dr. Damiano presented an experiment using intravenous infusions of both insulin and glucose, given to stress-induced-hyperglycemic pigs with diabetes. Hypoglycemia was completely avoided, even though dosage was not individualized except by each pig’s weight. Most impressively, Dr. Damiano overlaid data from the response of non-diabetic pigs to stress-induced hyperglycemia, and the curves matched almost perfectly.
  • ​Dr. Damiano said that although subcutaneous insulin will never have as good an action profile as intravenous, new rapid-acting insulins could substantially improve control. He noted that Biodel’s Linjeta VJ7 has a time-to-peak of 23 minutes, roughly one third of lispro’s 60 minutes (or 65 minutes, as Dr. Damiano’s group found). Based on results from the first trial, subjects with relatively fast PK had 18 mg/dl declines in average blood glucose when the algorithm assumed a 33-minute time to peak insulin concentration rather than 65 minutes – even though all of these subjects’ actual time to peak insulin concentration was considerably slower than 33 minutes. If an ultra-rapid acting insulin formulation like Linjeta were used, which really is more than twice as fast as lispro, Dr. Damiano said, “I’d expect at least this kind of benefit.”
  • Early results are favorable in Dr. Damiano’s team’s second inpatient trial of bihormonal closed-loop control. Several variables were changed from the first study: the experiments are longer (48 hours), the algorithm is based on CGM (Abbott Navigator) rather than reference blood glucose values (GlucoScout), subjects exercised moderately for roughly 30 minutes, and pre-meal priming boluses were given. After almost 200 patient-hours, there has been no hypoglycemia (minimum recorded value 72 mg/dl), and mean blood glucose is 136 mg/dl. Dr. Damiano closed his talk by asking the audience, given these impressive results, to imagine what would be possible with faster and less variable insulin pharmacokinetics.

THE DEVELOPMENT OF THE GEN2 INHALER FOR AFREZZA: BIOEQUIVALENCE BASED UPON THE SYSTEMIC DELIVERY OF INSULIN VIA THE LUNGS

Mark Marino, MD (MannKind Corporation, Paramus, NJ)

Dr. Mark Marino gave a technical summary of the redesign of MannKind’s inhaler for Afrezza. To start, Dr. Marino outlined the difficult factors involved in systemic delivery of insulin via the lung, including particle size and physics, device design, and user interface. Transitioning to the delivery device itself, he compared the old MedTone delivery device to the newer Gen2 device, which was developed using a bioequivalence strategy. The Gen2 retains the properties of the earlier MedTone device, but the new device requires 33% less powder (now 20 units in the Gen2 device) and lower patient inspiratory effort to deliver the same insulin dose. According to Dr. Marino, the development of the Gen2 device was novel in several respects. It involved an extensive collaboration between engineers and clinicians. More importantly though, it is the first inhaled drug for which a device switch was done based upon a bioequivalence approach. In using such an approach, MannKind applied a staged approach to clinical trials: (1) relative exposure to the principal excipient in Technosphere Insulin powder (FDKP); (2) dose finding and insulin variability; (3) insulin bioequivalence between devices. Although Afrezza is currently under review by the FDA, the improved Gen2 device appears to be a solid start to a commercialized product, especially relative to the poor design of the Exubera inhaler.

THE EVOLUTION OF PEN AND NEEDLE DEVELOPMENT

Timothy Bailey, MD, FACE, CPI (AMCR Institute Inc., Escondido, CA)

Dr. Bailey delivered an overview of insulin needles and pens, discussing developments in features and safety. After summarizing the many specifications and features of pens on the market (Bailey et al., Diabetes Technology and Therapeutics 2010), he reviewed pens’ general advantages over syringes: greater convenience, less obtrusiveness and social stigma, and shorter needles. However, Dr. Bailey noted that pen uptake has been low in the US relative to the rest of the world, a phenomenon he attributed to the pricing and perceived costs of pens, the potential administrative hassle of getting reimbursement, and patients’ unawareness of pens’ availability. Turning specifically to needles, Dr. Bailey noted that the progression toward smaller and smaller needles has improved not only comfort and efficacy, but potentially safety as well. He cited recent research on skin thickness funded by Becton Dickinson that revealed skin thickness varies minimally across adults (almost always between 2.0 and mm) regardless of gender, age, ethnicity, or even BMI – suggesting that essentially no one needs 8mm or 12.7 mm pen needles (Gibney et al., Curr Med Res Opin 2010). In a related clinical study that Dr. Bailey helped conduct, three-week comparisons were conducted between 4 mm needles (BD’s new 4 mm, 32 gauge Nano pen needle) versus 5 mm needles or versus 8 mm needles (Hirsch et al., Curr Med Res Opin 2010). The 4 mm needles demonstrated no significant differences in leakage or glucose control across insulin doses of up to 40 units, and patients in the unblinded study gave the shortest needle favorable responses on ease of use, pain, and overall preference. Closing on a higher-tech note, Dr. Bailey showed John Walsh’s vision for a durable, sophisticated pen with features like Bluetooth connectivity, insulin on-board calculation, and a needle that was retracted when not in use – the sort of progress that Dr. Bailey said would make therapy safer and more effective for patients.

PANEL DISCUSSION

Questions and Answers

Q: Could you imagine the possibility of a fixed-ratio formulation of a rapid-acting insulin with a delayed-action glucagon?

Dr. Ward: I have a feeling you couldn’t titrate the timing as well as you’d need to. To comment on the ratio, we’ve been trying to understand this. If there were an insulin dose, could you overcome it with glucagon? Dr. Muchmore referred to instances of using up to 1,000 units of insulin to induce coma in psychiatric patients. With glucagon, these patients could be rescued from coma after 90 minutes, from blood sugars of something like 10 mg/dl. Even with these massive insulin doses, glucagon still worked.

Dr. Damiano: What was the glucagon dose?

Dr. Ward: That was at a time when glucagon was only about 50% bioavailable. So they gave 4 mg, but that’s really more like 2 mg.

Dr. Damiano: To your question, any system using both glucagon and insulin would have to be timed; you couldn’t just deliver both at once and hope that the pharmacokinetics worked out.

Q: Fast and slow PK…I’m puzzled by this categorization. Once insulin is absorbed, it’s handled the same way. However, what has been clinically shown is that body mass index has a huge impact on absorption. Is it better to use a BMI adjustment in the closed-loop algorithm?

Dr. Damiano: We actually didn’t see any correlation between BMI and PK in our studies. Our studies are small (~12 subjects), but we saw no correlation. We did see a correlation with antibodies and insulin PK. We are about to begin a study to further investigate this .

Dr. Vigersky: To those using glucagon as a second hormone, I’m wondering about the connection with growth hormone. Have you looked at growth hormone with small-dose administration of glucagon? Could glucagon affect growth hormone over time? What about cancer risk?

Dr. Ward: Glucagon induces secretion of many things in the body, although I hadn’t thought of cancer issues. Growth hormone might also have an effect on glycemic control. It’s possible that repeated high doses of glucagon could cause insulin resistance. But I don’t think this is an issue because the doses of glucagon in bihormonal closed-loop studies are really small.

Dr. Klonoff: I have a couple questions for the those using glucagon in bihormonal systems. Are you expecting to identify a certain group of subjects that would be the first to receive bihormonal therapy? Would you target therapy to those with greater hypoglycemia or longer duration of diabetes?

Dr. Ward: This issue of glycogen stores is very important with glucagon.

Dr. Damiano: A JDRF study group identified three populations and placed them into a hierarchy. First are people who have lots of hyperglycemia – this might be subjects with A1cs of 9% or 10%. Then, there is the group with lots of hypoglycemia or even hypoglycemia unawareness. Finally, you’ve got a group of people in the middle. The first target population is the group of people in the middle. For any closed-loop system, you want to target the middle group first.

Dr. Klonoff: As an engineer, do you see problems building a pump with dual chambers?

Dr. Damiano: My concern is that the regulatory pathway for glucagon formulation will take a long time. Developing a dual chambered pump will only be driven by glucagon that’s on the market. When pump companies see a stable formulation coming to market, I believe those devices will come to market.

Q: Could you comment on how alcohol might affect the deployment of glucagon?

Dr. Ward: Alcohol is known to block some enzymes for glucose production, and this might inhibit the glucagon effect. So you would go easy on the nighttime wine.

Dr. Damiano: It would be incumbent on you to keep glycogen stores high; if you fast you can get into trouble. There’s a responsibility that the subjects wearing this device will have to the device and that they will have to be conscious of.

Dr. Klonoff: I was just going to talk about methods to keep glycogen stores high; you already addressed some of this. When insulin and glucose are high, they should build up stores of glycogen.

Novel Formulations of Insulin and Metabolic Peptides – Long Acting

RE-ENGINGEERING INSULIN: CAN WE IMPROVE ON MOTHER NATURE?

Alexander Fleming, MD (Kinexum, Harper’s Ferry, WV)

To lead off the session on novel formulations of insulin, Dr. Fleming, who formerly headed diabetes drug review at FDA, examined the challenges, opportunities, and candidates for better formulations of insulin. He noted that improving on Mother Nature’s design of the insulin molecule, secretion, and delivery is a formidable proposition. Despite advances in insulin analogs and formulations, unmet clinical need encourages continued efforts. Dr. Fleming believes convenience is becoming a much less important driver of this hunt in the modern era of small needle, pen injection. Meanwhile, the downsides of conventional insulin therapy such as hypoglycemia, weight gain, and dyslipidemia are becoming more important. The long sought oral insulin route and other fast-acting and ultra long-acting insulins approaches tantalize with the prospects of mimicking physiology and achieving better clinical outcomes, though each approach has its challenges and risks of unanticipated consequences. Turning to the clinical and regulatory challenges of insulin development, Dr. Fleming expressed serious concern that the cardiovascular outcomes trials now required for oral type 2 diabetes therapies would be extended to some insulin products. In his opinion, these regulations have already had devastating consequences on the development of oral type 2 diabetes therapies. He pointed out that alternately delivered insulin products as add-on therapies for T2DM may have an easier approval route if the FDA does not insist on head to head non-inferiority comparisons with conventional injected insulin, given that insulin is highly effective and tough to beat on the efficacy side. To close, Dr. Fleming made it clear that whether or not it is possible to improve on Mother Nature, “We need to try.”

• The endocrine system is very complex and the more we study it, the more difficult it seems to improve on Mother Nature. The right amount of insulin is needed at the right time, the right, place, the right receptor, and at a cost that’s acceptable in the coming era of analog biosimilars (generics). According to Dr. Fleming, improvements in this space will lead to more convenience, no needles, better metabolic control, positive bodyweight effects, less hypoglycemia, fewer safety concerns, and, in some cases, greater affordability. Biosimilar insulins will provide competition from a pricing standpoint, benefitting patients but making it more challenging to commercialize new approaches.
• New insulin formulations should address unmet clinical needs. Dr. Fleming noted that convenience is less important since we have almost painless modes of delivery with current technology. Oral insulin has been long pursued for its convenience and potential clinical advantages such as reduced risk of hypoglycemia and weight gain. However, unanticipated consequences could result. As an example, intraperitoneal pump delivery of insulin has been linked with subcapsular hepatic steatosis.
• There are a number of new insulins in development, including oral, ultra fast- acting, and ultra-long-acting. Oral insulins face challenges of low bioavailability and consistent performance within and across individuals, but they provide very rapid-acting prandial coverage (see review in Heinemann et al., Journal of Diabetes Science and Technology, 2008). Ultra fast-acting insulin formulations may offer less hypoglycemia and better glycemic control (e.g., Biodel, MannKind, Halozyme). Ultra-long acting insulins (e.g., Flamel’s FT-105, Novo Nordisk’s degludec) may have the benefits of lower day to day variability, improved glycemic control and less risk of hypoglycemia compared to glargine in addition to the potential added convenience of less frequent administration. Smart insulins – insulin molecules that modulatetheir glucose lowering activity based on ambient glucose concentration – are highly interesting. But smart insulins face challenges analogous to those seen with current insulin formulations used in closed loop pumps: both must be able to respond to very rapidly changing insulin requirements.
• New insulin formulations also face clinical design and regulatory challenges, problems that could become more difficult in the near future. The requirements for cardiovascular outcomes trials for oral type 2 diabetes therapies have not been imposed on alternately delivered insulins, including oral insulins, but it’s not clear that this will continue to be the case. This regulation has had a devastating impact on the development of oral type 2 diabetes therapies, according to Dr. Fleming. Additionally, it’s unclear if FDA will be satisfied with comparing insulin products to placebo as add on to background therapy, as is the primary study design for conventional oral T2DM therapies. Requiring non-inferiority comparisons to injected insulin, such as is the case for approving injected insulins, would be a more challenging proposition. Injected insulin analogs are hard to beat or even match on the efficacy side, but Dr. Fleming said that before any other benefits can be claimed by a novel insulin therapy, non- inferiority on glycemic control must first be demonstrated.

LONG-ACTING INSULIN ANALOGS: CLINICAL GOALS AND MOLECULAR STRATEGIES

Michael Weiss, MD, PhD (Case Western Reserve University, Cleveland, OH)

Dr. Weiss spoke on the aims, approaches, and future of long-acting insulin therapy. He described the mechanisms of sanofi-aventis’ Lantus (insulin glargine) and Novo Nordisk’s Levemir (insulin detemir) and degludec, and he introduced “zinc stapling,” a technique developed by Dr. Weiss at Case Western Reserve University and licensed (along with dozens of other insulin analogs) to Thermalin Diabetes, a company co-founded by Dr. Weiss.

• Dr. Weiss stated that the profile of the ideal basal insulin would include 1) reliable, “true” 24-hour (or longer) kinetics, 2) fewer episodes of severe hypoglycemia, 3) less nocturnal hypoglycemia, 4) larger improvements in A1c, 5) little or no weight gain, 6) longer shelf life at or above room temperature, and, in recent years, and 7) low mitogenicity.
• Insulin has many levels of structure that are relevant to therapy and drug development. Dr. Weiss said these structures include: 1) folding intermediates (relevant to manufacturability), 2) folded monomers (half-life), 3) active monomers (interaction with the receptor) – the actual structure of which Dr. Weiss said is still unknown (relevant to potency), 4) self-assembly (relevant to formulation and storage), 4) misfolding/modifications (relevant to chemical degradation), and 5) amyloids (the principle mechanism of physical degradation during storage). Because molecular changes that alter one property will likely affect others as well, insulin analog design involves trade-offs.
• The pharmacokinetics of insulin depend on how quickly zinc-stabilized hexamers dissociate into insulin dimers, as well as how quickly these dimers dissociate into monomers. Long-acting or short-acting analogs can be developed through complementary strategies of stabilizing or destabilizing these hexamers.
• One strategy of hexamer-stabilization involves altering insulin’s isoelectric point (the pH at which a molecule is electrically neutral and thus most likely to precipitate out of solution). sanofi-aventis’ Lantus (insulin glargine) is formulated at pH 4, where it is soluble. However, glargine is modified so that it precipitates at physiological pH to formsubcutaneous depots upon injection. The acid formulation, however, adversely affects the room- temperature shelf life of Lantus. Dr. Weiss said that Eli Lilly is reportedly developing a long- acting insulin candidate based on isoelectric precipitation, although this project does not appear in Lilly’s published pipeline.
• Other long-acting insulins achieve a slow action profile by binding to the body’s albumin. Insulin detemir (Novo Nordisk’s Levemir) is attached to fatty acid chains to enable association with albumin, achieving long action by extending half-life of insulin detemir in the blood rather than through a prolonged subcutaneous depot. Dr. Weiss noted that this approach results in shorter duration of action than Lantus.
• Insulin degludec, Novo Nordisk’s long-acting phase 3 candidate, uses a mechanism that Dr. Weiss called “novel and not straightforward.” Degludec is stored as a phenol- stabilized dimer of hexamers. Dr. Weiss highlighted the role of phenol: it diffuses away after injection, allowing zinc-stabilized multihexamer assemblies to form. Then over the next 24-72 hours, active insulin monomers are gradually released. Dr. Weiss showed phase 2 data that suggests degludec kinetics have high inter- patient variability, but that for some patients, Monday-Wednesday-Friday dosing might be possible.
• Dr. Weiss said that “zinc stapling,” his own group’s method of extending insulin life with multi-hexamer assemblies, is inspired by the mechanism of zinc finger domains. Zinc finger domains, best known for their importance in DNA- and RNA-binding proteins, bind zinc with two pairs of histidines that are four amino acids apart. To translate this mechanism to insulin, Dr. Weiss and his colleagues created an analog with histidines at positions A4 and A8 to create a “half” zinc-binding motif. Every disk-shaped hexamer thus has three half- sites on each face. This means that hexamers can stack in massive assemblies, with three zinc ions bound in the plane between successive hexamers.
• Dr. Weiss presented pre-clinical data on the pharmacokinetics of the zinc-stapled analog showing that in a rat model of diabetes (STZ-Lewis), it had an action profile as long as Lantus’.
• Turning to safety, Dr. Weiss explained that improved receptor selectivity is potentially important for reducing the documented link between insulin (endogenous or administered) and cancer. Among other possible mechanisms, the insulin- like growth factor 1 (IGF1) receptor has been implicated in mitogenicity: insulin analogs with higher cross-binding to IGF1 receptor than wild type human insulin, such as AspB10 insulin, have been shown to accelerate cancer cell growth in cell culture and to increase the incidence of breast cancer in SD rats. Dr. Weiss showed assay results of cross-binding study in which the zinc-stapled analog was identical to human insulin in affinity for the insulin receptor, but had 10-fold lower affinity than human insulin for the IGF1 receptor (with insulin glargine showing a threefold higher IGF1 receptor affinity than human insulin).
• Dr. Weiss closed his talk with an acknowledgement of other potential approaches to extending insulin action. He mentioned Insulet’s OmniPod and Altea’s transdermal PassPort (phase 2) as systems that use rapid-acting analogs to produce long-term effects. He also mentioned alternative delivery solutions including Flamel’s FT-105 (see Dr. Chan’s talk in this session), Aphios’ microspheres, amyloid-based depots (Gupta et al., PNAS 2010), and smart cells or polymers that could deliver insulin in response to ambient glucose concentration.

FT-105, A LONG-ACTING BASAL INSULIN OFFERS IMPROVED PHARMACOKINETIC AND PHARMACODYNAMIC PROFILES COMPARED WITH LANTUS

You-Ping Chan, PhD (Flamel Technologies, Venissieux, France)

Dr. You-Ping Chan continued the focus on insulin formulations by discussing Flamel Technologies’ ultra-long-acting FT-105 insulin. It employs the firm’s MEDUSA technology, which is based on poly- amino-acid nanocarriers that allow for the sustained release of biologics. The amino acids used are biodegradable and biocompatible; MEDUSA is also degraded to GRAS compounds, vitamin E and glutamic acid. Turning to the insulin product more specifically, it is a dry powder that is reconstituted, has a stability of two years, and can be injected through a small 30/31 gauge needle. According to Dr. Chan, Ft-105 offers benefits over Lantus and Levemir, insulins that he said must often be administered twice per day (this is less true for Lantus). Dr. Chan said that early, short-duration human clinical trials of FT-105 have favorably compared the drug to Lantus, offering “much flatter” PK/PD profile over 48 hours. In repeated dose studies in rats, FT-105 was administered once a day over a seven-day period and demonstrated longer duration of action, a flatter PD profile, and a lower peak-to-trough ratio compared with Lantus. The studies have also shown that after three or four daily administrations, a steady-state basal rate was reached in rats. However, we note the smaller size of rats increases the relative size of the subcutaneous space, so these results do not imply similar efficacy in humans. Nevertheless, we hope that further trials on FT-105 continue to progress, as an ultra-long acting insulin with a flatter profile would certainly offer benefits to patients: potentially less weight gain, possibly less hypoglycemia, less variability, fewer safety concerns, and fewer injections.

EXENATIDE ONCE WEEKLY FORMULATION: ENABLING A NEW THERAPY FOR TYPE 2 DIABETES

Christopher Rhodes, PhD (Amylin Pharmaceuticals, Inc., San Diego, CA)

Dr. Rhodes reviewed the development history, delivery platform, and clinical data of Bydureon, Amylin’s once-weekly exenatide formulation that was dealt a surprisingly negative complete response letter from the FDA earlier this year (see Closer Look, October 20, 2010). This talk included an overview on Byetta’s (exenatide twice daily) therapeutic successes, an outline of Alkermes’ Medisorb microsphere technology that enables Bydureon’s long action profile, and photos from the Bydureon manufacturing site in Wilmington, OH. Dr. Rhodes also presented summaries of Bydureon clinical trial data (the DURATION program), reminding us once again of how much potential this therapy holds for effective and tolerable care.

PANEL DISCUSSION

Moderators: Thomas Forst, MD (University of Mainz, Mainz, Germany) and Howard Wolpert, MD (Harvard University, Boston, MA)

Questions and Answers

Comment: I’m a simple electrochemist, and I learned about insulin formulations only recently. I realized that phenol and cresol are put into formulations. These molecules are electrochemically active. One goal for future formulations could be not to use electrochemically active molecules, since they could confound CGM or SMBG sensors.

Dr. Weinzimer: What’s the fate of vitamin E in your Ft-105 polymer? If I’m not mistaken, there have been issues of hepatic toxicity with vitamin E in the past.

Dr. Chan: The polymer degrades to glutamic acid and vitamin E. We use small doses that result in about 10 mg of vitamin E. It should not produce any toxicity and the vitamin E is systemically absorbed.

Dr. Kenneth Ward (Oregon Health and Science University, Portland, OR): I have a comment for Michael Weiss. Your last slide had a comment about insulin amyloid for long- acting therapy. In our lab we’ve found that amyloid formulations of glucagon get absorbed but are a little delayed. Still, my guess is that it’s very hard to get an amyloid formulation approved, given safety concerns including that amyloidosis is associated with Alzheimer’s Disease and prions.

Dr. Weiss: I share those safety and toxicity concerns – both short-term in the lungs and long-term promiscuity and seeding. But I wanted to include that publication in PNAS as a collegial acknowledgement of peers in the field.

Q: I have a question concerning your PK/PD results. I’m wondering if there was a significant difference between the 0.3 unit and 0.6 unit doses because they looked very similar.

Dr. Chan: Yes, there was a significant difference in PK/PD profiles between the two doses.

Telemedicine for Diabetes

TECHNOLOGY TO GO: EVOLVING THE NEXT GENERATION OF TELEMEDICINE

Suzanne Boren, PhD, MHA (University of Missouri, Columbia, MO)

Dr. Suzanne Boren gave a broad summary of how the Internet, mobile phones, and social networks are filling a void in diabetes. These technologies are enhancing traditional healthcare delivery by improving patient understanding, boosting patient-health care provider communication, reducing social isolation, and potentially, improving outcomes. Although there are many questions to consider in adopting these types of devices, there’s no doubt that they have significant benefits for people with diabetes.

• The Internet has allowed patients with diabetes to be more active participants in their therapy. Of all adults with a chronic disease, 93% ask their health care professional when dealing with a medical issue, 83% look online for health information, 20% of users create online health content, and 25% of users have used a social networking site. The most useful features of Internet technology vis-à-vis diabetes include ability to log and track blood glucose values, electronic reminders, scheduling physician visits, accessing relevant information, emailing the healthcare team, and social networking. The speaker also noted that half of Americans are interested in using a personalized health record accessible electronically through the Internet.
• Mobile devices offer a variety of benefits for diabetes management: patients can now access diabetes applications, get general information, receive reminders, and send information to their diabetes care team on the go. According to Dr. Boren, as of December 2009, approximately 42% of Americans owned smart phones. Such phones are important in diabetes, as they allow the use of sophisticated diabetes applications like insulin calculators, education, diet-logging, nutritional advice, exercise information and logging, alerts, and blood glucose logging.
• User generated content and social media sites are now a major source of advice for diabetes management. In an evaluation of the 15 largest Facebook groups related to diabetes. 66% of posts included unsolicited sharing of diabetes management strategies, 29% includedemotional support (we thought this was quite low), and 27% featured promotion of products. There are now a wide variety of online communities and social networking sites for diabetes: ones noted included Children With Diabetes, dLife (by far the largest), Diabetes Daily, Diabetes Friends, Diabetes Sisters, Talkfest, Connect, Juvenation, My Diabetes, TuDiabetes, We Are Diabetic, and Diabetic Rockstar. Common elements of these sites include blogs, forums, news, groups, and occasionally recipes, TV shows, journals, and articles.

NUMERACY AND LITERACY: THE ELEPHANTS IN THE DIABETES TECHNOLOGY ROOM

David Kerr, MD (Bournemouth Diabetes and Endocrine Centre, Bournemouth, UK)

Dr. Kerr called into question the practicality of telemedicine, suggesting that low rates of health literacy and numeracy might lead to disappointing results and adherence. He suggested that low rates of health literacy translate to worse diabetes care in both clinical trials and the real world, and he encouraged the audience to target their interventions and communications in light of patients’ abilities. Also, highlighting the importance of social media, Dr. Kerr urged healthcare providers and industry to develop better ways to monitor and use sites like Facebook to promote better health.

• Noting that technological advances have not led to ideal care, Dr. Kerr challenged a statement of Dr. David Klonoff in 2003: “telemedicine is rapidly maturing… sound, effective, cost-effective, and practical” (Diabetes Care). Dr. Kerr acknowledged that telemedicine was sound and (at least in the short-term) effective. Off-handedly remarking that it was probably not cost-effective, Dr. Kerr explained that his presentation would examine whether telemedicine is practical.
• Dr. Kerr cited several recent studies showing somewhat disappointing results and/or adherence, and he suggested that these problems might relate to low levels of health literacy, health numeracy, and/or technological proficiency. These studies included Larsen et al., Journal of Telemedicine and Telecare 2010; Stone et al., Diabetes Care 2010; McManus et al., Lancet 2010; Shea, American Clinical and Climatalogical Association, 2007.
• Dr. Kerr cited the National Assessment of Adult Literacy, a 2003 survey of over 19,000 US adults, to show that low health literacy is widespread. 14% of respondents scored “below basic” (e.g., able to circle the date of an appointment on a calendar), 22% were “basic,” 53% were “intermediate” (e.g., able to read a medication’s label and say what time they were supposed to take it), and 12% were “proficient” (Kutner et al., US Department of Education 2006).
• To illustrate the sorts of day-to-day tasks affected by health numeracy, Dr. Kerr presented selections from a questionnaire given to 398 adult US patients with type 1 or type 2 diabetes (Cavanaugh et al. Ann Intern Med 2008). One question from the study read: “Your target blood sugar is between 60 and 120. Please circle the values below that are in the target range (circle all that apply): 55, 145, 118.” 74% of patients correctly circled only 118. Another question asked, if you test your blood sugar three times a day and buy a box of 50 strips on March 5th, what day new strips will need to be purchased. 62% of patients correctly chose March 21st from among four multiple-choice responses.
• Low levels of numeracy have been found to be associated with adverse measures of health. A 2008 study found that numeracy, but not literacy, is statistically significantly inversely correlated with BMI (Huizinga et al, Obesity 2008). In a paper currently in press, Dr. Kerrpresents his findings of an inverse relationship between A1c and level of numeracy in 112 adults with diabetes.
• Dr. Kerr said that industry seems not to recognize these problems. He quoted a passage from an unnamed insulin pump manual that required division by a negative number to calculate dose, and he noted that CGM data readouts typically contain multiple graphs and charts of numbers. However, Dr. Kerr warned that half or more of adults in the UK cannot divide by negative numbers, and at least half cannot interpret pie charts (Kerr and Marden, Diabetic Medicine 2010).
• Noting that health literacy and numeracy might have unrealized effects on clinical trial data, Dr. Kerr reviewed the demographics of patients in the STAR-3 trial of sensor-augmented pumping. He compared adults in the trial and pointed out that the injection therapy group included fewer who were employed/volunteers than the sensor- augmented pumping group (127 vs. 139), higher mean weight (85.1 kg vs. 80.8 kg [187.2 lbs vs. 177.8 lbs]), and more students, a group he suggested tends to take sub-par care of their diabetes (21 vs. 9).
• Briefly addressing the importance of social media, Dr. Kerr said that information from providers is immediately circulated by patients on Facebook and similar sites. He recommended that healthcare providers track the flow of health information throughout the social media space. He also noted that industry was in general slow to make effective use of social media, and he suggested it as an area for further development (albeit one fraught with potential regulatory issues).
• Dr. Kerr summarized his message with a quote from pioneering physician Sir William Osler: “It is more important to know what patient has the disease than what disease the patient has.”

 

DESIGNING MOBILE DIABETES APPLICATIONS: CHALLENGES AND OPPORTUNITIES

Erik Arsand, PhD (University Hospital of North Norway, Tromso, Norway)

Dr. Arsand presented the main lessons learned from a user-involved design approach to a mobile diabetes application (Arsand et al., Journal of Diabetes Science and Technology, March 2010). In an interesting audience question to start the session, Dr. Arsand asked, “Have you recommended cellular phone-based self-management systems to diabetes patients?” Of those that answered, 66% said “No,” 22% had “recommended one type of system,” and 12% had “recommended more than one type of system.” According to Dr. Arsand, these types of mobile diabetes applications have low long-term rates of use; while patients might use them for a few weeks, they inevitably stop using them thereafter. As a result, Dr. Arsand and his team set out to develop a user-designed mobile diabetes application by utilizing focus groups, prototyping, thinking aloud, and sketching exercises. The research culminated in the Few Touch application, a mobile phone-based system that includes an off-the-shelf blood glucose meter (data automatically transferred to the phone), a tailor-made step counter (data automatically transferred to the phone), and software for recording food habits and providing feedback. Users found the system simple, motivational, and useful for changing medication, food habits, and physical activity. According to Dr. Arsand, the main lesson learned from the project is that simplicity (e.g., automatic data transfer) is the key to long-term use. He found that gradually introducing the application and using human-computer interaction methods were also useful strategies.

E-, I-, OR M-HEALTH? BLURRING BOUNDARIES BETWEEN PROVIDER-AND PATIENT- CENTERED MANAGEMENT

Stephanie Fonda, PhD (Walter Reed Army Medical Center, Washington, DC)

Dr. Fonda set out to define telehealth and its variations as they apply to diabetes care, a difficult task for such a fast-changing sets of fields. She suggested that telehealth could be broadly defined as the use of health information technology to augment care, including diagnostic, therapeutic, and educational aspects. She proceeded to describe three flavors of telemedicine: e(lectronic)-health (e.g., the remote evaluation of retinal images by eye care specialists), i(nternet)-health (e.g., a portal for outside-the- office provider follow-up), and m(obile)-health (e.g., educational text messages and reminder videos regularly posted to a patient’s cell phone). Dr. Fonda noted that a host of issues still need to be addressed by (or about) telemedicine’s clinical trials: questions of selection bias, secular trends of improvement in the general population, and the problem of how to distinguish “usage” from “engagement.” She advocated a patient-centered approach for all those working in the field of telemedicine, which she said should really be called u- (for user-, user-centered-, or ubiquitous-) health.

PANEL DISCUSSION

Moderator: COL Robert Vigersky (Walter Reed Army Medical Center, Washington, DC) and John Pickup, MD, PhD (Guy’s Hospital, London, UK)

Questions and Answers

Dr. Nathanael Paul (Oak Ridge National Laboratory, Oak Ridge, TN): I have type 1 diabetes and I appreciate what these apps offer for glycemic control. But we haven’t really heard about the risks of these technologies. If a patient is prompted to treat hypoglycemia when they’re not actually low, that’s fine, but if they’re prompted to treat a high that they don’t actually have, that could be detrimental. CGM is approved only for trending, but patients sometimes use it for exact dosing. Similarly when the phone tells them to do something, they might not actually do it. Also, while telemedicine is promising, I’m not sure the benefits really outweigh the risks. What’s your evaluation of whether telemedicine is really showing improvement?

Dr. Kerr: The risk is obvious, that’s why companies have only 10 Facebook pages. They’re worried something will go wrong. But whatever they desire, people will use them more and more. Peer support is a huge driver of behavior and will continue to be. You and I need to work out a system of what’s going on, how people communicate, to keep up with this. In terms of metrics, I think you make a good point. It’s been suggested that maybe we’re measuring the wrong things, and this is an area for continued discussion. But people will use these systems for communication, and we have to find out how often things go wrong out there.

Dr. Arsand: I think these applications can be important in many cases. People were encouraged to reflect when they used their phones in our study. When they’re recording what they eat, they’re forced to think more about their choices. This holds a benefit that we see as very important and that is not much communicated about.

Dr. Andrew Karter (Kaiser Permanente, Oakland, CA): David Kerr, thank you for bravely pointing out the elephants in the room. I was a co-author of one of the health literacy studies you referenced, and I think these issues are a huge challenge to developers of technology for diabetes care. It’s important to make sure all these technological advances do not forget about patients who are especially vulnerable. I urge everyone working on these products to incorporate consulting from adult learning and health literacy experts, because we’ve seen that people don’t understand it. If you want to maximize your reach, you have to do that.

Q: We’re all very ecstatic about shiny new technologies. If I recall, your systematic review of telemedicine in diabetes in 2008 showed only one study demonstrating an improvement in A1c. You mentioned you had just updated this study and virtually doubled the number of RCTs in the review. How many had clinically significant improvements in A1c?

Dr. Boren: I believe in the original review, there were many behavioral measures and fewer clinical outcome measures being studied. The search criteria were just completed last week, so I haven’t had a chance to go through the more recent studies, but there is definitely a trend of more outcomes measures being studied. It will be interesting to see what the data looks like.

Q: I worked for a telemedicine project from 1999-2002. We got two different projects funded by the European Commission and nobody wanted to use our system. Nobody wanted to pay for it and the main problem was with the physicians. There was no organizational change in the clinic to manage home-care patients. Another point: it’s very difficult to isolate the effect of telemedicine in a study. My conclusion is that we need organizational interventions, sort of a healthcare 2.0 that will take into account the presence and existence of technology and how providers use it.

Dr. Kerr: Certainly, it depends on which health care system you’re looking at because they may or may not be paid or reimbursed for using these technologies. Physician adoption of these systems is heavily influenced by reimbursement as well as the enthusiasm and expertise of diabetes centers. It could also be that the systems are not the sort of systems that patients desire. I have this feeling that if a system incorporated social media with communication between patients, you might get a more effective tool. Many, many systems have been tried and many have some behavioral change improvement, but only modest A1c declines.

A: I agree. Sustainment is a huge issue; what happens after the project or research ends? Does the application continue? Typically, we see that it doesn’t. There is just an unwillingness to pay for this. Patients are not going to pay for it and providers won’t pay for it if they’re not reimbursed. We provide great applications and believe that they’ll make work easier, but the reality is that providers get inundated with information and they have enough on their plates. Just because we say it’s great doesn’t mean it’s great.

Q: What is more important for you: Low health numeracy or low health literacy? Most of the examples you gave were on low health numeracy. You asked American people the question about when they will need to buy strips assuming one box 50 of strips and three strips per day. Isn’t it more important to know whether someone will know to buy more strips when they see the box is running out? Isn’t there a literacy issue embedded in the numeracy question?

Dr. Kerr: I don’t think there is a hierarchy. There are people with low health literacy, people with low health numeracy, and people with both. There are data to show that if you inform providers that they’re going to see patients with poor health literacy or numeracy, they may change their content. However, the providers rarely follow-up to see if the message was received and understood.

Dr. J. Geoffrey Chase (University of Canterbury, Christchurch, New Zealand): Thank you for a very stimulating set of talks. It was so stimulating that I conducted a short study in the back. My colleague and I whipped out our smart phones and typed “diabetes” into the app store. The first result is that my iPhone beat his Android, but the larger point is that most of these dozens of applications were just written by some user out there, even though a lot of talks here have been about some established entity providing content. Often, a patient goes out and finds something useless, but they like it. It seems like this would produce a lot of risk. Could you comment on this?

Dr. Fonda: I’m not sure I can answer your question, but on the one side we have someone looking at Wikipedia during our panel, and then in the back you and your colleague are comparing mobile phone apps… Was this panel actually stimulating?

Dr. Chase: It was. Many of these applications are free and others cost money; most are probably not clinically useful. When one thinks about telemedicine, how do you direct people to good resources?

Dr. Kerr: I think the onus is on people like you and others to review the apps yourself – human behavior is such that people listen to experts, KOLs, or whatever you call them. When I checked a few weeks ago, there were over 200 apps in the Apple Store specifically for diabetes. People are also looking more and more to social networks, and I think that we somehow need to bring all this information together so that we can maintain a grasp on the field.

Dr. Boren: Most people will still seek out their provider’s advice but will also look to user-generated content. I think that’s a subject where the discussion with the provider comes in, as well.

Q: What was the answer to your last audience question about patient-to-patient education? There was a recent study at the University of Michigan where patients educated each other. They did significantly better than usual care.

Dr. Arsand: I was actually just curious about what the audience thought. I personally haven’t looked at this, but it is worth examining in more detail.

Improving Adherence to Therapy

Moderator: John Pickup, MD, PhD (Guy’s Hospital, London, UK) A MECHANISM FOR NONADHERENCE IN TEENAGERS?

Gérard Reach, MD (Hospital Avicenne, Bobigny, France)

Dr. Gerard Reach provided a unique look at a real problem in diabetes: nonadherence in teenagers. To start, Dr. Reach gave a synopsis of some of the research on nonadherence in teenagers, concluding that depression, age, and maturity may all play a role. However, he also suggested that impatience may play a significant role in teenage nonadherence. Studies suggest that impatience is associated with worse glycemic outcomes overall. Further, on a neurobiological basis, the rate of development of the brain suggests that teenagers are more likely to be impatient. Although this argument is potentially hazardous, it could explain the high rate of nonadherence and worse glycemic outcomes in teenagers.

• Nonadherence to therapies is not just about medication. In a study of the factors associated with less than daily SMBG, criteria such as race, belonging to a deprived community, education level, difficulty with the English language, and smoking were all found to have an effect on blood sugar testing.
• Studies have shown that teenage nonadherence to diabetes therapy is a significant problem with a variety of potential causes. The Diabetes Audit and Research in Tayside, Scotland/Medicines Monitoring Unit Collaboration Study found that poor glycemic control in patients aged 10-20 years was associated with a significant reduction in adherence. Additionally, the JDRF CGM trial found that CGM sensor use of 6 or more days per week was observed in only30% of the patients in the 15-24 year old group. This group failed to achieve any significant improvements in A1c, whereas the 83% of patients in the 25-plus year old group who wore the sensor 6 or more days a week saw a decrease in A1c of 0.53%.
• There are a variety of potential causes of nonadherence in teenagers: peers and schools, depression, diabetes duration, and age. Although peers and schools would intuitively have a negative impact on adherence (e.g., friends negatively react to diabetes), empirical work suggests that friends tend to provide encouragement. However, depression has been associated with lower SMBG and A1c has been associated with more depressive symptoms and a lower level of SMBG. In a study examining seven chronic diseases in 700,000 adults, the data showed that the younger the patients, the poorer the adherence. Generally speaking, patients who are more adherent are more mature and have a better focus on long-term goals versus short- term goals.
• The role of impatience is also important in explaining nonadherence. Health is a long- term objective, but may be jeopardized by impatience. In a trade-off scenario between $100 today or $140 a year from now, impatient people will take the money now in lieu of waiting. This concept was examined in a study of 90 people with type 2 diabetes on oral antidiabetic agents, with a mean A1c of 7.6% and a diabetes duration of 12 years. The participants were offered $500 today, $800 in six months, or $1500 in one year. They were also given a questionnaire that measured adherence. It was found that patients with an A1c higher than 7% were monetarily impatient (i.e., not willing to wait one year for the $1500) and had suboptimal foresight (both results highly statistically significant).
• On a neurobiological level, the development of teenagers’ brains may play a role in nonadherence. Choices for delayed outcomes are related to the prefrontal cortex, while choices for immediate outcomes are related to the limbic brain regions. Data from fMRI studies has provided evidence that the balance between frontal and limbic circuitries is relatively late maturing during the brain’s development from childhood to adult age. This may explain the higher rate of risky behavior in teenagers as well as poor adherence to diabetes therapies.

Questions and Answers

Dr. John Pickup: It’s surprising that everybody isn’t addicted to present. I’m wondering if impatience has some survival value?

Dr. Reach: I totally agree with you; survival value could be tied to patience. There are experiments in animals on this. For instance, rats are more patient and have a high ability to wait. But obviously, some of us are not that patient.

Dr. Pickup: It’s not surprising that teenagers have no frontal lobes [laughter]. The brain is not yet fully developed at that age.

Dr. Ken Ward: If I’m meeting an adolescent for the first time in a diabetes ward, what is a simple question I could ask to classify him or her as impatient?

Dr. Reach: This is a very good question. There could be a way to do it, but the patient might be surprised or think it is random. Instead of the questions on money, we have other questions we use as well. For instance, “You go to a restaurant and you see a long queue, will you wait?” This question is perfectly associated with the monetary questions I discussed earlier. Another question is, “Do you fasten your seatbelt in the back of a taxi?” This is associated with adherence and behavior. In my mind, the people who fasten their seatbelts are more compliant to rules.

Dr. Ken Ward: When you take a taxi in Paris and you’re in the back seat, I recommend you fasten you seatbelt [laughter]!

ASSESSING AND ENCOURAGING DEVICE ADHERENCE IN DIABETES MANAGEMENT

Jen Block, BSRN, CDE (Stanford University School of Medicine, Stanford, CA)

Ms. Block gave a patient-centered talk focused on improving patient use of blood glucose meters and continuous glucose monitoring. Drawing on clinical trials, expert opinion, and personal experience, she encouraged providers to see their role as helping patients to manage their disease independently, using techniques such as open-ended “motivational interviewing,” objective data analysis, and affirming language. She noted that patients with the most adherence to devices tended to be those with problem- solving attitudes and strong social support, and she said that part of clinicians’ role is assessing which patients are ready to use devices in the first place. (For more detail on a more detailed talk delivered by Ms. Block at this year’s AADE, see Closer Look August 24, 2010). In closing, she shared her hopes that more and more of the day-to-day tasks associated with diabetes devices can be automated, saying, “I think that is what patients really want.”

• In her first of two audience response questions, Ms. Block asked, “How often do you download blood glucose meter, insulin pump, and continuous glucose monitoring data in your practice?” Audience members responded: 75% or more of the time (55% of respondents), 55% to 74% of the time (7%), 25% to 49% of the time (3%), or 0% to 24% of the time (34%).
• To follow up, Ms. Block asked, “What would make you more likely to download these devices?” Audience members responded: if it took less time (18%), if the data generated could be integrated into medical records directly (29%), if the data were easier to understand and interpret (11%), and if there were a universal software for the reports (42%). “If you’re in industry,” Ms. Block said to the audience, “this is what people are asking for.”

​Questions and Answers

Dr. Pickup: You talked about behavior and motivation in patients but not so much in physicians and other healthcare providers. Do you think there’s a strong correlation between demotivated doctors and how that spins off to their patients?

Ms. Block: I think maybe it’s not about provider motivation but about how you present issues. For example, open-ended questions seem more effective in motivating persistence.

Dr. Pickup: I’ve seen at least one study on this topic that showed a significant reduction in A1c regardless of whether patients received the sorts of motivation-promoting care you’re describing, suggesting that it didn’t make much difference.

Ms. Block: I think that the people who volunteer for clinical trials are more likely to be motivated and to have had those skills already.

Fotios Papadimitrakopoulos, PhD (Biorasis, Storrs/Mansfield, CT): We live in a world where compliance is enforced by rules and regulations. If you don’t put your seatbelt on, a policeman comes and gives you a nice ticket. Could you please comment on the possibility of penalties for not following therapy?

Ms. Block: I feel really challenged by the word “compliance,” and sometimes it’s hard for me to say “adherence” as well. I think it’s accurate that adherence is more challenging for some people than others,

and I think it’s important to individualize care for patients. I’m not sure what the impact of penalties would be; I’m not aware of literature on the subject.

Dr. Papadimitrakopoulos: For example, a company could fine patients for not taking their insulin as prescribed.

Dr. Reach: I think it would be completely unethical, even if we consider it’s best for patients’ welfare. Who are we to define their welfare? I know this might shock some of you. If I see someone who smokes three packs of cigarettes a day, I will do everything I can to convince him not to smoke, but at the end of the day I have to respect his decision. Under French law, it’s strictly forbidden to smoke in restaurants – because it’s dangerous for the lungs of the waiters, not to prevent people from making individual choices. In terms of therapies, it’s a conversation, and at the end of the day the patient has the last word.

Ms. Block: Patients are already penalized by their diagnosis, and to propose another penalty is something I don’t think I would even entertain.

ASSESSING MEDICATION ADHERENCE FOR NEWLY PRESCRIBED THERAPY

Andrew J. Karter, PhD (Kaiser Permanente, Oakland, CA)

Dr. Andrew Karter of Kaiser Permanente gave the audience a wake-up call with his presentation on measuring medication adherence. Typically, medication adherence is measured by medication possession ratio (MPR) or continuous medication gaps (CMG). While these measures aren’t terribly difficult to calculate, they don’t include patients who fail to have their prescription initially filled or those that fail to refill a new prescription. Kaiser’s new method, new prescription medication gaps (NPMG), is an effort to rectify this limitation. Using electronic pharmacy records and prescribing information, Dr. Karter explained that a study of 27,329 patients found that standard CMG measurement significantly underestimates poor adherence; inadequate adherence via CMG was 29%, while NPMG put prevalence at 48%. Although electronic data is required to perform this calculation, it has serious implications for health care delivery and outcomes. Identifying the patients that are most non-adherent and the approaches that boost adherence can improve diabetes management.

• Medication adherence can be measured in a variety of ways, most with significant limitations. Doctors can ask patients, but their self-reported adherence is usually an overestimate. Other methods to estimate adherence use medication possession ratio (MPR) or continuous medication gaps (CMG). MPR is very easy to calculate, and simply takes the number of days supply dispensed divided by the total number of days between the first and last fill, plus the number of days supply of the last fill. Gap measures such as CMG estimate the percentage of time without sufficient pill supply among ongoing users. Unfortunately, these measures have a serious limitation: they exclude patients who either fail to initiate or fail to ever refill a new prescription.
• More robust data analysis is now possible with electronic prescribing; Kaiser examined adherence by analyzing data from the Northern California Diabetes Registry (n~220,000). The study examined a subset of 27,239 patients who had been prescribed a new cardiometabolic (antihypertensive, antilipidemic, or antihyperglycemic) medication. Because of the Kaiser managed care model, the researchers were able to look at prescriptions and utilization through an electronic prescribing database.
• The study used a new measure, new prescription medication gaps (NPMG), and found that standard gap measures significantly underestimate poor adherence among a new prescription cohort. When looking at the percentage of time individuals lackeda medication, CMG estimated 16% from the data, while Kaiser’s NPMG method estimated 29%. In terms of the prevalence of inadequate adherence, data analysis via CMG reported 29%, while Kaiser’s NPMG method reported a much higher 48% prevalence of inadequate adherence. The study found that 5% of patients did not fill their prescription at all, 18% did not refill their prescription, 9% had their healthcare provider switch or discontinue their treatment before it was refilled, 25% discontinued use within 24 months, and 4% discontinued after 24 months. At the 24-month mark, 39% were still using the medication.

Questions and Answers

Dr. Pickup: Have you done any demographic breakdown of the registry data? How does non-adherence relate to the type of drug, age, etc.?

Dr. Karter: I didn’t have time to show that data, but the findings are consistent. We’ve found that the lipid lowering drugs have the lowest rates of adherence.

Dr. Robert Vigersky (Walter Reed Army Medical Center, Washington, DC): That’s fascinating data, Dr. Karter. In the US, we’re imposing performance measures on our physicians to which payment is going to be attached. We don’t want to penalize patients because of nonadherence, but we’re going to penalize our providers. Is Kaiser going to integrate this into your physician compensation package?

Dr. Karter: Not that I know of. However, if you do compensation based on adherence, you need the electronic prescribing data I talked about. Otherwise, you miss half of the non-adherent patients.

Dr. Fotios Papadimitrakopoulos: (Reading from his iPhone) I’m looking at a press release from Novartis. They are seeking regulatory approval for smart pill technology to measure adherence. I thought this was interesting. (Editor’s note – this was referencing Novartis’ technology from Proteus.)

THE MED-STEP STUDY: A PILOT STUDY OF HEALTH INFORMATION TECHNOLOGY SUPPORTED BLOOD PRESSURE MANAGEMENT

Richard Grant, MD, MPH (Massachusetts General Hospital, Boston, MA)

Dr. Grant reviewed the challenges of treating chronic conditions with primary care visits, and he described his work with an online portal to facilitate patient self-management of blood pressure. The Medication Self-Titration and Evaluation Program (Med-STEP) allows patients to remotely monitor their blood pressure, enter the data into an online portal, and modify their care based on a pre-written algorithm developed by their primary care physician. Although the system remains to be fully implemented or studied, Dr. Grant was optimistic that it could be expanded and modified to facilitate better control of hypertension and other chronic conditions like diabetes.

• Dr. Grant noted that primary care visits are compressed, that appointments are often missed, and that primary care physician-patient interactions occupy a minuscule fraction of the healthcare system and of patients’ lives. He presented data that adult primary care visits addressed an average of 7.1 items in 2005 vs. 5.4 in 1997 (Abbo et al., J Gen Intern Med 2008).
• The Medication Self-Titration and Evaluation Program (Med-STEP) study was designed to test the use of an online program for blood pressure management (although Dr. Grant noted that a similar system could be used for other chronic conditions such as diabetes and high cholesterol). Dr. Grant’s group found that patients like to know in advance the circumstances under which their medications would change, but thatthey don’t like an outsider imposing an algorithm on their care. Consequently, in Med-STEP, each patient’s own primary care physician develops a protocol for adjusting medications based on the patient’s blood pressure readings. During a pilot study of five primary care physicians and 20 patients, 19 unique pathways were developed. The physicians spent an average of 4.5 minutes per pathway entering instructions in their online portal, but this reflected a learning curve: discounting the first pathway developed, the average time was 3.3 minutes, and the last few pathways took roughly two minutes to complete.
• Patients monitor their own blood pressure at home, the results are transferred to the Med-STEP online portal, and the medication regimen updates accordingly. Blood pressure readings taken during the first calendar week of each month are used to determine medication changes. The first question asked by the online patient interface is whether the patient has been taking the prescribed medications – a question that Dr. Grant said patients answer honestly when they are the ones involved in their own titration.
• Dr. Grant noted that in TASMINH2, a large (n=527) randomized-controlled trial recently conducted on blood pressure self-management with telemonitoring, self- managing patients completed the yearlong study with a 3.7 mean placebo-adjusted systolic blood pressure reduction (McManus et al., Lancet 2010). He was hopeful that, with funding, his own group could conduct an RCT showing substantial improvement from their self-management system.
• In order to implement and scale telemedicine systems like Med-STEP, Dr. Grant said that the next steps would be building health IT infrastructure, redesigning healthcare delivery, and reforming payment systems. On the reimbursement side, Dr. Grant noted that accountable care organizations – where healthy outcomes are more important than specific treatments – offer a promising opportunity for telemedicine solutions.

Questions and Answers

Dr. Pickup: We need evidence from randomized controlled trials that this would make a difference for outcomes. How is progress on that?

Dr. Grant: It’s in process. We know the medications work; we know they work in RCTs. So we need RCTs on whether we can change practice.

Dr. Pickup: Do you think that a system like this could be used for self-titration of diabetes medications?

Dr. Grant: That’s already commonplace.

Dr. Pickup: Actually it’s not. People are meant to self-titrate, but there’s kind of a mismatch between provider expectations and what patients do.

Dr. Grant: My thinking on these systems is that if we build it, people will come.

Dr. Kenneth Ward (Oregon Health and Science University, Portland, OR): If a patient gets a high blood pressure reading, do they just follow the program’s recommendation to increase lisonopril, or does the physician have to sign off on the decision?

Dr. Grant: The way it’s meant to work is that once the physician has made the regimen, he or she doesn’t have to do anything else. More health IT needs to be built to get to this point, but the idea is that a note would get posted to the pharmacist, who would fill the prescription. In chemotherapy, a patient might go through different cycles and use different medications over different weeks. If things are going well, maybe they would hold medications for a few months. That’s what we want to happen with this system.

Christen Rees (Roche Diagnostics): In light of the obstacles you mentioned, maybe it makes sense to incorporate pharmacists more into the process.

Dr. Grant: The way we’ve modeled workflow is that many primary care practices have a diabetes nurse and some have a medications nurse. The primary care physician would set the pathway, and then when it’s time to make a change, the practice nurse can make sure that the process works. Another problem with what I showed is that there’s no direct patient education. It’s sort of implied, but it needs to be there.

Biomarkers

INDIVIDUAL RISK ASSESSMENT AND PERSONALIZED DIABETES TREATMENT BY MEANS OF LABORATORY BIOMARKERS

Andreas Pfutzner, MD, PhD (IKFE Institute for Clinical Research, Mainz, Germany)

In an interesting take on the management of type 2 diabetes, Dr. Andreas Pfutzner argued for the adoption of a biomarker approach in type 2 diabetes therapy. According to Dr. Pfutzner, measuring adiponectin, intact proinsulin, and high-sensitivity C-reactive protein gives more detailed information on cardio-diabetes status, provides routine risk assessment, helps select appropriate therapeutic interventions, and helps monitor treatment success or failure. A number of epidemiological and interventional trials with more than 20,000 patients have demonstrated the usefulness of this approach. Point of care tests have also been developed to enable the physician to determine the three biomarkers in a fast and economic way using capillary whole blood from a finger prick. In Q&A, Dr. Pfutzner made it clear that the biomarkers could be weighted differently based on individual patient characteristics and health risks.

• Although we measure diabetes with glucose or A1c, type 2 diabetes has complicated pathophysiological causes that are often ignored. Current treatment targets attempt to lower glucose or A1c, but these often do little to address mortality risk. The insulin resistance seen in type 2 diabetes leads to adipogenesis, a decrease in adiponectin, secretion of proinsulin, angiotensin, dyslipidemia, inflammation, oxidative stress, and, quite often, atherosclerosis. According to Dr. Pfutzner, we need to look at circulating functional proteins to identify the severity of the underlying pathophysiological disorder.
• Dr. Pfutzner proposed that treatments should focus on not only reducing glucose but also on improving three biomarkers: adiponectin, high sensitivity C-reactive protein, and intact proinsulin. Examining these three biomarkers can allow for more detailed information on cardio-diabetes status. For instance, increases in proinsulin levels have been shown to confer a 200% relative risk increase in mortality. When choosing between treatments with equivalent glycemic outcomes, we should always choose the one that improves biomarkers the most.
• Certain treatment interventions are more desirable when examined from a biomarker perspective. Treatments that improve multiple biomarkers include diet and exercise, metformin, insulin, GLP-1s, and TZDs. In a case study, a patient with an A1c of 6.6% and seemingly good health actually had very poor biomarker status. In this patient, glimepiride was stopped in favor of pioglitazone (Takeda’s Actos) and liraglutide (Novo Nordisk’s Victoza). A low- carb diet was adopted, and five months later the patient recorded a slightly better A1c (6.4%) and normal biomarker status.

NONINVASIVE DETECTION OF DIABETES-INDUCED RETINAL METABOLIC STRESS

Victor M. Elner, MD, PhD (University of Michigan, Ann Arbor, MI)

Dr. Elner presented on a non-invasive technology designed to measure diabetes-induced stress in the retina. Retinal metabolic analysis (RMA) quantifies metabolic stress on mitochondria by optically measuring the fluorescence of flavoproteins in the inner mitochondrial membrane. Dr. Elner showed evidence that RMA can detect early stages of the progression toward retinopathy, and that it is sensitive to changes in the extent of retinopathy. We are curious to see how RMA fares in the two longitudinal clinical trials currently being undertaken: a user-friendly, cost-effective means of assessing retinopathy would be a boon to patients throughout the progression of diabetes.

• Dr. Elner explained that apoptosis (programmed cell death) of retinal cells begins very early in the course of diabetes, long before retinopathy can be detected. Apoptosis is associated with increased metabolic stress in the mitochondria, which can be measured by the increased fluorescence of flavoproteins in the inner mitochondrial membrane. Flavoprotein fluorescence (FPF) is well accepted as a measure of metabolic stress in several physiological systems, and Dr. Elner showed data suggesting that FPF is more sensitive than a conventional apoptosis assay.
• Retinal metabolic analysis (RMA) measures flavoprotein fluorescence to non- invasively detect diabetes-induced metabolic stress in the retina. The RMA instrument shines a blue light into the retina and then measures the returning green signal from fluorescing flavoproteins. High-intensity signals with wide variability indicate increased metabolic stress differentially affecting retinal cells, a marker of disease. Dr. Elner showed images of increased fluorescence in a retinopathic eye relative to a healthy one, and he explained that imaging techniques can minimize the effects of potentially confounding physiological factors (e.g., advanced glycation end products [AGEs] or the fluorescent molecule lipofuscin).
• Researchers are currently undertaking two longitudinal clinical trials of RMA as a measure of treatment efficacy. In one, patients are measured before and after treatment for retinopathy, to see whether RMA correlates with the effects of local therapy. In another, newly diagnosed type 1 patients will be studied before and after insulin initiation to see whether RMA reflects systemic treatment.
• During the question and answer session, Dr. Elner, who has co-founded OcuScience, Inc. with plans to commercialize RMA, said that he expects the technology could be available for researchers to use within one to one-and-a-half years, with a clinical version potentially available one-and-a-half or two years after that. These timelines are aggressive in our view given the challenges in this field as well as relatively slow regulatory timelines overall.

GUT-MICROBIOTA-BASED BIOMAKERS TO PREVENT METABOLIC SYNDROME VIA NUTRITIONAL MODULATION

Liping Zhao, PhD (Shanghai Jiao Tong University, Shanghai, China)

Dr. Zhao described a hypothetical role of gut microbiota in insulin resistance during a well-received talk that ranged from traditional Chinese medicine to his personal weight-loss experience. He proposed a pathway whereby nutritional habits cause changes in the gut microbiota: decreases in the species that protect the lining of the large intestine and increases in species that produce endotoxins and corrosive chemicals. Consequently, harmful antigens are released into the blood, promoting the low-level inflammation that characterizes insulin resistance and the metabolic syndrome. Although the causality of these separate factors remains unclear, Dr. Zhao presented data from animal studies and a clinical trial that showed correlation between positive dietary changes, favorable shifts in the species composition of gut bacteria, improvements in gut barrier integrity, reductions in lipidosaccharide binding protein (a marker of antigen levels), lower levels of conventional inflammatory markers (e.g., C-reactive protein, adiponectin), and weight loss.

• Dr. Zhao proposed that lipidosaccharide binding protein, C-reactive protein, and fasting insulin be used as indicators of antigen load, inflammation, and insulin resistance, respectively. He noted that lipidosaccharide binding protein can bind to a variety of antigens, including some produced by gram-positive bacteria.

O-GLCNACYLATION AS A DIAGNOSTIC FOR PRE-DIABETES

Ronald Copeland, PhD (Johns Hopkins University, Baltimore, MD)

Dr. Ronald Copeland summarized the body of research on O-GlcNAcylation, arguing that it could be used as a diagnostic tool to detect the earliest stages of diabetes. O-GlcNAcylation is an endpoint of the hexosamine biosynthetic pathway and serves as a nutrient sensor to regulate cellular signaling, transcription, and phosphorylation. Dr. Copeland reminded the audience that there is substantial research linking diabetes with hexosamine metabolism. He then showed the results of a number of studies linking O-GlcNAcylation with insulin resistance and glucose toxicity. In their research, Dr. Copeland and his team have found that O-GlcNAc is significantly increased in samples from patients with pre-diabetes or diabetes. Additionally, the enzyme that removes O-GlcNAc, O-GlcNAcase is also significantly elevated in patients with pre-diabetes and diabetes. O-GlcNAc is a promising diagnostic for diabetes because it cycles rapidly, is highly sensitive to nutrients and stress, and most importantly, increases earlier than A1c. So far, Dr. Copeland has identified about 50 O-GlcNAc sites on human red blood cells that could be useful in diagnosing pre-diabetes. Such a strategy might help catch diabetes in its earliest stages, when lifestyle interventions can really make an impact on disease progression. Nevertheless, while the test may be promising, we note that it will likely prove to be more expensive than less technical diagnostic measures.

PANEL DISCUSSION

Moderator: COL Carl Castro, PhD (US Army, Fort Detrick, Frederick, MD) Questions and Answers

Dr. David Kerr (Bournemouth Diabetes and Endocrine Centre, Bournemouth, UK): I have a question for Dr. Pfutzner. Does each biomarker have the same weight in your risk assessment system?

Dr. Pfutzner: You are right that there’s a difference in the way biomarkers are weighted depending on different conditions. Proinsulin is good as a measure of beta cell function, whereas it’s not a helpful indicator of insulin resistance developing in people with type 1 diabetes who have no beta cell function left. The idea is to develop a simple system to let us see the underlying processes. It can require clever, individual interpretation, but it allows individualized medicine.

Q: Dr. Zhao, you presented an interesting concept. If you don’t change diet but lose weight by increasing activity levels, will the gut microbiota change?

Dr. Zhao: Our hypothesis is that when we eat, your own body takes out nutrients from the food first. Then whatever is left moves to the colon, where the bacteria there will use those nutrients to grow. If you exercise, you consume more nutrients and leave less for the gut microbiota. So though we don’t have the data yet, I expect that vigorous exercise would change gut microbiota.

Dr. Jens Christiansen: The concept that gut bacteria contribute to insulin resistance has been around for a couple of years. I would like it if the next time you give your talk, you would include a study on the subject from I think Holland that was presented at EASD. Patients with type 2 diabetes had their gut bacteria eradicated, and then half of them were re-fed with their own feces and half were fed with feces from lean people without insulin resistance. Those who received the new feces showed statistically significant improvement in insulin resistance, while those who had their own – sorry – shit fed back to them returned to insulin resistance.

Dr. Zhao: I know people are doing that experiment. It’s an interesting piece of evidence to show that gut microbiota play a role in insulin resistance.

Dr. Robert Burk (Albert Einstein College of Medicine, New York, NY): When you saw changes in biomarkers, were these due to weight loss or changes in the microbiome?

Dr. Zhao: We know it’s very complicated. When you perform an intervention, you can’t adjust a single factor. I’m not saying we proved a causal relationship. The way we did the clinical study was to analyze the plausibility of the relationship between antigen load, inflammation, microbiota, and the gut barrier. We were trying to connect these to link the effects. We need more studies to see whether that’s for sure; so far the data are encouraging. The transplantation study Dr. Christiansen mentioned added more weight to the causality relationship.

Dr. Burk: What sort of genetic analysis are you doing on the microbiota?

Dr. Zhao: We’re not doing whole sequencing yet; we’re just using the 16s region to determine composition.

Q: Dr. Pfutzner, I would’ve thought you’d be interested in the proinsulin/insulin ratio rather than just proinsulin. If you have insulin resistance but the beta cell is still normal, you’ll see an increase in both insulin and proinsulin. But if the beta cell is overstimulated or defective, then you’d see a preferential rise in proinsulin over insulin.

Dr. Pfutzner: We wanted to keep it simple. The ratio is most interesting to look at in the dynamic postprandial state. Postprandial proinsulin was studied in the ’80s and ’90s, but in addition to intact proinsulin, which has a half-life of 15 minutes, they looked at degradation products. These studies don’t tell us the reality of what beta cells are doing in a dynamic state. Also, our own analysis of thousands of patients suggests that proinsulin as a single factor is superior to the proinsulin/insulin ratio.

Everyone suspects that DPP-4 inhibitors improve beta cell function. All the companies publish the ratio and yes, it’s improving. But they rarely show a simple proinsulin level. Usually proinsulin doesn’t improve; the underlying deterioration is still there. After a few years of monotherapy, A1c goes up again. There’s dynamic improvement, but the underlying deterioration is not gone. Improvement in type 2 diabetes is an area that links my research to Dr. Zhao’s. There are drugs where you get an increase in body weight but not visceral lipids. Sure, you are allowed to be obese and happy, as long as it’s subcutaneous fat.

Dr. Zhao: I would like to add a comment. In the severely obese case I showed, he lost 30 kg (~66 lbs), but he still weighed 145 kg (~319 lbs). But his metabolism came back to the normal range. There’s a group of benign obese people.

Dr. David Klonoff (Mills-Peninsula Health Services, San Mateo, CA): How soon do you think we’ll see a clinical test for O-GlcNAcylation?

Dr. Copeland: It will depend on the development of a high-throughput screening mechanism. Right now we’re doing site-specific binding assays in the lab.

Dr. Klonoff: Same question for Dr. Elner, how soon until your diagnostic is in the hands of doctors and patients?

Dr. Elner: I would say researchers could have an instrument within a year to a year-and-a-half, with a clinical version in perhaps another one-and-a-half to two years after that.

Live Demonstration

MOBILE PHONES FOR REMOTE MONITORING: APPLICATIONS FOR CARDIOVASCULAR DISEASE INCLUDING DIABETES

Mani Chandy, PhD and Danny Petrasek, MD, PhD (California Institute of Technology, Pasadena, CA)

In an exciting demonstration to close the conference, researchers from Cal Tech demonstrated wireless technology that could help deliver healthcare in the developing world. Drs. Chandy and Petrasek purchased a standard stethoscope and placed a microphone inside with Bluetooth transmission capabilities. The device could then record a heartbeat and transfer the recording to a standard consumer cell phone. From here, the recording could be saved or uploaded to a secure Internet database. The idea is that a doctor could use the recording to make a remote diagnosis. Demonstrating the technology in front of the audience, the researchers recorded a student’s heartbeat and played it back from the web. While the quality was a bit questionable (the researchers said it was the large room), a recording from India (in a smaller room) was crystal clear, and an audience member quickly identified the Indian patient’s heart condition (mitral regurgitation). The researchers envision this technology helping provide healthcare in the developing world, which contains 64% of the world’s mobile phones. According to Dr. Petrasek, modern cell phones are relatively inexpensive, with good computing power and Internet connectivity. He and his team foresee the “10 cent medical checkup” using a tablet-based system. Such checkups would include heart sounds, temperature, blood pressure, visuals, ECG, and a multiple choice health questionnaire. Their aim is to have people in the communities administer these checkups, getting diagnoses from remote medical practitioners. We thought it was definitely notable technology, but clearly it has limitations and needs more research. We’d also like to see such technology applied more specifically to diabetes, perhaps through a blood glucose test that can be remotely uploaded.

Pre-Conference Workshops

Advances in Continuous Glucose Monitoring: Subcutaneous

A NEW GENERATION OF CONTINUOUS GLUCOSE SENSORS

Rajiv Shah, MS (Medtronic Diabetes, Northridge, CA)

“Many of you have been asking, ‘When is Medtronic going to develop a new sensor?’ Today, I hope to convince you that that day is coming soon.” In one of the more buzzed about early morning talks, Mr. Rajiv Shah discussed Medtronic’s newest innovation in sensor technology, the Enlite sensor. In developing the new sensor, Medtronic candidly admitted and addressed patients’ frequent problems with the current Sof-Sensor: performance, usability, comfort, and setup. Mr. Shah also acknowledged the great competition in the CGM field, noting that Medtronic’s goal has always been to bring the Enlite to market as fast as possible. He explained that the new Enlite sensor is easier to insert, has higher accuracy (MARD: 14.77%, 96% within A & B Clarke Error Grid Zones), smaller size (27 gauge needle, implanted volume down 75%), greater sensitivity to temperature extremes and drug interferences, better adhesion to the skin through a 90 degree insertion, backwards compatible with existing products, and featuring a seven day sensor life. Also quite noteworthy was the fact that better CGM algorithms can further reduce the MAD of the sensor (Paradigm Real-Time: 17 mg/dl, Paradigm Veo: 9 mg/dl, next generation algorithm: 5.6 mg/dl). Mr. Shah made it clear that the Enlite sensor represents a meaningful and significant improvement over the Sof-Sensor and we look forward to hearing about the progress on this new and exciting technology. At this time, the sensor has been submitted for CE Mark and feasibility studies are still ongoing – we aren’t clear on when this will hit the US but it looks to be a significant step forward for patients.

• Medtronic has a long history of CGM technology and innovation, but 73% of customers have experienced problems with their CGMs that could be addressed by improvements to the current Sof-Sensor. Medtronic developed a CGM sensor in 1999, with subsequent improvements in 2003 (CGMS Gold), 2005 (Guardian RT), and the more recent Revel and Veo sensor-augmented pumps. However, according to data from the 24-hour product helpline and CareLink, customers experience problems with setup, performance, usability, and comfort of their CGMs. A common criticism of the Sof-Sensor is that “It does not last long enough.” Fran Kaufman, a pediatric endocrinologist and Chief Medical Officer at Medtronic Diabetes, has even declared that “The needle is too big.”
• In developing the new Enlite sensor, Medtronic addressed specific challenges. The sensor had to be backwards compatible with existing installed MiniMed products, have the same sensitivity as the Sof-Sensor, use the same manufacturing methods and glucose oxidase chemistry, and get to market as fast as possible. The development program started in 2008 and arrived at a pivotal clinical trial in 18 months.
• The mean absolute relative difference (MARD) of the Enlite was 14.77% and 95.97% of sensor values were within Clarke A and B zones when used with current market Medtronic CGM devices; the algorithm used played a significant role in the accuracy of the sensor: the Paradigm Real-Time yielded an MAD of 17 mg/dl, the Veo algorithm achieved an MAD of 9 mg/dl, and a next generation algorithm achieved a very impressive MAD of 5.6 mg/dl. The median lifetime of the sensors was over 6.6 days.
• With the new Enlite sensor, the developers are striving for dramatic improvements over the Sof-Sensor: greater ease of use, accuracy, comfort, and reliability. The Enlite features a new 90-degree insertion and serter similar to that seen on the Quick Set Infusion sets. The sensor will be assembled onto a plastic “pedestal,” inserted via a user-friendly insertion device, and features a retractable needle that is not visible to the patient. Compared to the relatively large Sof-Sensor needle (22 gauge, 17.5 mm), the Enlite has a 27 gauge needle measuring 10.5 mm. The surface area of the new sensor has been reduced by over 50%, the implanted volume is down 75%, and the insertion force has decreased by 63%. This is very impressive from our view and we look forward to getting patient feedback. We believe the pain associated with the lower-gauge sensor is one factor holding back more frequent sensor changes – we believe the higher-gauge sensors could make a big commercial difference as we know from our dQ&A patient data that even patients who have sensors covered by insurance leave sensors inlonger to avoid the pain associated with changes (for more details, data, and subscription information, contact Richard Wood at richard.wood@d-qa.com).
• Because of the smaller surface area, the electrochemical cell in the Enlite had to be redesigned; the result is increased sensitivity over previous generations. The sensor also features a 60% reduction in temperature response and a 40% reduction in drug interference. Finally, the Enlite has been designed with sensitivity to oxygen in mind. As one of the major factors in predicting long-term sensor use, this greater sensitivity to oxygen will mean that the Enlite will have a longer sensor life.
• The Enlite sensor will feature a better securing mechanism than the Sof-Sensor. The 90-degree insertion will help prevent sensors from pulling out, while the adhesion used with the sensor will focus on securing the sensor base.

INNOVATIONS IN CONTINUOUS GLUCOSE MONITORING

Peter C. Simpson, MEng (DexCom, Inc., San Diego, CA)

“The best sensor is the sensor that patients wear.” With this patient focus in mind, Peter Simpson gave a whirlwind tour of what’s on the horizon for DexCom in the continuous glucose monitoring space. After a brief history of DexCom’s major innovations in the field (five-to-eight minute time lag, hypoglycemia accuracy), he looked towards the future. Mr. Simpson reminded the audience of the partnerships with Animas and Insulet, both exciting opportunities to fully integrate the company’s smaller (50% volume reduction), more accurate (MARD of 15%, and MAD of 12.1 mg/dl in hypoglycemia), and more comfortable fourth generation sensor with insulin pumps. The fifth generation sensor has shown even more promise, achieving phenomenal hypoglycemic accuracy (MAD of just 8.5 mg/dl) and a noteworthy 99.3% of values in zones A and B of the Clarke error grid. Finally, Mr. Simpson rounded out the presentation with a quick review of the Edwards partnership to develop an inpatient CGM, just submitted for approval in Europe early this year.

• In DexCom’s 10-year history, the company has made significant strides in CGM technology. In 2006, the STS system was launched and was the first CGM to display trend data. In 2007, the SEVEN became available, allowing a full week’s worth of wear. 2009 marked the company’s third generation system, the SEVEN PLUS. Early this year, Edwards and DexCom submitted their jointly produced inpatient sensor for approval in Europe. Amazingly, DexCom has brought four products to market in the last five years.

• “The best sensor is the sensor that patients wear.” In a recent survey, over 70% of DexCom patients use the company’s sensors 24/7 – dQ&A data show this as the highest 24/7 wear to date. Priority has been placed on making sensors small in size, using a comfortable, 26 gauge needle, and minimizing the size of variable components.
• The partnerships with Animas and Insulet represent very exciting developments for DexCom. This new sensor features more than a 50% volume reduction, more comfort, improved accuracy in the hypoglycemic range due to a new membrane, and an enhanced biointerface. The latter improvement has allowed patients to get more reliable information from this device early in the sensor life and has increased the possibility that sensors last a full seven days. It is still unclear if the combined Insulet/DexCom device will have the next-gen sensor, and we look forward to getting more information on this.

• The clinical trial results from DexCom’s fourth generation system are quite promising, demonstrating significantly better accuracy in the hypoglycemic range. In a study of 60 subjects with type 1 and 2 diabetes and in-clinic tracking on days 1, 4, and 7, the next generation sensor achieved a mean absolute relative difference of 15% when compared with YSI. More importantly, the sensor achieved an outstanding mean absolute difference in hypoglycemia of 12.1 mg/dl. This represents a 25% improvement over the current system. 95% of values were within zones A and B of the Clarke error grid.
• DexCom’s fifth generation system will aim to be on par with traditional SMBG meters, potentially allowing use in the artificial pancreas; early results suggest impressive accuracy and hypoglycemic sensitivity. In an early clinical trial (n=15, 30 sensors) of patients with type 1 and type 2 diabetes and in-clinic measurements on days 1, 4, and 7, the sensor showed a MARD of 12% (n=1291) and an impressive mean absolute difference of 8.5 mg/dl in hypoglycemia. Accuracy on Clarke error grid was exceptional as well, with 99.3% of points in the A and B zones.
• On the inpatient front, Dr. Simpson said that DexCom’s CGM project with Edwards will revolutionize the inpatient CGM space. The sensor is inserted with a peripheral IV catheter. The new system exceeds the ISO blood glucose monitoring recommendations, achieving a mean ARD of 6.6%. Mr. Simpson made it clear that this level of accuracy is truly needed in the ICU environment. The product has excellent accuracy, high reliability, no interference with over 30 common ICU drugs, a new membrane technology, and automatic calibration.

ROLE OF VASCULAR NETWORKS IN SENSOR FUNCTION IN VIVO

Ulrike Klueh, PhD (University of Connecticut, Farmington, CT)

Dr. Klueh outlined the results of a trial examining the use of localized gene therapy on CGM sensor life in mice. Over a 28-day period, mice either received no injection, an injection of saline, an injection of adenovirus LacZ, or an injection of adenovirus VEGF. The injections were localized around a modified Abbott Navigator CGM sensor site. In the experimental group (injection with adenovirus VEGF), sensor life and corresponding accuracy were significantly extended over four weeks. No such benefit was found in the three control groups (no injection, saline, injection of adenovirus LacZ). The experimental group also exhibited significant increases in both vessel density and vessel area at the sensor site, confirming results from previous research. While this therapy is still in the exploratory stage, these results are certainly encouraging for researchers and patients alike.

• Local gene therapy has the potential to enhance vascular networks, decrease fibrosis, and extend the life of sensors. Therapy with VEGF and other growth factors has been used to treat wounds in people with diabetes. The basic concept is that a vector (e.g., a virus) carries genes, the genes express proteins, and the proteins alter cell and tissue function.
• This small study using gene therapy in mice examined the potential of VEGF (carried via adenovirus) to extend sensor life. The study divided mice into four groups: no injection, saline injection, adenovirus LacZ injection, and adenovirus VEGF injection. Modified Abbott Navigator CGM sensors were implanted on the backs of the mice, and between five and eight days after sensor insertion, the animals were injected with the various solutions.
• Injecting mice with adenovirus VEGF resulted in a significant increase in sensor life, an effect that was not seen in the other groups. The sensors on mice treated with adenovirus VEGF exhibited sensor data that compared very well with reference blood glucosevalues, even after 28 days of sensor life. By contrast, sensor accuracy in the other three groups was poor after the first week, demonstrating significant excursions and unpredictable accuracy.
• When examining vessel density and vessel area at the sensor site, mice treated with adenovirus VEGF showed significant increases in both parameters; these increases were not shown in non-VEGF-treated mice. According to Dr. Klueh, this study provides proof of concept for using VEGF in vivo to extend sensor life. Further, there are many directions for future research, including using other angiogenic factors, other delivery systems (e.g., nanobubbles, nanotubes), and targeting inflammatory cytokines and fibrosis.

PANEL DISCUSSION

Moderators: COL Robert A. Vigersky, MD (Walter Reed Army Medical Center, Washington, DC) and Dariush Elahi, PhD (Johns Hopkins University, Baltimore, MD)

Questions and Answers

Q: Have you thought about the error associated with changes in hematocrit?

Mr. Simpson: I’ll refer you to Mike Higgins’ presentation later today. In this hospital product, we’ve looked at this for the blood-based sensor. Currently available disposable strips are influenced by hematocrit, but we’ve found no effect in our systems.

Comment: I think the issue with hematocrit is more about how you calibrate the sensors than with the CGM itself.

Q: I like the concept of your inpatient CGM in partnership with Edwards. However, do you need to have heparin in the system to return the sampled blood back to the patient? If so, is that a consideration for the US? ICU workers may have a lot of resistance to such an indication.

Mr. Simpson: The first generation does have heparin in the IV system. I agree, there may be some resistance in the field to the use of this.

Q: Can you tell us about the use of the VEGF vector – is there a dose dependent effect on the neovascularization? How much neovascularization is desirable?

Dr. Klueh: We have not looked at different dosages. What I want to focus on is that there is really not much data in this area. Most studies have focused on just one part of the VEGF equation.

Dr. David Klonoff (Mills-Peninsula Health Services, San Mateo, CA): A comment for Mr. Shah and Mr. Simpson. I thought that it was really impressive your companies have developed new generation CGMs. It’s great for patients. Do you guys have any ideas to extend sensor life?

Mr. Simpson: We’re definitely looking at this, but we’re still in the feasibility stages at this time.

Q: Is there a downside to 90-degree vertical insertion, and if not, why was the original Medtronic insertion done at a 45-degree angle?

Dr. Shah: In terms of downsides, there are definitely some to the 45-degree insertion. Now, we know more about how products are used. Back then, when we first developed the system, we didn’t know. The data was clear that the 90-degree insertion was better in many aspects.

Q: I’m interested in the data on in vivo oxygen. Does that implicate micro encapsulation?

Dr. Shah: We don’t know why this happens, we just know that oxygen levels decline with sensor life. It probably occurs more for biological reasons that anything else.

Q: I’m a pancreatic surgeon at UCSF. I have a question on soft tissue edema. We’ve been concerned that the surgically induced edema can lead to CGM inaccuracy. When can we allow these patients to go back to their sensors?

Mr. Simpson: We haven’t specifically studied that. I don’t know if I would anticipate much of a challenge there though.

Dr. Klueh: When you talk about edema or tissue destruction, any of this can affect sensor function. I would not put a sensor in the compromised tissue.

Q: You showed statistics about improved sensitivity. You said 40% decreased drug interference and 60% decreased temperature response. Are those clinically relevant? Can CGMs be used in hypothermic indications now?

Dr. Shah: With temperature, absolutely. Our research has definitely cleaned up some of the in vivo effects we’ve seen with temperature variation. But with drug interaction, the effects are minimal to begin with in the ambulatory situation. It’s probably only the hospital situation that would be a concern here.

Q: Can you speculate on the histological response of low dose gene therapy? What applications are there for wound healing? Is there a precedent for a tissue response over a long period of time? What would the systemic response be if we treated a site with VEGF and then came back to the sensor site later down the line?

Dr. Klueh: The adenovirus probably wouldn’t be used in the clinic. You would use an adeno-associated virus on the clinical side and that form is considered safe.

Q (Becton Dickinson): I’m wondering what happens with the newly formed tissue or vessels after VEGF is no longer administered?

Dr. Klueh: Once the trigger is gone, you go back to fibrotic tissue and the regular wound healing process. Other genes could be used, but we don’t have the data.

Q: How variable are the new sensors to the calibrations used?

Dr. Shah: What we see with the Enlite sensor is convergence to the in vivo sensitivity. You don’t see large calibration factors. You see something that looks like in vitro behavior.

Mr. Simpson: Our algorithm accepts a broad range of sensitivities. For us, it doesn’t affect our performance.

Dr. Bruce Buckingham (Stanford University, Palo Alto, CA): To comment on a previous question, we have in our pediatric ICU used CGM in kids with edema and we get the same MARD. To pose a question, I’m wondering about your new sensors and nocturnal sensing?

Dr. Shah: We have not looked at things like pressure on the sensor site like you might see while sleeping. However, the sensor excursions associated with sleeping may be better with new sensors.

Mr. Simpson: We’ve seen improvements in newer generations, but we haven’t explicitly studied it.

Comment (Animas): I recall an abstract a few years ago about subcutaneous sensors being used in patients with edema. I don’t think we should rule out using sensors in these patients.

Comment: The Endocrine Society is coming out with recommendations on CGM use in the ICU, and they will recommend against using subcutaneous CGM in the ICU because there is no data.

Comment: I’m a pediatric intensivist at Emory University, and we used CGM devices in 50 critically ill children. CGM showed very good correlations with blood glucose readings. In our experience, the smallest and sickest patients showed the strongest results.

Q: What was the calibration method used?

A: We used standard link in our pediatric setting.

Arleen Pinkos (FDA): In evaluating the effect of drugs on CGM performance, how many people are you evaluating and what type of study designs are you using?

Dr. Shah: In the validation of all of our sensor products, we use an in vitro model and we expose the sensor to drugs that are known to interfere with electrochemical systems. We’ve tested this in ambulatory setting, but the hospital setting has a higher standard, obviously.

Advances in Continuous Glucose Monitoring: Intravenous

INTRAVENOUS CONTINUOUS GLUCOSE MONITORING WILL BECOME STANDARD OF CARE IN THE HOSPITAL

Jeffrey Joseph, DO (Director, The Artificial Pancreas Center, Thomas Jefferson University, Philadelphia, PA)

Dr. Jeffrey Joseph gave a useful overview of the current and future status of inpatient glucose monitoring technology. He began by highlighting the importance of glycemic control in the hospital, casting serious doubts on the validity of NICE-SUGAR. Next, he delved into the limitations of currently available inpatient glucose monitoring, noting that traditional blood glucose monitoring takes up far too much nurse time (over four minutes per measurement and ~2,500 hours per month; Aragon, Am J Crit Care July 2006). Finally, he surveyed what’s on the horizon, briefly discussing a variety of promising continuous glucose monitor technologies, including those in development by Braun, GluMetrics, and DexCom/Edwards.

• IV sensors, subcutaneous sensors, and non-invasive devices will all have a place in the hospital environment. For instance, in the critically ill, IV accuracy may be best, but patients needing physical therapy might need the convenience of a subcutaneous sensor. Unfortunately, the technologies that most hospitals currently use are not sufficient to give good glucose control.
• Dr. Joseph argued that NICE-SUGAR was a dangerously flawed study, while more recent data suggests the benefits of improved glycemic control. In NICE-SUGAR, the investigators took blood from different sources (e.g., fingerstick samples AND arterial samples), handled the samples differently, and used different meters. These methodological issues made the study’s errors additive. According to Dr. Joseph, you can’t say there was a distinction between the two groups, and NEJM “messed up big time” allowing the study to be published. More recent data, however, have suggested that improved glycemic control can reduce the risk of nosocomial infections. Unfortunately, inpatient studies of glycemic control are hard to conduct, often because they require a large population of patients (~1500) to show mortality outcomes.

• Current blood glucose methods in the hospital have serious limitations. First and foremost, manual blood glucose monitoring is labor intensive and takes up a lot of nurse time. The GlucoScout, an IV-based system, requires a large catheter to support the substantial 6 ml sample volume for testing. The Biostator also has problems, including its large size and unfriendly user interface. Arterial blood samples are the gold standard, but these invasive catheters last for only one or two days.
• A number of new glucose monitoring devices have significant potential to enhance inpatient glycemic control. A new Braun system will rely on an MPC algorithm, IV insulin infusion, and IV glucose sensing to achieve glycemic control in the inpatient setting, with potential for closed-loop control in the future. GluMetrics’ GluCath is an intravascular continuous glucose monitoring system based on fluorescent chemistry. The system is optimized in the hypoglycemic region and will offer very accurate measurements every minute. The DexCom and Edwards Lifesciences product is another very exciting technology, with excellent accuracy and a small sample size. Dr. Joseph also looks forward to CMA’s Eirus microdialysis system (an alternative to glucose oxidase and optical measurements), OptiScan’s spectroscopy-based continuous glucose monitor, Luminous Medical’s Automated Blood Glucose Monitor (using a radial artery catheter), and Animas’ research into a long-term implantable glucose sensor (using spectroscopy and currently in long-term animal trials).

DESIGN AND CLINICAL CONSIDERATIONS FOR INTRAVENOUS GLUCOSE MONITORING IN CRITICALLY ILL PATIENTS

Michael Higgins, MA (Edwards Lifesciences LLC, Irvine, CA)

Mr. Higgins gave a brief overview of the DexCom/Edwards GlucoClear system, an inpatient intravenous glucose monitor. The system provides automatic, real-time blood glucose values at 7.5- minute intervals, with each blood glucose value viewed as a discrete event that is independent of previous results. The IV sensor is inserted into the peripheral vein with a standard 20 gauge catheter. A 40 ul blood sample is pulled up into the catheter, sampled, and flushed back into the patient. Advantageously, the GlucoClear system does not have the time lag seen in subcutaneous CGM and can give practitioners nearly real-time trends to better manage glycemia in the hospital. It has been evaluated in two major clinical studies, one examining accuracy and the other testing it for clinical use. In the first study, the system was evaluated in 50 healthy people with diabetes in diabetes centers (3000 data points collected); the GlucoClear achieved an impressive MARD of 6.6%, with 95.1% of readings in Clarke Error Grid Zone A. In the clinical use study, 22 patients (mostly cardiac surgery) were put on the system in the ICU. The GlucoClear recorded a MARD of 7.1% and 93.6% of readings within Clarke Error Grid Zone A. Thus, even in the hospital environment, where 40-plus drug interactions were tested and patients were much sicker, the system still maintained its accuracy: a very positive result in our view.

INTRAVENOUS HOSPITAL SENSORS

Courtney Harper, PhD (Office of in Vitro Diagnostic Device Evaluation and Safety, FDA, Silver Spring, MD)

Dr. Harper led off her talk on intravenous hospital glucose sensors by outlining the two major studies in the field. The game-changing Leuven I study conducted by Van den Berghe et al in 2001 found a decrease in mortality associated with tightly controlled blood sugar in the surgical ICU. However, the infamous NICE-SUGAR found an increase in mortality from tight glycemic control in intensive-care wards. According to Dr. Harper, it’s unclear what the reason is for the difference between these studies, but there’s no question that we need better ways to assess blood glucose in the hospital setting. Laying out the FDA’s perspective, she posed a number of important considerations that intravenous hospital sensors will have to address. For instance, what technology should be used – an IV sensor or one that draws on a peripheral vein? An external sensor that removes blood from the patient? A fluorescent sensor? Additionally, design considerations, sterility, calibration, interferences, and sensor structure and design must be an integral part of the development process. Dr. Harper also spent time addressing the familiar accuracy question, arguing that ISO 15197 standards (±20% at or above 75 mg/dl and ±15 mg/dl below 75 mg/dl) are not accurate enough for hospital use. Instead, she cited an audience poll from the International Hospital Diabetes Meeting where most agreed 10% was an acceptable error (for our coverage of that poll and the rest of Dr. Harper’s presentation, see Closer Look October 11, 2010). To close, Dr. Harper quickly addressed trial design, emphasizing incremental steps and development. Overall, she believed that IV sensors for the hospital environment hold great promise and looked forward to working with researchers to get them approved.

THE NEED FOR CONTINUOUS GLUCOSE MONITORING FOLLOWING PANCREATIC SURGERY

Kimberly Kirkwood, MD (University of California San Francisco School of Medicine)

Dr. Kirkwood discussed the serious clinical benefits pancreatic surgery patients would receive from CGM technology in the hospital. She started by educating the crowd on the different types of pancreatic surgeries that are often performed. The most serious operation, a total pancreatectomy, removes the entire pancreas and is typically the result of cancer. Unfortunately, these patients often have long lengths of stay, high readmission rates (over 25%), and low quality of life. Nurses also struggle to manage the blood sugars for these patients following surgery, a fact that leads to many complications and poor recovery following surgery. In one case study, Dr. Kirkwood showed a patient with a blood sugar range from 21 mg/dl to over 600 mg/dl in a matter of hours. The speaker made it clear that these patients are an ideal test cohort for inpatient CGM trials considering they already have frequent levels of blood glucose monitoring, long lengths of stay, and wide glucose variations. In her mind, there is no doubt that a CGM or even a closed-loop would have substantial clinical benefits for this cohort of patients.

PANEL DISCUSSION

Moderators: Ricardo Bellazi, PhD (University of Pavia, Pavia, Italy) and Gerold Grodsky, PhD (University of California, San Francisco, San Francisco, CA)

Questions and Answers

Dr. Grodsky: Is this now an area where we should look at non-invasive CGM?

Dr. Joseph: Of course, we would love to have an accurate, sensitive, and non-invasive CGM in the hospital. But it’s challenging, even in healthy diabetics. We’d love to have this though.

Q: I worked in the clinical area developing glucose sensors for chemistry and blood gas. I’m confused—the FDA already has a traceable glucose standard using spectral measurement. Your precision when measuring is in your pipette error. Thus, people comparing results to YSI are doing themselves a disservice. However, when you’re actually developing a sensor, you must look at accuracy in certain scenarios. There are a lot of things that need to be

looked at. I’m also curious why more people don’t use the HemoCue, as the precision is at least equal to or better than YSI. Why aren’t we just starting with clinical standards as opposed to making new ones?

Dr. Harper: We would accept the reference standards you proposed. We generally let companies choose. The issue here is not that we couldn’t apply lab standards to IV sensors, it’s that I’m not sure the technology would meet it. Is there a point at which the benefits outweigh the risks and we can ensure safety? I don’t know what that is, but there’s been discussion around a level of 10% error.

Q: It appears that with these new technologies for the hospitalized patients, you’re looking at whole blood measurements that are better than 10%. To expect better than 10% on the MARD side is difficult.

Dr. Harper: I agree. We weigh the risks and benefits and there may not be an absolute number right now. Point measurements may also not be the best way to look at this type of device. We’re certainly open to any of that discussion. It’s still up in the air and I hope I didn’t come across as stating that there is a clear standard, because we’re certainly open to discussion on this.

Dr. Joseph: Every sample has error. Over a number of days, this is magnified. A lot of the issues in the clinical setting surround whether a value is an outlier – is it contaminated? If you measure every several hours, you cannot be sure of the quality of that measurement.

Dr. Kenneth Ward (Oregon Health and Science University, Portland, OR): I have a question for Mike Higgins. You data is very impressive. I’m wondering about the catheter your device is using. Do you have problems with the withdrawal of blood from the vein if the catheter is pushed up against the wall? Is any heating required?

Mike Higgins: None of the studies use heating. Also, we minimized our catheter placement so the wall wouldn’t be a problem. But it cannot be placed at joints.

Q: What is the start up stabilization time?

Mike Higgins: The start up time is 16 minutes and the use time is 72 hours.

Q: What about using your IV sensor when a patient is in ketoacidosis?

Mike Higgins: Our system was tested across the physiological range of pHs.

Rick Thompson (President and CEO, Luminous Medical, Carlsbad, CA): Dr. Harper, is it correct to assume that there could be different standards of performance from point measurements?

Dr. Harper: I encourage manufacturers to be creative. People developing products make assumptions about what we want without talking to us. If there is a different way to evaluate these systems than we currently have, we are open to speaking about that.

Q: I work in pediatrics. Will the Edwards system have any small catheters suitable for children?

Dr. Joseph: We have done work with different catheters for sampling and flushing over time. The larger diameter catheter in a small diameter vein doesn’t work well because blood flow is essential. But you don’t want a catheter to be too small because a little bit of a clot would be a problem. You want to optimize blood flow and minimize clotting.

Q: I urge you to remember that when the FDA approved subcutaneous CGMs, their performance was not very good and they did not compare well with point measurements.

We worked through statistical methods and figured it out. But don’t get hung up on specific point standards. Think about what are the risks and benefits.

Dr. Joseph: Nurses are busy and must multitask. They want to come into the room and take a glance at a glucose trend screen. The trend data is the key to minimizing risk for hypoglycemia. Currently, what we call intensive or aggressive glucose monitoring is hourly. The frequency of monitoring does matter, especially when you are trending downwards. If you had an automated system, more samples would be better.

Q: For those who have had pancreatic surgery, it’s hard to figure out how many nutrients are being absorbed because of the enzymes. It seems to me that CGM in that patient population would help better understand pancreatic enzymes.

Dr. Kirkwood: That’s an outstanding point. The testing and methods are appalling. We don’t adjust enzymes based on anything rationally. If we had a way of delivering things automatically, it would be great.

Q: You do not know how accurate our glucose sensors need to be. Why should we measure glucose? We measure glucose to steer insulin infusion. One suggestion to test the accuracy of glucose monitoring is to use a controller. We still have to look at the whole picture of closed-loop control. The CGM technology we have may be accurate enough, we just might not know it.

Dr. Joseph: The accuracy of current devices is good enough. It should be an integrated system with CGM, pumps, and the critical care team. The system should save time, be safe and effective, and user friendly. An integrated sensor is crucial in this.

Q: Where are we in regards to closed loop in the hospital setting?

Dr. Joseph: There’s a whole lot of work towards making sensors work in the hospital setting. A product will come out in the next months to years that will give automated glucose measurements. Then, we’ll incorporate insulin-dosing algorithms.

Q: You’ve talked about using automated insulin and glucose together. I recall studies also including potassium. Can you comment?

Dr. Joseph: Glucose, insulin, and potassium infusions have been a hot topic. The concept was that insulin would drive potassium into cells and improve nutrition. But they weren’t really looking at glucose control in these studies. What we need though are better tools for improving glycemic control in the hospital. This will improve outcomes and we can rerun the studies. The key concept is avoiding hypoglycemia. We say hypo is below 60 or 50 or 40 mg/dl, but it’s the duration that’s also important. Also, the concept of glucose delivery to the tissues matters. We have no data about the concept of flow over time.

New Ideas for Blood Glucose Monitoring: Industry Panel

AUTOMATIC ALL-IN-ONE BLOOD GLUCOSE MONITORING

Ron Nagar, MSc (InsuLine Medical Ltd., Petach Tikva, Israel)

Mr. Nagar reviewed the benefits and drawbacks of current SMBG and CGM devices, evaluated progress that has been made toward non-invasive blood glucose monitoring, and looked forward to devices that incorporate all the components of SMBG into one unit and minimize the steps required to test. He said that although developing noninvasive monitoring is not impossible (“chasing rainbows”), it is still quite difficult (“climbing a slippery iceberg”); hurdles include glucose specificity, engineering, regulation, and marketing. Mr. Nagar was most optimistic about all-in-one glucose sensing devices. Mentioning Mendor’s Discreet (available in Europe) and Intuity’s Pogo (still under development), he went on to play a short animated video featuring a novel BGM system that uses mobile-phone-activated, self-contained patches to provide alternative site testing at pre-set times. Although this system seems to be in early stages and would doubtless face serious regulatory challenges given its cell phone connection, we are excited and intrigued by the possibilities (e.g., nocturnal hypoglycemia detection) that this system could offer.

• Mr. Nagar noted that although SMBG is reliable, accurate, and approved to communicate with insulin pumps and inform dosage decisions, the technology has many drawbacks. He said testing involves too many steps and too many pieces of hardware, and for many patients, testing is painful. Moreover, Mr. Nagar mentioned that SMBG does not offer low- or high-glucose alarms, and there is not a simple way to communicate the data. CGM, on the other hand, offers continuous trending information and alarms, and it too can communicate with insulin pumps (although it is not approved for use to titrate insulin). However, Mr. Nagar said CGM’s downsides include its relative inaccuracy for point measurements and its unwieldiness.
• Mr. Nagar believes the pursuit of a noninvasive glucose monitoring device is not quite “chasing rainbows,” but is similar to “climbing a slippery iceberg” – feasible but extremely difficult. Having previously worked in the field, Mr. Nagar explained that most noninvasive monitoring technologies are sensitive to water as well as glucose. Thus, during an OGTT (often used to validate these systems), the device might seem to show positive results even if it is really responding to OGTT-induced movement of water around the body. Besides these osmosis-related concerns, Mr. Nagar said that noninvasive devices also face significant hurdles with regard to engineering, regulation (probably the devices would require PMA approval), and marketing (especially price). In sum, he stated that these systems seem to be far away from market.
• All-in-one blood glucose monitoring devices incorporate all the components of SMBG into a single unit. According to Mr. Nagar, the Discreet, produced by Mendor, still requires several steps in order to check blood glucose, whereas Intuity Medical’s Pogo allows testing with a single action. He noted that accuracy data are not available for either device. Currently, the Discreet is available for sale online in Europe, while the Pogo is still in development.
• Mr. Nagar closed his presentation by describing a novel blood-glucose testing system under development by Mon4D, a new company that Mr. Nagar has helped found. The system involves small, partially disposable “measurement buttons.” These small, flat buttons include a conventional, disposable lancet and test strip, along with a re-usable part containing electronics that can communicate with a mobile phone. The measurement buttons, worn on alternative testing sites (e.g., arms, thighs), are set via cell phone to activate at specific times. When the time comes, the lancet sticks the patient, testing occurs within the measurement button, and the results are sent back to the cell phone. Mr. Nagar explained that a patient could set one or more measurement buttons to activate during the night, checking blood glucose and activating a cell phone alarm in the event of hypoglycemia. He also noted that the system could communicate directly with CGM to enable automatic calibration. Although this incredible- sounding system seems to be very early in development, we are excited to see such an innovative mobile-health reimagining of BGM.

IMPROVING PATIENT AND PROVIDER UTILIZATION OF BLOOD GLUCOSE MONITORING DATA

David Horwitz, MD, PhD, FACP (LifeScan, Inc., Milpitas, CA)

Dr. Horwitz discussed techniques of motivational interviewing as taught to healthcare providers at the Johnson & Johnson Diabetes Institute. Noting that he would address several of the topics addressed earlier by Dr. William Polonsky (see below), he gave several examples of good and bad patient-provider interactions along with general tips on how to use data and psychological techniques to help patients manage their glucose better.

• “If you’re working harder than the patient, you’re not doing your job properly.” Dr. Horwitz advocated the technique of motivational interviewing, which involves asking open-ended questions about objective glucose monitoring data, allowing the patient to do most of the talking. He explained that it is best to start a conversation in a way that lets the patient direct the flow (e.g., “What happened?”) rather than by setting a particular tone from the outset (e.g., “So, looks like you had a bad time here.”) He noted that self-monitoring software makes it easier for patients and providers to find potential problems, setting up productive discussions.
• Dr. Horwitz listed several other examples of good and bad interviewing techniques based on data visualization readouts. For a patient with well-controlled fasting glucose but fluctuations around meals, a good approach might be, “You do a good job in the morning; what happens later in the day? How do you feel?” A less productive approach would be “You don’t seem to get things under control in the afternoon. Do you do things differently every day?” For a patient who typically has hyperglycemia throughout the day, a good question might be, “Tell me about this day you stayed at target,” followed by “What happened on the other days?” A bad question would be “Look at how much your blood glucose goes up when you eat. You must be eating too much.” Dr. Horwitz noted that although such recommendations seem obvious, even experienced providers often negatively tinged questions and recommendations.
• The Johnson & Johnson Diabetes Institute offers two-day onsite courses to healthcare providers, who then continue to participate in an online social community. According to self-reported attitudinal and behavioral changes, the courses help providers offer better diabetes care. Compared to their attitudes before the course, providers feel more confident in areas such as applying motivational interviewing (from 40% to 77%), explaining to patients how to use the results of blood glucose monitoring (77% to 96%), and analyzing readings and predicting patterns with diabetes management technology (25% to 79%). The program’s analysis also involves self-reported behavior at baseline, three months after the course, and six months after the course. An increased number of providers reported offering motivational interviewing (from 47% at baseline to 78% at three months to 77% at six months), using blood glucose management software to analyze patterns (34% to 46% to 47%) and specifying glucose testing frequency based on type of therapy and patient need (89%, 93%, 100%).

DESIGNING A BLOOD GLUCOSE MONITORING SYSTEM TO MEET NEW NEEDS FOR ACCURACY AND EASE OF USE

Ronald Ng, PhD (Abbott Diabetes Care, Alameda, CA)

Dr. Ng discussed how to build a blood glucose monitoring system in light of technological improvements and user needs. Focusing on Abbott’s new FreeStyle Lite strips, released in the summer of 2010, Dr. Ng talked about minimizing interference, enhancing accuracy, and increasing ease of use.

• Dr. Ng began by reviewing a list of “significant improvements” that Abbott had made with its previous FreeStyle Lite system, including faster testing time, smaller blood samples, easier-to-read meter display (e.g., larger readout, backlighting), ergonomic design, a fixed unit of measure, no need for coding, wider temperature range, the ability to flag test results (e.g., with meal markers), user feedback (e.g., reminders to test), underfill detection, and clear error messages (e.g., for degraded strips).
• The new FreeStyle Lite test strips’ benefits include reduced interference, tighter accuracy, and greater ease of use. Dr. Ng explained that by switching the strips’ enzyme from GDH-PQQ to GDH-FAD, interference from agents such as maltose (an ingredient in certain dialysis solutions) could be reduced. (He noted that xylose, another sugar, still caused interference with the new strips, but said that xylose is much less commonly encountered than maltose).
• In a recent trial of accuracy, the strips with the new chemistry compared favorably to reference measures: 68% of the FreeStyle Lite readings were within 5% of the reference value; 96% were within 10% of the reference value, and 99% were within 15% of the reference value.
• In terms of user-friendliness, the new strips are backwards compatible with old Abbott meters (which means, among other things, that the strips require no coding), they require only a 0.3 ul blood sample, average testing time is five seconds, additional blood can be added for up to a minute after the first exposure, and the strips’ ZipWik tabs draw in blood more readily than previous strips.

ACCURACY METRICS FOR BLOOD GLUCOSE METERS

Scott Pardo, PhD (Bayer HealthCare, LLC, Tarrytown, NY)

Dr. Pardo presented a statistical model for simulating a blood glucose meter’s performance. He explained that in a linear regression model, the accuracy of both the meter and the reference value can be characterized by three parameters each: slope, intercept, and coefficient of variance. By using experimental data to estimate these parameters, researchers can create a simulation that indicates how a blood glucose meter would likely fare under different accuracy criteria.

• Dr. Pardo reviewed the ISO 15197 guidelines, which set standards for blood glucose meter accuracy. For blood glucose reference values at or above 75 mg/dl, the meter must read no more than 20% higher or lower than the reference value. For glucose reference values below 75 mg/dl, the meter reading must be within 15 mg/dl of the reference value. A meter is considered accurate if 95% of its readings meet these criteria. The ISO standards are currently under review, however, and it is expected that the updated version will set tighter accuracy criteria.
• Blood glucose meters and reference measurements have associated error that can be estimated with a linear regression model using three parameters each: y- intercept, slope, and coefficient of variance (a measure of “noise”). (Dr. Pardo noted that this model assumes errors to be linearly distributed, an assumption that he said is usually valid.) These six parameters can be estimated based on clinical data, and the estimated values can be used to model the likelihood of passing the ISO 15197 guidelines (i.e., the likelihood that 190 out of 200 measurements fall within the acceptable level of accuracy). Dr. Pardo explained that such a linear regression model would be useful for companies to see how a particular meter would fare under a variety of potential updated ISO standards. A company could also model theprobable distribution of reference and meter values before a clinical trial and then compare it to the observed data; large discrepancies would suggest the presence of “odd, highly influential points that some people like to call outliers.”

 

SELF-MONITORING OF BLOOD GLUCOSE IN PATIENTS WITH DIABETES: MODE OF ACTION AND THE CONSEQUENCES

Matthias Schweitzer, MD, MBA (Roche Diagnostics GmbH, Mannheim, Germany) 

Dr. Schweitzer discussed a progression of improvements in blood glucose testing, focusing especially on Roche’s STeP study. He explained that previously, patients recorded random blood glucose testing values in a handwritten diary. One improvement was the advent of downloadable meters, and another was the dawn of “structured testing,” the use of very frequent and regular testing over a period of several days, as popularized by Roche. In the STeP study of structured testing in insulin-naïve patients, quarterly use of structured testing was associated with an increased number of treatment change recommendations and a statistically significant 12-month A1c reduction of 0.3% compared to patients testing blood glucose randomly. (For more on STeP results, see our coverage of this year’s ADA poster in July 31, 2010 Closer Look and Dr. William Polonsky’s DTM talk below). Dr. Schweitzer said that the next step in self-monitoring of blood glucose would be better data visualization on meters, as exemplified by Roche’s Accu-Chek Combo system (expected to launch in the US in 2011). Subsequently, he looked forward to meters that use automated decision support.

PANEL DISCUSSION

Moderators: Robert D. Burk, MD (Albert Einstein College of Medicine, New York, NY) and Christine Kelley, PhD (National Institutes of Health, Bethesda, MD)

Questions and Answers

Dr. Holly Schachner (Bayer): I’m a pediatric endocrinologist, and I know from experience in the clinic and with my family that many kids have trouble getting blood from their fingers, and so do adults. My question is about the accuracy and method of refilling. Do you have accuracy on the results you obtain from a repeat application of blood?

Dr. Ng: Not all test strips are designed in the same way, and some are limited by the technology of the strip. If you compare those on the market today, with some you can reapply blood for maybe 15 to 30 seconds. For the FreeStyle it happens to be 60 seconds. With some others you are not supposed to reapply because the result would be interpreted incorrectly.

Dr. Klonoff: I thought this was a good way to bring out different advances that will help get benefits into patients’ hands. I think all these approaches – technological, educational, psychological – are needed. Earlier, Dr. Heinemann suggested that 2 ul is the ideal size of a blood sample. Does anyone want to comment on the drop size potentially getting too small?

Dr. Ng: We use 3 ul for Abbott. We regularly survey people in our user studies, and we’ve gotten only positive feedback. In terms of usability, it seems to be well accepted.

Dr. Schweitzer: What we know from development is that there might be a threshold where a lower blood drop volume might not be beneficial. I think we’ve done a good job lowering volume. If it’s combined with an appropriate failsafe, I think this is a good balance.

Dr. Laurence Hirsch (Becton Dickinson, Franklin Lakes, NJ): As a healthcare provider who still sees patients and as someone who’s been taking insulin more than 50 years, I have a simple solution for all of you who make blood glucose meters – put a permanent time clock in the meter that is permanently aligned with a standard clock. With 30-50% of my patients, the readings are useless because the time and date in the meters are wrong. Then you’re left trying to look at single readings and you can’t make sense of the measurements. This isn’t very high-tech, but it would make an enormously valuable contribution to the practicality and usability of the information we’re collecting.

Dr. Klonoff: As we get into closed-loop control, all the devices involved have to be synched to the same time. Maybe it would be helpful to have all diabetes devices linked to an atomic clock.

Claudia Harris (Abbott Diabetes Care): To follow up on Dr. Horwitz’s talk, some people might not be aware that the ADA has a program called “Facilitating Behavior Change” that focuses on training health care providers in motivational interviewing for diabetes.

Dr. Horwitz: I think the challenge is changing not just patient behavior but also provider behavior, and I think our course is one way to do that.

Dr. Burk: If we look toward the future, what do you see as what we need in the next 20 or 30 years?

Dr. Schweitzer: First of all, for me it’s important that we test all medical steps in terms of added clinical and medical value. Take spot monitoring vs. CGM – for some segments of the population, it seems there is limited added value. From a product focus, we might add the convenience of an integrated device, which Roche has developed. These steps might have value, but we have to have medical proof that they improve outcomes.

Dr. Burk: So you’re saying we need cost-benefit analysis, clinical review. (Dr. Schweitzer nods.)

Dr. Ng: I think the technology has made major improvements in recent years. I think in terms of helping patients to take the right action, there’s something to be done on that.

Dr. Horwitz: I think from a societal perspective, the number one goal for our companies is to go from managing diabetes to curing diabetes. Short of that, I think we should be trying to simplify life for patients, who right now have so much to think about, and helping them get on with their lives.

Mr. Nagar: I’m for simplifying. Keeping it simple is better for the patient. I agree that it’s not just having the data, but knowing what to do with it.

Dr. Pardo: I would like to reiterate that. Aside from getting a more accurate device, I think it will be about exploiting devices like cloud computing to give patients something actionable, as Dr. Polonsky was discussing. You could access Internet-based software through your cell phone, and it would evaluate the numbers and tell you what you need to do.

New Ideas for Blood Glucose Monitoring: Academic Panel

STICKING WITH SAFETY: ELIMINATING BLOODBORNE PATHOGEN RISKS DURING BLOOD GLUCOSE MONITORING

Joseph Perz, DrPH, MA (Centers for Disease Control and Prevention, Atlanta, GA)

Dr. Perz discussed assisted blood glucose monitoring (AMBG) and associated risks of infection from hepatitis B and other pathogens. He defined AMBG as the practice whereby a care provider assists with the glucose testing of many patients; since 1990 it has been associated with 18 hepatitis B outbreaks, mostly in long-term care settings. The CDC has recommended that providers minimize risks by using single-use safety lancets for fingersticks, with blood glucose meters either a) never shared by patients or b) thoroughly cleaned and disinfected between use by different patients. Dr. Perz presented several studies suggesting that unsafe AMBG practices are more widespread than previously thought, and he called on industry to promote safety with clearer labels and cleaning instructions.

• Dr. Perz warned about the risk of blood-borne pathogen transmission associated with assisted monitoring of blood glucose (AMBG). He noted that the primary concern is the hepatitis B virus, which can remain viable for over seven days at room temperature, even in the absence of visible blood. Since 1990, 18 hepatitis B outbreaks have been associated with AMBG – roughly two thirds due to shared fingerstick devices, and another third simply from shared meters (Thompson and Perz, J Diabetes Sci Tech 2009). Two of these outbreaks were reported in hospitals (1990 and 1996); eight in nursing homes (between 1996 and 2003), and eight in assisted-living facilities (between 2004 and 2008). Dr. Perz observed that the level of oversight is lower in assisted-living facilities than hospitals, a plausible reason for the higher observed rates of infection. However, the CDC believes that outbreaks may be a problem in other settings as well; Dr. Perz pointed out that long-term care facilities, with their stable patient population, make it relatively easy to detect outbreaks.
• According to the CDC’s guidelines on glucose testing and insulin administration, fingerstick devices should never be re-used; single-use safety lancets should be the standard. Care facilities are encouraged to assign each patient a single glucose meter; alternatively, meters should be cleaned and disinfected between use by different patients. The recommendations also address changing gloves and hand hygiene; a full list is available in Klonoff and Perz, J Diabetes Sci Tech 2010, or on the CDC’s website at http://www.cdc.gov/injectionsafety/blood-glucose-monitoring.html#Best.
• Dr. Perz said that the safety of assisted blood glucose monitoring requires attention from industry on issues including: validated instructions for blood glucose cleaning and disinfection, improved labeling and marketing, innovation in equipment design, and education to raise awareness of disease risk and support safe use.

THE FUTURE OF LANCING DEVICES

Lutz Heinemann, PhD (Profil Institute, Neuss, Germany)

Dr. Heinemann spoke on the benefits and frustrations of fingerstick blood checks, and he shared his ideas of how the field can move forward. He reminded the audience that although traditional SMBG is painful, fingersticks still often make more sense than alternate site testing (AST). He noted that reimbursement for pain-free lancing will improve only with long-term clinical trials showing benefits, and he called for independent, blinded trials of blood-checking devices. Dr. Heinemann believes that big improvements can still be made in lancing; he said the trick will be to develop technology at today’s standards of cost and size.

• Dr. Heinemann said that the pain of fingersticks, along with the cost of SMBG, is a major reason that patients do not check their blood glucose more frequently. He pointed out that although the fingertips are sensitive to pain, their high perfusion rate means that patients consistently draw enough blood when the lance their fingers. (Although the volume ofblood required for SMBG has declined dramatically over the years, Dr. Heinemann proposed that 2 ul is a realistic limit: below that, he said blood samples get too hard to see.) Factors contributing to the pain of lancing include insertion depth, needle shape, friction, and vibration/skin fixation. Dr. Heinemann displayed images to show that needles become much rougher after repeated use, and he suggested that the common practice of reusing lancets might dissuade patients from testing as often as they should.
• Dr. Heinemann noted that alternate-site testing (AST) – taking blood samples from areas like the palm, arm, abdomen, calf, or thigh – was popular 8-10 years ago as a way to reduce the pain of SMBG. Downsides of AST include the slow responsiveness of these areas to changes in blood glucose, the likelihood of staining clothes with blood, the conspicuousness of the procedure, and the odds of drawing insufficient volume.
• Several alternative fingerstick devices have come to market, but their drawbacks (notably expense) have prevented mainstream success. The Lasette, produced by Cell Robotics International, uses a laser beam to cut through skin with minimal pain. However, Dr. Heinemann noted that the expensive, bulky device emits a bang and odorous smoke when it’s used; he said the Lasette can be purchased from China, but he has not seen data that it is widely used. He said many patients loved Pelikan Technologies’ Pelikan Sun and CanAm Care’s Renew Lancing System, two electronic lancing devices that minimized pain. However, with high price tags and without strong data showing benefits beyond standard SMBG, Pelikan Technologies folded and the Renew Lancing System was discontinued.
• Dr. Heinemann suggested that the best two available lancing devices are Roche’s Accu-Chek SoftClix and LifeScan’s OneTouch Delicia. He cited a study showing the superiority of Accu-Chek lancets to those of other companies, although he pointed out that the research had been sponsored by Roche (Kocher et al., J Diabetes Sci Technol 2009). Dr. Heinemann called for independent clinical trials with (at least) single-blinded design, and he also said that developers of educational programs should increase their focus on proper lancing technique (e.g., inserting the needle at the right depth to minimize pain).
• The ideal testing device would be a continuous glucose monitor that could maintain accuracy without the need for calibration, Dr. Heinemann said. Other potential improvements include combining the steps of SMBG into one device (e.g., Accu-Chek Compact Plus, Mendor Discreet) and conducting a long-term clinical trial to show insurance companies that pain-free lancing is worth the expense (although Dr. Heinemann acknowledged that he wasn’t sure what the endpoints of such a trial would be). In closing, he said the goal of blood glucose monitoring development should be a system with high performance, low cost, and a size similar to current options.

WHEN BLOOD GLUCOSE MONITORING IS PERSONALLY MEANINGFUL: A NEW APPROACH FOR PATIENTS WITH TYPE 2 DIABETES

William Polonsky, PhD, CDE (University of California, San Diego, San Diego, CA)

Dr. Polonsky emphasized the psychological barriers to self-monitoring of blood glucose, and he suggested structured testing techniques to overcome these challenges. He addressed four common beliefs that decrease the “perceived worthwhileness” of SMBG: 1) that the data are not actionable, 2) that behavior changes in response to testing are not useful, 3) that the data are discouraging or shameful, and 4) that testing is too burdensome, inconvenient, or painful (a concern that Dr. Polonsky noted is being continually addressed by technological and manufacturing improvements). Focusing mainly on type 2 diabetes patients not taking insulin, Dr. Polonsky discussed two paper-based techniques advocated by Roche to make SMBG more useful and user-friendly: testing in pairs (SMBG before and after a daily activity on seven consecutive days) and 360° View (seven-point testing for three consecutive days). He also reviewed data from the STeP trial showing the benefits of structured testing (360° View). To close, Dr. Polonsky briefly addressed CGM as another way to foster problem-solving attitudes in patients, as long as it is used “as a compass” (i.e., trend arrows, alarms) as opposed to a GPS (precise numbers).

• Dr. Polonsky discussed key psychological factors that decrease the “perceived worthwhileness” of SMBG. He began with the problem that many patients conclude blood glucose data are not actionable. This is an especially common issue for patients with type 2 diabetes who are not taking insulin, Dr. Polonsky noted. Another challenge occurs when patients try to make data-driven changes but still struggle with glucose control, leading to the conclusion that SMBG data are not useful.
• A further obstacle to SMBG is that patients often find the data discouraging or shameful. To illustrate this point, Dr. Polonsky cited data from the STeP study of SMBG in poorly controlled, insulin-naïve type 2 diabetes patients. At baseline, the 483 patients had a mean A1c of 8.9% and a mean testing frequency of 6.6 times per week. When asked about their attitudes towards SMBG, 32% said that their SMBG results often make them feel bad, 21% said they feel there’s nothing they can do about the results, 22% said they felt there was no rhyme or reason to the results, and 81% said that they blame themselves when SMBG readings are high.
• Dr. Polonsky proposed that the perceived worthwhileness of SMBG could be enhanced with two “simple paper tool approaches” being publicized by Roche. The first is called “testing in pairs,” and it involves checking blood sugar before and after a regular activity (e.g., meal, walk) for seven consecutive days. Testing in pairs is an exercise in discovery and learning, Dr. Polonsky explained. He told the story of one of his patients who previously hated exercise, but, after using the technique, proudly announced that he had “discovered exercise lowers blood sugar.”
• The second paper-based tool, called 360˚ View, involves testing seven times a day for three consecutive days. This form of structured testing was used in STeP, where the intervention group was instructed to use 360˚ View five times over a year, and the active control group used traditional SMBG. The A1c difference between intent-to-treat populations was 0.3%, and patients in the structured testing group who used 360˚ View at least four times in 12 months experienced a 1.3% mean A1c reduction compared to 0.8% for those in the active control group. Dr. Polonsky speculated that the trial protocol, which included giving all patients free glucose meters, may have encouraged the control group patients to have more productive conversations with their physicians – and thus better glycemic control – than they otherwise would have.
• “Promote continuous glucose monitoring as a compass, not a GPS.” Dr. Polonsky encouraged providers to “address trust issues head-on” and tell patients to focus on their CGM’s alarms and arrows rather than its numbers. He noted that by helping people to see how their actions influence blood glucose changes, CGM can foster a problem-solving attitude

 

NEW IDEAS FOR BLOOD GLUCOSE MONITORING: INTERACTIVE SUPPORT FOR SELF- MONITORED BLOOD GLUCOSE GUIDED INTENSIFICATION OF MEDICATION

Andrew Farmer, MA, DM, BM, BCh, FRCGP (University of Oxford, Oxford, UK)

Dr. Farmer reminded the audience that SMBG in non-insulin-using patients with type 2 diabetes typically shows only slight A1c benefits (0.2-0.4% drop). To increase these benefits, he proposed that researchers a) determine how to make SMBG information more useful to patients and b) better target the patients and situations where SMBG will be most helpful. He said that mobile phones offer a platform for leveraging SMBG data, and he delivered promising results from his group’s feasibility study of telehealth decision support for insulin dosing. Dr. Farmer forecasted that telehealth decision- support systems are on the way, and he noted they will likely provide a range of feedback rather than a specific treatment recommendation.

• Dr. Farmer described his group’s feasibility study of type 2 diabetes patients who were poorly controlled after recently starting insulin therapy. The nine-center study included 23 patients with mean age 58 years, mean diabetes duration 6.4 years, and mean baseline A1c 9.5% (Turner et al., Informatics in Primary Care 2009). Patients used a telehealth platform that included a standard self-titration algorithm, a glucose meter linked via Bluetooth to a mobile phone, an integrated diary of insulin dosage, and automated feedback including charted SMBG data. A telehealth nurse reviewed their data every two-to-three days, giving telephone follow-up two-to-four times per week. Mean A1c decline at three months was 0.52%. Among the 22 patients who completed the trial, mean A1c declined by 0.66% and mean insulin dose increased by 17 units per day (Larsen et al., Journal of Telemedicine and Telecare 2010).

ACCURACY OF GLUCOSE METERS: ARE STANDARDS ADEQUATE FOR OPTIMAL PATIENT CARE?

David Sacks, MD (Harvard University, Cambridge, MA)

Dr. Sacks argued that the current ISO standards are not sufficiently strict and that they should be replaced by guidelines based on biological variation. In one version of such criteria, a test would be permitted imprecision equal to no more than half of the within-subject biological coefficient of variation, which works out to a maximum allowable total error of 6.9%. Although he did not explain his calculations in detail, Dr. Sacks said that, at minimum, biological-based criteria would require accuracy within 15% (or 15 mg/dl for values less than 100 mg/dl). “Desirable” criteria would demand accuracy within 10%, and “optimal” criteria would require accuracy within 5%.

• Dr. Sacks noted that blood glucose monitoring is a rich field: he cited market research data that point-of-care testing is likely to be an $18.4 billion/year industry in 2013, with diabetes continuing to account for the largest share (Scientia Advisors, September 15, 2010).
• According to the 2003 International Organization of Standardization recommendations (ISO 15197), 95% of meter readings must be accurate within 15 mg/dl for a value less than 75 mg/dl or within 20% for a value equal to or greater than 75 mg/dl. The Clinical and Laboratory Standards Institute (CLSI) published a similar set of standards in 1994: 95% of readings had to be within 15 mg/dl for values at or below 100 mg/dl and within 20% for values above 100 mg/dl. Dr. Sacks said that despite the development of these and other standards (e.g., based on expert opinion, clinical endpoints, state of the art technology, regulatory bodies, or biological variation), there is still not a consensus on how accurate glucose meters need to be for clinical use.
• Dr. Sacks argued that the current ISO standards are too lax. Stating that SMBG’s greatest importance is probably to prevent and detect hypoglycemia, Dr. Sacks applied the ISO standards to the example of a 50 mg/dl blood sugar, for which a reading would be considered accurate if it fell anywhere in the range of 35 to 65 mg/dl. Moreover, one in twenty results (onetest every five days, for a four-times-daily tester) is permitted to fall outside this range. Dr. Sacks argued that this level of accuracy does not permit the reliable detection of hypoglycemia, and he proposed that future criteria should apply to more than 95% of values. He also noted that meter accuracy is lower in the real world than in clinical trials, where measurements are taken by medical technologists.
• Dr. Sacks argued that standards for testing accuracy should be based on biological variability – how much blood glucose fluctuates for an individual. Specifically, he said that a test’s imprecision should not exceed one-half of typical within-subject coefficient of variation (i.e., standard deviation divided by mean). He suggested that the ideal test would have imprecision (i.e., coefficient of variation) of no more than 2.9%, bias of no more than 2.2%, and total error of no more than 6.9%. The current gold standard, clinical laboratory tests, have a coefficient of variability under 1.50%, according to Dr. Sacks’ research in his own lab.
• Looking to future guidelines, Dr. Sacks said that the minimum accuracy criteria should be within 15% (at or above 100 mg/dl) or within 15 mg/dl (below 100 mg/dl). Accuracy criteria within 10% would be “desirable,” and within 5% would be “optimal” (though he acknowledged that a 5% standard would not be a feasible “in the immediate future”). Dr. Sacks chairs a subcommittee that is updating the CLSI guidelines for meters in acute and chronic care facilities, with publication anticipated in late 2011. The ISO guidelines are also under revision; though Dr. Sacks noted that this process is still at a relatively early stage.

PANEL DISCUSSION

Moderators: Patricia Beaston, MD, PhD (Office of in Vitro Diagnostic Device Evaluation and Safety, FDA, Silver Spring, MD) and Kong Chen, PhD (National Institutes of Health, Bethesda, MD)

Questions and Answers

Dr. David Horwitz (LifeScan): Meters have become more accurate over time for competitive reasons if nothing else, and this will continue. Does the accuracy of any number depend on how the number will be used? In essence, there’s not an evidence base that says a more accurate meter will give more positive results. There have been very few outcomes studies since DCCT. Do you think we need to look at whether better accuracy will give better outcomes? Is that the kind of data we really need before we say what the accuracy should be?

Dr. Sacks: I think the problem is that it’s very difficult to do a study of that nature. I’m not aware of anyone who’s talking about doing one or who would pay for it. I think if your company is willing to do it, it would be a very valuable contribution. To say “The DCCT showed people got better, and that’s good enough,” I think that’s probably incorrect, because we can do better. In an absence of large patient studies, the evidence is based on in silico analysis. One can criticize this method, but it’s better than nothing. This evidence suggests that there are fewer errors in insulin dosing with more accurate meters.

Dr. Fran Kaufman (Medtronic): Dr. Perz, do you think that hepatitis B immunization might be useful to reduce the risk to people with diabetes?

Dr. Perz: A CDC Advisory Committee has been considering the expanded use of hepatitis B vaccine among people with diabetes for the past two years. While the vaccine is very useful and could be effective in a long-term care setting, we’re also seeing increased risk of hepatitis C, especially in aging populations. Vaccination in general is appealing, but cost and other concerns are being considered.

Dr. Kaufman: Perhaps I’m glad that I’m a pediatric endocrinologist and at least my patients are protected from hepatitis B.

Dr. Perz: They can still be at risk of other infectious diseases if they’re exposed to contaminants.

Q: Dr. Polonsky, can the testing protocols you showed be downloaded?

Dr. Polonsky: Yes. I think you’d have to go to the Roche website, www.diabetesbehaviorchange.com.

Dr. Robert Bernstein (Private practice, Santa Fe, NM): For Dr. Perz, as a clinician, I often do random blood glucose testing in the office because my patients tend to bring me just their fasting sugars. I was very impressed by your data, and I’m worried because I usually use the same meter. Is there any way to disinfect between patients?

Dr. Perz: If you’re using single-use safety lancets, that offers a certain level of protection. For cleaning and disinfection of the meter, it’s necessary to follow the manufacturer’s guidelines. But often the manuals don’t specify these clearly. I’d begin by looking carefully at the meter and instructions, and after that call the company. Going forward, I think we’ll have more clarity due to the FDA’s letter in September. There’s no one-size-fits-all in terms of approved disinfection for meters.

Dr. Bernstein: It seems like each patient could be given their own meter.

Dr. Perz: Yes, that’s a very good point. I think with the increasing use of blood glucose meters in acute care, it deserves attention. I showed data from a Florida hospital; what I didn’t show was that they intervened by increasing their inventory and moving toward assigning meters to individual patients. I think when you factor in the costs associated with testing, you can make a good argument that this is perhaps a better way to go than investing in the alternative, which is 100% adherence to cleaning and disinfecting between each use. As we contemplate CGM, it seems inherent there and will become the default.

Dr. Marc Torjman (University of Medicine and Dentistry of New Jersey, Camden, NJ): Dr. Sacks, how did you derive the recommendation of 15%?

Dr. Sacks: I didn’t go into it. It’s based on biological variation; it’s just a mathematical formula. I can discuss it later.

Dr. Torjman: Was that also the case for the 10% recommendation?

Dr. Sacks: The idea is having the variability expressed as a percentage of intra-individual variation.

Dr. Torjman: With present technology, my feeling was that 10% is probably the right target, so I appreciated that. Does ISO have a way to deal with outliers?

Dr. Sacks: This issue has been raised and discussed. The manufacturers state, and justifiably, that it’s very difficult to get 100% within 10% or 15%. Let’s say 95% within 10%, and 99% within 15% so that the manufacturer can realistically reach the goal. Otherwise you’d have one outlier and then be done.

Dr. Torjman: Companies report all of the values sometimes, and then when they look at error it’s the error on all of the outliers. As I think the Edwards speaker noted, they included all of the points.

Dr. Sacks: Usually there are certain samples that can be excluded. For example, if someone has hematocrit outside a predefined range, those samples wouldn’t be used. This doesn’t necessarily mimic a real-world situation.

Dr. Bernstein: Is there any harm in a patient re-using the same lancet other than increased discomfort? At times, it’s economical.

Dr. Heinemann: If it hurts too much, patients will not measure. There’s clearly a correlation in frequency and metabolic control. If lancing pain is a barrier to more frequent measurement, we should remove this barrier.

Dr. David Klonoff (Mills-Peninsula Health Services, San Mateo, CA): Dr. Sacks, in your talk you focused mostly on analytical accuracy. What do you think is the role of clinical accuracy, and how should it be measured?

Dr. Sacks: What do you mean by clinical accuracy?

Dr. Klonoff: Does the information you receive lead you to make the correct decision? That’s what led to the Clarke Error Grid and later the Parkes Modified Error Grid.

Dr. Sacks: Everyone intuitively knows that 20% error at 600 mg/dl doesn’t matter, but at 40 mg/dl it matters. The way this has been addressed by the ISO guidelines is to have a 20% percent range at or above 75 mg/dl and a 15 mg/dl range, that is, a fixed number, below 75 mg/dl.

Dr. Klonoff: That sounds as if they’re really focused on analytical accuracy but not clinical accuracy.

Dr. Sacks: Yes.

Insulin Pump Safety and Security: Technical Panel

MICROELECTROMECHANICAL SYSTEMS: A NEW BENCHMARK FOR INSULIN PUMPS

Frederic Neftel, MD (Debiotech SA, Lausanne, Switzerland)

In a data-driven lecture, Dr. Frederic Neftel showed the potential of the Debiotech Jewel Pump to revolutionize insulin-pumping therapy. Reviewing a series of experiments on the accuracy of insulin pumps, Dr. Neftel showed that the Jewel Pump infused insulin at an accuracy of ±5% at all times, while traditional pumps had tremendous variance in accuracy at different doses and time periods, all falling well out of the ±5% range. Dr. Neftel also made to sure to emphasize the Jewel’s benefits over the OmniPod patch pump—better accuracy and smaller size/volume. The presentation was convincing from an engineering point of view, but we remain concerned about the company’s ability to get the Jewel Pump approved with a Bluetooth cellular phone controller – especially given the current regulatory environment.

• Microelectromechanical systems (MEMS) will make insulin pumps more precise, reliable, and inexpensive. MEMS are used in many of our daily life applications, including automotive, GPS navigation, cellular phones, and aircrafts. MEMS chips are manufactured through a batch wafer process. Each eight-inch wafer can produce 400 pumps, allowing a vastly greater scale of production.
• The Jewel Pump is based on a MEMS chip weighing only two grams and incorporating the entire pumping engine and multiple sensors. The pump can deliver single strokes of insulin measuring 0.02 units.
• The Jewel Pump offers significantly greater accuracy over durable pumps. Measured based on accuracy over time (trumpet curves), the Jewel Pump achieves a constant 5% accuracy over all time periods observed, compared with up to ±30% accuracy with current pumps on the market. When accuracy is measured based on the TRAC curve (i.e., what’s actually coming out of the catheter), the Jewel Pump achieves ±5% accuracy at 1.0 unit per hour over a six-day period or at 0.5 units per hour delivered over a 24-hour period. By comparison, durable pumps exhibitsignificant variation in the amount of insulin delivered, ranging from +36% to -32% error over 114 minutes and +38% to -25% error over 30 minutes. According to Dr. Neftel, this variation is due to the barrel piston motor used by traditional pumps.
• The Jewel Pump outperforms the accuracy of patch pumps. At a delivery rate of 1 unit per hour over three days, the patch pump recorded a +28% error at 15 minutes, a -45% error over 15 minutes, and a -35% error over 40 minutes. As noted above, the Jewel achieves an error within ±5% at all time periods.
• Occlusion detection on the Jewel Pump provides faster detection relative to traditional pumps. Durable pumps can take 10-40 times as long to detect occlusion versus the Jewel Pump. Of those studied, patch pumps took the longest to detect occlusion.
• Thanks to MEMS technology, the Jewel Pump will be significantly smaller than the OmniPod. The Jewel is 39% lower in weight and 30% smaller in volume. The new pump will also feature the ability to communicate with a cell phone (telephone capabilities are automatically deactivated during pump use).

BETWEEN PUMPS AND PENS

Jim Peterson, MBA (CeQur SA, Montreux, Switzerland)

In a late afternoon session, Jim Peterson gave the audience a glimpse into the CeQur insulin infuser, a novel device that attempts to bridge the gap between the insulin pump and insulin pen. One of the ideas behind the device is to lower the barrier for insulin therapy, both in terms of cost and training time. The CeQur device is completely mechanical (keeping costs low), has a simple design (just one button for bolusing), and delivers a constant basal rate throughout the day. Although it may be targeting a more niche market, the device seems to hold promise for aiding those who struggle on MDI and are hesitant to switch to a full pump.

• Although, 80% of all insulin is consumed by the type 2 population, those with type 2 diabetes struggle with their insulin therapy. In a February 2010 study appearing in Diabetes Care, it was found that intentional omission of insulin injections occurs in over 50% of adults using insulin to treat their diabetes and is common in roughly 20% of these individuals (Peyrot, et al., “Correlates of Insulin Injection Omission.”)Technology can help people with type 2 diabetes achieve better glycemic outcomes. In the U.S., 350,000 patients move to MDI every year, while another 50,000 patients are resistant to MDI therapy. These patients can be helped by technology. Indeed, studies have shown that insulin pump therapy can improve blood glucose control in type 2 diabetes. At ADA 2010, a study of 16 patients achieving suboptimal control on MDI concluded that CSII therapy using a simple dosing regimen significantly improved glycemic control. For 90% of patients, only one basal rate was required.
• The CeQur insulin infuser falls somewhere between insulin pens and pumps, offering the benefits of both technologies in a user-friendly interface. The device is completely mechanical, provides continuous subcutaneous insulin infusion, and is targeted at the type 2 diabetes audience. Boluses are administered with the simple push of a button while a constant basal rate is delivered around the clock (choice of seven rates). Notably, the cost profile and training time required for the CeQur insulin infuser is similar to standard insulin pens. Finally, the user is notified when the three-day reservoir is depleted or disrupted and the small battery that powers the unit lasts for one year.
• The CeQur device will be in clinical trials in Europe this January, with data coming sometime in 2011. The hope is that the user-friendly device will boost treatment compliance and result in better outcomes.

BOLUS CALCULATORS IN INSULIN PUMPS TO PROMOTE SAFETY

Ofer Yodfat, MD (Medingo, LTD, Yoqneam Illit, Israel)

Dr. Ofer Yodfat gave an engaging presentation on the basics of bolus calculators and the challenges that pump patients often face while using them. Starting with the fundamentals, Dr. Yodfat described the types of boluses and the typical calculations they perform. Quickly transitioning to challenges, he identified the typical issues with bolus calculators that patients encounter, including errors in entering or setting formula parameters, pump algorithm variations, and carbohydrate estimation errors. In an interactive test, Dr. Yodfat put several pictures on the screen and asked the audience to estimate the carbohydrate content of the meals. It was clear from the results that audience members would be poor patients indeed! Additionally, when the pictures were shown in a study to CDEs, the educators’ estimates varied considerably and had a strong impact on the dose taken by the patient. Dr. Yodfat also went through a patient study of carbohydrate estimation – 60 people with type 1 diabetes were asked to estimate the carbohydrate content of eight meals. The estimated error distribution (coefficient of variation) of typical meals ranged from 27.9% to 44.5%. Moreover, the data showed that patients tend to significantly underestimate the carbohydrate content of higher carb meals. Turning to algorithm parameters, Dr. Yodfat encouraged using 500 divided by total daily dose for calculating insulin to carbohydrate ratio and between 1600 (more aggressive) and 2200 (more conservative) divided by total daily dose for calculating insulin sensitivity. He closed by revealing the surprising differences between bolus calculators on different pumps. Interestingly, target blood glucose, correction and meal boluses, and insulin on board calculations can all vary drastically between pumps. In a case study of a Medtronic, a Cozmo, and an Animas pump, the recommended doses for 30 carbohydrates varied by up to three units based on the different bolus calculators in the pumps. Dr. Yodfat’s presentation once again reminded us how difficult managing type 1 diabetes is, even with the great technology patients have access to.

INSULIN PUMP SAFETY FOR TYPE 2 DIABETES

Poul Strange, MD, PhD (Integrated Medical Development, Bridgewater, NJ)

Dr. Strange gave a nice overview of the safety issues that accompany insulin pump use in type 2 diabetes. Starting broadly, Dr. Strange outlined some of the statistics on pump use and concluded that pump use in the type 2 diabetes population will likely increase in the coming years. Transitioning to the concept of intensive insulin therapy, he noted that the major barriers are fear of hypoglycemia, stigmatization and hardship associated with insulin therapy. Dr. Strange also argued that pumps may offer benefits for patients that don’t like insulin injections. In an interesting survey of 700 patients (90% with type 2 diabetes), only 20% of those on intensive therapy (three or more injections per day) and only 6% of those on conventional therapy injected away from home. From the perspective of safety, use of insulin pumps in type 2 diabetes carries less risk of hypoglycemia and DKA. While there are human interface issues (e.g., training and troubleshooting), Dr. Strange believes that there is no indication that safety with insulin pumps is more of a concern in type 2 diabetes. He closed by reiterating his belief that this population will increasingly turn to insulin pumps in the coming years.

PANEL DISCUSSION

Moderators: Athanassios Sambanis, PhD (Georgia Institute of Technology, Atlanta, GA) and Jan Wojcicki, PhD, DSc (Institute of Biocybernetics, Warsaw, Poland)

Questions and Answers

Comment: To Dr. Yodfat; on the 450 and 500 rules, we’re no longer using that to determine insulin to carbohydrate ratio. In a study by Paul Davidson, it was shown that carbohydrate factor is related to weight and total daily dose. We’re currently recommending 2.6 times weight divided by total daily dose as the carbohydrate factor. I think that’s more appropriate. For the correction factor, it’s hard to use an exact rule. Someone with poor control needs more insulin than someone with lower insulin requirements.

Q: In the Debiotech Jewel Pump, you have some kind of a monitoring device that works by pressure distribution. Have you controlled the potential for a catheter disconnection or is this undetectable?

Dr. Neftel: It’s a different issue. I would not make the claim that we can detect it, but we are closer than we’ve ever been.

Q: I think there might be an application for the iPhone that allows a user to take a picture of their meal and it will count the carbohydrates. Would this enable more accurate estimation?

Dr. Yodfat: I don’t think that will work. It’s very hard to estimate the volume of carbohydrates in a picture. Think of rice in a bowl as an example. It’s tough to really gauge the volume of food and accurately estimate the carbohydrates.

Q: Do you have plans to compare your pump against another pump in a clinical trial?

A: Looking at a trial to compare pumps with a daily shot of Lantus, you’d wonder which produced the flatter profile. Even given the trumpet curves, the standard of today’s pumps produces a flatter response than long-acting insulin. At the end of the day, when you talk about accuracy, the pharmacokinetic link is missing. It really is the effect on the glucose control of the patient over time.

Dr. Neftel: Regarding accuracy, there are several factors that play a role—the insulin you inject and the amount. You want to tightly control the amount. Industry is working on faster acting insulin and different modes of delivery. Accuracy is also important for the closed loop in the future. If you realize that the pump is inaccurate, the whole algorithm is wrong. We hope to show that tighter control with our pump will improve closed-loop algorithms and control over time.

Insulin Pump Safety and Security: Clinical Panel

INSULIN PUMP SHUT-OFF FOR HYPOGLYCEMIA

Pratik Choudhary, MD, MBBS, MRCP (King’s College London, London, UK)

In an anticipated talk, Dr. Pratik Choudhary built a strong case for the safety and effectiveness of the Medtronic Paradigm Veo system. From a physiological perspective, Dr. Choudhary argued that low- glucose suspend systems make perfect sense, as shutting off insulin mimics the non-diabetic’s response to hypoglycemia. He presented data from a small study of the Veo in the United Kingdom; the low- glucose suspend feature of the pump was effective at reducing time spent at < 70 mg/dl with no significant hyperglycemia reported. The vast majority of low-glucose suspend alarms occurred during the day and were responded to within ten minutes. Overall, the system was well tolerated and liked by patients.

• Current technology is insufficient to prevent hypoglycemia. The JDRF CGM trial and STAR-1 studies found that CGM and sensor-augmented pumps do not prevent hypoglycemia. Additionally, a study by Bruce Buckingham found that 71% of patients seem to sleep through nocturnal alarms. In addition to the acute danger of hypoglycemia, low blood sugar may also affect the heart.
• Insulin pump suspension when the patient is already hypoglycemic is less effective than suspending when hypoglycemia is predicted. Studies by Drs. Bruce Buckingham and Roman Hovorka have demonstrated that a 90 minute or 165 minute suspension of insulin can dramatically prevent hypoglycemia.
• A low-glucose suspend (LGS) system makes perfect sense, as the body physiologically responds to hypoglycemia by shutting down insulin production. The Medtronic Paradigm Veo pump has the same features as the Revel, but will also automatically suspend insulin delivery for two hours if the patient fails to respond to alerts. Patients can restart insulin delivery at any time during the two-hour period.
• Audience poll question: What is your biggest concern about using automated insulin suspension?
• Risk of sensor malfunction causing insulin suspension in absence of hypoglycemia: 45%Is a two hour suspend enough to prevent hypoglycemia?: 30%Rebound hyperglycemia: 19%Risk of ketosis: 6%Ketosis associated with insulin pump suspension is not a concern. Even after five hours of insulin suspension, ketone levels don’t get nearly high enough to be associated with diabetic ketoacidosis (DKA). Additionally, if the pump is suspended during hypoglycemia, a less rapid rise in blood sugar occurs because insulin levels are already high.
• A recent study of 29 subjects in the UK demonstrates the benefits of the Paradigm Veo. The mean age of participants was 41 years, with A1c ranging from 6.5% to 9.2%. The study included a two-week run-in period with use of CGM. This was followed by four weeks of therapy with the Veo LGS system.
• 75% of LGS events occurred during the day (8AM to midnight), and in 70% of cases, the patient resumed insulin infusion within ten minutes of the alarm. The remaining 25% of LGS events occurred between 12AM and 8AM. In total, 166 events of LGS occurred throughout the study. Events with a longer alarm response time were more associated with nighttime occurrence. These data demonstrate that the LGS typically occurs during the day, when the patient is awake and aware of what is happening.
• The Veo’s LGS led to a significant reduction in overall time spent below 70 mg/dl. Across the whole study, those with LGS turned off had a mean duration of 1.7 hours per day of hypoglycemia. With LGS turned on, the mean duration of hypoglycemia was 0.6 hours per day.
• LGS was not associated with rebound hyperglycemia. Mean sensor glucose on resumption of basal insulin was 75 mg/dl in the study. In looking at the ten events where pumpsuspension lasted for two hours, glucose ranged from 4.9 mmol/l (88.2 mg/dl) to 8.9 mmol/l (160 mg/dl) upon basal resumption. Ketones were not measured in the study, but there were no reported clinical episodes of ketosis.

RECOGNITION AND MANAGEMENT OF CLINICAL RISKS ASSOCIATED WITH INSULIN PUMP THERAPY

Lori Laffel, MD, MPH (Joslin Diabetes Center, Harvard University, Boston, MA)

Dr. Lori Laffel gave a very patient-focused overview of the risks associated with pump therapy. She started by outlining many of the benefits of pump therapy, including better glycemic outcomes and greater flexibility. She transitioned to the serious problem of missed boluses, which can have an astonishing 1% adverse effect on A1c after just one missed bolus every other day. Dr. Laffel concluded by emphasizing that insulin pump therapy is more than just button pushing; it requires multi-disciplinary education and support, consistent and frequent blood glucose monitoring, exercise management, hypoglycemia knowledge, and emotional and psychological stability.

• Insulin pump therapy offers tremendous benefits and opportunities for patients. These include greater flexibility, reduced blood glucose fluctuations, a more physiologic delivery of insulin, better quality of life, and reduced severe hypoglycemia. To illustrate the benefits of insulin pump therapy, Dr. Laffel quoted a survey that found 63% of patients used a pump to improve their glycemic control, 43% wanted increased flexibility, 9% wanted fewer injections, 8% of patients were on pumps for food-related reasons, and 2% preferred insulin pumping because of summer camp.
• Insulin pumps can improve glycemic outcomes over standard therapy. A study by Doyle et al. in Diabetes Care (July 2004) compared glargine-based MDI to CSII in 32 patients ages 8-21 years. The glargine group recorded no significant improvement in A1c (8.2% at baseline to 8.1% after 16 weeks) while the CSII group recorded a significant drop in A1c (8.1% at baseline to 7.2% after 16 weeks; p < 0.02 vs. baseline and p < 0.05 vs. glargine). An analysis of 1,692 pumpers at the Joslin Clinic found that pump therapy was associated with greater attainment of A1c goals. Of those using pumps, 83% met the A1c goal of 8.5% vs. 60% of those on MDI; 51% met the A1c goal of 8.0% vs. 32% on MDI. However, for adolescents between 13 and 19 years old, the study found that pump therapy demonstrated no improvement in terms of meeting glycemic goals.
• Pump therapy brings with it challenges: missing boluses, device failure, the need for more frequent blood glucose monitoring, ketone measurement, potentially decreased family involvement, and more personal responsibility. Dr. Laffel presented data that showed missing just one bolus of insulin every other day increased A1c by 1%. In her mind, this makes it “extraordinarily impossible to be human with diabetes,” as we all have an easy time forgetting things. A study by Peter Chase provided reminder bolus alarms, which helped with remembering boluses after three months, but after six months, patients got tired of alarms and the benefits were lost. Another study, appearing in Pediatric Diabetes in 2009 (Olinder et al.), showed that 38% of 12-18 year old adolescents missed insulin boluses. The youth who forgot to bolus had a 0.8% higher A1c (7.8% vs. 7.0%). Analysis showed the variations in A1c could be explained by the frequency of bolus doses (p=0.013).
• Hypo- and hyperglycemia experienced on pump therapy generally have distinct causes. The major factors leading to hypoglycemia are impatience, insulin stacking, wrong carbohydrate estimation, no adjustment for exercise, and incorrect dosing. The major causes ofhyperglycemia while on pumps include missed boluses, incorrect basal rates, improper time settings, and occluded catheters (about one per year per patient). In Dr. Laffel’s opinion, extremely accurate carbohydrate counting is much less important than a pre-meal bolus. She reported that as long as a patient is within 15% of the true carbohydrate count, postprandial glucose will be the same.

REGULATORY CONSIDERATIONS FOR INSULIN PUMP SAFETY AND SECURITY

Paul L. Jones MSCE, CDP, CSQE (FDA, Silver Spring, MD)

Mr. Jones, whose PowerPoint slides were unfortunately not working, gave the FDA’s perspective on the risks of digital and wireless communications technology. While these advances have enhanced patient safety and quality of life, they also have many issues regarding safety. Mr. Jones was specifically worried about wireless communication between devices; he mentioned that the FDA would want to see that a manufacturer had considered potential safety and security issues in these insulin pump systems. In the best case scenario, a hazard analysis submitted by manufacturers would help the FDA more rapidly understand the security risks associated with a medical device. During Q&A, it was abundantly clear that Mr. Jones and moderator Carol Herman, RN (Center for Devices and Radiological Health, FDA, Silver Spring, MD) – and by extension we assume the FDA – were way, way behind the curve on these threats and what constitutes an adequate response from a design point of view. From a patient perspective, we also lament these potentially troubling requirements. The artificial pancreas and closed- loop control will not become a reality unless devices CAN talk to one another.

MITIGATING SECURITY SOLUTIONS FOR ATTACKS ON INSULIN PUMP SYSTEMS

Nathanael R. Paul, PhD (Oak Ridge National Laboratory, Oak Ridge, TN)

Dr. Paul, a type 1 diabetes patient himself, gave a very general talk on some of the security vulnerabilities that insulin pump systems can present. He was especially concerned with closed-loop systems, where many devices communicate with one another wirelessly and messages could be intercepted or changed. In his opinion, as the system gets more complex, it gets more vulnerable to malicious attacks from outsiders. He seemed convinced that there was potential for such actions and encouraged the future use of encryption and other security strategies to protect patent safety, confidentiality, and data.

SECURITY AND PRIVACY FOR WIRELESS PERSONAL MEDICAL DEVICES: EXPERIMENTAL RESULTS, CHALLENGES, AND NEW DIRECTIONS

Tadayoshi Kohno, PhD (University of Washington, Seattle, WA)

In one of the more off-the-wall presentations at the conference, Dr. Tadayoshi Kohno presented some fascinating research on the security of wireless technology. To start, Dr. Kohno gave an overview of the technological trends towards greater computational capabilities, wireless capabilities, and integrated, multi-agent systems. However, before delving into his research, Dr. Kohno emphasized that the current risks to patients are small, and researchers want to keep them small. His first study involved an FDA- approved implantable cardiac defibrillator made in 2003. Using their own wireless equipment, Dr. Kohno and his team were outrageously able to obtain patient information, change therapies, disable therapies, and discharge a large electric shock. In a second example, he and his team were able to remotely access a car and interface with the internal computer. The researchers were able to set the speedometer to 140 mph while in park, put messages on the dashboard, turn off all the lights remotely, wirelessly apply and disengage the brakes, and even install a virus in the car that would activate once certain conditions were met (e.g., speed exceeds 20 mph and the windshield wipers turn on). Dr. Kohno noted that with medical devices, we must anticipate the threat rather than follow it. By assessing the range of threats and understanding how they can be addressed, researchers and regulators can make patients safer.

THE ALGORITHM OF INSULIN PROGRAMMING FOR CONTINUOUS SUCBCUTANEOUS INSULIN INFUSION: THE WARSAW PUMP THERAPY SCHOOL

Ewa Pankowska, MD, PhD (Institute of Mother and Child, Warsaw, Poland)

Dr. Ewa Pankowska gave a simple overview of the basics of pump therapy. She commenced by citing the physiological basis for continuous subcutaneous insulin. She reviewed a study that showed meal insulin secretion is more than fifteen times higher than background insulin needs in non-diabetic subjects. Additionally, she cited studies that uncovered the major factors that affect basal insulin requirements: age, duration of diabetes, BMI, and C-peptide levels. Studies have also found that the percentage of total daily dose as basal insulin tends to increase with age. As a result, the proportion of basal-to-bolus insulin could be differentiated from 10% to 60% when CSII is introduced. In terms of achieving better outcomes, she emphasized that lower basal rates and using more dual-wave boluses are both factors that are correlated with better A1c outcomes. Dr. Pankowska wrapped up by summarizing a study conducted in her center. Patients ate pizza for dinner (45 carbohydrates and 400 calories from fat-protein products) and were asked to use either a normal bolus (only account for carbohydrates) or dual-wave bolus (accounts for carbohydrates, protein, and fat). Those using the dual- wave bolus were better able to control postprandial glycemia (after two hours: 88 mg/dl vs. 163 mg/dl; after four hours: 93 mg/dl vs. 196 mg/dl). According to Dr. Pankowska, individualizing insulin programming in pump therapy is one way important way to achieve metabolic control.

PANEL DISCUSSION

Moderators: Dorian Liepmann, PhD (University of California, Berkeley, Berkeley, CA) and Carol Herman, RN (Center for Devices and Radiological Health, FDA, Silver Spring, MD)

Questions and Answers

Dr. Darren Wilson (Stanford University, Palo Alto, CA): I don’t like what I’m hearing about isolating systems. This will really hinder the benefits of these systems. We don’t want these devices to be silos. We shouldn’t get to a point where the only safe computer is the computer that’s turned off.

Q: There’s a need of these devices to interact with each other across manufacturer platforms. We also have big issues downloading devices. I’m concerned you’re going to make the problem worse by adding layers of security. I actually see the download and communication problems as a much more clear and pressing problem than security. We don’t just want to make things harder.

Dr. Kohno: That’s a great insight and observation. With multiple manufacturers and multiple devices, integration is hard as it is and you don’t want to make it harder. Integration brings about vulnerabilities. If we can make it easier to integrate components, we may be able to address security issues too. It needs to be an industry-wide effort.

Q: Can you hack a Bluetooth phone?

Dr. Kohno: We haven’t done it that quickly in our lab, but other researchers have succeeded. Dr. Paul: A mobile phone worm was released a few years ago as a proof of concept.

Comment (Medtronic): No system in the world is secure. It may be helpful if the agency adopts security standards instead of leaving it open-ended.

Mr. Jones: I would agree with that, and I would raise the notion of acceptability criteria. It depends on the state of the technology and what reasonable claims can be made about the security, As the standards bodies come up with technology that will make security more rigorous and robust, we would use the new standards and control measures.

Dr. Kenneth Ward (Oregon Health and Science University, Portland, OR): Consolidating devices makes things easier for patients. Carrying all the different devices frustrates patients. The idea has been proposed that we put everything in the cell phone. As I hear you talk, however, I’m thinking you want all this separate to preserve security.

Mr. Jones: The basic architecture I discussed was a crude model to illustrate the issues. It’s really up to the manufacturer to come up with something.

Dr. Paul: If you put an artificial pancreas controller mechanism in the phone, you could have someone around the world controlling your pump. These are questions about using untrusted devices in a medical device system.

Dr. Kohno: Keeping things isolated is a simple approach to security. But I think with appropriate security mechanisms we can consolidate. But we need appropriate standards. Who will set the bar for standards? And let’s remember that there is no such thing as completely secure. And finally, how do we patch the software for medical devices when there is a breach?

Ms. Herman: Phone companies are not that interested in the medical device space. They don’t want to go through the FDA ,and phone companies don’t want the medical device aspect of the device to detract from use of the cell phone. If the battery on the phone is dying, the medical device on the phone takes priority over calling, texting, etc... No company wants this.

Q: I didn’t hear any discussion about the insertion of malware at the manufacturing source of chips. These chips are made outside the United States.

Dr. Kohno: I can’t answer that from a medical device perspective, but it could be a big problem and a broad problem. It’s a problem that needs to be addressed.

Dr. Paul: One approach is to keep devices isolated so that collusion would be required for manufacturers to compromise the system.

Comment (Medtronic): There are a few techniques for checking the integrity of chips. It is a complicated question. FDA is not there yet and doesn’t check this. But it involves tests on the chip level to ensure that the chip is responding correctly.

Dr. Jeff Reynolds (Bayer Diabetes Care): When you look at risk analysis, you are looking at risks to a patient. But for malicious intent, you are talking about groups of patients. Is there a way to handle that formally?

Dr. Kohno: Doing risk analysis in the academic computer space is difficult because malicious intent is very unpredictable.

Dr. Paul: We don’t have a way to articulate all the issues. All we can do is publish the design and get comments about it. And some threats are more serious than others.

Dr. John Mastrototaro (Medtronic): To the FDA, when you’re evaluating medical devices, you are evaluating the security of a device in a standard situation. For instance, pumps communicating with controllers have unique IDs so they don’t get mixed up with other pumps. But what about outside standard situations? Is the FDA going to institute requirements to prevent malicious intent?

Mr. Jones: I can’t really talk about whether it’s going to become a requirement or not. At some level, all we’re expecting is that the design will effectively take into account potential security risks.

Ms. Herman: The answer is we don’t know. Technology will determine how far we need to take it.

Lane Desborough (Medtronic): Why don’t we look to other domains for regulations and standards? This happened in the oil refining industries. Other domains have wrestled with these questions and have come to an understanding.

Ms. Herman: In the past, we’ve had healthcare specific standards. We’re starting to look at standards in other industries.

Mr. Jones: You can borrow from other technology. For instance, separating the safety system from the control architecture.

Q: I’m a physician from North Carolina, and I sign in to my office computer from home with access to a patient database. My only way of getting in is a six-digit rotating code that changes every minute. Can we use something like this?

Ms. Herman: We actually use this at FDA, and we’re very familiar with it.

Dr. Kohno: This does have potential. But in medicine, if you’ve secured it, you’ve prevented emergency teams from accessing the device. So we may get more security, but there may be things about the field of medical devices that make it unique from other industries in terms of security.

The New Bundled World – Monitoring Diabetes Following Hospital Discharge

Moderators: John Ryan, MBA (Onset Ventures, Menlo Park, CA) and Charlene Quinn, RN, PhD (University of Maryland, School of Medicine, Baltimore, MD)

REIMBURSEMENT

Edward Dougherty, MS, MA (B&D Consulting, LLC, Washington, DC)

Addressing the short-term future of America’s healthcare system, Mr. Dougherty said that the fee-for- service model doesn’t work (especially for chronic care) and that it is going away. He criticized Medicare’s attempt to lower costs by cutting payments across the board, a “blunt instrument” approach that he said has led providers and insurers to create inefficient and costly workarounds. He believes that coordinated care models are likely to succeed; although early studies have suggested they may not save money, Mr. Dougherty said the patients who enrolled in these studies tended to be sick and would have required care anyway. He also anticipated benefits from broadening the definition of “episode of care,” a process in which he encouraged the audience to engage with policymakers. Closing with several pieces of advice for companies and individuals, Mr. Dougherty said it’s clear that payment reform is happening and that there will be winners and losers in the transition.

• The presentation addressed several topics that Mr. Dougherty said would likely play important roles in the new healthcare system, including: quality reporting and payment incentives; prospective, bundled payments with shared risk (and the aforementioned quality bonuses); decision support (e.g., a provider remotely discussing treatment options with a peer); medical homes (likely to be increasingly used for chronic care); accountable care organizations (though Mr. Dougherty noted a government analyst had recently challenged the extent of their proposed cost savings); and redefining episodes of care.
• Mr. Dougherty outlined several key success factors likely to characterize a successful healthcare model: 1) predictable course of care, 2) clear, effective clinical guidelines, 3) information technology capable of supporting interaction and analysis, 4) clear roles and responsibilities, and 5) appropriate documentation.
• Mr. Dougherty forecasted that successful companies will link data to patient demographics, work well with competitors and collaborators, meet with payers early and often (once they are ready), and have advocates to Congress. He recommended that investors challenge their companies to develop good data; he said that small companies planning to do nothing more than get FDA approval are unlikely to succeed in the long term.

Thomas Foels, MD, MMM (Independent Health Association, Williamsville, NY)

Dr. Foels explained that pay-for-performance is evolving from evaluating process measures (e.g., whether the provider ordered an A1c test) to outcome measures (e.g., whether the patient’s A1c decreased over time), a transition that requires a move toward electronic medical records. He also suggested that tiered fee schedules and public transparency would be strong motivators of quality improvement. Noting that data on the benefits of pay-for-performance are limited and inconclusive, Dr. Foels discussed the implementation of a pay-for-performance system for diabetes care at his own institution. Although quality improved across a series of metrics, Dr. Foels questioned whether the changes to provider behavior would be deep or long lasting. To close, he proclaimed that he remains a pay-for-performance agnostic.

• Dr. Foels listed seven steps that Independent Health used to design its pay-for- performance model of diabetes care: 1) define the denominator (i.e., the total number of patients), 2) define the numerator (i.e., the number of patients who should be measured for a particular goal), 3) assign physicians, 4) set performance thresholds (e.g., should there be a rigid goal of A1c below 7.0%, or multiple thresholds?) 5) determine incentives, 6) develop a system for reporting results, and 7) establish a process for correcting data errors.
• The Independent Health “Practice Excellence” program rewards providers with a bonus of up to 10% for improving diabetes care. The program has now passed through 10 six-month assessment cycles, and providers’ scores have risen across a combination of process measures (e.g., checking A1c regularly, examining retinas annually) and outcome measures (e.g., A1c less than 7.0%, blood pressure less than 130/80 mm Hg). Dr. Foels noted that these improvements did not simply reflect US trends; his institution’s Healthcare Data and Information Set (HEDIS) score moved from the 50th to the 90th percentile nationally.
• Calling himself a pay-for-performance agnostic, Dr. Foels looked forward to learning more over time. He said that although the Independent Health “pay-for-compliance” program was intended to foster a culture of learning and discovery. However, many providersseem to have taken a narrow, “paint-by-numbers” approach that may not indicate fundamental, long-term change.

Jay Shubrook, DO, FACOFP, FAAFP (Ohio University, Athens, OH)

Dr. Shubrook shared a clinician’s perspective on the intricacies of quality measurement systems, saying “I’ll be the first to tell you, physicians don’t get it.” He noted that reimbursement based on outcome measures can be problematic, since some physicians (e.g., specialists) tend to see patients who are already sicker or less compliant; Dr. Shubrook called for a system that accounts for the complexity of care required. He discussed bundled reports of care as a way to broadly assess provider performance, saying that he favored indicator-level bundles (e.g., what percentage of patients meet their A1c goal) instead of patient-level bundles, which assign each patient an all-or-none score based on whether all the goals for that patient were met. He also highlighted the importance of transparency and fairness in patient assignment and payment, and he emphasized that systems should address extra-clinical factors (e.g., race, insurance status, absenteeism) that correlate strongly with glucose control.

• Dr. Shubrook explained that he favored measuring performance based on how often indicated care was delivered (indicator-level bundles) rather than how many patients received all the indicated care (“all-or-none” patient-level bundles). To illustrate the difference, he showed a figure representing five patients who respectively met three, three, four, four, and five of five possible outcomes measures. By the indicator-bundle method, the provider’s score would be 80%, because 20 out of 25 total indicators were met. Alternatively, the patient-bundle method would give a score of 20%, because only one of the five patients met all the outcome measures. (In a retrospective study of diabetes care delivered to 7,333 patients at 95 centers, the discrepancies between indicator- and patient-level analysis were almost this great, showing the real-world importance of defining how care is measured [Shubrook et al., Am J Managed Care 2010].) Dr. Shubrook favored the indicator-level method as more comprehensive and more reasonable for providers, given that many patients may be unable to meet every single outcome measure despite receiving quality care. He noted that when using indicator-level analysis, it is important to decide whether separate measures should be treated as equivalent (as in the example he showed) or weighted differently.

Judy Chen, MD (IMS Health, Woodland Hills, CA)

Dr. Chen reviewed years of clinical literature about reducing hospital readmission for patients with diabetes, and she discussed a number of factors associated with rates of readmission. Several of these factors were patient-specific: age, race/ethnicity, and prior utilization of care. (In one study being analyzed by Dr. Chen’s group, patients who had been seen by multiple primary care physicians in the prior year had an 80% higher risk of readmission – evidence for the value of coordinated care, she noted). She explained that readmission rates have been shown to differ depending on quality of outpatient care, inpatient care (e.g., Levetan et al., Diabetes Care 1999), and discharge procedure (e.g., Cook et al., Endocr Pract 2009). Dr. Chen concluded that “you can start anywhere” to minimize readmission, but she said that it’s ultimately important to have quality at every stage of care.

PANEL DISCUSSION

Questions and Answers

Q: When my son was diagnosed with type 1 diabetes a year ago, our first thought was, “Dang, that’s 20 years off his life.” The second was, “We’ll do anything to shorten that number.” The average is one way to characterize a distribution, but my question is around glycemic variability and whether any work is being done to correlate the shape of the distribution to outcomes.

Dr. Shubrook: I think most people who treat diabetes routinely know that A1c is a wonderful way to communicate but also to some extent a disadvantage because it doesn’t tell the glycemic experience. We routinely do look at downloads of meters, pumps, and sensors; we talk about reducing variability because it does make a difference. Right now all I can tell you is that hypoglycemia is bad. But I think you will see that the distance traveled over a day – that is, greater glycemic variability – even in the normal range, may be a bad thing.

Q: An episode of care was defined as three days prior to admission until 30 days after discharge; maybe glycemic variability is an important marker of why they came to the hospital.

Dr. Shubrook: There are lots of measures of variability, but none has become standard.

Q: I have a question for Mr. Dougherty. You used an interesting phrase: patient demographics. Could you explain how you define this in the context of payment?

Mr. Dougherty: It has to do with the predictability of the intensity of service you might be required to provide. Anyone entering into a bundled payment agreement needs to understand whether they have sicker or healthier patients and therefore what level of service is required. If you get that wrong, you’re guaranteed to lose significant revenue. Provider systems are looking at their own data sets closely now, and they don’t have a good handle on what the actual range of patients is. At the end of the day they’re likely to agree on a prospective payment system with Medicare or a commercial payer and some sort of stop-loss agreement for patients at the far end of the range. But you can’t really do this without understanding what the patient population will look like in the coming year.

Q: So you’re not really talking about demographics in the anthropological sense of community; you’re over here in compliance and accountability.

A: Our blunt instruments a few years ago were age and gender. We had the data, but the correlation with risk was pretty small. More recent models using comorbidities probably get you to a correlation coefficient of 0.4 or 0.5, considered excellent. What we don’t have good data on is the socioeconomic status factors that probably account for the rest of the variability. Capitating failed 15 years ago, but it will likely return with better tools and monitoring – though given the lack of some data and other subtle factors, it’s not exact, as you can probably guess.

INFORMATION TECHNOLOGY FOR RECENTLY DISCHARGED PATIENTS

Chris Mannasseh, MD (Boston University Medical Center, Boston, MA)

Dr. Mannasseh described the development of Project Re-Engineered Discharge (RED), Boston University Medical Center’s modified discharge process to improve provider follow-up and patient self- management. The intervention involved two key players: the discharge advocate (a hospital nurse who interacts with the care team and trains patients to manage their own care) and a clinical pharmacist (who reviews medications and addresses other concerns in the 72 hours following discharge). Patients were also given customized, comprehensible after-hospital care plans in spiral-bound booklets. In a randomized controlled trial of 749 patients, the intervention saved $412 per patient compared to standard care, but it was also time-intensive. Clinical pharmacists spent roughly 30 minutes with each patient, and discharge advocates spent roughly 90 minutes – unfeasible in a real-world setting, where inpatient caregivers spend an average of eight minutes per patient reviewing discharge instructions and answering patients’ questions about their hospitalization. Consequently, Dr. Mannasseh and his colleagues developed virtual discharge advocates: animated characters that can engage patients in basic conversation and teach them the hospital discharge plan. Notably, 68% of patients preferred the virtual discharge advocates to human caregivers, saying that physicians tend to be rushed and to not listen to the patient’s concerns. Dr. Mannasseh concluded that computerized interventions of this kind could be a cost-effective way to improve care delivery, and he looked forward to stronger business cases for health information technology.

• The Project RED intervention included spiral-bound notebooks containing customized after-hospital care plans. Designed in collaboration with the Rhode Island School of Design to reach patients with limited health literacy, each booklet contains: pictures of the patient’s discharge advocate and primary care physician, a color-coded medication guide, a color-coded appointment page that includes all of a patient’s specialists, an appointment calendar with a magnet (for placement on a refrigerator), a list of tests for which results were still pending at time of discharge, and an educational page (titled “My Medical Problem”) that includes an illustrated description of the patient’s health condition(s).
• To reduce the caregiver time required to prepare patients for after-hospital care, Dr. Mannasseh and his colleagues worked with an MIT fellow to develop virtual discharge advocates. These consisted of cartoon healthcare providers that were displayed on kiosks at a patient’s bedside; they were designed to simulate the empathetic gaze and gestures of human nurses. The virtual characters talked to patients to prepare them for discharge and assess their competence, with patients able to interact by selecting from predetermined responses. (The characters even make small talk. An example (best if spoken aloud in a female robot voice): “So, are you a Red Sox fan?... That is great, I really hope I see a game someday, but they don’t allow computers at Fenway Park.”) Notably, 68% of patients studied said they preferred the virtual discharge advocates to human caregivers, noting that the computers (unlike many physicians) took the time to speak with them and address their concerns. Following discharge, patients could access the virtual discharge advocate online, along with other web-based resources. Dr. Mannasseh and his colleagues are assessing this version of the intervention in an ongoing trial.

Kathleen Dungan, MD (Ohio State University, Columbus, OH)

Dr. Dungan discussed the little-studied relationship between glycemic control and hospital readmission rates. Although diabetes is not considered one of the top causes of readmission, Dr. Dungan noted that this doesn’t mean it is not an important contributor to risk. She reviewed a handful of studies, including a retrospective analysis of nearly 10,000 managed care patients in which A1c predicted readmission only at levels above 10.0% (Menzin et al., J Manag Care Pharm 2010) and a retrospective analysis of nearly 2,000 renal transplantation patients in which post-transplant hyperglycemia was associated with a 2.9 risk ratio for readmission (Hosseini et al., Ann Transplant 2007). She also discussed a retrospective study conducted by her own group in which both A1c and inpatient glycemia were associated with 30-90 day readmission rates (Dungan et al., Endocr Pract 2010). Noting that cardiac journals had not been interested in this research, Dr. Dungan called for greater attention to glycemic control outside the diabetes community.

• Dr. Dungan discussed a retrospective study conducted by her own group that included 749 congestive heart failure patients who had received a fingerstick glucose measurements while hospitalized (a proxy for diabetes, which is often under- recorded during inpatient care). Significant predictors of 30-90 day readmission included time- weighted mean glucose (based on 9,236 total measurements, an average of 12.3 per patient), A1c, renal disease, African American race, and year of hospital admission (2005 or 2006). Glycemic variability as measured by glycemic lability index was not predictive of readmission rates (Dungan et al., Endocr Pract 2010).

Patricia E. Sokol, RN, JD (Institute for Healthcare Improvement, Cambridge, MA)

Ms. Sokol discussed proactive systems for monitoring risk after patients leave the hospital. Saying that the information technology tools needed could be delivered in a variety of platforms, Ms. Sokol focused on the general goals of risk management rather than specific implementation issues. She said that patients are typically sent home with suboptimal resources for self-management and that providers typically know little about their current or ongoing risk factors. Instead, she advocated a proactive, bidirectional “safety science module” so that providers can monitor patients’ status and know if home visits are needed.

• Ms. Sokol said that a good post-discharge plan should proactively identify, monitor, and address risks. She identified several considerations for a good risk monitoring platform, including: 1) reliance and trust (so that the process can be “shame-free” for patients), 2) timeliness (with faster response when risks are higher), 3) reliability (which she said depends largely on the equipment used), and 4) documentation (a topic familiar to her as an attorney – she said that the discharge procedure should be considered just as structured as other medical records).

PANEL DISCUSSION

Questions and Answers

Q: What were the compliance rates at baseline with the post-discharge program at BU?

Dr. Mannasseh: It was 72% for the intervention group and roughly 40% with traditional care.

Q: For the BU hospital discharge program, did you consider putting some of that feedback on patients’ cell phones rather than in a book or online?

Dr. Mannasseh: We didn’t, but that’s a great idea.

Dr. Sokol: The younger people we’ve asked said they don’t know if they want medical information on their cell phone. It’ s not password protected, and they have it in public a lot.

Mr. Ryan: Are the three of you familiar with a process whereby hospitals monitor A1c or other things proactively in the post-discharge period?

Dr. Mannasseh: I don’t know if much is being done specifically for diabetes. In terms of the usual subjects, we get information back on surgical complications at BU.

Dr. Dungan: We do use our information warehouse to get back information of that nature. We are in the process of implementing programs to follow patients with the worst glycemic control, but it’s been very difficult for us using phone-based follow-up to even get patients on the phone.

Q: I know multiple groups are working with combining cell phones and glucose monitoring tests. Do you think that would be useful?

Dr. Dungan: Yeah, I would think so. We’re still in the start of actually implementing our information programs. I don’t know – at least in our population we tend to have a lot of low-income patients, though I suppose a lot of them have cell phones.

Dr. Sokol: Dr. Mannasseh, you said your discharge evaluation saved how much – $412 per patient? I’m in the school of thought that says just buy patients a computer and hook it up to the system. We give them other types of technology; we prescribe glucometers and so forth. I think that’s one of the best investments we could make. I don’t see why not to do it.

Dr. Mannasseh: We try to use the telephone for follow-up, but we had to cancel our pilot study because compliance was less than 50%. It’s close to 80% in our current pilot study for follow-up with the Internet, though. At least in our inner-city population, people seem to prefer the Internet to the telephone for medical follow-up.

Dr. Sokol: Several big cities have done studies on computer use. Chicago has found that although more people have cell phones, the trend is that more people are using computers to gather and exchange health information.

Q: I was hoping each of you could comment on the need for sharing of a broad set of patient biometric data. Dr. Sokol, to your point about buying patients a computer: there are a number of startup companies that will send patients home with devices that transmit data like pulse oximetry and body weight to the case manager. I was surprised not to hear this as a theme in your talks that you want to see how patients are doing especially in the first day, the first week. Is this a need you recognize, and is the healthcare system ready to digest this data if it were easy to get back to?

Dr. Sokol: That’s a big issue of informed consent. How does the patient understand how the data will be collected, used, sheltered, and whether it will be used in some other capacity? Let’s suppose that’s all wonderful and good. If the patient consents, I would be in favor of limited use. There are so many things that play into that; we’re almost cavalier about the hurdles now. The most important thing to me is not only transmitting out, but timely feedback and use: there has to be really tight control in the first place and then again in its use.

Dr. Quinn: We’re involved in lots of studies, and we’re finding that it’s not only sending results. What people do want is an interpretation of the data, what does it mean. To me as a healthcare provider, I think that there’s going to be an evolution of information management with regard to managing diabetes and other diseases. I just came from the mHealth Summit and the technology is moving very quickly. A recent Harvard Business Review article says this will completely change how we do science. It’s not just patients or providers, but systems to analyze.

Dr. Dungan: I think having a way to send glucose logs electronically does remove one big barrier. With a lot of these patients, it’s not just a set of numbers, but also a way for patients to implement other things like insulin and blood pressure. If we have an insulin pump patient we’re downloading, they can’t lie about when they’ve overridden the pump. A more integrated way to deliver information would be very useful, I think. In my clinical practice, the major limitation is getting glucose logs. Also, it’s not reimbursed. I have the luxury of doing that, but a busy provider won’t be able to unless it’s reimbursed.

Dr. Nathanael Paul (Oak Ridge National Laboratory, Oak Ridge, TN) : I’m less concerned about privacy and more concerned about intentional or unintentional changes in information leading to a change in treatment that could induce hypoglycemia or hyperglycemia. We have recent examples of malware affecting heart machines. Have you thought about these issues, and if so how are you addressing them?

Dr. Sokol: I haven’t heard that as the number-one problem. I think we’re most concerned with having structure and pay to match the workflow. There’s also the time issue and the valuing issue – promoting interpersonal relationships. The technology has to have ways to deal with the problems you’re talking about: if an odd number is put in, it questions that. That may generate an old-fashioned phone call or office visit.

Mr. Ryan: I have a question about the structure of the healthcare system. Assuming we can put the tools in place to monitor patients post-discharge, who will be responsible for that? Dr. Dungan, who will get the data and evaluate following congestive heart failure discharge?

Dr. Dungan: This often gets relegated to the primary care physician or the subspecialist taking care of the specific conditions. CHF and diabetes may be tightly integrated, and it would be nice to have a follow-up model that integrates these comorbidities together. It seems inefficient to have to send weight to the cardiologist, blood glucose to the endocrinologist, and whatever other information to the primary care physician.

Dr. Mannasseh: Both sides have to know who’s responsible. Patients have to know who’s taking care of them, and so do providers. In our situation, it’s the primary care physician. After that, we have to make sure that ongoing real-time communication is happening. Within 24 hours of admission, we have the coordinator communicate with a triage nurse identified in each health center. A registered nurse puts the patient in the chart so that if the information needs to be given to the PCP, someone is responsible even if the PCP isn’t there. The same thing happens at the time of discharge. I think you need a tightly controlled system to make sure this all happens.

TELEMONITORING FOR RECENTLY DISCHARGED PATIENTS

Jim Callaghan, MT (ASCP) (FDA, Silver Spring, Maryland)

Mr. Callaghan reviewed the FDA’s general classes of medical device approval and also broadly addressed several areas of concern for mobile health management of diabetes. He said that the FDA is excited about the future but that the coming years will offer some challenges, and the question and answer session gave him a chance to elaborate on the Agency’s arguably ambivalent perspective.

• Mr. Callaghan highlighted several areas of FDA concern around telemedicine including: cybersecurity and data integrity (although he noted that information issues are enforced by HIPAA rather than the FDA), lay use, rural connection, bandwidth required, information that drives therapeutic decisions (a question of particular importance, he noted), and critical use (another key topic). Mr. Callaghan also mentioned specific segments of the regulatory process, including human factor studies and software overviews.

Karen Gilberg, MD, FRCP(C) (Robert Bosch Healthcare, Inc., Palo Alto, CA)

Dr. Gilberg began her presentation by noting that “telemonitoring” is already an outdated term, since it implies a passive role for the patient. She outlined a more dynamic, plausible model for mobile health that she referred to as Proactive Integrated Care (PIC), whereby patients and providers can interact remotely to manage a variety of health conditions. She noted that although studies have shown PIC to be effective for the management of congestive heart failure (CHF) and chronic obstructive pulmonary disease (COPD), studies of PIC for diabetes have thus far been inconclusive. She attributed these unclear results to the high variability of patients with diabetes, and she recommended that study designers examine PIC for specific subpopulations with diabetes. Dr. Gilberg closed by looking forward to a new stage of health management, when hospitals’ business models involve keeping patients out of the hospital and when “home” care can occur while a patient is sitting in a Starbucks Wi-Fi zone.

PANEL DISCUSSION

Questions and Answers

Q: So on my BlackBerry I can get any number in the world except for those related to healthcare information. At what point does a number become just a number and cease to make something a medical device? Over the counter I can print out my son’s glucose data on an HP printer. Why isn’t that regulated?

Dr. Callaghan: We expect that any company producing a device to support a medical device is being produced at a level sufficiently rigid to support that use. If you are going to use devices that will recommend insulin dosages, we will audit those devices to see that they are valid to manage that information.

Dr. Quinn: It sounds like you’re asking about just having that information, for example to monitor a family member, not that there would be a medical decision based on it.

Q: Something like that. Maybe it’s incorrect to say that a number is just a number. But in this world where I can get the weather forecast from Timbuktu, why isn’t health information just available to me once it’s in the cloud?

Dr. Callaghan: There is a cellular-phone-integrated glucose meter (HealthPia’s GlucoPhone) that allows the transfer of data. If you have the resources to subscribe to that service, you can do so.

Q: At what point does software become a medical device? Do you regulate the software- only pieces on a device like an iPhone?

Dr. Callaghan: It depends. If the software can download directly from a glucose meter, we consider it an accessory. However, if you manually enter the data yourself – it’s not necessarily that we don’t consider it a medical device, but we have not been reviewing those.

Q: I think someone makes an application that talks to OneTouch Ultra meters and then uploads the data to a Google Android phone (https://sites.google.com/site/glucosemeterandroid/).

Dr. Callaghan: If something is integrated into a device, we ask the company to submit a 510(k). We have limited resources, but if you have information on these matters, you can let us know.

Q: I’m not very knowledgeable concerning the issue of when data becomes personal property. If I have a medical device with my personal property on it, isn’t that just a transfer of personal property independent of the device? I don’t understand the junction between what you as the FDA do and what a patient does with HIPAA-protected information.

Dr. Callaghan: I’m not very familiar with HIPAA regulation, but from our perspective it depends on the claims of applications. If they say you can email clinicians with the results of diabetes, then we’re not going to regulate that. But if you say you can download results from a particular meter and then email them, we have concerns over that. It depends on where the claims start and where they end.

Q: Maybe the field is so new that we haven’t established well when that information is the patient’s.

Dr. Callaghan: It’s the patient’s data, no question. But it’s still a medical device when it’s claimed to link to other medical devices. The data is still the patient’s property. We are concerned with ensuring privacy and security, and we ask companies to address these concerns and review them in a clear and appropriate process.

Q: You mentioned a platform that manages comorbidities. What class of device are these?

Dr. Gilberg: They are Class 2 devices.

Q: What are some examples of devices that manage diabetes and other conditions?

Dr. Gilberg: I know at Robert Bosch, the Health Buddy system does that. There’s also the ViTel Net’s Turtle 400, which manages multiple devices at the same time. GE and Intel have a device that I think manages diabetes and comorbidities, as well.

Dr. Quinn: They do, I just saw it at the mHealth Conference. It’s a pretty large box, and Intel and GE are investing in it heavily. They are very much going after comorbid conditions, treating multiple diseases at once and not just diabetes.

Q: And why is it that no one calls these Class 3 devices if there are decisions being made based on them?

Dr. Gilberg: I know the box itself is Class 2, and the software – maybe you can explain?

Dr. Callaghan: The software guidance is five years old, but software is part of medical device. If you have physician intervention, the physician will drive whatever decisions are being made based on the device. So that would lower the risk associated. If the device were overriding the physician’s decision, that would be different.

Dr. Gilberg: And none of these devices do that. They ask questions about vital signs, blood glucose readings, and medication use, and number of other questions that can be put in there. For patients with chronic heart disease and COPD (Chronic Obstructive Pulmonary Disease), they will ask if the patients have been feeling short of breath. If a device uses branching logic, it will ask follow-up questions. The information is then transmitted to the healthcare provider who makes decisions.

Q: I have a question about the machine-to-machine space. Has the FDA looked to standards from cybersecurity firms or other domains connected to telehealth?

Dr. Callaghan: We’re working on a consensus standard. That doesn’t mean you can’t submit along with a standard that you think is appropriate for your device, although we would want to know more information about that standard than would a recognized standard.

 

FDA/NIH Public Workshop – Innovations in Technology for the Treatment of Diabetes: Clinical Development of the Artificial Pancreas (an Autonomous System)

On November 10, 2010 in a nondescript Hilton Ballroom in Gaithersburg, MD, the FDA and NIH assembled some of the most well-known and prominent diabetes clinicians and researchers to discuss the development of the artificial pancreas (AP). The day truly embodied the definition of the word “workshop,” as a wide variety of perspectives were shared, lively debate occurred between clinicians and the FDA, and audience participation from children with diabetes added a real patient flavor to the discussion.

Clinicians were overwhelmingly in favor of low glucose suspend and artificial pancreas systems and their frustration was palpable in the context of an increasingly restrictive regulatory environment. The FDA, on the other hand, placed tremendous emphasis on safety. Dr. Patricia Beaston, who frequently reiterated the inaccuracy and limitations in current continuous glucose monitoring technology, spearheaded many of these safety concerns. However, such points were strongly disputed by renowned Drs. William Tamborlane (Yale), “You’re absolutely wrong!”; Dr. Bruce Buckingham (Stanford), “When would you consider the system safe to the point where we don’t need to be there 24/7? I think it’s safe!”; and Dr. Aaron Kowalski (JDRF), “In the near term, we can use the information from CGM to make small but clinically significant changes in insulin dosing.” What is amazing from a patient view is that the FDA doesn’t seem to view insulin as a particularly dangerous drug or diabetes management as a particularly complex or complicated process that can be aided by technology that is very helpful, if itself imperfect.

A common theme throughout the day was the importance of the stepwise approach to the development and approval of the artificial pancreas. Dr. Aaron Kowalski and the JDRF’s roadmap to an AP was cited early and often throughout the day and has become an industry standard of sorts when thinking about closed-loop development. Likewise, speakers emphasized that research should progress in a step-wise fashion, with simpler inpatient studies first and movement towards more complex outpatient studies once sufficient data is gathered.

The JDRF Clinical Recommendations Panel on Closed-Loop Systems also released its recommendations for the advancement of artificial pancreas systems outside the hospital setting. The distinguished panel of leading diabetes clinicians and researchers laid out the most important considerations as closed-loop studies move to the “real world,” and the hope in everyone’s mind is that the FDA will take these to heart. Next steps for Artificial Pancreas clinical testing can be found on the JDRF’s website at http://www.jdrf.org.

We truly appreciated the women and men who courageously approached the microphone and documented their struggles with diabetes and desire for better therapies. We were also encouraged by Dr. Charles Zimliki’s (FDA/CDRH) closing remarks, where he confidently stated that an AP system within the next four-to-five years is “definitely possible” although we also remember two other meetings co-organized by the NIH/FDA/JDRF, in December, 2005, and July 2008, where the field was pushed to go faster and where various ambitious goals that have not been achieved were put forward. Given the pace of innovation at the moment, we certainly hope that regulatory bottlenecks in particular can be addressed.

Clinical Expectations for Low Glucose Suspend Device Systems

FDA: LOW GLUCOSE SUSPEND (LGS) DEVICE SYSTEMS: GOALS AND CONSIDERATIONS

Patricia Beaston, MD, PhD (FDA/CDRH, Washington, DC)

To round out the early morning sessions, Dr. Patricia Beaston (FDA/CDRH, Washington, DC) laid out many of the FDA’s concerns with low glucose suspend systems such as the Medtronic Paradigm Veo. In evaluating these systems, Dr. Beaston made it clear that the FDA has reservations about the accuracy of CGM. She noted that in the hypoglycemic range, continuous glucose monitors have high false alert rates of 60%. As a result, an inappropriate shut-off could result from an inaccurate CGM reading, eventually causing a very high blood sugar. Dr. Beaston argued that low glucose suspend systems with predictive capabilities offer distinct advantages over low glucose suspend systems that use thresholds because predictive systems use the most successful components of CGM: tracking and trending. To close, Dr. Beaston highlighted the significant scientific challenges associated with approving low glucose suspend systems, paying particular attention to the need to externally validate the system. In her words, “you can’t use the system to evaluate if the system works.” According to Dr. Beaston, more consensus will be needed in the future on what glucose level constitutes hypoglycemia. Overall, we found this talk relatively negative on the potential near-term approval of low glucose suspend systems, a fact that is even more frustrating considering the Veo is already approved in Europe. From our view, the FDA appears to be fine letting patients fail on their own, as long as it is not their fault. The fact that patients are dying should be motivating the agency in our view, but this does not appear to be the case.

Table Discussion: Clinical Expectations for the Low Glucose Suspend Device Systems

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

Dr. Klonoff gave the first presentation on the clinical aspects of low-glucose suspend (LGS) systems, highlighting their purported benefits and possible risks. On the plus side, LGS systems ought to cause fewer nocturnal hypoglycemic events and consequently fewer acute CV events and less rebound hyperglycemia (and thus, possibly, lower A1c and lower glycemic variability). However, he noted that inappropriate suspension might cause rebound hyperglycemia, contributing to ketonemia and possibly higher A1c. He also warned that even proper suspension might not provide protection from hypoglycemic episodes. These questions translate into potential trade-offs of short-term and long-term benefits – is it more important to reduce hypoglycemia and address short-term CV risk, or focus on hyperglycemia and long-term complications? Dr. Klonoff closed his presentation with data from research by Dr. Choudhary on the benefits of approving LGS and Dr. Tannenberg on the risks of keeping the technology unavailable.

• Dr. Klonoff drew attention to unpublished LGS data from Choudhary that will be formally presented later in the week at the Diabetes Technology Meeting. The X54 Veo study included 31 patients at six sites in the UK. For the first two weeks of the study, patients used sensor-augmented pump therapy without LGS; for the next four weeks, LGS was active. 166 suspension events occurred over 812 total patient-days of LGS, with 20 episodes in which insulin was cutoff for the LGS system’s full duration of two hours. Of these 20 episodes, the mean blood glucose at time of suspension was 88 mg/dl and the mean glucose after two hours of suspension was 160 mg/dl – not a significant amount of rebound hyperglycemia in Dr. Klonoff’s view. Ketones were not tested, but no DKA was reported. No meaningful A1c changes were possible given the study’s six-week timeframe, but data on mean glucose data are under review. Hypoglycemia was significantly reduced from 1.7 hours without LGS to 0.6 hours with LGS (p=0.03), and quality of life improved with LGS as well. Although he said it would be nice to have more data, Dr. Klonoff was optimistic about the benefits of LGS based on this study. We are surprised the levels of blood glucose were as high as they were at the time of suspension and we wonder if an outlier lifted the average, since it would seem more likely that suspensions would be made during hypoglycemia.
• Dr. Klonoff closed with a recently published case study in which a patient was found dead due to “stacking” boluses (Tannenberg, Endocrine Practice 2010). Dr. Klonoff suggested that this death might have been averted with a safe and effective LGS system, and heemphasized that the risks and benefits of not approving the technology must be considered as well as the risks and benefits of approving it.

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

In an impassioned presentation, Dr. Bruce Buckingham outlined the safety, effectiveness, and critical need for an approved low glucose suspend device in the United States. In Dr. Buckingham’s opinion, the substantial benefits of avoiding a hypoglycemic seizure far outweigh the potential hyperglycemia resulting from a two-hour pump suspension – in his experience, the hyperglycemia is usually an inconsequential 160 mg/dl. He made it clear that nocturnal hypoglycemia is still a significant problem and patients don’t wake up to alarms, highlighting the need for a low glucose suspend device. Additionally, Dr. Buckingham demonstrated that the risks of hyperglycemia, ketonemia, and increases in A1c are minimal or nonexistent with low glucose suspend devices. He argued that such a system should not be tested in the inpatient environment because it is so basic and the outpatient trials would be safe and easy. Instead, Dr. Buckingham proposed testing a low glucose suspend system in a three- month randomized outpatient trial, with hypoglycemia detected via CGM, hyperglycemia evaluated via A1c, and ketones measured every morning. He felt that an A1c increase of 0.5% would be undesirable with such a system. In a proposed study of predictive alarms, subjects would be randomized on a nightly basis, so a 3 month A1c measurement could not be used, and for this study Dr. Buckingham urged the use of mean overnight glucose as a surrogate outcome measure for an A1c result. We were struck by Dr. Buckingham’s persuasive arguments, use of logic and previous research, and passion for getting a low glucose suspend device approved. His presentation renewed our sense of hope that something as simple as the Paradigm Veo might actually be approved in the current regulatory environment.

William Tamborlane, MD (Yale University, New Haven, CT)

Dr. Tamborlane gave an argument for LGS systems in light of what he sees as their greatest clinical benefit: rescuing people who go unconscious from nocturnal hypoglycemia. He reviewed the response to hypoglycemia in people without diabetes, which includes a stop on endogenous insulin secretion at 80 mg/dl (“sound familiar?” he asked) and the release of glucagon and epinephrine at 70 mg/dl. Unfortunately, people with type 1 diabetes do not secrete glucagon in response to low blood sugar. Although people with type 1 diabetes are usually still partially protected by the epinephrine response, a study by Dr. Tim Jones suggests that the epinephrine response is suppressed in sleeping adolescents – apparently one of the reasons pediatric patients are notably prone to nocturnal lows. Dr. Tamborlane also reviewed a study by Dr. Bruce Buckingham to show that nocturnal hypoglycemic seizures are tend to be preceded by long (and thus, in theory addressable) periods of blood glucose, and he presented data from his own group suggesting that two hours of insulin suspension raises blood glucose by roughly 36 mg/dl and has negligible effects on ketonemia. He told the audience that a product like the Medtronic Veo should be considered as an entire system that combines a sensor-augmented pump (SAP) with low- glucose suspend. On this note, Dr. Tamborlane said that any system incorporating a more consistent sensor will reduce frustration, increase use, and improve outcomes, and he suggested that feedback in Veo-approving countries has been positive on the monitoring system’s user-friendliness. It is a very big positive in our view that the highly respected Drs. Tamborlane and Buckingham argued so persuasively in favor of the low-glucose-suspect Veo system.

Francine Kaufman, MD (Chief Medical Officer Medtronic Diabetes; Children’s Hospital Los Angeles, Los Angeles, CA)

After giving an overview of the Medtronic Paradigm Veo, Dr. Kaufman reviewed three-month data from the Interpret study. The study is being conducted in Europe in 44 subjects using the Paradigm Veo system with the low glucose suspend feature turned on. Dr. Kaufman emphasized that those using the low glucose suspend exhibited a slight decrease in A1c (baseline: 8.4% vs. 6 months: 8.2%, not statistically significant), debunking the assumption that low glucose suspend systems will lead to an increase in A1c. Along the same lines, impressive data from a study of CareLink in 935 patients showed that with the low glucose suspend feature of the Veo turned on, a reduction in hypoglycemia was achieved with no evidence of an increase in hyperglycemia. Moreover, data mining showed that most suspensions occur during the day (many before lunch) and the vast majority are 0-5 minutes (10% of suspends are two hours long). According to Dr. Kaufman, the data suggest that the low glucose suspend feature is typically being activated when the patient is aware of the alarm and can react to it. Moreover, the fact that no increase in A1c or hyperglycemia has been observed with the Veo system makes it clear that the system is safe and effective for use. We would ask the additional question what the risks are of not having such systems; it is clear from our view that hypoglycemia is an accepted fact of life for most patients and we would like the FDA to consider more of the safety problems associated with insulin.

PUBLIC COMMENTS

We are not publishing the full patient names associated with the public comments because we did not ask permission from the patients, except in cases where the full name is noted.

Katherine U.: My glucose is relatively well managed; my A1c is in the low sixes. But the risk of death by hypoglycemia is always there. Please support these technologies, because first, people can always be doing more to manage their diabetes better, and secondly, for peace of mind – now, there’s always that risk. I am an adult with type 1 diabetes, and I live alone. LGS devices are very exciting. I urge the FDA to adopt reasonable standards and encourage companies to keep developing these technologies.

Charlotte L.: Thank you for letting me speak to you about the importance of the artificial pancreas to kids like me. I’m nine years old, and I was diagnosed with diabetes when I was six-and-a-half. It’s hard to remember eating when I was hungry and drinking when I was thirsty. You probably already know I prick my finger every time I eat and many other times each day. I have to call my parents each time I eat when I’m on a play date. I never ever get a break. Testing blood glucose is extra hard when you’re a kid, because sometimes other kids see it and say it’s gross, and it hurts my feelings. I go low almost every day, sometimes down to 30 mg/dl before I realize it. If a grownup goes to 30 mg/dl, they’re in big trouble. My parents, my teachers, and the school nurse are very worried about my lows. I’m supposed to be careful because I could pass out or have a seizure. I don’t know what that would be like, and I don’t want to. The artificial pancreas could do a lot to prevent this. My doctor has 700 patients with diabetes and says mine may be the best managed. My mom is a healthcare professional, and my dad is a partner in a law firm. My A1c has been in the fives ever since I got an OmniPod a couple years ago. Can you imagine how hard it is for kids without my resources? My mom says if she were speaking, she’d get on her knees and ask you to move the artificial pancreas forward. I’m definitely not comfortable getting on my knees, but I want the artificial pancreas as much as she does, and maybe more. I will be the first volunteer to test it out, and my parents can double-check everything.

There’s a song by Nick Jonas called “A Little Bit Longer” that goes, “You don’t know what you got ‘til it’s gone, / And you don’t know what it’s like to feel so low.” If more people knew what it was like to live with diabetes, I don’t think we’d be having this meeting. I told my parents I was too young to get diabetes when I did, but some kids get it when they’re one or two. I asked my parents to make my diabetes go away. You can do what they can’t. I’m almost never scared anymore, and I want to work with you to get this technology to those who need it. Thank you very much for listening to me.

Caitlin R.: My name is Caitlin and I am 10 years old. I have had type 1 diabetes for five years. My Dad and my brother have it too, and so does my cat. It’s not easy. My mom wakes me up with a blood glucose check, and then we decide what to have for breakfast. At 10:30 am I meet with the school nurse I check, and then again at lunchtime. (I eat a packed lunch; my mom writes down the carb count.) Then I check after school. If I have ballet or soccer practice, I check before and after, and the same at dinner and bedtime. Mom checks me before she goes to bed. I’ve probably checked my glucose 13,000 times in the last five years, and used even more test strips trying to get it right. Even with a pump you don’t get a break. I know this isn’t a cure, but it could lead to a more normal life and I wouldn’t have to worry every minute and every day. In six years I want a driver’s license, in eight years I want to go to college, and I don’t want diabetes to get in the way. Please do everything you can to speed up the artificial pancreas.

Comment: A rough analogy that might be useful is to a pressure vessel in an industrial situation. The rules now are very clear: you cannot just put more pressure sensors, you must have some secondary safety system. Usually this means rupture valves: always something stops the vessel from blowing up. From the perspective of the closed loop, work with Tim Jones has shown there are features in the nighttime that can be used to shut down the pump.

Michel B.: I have a comment on redundant sensors in the artificial pancreas. Redundant sensors should be more effective in the low range compared to conventional CGM, someone said. And they could also be of two different technologies. Two different CGMs measuring the same thing should lead to the same result. A practical way to have redundancy would be, when there’s a discrepancy, to have user intervention. This would be a great way of improving the technology.

With regard to the low-glucose suspend and alarm, we’ve seen in Dr. Buckingham’s presentation that many times users don’t wake up at night. There’s an issue of having the alarm transmit to another person using cell phone; I would like to know the view of FDA on this, since cell phones are not FDA approved. Is there a path to get an alarm sent to a cell phone and having that approved?

QUESTIONS AND ANSWERS

Moderator: Jeffrey Joseph, DO (Thomas Jefferson University, Philadelphia, PA) Dr. Joseph: Now we’ll have a discussion of the panel members based in part on the questions posed earlier. I’m hearing that the data people are looking for is more outpatient, ambulatory. What is the real risk to the patient in such a study, and is there something you’re comfortable doing with low risk to the patient?

Dr. Buckingham: The length that the pump would be off is two hours, which is not enough to cause dangerous hyperglycemia in the short term. The only way to assess the risk for long-term hyperglycemia is an outpatient study; you can’t do that in the CRC. If we need to know how many nights someone is high, we can’t go with an inpatient trial. In terms of the algorithms, when someone goes past the nadir of blood glucose we know the pump will turn back on. We have the technology worked out, and now we need to test in the home environment where it will be used. The risk of hypoglycemia is huge, but the risk of transient hyperglycemia is very low.

Dr. Tamborlane: I’d like to clarify the issues on CGM as an outcome measure. We know that the relative absolute difference for the current sensors is 14% to 15%, but the actual difference is effectively 0% because the highs offset the lows. There’s no bias among the three current devices. You end up with the

same mean but higher variability, which is why you need a larger sample size. Dr. [Roy] Beck and I had a commentary in The Lancet because this is what we looked at in the JDRF study. I think if you talk to statisticians … [they’ll say] if you have blinded controls, there’s no reason you can’t compare the CGM values to a control group.

Dr. Klonoff: The main advantage of an inpatient study would be to teach patients, but there isn’t much to teach in this kind of trial, so you could do it in an outpatient setting. You need to teach them how to check ketones in the morning, but not much else.

Q: There is a clear sense that the concept of a low glucose suspend system is a good idea. A no-brainer. Yet, the people at the FDA are concerned that the device will shut off at times when it shouldn’t shut off. The low glucose suspend is great when it’s correct (i.e., the patient’s blood sugar is actually low), but what about times when it isn’t correct (i.e., the patient’s blood sugar is in target range or high)?

Dr. Beaston: As I stated before, you can’t use a system to measure its own efficacy. The CGM and glucose meter are part of the low glucose suspend system and have inherent errors. The fact that they agree with themselves makes it hard to make determinations about the accuracy. We really need to evaluate the system with an outside truth. The accuracy of the sensor varies widely between individual patients. It is dependent on how often the calibrations are done, the quality of the glucose meter, and the type of CGM used. Each CGM has its good and bad points. And we have talked to the statisticians. You can’t average out the error and say that you have an effect. You have to have an outside truth by which to measure the benefits of the system. Don’t get me wrong, we’re not saying that the outpatient studies can’t be done. But we need to think about how to do them correctly. For instance, maybe during the study you have a time during the sensor use period to go in, get a measurement, and use that information to really verify where the CGM was at the time it made a determination. It’s a very complex issue. We would welcome people to come in and bring us models and information. We know that CGMs are very good for tracking and trending. They’re good at showing ups and downs, but you don’t always know where the blood sugar is at a specific point in time. The other thing is the discussion of using A1c. At this point, we haven’t seen data or a plan that shows us how this can be used as a validation tool.

Dr. Tamborlane: Show me studies that found the three currently available CGMs read significantly different values than reference glucose. I’ve never seen such a study.

Dr. Beaston: There is huge variability in YSI and CGM pairs. If you take the means and average it out, it averages out. But you cannot do this. We’ve talked to the statisticians and this aggregation just isn’t correct.

Dr. Tamborlane: You’re getting the wrong statistical advice. We’ve given this a lot of thought as well. If you want to look at hypoglycemia exposure, the only way to get a quality measurement is CGM. You’re absolutely wrong.

Dr. Buckingham: You say you need an external validation of the system. But the system is looking at individual events. Additionally, you can compare nights when the low glucose suspend system is turned on versus nights when it is turned off.

Dr. Beaston: If you have a change of 5 mg/dl, you don’t know if it’s because of the calibration or the accuracy of the CGM. The CGM may be tracking at a value that shows the patient is hypoglycemic, but the YSI value will show that the patient is not in a state of hypoglycemia. We’re talking apples and oranges about a number of different things. The low glucose suspend system is reactive and has lots of challenges because of the accuracy at a single point of time. At 70mg/dl, it’s a 50/50 shot that you have a true event. If you do a predictive low glucose suspend, it’s an entirely different story—you’re using the CGM at its best capabilities. When it’s predictive, I may be at 110 mg/dl with a downward trajectory. In this case, there’s an opportunity to intervene and you have more of an opportunity for success. When we talk about these systems, we have to be very careful about which system we are talking about.

Dr. Buckingham: To overcome the statistical problems you mentioned, you just need a larger sample with either more study days or more patients. These are people that get low for hours and we need to protect them. If you assess the number of events on the nights the hypoglycemia prevention is activated compared to the nights it is inactive, you can use CGM to assess the number of nights with events. Since the CGM is less accurate than a reference glucose value, you do need to increase the number of nights in the study with either a larger number of people or over a larger number of days.

Dr. Tamborlane: I don’t want to mix up what we’re talking about. We started with accuracy of CGM as an outcome measure. I absolutely agree with you that we want to do predictive suspend based on a predicted low. The Veo was considered the least possible, the smallest setting, to rescue from death or seizure. But the predictive alarm is fabulous, I think we’re very strong with that. It’s good for sensor fatigue. When we reanalyzed closed-loop studies based on predictive lows in Diabetes Technology and Therapeutics, we saw dramatic reductions in hypoglycemia.

Dr. Irony: I think I’m in a middle ground position between Patricia and Dr. Tamborlane: looking at indications of CGM tracking and trending and not acting on data from that, but instead using the glucometer as the means to take action. If your blood sugar is 300 mg/dl take a bolus, and if it’s 40 mg/dl take food or glucagon immediately. The glucometer has a mandate and an indication to take action. There is no reason we cannot use CGM in outpatients with the recommendation that the alarm should be confirmed with a fingerstick.

Dr. Buckingham: The problem is that people don’t wake up to do a fingerstick. People wake up to the first alarms only 11% of the time. If the alarm persists, often for hours, you can get 30-40% to respond to alarms. This is therefore not a good outcome measure. You have to use CGM at night.

Dr. Beaston: Lots of the Veo data is 24-hour data, when LGS is used throughout the day. We have to be careful when talking about our expectations. What are the studies and expectations at night? Are those different than if we’re developing a system that worked throughout the day? Everyone’s presented a slightly different version and I think it’s important that we be clear when we talk about the hopes for each approach.

Dr. Kaufman: The Veo data we can segment by night and day. The use is increasing around world except in the US. We’ve seen no deterioration of A1c, and we would be happy to do an outpatient trial to validate this. So you get the same mean with apparently less glycemic variability and hypoglycemia. If the sensor is 100% on all the time this would be different study. The Veo algorithm is more accurate in the low range than the soft sensor Medtronic has now. The other issue is that if it’s not 70 mg/dl, what’s the range? Is it 72 or 75 mg/dl? So that even if it’s triggering above the intended threshold, there’s not a seemingly significant difference. Most of the data that’s off is minimally off, still in a glucose range where patients don’t want to be. To stop insulin for 2 hours seems reasonable in this range. We’ll go forward with the studies as required.

Dr. Beaston: There’s information about the Veo that hasn’t been discussed. Do you want us to talk about our concerns with regard to the data you submitted?

Dr. Kaufman: I think we don’t want to get into the specifics of the study. We’re going forward as has been decided and developed. We’re willing to do what we need to make the Veo available in the US.

Dr. Beaston: The Veo is a SAP/LGS system. I’m not going to talk about data that we have but that you haven’t presented. But the majority of Veo data has little to do with the capability of LGS, but involves patients who rely on alarms. Most patients deactivated the pump or did something else, and we’re not sure why – they did a fingerstick and it didn’t confirm the sensor, or they treated in some other way. They chose not to suspend. And it wasn’t an issue of an alarm and the patient not responding, but the patient making improvements based on alarms. So you have to be very clear on the effect of system due simply to sensor-augmented pump use. If you take a patient from MDI and take them to SAP, the literature shows they will do much better. The question is, if they have SAP and this is how they are doing, what are we gaining and what are we possibly losing by adding LGS? We have to be clear with the Veo. Is the benefit from alarms that patients respond to, or is there an additional benefit for patients who don’t respond? I believe there’s a patient population out there that really can benefit from nocturnal LGS, but I think it’s important to understand who those patients are so we can provide the best information to those patients, their physicians, and their families on what these systems can an cannot do, so they understand the difficulties and challenges. Fingersticks are a burden, but none of the technology to date has shown that it can replace the need for fingersticks during the day. For one, sensors continue to need calibration. Also, there is a need to do confirmation of alarms. We have great hopes for the systems, but we also have to be realistic and responsible in describing them to people so we don’t create false expectations.

Dr. Kenneth Ward (Oregon Health and Science University, Portland, OR): It’s really frustrating to not have an outside reference to really comprehend the accuracy of sensor data. I understand where Dr. Beaston is coming from. There’s no question that the pump will shut off inappropriately at times. But Drs. Tamborlane and Buckingham’s studies show that A1c and ketones are not adversely affected. Even though the pump will definitely shut off inappropriately at times, we’ll be better off than where we are right now. Big picture, I see a definite benefit if the studies can confirm no adverse effects from a two- hour low glucose suspend.

Dr. Tamborlane: I want to address the inappropriate shutoff. We’re talking about crossing a threshold, which sets off an alarm. Let’s say the threshold is set at 70mg/dl. If error is 40%, that’s 28 mg/dl, so the system would shut off glucose at 98mg/dl. Who cares! That’s not an inappropriate shutoff. You don’t want to over-exaggerate the false negative.

Dr. Ward: Even if it’s 120 mg/dl or 130 mg/dl, Dr. Buckingham showed data that if your pump is suspended for 2 hrs, there are no ill effects. Inappropriate shutoff won’t be a problem.

Dr. Beaston: But the issue is that CGMs are telling the pump to turn off when the true plasma glucose is in excess of 180 mg/dl. Our data suggest that this is not a rare event, but is occurring a significant number of times. Even if you disable the pump for two hours and you turn it back on, going back to the basal rate is not sufficient to prevent further loss of glycemic control. I’m not saying that people are going to go into DKA, but we have to have a real understanding of what the system can do and how well it performs.

Comment: I’d like to address the use of CGM as an outcome measure. We’re in the middle of a 1,000 patient study examining tight glycemic control versus standard of care treatment. We concluded that we couldn’t use CGM as an outcome measure. In our study, we found that the statistical inaccuracy of the CGM is heavily influenced by the calibration. This has made me realize that we really need to be careful with the calibration of CGM with SMBG devices! Despite this problem, however, I think we can use CGM as an outcome measure.

Q (University of Alberta Hospital, Alberta, Canada): There seems to be a schism between the clinicians and the FDA. The way to reconcile is the science of medical decision-making. Physicians are aware of the risks – the risk of death vs. minor biochemical changes. In a patient with a higher prevalence of hypoglycemia, the monitoring and the LGS will work very beneficially. Maybe the clinicians could make this adjustment by prescription only and then the FDA can wonder how best to do this experiment.

Dr. Irony: I think there’s always a balance the FDA tries to strike between how much information we think physicians and patients need to make decisions versus just a free market. Depending on the patient population, we already do this. If someone is very brittle and hypoglycemia unaware, our target A1c is already higher. I think that the unlikely event of DKA is well balanced by the risk of hypoglycemic events that could lead to death or seizure. But the long-term risk balancing is something we already do in medical decision-making, for example by backing off on intense insulin therapy for someone who’s prone to hypoglycemia. I think LGS will add another step to medical decision-making.

Dr. David Rodbard (Biomedical Informatics Consultants, Washington, DC): We’re beating around different statistical questions here. We simple need to know the precision and accuracy of CGM, of SMBG in all its flavors, of YSI, the effect of the calibration, and what is the effect of the time lag. My suggestion is that someone convene a panel of statisticians, clinicians, FDA representatives, and academics and come to a consensus. I would suggest that this could be solved in a day if it’s not solved here today.

Q (Germany): I missed the change in insulinemia if you switch off the pump for 30-60 minutes. I would expect that the difference in insulinemia is not that pronounced. Could you comment on this?

Dr. Buckingham: We’ve studied the time it takes for glucose to recover following pump suspension for hypoglycemia, and it takes 75 minutes to come back to a positive rate of change.. I think that we are giving much smaller doses of insulin at night as basal insulin, which results in a lot less insulin on board.

Q: But I imagine you would reduce fear if you report the insulin data.

Dr. Tamborlane: It was actually the top panel on my glucose suspend slide. I don’t remember the number off-hand, but I think it was a 25-30% reduction over the two-hour period of suspension.

Q: The FDA should ask for this data in studies.

Comment: I think there has been negligence in the alert system for CGM. These sensors could alert other things in the home. You could do a host of different things that could increase safety. If the issue is waking people up, we can definitely do that. The real issue though is the inaccuracy of the sensors. Discussing the alarms may not be the best discussion right now. Nevertheless, I was interested in your thoughts on an improved alert system.

Dr. Beaston: Could the system call someone else if the person does not respond to a low glucose alarm? It’s certainly within the realm of reasonability that that could happen. It would require discussion with a software and device group. But we’re very careful about linking devices. You don’t want a software glitch that causes the system to fail. I had a problem with my car and phone because of a software issue that took days to resolve. We are very cautious when people want to start linking more and more technology. We need to understand what the linking means and ensure that we’re not interfering with the functionality of the system.

Q: I think the whole debate on LGS comes from the high false alarm rate. But in the case where the user is not reacting to the alarm, chances are that the alarm is a good one – a genuine instance of hypoglycemia. This would be a reason to use the suspend function, I believe. In the case where the user doesn’t respond, you suspend insulin – simple as that.

Dr. Beaston: I respectfully disagree. I have plenty of friends and family members who can sleep through anything. Just because they don’t react to an alarm when sleeping doesn’t mean that they are hypoglycemic and incapable of responding. There are plenty of people who don’t feel the vibrations or don’t hear the alarm no matter how loud it is. I do believe there are people who don’t respond because they are hypoglycemic and incapable, but I also believe there’s a high false positive rate and the reason for not responding could be a myriad of reasons, including the fact that they would sleep through it anyway.

Q: If you’re not in hypoglycemia, your alarm clock will wake you up 90% of the time.

Dr. Buckingham: I want to agree with Patricia on this. We videotaped people sleeping, and generally those with hypoglycemia eventually responded to the alarm, while people with hyperglycemia often did not.

Dr. Tamborlane: I think that’s one of the reasons why I like the automatic suspend for projected low. A question for the agency: would you accept a study for a projected low with an automatic suspend and no alarm until the patient reached a certain threshold. Is that an acceptable study?

Dr. Beaston: You’re making the assumption that the CGM is correct. If the sensor is incorrect and it suspends when the blood sugar is actually 200 mg/dl, then the patient will not have the opportunity to reject that suspension. It’s all going to come down to what the rate of success is and how reliable the information is. It really comes down to having an idea, developing a study, and showing us the data.

Dr. Tamborlane: You can’t get the study unless you guys sign off on it! The idea of linking high glucose values with the risk of DKA is simply untrue. We have children that are 200 mg/dl all the time – 40% of their CGM values are over 180 mg/dl. They don’t get DKA.

Dr. Beaston: You’re being unfair. I am concerned about the loss of glycemic control. If you look at what average glucose values mean to A1c, things make more sense. A 15 mg/dl change in mean glucose equals a 0.5% change in A1c. If you have a system that is inappropriately turning off the pump, you still have a very real potential to affect glycemic control by having these extreme excursions into hyperglycemia. A study from a couple years ago found that if you shut the pump off, you will get a 1 mg/dl/minute increase in blood glucose.

Clinical Expectations for the Artificial Pancreas Device Systems

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

Dr. Klonoff summarized the basics of current artificial pancreas systems, suggesting that future systems would incorporate more sensors and more information technology. Saying that GPS and remote alerts could change diabetes care to make it look more like GM’s OnStar emergency response service, Dr. Klonoff presented the outline of diabetes alert system he called KlonStar. Although the name drew its intended laughter, the panelists took seriously Dr. Klonoff’s predictions about mobile health. Indeed, Dr. Klonoff cited a 2009 paper by Zisser in Diabetes Science and Technology that featured a similar telemedicine system for rescuing people from hypoglycemia.

• Dr. Klonoff described the components of current and future artificial pancreas systems. The most basic artificial pancreas systems involve CGM, an insulin delivery system, and a local controller connected by radio. Some versions currently being studied are more complex, also including a daily insulin sensitivity detector, a glucagon delivery system, and a remote controller. Going forward, Dr. Klonoff predicted that these six components would be supplemented with information technologies like food and exercise sensors, telemedicine management, and an integration into an emergency medical response system.

• In future versions of the artificial pancreas, Dr. Klonoff predicts that sensors will deliver input on far more than just glucose levels. He reviewed that current systems use CGM point data (whether subcutaneous or intravenous), SMBG calibration data, direction of change data, rate of change data, rate of change of rate of change data, and predictive modeled data. Soon, however, he said that closed loop systems could include measurement of such variables as meal size, the time that meal carbohydrates appear in the bloodstream, amount of exercise, variable metabolic profiles, stress, personal risk of hyper- or hypoglycemia, variations in insulin sensitivity, amount of insulin on board, accuracy of CGM calibration, and the pharmacokinetics and pharmacodynamics of subcutaneous insulin, subcutaneous glucagon, and/or intraperitoneal insulin.

Richard Bergenstal, MD (President, Medicine and Science, American Diabetes Association; Executive Director, International Diabetes Center, Minneapolis, MN)

In an important and anticipated afternoon session, Dr. Richard Bergenstal outlined the JDRF’s newly agreed upon inpatient clinical recommendations for closed loop systems. The recommendations were developed by an all-star panel that included Drs. Robert Sherwin, Richard Bergenstal, Patricia Cleary, Irl Hirsch, Roman Hovorka, David Klonoff, David Nathan, William Tamborlane, and Robert Vigersky. Dr. Bergenstal and the panel believe that the first studies of the closed loop should start in the inpatient setting, emphasize safety, and study ideal patient populations (i.e., healthy type 1 patients with experience using a pump and sensor). According to Dr. Bergenstal, the system should be stressed by meals and exercise in inpatient trials and a trained nurse or staff should closely monitor subjects. Generally speaking, the panel felt that inpatient trials will not require randomized control groups. Initially, it will be incredibly important to test the safety of the devices and one or two day studies should suffice. However, as the system begins its transition to the outpatient space, up to two-week studies will likely be required where a wider population of patients will have increasingly greater control over the system. It was refreshing to see clear and concise recommendations to take closed-loop research out of the lab and into the home; we can only hope that they FDA will take these recommendations seriously in the quest to develop and commercialize an AP.

Robert Vigersky, MD (Walter Reed Health Care System, Washington, DC)

Dr. Vigersky reviewed the JDRF Panel’s recommendations on outpatient trials of the artificial pancreas, focusing on the selection of patients and efficacy endpoints.

• Dr. Vigersky outlined a progression of the patients who would enroll in outpatient artificial pancreas trials, from the easiest to study to the hardest. He said that the ideal patient population with the least likelihood of treatment failure would be C-peptide-deficient people with well-controlled type 1 diabetes (A1c < 7.5%) who are already pump-treated and/or trained and competent in CGM use. Exclusion criteria would include severe hypoglycemia in the past 6-to-12 months and behavior problems such as eating disorders and ADHD. Subsequent trials might include adolescents and children, patients with severe recurrent hypoglycemia and/or hypoglycemia unawareness, and people unable to achieve A1c below a certain number (e.g., 7.5%) over the course of the previous year. Even more-challenging populations would include patients who have received pancreatectomies, who have gastroparesis, who use steroids, or who are blind.
• Dr. Vigersky defined two main types of patients who could benefit from the artificial pancreas: those with severe recurrent hypoglycemia and those who haven’t achieved their metabolic goals but don’t suffer from severe hypoglycemia. To assessbenefits for people with severe hypoglycemia, he suggested crossover studies that compare closed-loop to open-loop control with respect to nocturnal hypoglycemia (primary endpoint) and diurnal hypoglycemia (secondary endpoint). Other secondary outcomes might include non- inferiority of maintaining A1c, as well as validated measures of quality of life. For patients with generally poor control, Dr. Vigersky recommended that the primary endpoint be A1c change , with possible secondary endpoints to include non-inferiority with respect to hypoglycemia and glycemic variability. He said that the studies should be at least two-to-three months long, with safety endpoints such as lack of increase in DKA, symptomatic hypoglycemia, or other adverse events like hospitalizations and catheterizations.
• Dr. Vigersky noted that the FDA asked the panelists how to define standard of care for the artificial pancreas, but he said that none existed. Noting that there are presently not even standard-of-care guidelines for CGM, he looked forward to hearing the CGM recommendations from an Endocrine Society panel chaired by Dr. Klonoff. Dr. Vigersky said he couldn’t identify what the standard of care should be, except that it has to be better than what we’re doing now.

David Nathan, MD (Harvard University, Cambridge, MA)

Dr. David Nathan led off the final JDRF clinical recommendations presentation with a substantial claim, stating his belief that we’ve made more progress in diabetes in the last three years than we have in the last 30 years. Yet, when looking towards closed-loop studies of the future, care must be exercised, as different groups of patients will require different approaches and methods of measurement. In his view, the patients with the most to gain from a closed-loop system are patients with recurrent severe hypoglycemia. For this population of patients, the primary goal of a closed-loop system would be a reduction in CGM-measured hypoglycemia with a secondary goal of improvement or maintenance of A1c. In patients with well-controlled type 1 diabetes (i.e., A1c < 7%, little or no severe hypoglycemia), studies would have the primary aim of achieving a target A1c with a secondary goal of no increase in the frequency of hypoglycemia. The panel also strongly recommends examining quality of life in this group of patients, as the closed-loop devices should perform at least as well as open-loop devices with significantly less burden on the patient; the hope is that the artificial pancreas would significantly improve patients’ quality of life. For the third group of patients, those not achieving their A1c goals, the primary aim of a closed-loop system would be to decrease A1c and area under the curve. Studies in this group should also measure quality of life, with the goal of maintenance or improvement. Dr. Nathan made it clear that different patients will garner different benefits from closed-loop studies and researchers should keep this in mind as they move forward with artificial pancreas research.

John Mastrototaro, PhD (Vice President, Global R&D, Medtronic Diabetes)

Dr. Mastrototaro shared his own answers to the FDA’s seven questions about clinical expectations for the artificial pancreas. He began by presenting closed-loop feasibility studies by Dr. Stuart Weinzimer of Yale, and he reviewed the progression of insulin delivery devices: first CGM-guided (with a CGM device offering recommendations on delivery, SMBG testing, and adjusting basal profiles), then CGM- protected (e.g., the Medtronic Veo’s LGS system, predictive suspension systems, or “treat-to-range” systems that prevent both hyperglycemia and hypoglycemia), then CGM-controlled (e.g., overnight closed loop), and then a fully closed loop. On the FDA’s clinical questions, Dr. Mastrototaro said that acceptable target ranges depend on the accuracy of sensors, and that the best way to assess this is by evaluating real-world data like the 5 million paired readings stored in Medtronic’s CareLink database).

As an efficacy endpoint, he recommended a “goodness score” that weights glucose values according to their danger, so that 70 mg/dl would be given a much higher risk score than 230 mg/dl even though both scores are equally far from target range. For assessing safety and efficacy, he was generally skeptical of inpatient settings, saying they are too artificial to be informative and in the case of simulated emergencies to test safety, too hard on patients. Instead, he advocated simulation, modeling, and data-mining prior to randomized outpatient studies of closed-loop vs. open-loop control.

• As examples of closed-loop feasibility studies, Dr. Mastrototaro reviewed work by Yale University’s Dr. Stuart Weinzimer. In one study, 14,779 blood glucose measurements from CSII users included 9% of measurements below 70 mg/dl, 33% above 180 mg/dl, and 59%between 70 and 180 mg/dl. Of the closed loop measurements, remarkably, only 3% were below 70mg/dl, 12% were above 180 mg/dl, and 85% were in zone (Diabetes Care 2008). In a short, unpublished follow-up study, Dr. Weinzimer’s group showed that closed-loop control reduced the average number of exercise-related hypoglycemia treatments from 2.5 to 0.625 per patient and reduced the average number of nocturnal hypoglycemia treatments from 1.125 to 0.125 per patient. Acknowledging that these closed-loop systems are still not perfect, Dr. Mastrototaro said “we shouldn’t let perfection be the enemy of improved glycemic control.”Henry Anhalt, DO (Medical Director, Artificial Pancreas Program, Animas Corp/J&J, NJ)Dr. Henry Anhalt gave a useful overview of one of the promising Animas-JDRF partnership, a joint effort to bring an artificial pancreas system to market within the next few years. The Animas partnership with JDRF will focus on the third step in Dr. Aaron Kowalski’s roadmap to an AP, a hypo- and hyperglycemia minimizer. Before this is developed, a low-glucose suspend (e.g., Paradigm Veo) and predictive hypoglycemia minimizer systems will probably be approved. The Animas artificial pancreas system will use Model Predictive Control and feature robust safety constraints. For Animas, the development process has been quite atypical; going forward, constant feasibility testing, feedback, and tweaking of the system will be required. The development process has been informed by research from major well-published closed-loop groups (Sansum, UCSB, UVA, Boston University, Cambridge, etc…) and will attempt to control blood sugar using microboluses of insulin rather than traditional basal/bolus pump therapy. While no updates were given on the progress of this system, in our view, it seems as though the right ideas are in place and execution and the regulatory pathway will ultimately decide the fate of the Animas AP.

QUESTIONS, ANSWERS, AND PUBLIC COMMENTS

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

Dr. Klonoff: Dr. Beaston, how has this session gone? Have your concerns been addressed?

Dr. Beaston: It’s obvious that everyone put a lot of thought into this. The recommendations are really well-balanced and formative. And I appreciate the higher-level approach. But there is one remaining unanswered question: given the limitations in the currently available technology, what should we deem acceptable? We could potentially develop a closed-loop system that can keep patients between 80-120 mg/dl, but this might take time. Is 100-180 mg/dl acceptable? As you try to identify patient populations and what your expectations are, we’d like a little more information about what the specific goals are.

Dr. Vigersky: I think we have data that suggest that A1c of 7.5% is an achievable goal. This represents an average blood glucose of about 150 mg/dl. In this paradigm of treating to a target range, that’s a number we can start with. Then, it could be a stepwise approach. After that has been shown to be safe and effective, we can ratchet the range down. So we could start with something around 150 mg/dl. I think we should avoid being so A1c centric for the time being.

Dr. Klonoff: How difficult do you think it will be to get patients educated to use this technology in the hospital and then to get them in the transition to outpatient care?

Dr. Bergenstal: For the folks we were talking about selecting, the motivated, experienced patients, I think we’ll be surprised at how fast they catch on. I think they will pull us along.

Dr. Mastrototaro: In a TrialNet study, patients were put on a closed-loop system for a couple days, taught about the insulin pump and CGM. All the training is done then and they’re sent home, perhaps Bruce Buckingham or someone could comment on their ability to learn the technology – these are people who are pump and sensor naïve.

Dr. Buckingham: We have them as captives in the hospital for three days, where we educate them and their families. It seems to work well.

I have a very personal question for the FDA. We’ve done closed-loop studies for over 1,200 hours and achieved a mean glucose of 138 mg/dl with a standard deviation of 12 mg/dl. 12 mg/dl! When will you consider these systems safe? When can this come out of the inpatient environment? When would you consider the system safe to the point where we don’t need to be there 24/7. I think it’s safe!

Dr. Klonoff: For outpatient trials, I would like to ask what the panel thinks are good methods for following up, communication, monitoring, so Bruce can get some sleep.

Dr. Mastrototaro: I recommend using subjects who don’t live alone, who sleep close to someone else. The need for remote monitoring is something that should be considered, as well as frequent interaction with the healthcare system every week or two to make sure they are doing well.

Dr. Klonoff: Do you expect many different types of artificial pancreas systems? Dr. Nathan, you said it’s not a one-size-fits-all situation, but will there be any way that one size can fit all? Will there be any overlap at all among closed loop systems?

Dr. Nathan: I said I didn’t think one size would fit all in terms of measuring outcomes. I think one pump could be developed that would fit all ‘sizes’ of patient, but I don’t think that’s the way the work is going on. I think with the ongoing pattern of work going on in individual bunkers, we’ll come up with a variety of devices. A single approach could be feasible, but not that’s not how the work is proceeding.

Dr. Klonoff: I’m going to call the scheduled speakers. First, Mr. David C.

Mr. David C.: Thank you for the opportunity to speak. I’m a high school junior in Washington, DC. I have a disclosure: my mom works for Medtronic. But that’s not why I’m here; I’m here because I was diagnosed with diabetes at age nine. Picture you are 16 years old. You’re facing worries about standardized tests and getting into college, peer pressure, parental pressure, and on top of that you have a debilitating chronic disease. Add six-to-eight times a day pricking your finger – and, before that, remembering to prick your finger, because nobody’s perfect. Guess what? You forgot to test before lunch. You have a high later, and then you’re disappointed, your parents are disappointed, and you’re left with fears about what this will mean for your health down the road. Whenever you’re unable to tell how many carbohydrates are in some piece of food, either you can’t eat it or you guess – and guessing is bad game to play when you’re diabetic. Maybe while you’re asleep, your pump tubing comes loose and you go hyperglycemic. You’re throwing up in the morning so you have to miss school, and it’s difficult for a high school junior to miss. I have made promises I can’t keep if I’m not healthy. My brother often asks me to take care of myself, because he wants us to be able to grow old together. Panel, I beg you please to consider benefiting me and other young people with diabetes – we all just want to live normal, happy lives. Thank you.

Ms. Elna N.: Thank you. I’m the mother of a daughter who’s had diabetes for ten years, and also a registered nurse. I’ve had many interactions with providers over the years, and now I myself work in diabetes research, where I find hope. There have been great gains in the past 10 years, but CGM is not enough. We need to take the incessant 24-7 burden off kids and families. We need the artificial pancreas until we get an absolute cure. The first night my daughter used a CGM at night, I had a deep realization: for first time in so many years, I didn’t have to be on alert. I think all parents of those with type 1 diabetes should have access to CGM alarms. Then the next night, I was frustrated because the alarm wouldn’t stop going off even though she wasn’t very low, which shows that we need better systems, and soon. With regard to safety, I understand the need to have the strongest possible safeguards. Frankly, it’s too dangerous just living with type 1 diabetes. Kids are always at risk of going too low or too high. Recently, two members of the Children With Diabetes community died in their sleep, and since then I have learned of other cases too. It’s scary: a kid goes to bed at night, doesn’t wake up. Is it because the went too low, because they got arrhythmia? We don’t know. This scares parents and makes us more determined to do what we can. My daughter is in an artificial pancreas trial, and she talks about it with whomever she can. Many want to be part of these trials. I hope you incorporate these perspectives into your regulatory guidelines. People have said anything is better that what we have, and I know many out there feel that way. Thank you.

Dr. Klonoff: I think people at the FDA would like to see products on the market quickly, but they also have an obligation to show that products are safe. Many times in the past year we have read about things that are out there but not safe. It’s really a delicate balance. We all want these products, but they’ve got to be the right products.

Dr. Kenneth Ward (Oregon Health and Science University, Portland, OR): I wanted to comment on this issue of sensor redundancy. This issue has come up a few times. We’ve done studies on this. If you’re able to choose between two sensors at the time of calibration, you can get overall better data and more accurate data. The magnitude of that improvement is about 2.5 MARD points. Looking at wearing four sensors, if you’re allowed to choose the best of the four after an observational period, that improvement is about 5 MARD points. Also looking at sensor distance, you get a benefit of the redundancy even if the sensors are very close.

Q: Dr. Ward, do you think that AP systems will all have multiple sensors?

Dr. Ward: It’s not a hassle for the system to switch back and forth between the sensors. Every time a calculation is done, you calculate an error of magnitude between the CGM and blood glucose. Then, you switch to running the pump off the sensor that is reading more accurately. This can be automated. I think it’s the way to go.

Dr. Vigersky: I wanted to ask Ken (Ward) a question. What would the expected clinical difference be?

Dr. Ward: We haven’t done this yet. We have to go back and put the data into our simulators.

Dr. Joseph: We did an in-hospital study with as many as six, and the more sensors you have, the better your estimation. Obviously there’s a tradeoff of complexity versus accuracy. Many times we saw two sensors going in one direction and another going in another. We couldn’t understand why this was, but if you have an array there’s a way of voting – if two go in parallel, you can use that data to input into your algorithm.

Dr. Boris Kovatchev (University of Virginia, Charlottesville, VA): Thank you for giving me a chance to talk. I would like to address the issue of low glucose suspend systems. The case for such systems was made by Drs. Buckingham and Tamborlane. There is no question that this is a very useful safety feature. I would like to focus on a less discussed fact. The efficiency of low glucose suspend systems is only as good as the algorithm used. If a threshold is used, the algorithm is very vulnerable to sensor error because it is relying on a single data point. However, if a trend is used, this is better, as it relies on a series of data points. However, the most optimal solution may be the system approach. Such a method would use information from all available sources: pump, sensor, and other possible sources. The bottom line is that a system can be better than its components. A systems approach can make the pump safer and CGMs more accurate. But we need to discuss system architecture, system metrics, and other system-level parameters.

Dr. Klonoff: Since the topic of low glucose came up, I’m going to ask Dr. Mastrototaro a question. You said you’re looking for an equation to penalize hypoglycemia and severely penalize severe hypoglycemia. How would it compare to the one Dr. Kovatchev uses?

Dr. Mastrototaro: In our system, there is a histogram of blood glucose values, each with an associated risk score. As you get below 70 mg/dl or closer to 40 mg/dl, the risk goes significantly higher. You don’t see those same risks with hyperglycemia until you get very high, around 400 mg/dl. It’s a way of showing equivalence, or asking if they are equivalent at all. Boris has a curve on this; we have one; another group has developed one. They all have similar shapes, though the numbers are somewhat different. Still, it’s a similar concept. If you compare a closed loop group to a control group (open-loop SAP) you could use this score to come up with which has a better risk score.

Dr. Bergenstal: How would you factor in a sustained level of 400 mg/dl, for example?

Dr. Mastrototaro: If you have continuous sensing, you’ll have a continuous level. The numbers all get added together.

Dr. Richard Mauseth (Private Practice, Woodinville, WA): I’m a pediatric endocrinologist, and hearing these studies with an HbA1c of 7.1 is concerning to me. I think we can get a lot of benefit out of not having such tight control. I have teenage patients with A1cs of 10% or 11%. They would get a huge benefit from lowering A1c to 8%. Also, many young kids just can’t tolerate such intensive control.

Dr. Tamborlane: We had an argument in the JDRF group concerning the move from inpatient to outpatient studies. We wanted to have the best possible results with patients that we didn’t have to worry about regarding compliance issues and training. I agree with the idea that we have 13-18 year olds with families that could really benefit from closed-loop therapy. But we want initial outpatient studies with people who would do an outstanding job. Patients with an A1c of 9% or 9.5% are not able to use the equipment or methods of treatment that we have. However, I’m less committed to the idea of only over-21 year-old patients in these studies.

Dr. Beaston: To Dr. Tamborlane – I don’t think we ever said you can’t enroll adolescents. Adolescents are being enrolled. For ethical and safety reasons, technology should first be studied in adults. Once we have early data, we can proceed to adolescents in a stepwise fashion.

Dr. Tamborlane: I wasn’t talking about the FDA, I was talking about the JDRF working group.

Dr. Klonoff: Next is Dr. Aaron Kowalski.

Dr. Aaron Kowalski (JDRF, New York, NY): Good afternoon. I’m a scientist who’s worked on the artificial pancreas project for about six years at JDRF, which this year will fund over $100 million in research for treatments and a cure. I would like to thank the FDA and NIH for putting this meeting on, and to thank the JDRF clinical recommendations panel.

I would like to focus on the urgency of the problem we have right now. People with diabetes need better tools now. The DCCT data were published in 1993, with less than half of patients achieving their target as measured by A1c. That’s just one measure; I personally prefer time in target. Even aggressively managed patients spend only a fraction of the day in their euglycemic target range. Kids in the JDRF study spent only half the day under 180 mg/dl. People with the target A1c spent up to 90 minutes per day hypoglycemic.

Recently, a wonderful JDRF family lost their 18-year-old son. He was at a church camp, and every precaution was taken. He was surrounded by friends and he was testing, including right before bed. He died in his sleep. Systems are available that could have saved his life. They are available in Europe, but not in the US. Research on the closed loop has shown fantastic results; we need to move to real-world trials. Until patients benefit from the systems, we’re not there yet. I strongly urge the FDA to accept the recommendations of the JDRF panel, and to define a clear and reasonable regulatory pathway that will allow studies to proceed and help make these technologies available for the people and families affected by this devastating disease.

Q: Dr. Kowalski, there’s certainly a need for better technology. Do you think the problem with the artificial pancreas is that we need better science, that we need tools that don’t currently exist, or is it the regulatory environment?

Dr. Kowalski: The biggest thing we need is to move these studies out of the inpatient environment and into the real world. We need to do this for the low glucose suspend system, for the predictive glucose suspend systems, and the treat to range approaches that JDRF is currently funding. I’m completely convinced that if we do this in the real world, the systems will show dramatic improvements in the safety and efficacy over standard care.

Dr. Frank Schwartz (Ohio University, Athens, OH): Thanks, Chip (Zimliki) and David (Klonoff), for putting this meeting on. Before I give my canned speech, I would like to comment on LGS. I can count eight patients I know of who died in motor vehicles because they went low and had hypoglycemia unawareness. I was initially skeptical about daytime LGS because of the override, but I think this technology could save lives.

I would like to talk about the human factor, following Henry (Anhalt). We have algorithms that can’t even predict a pre-meal bolus. How will they predict stress, exercise, or menstruation? Individual patients tend to respond similarly to stress. There’s a catecholamine response leading to hyperglycemia in some people, whereas stress causes hypoglycemia in others. Most athletes go low during exercise, but often they will go high on the day of competition in anticipation of their event. I really advocate the use of case-based reasoning to look at each patient’s responses and add that to the algorithms we’re developing now.

Dr. Anhalt: I think that’s a wonderful way to approach these problems. However, I think that some algorithms can detect unannounced meals as well as open-loop control does.

Adam Brown (diaTribe editor/Close Concerns): Thank you for the opportunity to comment. My name is Adam Brown, and I've had type 1 diabetes since 2000, when I was just 12 years old. … Our main request today is to ask FDA to move faster in approving technologies that could help people with diabetes and to renew its commitment in promoting medical innovation – as your Mission Statement promises. It is clear that the regulatory climate is choking investments for new products in diabetes. I attend the Wharton School, and I know investment firms that now include as a point of pride in their literature that they do not invest in diabetes. Given our needs, this is a travesty. As you know, the most harrowing risk about type 1 diabetes is hypoglycemia. It can strike at any time, it strikes suddenly, and it kills. I am aware of three teenagers who this year alone have died of "dead in bed" – they never woke up because of severe low blood sugar. Those three teenagers deserved better.

Hypoglycemia also affects – and kills – people with type 2 diabetes. A recent study in the United Kingdom concluded that 40 elderly people each year die of hypoglycemia from sulfonylureas. No diabetic therapy is perfect, and we don’t expect perfection. But we know that low glucose suspend pumps and the artificial pancreas can reduce hypoglycemia, and their increased use will save greater numbers of patients from death.

The FDA's inaction in this area has resulted in real patient loss – more uncertainty, more ER visits, more deaths. We ask the FDA to consider conditional approvals. You may feel you are saving patients from products that aren't perfect, but for people with diabetes, the price of perfection is too high. We just want better. Thank you very much for your consideration.

Dr. Klonoff: You’ve lost patients to hypoglycemia, you’ve seen the stacking phenomenon. With real-time monitoring, patients in a hurry will see their glucose results not change, and they are prone to overdosing if they really want to get that blood glucose level down. We hadn’t heard about stacking before CGM became available. Now every CGM produces information that could put an impatient person at risk. Maybe the FDA should mandate that companies provide safety measures to protect patients: for example, low-glucose suspension. Maybe you could argue, as I did in my last slide earlier, that LGS is a critical safeguard against insulin stacking.

Dr. Schwartz: I agree. We’ve done studies of machine-learning for CGM hypoglycemia and hyperglycemia prediction. We don’t have enough data to present, but I think this is another tool to use with CGM and make incremental basal changes instead of the sudden ones we discussed. I think incremental changes are very important to the process.

Dr. Klonoff: I would like the panel members to comment on the FDA’s question number two: What are appropriate effectiveness endpoints? I think this is something the FDA would really like to hear from the panel.

Dr. Bergenstal: I also really like the percent of time in target as a measure of effectiveness. I think it’s much better than A1c; it gets at both highs and lows. I think highs and lows need definition so we can compare studies, but that’s a good target I think.

Dr. Vigersky: The epinephrine response kicks in around 70 mg/dl, so below 70 mg/dl should be one of the metrics. And then of course, 50 mg/dl being the point at which it becomes in most people a clinical event, 50 mg/dl should also be used. We discussed these as metrics that should be standardized and reported by everyone so that devices can be compared with equal measures.

Dr. Beaston: I’d like to ask a follow-up question for the area under the curve. How would you evaluate area under the curve in the case of frequent but small hyperglycemic excursions versus one large, extreme episode of hyperglycemia?

A: I think we need a blood glucose index. This would involve penalizing severe excursions compared with less severe excursions. Such a scoring system would give us a more accurate picture of how the system performs.

A: An episode under 50 mg/dl should be scored severely, whereas one at 70 mg/dl should be penalized less severely.

Dr. Klonoff: I think the panel this afternoon has discussed how different types of patients with type 1 may need different types of goals and protocols. I’ll give my sense of what was said, going through those seven questions.

• What is an acceptable target range depends on where you’re starting. I think Bob said A1c around 7.0% and mean glucose around 150 mg/dl. But for teenagers, it would probably be something less ambitious.
• For effectiveness endpoints, the big ones seem to be time spent in hypoglycemia and severe hypoglycemia, mean glucose, and A1c.
• How can effectiveness be demonstrated in inpatient vs. outpatient clinical trials? The JDRF panel, which I was a part of, recommended that the inpatient setting is most valuable for teaching about a technology. Then there can be a transition into a semiautonomous environment, e.g., an apartment on the hospital grounds. Then there can be a move to the outpatient setting, where patients are on their own but need to be observed. They have a loved one with them, frequent consultations with a nurse or CDE.
• What are appropriate safety endpoints? I’ve heard one method could be dual sensors, another could be dual hormone therapy, or triple hormone therapy with pramlintide. Also, telemetry is coming. I predict it will be a part of the first artificial pancreas to be FDA-approved.
• How can safety be demonstrated? We discussed a progression of endpoints.
• Safety mitigations we’ve already basically addressed.
• For the effective balance of effectiveness and safety, we discussed two main categories of patient: the person with lots of hypoglycemia problems and the person with high A1c. If someone has lots of hypoglycemia, the main goal is to decrease hypoglycemia. For the person with poor metabolic control, the goal is to improve that. Always you want to improve quality of life. That’s how I see the situation, that’s how I would respond to the FDA’s questions.

Comment: I just want to make a quick comment about the use of other hormones. I’m wondering about the potential for an investigational device exemption from the FDA. We would like to receive this as early as possible because of our interest in studying the use of glucagon and pramlintide in closed-loop therapy.

Dr. Tamborlane: If I recollect properly, to get a new drug approved by the FDA, the drug must be deemed effective and safe. To test effectiveness, we used to conduct a placebo-controlled trial. Now, we use an active comparator that demonstrates whether the new drug is as effective as the current approach. I believe we’re putting a higher standard on the artificial pancreas. We’re asking for a trial of a closed-loop group versus standard therapy, the active comparator. This study will help us see if the therapy is effective and safe compared to what we currently have; namely, non-inferiority not necessarily superiority. Superiority in lowering A1c or in reducing the exposure to hypoglycemia should not be the requirement for approval. Everything that we’ve done to add to diabetes technology has added a burden on patients and families. The question is whether we can reduce the burden on families and achieve the same glycemic outcomes. If so, that should be sufficient for approval.

Dr. Klonoff: Bill (Tamborlane), I think that what you said suggests a big improvement in quality of life with the closed loop system. If we showed a non-inferiority of outcomes, then improvements in quality of life would make it a candidate for approval.

What Safety Information is Needed in Clinic Studies Prior to Adding Outpatient Studies?

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

In a presentation focused on transitioning closed-loop research from the inpatient to the outpatient setting, Dr. Roman Hovorka laid out some important guidelines that the researchers and the FDA must keep in mind. To start, Dr. Hovorka agreed with previous speakers in that low-glucose suspend systems should be tested in the outpatient setting. However, for closed-loop research, studying in the clinical research center environment cannot be avoided. Risk and hazard analysis must be performed in the inpatient setting to identify the myriad of system level issues that can arise in the artificial pancreas. Evidence must be collected in the clinical environment on challenging lifestyle conditions such as large meals, stress, and exercise. Simulations will also play an important role in gathering data, permitting edge-of-the-envelope tests that would be impossible in the clinical environment. Following sufficient collection of data in the clinical research center environment, transition studies could take place that would allow testing of more human factors in this system. Finally, after a brief review of his promising closed-loop research, Dr. Hovorka postulated that outpatient prototype development will take approximately one-to-two years, bringing the time to the first outpatient studies in the neighborhood of two-to-four years depending on system complexity.

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

Dr. Damiano described the process of transitioning from inpatient to outpatient studies of the artificial pancreas. He drew examples mainly from his own group’s research and presented a timeline of preclinical and preclinical studies, starting with a series of swine experiments completed by his group in 2009 and looking as far ahead as 2014, the hopeful start date of a year-long clinical trial of an outpatient closed-loop device that would be submitted for regulatory approval. (As part of his group’s early studies, the DexCom Seven and Seven-PLUS, the Medtronic REAL-Time Guardian, and the Abbott Navigator were compared against each other. Dr. Damiano noted that the Abbott product was the most accurate, and he lamented that the product’s current unavailability in the US is “a terrible situation for people with diabetes currently and for closed loop control.”) He explained that successive trials would have less monitoring by health care professionals and more freedom for patients, with increasingly smaller and more integrated devices. His group is currently performing a second round of inpatient feasibility tests, featuring a CGM-driven system worn by patients who have six meals and 40 minutes of structured exercise over 48 hours. The third and hopefully final set of feasibility studies is set to begin in nine-to-twelve months, when the artificial pancreas will be worn on a pouch and meals and physical activity will be unrestricted. Although he plans to study bihormonal and insulin-only pumps in parallel, he expects that since glucagon is not yet approved for ongoing treatment of hypoglycemia, an insulin- only system has better prospects for approval. Noting that he hopes a closed-loop system is on the market by the time his 11-year-old son goes to college, Dr. Damiano also emphasized the importance of ensuring the quality of the first artificial pancreas: an inferior technology could “cast a shadow” over all subsequent closed-loop efforts.

• Following a successful study in swine (2005-09), Dr. Damiano and his team conducted their first human feasibility study of closed-loop control from 2008 to 2009. In this small inpatient trial of C-peptide-negative patients, the researchers used a bihormonal (insulin and glucagon) pump and no pre-meal boluses (even though Dr. Damiano said he expects the first approved closed-loop system will include these manual priming boluses). Dosage was driven by venous glucose measurements rather than CGM. Three large, weight- standardized meals were administered over 27 hours without any rescue snacks allowed (except in cases of dangerous glucose excursions, in which case health care providers were standing by). Patients were not allowed to exercise. Dr. Damiano noted that his group tried to make use of all the data available by measuring hormone levels in addition to blood glucose.
• Dr. Damiano explained that successive trials would have less monitoring by health care professionals and more freedom for patients, with increasingly smaller and more integrated devices. His group is currently performing a second round of inpatient feasibility tests, featuring a CGM-driven system worn by patients who have six meals and 40 minutes of structured exercise over 48 hours. The third and hopefully final set of feasibility studies is set to begin in nine-to-twelve months, when the artificial pancreas will be worn on a pouch, and meals and physical activity will be unrestricted for five days. The first outpatient study, involving five days of “house arrest” and a nurse on site, could begin as early as 2012. The first “leap-of-faith” study not to involve a nurse could start in 2013, and it would involve six weeks of open loop control compared to six weeks of closed-loop control during routine living conditions. The “pivotal study,” possibly coming in 2014, would enroll roughly 200-300 patients in a comparison of six months of closed-loop control and six months of open-loop control. The device used in this study would be ready for FDA submission, with some cohorts using a bihormonal version and some using insulin only.

QUESTIONS AND ANSWERS

Moderator: Robert Vigersky, MD (Walter Reed Health Care System, Washington, DC) Dr. Vigersky: What are the patient selection criteria in your studies? Is there a difference between clinical research centers and outpatient closed-loop studies? Which groups of patients will you study first?

Dr. Damiano: When we started our first study, we required A1cs below 8.0%. We did not admit subjects with severe hypoglycemia. Basically, we wanted people with fairly well-managed diabetes on a pump for five years. We also limited it to 18 years and older. We relaxed these requirements in the second study, admitting subjects as young as 12 years old. We also relaxed our A1c and pump requirements slightly. This relaxation in exclusion criteria was intended to challenge the system. As we move on to the outpatient setting, we’ll recoil and look towards a smaller population of patients that will ensure better safety and effectiveness – patients with better managed diabetes and more commitment to the technology.

Dr. Hovorka: Our inclusion criteria was an A1c below 10%, six months or more of insulin pump use, and type 1 diabetes for one year. For our upcoming home studies, we are also requiring good hearing and good vision to ensure subjects respond to alarms. Finally, we’re only looking at patients without complications and frequent severe hypoglycemia patients. These are not the ideal patients that would most benefit from the system, but we’ll reduce the requirements down the road as we expand the outpatient studies.

Dr. Tamborlane: We’ve been studying in subjects 15-30 years old.

Dr. Markham Luke (FDA, Washington, DC): I just want to comment – it is appropriate to start with a well-selected population that will be good subjects. This is really important for the foreseeable future if the product gets to market. Those will be the kinds of patients selected to receive those devices; you don’t want to throw a device at someone who’s going to fail it. Some of that can be measured by compliance with other therapies.

Also, I would like to address a perception I’ve heard throughout this meeting – that we at the FDA are stalling or preventing development in this area. I would like to propose that we’re actually here to guide safety and efficacy investigations for medical products going forward and to provide an environment for discourse. There is some polarity even among the research community, and that’s also reflected among FDA reviewers. It’s important not to let the discourse generate polarization that in itself can lead to stalling development. We need to work together to get goals accomplished. There have been past cases like this, for example the artificial heart. Even at the time I began my medical career, it was a glint in someone’s eye. Now it is a reality in patients, not just as a bridge to transplant but as a destination therapy.

The evidence submitted to the FDA needs to be given to help us move along with planning. As Dr. Damiano mentioned, these should be not just summaries but subject-level data. There should also be good clinical practice in the trials; the quality of the study colors how beneficial the evidence is. Also, we like to see validation of endpoints when we’re talking about patient-reported outcomes. They should be validated in the population you are studying. I’m not sure why data is often held so tightly by investigators. Maybe there’s a feeling that it’s proprietary. I think it’s important that it be shared in the research community as much as possible. A group like the JDRF can help with that. Thank you for hearing my comments.

Dr. Buckingham: You mentioned that one of the first groups to be targeted would be those with nocturnal hypoglycemia. How will you monitor patients at night and let them sleep? How will this transition to the outpatient setting?

Dr. Damiano: In our current studies, we’re not monitoring overnight. In the past, we’ve used a GlucoScout every 15 minutes to check IV blood glucose. Going forward to the outpatient environment, we can’t use an IV system. This is why I think the HemoCue glucose testing system is good – it’s very accurate. We will not be able to evaluate the period from 11pm to 7am. However, one approach might be to use redundant CGM systems to verify that the CGM that is driving the algorithm is reading accurately.

Dr. Hovorka: In our studies, we will be using CGM as a primary outcome. In terms of using multiple sensors, much of the error is due to calibrations. Thus, even using two sensors, you may still have the same sensor error if you use the same calibration.

Dr. Buckingham: When you go to in-home studies, how are you going to assess outcomes at night? Dr. Hovorka: As I mentioned, the primary outcome is time in target, assessed by CGM overnight.

Dr. Aaron Kowalski (JDRF, New York, NY): I want to point out something interesting about this entire discussion. We have a goal of a fully automated closed-loop system without patient input. In the near term, however, patient interaction with the system will be critical. Let’s remember that insulin pumps today are making automated decisions throughout the day based on no information at all! Now, I’m exaggerating a little, but the decisions are based on preset basal rates from a visit to an endocrinologist three months ago. Dr. Buckingham showed data from a patient that had a seizure due to a pre- programmed increase in her preset basal rate while she was hypoglycemic; the pump didn’t know she was low and couldn’t turn off insulin infusion. In the near term, we can use the information from CGM to make small but clinically significant changes in insulin dosing and potentially avoid these dangerous scenarios.

Dr. Frank Doyle (University of California, Santa Barbara, CA): Two terms keep popping up today that I think are potentially incongruent: the closed loop and mealtime priming boluses, which of course brings humans back into the loop. I was wondering if the panel could talk about this.

Dr. Tamborlane: First of all, the human right now is always in the loop. Managing the system always involves the patient. To me, adding a priming bolus doesn’t change the idea that most of the work would be done by the system. The difference between auto-suspend and a fully closed loop is that, as Dr. Knight said, for the auto-suspend you can go right to outpatient trials. The system can either turn insulin on or off. To me, turning insulin on is a major obstacle. All it takes is one catastrophic event, when a pump delivers 300 units and kills a patient. We’ve barely touched on mitigation issues today with respect to over

delivery issues. Also, as we heard this morning, a simple thing like shutting the system down during hypoglycemia doesn’t take any fancy algorithms; that we believe is fail-safe. We need to pay more granular attention to things that will prevent over-delivery. Maybe that means putting a maximum delivery rate per hour, or using dual sensors – not only to pick the best one, but to have two or three reading the whole time. Then if the sensors diverge, maybe you want to be able to shut the system down and switch over to pre-defined basal rates. There are many of these issues and they will drive outpatient studies. We have to look at the patient end, how they can screw things up, as well as the system end, how it can fail. Another thing from aviation: engines are inspected and repaired periodically. Maybe these closed-loop systems would be regularly evaluated.

Dr. Damiano: When we design a pre-meal priming bolus, the system is made aware of it. The intent is for subjects to provide a little bit of input, but the controller projects the insulin rise in its calculation of subsequent doses. The system will be a little more cautious. You can have the subject involved in this and at the same time, the system is aware of everything they’ve done.

Dr. Beaston: We’ve seen some pretty amazing correlations between CGM and reference methods. They’re actually using reference glucose measurements rather than fingersticks to calibrate CGM in many of these studies, however. A big limit to success is a really accurate, portable device to get good calibration and improve any system we use.

Dr. Hovorka: I did some back-of-the-envelope calculations on this. Half of the daily dose is pre-meal and half is basal. If you look at the daytime, then, three quarters of the insulin is coming from boluses and one quarter is coming from basal. So meal bolusing is a big challenge. I think priming boluses are the right way to do it, but if we could have a functioning system without them, that would be great.

Dr. Luke: As humans, we’re not closed systems, we’re open systems. We get fevers; we take medications. What if patients get H1N1? It’s something to be explored as you develop these products.

Comment: I’d like to point out what is going on in aircraft engines. Aircraft engines are in permanent communication with the ground. With respect to the artificial pancreas, it might be feasible to have a real time link to experts who can monitor patients.

Dr. Vigersky: I think, in line with Dr. Klonoff’s comments, that telemetry in the closed loop will happen in the near future.

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