American Diabetes Association 74th Scientific Sessions

June 13-17, 2014; San Francisco, CA – Diabetes Technology – Draft

Executive Highlights

This report contains our coverage of all-things Diabetes Technology from ADA 2014 – artificial pancreas, CGM, SMBG, insulin delivery, and data/connected devices. Below, we enclose our themes from the meeting, followed by detailed discussion and commentary from each of the aforementioned sections.

Talk titles highlighted in yellow were among our favorites from ADA 2014; those highlighted in blue are new full report additions from our daily coverage.

  • The profile of automated insulin delivery (also known as the artificial pancreas or the bionic pancreas) continues to rise at each ADA, and this year’s conference truly exemplified that theme. That said, ADA did not bring any major breakthrough data; partly this is timing and partly this reflects the fact that ATTD is increasingly becoming the go-to hub for new diabetes technology data. Visibility aside, we do point out there is a huge spectrum of approaches to automating insulin delivery, with very different demands on the person with diabetes as well as different levels of perceived risk. There is also more disagreement in the field; though reimbursement has long been a source of uncertainty, there are also more commercial questions that have come to the fore that weren’t as visible even a year ago.
  • A major positive headline in automated insulin delivery came during Dr. Steven Russell’s (MGH, Boston, MA) presentation on the bionic pancreas – results from the Beacon Hill and Summer Camp Studies were published in the NEJM (topline results were originally shared at ATTD 2014). The paper received a veritable media blitz and quickly became the most popular article of the last month on by a very wide margin (195,000+ views vs. 42,000 views for the #2 article). As we understand it, the impressive popularity will put it among NEJM's most viewed articles of 2014. This publication itself – and the resulting wave of attention – was a clear reminder of just how far the field has come in less than a decade, moving from a bedside research project at just a couple of institutions to making major national and international headlines.
    • The single oral session on the artificial pancreas was headlined by the Cambridge team, who presented some of the longest, largest, and most real-world studies we’ve seen to date: Dr. Hood Thabit shared data from a four-week overnight home study, while Dr. Lalantha Leelarathna highlighted results from a seven-day home study testing unsupervised 24-hour closed-loop control. Both studies showed that closed-loop outperformed open-loop therapy as measured by time-in-target range (hypoglycemia was very low at baseline and did not improve significantly). The major area for improvement was also clear in both studies: connectivity. The seven-day trial has paved the way for the AP@home04 study, which will include a striking three months of day+night closed-loop control in up to 42 adults. currently indicates an anticipated primary completion date in June 2015, which of course also underscores the runway needed to enroll and complete these longer studies.
    • We’d emphasize that the catch-all term “artificial pancreas” is becoming increasingly insufficient to describe all systems, as systems differ greatly in their level of automation and need for user interaction. Though the Cambridge and Boston systems were both discussed in the “artificial pancreas” oral session, they are very different – the Cambridge approach modulates basal insulin delivery and requires user input for meals, while the Boston system controls basal delivery and can handle meals without any user announcement, although the average glucose will be better if meals are announced. The Cambridge approach itself is still a dream to many patients (we’d have our basals modulated any day!), so we are less concerned with “Which system is better?”, and more focused on, “When are these different systems coming and what will commercial embodiments look like?” To what extent the Cambridge vs. Boston approaches will reduce the burden of managing diabetes will vary from patient to patient, and family to family, of course – in any case, both stand to make huge gains in the area of diabetes burden. As we noted in our ADA 2013 themes, we would not be surprised to eventually see different closed-loop products for different types of patients – those already in tight control might benefit the most from a predictive low glucose suspend system, while those who often forget meal boluses might see the most positive benefits with a 24-hour closed-loop system. Like in other market spaces, we expect to see variability in what patients and families want from their technology.
  • Challenges aside (which all first generation products surely have), the excitement is palpable – and the pressure is now on to sort out commercialization. Said JDRF’s Dr. Aaron Kowalski at the JDRF/NIH Closed-Loop Meeting: “We’re right on the cusp. People are wearing closed loop at home, and these systems are safer than what we’re doing right now. We’ve got to drive towards commercialization.” This is an invite only event, but we go into detail in this report on the meeting. This quote from Aaron served as an excellent reminder of where the field stands right now and where it must clearly go in the next couple of years. Now that closed-loop control has demonstrated safety, efficacy, and feasibility in inpatient and transitional studies of increasing rigor, the big unanswered questions surround commercialization – What will pivotal studies look like (size, length, comparator group)? What will commercial products look like (i.e., level of automation, user interface, etc.)? Who will commercialize automated insulin delivery (industry, academic investigators, partnered companies/institutions, etc.)? Indeed, the JDRF/NIH Closed-loop evening shared industry perspectives from Medtronic, Animas, and Dose Safety (an artificial pancreas software startup) in an aptly named session, “The Last Mile – Bridging From Academia to Industry.” The tone of the industry presentations and subsequent panel discussion was somewhat negative, centering on the challenges around safety, robustness, and everything that can go wrong (failed infusion sets, calibration issues, skipped boluses, dead batteries, etc.). This contrasted markedly with the second panel, which featured off-the-charts enthusiasm from patients and closed-loop researchers. All were highly enthusiastic about the systems they’ve experienced first-hand – said patient Miss Willa Spalter (age 12), “Nothing is perfect, but I definitely would sign up [right now]. It’s much better than what everybody else is wearing now.” There is clearly a hard balance to strike between patients’ interests in systems to come now, and industry’s concerns over liability and investment and who will own what.
  • ADA 2014 left us with many outstanding questions in automated insulin delivery: Following predictive low glucose suspend, what will be the next commercial product – overnight-only closed-loop, daytime treat-to-range, 24/7 hybrid (or fully) closed loop? What level of automation will be meaningful enough to get a sizeable number of patients on automated insulin delivery? Will academic groups pursue commercialization of closed-loop devices? How will industry balance patient demand to close the loop with liability concerns and R&D constraints? How large and long will pivotal studies be? What’s an appropriate comparator group – sensor-augmented pump therapy or whatever therapies patients are currently using? What data on automated insulin delivery will it take to support broad reimbursement and access? What level of CGM accuracy and reliability is needed for safe and effective overnight treat-to-target/daytime treat-to-range control? How will chronic use of glucagon be defined and what will trial requirements be? Will Xeris’ stable glucagon be ready for the Bionic Pancreas pivotal study?
  • This was a lighter year for new CGM data and systems, perhaps a testament to the field’s laser focus on automating insulin delivery – indeed, we thought it was notable that ADA 2013 had an entire oral session devoted to CGM, while this year’s conference focused all new CGM data into poster presentations. One could argue this was an in-between year for Dexcom and Medtronic, as both have recently commercialized their next-gen sensors (G4 Platinum and Enlite, respectively) and are working diligently on future add-on products or new sensors (Dexcom Share, Gen 5, Gen 6; Medtronic’s Enlite 2, 3, redundant sensors, MiniMed 640G).
    • However, notable new CGM data did emerge in posters from Dexcom, Roche, and Senseonics – accuracy of these systems hit or exceeded the ~10% MARD threshold. Dexcom had some of the biggest CGM news of the meeting in late-breaking poster #75, which shared clinical data (n=51) on a version of the G4 Platinum with an improved algorithm (“G4AP”). Overall G4AP MARD vs. YSI was an impressive 9.0%, compared to the Contour USB meter’s MARD of 5.6% vs. YSI. This poster, along with several accompanying posters, made a case that the G4’s accuracy is approaching that of fingersticks – a smart move from Dexcom as it seeks to eventually obtain an insulin-dosing claim for CGM. Meanwhile, in poster #846, Roche shared data comparing its prototype CGM to the Dexcom G4 Platinum – the mean seven-day MARD was 10.9% for the G4 and 8.6% for the Roche prototype. Senseonics also shared new data on its implantable CGM – over 90 days of use, overall MARD vs. YSI was 11%, ranging from a low of 7.7% in one patient to a high of 17.7% in another patient. The Senseonics and Roche posters did not divulge a lot of study design details, so it’s hard to know how real world the accuracy is. Still, we are encouraged to see new CGMs in development from established players, along with new entrants.
    • Aside from the recent EU launch of the MiniMed Duo combination insulin infusion-CGM device, Medtronic was notably stealth on the next-gen sensor front. It was nice to see the MiniMed Duo in the international section of Medtronic’s ADA exhibit hall booth, though we continue to believe commercialization will be challenging for this product (see our report for more details). In terms of the company’s CGM pipeline, Enlite 2 was launched in Europe in conjunction with ATTD 2014, though no accuracy data has ever been shared on this product. Meanwhile, the US study of the MiniMed 640G predictive low glucose suspend device will use the Enlite 3 sensor – according to Medtronic’s June 5, 2014 Analyst Day, this product will include “intelligent diagnostics” and “improved accuracy and comfort.” The company also has an orthogonally redundant CGM (data last shared at ATTD 2014) and a redundant glucose oxidase sensor system (last discussed at DTM 2013).
  • ADA 2014 included three notable partnership developments in diabetes technology: Medtronic/Sanofi, Dexcom/Insulet, and Joslin/Glooko. We see all three partnerships as valuable news for patients, since there is so much room to expand device penetration, to get more patients to goal, and to make providers’ lives easier.
    • Medtronic and Sanofi jointly announced a “strategic alliance” focused on improving the management of type 2 diabetes. The partnership will pair Sanofi's insulin and GLP-1 portfolio and drug development expertise with Medtronic’s background in insulin pumps and CGM – a particular priority is new drug-device combinations, including new form factors that are affordable, convenient and easy to-use. We could imagine multiple products that could come out of the alliance, especially prefilled patch pen-like wearable devices (like Valeritas or CeQur) or simplified prefilled insulin pumps; too, we think the information from CGM will be hugely valuable for Sanofi as it expands and begins to serve patients across a broader spectrum of diabetes.
    • Dexcom announced that its upcoming Gen 5 mobile app will pull data from Insulet’s next-gen PDM via Bluetooth. This was major and fairly unexpected news following the dissolution of their PDM-CGM integration partnership in 4Q12 (a move that was ironically motivated by Dexcom’s desire to move to the smartphone in the first place).
    • Joslin and Glooko debuted their HypoMap software to identify and improve hypoglycemia unawareness, which we covered in detail just prior to ADA 2014.
  • Insulin pumping for type 2 diabetes was not a major focus of ADA 2014, though two trials were encouraging, headlined by data from Medtronic's long-awaited Opt2mise trial – the randomized, six-month study compared insulin pump therapy (n=168) to MDI (n=163) in type 2 patients in poor control. From a baseline of 9.0%, A1c declined by 1.1% in those in the pump group vs. a 0.4% decline in the MDI group (p<0.001) after 27 weeks; 55% of the pump group achieved an A1c <8% vs. 28% of the MDI group. Given the high starting A1c of 9.0%, the magnitude of reduction (-1.1%) was perhaps not quite as high as some would have expected – we wonder if insulin titration could have been better, if a simpler device with on-body bolusing (e.g., Valeritas’ V-Go or CeQur’s PaQ) could have helped drive patients even lower, or if this simply underscores what a challenging population this is to manage. Certainly a mean of 8% is impressive and some patients were likely well below 8%; virtually all patients were also taking less insulin, a major cost and personal victory for patients. Results from this trial were published in The Lancet (Reznik et al.) shortly after ADA on July 3. We were also intrigued by a study from Dr. Anand Velusamy (King's College Hospital NHS Foundation Trust, London, UK), which examined the impact of pumping U500 insulin in very poorly controlled, highly insulin resistant type 2s. A1c declined by 1.9% at six months (baseline: 10.4%), 2.3% at 12 months, and was maintained out to 36 months. At the same time, total daily insulin requirements declined by ~20%. Perhaps most notable were the cost implications – using U500 in the pump vs. U100 insulin was estimated to save ~2,200 British pounds per patient per year (~$4,000 USD). Though the study was uncontrolled and patients did receive nursing support, we thought these were very strong clinical results in a highly challenging population. This clinical data lines up well with efforts from pump companies that are now actively pursuing type 2 focused products (Insulet’s U500 OmniPod with Lilly; Tandem’s 480-unit reservoir t:slim; Medtronic’s new type 2 business unit). The need is substantial in the severely insulin resistant population, and it’s terrific to see movement from industry that corresponds to encouraging clinical data that continues to accrue on this front.
  • This year’s ADA highlighted several key issues in self-monitoring of blood glucose (SMBG), most notably the FDA’s recent draft standards on BGM accuracy and DTS’ new post-marketing error surveillance grid. Dr. David Sacks (NIH, Bethesda, MD) discussed the recent FDA draft guidance, which he described as “really narrowing the range for error” in comparison to the 2013 ISO and CLSI standards. He described the particularly stringent guidelines for point-of-care meters as “obviously not feasible in most circumstances” (in line with his comments from the EASD Diabetes Technology Conference in February), pointing out that even some central lab methods cannot meet the new point-of-care accuracy bar. The concern that many in-hospital meters won’t meet the draft accuracy standards is not a new – we have heard similar concerns from Dr. David Klonoff (Mills Peninsula Health Services, San Mateo, CA), who addressed another SMBG hot topic – the new DTS Surveillance Error Grid (SEG) for post-marketing surveillance. Compared to the older Parkes and Clarke Error Grids, the SEG now accounts for DCCT trial results, analog insulins, new information about hypoglycemia, and raised BGM accuracy standards. Dr. Klonoff noted that the FDA has already begun using the SEG as a model to assess other measuring devices, and the hope is that the FDA will use the SEG as a post-market surveillance tool for BGM (definitely not for pre-market use). Currently, the FDA doesn’t conduct post-market surveillance, but assessing and enforcing meter accuracy remains a concern for both patients and providers. As of a June 12 email we received, DTS announced that a Steering Committee had been assembled for its post-market Surveillance Program for cleared BGMs. The first committee meeting will take place in July in Washington, DC. This follows the May announcement that DTS had kicked off the surveillance program planning process with funding from Abbott.
    • Since ADA 2013 largely ignored last year’s issues of competitive bidding and the increasingly challenging environment of blood glucose monitoring, we were pleased to see renewed interest in SMBG. We counted a total of 17 abstracts, three orals, and four late-breaking posters this year related to BGM, a solid increase from 2013, which featured 13 abstracts, one oral, and no posters by our count. However, similar to last year, the Big Four BGM companies did not make a strong showing at the exhibit hall; of the Four (Bayer, Roche, Abbott, and J&J), only J&J was present. The trend worsened from last year, where both J&J and Roche were present. 
  • The value of SMBG for non-insulin users remains controversial, but speakers did provide evidence for its use, assuming it is bundled with education. Dr. Richard Grant (Kaiser Permanente Northern California, Oakland, CA) noted that in a review of 12 RCTs of patients with type 2 diabetes, SMBG reduced mean A1c by a marginal 0.26%. However, he emphasized that the “mixed bag” of results for RCTs speaks to the necessity of prescribing SMBG for patients with type 2 diabetes in the context of a larger educational effort and as a tool to effect change in self-care or medication. An oral presentation by Dr. Yi Sun Yang (Chung Shan Medical University, Taichung, Taiwan) on SMBG models in type 2 diabetes supported Dr. Grant’s emphasis. All three SMBG models (six-pair, three-pair, and seven-point testing) all demonstrated substantial reductions in A1c (-1.7%, -1.8%, and -1.1%, respectively) in combination with thorough education and protocols for translating results into treatment changes.
    • These presentations were encouraging, particularly in light of the recent legislative effort in Oregon to restrict test strips for patients with type 2 diabetes not on insulin. Audience members during Dr. Grant’s presentation were particularly interested in this issue, bringing up during Q&A that the Oregon legislation used Dr. Grant’s DISTANCE study to support test strip restrictions. For background, Dr. Grant’s study suggested that 15% of patients reported that their SMBG results were not used by anyone to make adjustments to diet, exercise or medicine. Notably, Dr. Grant was quick to clarify that “he would never have come to the conclusion that test strips should be restricted for all patients with type 2 diabetes not on insulin.” Rather, he would focus on individualizing care and on prescribing SMBG to patients who will benefit from it. With regard to the Oregon legislation, Dr. Grant commented, “Using population-based prescriptions to restrict strips doesn’t make any sense... I do not agree with it at all.”
    • There was no data on Abbott’s Flash Glucose Monitoring at this year’s ADA, though there is certainly clear momentum behind it, as we could tell from multiple hallway conversations; we expect to see a major presence at EASD 2014. The last data on Flash Glucose Monitoring was shared in a symposium at ATTD 2014 in February. As a reminder, Abbott’s Flash Glucose Monitoring system is intended to overcome some of the limitations of both BGM (pain, inconsistent and hard to interpret data) and CGM (alarm fatigue and cost) – the factory calibrated two-week subcutaneous sensor is expected to have an insulin dosing claim, and for the most part, patients will not need to use any test strips – if it works as advertised, it could the change the paradigm of glucose monitoring. The device is still pending a CE Mark, and management expects the technology to launch in late summer in the EU, according to Abbott’s 1Q14 financial update. The system’s sensor patch, which will be the size of a €2 coin, could be particularly attractive for type 2 patients desiring a low-hassle and discreet option for measuring blood glucose. There is still no US timeline on this device, but we are optimistic.
  • Though still in the “early adopter” phase, ADA 2014 reminded us of the increasing demand for more connected diabetes devices. The desire for better connectivity has long been a theme of many device presentations (particularly those related to the artificial pancreas), though the DiabetesMine D-Data Exchange event  showed us how some smart and driven patients are taking matters into their own hands (#WeAreNotWaiting on Twitter). Most notably, we got a look at Nightscout/CGM in the Cloud, which has rocketed in popularity on Facebook – the project allows anyone to download instructions to “hack” their Dexcom CGM and send the data to the cloud and then to any device. The patient enthusiasm for this approach underscored just how much demand there is for greater connectivity, and correspondingly, we continue to be encouraged by industry’s progress: Dexcom Share was advertised on brochures in the exhibit hall (still awaiting FDA approval); Glooko attracted a number of interested attendees in the exhibit hall with its MeterSync cable, HypoMap software, and in-development Bluetooth product; LifeScan’s OneTouch VerioSync, Sanofi’s iBGStar, and Telcare’s BGM were all being demoed for eager attendees; Tidepool’s Blip (web-based universal data platform) started a clinical trial at UCSF (Adam and Kelly are both dying to get into it) and drew lots of interest at the D-Data Exchange; and smartphone BGMs from Dario and iHealth are now out in the marketplace. In our view, the major question is whether patients and providers will embrace these systems, particularly given the poor reimbursement for non-face-to-face patient/physician interactions. In addition, we’d note that these new systems are at the stage of simplifying data upload (a very laudable goal, given historical challenges on this front!) – we think the major inflection point will come when software takes the uploaded data and delivers actionable recommendations to patients and providers. Baby steps first, but the future looks bright in our view.


Table of Contents 

Artificial Pancreas

Oral Presentations: Constructing an Artificial Pancreas

Multiday Outpatient Glycemic Control in Adolescents with Type 1 Diabetes Using a Bihormonal Bionic Pancreas: The Barton Center Summer Camp Study (237-OR)

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

Dr. Steven Russell shared topline results from the bionic pancreas Summer Camp and Beacon Hill studies, which were simultaneously published online in the New England Journal of Medicine (“Outpatient Glycemic Control with a Bionic Pancreas in Type 1 Diabetes”) – this publication received significant press (New York Times, Wall Street Journal, Washington Post, Boston Globe, Time, Bloomberg, USA Today, Huffington Post, NPR, CBS, Popular Science, US News & World Report, Fortune, and more listed here) and quickly rose to become the most popular NEJM paper of the past month. Though Dr. Russell’s topline results presentation was similar to those given at ATTD 2014, it was terrific to see the excitement in the room among fellow researchers – said Dr. Roman Hovorka, “Congrats on the publication; it’s brilliant for the field and for you as well.” In his talk, Dr. Russell emphasized the impressive average level of glucose control during closed loop (133 vs. 159 mg/dl in adults; 138 vs. 157 mg/dl in adolescents), which was simultaneously achieved with no increase/significant reduction in hypoglycemia (4% vs. 7% in adults; 6% vs. 8% in adolescents). Dr. Russell also noted the challenging circumstances of these studies (strong usual care control in the camp environment; 45% of the Beacon Hill adults wore their own CGM during usual care) and the robustness of the control algorithm (adaptable over time; initializes only based on weight). The team’s first home use study began on June 16 – these randomized, crossover experiments in adults with type 1 diabetes will compare 11 days with the Bionic Pancreas to 11 days of usual care. The multicenter study will take place at MGH, UNC, Stanford, and UMass, with 12 subjects expected per site. Patients must either work or go to school at the institutions, and their home must be within 30 minutes of the center. Notably, they are allowed to travel as far as 60 minutes driving time away (including driving their own car!). Remote monitoring will be quite minimal. This study is certainly one of the more ambitious and real-world outpatient studies to date, and we cannot wait to see it get off the ground.

  • The results from the bionic pancreas Summer Camp and Beacon Hill studies were published online in the New England Journal of Medicine (“Outpatient Glycemic Control with a Bionic Pancreas in Type 1 Diabetes”) on June 15. The paper impressively combines both studies into a single manuscript. The publication has rich detail on the performance of the bionic pancreas and provides lots of illustrative data and statistics.
  • The randomized, crossover Summer Camp study compared five days on the bionic pancreas to five days of supervised camp care. The study took place at Camp Joslin (n=16 boys) and Clara Barton (n=16 girls) in 2013. Point of care capillary blood glucose checks occurred during the day and night (no venous glucose monitoring!). The same mobile platform was used as in Beacon Hill – two Tandem t:slim pumps (insulin and glucagon), a Dexcom G4 Platinum sensor and transmitter, and an iPhone 4S controller. Study staff and camp staff provided 24-hour, round-the-clock telemetry to monitor glycemia. A total of 160 days on the bionic pancreas were accumulated. For more information and interviews with trial participants, please see our detailed Closer Look write-up after we visited the study site this past summer.
    • Relative to usual care, the bionic pancreas improved mean blood glucose (158 to 142 mg/dl) and simultaneously reduced hypoglycemia (2.2% to 1.3% of time <60 mg/dl). Dr. Russell showed a plot charting individual patients’ mean glucose (days 2-5) in the control condition vs. with the Bionic Pancreas. Notably, thirty-one out of the 32 patients had a mean glucose <168 mg/dl (the ADA goal of <7.5%) on the bionic pancreas. Most patients experienced a striking decline in mean glucose, with some going from over 215 mg/dl to <145 mg/dl. In the handful of patients who did see a rise in mean glucose after wearing the bionic pancreas (5/32), the system reduced high baseline hypoglycemia and still brought patients to goal (<168 mg/dl).
    • One patient did have a mean glucose exceeding 168 mg/dl on the bionic pancreas, a finding attributed to the adaptive algorithm. In most patients, it takes the algorithm ~18 hours to adapt to patients (initialization only requires weight) – this is why the team focuses on study data for days 2-5 (i.e., day one is not representative of how the system would perform ad infinitum). In the case of this single camper, it took the algorithm two days to adapt to the patient (instead of the typical 18 hours). Indeed, the patient’s average on days 3-5 was a solid 142 mg/dl, well below goal. The team has since modified the algorithm to allow the system to begin adapting immediately once it first comes online.

Summer Camp Study Results, Days 2-5
(n=32 adolescents, 160 bionic pancreas days)


Bionic Pancreas

Supervised Camp Care




% CGM <60 mg/dl


% CGM <60 mg/dl



142 mg/dl


158 mg/dl


189 mg/dl

Projected A1c




  • The randomized, crossover Beacon Hill study compared five days on the bionic pancreas to five days of “usual care” (what a patient would normally do, though with the addition of blinded CGM). The study included 20 adult type 1 patients >21 years. The bionic pancreas mobile platform consisted of two Tandem t:slim pumps (insulin and glucagon), a Dexcom G4 Platinum sensor and transmitter, and an iPhone 4S controller. Patients had free run of a three-square mile area of the Boston peninsula. Point of care capillary blood glucose checks occurred during the day via 1:1 nursing. At night, patients slept in a hotel with venous blood glucose monitoring and 1:2 nursing. A total of 100 days on the bionic pancreas were accumulated. For more information, read our diaTribe test drive on the Beacon Hill study.
    • Similar to the summer camp study, the bionic pancreas improved mean blood glucose (159 to 133 mg/dl) and substantially reduced hypoglycemia (3.7% to 1.5% of time <60 mg/dl). Dr. Russell emphasized that the comparison to “usual care” was quite challenging, since 45% of patients wore their own CGM in addition to blinded CGM. In addition, 100% of patients were on insulin pumps in the control condition. This contrasts significantly with “real world” care of type 1 diabetes, where it is estimated that ~30% of patients are on pumps and ~10% are on CGM in the US.
    • All 20 patients in Beacon Hill had a mean glucose <154 mg/dl (the ADA goal of <7%) on the bionic pancreas. Most patients experienced a striking decline in mean glucose, with one patient going from an average of 215 mg/dl to <120 mg/dl on the bionic pancreas. Only one patient saw a rise in mean glucose after wearing the bionic pancreas (approximately +10 mg/dl), and in that case, the system reduced a high level of baseline hypoglycemia. Under usual care, there was also a wide dispersion in mean glucose between patients, with some having a mean under usual care of >210 mg/dl and others at <120 mg/dl. After wearing the bionic pancreas, mean glucose levels converged to 115-153 mg/dl.

Beacon Hill Study Results, Days 2-5
(n=20 adults, 100 bionic pancreas days)


Bionic Pancreas

Usual Care



% CGM <60 mg/dl


% CGM <60 mg/dl


133 mg/dl


159 mg/dl


Projected A1c



  • Dr. Russell emphasized the robustness of the bionic pancreas algorithm, which only requires weight for initialization, adapts over time, and does not mandate pre-meal priming boluses. Patients can optionally announce meals to the system, but they only enter qualitative information using a slider – is this “more,” “about the same,” or “less” than the amount of carbs that you typically eat? At ATTD 2014, Dr. Damiano called this, “Diabetes without numbers.” In Beacon Hill, patients could use the meal announcement feature if they desired, but were not reminded about it if they forgot. Only about 70% of the meals were actually announced to the system. As a reminder, the algorithm consists of three insulin controllers (basal, bolus, and meal priming) and one glucagon controller. The algorithm’s adaptive capabilities are described in El-Khatib et al., J Clin Endocrinol Metab 2014.

Questions and Answers

Dr. Roman Hovorka (University of Cambridge, UK): Wonderful stuff. Congrats on the publication – it’s brilliant for the field and for you as well. In the camp study, you compared closed-loop to standard treatment. Did standard treatment have real-time CGM?

A: In both studies, the bionic pancreas was compared to usual care for that patient. In the adult study, 45% of the time subjects were using their own unblinded CGM. All patients wore blinded CGM across the board. But if they used CGM, they were allowed to wear it during the usual care period of the study. Usual care in the camp study was much better than at home. But only about 9% of them used CGM during the camp usual care arm.

Dr. Hovorka: So some of the benefit is due to closed-loop, but also some could be due to adding CGM…?

A: It’s very important to highlight, “What are we aiming to find out?” You take a population of people with type 1 diabetes and provide them with this technology. I think this study underestimates that effect size. We know only about 9% of the US population uses CGM. It is possible that some of the achieved benefits are from using real-time CGM. On other hand, that would underestimate the effect, because CGM is not widely used.

Dr. Hovorka: You could look at it two ways. One should use the best possible treatment vs. the standard of care. Another is to use what people are using. The argument NICE would take is that you must go against the best possible treatment.

A: Granted, but it’s unlikely we’re going to convert everyone to CGM.

Dr. Bruce Buckingham (Stanford University, Stanford, CA): If you look at the adults, there was no remote monitoring in the control arm. There was a significant benefit in hypoglycemia. In the camp study, you remotely monitored, and the camp control group was very closely watched. You really set yourself up to a disadvantage in showing a benefit...

A: That’s right. And we did see a benefit. In the next study, monitoring will really be reduced – alerts will only trigger for hypoglycemia that persists for more than 15 minutes.

Dr. Irl Hirsch (University of Washington, Seattle, WA): You’ve been doing subcutaneous glucagon infusion. How often did you change the sites, and did you see any problems with skin reactions?

A: Glucagon was changed every day. As has been well pointed out in this session, current formulations of glucagon are not very stable. Many companies are working to develop stable glucagon, and some are quite far. We’re doing a clamp study with the Xeris glucagon right now. Preliminary data suggests it is equivalent in PK/PD in microdoses. There was no difference in the rate of skin reactions at infusion sites.

Four Weeks’ Home Use of Overnight Closed-Loop Insulin Delivery in Adults with Type 1 Diabetes: A Multicentre, Randomised, Crossover Study (233-OR)

Hood Thabit, MD (University of Cambridge, Cambridge, United Kingdom)

Dr. Hood Thabit shared notable results from a 28-day, crossover, home-use study of overnight closed-loop control in adults with type 1 diabetes (n=24). At baseline patients had mean age 43 years, A1c 8.1%, duration of diabetes 29 years, duration of pump use of six years, and BMI 26 kg/m2. Compared to the period with open-loop CGM/pump therapy, the overnight closed-loop period – including all data whether closed loop was turned on or not – included statistically significantly more overnight time in target 70-144 mg/dl (53% vs. 39%), lower overnight mean glucose (148 mg/dl vs. 162 mg/dl), less overnight time >144 mg/dl (44% vs. 57%), lower mean glucose at 7 am (130 vs. 158 mg/dl), a lower 24-hour mean glucose (157 mg/dl vs. 167 mg/dl), and more 24-hour time in target (66% vs. 59%). Dr. Thabit explained that the improved glycemic control was due to overnight insulin delivery that was significantly higher-dose (6.4 vs. 4.9 U/night) and more variable (SD 0.6 vs. 0.1 U); however, thanks to better glycemic control at the start of the day, total insulin dose was not significantly higher (34.5 vs. 35.4 U/day). Rates of nocturnal hypoglycemia <70 mg/dl were low (1.8% vs. 2.1%) and not significantly different between groups, due to optimization of open-loop therapy. Closed-loop control was interrupted for technical reasons on roughly 20% of nights; the main problem was a loss of connectivity with the pump (81%). Two severe hypoglycemic episodes occurred, both overnight during interruptions of closed-loop connectivity; both patients recovered fully. Shortly after this presentation, trial results were published (Thabit et al., Lancet Diabetes Endocrinol 2014).

  • This trial of overnight closed-loop control included adults (≥18 years) with type 1 diabetes who use insulin pumps (n=24). Patients were enrolled at three sites: Cambridge, Sheffield, and London. Mean baseline data were as follows: A1c 8.1%, age 43 years, diabetes duration 29 years, pump use duration 6.3 years, total insulin dose 0.5 U/kg/day, BMI 26 kg/m2. After a two-to-four week run-in period in which patients wore blinded CGM, sensor glucose data were downloaded to evaluate compliance and to optimize pump therapy. Patients then participated in a 28-day period of either 24-hour sensor-augmented pump use, or overnight closed-loop control with daytime sensor-augmented pump use. They then went through a four-week washout period and crossed over to the alternate condition. The first night of the overnight closed-loop experiment was spent at the clinical research center, for training and competency assessment with the closed-loop system. While at home patients had unsupervised access to a 24/7 support line. 
  • Closed-loop control was performed with the FlorenceD2 prototype system, which includes an Abbott FreeStyle Navigator 2 transmitter, Abbott FreeStyle Navigator 2 receiver, Dana Diabecare pump, and an ultraportable PC that hosts a model-predictive control (MPC) algorithm and communicates with the pump wirelessly.
  • During the period between midnight and 7 a.m., sensor glucose values fell in the target range (70-144 mg/dl) significantly more often with overnight closed-loop control than sensor-augmented pumping (53% vs. 39%, p<0.001).  Rates of overnight hyperglycemia >144 mg/dl were also significantly reduced with closed-loop control. However, overnight hypoglycemia was quite rare in both groups (mean <10 minutes per night), and the rates were not significantly different between groups. By way of partial explanation, Dr. Thabit said that each patient in the open-loop group had their overnight basal rate was “optimized” after the run-in period. Basal rate was then adjusted as needed in the weekly conversations with study clinicians. 

OVERNIGHT (00:00 – 07:00)

Overnight Closed-Loop

Sensor-Augmented Pump


% Time 70-144 mg/dl




% Time 70-180 mg/dl




% Time >144 mg/dl




% Time <70 mg/dl




% Time <50 mg/dl




  • Overnight closed-loop led to lower mean sensor glucose in the overnight period (148 vs. 162 mg/dl, p<0.05), lower mean glucose at 7 a.m. (130 vs. 158 mg/dl), and a trend toward less between-night coefficient of variation (CV). Dr. Thabit noted that within-night CV was actually higher with closed-loop control, because patients who began the night hyperglycemic were more consistently brought to a lower range by the closed-loop system. Sensor glucose fell below 63 mg/dl on 36 nights in the closed-loop period and 58 nights in the open-loop period; this difference did not reach statistical significance (p=0.18). 

OVERNIGHT (00:00 – 07:00)

Overnight Closed-Loop

Sensor-Augmented Pump


Mean glucose (mg/dl)




SD glucose (mg/dl)




Within-night CV glucose (%)




Between-night CV glucose (%)




AUC <63 mg/dl (mmol/l*min)




Glucose at 7 am (mg/dl)




SD = Standard Deviation; CV = Coefficient of Variation; AUC = area under the curve

  • The benefits of overnight closed-loop control extended beyond the night, such that 24-mean glucose and 24-hour time in target were improved relative to sensor-augmented pumping. Dr. Thabit presented a modal-day chart showing the median and inter-quartile range of sensor glucose for each treatment condition. Median sensor glucose was lower with overnight closed-loop control through the morning and into the mid-afternoon, even though these patients used open-loop control during the day. We assume that the closed-loop group was able to achieve better daytime control because they were more likely to wake up in their target range.   


Overnight Closed-Loop

Sensor-Augmented Pump


% Time 70-180 mg/dl




Mean glucose (mg/dl)




Change in A1c from baseline (7.9%)


No change


  • The closed-loop controller delivered overnight insulin doses that were higher and more variable than with sensor-augmented pumping. Dr. Hood observed that the system increased insulin variability in order to reduce glycemic variability – an equation often described in talks on overnight closed-loop control. 


Overnight Closed-Loop

Sensor-Augmented Pump


Mean insulin dose, overnight (U)




SD insulin dose, overnight




Mean insulin dose, 24-hour (U)




  • Closed-loop control was used on 555 nights (86% of the full intervention period) for a median of 8.3 hours per night. The control system was turned on and off at median times of 10:52 pm and 7:23 am, respectively.
  • The total number of interruptions to closed-loop control was 112 (an average of once every five nights). These interruptions were due mainly to lack of connectivity between the control device and the insulin pump (61%). Other causes of interrupted closed-loop control were inability to start the closed-loop cycle within 30 minutes (19%), changes in pump settings by the user (10%), unavailability of sensor data (6%), operating system malfunction (3%), and error in the control device’s software system (1%). 
  • Two episodes of severe hypoglycemia occurred during the study, both at night and both when closed-loop control was interrupted. Both patients had a history of hypoglycemia unawareness. One of the episodes was attributed to a suspected overbolus of insulin while the pump was being primed. The cause of the other episode was not clear, but risk for hypoglycemia had been raised by increased physical activity during the day. Both patients recovered fully. Both patients also reduced their participation in the closed-loop experiment to only 14 days, with advice from the Steering Committee. (The other 14 days were not counted toward the total number of closed-loop days. Thus the total number of closed-loop days in the intent-to-treat analysis was 2*14 + 22*28 = 644). No episodes of hyperglycemia with ketosis were observed during the study.

Questions and Answers

Dr. Yogish Kudva (Mayo Clinic, Rochester, MN): There were 112 interruptions of closed-loop control over 555 days. What was the duration of interruptions?

A: I don’t have those data, sorry.

Q: What did it mean to optimize the non-closed-loop pump group’s overnight basal rate before starting the study?

A: Following the run-in period, the CGM was downloaded. We tried to optimize control as much as possible. Participants were also in conversation with clinicians once a week during the study. Both sides tried to optimize as much as possible, within limits.

Q: These results looked better than optimized, there was so little hypoglycemia with sensor-augmented pumping.

A: This was one reason we couldn’t see a difference. The baseline rate of hypo was very low, as I said – less than 10 minutes per night. To see a difference from this baseline, one would either need to recruit more hypoglycemia-prone patients or get a higher number of patients.

Dr. Nancy Bohannon (San Francisco, CA): The two severe episodes you had were when closed loop was turned off. Did these occur during the day, or during the night?

A: Both occurred at night. The closed-loop system wasn’t functioning; it wasn’t because the patient turned it off. There was a loss of connectivity, so the system reverted to the patient’s own basal pump setting. We wanted patients to use closed-loop control every night it was available, but sometimes they wouldn’t. If CGM was available, they used it close to 92% of the time; only 8% of the time they didn’t want to use it.

Day and Night Home Closed-Loop Insulin Delivery in Adults with Type 1 Diabetes: Three-Centre, Randomised, Crossover Study (235-OR)

Lalantha Leelarathna, MD (University of Cambridge, Cambridge, UK)

Dr. Leelarathna shared exciting results from a feasibility study of a closed loop system under free-living home conditions for seven days and nights in adults with type 1 diabetes (n=17). Patients were randomized to receive either the FlorenceD2 closed loop system or sensor augmented pump therapy. After seven days of treatment, patients in the closed loop arm spent significantly more time in the target range (defined as 70-180 mg/dl) compared to patients on sensor augment pump therapy (75% vs. 62%). The closed loop system also outperformed sensor augmented pump therapy on multiple secondary outcomes, including mean glucose (146 mg/dl vs. 158 mg/dl) and standard deviation of glucose (52 mg/dl vs. 59 mg/dl), without any significant difference in time spent in hypoglycemia. Operationally, however, Dr. Leelarathna emphasized that the portability and connectivity of the closed loop system needs to be improved going forward. Nevertheless, this study validated the feasibility of day and night closed loop and supported the initiation of the AP@home04 study, which includes three months of day and night in 30 adults ( currently indicates an anticipated completion date in the second half of 2015).

  • The objective of this study was to evaluate the feasibility of day and night closed loop system under free-living home conditions for seven days in adults with type 1 diabetes. The trial was conducted under the AP@home consortium and recruited 17 patients from Germany, UK, and Austria. Type 1 diabetes patients were eligible for the study if they were on insulin pump therapy, had an A1c <10%, and did not have any significant comorbidities or hypoglycemia unawareness. Patients were randomized either to the FlorenceD2 closed loop system or an open loop treatment (consisting of insulin pump therapy combined with real-time CGM). The FlorenceD2 system contains three components: the Dana R insulin pump, the Navigator II receiver, and the control algorithm device. Each treatment phase included 23 hours in a clinical research facility followed by seven days at home. After a 1-3 week washout period, patients were crossed-over to receive the other treatment. During the research facility phase of the study, patients were trained on using the closed loop system; however, after the 23-hour inpatient stay, participants went home and used the system without any supervision – they could consume any meals of their own choice. Patients were encouraged to engage in moderate physical activity, but were advised to avoid strenuous activity or driving.
  • The study recruited 17 adult patients with type 1 diabetes. The participants included 10 males and 7 females, with an average age of 34 years and an average duration of diabetes of 19 years. At baseline, patients were reasonably well controlled, with a mean A1c of 7.6%.
  • Patients on closed loop treatment experienced significantly greater time in target range (defined as 70-180 mg/dl [3.9-10 mmol/l])] during the seven day home phase, compared to the open loop arm (75% vs. 62%). When using YSI reference glucose values, the target zone remained consistent (74% vs. 61%). The total daily dose of insulin infusion was lower in the closed loop group, but did not reach statistical significance (total basal insulin was slightly higher with closed loop, but boluses were significantly lower).
  • There were a total of 194 operational interruptions, translating to an interruption event every 12 hours (out of 2,333 total hours of closed loop operation in this study). The two most common reasons for these events were lack of pump connectivity and CGM unavailability. One severe hypoglycemia episode occurred in the closed loop arm because the sensor stopped working and the patient administered two manual boluses. Dr. Leelarathna highlighted this as an opportunity for improvement and mentioned that the next study (AP@home04) has moved to a mobile home platform with wireless communication between devices.

Questions and Answers

Dr. Irl Hirsch (University of Washington, Seattle, WA): Regarding the differences in bolus insulin on closed loop, what can I take away from that? Does that mean we are we doing too much bolus insulin or were patient eating differently on closed loop?

A: One of the reasons was that glucose was more in target on closed loop. Therefore, the correction bolus was lower in patients on closed loop.

Dr. Hirsch: When you say bolus, you’re including the correction as a bolus?

A: Correct.

Dr. Hirsch: I would suggest for reporting in the future, we need to separate bolus from correction from basal.

Ms. Arlene Pinkos (FDA, Silver Spring, MD): Can you confirm if the subjects kept diaries during the study? If so, were you able to trace the glucose excursion on closed loop to specific activities?

A: We did give patients diaries, but unfortunately very few patients followed our advice. So, unfortunately I can’t answer that. Anecdotally, I would say the excursions were due to a miscalculation of carbohydrate bolus or exercise.

Hypoglycemia Reduction Capability and Insulin Dosing Behavior of a Predictive Controller (232-OR)

Daniel Finan, PhD (Animas Corporation, Westchester, PA)

Dr. Daniel Finan presented an update on Animas’ efforts to automate insulin delivery – the company has not been able to move quickly to larger studies or a commercial product, which is perhaps understandable given the challenges in the LifeScan/Animas business. This non-randomized, in-clinic, uncontrolled feasibility study examined 24-hour overnight use of a predictive low glucose suspend algorithm running on a laptop computer with the Dexcom G4 Platinum CGM, OneTouch Ping pump, Sansum APS system. The study was not powered to assess glucose outcomes and merely tested the algorithm’s insulin dosing decisions based on three different “aggressiveness factors” in 12 patients – conservative (n=4), medium (n=4), and aggressive (n=4). [We’d note that Animas reported on a slightly more ambitious hypoglycemia-hyperglycemia minimizer at ADA 2013.] The conservative predictive suspend approach activated only 7% of the time, and on average only reduced basal insulin by 4%. By contrast, the medium and aggressive approaches activated 21% and 23% of the time, on average reducing basal insulin dose by 19% and 21%, respectively. Animas was “encouraged” by the CGM results in this short and small study, as a median 0% of the time was spent <70 mg/dl and 72% of the 24-hour period was spent in the range 70-180 mg/dl. Further work will include appropriately powered clinical studies, developing this “study tool into a target product,” mapping out use cases, and validating human factors. There is clearly a long road ahead for this to turn into a commercial product, though it’s encouraging to see that Animas is still pursuing it.

Questions and Answers

Dr. Yogish Kudva (Mayo Clinic, Rochester, MN): What was time horizon you looked at for prediction? And was there an IOB setting you used, or did you just use that which the patient came into study with? Why did you choose to show just one standard deviation?

A: We look pretty far into the future – eight hours. That allows us to get an idea of all the dynamics that will take place in the near future. We want to map out our model’s best guess on what glucose will do. Regarding insulin on board, it’s a special IOB calculation that is done within this predictive model. I could have showed two standard deviations, and that would certainly show that some patients did go below 70. But this was a 12-patient study. I know that the data is so variable in this disease, I thought I would keep simple.

Dr. Roman Hovorka (University of Cambridge, UK): It’s good to see this project moving on. I take issue with the study design – the number of subjects, non-randomized, non-controlled. There are problems taking conclusions from the study design. Hypoglycemia might just happen by chance. I’m sure your team thought about it...

A: Point well taken. It’s tricky in these phase of feasibility studies. You don’t want to invest too much of anything into the development effort, so you have to do these small studies. It was not statistically powered. It’s a tricky thing. We thought this was our best way to do the diligence.

Dr. Hovorka: You mentioned a 20% insulin reduction. Over how long of a period was that?

A: We collected all the data from the four patients at one of the aggressiveness factor values. It’s across the whole study, all in the same bucket.

Effect of Glucose Concentrations on Hepatic Glucagon Sensitivity and Glucagon Clearance in Type 1 Diabetes (234-OR)

Ling Hinshaw, MD, PhD (Mayo Clinic, Rochester, MN)

Dr. Ling Hinshaw shared results from a physiology study demonstrating that hepatic glucagon sensitivity does not vary with the prevailing glucose concentration in patients with type 1 diabetes. The rationale for the study stems from bi-hormonal artificial pancreas work – does a glucagon controller need to account for increased endogenous glucagon production (EGP) when patients are hypoglycemic? In this study, 27 people with type 1 diabetes were randomized to receive either a euglycemic or a hypoglycemic clamp overnight on three separate occasions, and their response to low, medium, and high doses of glucagon (0.65, 1.5, and 3.0 ng/kg/min, respectively) was measured. There was no statistical difference between the euglycemic and hypoglycemic clamp groups in the level of endogenous glucose produced in response to the three doses of infused glucagon, suggesting that the hypoglycemic subjects were not more sensitive as originally hypothesized. The researchers also measured glucagon clearance, which increased linearly with the dose of glucagon in both groups. This means that although the glucagon controller in an artificial pancreas would not need to vary its dose based on the prevailing glucose concentration, it would need to account for increased glucagon clearance at higher doses.

  • The goal of this study was to determine the dose response of endogenous glucagon production (EGP) to glucagon in patients with type 1 diabetes under hypoglycemic and euglycemic conditions, and to measure the rate of glucagon clearance at different concentrations. The relationship between hepatic glucagon sensitivity and prevailing blood glucose concentrations was previously unknown and would have important implications for the design of bi-hormonal artificial pancreas control algorithms. The authors of this study hypothesized that hepatic glucagon sensitivity would be increased in hypoglycemic subjects compared to euglycemic subjects based on evidence from animal studies.
  • The study involved 27 subjects with type 1 diabetes who were randomly assigned to either a hypoglycemic or a euglycemic clamp overnight for a total of three nights. The average duration of diabetes for all participants was approximately 20 years. The hypoglycemic clamp group (n=14) had an average age of 45 years, an average BMI of 27 kg/m2, and an average A1c of 7.5%, and their plasma glucose was maintained at 59 mg/dl. The euglycemic clamp group (n=13) had an average age of 38 years, an average BMI of 28 kg/m2, and an average A1c of 7.4%, and their plasma glucose was maintained at 92 mg/dl.
  • EGP was measured in all subjects in response to infusions of low (0.65 ng/kg/min), medium (1.5 ng/kg/min), and high doses (3.0 ng/kg/min) of glucagon, and both groups displayed similar dose responses – in other words, hepatic glucagon sensitivity was statistically comparable during hypoglycemia and euglycemia. The infusions were given on separate days in a random order, and plasma insulin levels were held constant with a low-dose insulin infusion. All patients received an infusion of [3-3H] glucose to measure EGP. There was no significant difference in the dose response curve to glucagon between the hypoglycemic and euglycemic groups. In the hypoglycemic group, EGP was 0.5 mmol/kg/min with the low glucagon dose, 12.5 mmol/kg/min with the medium dose, and 14.4 mmol/kg/min with the high dose. In the euglycemic group, EGP was 0.6 mmol/kg/min with the low dose, 11.9 mmol/kg/min with the medium dose, and 14.6 mmol/kg/min with the high dose.
  • Glucagon clearance was also similar between the two groups, though it increased in a linear fashion as the dose of infused glucagon increased. This implies that a glucagon controller in a future artificial pancreas would need to account for the increased clearance of glucagon at higher doses. Dr. Hinshaw mentioned in passing that the researchers also measured plasma epinephrine concentrations, which were higher and more variable in the hypoglycemic group.

Questions and Answers

Q: Congratulations! I was well aware of the paper in dogs showing a bigger response from the liver to glucagon in hypoglycemia and I always assumed it would happen in people too. We did a similar study in which we varied the insulin level, not glucose. Everything was done at euglycemia and we saw a similar dose response curve. At high insulin levels, there was a decreased response to a higher glucagon dose. How high was the insulin level here?

A: The insulin level in this study was 200 pM, which is a middle to high physiological concentration in someone with type 1 diabetes. To measure glucagon response, we gave three doses, achieved three glucagon concentrations, and the dose response was not different between hypoglycemia and euglycemia.

Q: Very beautiful study. That epinephrine response is common with long-duration type 1 diabetes, but I got the impression it was different at different doses. Is that correct?

A: Many studies have observed a variable and higher epinephrine concentration in patients with type 1 diabetes, which was one reason we hypothesized the glucagon response might be higher at hypoglycemia, but the results show no glucagon sensitivity difference between hypoglycemia and euglycemia, and also we think the primary regulator of EGP is glucagon, not epinephrine.

Q: Right, but my question was whether different glucagon doses affected epinephrine.

A: We didn’t look at the data for that.

Q: Your data on glucagon clearance showed that it increased with different glucagon doses, which is very unusual. Can you speculate why?

A: It was reported in the 1980s that glucagon clearance differed between hypoglycemia and euglycemia. Our study showed different results because of the method of the glucagon assay and a different insulin/glucagon ratio. Our study was similar to a recent report saying that the rate of clearance did not differ between euglycemia and hypoglycemia.

Q: But it differed at different glucagon levels. Isn’t that unusual?

A: I don’t have an answer for that now. We can look at the data and try to answer it later.

Hypoglycemia Reduction and Changes in A1c in the ASPIRE In-Home Study (231-OR)

Timothy Bailey, MD (AMCR Institute, Escondido, CA)

Dr. Timothy Bailey presented a subgroup analysis of the ASPIRE in-home study, a randomized controlled trial of the Medtronic MiniMed 530G pump with vs. without its threshold suspend feature enabled (n=121 vs. 126). (For our primary coverage of the ASPIRE in-home study, see our ADA 2013 coverage or the NEJM paper.) In this analysis Dr. Bailey and colleagues stratified both groups according to whether A1c decreased by >0.3% (n=25 with threshold suspend, 28 without), remained within 0.3% of baseline (n=65, 77), or increased by >0.3% (n=26, 19). Mean end-of-study A1c values in the “decreased,” “stable,” and “increased” subgroups were 7.1%, 7.1%, and 7.7%, respectively. The researchers also looked at nocturnal hypoglycemia, defined as ≥20 minutes of sensor glucose ≤65 mg/dl with no evidence of user-pump interaction, between 10 p.m. and 8 a.m. Patients with threshold-suspend enabled had 1.5 nocturnal hypoglycemic events per week regardless of A1c-change subgroup; this was lower than the corresponding rates in the “decreased” or “stable” A1c subgroups without threshold suspend. In all three threshold-suspend subgroups, patients experienced several benefits in nocturnal-hypoglycemia statistics: higher nadir sensor-glucose value, shorter event duration, and smaller area under the curve below 65 mg/dl. Also of note, sensor-glucose coefficient of variation (a measure of glycemic variability) decreased by a statistically significant margin in patients whose A1c decreased or was stable.

Questions and Answers

Dr. Irl Hirsch (University of Washington, Seattle, WA): Given what we’ve learned about hypoglycemia in patients older than 50 and 60 years old in the T1D Exchange, did you look at hypoglycemia and coefficient of variation in these groups?

A: We should look at that, but I think the numbers will be small. We didn’t have that many older folks in study.

Q: In the group with increased A1c, could you identify them up front? Are there any characteristics to define this group that would not benefit from threshold suspend?

A: In some patients, A1c went up in trial. Obviously this happened in both groups; it wasn’t a feature of threshold suspend. We haven’t looked yet at what would predict failure.

Dr. Hirsch: You didn’t mention how much A1c went up in the increased-A1c group or down in the decreased-A1c group. What was the baseline? What was the mean of the increase or decrease?

A: Baseline A1c was 7.3% in the threshold suspend group and 7.2% in the control group. I don’t have the mean values of change for the groups that increased or decreased by more than 0.3%.

Q: Could you extrapolate certain types of patients in whom to use this technology, or do you think there was no relationship between the technology and A1c?

A: I wasn’t sure I would see significant effect in all the A1c groups. I was surprised to see benefits. I was also surprised to see how much hypoglycemia is going on in our patients. A lot goes undetected, especially at night. The technology is potentially a benefit to all patients; the question is how much benefit for how much money. Coverage for this device represents an important step to increase access to a device that helps protect patients from hypoglycemia.

The Quest for a Pumpable, Liquid Glucagon: A Novel Use for Ferulic Acid (236-OR)

Parkash Bakhtiani, MD (Oregon Health and Science University, Portland, OR)

Dr. Parkash Bakhtiani presented a series of promising preclinical experiments with a new formulation of glucagon, which uses ferulic acid as a stabilizing agent. (Ferulic acid is a highly stable phenolic compound found naturally in many foods.) Dr. Bakhtiani reviewed that the current formulation of synthetic human glucagon quickly degrades and forms insoluble fibrils, especially at high temperatures. However, when a ferulic acid formulation of glucagon (FAFG) was aged for seven days at body temperature (37°), no fibrillation was seen, bioreactivity was maintained, and degradation was low. (These attributes were assessed by transmission electron microscopy, a glucagon receptor / protein kinase A assay, and reverse-phase high-performance liquid chromatography, respectively.) Dr. Bakhtiani also described a pharmacodynamics experiment in Yorkshire swine treated with octreotide (to suppress their native glucagon secretion). Aged FAFG was compared to fresh FAFG and fresh regular human glucagon (n=8 for each condition); the glycemic rises over baseline were 91±12 mg/dl, 93±23 mg/dl, and 84±30, respectively – not statistically significantly different from each other. Dr. Bakhtiani looked forward to research on this new formulation’s pharmacokinetics, shelf-life studies at various temperatures, leachability/extractability, toxicity, dose-response curves, and clinical safety and effectiveness.

  • Dr. Bakhtiani briefly reviewed the challenges associated with glucagon in the setting of bihormonal closed-loop control. He reminded the audience that liquid glucagon is chemically unstable; it quickly degrades and also forms insoluble aggregates. Dr. Bakhtiani noted that even when a liquid-stable formulation becomes available, glucagon rescue therapy could fail if the CGM overestimates a patient’s glucose levels or if insulin levels are so high that hepatic glucose output is suppressed.   

Questions and Answers

Dr. Yogish Kudva (Mayo Clinic, Rochester, MN): Ferulic acid has got a phenol group. Is there any concern that it might cause a false signal on CGM, as acetaminophen does? Or is the concentration so low that this is not a concern?

A: It’s possible. I think that’s a great point. We should look at that.

Q: Is ferulic acid used as a stabilizer for human-use products already?

A: It is used as a therapy for some conditions, and it is present in various foods and beverages.

ADA Diabetes Care Symposium – New Drug Therapies, Innovative Management Strategies, and Novel Drug Targets

Safety and Efficacy of Outpatient Closed-Loop Control – Results from Randomized Crossover Trials of a Wearable Artificial Pancreas

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

Chosen to speak for the second year in a row at the prestigious ADA Diabetes Care Symposium, Dr. Boris Kovatchev discussed recent and upcoming closed-loop studies involving the University of Virginia’s Diabetes Assistant (DiAs) platform. His main focus was a 40-hour crossover study (n=20) in which partially closed-loop control reduced low blood glucose index (effect size of 0.64) and time spent below 70 mg/dl (1.25% in open loop vs. 0.64% in closed loop). The downsides of this hypoglycemia reduction were slightly higher mean glucose (152 mg/dl in open loop vs. 161 mg/dl in closed loop) and less time in target (70.7% vs. 66.1%), though these results were not statistically significant. At a higher-level view, Dr. Kovatchev emphasized that “AP systems are mobile medical networks,” and he explained how the DiAs system’s “network” is distributed among the CGM, the pump, and a smartphone, so that if any component fails, the patient can remain safe. To the question of “are we there yet?” with the artificial pancreas, Dr. Kovatchev said that the control algorithm “is probably there yet” and that “there are several viable algorithms around the world.” He said that today the main limiting factors for the artificial pancreas are hardware and clinical research – especially the latter. Looking to the next year or so, Dr. Kovatchev whetted our appetites for a month-long study of round-the-clock closed-loop control at home and a weeklong study of round-the-clock closed-loop control at a diabetes camp.

  • Emphasizing that “AP Systems ARE mobile medical networks,” Dr. Kovatchev explained that his group uses a distributed-computing model designed for redundancy in case one or more components fail. The insulin pump contains safety algorithms, and the smartphone “hub” contains the Diabetes Assistant (DiAs) graphical user interface and the main control algorithm; the pump and hub communicate with each other (and the CGM) on a Medical Android network. Also, a cloud-based component can run various functions such as GPS location, remote data transmission, and alert calls.
  • The main subject of Dr. Kovatchev’s talk was a 40-hour crossover trial of closed-loop control in a semi-outpatient setting, on which Dr. Kovatchev had shared background and top-line results at ADA 2013’s ADA Diabetes Care Symposium. The multi-site trial enrolled five patients at each of four sites: University of Virginia, Sansum Diabetes Research Institute, University of Padova, and University of Montpellier (n=20 total). Each session included a 45-minute walk through town and mandatory restaurant dinner; meal size was not restricted as long as patients estimated carbohydrate count and announced the meal to the closed-loop system. Alcohol consumption was allowed. Patients performed fingerstick blood glucose tests before and after meals and at bedtime, but not overnight. 
    • The trial’s primary outcome was reduction in hypoglycemia as measured by low blood glucose index; this goal was met with a statistically significant effect size of 0.64. Compared to open-loop control, closed-loop control led to lower percentage of time with glucose below 70 mg/dl (1.25% vs. 0.7%), lower number of hypoglycemic episodes requiring treatment (2.39 vs. 1.22 episodes per person per session), and lower amount of carbohydrates required for treatment (39.7 vs. 17.6 g per person per session). All of these differences were statistically significant.
    • The chief downsides of closed-loop control were that, compared to open-loop control, patients spent a lower percentage of time in target (70.7% vs. 66.1%; p>0.1) and had a higher mean glucose (152 vs. 161 mg/dl). The p-value for the mean glucose difference was 0.04, but the result was not considered statistically significant because it did not meet the threshold of p=0.01 (which was used to correct for multiple comparisons, because conducting multiple comparisons increases the chances of differences arising by chance alone).
  • Dr. Kovatchev mentioned three other closed-loop trials using DiAs that have been published in the past year, including two being presented at ADA 2014 (late-breaking abstracts 104 and 106).
    • First he reviewed a study of overnight closed-loop control at diabetes camp (n = 54 nights of closed-loop control vs. 52 nights of open-lop sensor-augmented pumping). Dr. Kovatchev described that overnight closed-loop control “virtually eliminated” nocturnal hypoglycemia while also decreasing time spent above 180 mg/dl or above 250 mg/dl (Buckingham et al., Diabetes Care 2014). 
    • Dr. Kovatchev then showed results from an eight-hour, crossover trial of daytime closed-loop control in teenagers (n=16), to see how the system might handle a missed meal bolus. The adolescents ate a 30-g snack at 9 a.m. with no bolus, and then they consumed an “underbolused” lunch. Patients in the closed-loop condition had statistically significantly lower postprandial excursions compared to open-loop control. This research was presented at ADA 2014 as 106-LB.
    • The other recent trial Dr. Kovatchev mentioned was a five-day study of overnight closed-loop control (50 vs. 50 patient-nights). He said that the overnight controller improved mean nighttime glucose to 139 mg/dl (vs. 168 mg/dl with open loop), increased nighttime time in target to 85% (vs. 60%), and reduced nocturnal hypoglycemia to 0.6% (vs. 2.1%). The system even improved glucose control on the following day, Dr. Kovatchev noted. This research was presented at ADA 2014 as 104-LB.
    • Dr. Kovatchev briefly previewed four closed-loop trials scheduled for this year. A one-month multi-site trial of round-the-clock control at home funded by JDRF; a week-long summer-camp study of round-the-clock control funded by the Helmsley Charitable Trust; a five-day multi-site trial of bedside closed-loop control funded by the NIH; and a study using closed-loop control to reduce hypoglycemia unawareness funded by the NIH.

Best of Diabetes Care 2013-2014 – Artificial Pancreas Developments

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

Dr. William Tamborlane provided a concise and organized overview of the past year in artificial pancreas research, focusing on the safety of low glucose/predictive suspend and overnight closed-loop. Regarding the latter, he noted that “the holy grail of unsupervised full closed-loop control may be just around the corner, given the results from Dr. Hovorka’s recent study. Dr. Tamborlane’s presentation covered four main studies: Beck et al., Diabetes Care 2014 and Sherr et al., Diabetes Care 2014 (both demonstrating the safety of low glucose/predictive suspend); Hovorka et al., Diabetes Care 2014 (demonstrating the safety of unsupervised overnight closed loop); and Buckingham et al., Diabetes Care 2013 (no impact of short-term hybrid closed loop immediately following diagnosis). Though closed-loop studies have demonstrated feasibility in the inpatient setting, Dr. Tamborlane explained that there are still some key obstacles to broad outpatient use: making the device user interfaces easy to use and establishing redundant safeguards to minimize the risk of excessive insulin administration.

  • Why do we need an artificial pancreas? First, too many type 1 diabetes patients fail to achieve A1c goal – Dr. Tamborlane highlighted the latest data from T1D Exchange, which suggests that adolescents have an average A1c of 9.0%! Second, he noted that severe hypoglycemia remains an ever-present danger, and based on T1D Exchange data, patients with a high A1c are not at reduced risk. Third, the burden of care is “extremely high” and has increased with the translation of new diabetes technologies into clinical practice.
  • Dr. Tamborlane described the iterative progression to an artificial pancreas, starting with low glucose suspend and predictive low glucose suspend, followed by nighttime closed-loop + daytime open-loop – the latter has emerged as a debate in the field, as some believe the regulatory path is easier for a 24-hour automated system (daytime treat-to-range + nighttime treat-to-target) vs. an overnight-only system. Said Dr. Tamborlane, “If you told a parent that their child could go to bed and reliably wake up at 120 mg/dl with no hypoglycemia, parents would take this is a moment.”
  • Dr. Tamborlane covered two papers relating to the safety of low glucose/predictive low glucose suspend:
    • Beck et al., Diabetes Care 2014 – Frequency of Morning Ketosis After Overnight Insulin Suspension Using an Automated Nocturnal Predictive Low Glucose Suspend System.  The study concluded that two-hour pump suspends are safe and won’t result in DKA or excessively dangerous ketone levels.
    • Sherr et al., Diabetes Care 2014Safety of Nighttime 2-Hour Suspension of Basal Insulin in Pump-Treated Type 1 Diabetes Even in the Absence of Low Glucose. Dr. Tamborlane noted that one of the FDA’s main concerns with the Veo/MiniMed 530G was the safety of two-hour suspends when the CGM was reading inaccurately low. This clever study had patients program a two-hour zero basal rate on random nights, regardless of the current glucose level. The study concluded that systems that suspend basal insulin for two hours are safe and do not lead to clinically significant ketonemia even if the blood glucose level is elevated at the time of the suspension.
  • Hovorka et al., Diabetes Care 2014 examined overnight closed-loop under free-living conditions in 16 young people with type 1 diabetes. Dr. Tamborlane highlighted that the study was done in the “real world” and demonstrated the efficacy and safety of nighttime closed loop + daytime open loop. Indeed, overnight closed loop increased time in zone by a median 15% and reduced mean overnight glucose by a mean of 14 mg/dl. Said Dr. Tamborlane, “The holy grail of unsupervised fully closed-loop control (i.e., overnight) may be just around the corner.”
  • Buckingham et al., Diabetes Care 2013Effectiveness of early intensive therapy on beta-cell preservation in type 1 diabetes. This ambitious study examined whether three to five days of inpatient hybrid closed loop at the time of diagnosis could preserve C-peptide one year later. There was “absolutely no difference” in A1c, CGM, or the rate of C-peptide decline between the intervention and control groups. This was a major disappointment when these results were first shared, as many had high hopes for this study. Dr. Tamborlane commented that in new onset type 1 diabetes “it appears we’ve already achieved about as much as we can achieve in slowing the loss of residual beta cell function by maintaining strict glycemic control,” regardless of using closed loop or open loop.  

Symposium: Closed-Loop Insulin Delivery — One Step at a Time (Sponsored by The Helmsley Charitable Trust)

Proof-of-Concept Trials

Stuart Weinzimer, MD (Yale University School of Medicine, New Haven, CT)

Dr. Stuart Weinzimer provided an in-depth review of recent advances in artificial pancreas research over the past two years, focusing on predictive low glucose suspend, hybrid closed loop, and full closed loop. While he spent most of his time reviewing early feasibility data, he emphasized that these trials will only provide preliminary safety and effectiveness information, but will lay the groundwork for more rigorous transitional (and eventually pivotal) clinical trials. Given the pace of research over the past few years, Dr. Weinzimer was encouraged and optimistic that the “future will show great promise” for closed loop.

  • Dr. Weinzimer began by reviewing the types of closed loop studies and the caveats of comparing data across trials. In general, closed loop studies fall into three categories, starting with small feasibility studies (to demonstrate preliminary safety and effectiveness), followed by transitional studies (with a greater number of patients and/or a greater duration), followed by pivotal studies, which will be designed for regulatory approval. Given the variation in trial design, Dr. Weinzimer cautioned against comparisons across trials without taking into account key factors, such as the presence/absence of controls, age of patients, hybrid vs. full closed loop use, size of meals, glycemic targets, and the treatment/definition of hypoglycemia.
  • There has been considerable progress in predictive low glucose suspend. Dr. Weinzimer focused on two major studies published in 2014: Danne et al., Diab Technol Ther 2014 and Maahs et al., Diab Care 2014. The Danne et al., study induced hypoglycemia with an exercise regimen in type 1 diabetes patients (n=16). The system shuts off insulin delivery when hypoglycemia is predicted to occur, a feature that averted actual hypoglycemia in 13 patients. Dr. Weinzimer noted that larger studies of predictive low glucose suspend are currently being planned and conducted. In the second study (Maahs et al., Diab Care 2014), the predictive low glucose suspend was shown to decrease both the percentage of nights with hypoglycemia (across all definitions) as well as the duration of hypoglycemia.
  • Dr. Weinzimer then reviewed overnight closed loop trials, which represent the next level of automation, after the predictive low glucose suspend. The longest duration studies of the closed loop come from the Cambridge group (Hovorka Diab Care 2014), which published last month the results of a three week study in 16 adolescents – this study showed a significant reduction in glucose variability overnight and a significant improvement in glucose levels within target range (70% of glucose levels within range). Dr. Weinzimer also reviewed the results of the DREAM project collaboration, which studied an overnight closed loop in 56 children at a diabetes camp. While he characterized this study as a feasibility study due its short one-night duration, he did note the “impressive” number of patients included in the study. This consortium is moving to in-home studies; recently published data demonstrate improvements in time in target, glucose variability, and exposure to hypoglycemia in the home environment with four consecutive nights (Nimri et al., Pediatr Diab 2014).
  • Finally, Dr. Weinzimer briefly touched on a few recently published studies of fully closed loop systems. In one 24-hour trial (Harvey et al., Diab Technol Ther 2014), a fully closed loop system was studied without manual meal boluses and no announcements to the patient (n=12). Dr. Weinzimer showed patient glucose profiles to demonstrate the slight increase in glucose excursions that would be expected without announcements. He also shared a study from his own group (Weinzimer et al., Diab Care 2012), in which a fully closed loop was tested in eight patients. After 48 hours of use, 71% of glucose levels were in target zone with no incidences of hypoglycemia.

Special Meeting: JDRF/NIH Artificial Pancreas Evening

A who’s who of closed-loop researchers, industry, non-profit organizations, and patient advocates gathered at the annual JDRF/NIH Closed-Loop Research Meeting on Sunday night of ADA 2014. This engaging evening featured a presentation from JDRF’s Dr. Aaron Kowalski on the past year of closed-loop research, followed by three industry perspectives (Medtronic, Animas, Dose Safety), and a panel that included researchers Drs. Stacey Anderson, Bruce Buckingham, Roman Hovorka, Moshe Phillip, and type 1/experienced closed-loop patients Ms. Kelly Close, Tia Geri, and Willa Spalter.

Brief Overview of AP Highlights from the Past Year

Aaron Kowalski, PhD (JDRF, New York, NY

“We still need better tools to treat patients with diabetes...someone on my airplane had a full-blown hypoglycemic seizure on my flight out here,” said Dr. Aaron Kowalski in his opening remarks to the JDRF/NIH closed-loop night. He provided a quick review of the past year of artificial pancreas (AP) research, noting that there are currently outpatient trials running in every bucket of the six-step JDRF roadmap. Notably, since ADA 2013, there were more than 14 approvals of new or significantly modified studies by FDA, MHRA, and other regulatory bodies. In addition, the past year saw over 34 peer-reviewed manuscripts and abstracts related to closing the loop. Dr. Kowalski rapidly highlighted recent work at more than 13 artificial pancreas research institutions around the world (see below). Similar to his comments at ATTD 2014, he shared that “predictive low glucose suspend is right around the corner” and “is going to be a huge step for this field.” He concluded his talk with lots of optimism: “We’re right on the cusp. People are wearing closed loop at home, and they are safer than what we’re doing right now. We’ve got to drive towards commercialization. JDRF is working with industry, working with the FDA, and already working with payers, to drive closed loop systems into commercial embodiments.”

  • UVA – JDRF-2 (outpatient closed-loop control), NIH 1 (five-day bedside closed-loop), Helmsley 2 (overnight summer camp studies), UVA Launchpad (adolescent missed bolus during the day), JDRF VCU (heart-rate monitoring). Said Dr. Kowalski, “This is just remarkable stuff, and it’s consistent – the benefit of waking up in the morning at 110 or 120 without hypoglycemia.”
  • Stanford – Overnight closed-loop camp studies (see our ATTD 2014 coverage; upcoming camp studies will test UVA’s DiAs and Medtronic’s system); full closed-loop at camp using UVA’s DiAs; in-home predictive low glucose suspend; predictive nocturnal hypo and hyper minimizer; a hotel-based study of Medtronic’s Android-based Hybrid Closed Loop during the day and full closed loop at night; drug eluting insulin infusion sets; the bionic pancreas multi-center study; and studies to detect sensor and infusion set failures.
  • Yale – Safety of nighttime suspension, reducing hypoglycemia following exercise, ultra-fast insulin (InsuPatch warming device). Current/upcoming inpatient closed-loop studies will examine injected liraglutide, hyaluronidase, and pre-exercise snacking. Upcoming outpatient studies will examine an ambulatory closed-loop device (collaboration with Stanford & Barbara Davis, initiating this summer), a Medtronic ambulatory closed-loop device (initiating this summer), and Medtronic’s predictive low glucose suspend (initiating this summer).
  • AP@homeTwo transition trials completed (2-7 days); final trials started (2-3 month home-studies testing both overnight-only and 24/7 control).
  • MD-Logic – Ongoing bolus calculator study (n=20), ongoing weekend 60-hour home study (n=24).
  • UCSB – Closed-loop with Afrezza (“looks like it will be approved in July as a very, very rapid-acting insulin”), control-to-range, outpatient clinical trials with zone MPC, predictive pump suspension, exercise detection.
  • RPI – Advanced closed-loop algorithms that have supported clinical sites in Colorado and at Stanford.
  • Cambridge – Over 1,200 home study nights and 100 day/nights of unsupervised, free-living use. Said Dr. Kowalski, “Cambridge has blazed the trail in terms of home studies.”
  • Dose Safety – Algorithm work to challenge closed-loop systems (e.g., high fat meals, exercise).
  • Joslin/Boston Children’s – PID control, skipped meals, bolusing for meals with high fat content.
  • IIT – Algorithm research on fault detection; control systems for AP use during and after exercise.
  • Dual hormone –Drs. Ed Damiano and Steven Russell were published in NEJM just a couple hours prior to Dr. Kowalski’s presentation; the latest bi-hormonal closed-loop research from Montreal shows that the addition of carb counting does not add much benefit to closed-loop control.
  • Australia (Dr. Tim Jones) – At-home and overnight closed loop studies using Medtronic’s Android-based research system. Studies that are underway include: ambulatory all day home studies (Medtronic); at home overnight studies (Medtronic); and a six-month multicenter predictive low glucose management randomized controlled trial.

The Last Mile – Bridging from Academia to Industry – Medtronic

Lane Desborough, MSc (Chief Engineer, Insulin Delivery, Medtronic Diabetes, Northridge, CA)

The insightful Mr. Lane Desborough addressed what he sees as the biggest issue in artificial pancreas development – how do we take closed loop from a narrow use case (CRC, transitional studies) to an extremely broad use case (running on anyone with diabetes)? His presentation was very similar to that given at ATTD 2014, though it was valuable to hear his perspective once again. He had a compelling slide summarizing the magnitude/scale changes on the path to commercialization (see below) – “the last mile between a sponsored study and a commercial product is vast by just about any measure.” To get there, Medtronic is learning from other fields (e.g., aerospace, automotive, nuclear power, oil refining, petrochemical) and thinking strategically about human-centered automation, information flow/display, and the components in closed-loop “systems.” Mr. Desborough concluded that the flow of information and the flow of insulin are both key to ensuring that we have a deep understanding of the way closed loop will operate before it is released as a commercial product.

Study Type




Sponsored Studies


3 days


Supervised Studies


7 days


Home Studies


14 days


Commercial Product


4 years


  • “By my estimate there are more control loops then there are people on the planet - billions.” Mr. Desborough highlighted that closing the loop is not a new concept, and some industries like aviation have been using feedback control for over fifty years. “Why start with a blank sheet of paper,” he asked, “when we can instead leverage the hard-learned lessons of successfully closing the loop in cockpits, control rooms, and driver’s seats?” A number of engineers from these industries work at Medtronic and are helping the company think about closing the loop.
  • Closed-loop systems are comprised of far more than just a CGM, algorithm, and insulin pump. They include the BGM (strips, calibration), insulin, infusion set, batteries, the person wearing the system (health, activities, competency, training, the surrounding environment), rescue countermeasures (carbs, glucagon), the person(s) adjusting the system, and other stakeholders (regulators, payers, insurers, care partners). Mr. Desborough highlighted in a big bold red circle, “Safety, efficacy, and burden are properties of an entire system” – the individual components and how they interact determine the outcomes.
  • Mr. Desborough emphasized the importance of human centered automation – the human and the automation must cooperate to succeed. He emphasized that the human must be at the center of the automation, rather than on the sidelines. The key is to avoid “de-skilling,” where the human doesn’t know what to do when the system breaks down.
  • The flow of information from the system to the human must allow completion of three tasks: maintenance, context, and supervision.
    • Provide maintenance to the system: calibrate CGM, change insulin/reservoir/infusion site; replace CGM sensor; recharge/replace batteries; adjust therapy settings.
    • Provide input to the system about significant events: meals, exercise, illness, manual injections of insulin.
    • Resume direct management when desired and/or necessary: maintain situational awareness, prevent mode confusion, enable safe experimentation, avoid de-skilling.

The Last Mile – Bridging from Academia to Industry – Animas

Krishna Venugopalan, PhD (Head of R&D, Animas, Westchester, PA)

In a rather corporate strategy-like presentation, Dr. Krishna Venugopalan described Animas’ approach to commercializing the artificial pancreas. He noted, “academia has the skill and inclination to develop the technological/scientific kernel, and industry has the resources and system to produce a product.” Dr. Venugopalan emphasized the importance of designing systems for “robustification” – sensor changes, start-up time, CGM communication errors, infusion set changes, insulin refills, battery changes, stopping closed-loop control, etc. Like Mr. Desborough, he also addressed the critical importance of component integrations and human factors. The latter must “minimize user errors” he said, pointing to Google’s driverless car – the vehicle only includes start and stop buttons (i.e., no brakes or gas). The rationale is that “an abrupt shift to driver-controlled piloting would be unpredictable and potentially dangerous.” [While an instructive analogy, the direct relevance to closed-loop may be a bit of stretch, since a shift to human control will be necessary in cases of system component failure.] He concluded that as we drive to close the loop, “communication and collaboration are critical” between HCPs, regulators, and payers. “We need investment in healtheconomics data,” said Dr. Venugopalan.

The Last Mile – Bridging from Academia to Industry – Dose Safety

Bob Kircher (VP Engineering and Regulatory Affairs, Dose Safety)

Mr. Bob Kircher, a former engineer at Boeing, highlighted the perspective of Dose Safety, a Seattle-based company developing artificial pancreas control software. We appreciated the company’s mission: “A holiday from diabetes.” Like the preceding presentations from Medtronic and Animas, Mr. Kircher’s remarks focused on appropriate human-centered design that takes automation learnings from other fields into account.

  • “People naturally resist change. You must work from the known and trusted to the new.” From a human factors perspective, early closed-loop users will likely be current users of sensor-augmented pumps. As a result, artificial pancreas systems should provide situational awareness mechanisms and tactile user interfaces that are consistent with users’ past experience.
  • Automation should default to hide the complexity from the user. We thought this was an important point, since there is always a temptation to show users everything that is going on. Ideally, we believe systems would hide the complexity, but also allow users to look “under the hood” and customize if they desire.
  • Automotive cruise control offers human factors learnings about user adoption. At first, cruise control was a discrete feature that could be added to a car’s feature set (i.e., much like automated insulin delivery will be an available feature on pumps). Mr. Kircher noted that cruise control “sometimes worked terrifically – when no cars are on the road” (i.e., automated insulin delivery at night). But for maximum benefit in more challenging cases, cruise control requires user interaction. Similarly, closed-loop control will require user interaction for optimal glycemic control in more challenging circumstances (meals, exercise). Like cruise control, closed loop should have an on-screen indicator to reflect the status of automated vs. manual insulin dosing. Mr. Kircher proposed that an AP’s software goal could be to maximize time in auto-dosing.
  • To see rapid market uptake and improvement of AP systems, Mr. Kircher believes the following are needed: adequate AP reimbursement by payers; CGM and pump data input/output data standards; and regulatory approval for AP-ready pump and control algorithms (i.e., any AP-ready control algorithm could run on any AP-ready pump).

Panel Discussion: The Last Mile – Bridging from Academia to Industry

Dr. Aaron Kowalski (JDRF, New York, NY): When I think about this, there is one key component – it’s an evolutionary process. We will get to better and better systems. But we must get to a first system. As you think about the risk-benefit from an industry perspective, how do you judge when you’re ready to pull the trigger? We need to appreciate that there is a need to learn about the first commercial products. In the fields you talked about, where control loops matured and gained steam, where was the risk-benefit before a launch happened?

Mr. Lane Desborough (Medtronic Diabetes, Northridge, CA): I was speaking with a doctor about a month ago, and he said, “My hat goes off to Medtronic to take on the responsibility for the closing loop.” Right now, all mistakes are made by the patient or doctor. Now, we’re taking on that responsibility – it’s kind of like Uncle Ben from Spider Man – with great power comes great responsibility. Industry is taking on the tasks that have traditionally been assumed by the human. We need to be confident in our ability to take on that responsibility.

Mr. Bob Kircher (Dose Safety, Seattle, WA): People with diabetes use sensor-augmented pump therapy today to control diabetes. Many do really, really well. The challenge from a software standpoint is to do what they do. Dr. Venugopalan used the term, “Robustification.” He showed a week in the life of a person with diabetes with stop signs (calibrate CGM, change reservoir, etc.). Those are opportunities for implementing robustness where the system turns off auto-dosing when it encounters something.

Dr. Kowalski: How do you judge success – weighting hypoglycemia vs. hyperglycemia risk?

Mr. Desborough: When automation is applied, there is usually a failsafe button you can press – you can bring the oil refinery down to a safe place. But there is no fail safe in diabetes. You are managing between two extremes constantly and trying to understand the tradeoff between those. So you come up with what control engineers call an objective or cost function – a careful consideration between the balance of hyperglycemia and hypoglycemia. At the end of the day, the control loop is transferring variation from where it hurts to somewhere it doesn’t – in this case, variation in glucose to variation in insulin dosing. The application of these objective functions, which balance these risks, is a very important input to the design of controllers.

Mr. Brandon Arbiter (Tidepool, Palo Alto, CA): I love the analogy of cruise control. Sometimes you turn it on, sometimes you don’t. Dr. Venugopalan mentioned how often things are complicated. As a first step, I don’t want to wake up in the morning above 150 mg/dl. Sometimes the system is on, sometimes it’s off. But what are practical applications of that? Can it really mitigate enough risk?

Mr. Kircher: I think it can. It requires the user to interact with the system when the system isn’t controlling glucose well enough. Even if the blood glucose is 250 mg/dl, if the system is green (auto dosing), the user could say, “I will leave it alone and see what happens.” If the glucose doesn’t come down, you could take a manual bolus. Cruise control is a good analogy.

Dr. Venugopalan: For hyperglycemia and hypoglycemia management, it’s about risk and responsibility and your level of confidence. Systems are still single point of failure devices – that is the quandary. I don’t have an exact answer. There is a lot lower bar in terms of responsibility for hypoglycemia minimization as opposed to hyperglycemia minimization. We need more real world experience on how the system will function.

Dr. Stu Weinzimer (Yale University, New Haven, CT): We heard from all of you a recurring theme: the first devices aren’t going to be perfect. That’s okay when a lot of technology is being developed in academia. Once you start upscaling to a major manufacturing level, you have huge corporate investments in a product. Isn’t there a disincentive to make rapid improvements to a system? You must almost re-envision how you roll out pipelines.

Mr. Desborough: Right – you need the governance and structure to successfully manage change over very wide scales.

Mr. Kircher: The example of Boeing’s 777 is a phenomenal one. Three years went into the deployment of the aircraft. The goal was to let the aircraft land itself. It turned out that the length of life of the tires was so much better when the computer landed the aircraft

Mr. Desborough: There are a variety of ways to build trust in automation. I don’t know anybody who cedes life critical tasks the next day to an automated device. There is a gradual process of trust building.

Mr. Kircher: In a cockpit, there is a notion of piloted command – there is always at least one other pilot if that person wants to give control to someone else.

Mr. Desborough: We do this every night as parents. Who has the first shift, who has the second shift? Now you’re adding another agent into the team to perform tasks.

Dr. Hovorka: Mr. Kircher, you mentioned that your company’s mission is a “holiday from diabetes.” And then Mr. Desborough, you mentioned “de-skilling” – these two things refer to same thing. You can see one as a good thing, and another as a bad thing. What are we going to call it, and how are we going to manage it?

Mr. Kircher: They may be automatically dosed, but they must still own that responsibility.

Dr. Kowalski: There has been progress, but are we being too conservative? Someone has a full tilt seizure on my airplane this week. We have teens with an average A1c over 9% in the US. And we have airplanes that are landing themselves. Lane and I went on a ride in a new Ford car. It has lane recognition and it can parallel park itself. If a car can auto-brake at 70 mph, are we being too conservative here with automated insulin delivery?

Mr. Kircher: CGMs and pumps work well enough. People mainly control their own blood glucose. The finish line is in sight.

Mr. Desborough: It’s very difficult to answer that. If I put on my parental hat, I want this desperately. I want it yesterday. I want to get some sleep, I want my family to get some sleep, and I want my child to be safe. On the other hand, I understand the incredible responsibility that this places on device manufacturers. The question is how to resolve that? The urgency is there – there are unsafe things happening every day in diabetes. We have to be extremely careful about how we take on that responsibility.

Q: What lessons can we take from the automation of automobiles and planes – how do those industries manage the liability question? Clearly there are big liability issues when those systems fail...

Mr. Desborough: One is playing out in the news right now – Toyotas with unintended acceleration. These examples have many common factors with what we’re trying o do. We need to internalize and understand what these industries are facing.

Dr. Kowalski: Even implanting defibrillators in people, those systems are running control loops...

Dr. David Kerr (Sansum Diabetes Research Institute, Santa Barbara, CA): In terms of commercialization from a payer and clinical perspective, where will this lie in the hierarchy? Is your vision that the numbers will stack up, and a large volume of people with type 1 diabetes will use this product? Or do we need to bring this to the marketplace at an earlier stage for specific groups: hypoglycemia unaware, those with recurrent severe hypoglycemia, and those with early retinopathy. From a commercialization point of view, where do you see this going?

Dr. Venugopalan: When you start looking at trying to advance these technologies, there is a very diverse set of users – in terms of both skill level and interest level. That’s one of the reasons why we struggle. Some people are more or less tech savvy, and that may or may not be the person that most needs the technology.

Ms. Tamar Sofer-Geri (CarbDM, Mountain View, CA): I am a mother of a daughter with type 1 diabetes. I’m wondering about how to get through the last mile in the regulatory process. We now have FDA guidance on the artificial pancreas, and we made huge progress in getting there. But technologies are still approved in Europe way before the US. To me, that’s the biggest barrier to get to a closed loop.

Mr. Desborough: My perspective is that this is a new frontier for our regulatory partners. There was no department of the artificial pancreas at the FDA because there hasn’t been an AP. How do we work with our regulatory partners to build the knowledge that systems are safe and effective and less burdensome? We should also reach out to the FAA, who has decades of experience in similar ventures. We must bridge gaps to understand what is safe and effective.

Dr. Kowalski: Having spent a lot of time with Dr. Stayce Beck and her team at the FDA, the pathway is pretty reasonable right now. The trials laid out in the guidance are reasonable. We need to get at them. There are things to solve on human factors and commercial embodiments. One thing to end on, which you said Lane – and I’ll take that back to the team at the FDA – is this idea of iterating. How do we iterate and not get hung up in PMAs that take a year at a time, especially when we can add on things that are better? We need to leverage other fields and examples and do it safely. This is going to be an area where we’ll learn quickly and improve. But we need a pathway to improve.

Panel Discussion – The Real World Challenges and Successes of AP Clinical Studies

Mr. Adam Brown (Close Concerns, San Francisco, CA): When designing closed-loop systems for younger patients, what are the most important device design aspects to keep in mind?

Dr. Bruce Buckingham (Stanford University, Stanford, CA): It’s got to be small – some of these patients don’t have a lot of on-body real estate to work with. The research systems right now still require carrying around a lot of stuff. We also need more research into infusion sets, since they fail so frequently.  

Mr. Brown: In addition to your work on infusion sets, you've done lots of incredible work on alarms, showing that patients often don’t wake up at night to them. When we think about designing alarms for real-world AP systems, what can we learn from your experience? How do we tradeoff more aggressive alarms with annoying users?

Dr. Buckingham: One alarm a day is too much for me. So we want to minimize alarms. I’m not too worried about highs, even for a few hours – it’s the lows that I am worried about.

Mr. Brown: The development of the UVA DiAs system included a lot of focus group work with patients, as I understand it. What was learned from this work that your team applied to the design of the interface?

Dr. Stacey Anderson (University of Virginia, Charlottesville, VA): Yes, we conducted many interviews with patients to understand how to design the user interface. We went through several iterations before arriving at the traffic-light user interface you are familiar with. For instance, we learned that patients wanted to see the glucose number and trend, even though the system was automating insulin delivery. We also made sure to keep the buttons on the home screen simple. Clear buttons to start and stop closed loop, a button to enter food, a button to enter exercise, a button to enter a fingerstick glucose value, and indicators on the top for connectivity to the CGM and pump.

Mr. Brown: I understand you recently changed the DiAs system based on recent closed-loop studies. Can you talk about those changes?

Dr. Anderson: Yes, we made some changes for prolonged home use. Portability is always an issue. Nobody wants to wear a fanny pack with three phones. There have been some advances thanks to Bryan Mazlish, and we are minimizing the things we need patients to carry. On the pump, patients have asked for temporary basals. There was some alarm fatigue, so our alarm now turns off for 15 minutes if the patient has given carb treatment and then re-checks. Our system now fail-safes in an intelligent way to the pre-programmed basal on the Roche pump. For example if the light was red, then it fails to a temp basal of zero for sixty minutes. If it was a yellow light, it’s zero for thirty minutes. This is going to be safer for the user at home.

Mr. Brown: Dr. Phillip, you've had tremendous experience doing home studies in Europe. What was the biggest challenge or biggest learning in taking MD-Logic outpatient, first to diabetes camps and then into patients’ homes?

Dr. Phillip: In 2011, when we did the diabetes camps, we thought we understood how patients react at night. But home is a totally different game. Patients do what they want, despite what we say. They don’t care about your instructions – they exercise, they eat whatever, and they disconnect when it’s convenient for them. You have to address this all the time. Now that we use algorithms instead of basal/bolus, we have to explain a whole new way of thinking to the patient. And nobody talked about the parents. We don’t have a parent who sleeps a full night if their child has diabetes. That’s why we call MD-Logic the Glucositter – like a babysitter. You build confidence with time. You start with surveillance and then get rid of it gradually. Parents like automatic control – in six weeks at home at night we lose patients in the control group, but the intervention group keeps using MD-Logic. Those were the lessons we learned. We need to bring this to them as soon as possible because it seems like it is safer and we get them closer to goal.

Mr. Brown: We’ve mentioned designing for patients quite a bit, but how do we design for parents?

Dr. Phillip: The same as for the children  - keep it simple. Don’t make them go into engineering school. Make the device small. Try to make one device, not ten. Make it strong. They might throw it into the corner of the shower.

Mr. Brown: Just to reiterate the point about simple – the metric I always like to use is whether I need an instruction manual to use the product. I shouldn’t need an instruction manual to use the device! Turning to you, Dr. Hovorka, your recent paper in BMJ (Barnard et al., 2014) does an amazing job of capturing the patient experience on closed-loop technology, both positive and negative. You cited four main thematic areas that were negative: Calibration difficulties/frustration when equipment ‘fails’; size and alarms; accuracy/trust; and discomfort/painful. As you think about the move to commercially available closed loop systems, which of these areas is most important for industry to keep in mind? If you were in industry and had to pick one, where would you put your resources?

Dr. Roman Hovorka (University of Cambridge, UK): This work was done by Dr. Katharine Barnard at Southampton. I think the psychosocial factors are under-researched. If you give closed loop to people, they see it as a single system, not individual components. What is new to us is the concept of “building trust.” If I was a manufacturer, I would invest in two things: connectivity should be good, and size matters to most people.

Mr. Brown: One other commercialization question for you, Dr. Hovorka. I remember a great exchange between you and Dr. Aaron Kowalski at the EASD Diabetes Technology Conference earlier this year. It revolved around the regulatory path for an overnight-only system vs. a full 24-hour system? Can you share your perspective on that?

Dr. Hovorka: The technology can handle 24/7, but the manufacturers want to play it safe.

Dr. Phillip: The enemy of the good is the best. We can answer an unmet need. The night is the most dangerous time. This is when parents and patients are afraid. If we can solve this, then it’s a huge value for patients. But we’re also studying 24-hour control to show that it is safe as well.

Mr. Brown: Kelly, you've had a chance to interact with two very different closed-loop systems - the Bionic Pancreas in the Beacon Hill study and the UVA Diabetes Assistant. What aspects of the device design and user interface did you like about each? If you were to start using either tomorrow, what would you want to see improved on each, if anything?

Ms. Kelly Close (The diaTribe Foundation, San Francisco, CA): I tend to be enthusiastic about any progress towards insulin automation, so I was so thankful to have an opportunity to be in both of these studies. In Beacon Hill, I didn’t actually care that the system was clunky. I was my best self – I didn’t have to think about diabetes and all the stress it brings every day. And it wasn’t until that was removed that I realized how much mindshare diabetes consumers. For me, having glucagon felt like something safe.

When I went to the overnight system, I didn’t think it was going to be as good, but I underestimated the power of having a really great night and waking up in the right place for the day. It also seems that this might be easier from a regulatory perspective. For me, whatever can get us there the fastest, obviously in a safe way, would be fantastic. I like that there are many different kinds of systems being created – patients have one thing in common, a diagnosis. But the spectrum of patients out there should be reflected in the variety of systems that come to market.

Mr. Brown: You and I have often discussed the need for incremental steps in automated insulin delivery, which means early products will improve with subsequent products. The major challenge, of course, is that early adopters are more likely to be intensively managed patients, who are doing pretty well already. With that in mind, how do we set patient expectations appropriately? What should industry keep in mind as they bring these systems to market?

Ms. Close: We are so lucky to be in this environment. We can ask for all these bells and whistles, but we must bear in mind that it is going to get better and better over time. We just need to get to a first product and get experience. Patients are getting more demanding all the time, and that’s just reality. But perfect should not be the enemy of the good.

Mr. Brown: Willa and Tia, You’ve been in so many closed-loop/artificial pancreas studies! What did you most like about the design of these systems? What do you think is the most important thing to improve on?

Ms. Willa Spalter: Probably a year ago, I was in an overnight study in a hospital, and I didn’t go over 128 mg/dl the whole night, which was great. Sometimes some of the sensors aren’t accurate and they shut off when you are not low, and then you go super high. It’s annoying when you are high.

Ms. Tia Geri: I did a study testing predictive low glucose suspend at night. Right now I wear the MiniMed 530G – it’s only half a closed loop. When I went low in the predictive study, I didn’t have to eat as much. It’s kind of like a backup safety net, and I like that about it.

Mr. Brown: If I said that you could have that system to wear tomorrow, what would you expect out of it?

Ms. Geri: It would shutoff before you I go low, and prevent me from going high. Insulin isn’t extremely fast right now, so you have to take it in advance. And I would expect the sensor to be accurate and reliable.

Ms. Willa Spalter: I would want it to shut off before you are going low. It’s really important to have an accurate sensor – it’s not good when it’s wrong.

Mr. Brown: Did you feel and act differently when you were on these systems? Were you scared?

Ms. Spalter: When I did my first study, it was literally a week after I got diagnosed. I was kind of scared because I didn’t know anything. It’s cool when you don’t have to treat in the middle of the night – usually if I eat, it’s hard for me to go back to sleep, because I have all this sugar in me.

Ms. Geri: I was excited to see how it would all work. So I almost wanted to test the system and make myself go low to see what would happen. [Laughter]

Ms. Tamar Sofer-Geri (CarbDM, Mountain View, CA): We already have an accurate device. Most people on CGM treat without checking their blood sugar. I say go for it. The predictive low glucose suspend is a no-brainer. I understand the risks. It’s never going to be a true vacation.

Ms. Close: It is amazing to be on closed loop. It’s like taking a vacation. It’s so special. But you must still interact with the system – no question about it.

Ms. Spalter: It’s really cool, but most of my studies are overnight. So it’s benefitting me, but it’s also for my Mom and my Dad who come and test me in the night. They wake me up if I am low, and it’s good if I don’t wake up in the middle of the night.

Ms. Geri: After I had finished one study, I was waiting for the pump to shutoff when I was going low. But it didn’t, because I was no longer in the study. I had to eat, and it was strange because I had grown so used to it doing its job.

Brandon Arbiter (Tidepool, Palo Alto, CA): Based on your most recent trial experience and what actually went wrong, would you be comfortable commercializing the closed-loop device you used?

Dr. Moshe Phillip: I think we are ready to commercialize the night. No question about it. It prevents hypoglycemia, provides tighter control, and influences the entire day. The night is ready. Waiting to solve the challenges of the day before seeking approval is wrong.

Dr. Buckingham: We would be ready if the connectivity was ready. If it was, I would go for 24-hour closed-loop control. We aren’t aiming for perfection during the day, but a treat-to-range system would be good.

Dr. Hovorka: We are ready now. It doesn’t mean it will happen soon, because there are other considerations.

Ms. Close: We have made amazing progress, but people ask about ‘edge cases’ – well, have conditional approval or make people sign something!

Ms. Spalter: Nothing is perfect, but I definitely would sign up. It’s much better than what everybody else is wearing now.

Ms. Geri: Things go wrong with what we use every day. So why not give people the best things that we have?

Dr. Kowalski: This year will be a tipping point in this field, don’t you think?



CGM Is Not a Limiting Factor in Artificial Pancreas Systems (75-LB)

T Bailey, K Nakamura, A Chang, M Christiansen, D Price, A Balo

This exciting poster shared in-clinic data from 51 patients that wore a version of the G4 Platinum with an improved algorithm (called “G4AP” in previous Dexcom presentations). The device’s accuracy was compared to YSI and fingerstick values (Bayer Contour USB) on days one, four, and seven. The poster also compared the accuracy of Bayer Contour USB values to YSI – a clear move from Dexcom to demonstrate that its next-gen CGM accuracy is approaching fingersticks. Overall G4AP MARD vs. YSI was an impressive 9.0%, compared to a fingerstick MARD of 5.6% vs. YSI. Notably, G4AP and fingersticks had a similar mean absolute difference (MAD) in hypoglycemia vs. YSI: 6.4 mg/dl and 4.2 mg/dl, respectively. In addition, the Clarke Error Grid data vs. YSI suggested G4 AP is really approaching the clinical accuracy of fingersticks– A+B Zone data was nearly identical (99.5% with G4AP vs. 99.6% with the Contour USB) and A-Zone accuracy was quite similar (92% vs. 99%). Overall, we thought the data were very, very strong and showed highly impressive accuracy using Dexcom’s existing G4 Platinum sensor and an improved algorithm – this hits the “holy grail” bar of a sub-10% MARD for CGM, a level of accuracy that some have called for to safely run tight closed loop control. This poster also underscored how much inherent inaccuracy there is in SMBG, and it makes us even more encouraged about the possibility of an insulin-dosing claim and factory calibration. A presentation later in the day noted that the “Share AP receiver” with the G4AP algorithm will be available for artificial pancreas research use in December 2014 (US) and 1Q15 (EU). We’re not sure if this would be rolled out to consumers, but are optimistic.

  • The poster concludes, “The clinical performance of this CGM is approaching that of current SMBG systems, particularly after the first day of use and in hypoglycemia ranges. The system could be adequate for use in diabetes management decisions without the need for SMBG tests, in particular for reducing hypoglycemia. Accordingly, the CGM accuracy should not limit AP development.” Given how many patients already use their existing G4 Platinum CGMs to dose insulin (technically “off label”), we agree and believe that G4AP surpasses the bar for independent diabetes management decisions.
  • This clinical trial enrolled 51 patients at three US centers. Patients inserted and wore one sensor for seven days and participated in three 12-hour clinic sessions (days one, four, and seven) with YSI every 15 minutes and SMBG capillary tests every 30 minutes. Glucose was manipulated to provide sufficient data in low and high glucose ranges during the clinic session. The CGM was removed at the end of the seven-day wear. The closest matched data point between CGM, SMBG, and YSI were used to assess CGM performance. The fingerstick meter used was a Bayer Contour USB. The CGM calibration scheme was twice daily fingersticks, prospectively calibrated.
  • The science behind the G4AP algorithm was described by Garcia et al., JDST 2013. The G4AP employs the same sensor and transmitter as the G4 Platinum, but contains updated denoising and calibration algorithms for improved accuracy and reliability. The JDST study used a retrospective G4AP algorithm application to the G4 Platinum pivotal study data. This poster reports on the prospective, clinical use of the G4AP algorithm – as we understand it, the G4AP clinical data (overall MARD: 9.0%) is even better than the retrospective data (overall MARD: 11.7%) because the study execution was better.


G4AP vs. YSI



Matched pairs




Overall MARD
  On Day 1
  On Day 4
  On Day 7




MAD in Hypoglycemia(<70 mg/dl)

6.4 mg/dl

4.2 mg/dl

7.8 mg/dl

Overall Clarke Error Grid

A+B Zones: 99.5%
A Zone: 92.4%

A+B Zones: 99.6%
A Zone: 98.5%

A+B Zones: 99.6%
A Zone: 98.5%

% within 20%/20 mg/dl




Rate-of-Change Dependence of the Performance of Two CGM Systems During Induced Glucose Excursions (846-P)

The authors compared the accuracy of two CGM systems: the Dexcom G4 and a prototype CGM system developed by Roche. This Roche-funded study enrolled 10 patients with type 1 diabetes who each spent about a week wearing four sensors simultaneously (two G4, two prototype). In an interesting wrinkle, the authors compared the performance of the sensors during two induced glucose excursions, which occurred roughly 40 hours and 70 hours after sensor placement. Measurements were compared to reference blood glucose readings drawn every 15 minutes during the excursions. (According to the poster these blood glucose measurements were also used for calibration; we are not sure exactly what this means or how it affected the results.) Notably, the G4 had numerically higher MARD than the prototype in every category of glycemic rate of change assessed, suggesting that the Roche sensor could be more clinically useful while glucose levels are rising or falling. The mean seven-day MARD was 10.9% for the G4 and 8.6% for the prototype. More than 80% of the prototype sensors had overall MARD below 10%, as compared to 20% of the G4 sensors.


Dexcom G4

Roche Prototype

Rate of Change (mg/dl/min)

MARD (%)

SD (%)


MARD (%)

SD (%)


< -3







≥ -3 to < -2







≥ -2 to < -1







≥ -1 to < 0







≥ 0 to < 1







≥ 1 to < 2







≥ 2 to <3







≥ 3







Clinical Benefit in Glycemic Control Using a Long-term, Implantable, Continuous Glucose Monitoring System in a 90-Day Feasibility Study (837-P)

C Mdingi, R Rastogi, A Dehennis

In this poster, Senseonics presented 90-day data (45 days blinded + 45 days unblinded) on its implantable CGM system (fluorescence-based sensor, body-worn transmitter with Bluetooth connectivity, and a mobile smartphone app). Twelve patients took part in the three-month study, and sensor accuracy was compared to YSI at in-clinic visits every ~14 days. Overall MARD vs. YSI was a strong 11%, ranging from a low of 7.7% to a high of 17.7%. The Clarke Error Grid showed 85% of points in Zone A and 14% in Zone B (n=1,890 paired CGM-YSI points). However, the poster did not divulge the calibration scheme, in-clinic glucose ranges, or specific study design details/protocols, so it’s hard to know how real-world this accuracy is. [From the Clarke Error Grid, the vast majority of points appeared to fall in the 70-180 range.] There was also no mention of the percentage of sensors lasting 90 days or any details on explantation. Indeed, the poster was really focused on comparing the 45-day blinded period of sensor wear to the 45-day unblinded (i.e., real-time) sensor wear – average glucose significantly improved from 175 mg/dl (blinded) to 156 mg/dl (unblinded), which included a 7% reduction in hyperglycemia, a 1% reduction in hypoglycemia, and an 8% improvement in time in range (75-180 mg/dl). Overall, these feasibility results are encouraging, but we would like to see a longer, larger, and more real-world study, along with more details on the sensor’s calibration scheme.

Symposium: Closed-Loop Insulin Delivery — One Step at a Time (Sponsored by The Helmsley Charitable Trust)

Present State of Sensor Technology

Jessica Castle, MD (Oregon Health and Science University, Portland, OR)

Dr. Jessica Castle provided a thorough overview of CGM sensors and their use in artificial pancreas (AP) systems, including a summary of current CGM accuracy/performance, interfering substances, telemetry, and connectivity. She compared Medtronic’s Enlite, Dexcom’s G4 Platinum, and Abbott’s FreeStyle Navigator by referencing the Damiano et al., head-to-head-to-head study recently published in JDST – Dr. Castle emphasized that the G4 Platinum had the best accuracy (MARD: 10.8%) vs. the Navigator (12.3%) and Enlite sensors (17.9%). Similarly, the G4 Platinum had the lowest egregious error rate (MARD >50%) – 0.5% vs. 4.3% with the Enlite and 1.4% for the Navigator. She emphasized that CGM performance is significantly less accurate on the first day compared to subsequent days, providing a key area for future sensor improvements. Dr. Castle also highlighted the critical importance of calibration accuracy, as MARD is significantly higher when a sub-optimal meter is used for calibration – we would note this presents both a challenge (poor SMBG technique and suboptimal meters negatively influence sensor accuracy) and an opportunity (factory calibration!). Towards the end of her talk, she reviewed the ASPIRE in-home study testing Medtronic’s MiniMed 530G, noting its ability to reduce hypoglycemia (we would note that the current label does not include a hypoglycemia claim, since the ASPIRE data was not incorporated into the approved label). She concluded by looking forward to the future, highlighting Roche’s prototype sensor (MARD <10%) and Dexcom future G4AP algorithm – see elsewhere in this report for updates on both products.  

  • Dr. Castle highlighted the importance of accurate CGM calibration, noting that meter can have a MARD almost double that of YSI (16% vs. 8.5%). She pointed out how powerful it is to wash hands prior to calibrating and to make sure no substances from the outside environment are interfering. We think this is an underappreciated piece of real-world CGM accuracy.
  • Dr. Castle highlighted the issue of telemetry, in which sensors momentarily stops transmitting data to the receiver. The Dexcom G4 Platinum made a particular advance on this front – the G4 Platinum captures 97% of all data on average compared to the Seven Plus’ 90%. We’d note that the improvements in communication reliability and transmission range were widely lauded by patients when the G4 Platinum came out.
  • Accuracy was shown to improve with the new algorithm G4AP compared to G4 Platinum. G4AP had a MARD >20% only 7% of the time compared to the G4 Platinum’s 20%.
  • Based on OHSU’s bi-hormonal AP work using the G4 Platinum, using two sensors does not seem to significantly boost accuracy more than using one single sensor. In the past, OHSU had seen a benefit to using two Dexcom Seven Plus sensors.

Questions and Answers

Q: Regarding calibration, I’d like to highlight the timing of calibrations. When is the best time to do this?

A: This is definitely an important concept when you’re calibrating a device. Do it when glucose is relatively stable so you can get around the delay issue.

Q: You focused a lot on commercially available sensors, but what are your thoughts on long-term implantable sensors?

A: I know great research is being done in that arena. I think some of the concerns are the invasiveness of having implanted sensors as well as the long-term complications of that. I don’t foresee long-term implantable sensors as the perfect solution for the AP system.

Q: Can you comment on how the number of times you calibrate the meter affects the accuracy of the sensor?

A: It depends on the device. Data from Medtronic showed improved accuracy when you calibrate three to four times compared to less frequently. But really it also depends on change in device’s sensitivity over time. Medtronic’s a little more sensitive to calibration, so that improves accuracy. It also depends on the situation, if the patient is having more drift or not. If the sensor is drifting over time, more calibration is going to improve accuracy.

Symposium: New Frontiers in Inpatient Diabetes Management

Is the Hospital Ready for Continuous Glucose Monitoring?

Michael Agus, MD (Boston Children’s Hospital, Boston, MA)

Dr. Michael Agus wittily prefaced his presentation as an entirely “off-label discussion,” since using CGM in the hospital has not been approved by the FDA. Despite the “near insurmountable hurdle” of FDA approval, Dr. Agus stressed that CGM in hospitals can offer an “enormous addition” to the data available at the bedside, allowing healthcare teams to capture useful trending data and extreme glucose variability that results in earlier blood glucose checks and subsequent clinical decisions to prevent adverse events. He called an MARD of <10% “outstanding” and safe enough for closing the loop, 10-14% “pretty good,” and 14-18% “mediocre” (though still good enough to get “terrific” avoidance of hypoglycemia). Indeed, in reviewing the ongoing HALF PINT study in pediatric ICU patients (n=1,900), Dr. Agus highlighted that the CGM protocol detected 45% (18/40) of hypoglycemic events (<60 mg/dl) before nurses/glucometers, thus triggering blood glucose checks and subsequent insulin infusion – this was with a MARD of 17.6% with an unspecified sensor (we assume Medtronic Guardian or Dexcom Seven Plus). Dr. Agus urged CGM manufacturers to “make the move” to producing hospital-based CGMs, and equally important, for the FDA to provide guidance for hospital CGMs rather than “stonewalling.”

  • Dr. Agus also shared his view in Q&A if companies get engaged in closed loop in the hospital, it will “be a standardized part of ICU care in the future.” Medtronic is certainly the most logical company to take this on – it already has a hospital-based CGM in the EU (Sentrino), and the ability to pair that with insulin dosing algorithms in-house.
  • While the “single greatest benefit for patients” of CGM use in hospitals is managing and preventing hypoglycemia, Dr. Agus added that hospital-based CGMs also help manage hyperglycemia, prevent diabetic ketoacidosis, evaluate the safety of insulin or anti-hypoglycemia medication, and save nursing time.

Questions and Answers

Ms. Arleen Pinkos (FDA, Silver Spring, MD): We know that alarm fatigue has been problematic. Do you have any recommendations for hospitals on where to set these alarms and schedules? Second, as these CGMs get better, what kind of performance criteria do you think will be necessary to use devices in hospital settings for dosing?

A: I don’t have specific recommendations for alarm fatigue, though we do worry about it. However in my world of pediatric endocrinology, nurses are more worried when there is no alarm. Adding alarms and finger sticks help them feel much more confident in the care they deliver to our young patients. Most of the fatigue in our department comes from arrhythmia monitors that are completely irrelevant. Secondly, to pick a number for dosing based of CGM, I believe we need MARD <10%. MARDs in the range of 10-14% add an enormous amount of value as a safety value and trigger for obtaining value. The blood measure is accurate but invasive and an enormous number of patients could benefit from CGM-based dosing.

Q: Do you have any experience in non-ICU pediatric patients with use of CGM? Also, if you could have a crystal ball and look into the future, when do you think we’ll be looking at closed loop?

A: We don’t have much hyperglycemia in pediatrics outside of diabetics and the ICU, but we have a fair amount of hypoglycemia in NICU. We are starting a trial in NICU this summer. If we can get companies engaged in closed loop, my crystal ball says that this will be a standardized part of ICU care in the future.

Q: What is the cost for training the nursing staff to react to CGM alarms?

A: There is some degree of training but it’s absolutely trivial. This is a decrease in nursing burden and nurses appreciate knowing what’s happening in between drawn blood values.

Q: I appreciated your enthusiasm and description of learning, but I want to add that there are still patients for whom subcutaneous samples aren’t terribly accurate. My advice: get an arterial sample please! Also, based on CGM use in outpatients, accuracy of hypoglycemia detection is at 50%. Can model prediction based on trends make this much, much better?

A: As we bring it in-house, there is absolutely modification in the algorithm that we can do to improve accuracy. Secondly, I absolutely share your bias that in sicker patients, CGM wouldn’t perform as well. However, that’s not what we found. Yet I agree that we have to use arterial samples in those sicker patients.

Q: With the increase in IV acetaminophen and the spike it causes in CGM monitoring, how do you train your team to ignore that glucose result?

A: In the ICU trials, it turns out that acetaminophen isn’t used that much. Through luck we’ve avoided it in our big trials. Since we’re bringing patients in, we tell them to avoid acetaminophen. Dexcom is trying to address this currently, and Medtronic did address this in Europe. The real answer to your question is that companies must address this issue.

Q: How often do you address blood glucose with CGM; for day-of calibration, is there a different protocol?

A: In the first 24 hours, we know that performance isn’t as good, but we haven’t adapted our algorithm to that. We don’t make any clinical treatment decisions unless we have a blood value. We used to do an early blood glucose check if the CGM was telling us that something was awry. We set up a schedule as if there was no CGM; we haven’t had the confidence to reduce blood drawing yet.

Corporate Symposium: Clinical Application of Real-Time CGM: Professional Use, Pediatrics, and the Pathway to the Bionic Pancreas (Supported by an unrestricted educational grant from Dexcom)

Pathway to the bionic Pancreas

Steven Russell, MD, PhD (Harvard Massachusetts General HospitalUniversity, Boston, MA)

Dr. Steven Russell updated the audience on his group’s development of a bionic pancreas that delivers both insulin and glucagon using input from the Dexcom Gen 4 CGM. Dr. Russell announced that they would begin a two-week home-use study of the system in just two days. Monitoring in the study is minimal, and freedom is high: patients are even allowed to drive while under closed-loop control! For details on this latest trial, see our coverage above of Dr. Russell’s oral presentation on the Summer Camp and Beacon Hill Studies (237-OR). Dr. Russell and his team are also conducting clinical PK/PD research on Xeris’ glucagon, which is liquid-stable for up to two years at room temperature. The hope is to use this stable glucagon in a pivotal trial (more than three months long) in ~2015, with FDA review “as early as the end of 2016” and possible approval in 2017. Dr. Russell was positive on his group’s relationship with the FDA, noting that all of their investigational device exemptions (IDEs) have been approved within the initial 30-day window: “We sometimes joke that our regulatory consultant has been the FDA itself.”

CGM Use in Pediatrics

Bruce Buckingham, MD (Stanford University, Packard Children’s Hospital, Stanford, CA)

Dr. Bruce Buckingham emphasized that age, BMI, and subcutaneous sensor location do not seem to affect CGM accuracy, but the size of the sensor and transmitter do matter to pediatric populations (particularly adolescents who face competing priorities of peer acceptance and body image concerns). Citing two different studies, one using the Medtronic Enlite sensor and the other using the Dexcom G4 Platinum sensor, Dr. Buckingham highlighted that sensor accuracy (MARD) remained similar across varying age groups. In his opinion, “CGM accuracy is dependent on the quality of the meter glucose reading,” (i.e., no dirty fingers allowed – a bad calibration results in bad data). Additionally, in a short plug for his poster presentation on Sunday, Dr. Buckingham also mentioned his current study assessing the effect of lipohypertrophy on CGM accuracy; interim data showed that median ARD for lipohypertrophy sites was actually better than median ARD for normal sites (10.0% vs. 12.2%). Looking ahead at innovations that could widen pediatric adoption of CGM, Dr. Buckingham was very enthusiastic about the peace of mind benefits that Dexcom Share will provide to parents and spouses of people with diabetes. However, he provided no timeline updates for the Share other than, “FDA approval is pending… with a pregnant pause.” As of Dexcom’s 1Q14 call, approval was characterized as “in the final stages of review.”

  • Dr. Buckingham emphasized that CGM use in adolescents is challenging due to the competing priorities of social acceptance, body image concerns, sports schedules, etc. To much laughter in the audience, he added, “Adolescence is a stage of temporary insanity… I tried to teach a high school class once and it was the hardest thing I’ve ever done. The number one priority at that age is friends. Priority two is friends, priority three is friends, and priority four is… sports. Sadly, you won’t find diabetes in that top priorities list.”
  • Dr. Buckingham highlighted the results from a recent multicenter study using the Medtronic Enlite sensor – accuracy was not affected by age or BMI. Overall MARD was 15%: the youngest of the five different age groups assessed (3-7 years old) exhibited a MARD of 16%, while the oldest age group (25-46 years old) saw a MARD of 15%. Similarly, MARD based on BMI hovered around the median of 15% for each of the age groups.
  • Similarly, accuracy data on the Dexcom G4 Platinum also showed that patient age does not significantly affect sensor performance. In patients ages 2-5 years, average MARD was 17%; in patients ages 6-12 years, average MARD was 16%; and in patients ages 13-17 years, average MARD was 15%. See our detailed coverage of the data from ATTD 2013. As a reminder, the FDA approved the Dexcom G4 Platinum in February for use in patients as young as two years old. Previously, marketing was only allowed to patients 18 years and older.
    • In a comparison to adult G4 Platinum performance, accuracy in pediatrics appears worse. In camp studies (n=740), MARD was 17.5% and median ARD was 13.5%. In comparison, inpatient studies on adults (n=201) suggested an MARD of 10.4% and median ARD of 7.7%. Dr. Buckingham hypothesized that the discrepancy in accuracy between pediatric and adult populations could be explained by incorrect meter calibration values used at camp.
      • Put simply, a real challenge for pediatric patients is clean fingers when testing. Indeed, testing himself, Dr. Buckingham noted that a clean finger resulted in glucose levels around 94 mg/dl, blood + sugar water = 94 mg/dl, blood + milk= 310 mg/dl, blood + jam= 361 mg/dl, and finally blood + pancake syrup= 526 mg/dl.
  • Dr. Buckingham also mentioned his current study looking at the effect of lipohypertrophy on CGM accuracy. The preliminary data from eight subjects (though total study n=30) was presented at a poster session during ADA 2014. Sensors’ median ARD at lipohypertrophy sites was 10.0% vs. a median ARD at normal sites of 12.2% (precision ARD= 9.5%). Dr. Buckingham optimistically stated, “Patients do have a place for sensor insertion in the ‘wasteland’ of lipohypertrophy.”
  • Dr. Buckingham was very optimistic about Dexcom Share, which will use a cradle to send G4 Platinum receiver data to the cloud and up to five “followers.” Using the Dexcom Follow App, followers can then view the receiver data and receive notifications on an iOS device. He emphasized the peace of mind that this device will bring for families with diabetes. In Dr. Buckingham’s examples, the Dexcom share would make sleepovers, business trips, and travel much less stressful. No updates were provided on FDA approval; the PMA supplement was filed in July 2013, and as of Dexcom’s 1Q14 call, approval was characterized as “in the final stages of review.” In Dr. Buckingham’s words, “FDA approval is pending… with a pregnant pause.”
  • Dr. Buckingham noted that Medtronic’s mySentry remote monitoring is “really loud,” but it is “several thousand dollars” and “very few people can afford it.” The FDA approved this system in January 2012 – and it was certainly a big advance to get it through the FDA, as we understand it. Since that time, the mySentry has not taken off to our knowledge, and it certainly has not been mentioned on any Medtronic earnings calls. We assume Medtronic has not prioritized obtaining reimbursement for the device, given plans to commercialize the Connected Care device/smarter transmitters, which will send pump and CGM data to the cloud and mobile phones.

Panel Discussion

Q: If accuracy increases with sensor use, why do you need to change it every week?

Dr. Bruce Buckingham (Stanford University, Stanford, CA): You don’t. (Laughing) I’m sorry. 

Dr. Jay Skyler (Miami University, FL): If you follow the label you do!

Dr. Buckingham: You should absolutely follow the label and change it every week.

Q: Given the tendency of your current prototype to rely on a wireless system, do you have concern about patient going into areas with heavy radio frequency?

Dr. Steven Russell (Massachusetts General Hospital, Boston, MA): I should mention that the current version of the device is dependent on Bluetooth for communication with pumps. We monitor how many dose commands didn’t get to the pump at the initial try and this was 4-7% of time, depending on the study. The good news is that the Bluetooth almost always spontaneously reconnects. We think the system could do better if it could deliver the dose command immediately. We’re looking forward to a fully integrated version, which won’t rely on wireless connectivity to the pumps.

Q: I’ve heard a rumor about tachyphylaxis with glucagon. Is that the case?

Dr. Russell: I’ve hard rumors of that too, but I’ve never seen evidence of that. It may be, in part, because we’re giving microdoses of glucagon. We’re not giving the huge doses that strip the liver of glucagon. In fact, serum levels of glucagon appear to be in the normal range most of the time despite our dosing.

Q: How much glucagon is typically infused in 24 hours by the bionic pancreas? How does this compare to the non-diabetic pancreas?

Dr. Russell: Our total daily dose of glucagon is typically 750-800 micrograms. The FDA-approved rescue dose is one milligram.

Q: If glucagon must be reconstituted and mixed, how does that complicate matters?

Dr. Russell: The stability concerns us a lot – it is probably not commercially viable to reconstitute every day and the current glucagon formulation probably would not be approved for even one day. Fortunately quite a few small companies, and some of the big ones, are interested in glucagon. A company from Austin, TX called Xeris is a bit further ahead of the rest. Xeris’ formulation is stable for more than two years at room temperature. Their primary interest is an EpiPen-like device. However, their glucagon could be used in a pump. We are now doing a clamp study comparing freshly reconstituted Lilly glucagon to Xeris glucagon at low doses (50 micrograms). We have seen no difference in PK or PD in preliminary use. We hope to substitute Xeris glucagon for the pivotal trial.

Q: For Ms. Kruger, what is the effect of acetaminophen on accuracy of the Dexcom G4? If you are taking Tylenol, how long should a patient wait to calibrate the CGM?

Ms. Davida Kruger (Henry Ford Health System, Detroit, MI): In our clinical experience, when we use the G4, we don’t see as much interference with Tylenol (acetaminophen) but we do tell our patients to be aware of this interference and to expect two to four hours of hyperglycemia readings, but to ignore this because it will correct itself afterwards.

Q: Have you considered incorporating a threshold-suspend feature into your bionic pancreas?

Dr. Russell: If the blood glucose is falling fast enough, that will actually happen. That’s just part of the control algorithm. If blood glucose is falling fast enough, it will stop giving insulin and give some glucagon.

Q: With relying on the sensor to pick up a meal-related rise, I’m surprised you don’t have more hyperglycemia. Even if patients pre-bolus based on accurate carb counts, they get hyperglycemia.

Dr. Russell: We do have a postprandial rise. It can be diminished with meal announcement, when the system gives 75% of the calculated dose. We have good control in the fasting period and very good control at night; that contributes to the average blood glucose. But there’s no way to avoid some postprandial rise without the pancreas dumping insulin into portal vein. Even the pancreas doesn’t prevent postprandial rises. If I eat a bowl of ice cream I’ll come up to 150 mg/dl. Now, I’ll come back down spontaneously. But I don’t think even the normally functioning pancreas is capable of clamping blood glucose without any excursions.

Q: In non-pump patients who want to bolus for post-prandial glucose levels, what guidelines do you have for bolusing safely?

Ms. Kruger: It’s common for them to be taking insulin without a pump. We will give them a very similar algorithm with a correction factor, treatment goal, and carb ratio. It’d be very similar to what a pump does but they have to do this in their head.

Dr. Buckingham: With insulin you have to prevent stacking. When you’re on MDI, we need smart pens to incorporate that feature. This could be solved pretty quickly. Also, in our practice, we tell patients not to do correction until two hours after a meal.

Dr. Russell: I just had a meeting with a company that is making this smart pen. It is on its way.

Q: How much insulin are you putting in these bionic pumps? How long does it retain effectiveness?

Dr. Russell: We use a Tandem t:slim pump, which has a capacity of 300 units of insulin that we fill up entirely. Typically, we switch out the insulin every two days just to make sure that loss of insulin efficacy isn’t affecting our results. However, the insulin is going to be OK longer than that in most cases. The question I thought you were going to ask is how much insulin we use. People who already have their blood glucose under target don’t use any more insulin on the bionic pancreas relative to usual care. However, adults who have an average blood glucose above target did use higher amounts of insulin on bionic pancreas than usual care in the Beacon Hill trial – they may just not have been getting enough insulin during usual care. Interestingly, that wasn’t the case for kids with high averages in the Summer Camp study – there was no difference in insulin usage between the bionic pancreas and control arms.

Q: What criteria of patients using professional CGM predict future continued success of personal CGM?

Ms. Kruger: I think after a week the patient understands what CGM is and isn’t. They understand that it will give alarms, what it’s like to sleep and shower with it. As long as patients come back with a positive: “I really want to own this.” Often people want the system for another week, or not to give it back. These are positive signs. If patients opt to own it, they usually do very well.

Dr. Skyler: The numbers you showed were that of 402 people, 205 went on personal CGM – that’s about half, right? That’s a good predictor.

Ms. Kruger: Four-to-five years ago people said there was no value to professional CGM, because we wouldn’t know if people would be good candidates to own systems. Based on our experience we didn’t agree with that. Patients have no concept of what CGM is unless they’ve used it. We don’t want to spend $1,000 in insurance money for the system to sit on the shelf.

Dr. Skyler: That makes good sense.

Q: Dr. Russell, have you tried using faster acting insulins? Do you think you would have to change algorithms? Would you see better results?

Dr. Russell: We use lispro in all our studies, and haven’t tried an ultra rapid-acting insulin. We think we would get significantly improved results with faster acting insulins. I should emphasize, though, that we clearly don’t need faster acting insulin to get good glucose control with the bionic pancreas. One thing that was surprising to us was just how variable the absorption of insulin could be. Patients absorbed lispro at Tmax values ranging from 30 to 180 minutes. There was also significant variability within a person, but despite that variation those who were slow absorbers tended to absorb at a slower than average rate each time we looked and fast-absorbers tended to absorb faster than the overall average. We had to make an adjustment to the PK assumptions in our algorithm to account for this variability. Right now the algorithm is set conservatively to handle slow absorbers. We could use rapid acting insulin right now, and it wouldn’t harm anyone due to our conservative algorithm, but we wouldn’t take full advantage of the more rapid absorption without an adjustment to the PK assumptions. The nice thing is that we would just have to turn one knob in the algorithm to adjust for more rapid absorption if it were available.

Q: It would for sure be more physiological though, right?

Dr. Russell: Of course. The delay in insulin absorption is a big challenge and this would make it a lot easier.

Dr. Skyler: You do so well already though.

Dr. Russell: I think if insulin were absorbed faster with a Tmax at 30 min, we could have had an overall mean blood glucose in our Beacon Hill study cohort of 120 mg/dl instead of the 133 mg/dl we actually saw.

Q: Studies have used GLP-1 therapy to reduce glucose variability. What does that say about the etiology of glucose variability? Would you ever go and use three hormones, adding a GLP-1 agonist, in a bionic pancreas?

Dr. Buckingham: We’re not supposed to use GLP-1 agonist in pediatrics. But I think that’s a great idea.

Dr. Russell: If you’re going to have a second hormone, I think the question is whether you want to use glucagon or pramlintide. There might be some benefit in using a once-a-day or once-a-week GLP-1 analog to slow gastric emptying and suppress endogenous glucagon secretion. In fact, there is some paradoxical endogenous glucagon secretion occurring in response to meals, and we would need less insulin if we could suppress that. However, given the performance we have observed with the bionic pancreas, we don’t need to do that.

Q: Would the device that keeps track of insulin dosing have to be on a smart pen, or could we maybe even have it on CGM – to tell it that you’ve had your insulin dose?

Dr. Russell: The nice thing about the smart pen is that then the patient doesn’t have to remember to enter the dose into anything. What you have with the pump is that it knows every dose it gave. But if you are using a pen, you have to put that information into an app. If you build a pen that automatically transmits, you get that functionality. That is probably the main benefit of pumps, honestly. There are some tricks you can do with different patterns of dosing, but I think the big benefits [with pumps] come from the built-in bolus calculator, insulin on board, and never forgetting which doses are given when.

Dr. Skyler: What proportion of your patients are on pumps vs. MDI?

Ms. Kruger: My patients include a pool of type 1 and type 2 patients, so less than half are on pumps.

Dr. Skyler: Certainly CGM works in MDI as well?

Dr. Kruger: Though we’d like everyone to use bolus wizards and we explain how to use it and even set it up for them, sometimes they miss the insulin on board. That’s a big risk to the patient. A lot of patients can do very well on CGM without moving to pumps.

Dr. Skyler: The nice thing about CGM compared to pumps is that you don’t have to take it off when you’re swimming.

Dr. Kruger: Well there are pumps that you don’t have to take off in water either. But sometimes patients have to think about what they can afford and what their insurance will reimburse and cover. If a patient is faced with the choice of only one, while we are strong proponents of pumps, CGM may be the better choice.

Q: Do you worry about the battery of the smartphone draining out? Do you have a backup system in place? Regarding safety mechanisms and malfunctions in general, what do you do if there are erroneous glucose readings? How do you ensure that everything works in the bionic system?

Dr. Russell: For this first device, we’re cobbling it together from parts, so it did have battery issues. We had to plug it in at night and once during the day. After all, it was on all the time, and an iPhone can’t last all day. However, we think that a couple of AAA batteries in the final device will let it run for weeks, once we optimize the system. In terms of what the bionic pancreas would do if it got erroneous glucose information, it would give the wrong dose of hormones. However, you can find ways to deal with that. Let’s say the device has been mis-calibrated. If the CGM is running high relative to blood glucose and that leads to the patient being low, they would feel that and be motivated to recalibrate.

What about people with hypoglycemia unawareness? What might not be obvious is that this device keeps people out of hypoglycemia almost all day, so even people who have lost hypoglycemic awareness are going regain it. So they’ll know if they are low. If the device was recalibrated low relative to blood glucose and that leads to running high, you’ll tolerate that for 12 hours until the next calibration, so not too much to worry about. And finally, what if the infusion set pulls out? It’s going to have two adhesive patches and two steel cannulas that are connected so if one pulls out, the other will as well. So you’ll be unlikely to lose just one hormone. Finally, the system now has an option to move the target glucose up by as much as 30 mg/dl, so the user can opt for a larger margin of safety and accept a somewhat higher average blood glucose.


Symposium: Joint ADA/AACC Symposium – Self-Monitoring of Blood Glucose – 21st-Century Issues

Update on Blood Glucose Meter Accuracy

David Sacks, MB, ChB (NIH, Bethesda, Maryland)

Dr. David Sacks provided an overview of BGM accuracy, focused on recent guidelines from ISO, CLSI, and FDA. He focused on the FDA’s January release of draft guidance for home and point-of-care glucose monitoring, which has since been met with significant pushback from the scientific and industry communities. Dr. Sacks was slightly more conservative in his opinions on the FDA draft guidance compared to the last time we heard his thoughts in February at the EASD Diabetes Technology Conference – at the time, he was quite frank, “There will be no glucose meters approved in the future.” However, he remained consistent that the FDA’s guidance for point-of-care meters is “much tighter than anyone has proposed before,” and proving this level of accuracy is not feasible given the inaccuracy in lab methods.

  • Dr. Sacks emphasized that the point-of-care FDA draft guidance leaves impossibly little room for error. In order for 99% of results to be within ±10% of the reference method, the coefficient of variation must be 3.8% and bias must be 0. For context, the plasma glucose measurements in central labs very rarely provide coefficients of variation below 3%. Furthermore, the imprecision of the reference method must also be zero, and according to Dr. Sacks, “no reference method is at zero.” To give weight to the statistical improbability of the guidance, Dr. Sacks consulted a statistician on the point-of-care guidance, who argued that to prove 99.9% accuracy, one would need 30,000 device measurements – a requirement that Dr. Sacks described as “obviously not feasible in most circumstances.”
  • Dr. Sacks reminded attendees of January’s FDA draft guidance, commenting that the guidance for SMBG meters for home use is “really narrowing the range for error” and the guidance for point-of-care meters is “much tighter than anyone has proposed before.” For more detail and comparison to the ISO and CLSI standards, see our report on the January guidance.
    • Regarding SMBG devices (home use), 95% of measured values must be within ±15% of reference (across the entire glucose range) and 99% of SMBG values must be within ±20% of reference (across the entire glucose range).
      • The home standards are much tighter than the updated 2013 ISO standards, which Dr. Sacks illustrated with an example of a patient with a true glucose value of 45 mg/dl. ISO would accept a 30 - 60 mg/dl range, which “is clearly not reliable to detect hypoglycemia.” However, with FDA proposed criteria, acceptable results are 38 - 52 mg/dl, providing a much tighter, and range.
    • Regarding point-of-care devices (healthcare facility use), 99% of measured values must be within ±10% of reference for >70 mg/dl and within ±7 mg/dl for <70 mg/dl. Additionally, no individual result should exceed ±20% of the reference method for samples >70 mg/dl or ±15 mg/dl for <70 mg/dl.
      • These new standards are much tighter than the updated 2013 CLSI standards, and have received “a lot of feedback, probably from many in this audience.” Due to the guidance’s extremely stringent accuracy requirements, Dr. Sacks anticipated that “it will probably be some time” before the official FDA guidelines are released.
  • For additional expert opinion on this controversial topic, see Dr. Irl Hirsch’s recent comments at AACE 2014 and commentary from the Diabetes Technology Society’s Hospital Diabetes Meeting.

Questions and Answers

Comment: I think its important to emphasize that what we see is only a partial picture because it only refers to error that comes from devices itself and does not take into consideration the overall system of errors. For instance, the human factor is a bigger contributor than even the reference method. I think I would emphasize that this picture you presented is even more difficult in reality.

A: Yes, that is a good point. These studies were done in optimum circumstances, and very few guidelines actually have account for patients doing these measurements.

Comment: I find it interesting this idea that hospital meters should be much more accurate than home meters. Inpatient glucose control is actually less stringent than for home control. In-hospital patients are surrounded by HCPs, while at home they can be driving cars, etc. Do you want to comment on this idea that it’s okay to have different standards at home and in the hospital?

A: Yes this is an interesting idea. Some people think that they should be different and each care environment should have different meters. Some people think that all meters should be equally accurate. The FDA, I think – I don’t know for sure – but I think their draft guidelines were driven in part by the meters used in ICUs. That is off-label and should not be done, but people still do them.

The New Error Grid – Rationale, Development, and End Product

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

We had trouble finding seats in the packed lecture hall where Dr. David Klonoff unveiled the new Surveillance Error Grid (SEG), a new alternative to the Clarke Error Grid (CEG) and Parkes Error Grid (PEG). The SEG was published today in the Journal of Diabetes Science and Technology. We’ve been looking forward to this presentation since April when Dr. Klonoff hinted at the error grid during the 2014 Clinical Diabetes Technology Meeting. To start, the SEG looks very different from the CEG and PEG, with a tie-dyed look, fading from combinations of green to orange to yellow to red based on averaged risk across survey takers (see our Diabetes Technology Meeting 2013 Day #1 report for more details). Dr. Klonoff explained the rationale behind developing a new error grid for surveillance, namely that the treatment of diabetes has changed, accuracy standards for BGM have become tighter, and our understanding of hypoglycemia has increased making it necessary to develop a new grid that will incorporate the new treatment and clinical treatment of diabetes care. To support this, Dr. Klonoff noted that, when comparing the attributed risk results between graphs, there was a 0.58 correlation between the CEG and PEG, but only a 0.36 correlation between the CEG and SEG results and a 0.31 correlation between PEG and SEG. Dr. Klonoff concluded from this that the SEG is similar enough to the CEG and PEG to conclude that a similar metric was being used but dissimilar enough to show that the SEG is useful in measuring something different – Dr. Klonoff called this the “sweet spot.” Looking at how this translates to surveillance, Dr. Klonoff noted that BGM standards of accuracy in SEG increased from CEG and PEG. Dr. Klonoff concluded by commenting that the new software accompanying the SEG can be downloaded for free at, allowing people can take their own reference BGM data points and find out what zone their data lie in.

  • According to the “Computing the Surveillance Error Grid Analysis” that developed the software mentioned above (and was also published today in the Journal of Diabetes Science and Technology), having more than 3.2% of data points in the at-risk zone (outside of the green), corresponds to more than 5% of data points outside the ISO 2013 standards. While this should not be taken not as hard evidence that a meter is not as accurate as it should be, we appreciate Dr. Klonoff’s efforts to put more responsibility into patient hands.
  • Dr. Klonoff explained that the SEG allows for clinician flexibility in how risk is assessed. After HCPs ranked the risk factor for reading vs. reference value on a nine-point, whole number scale, Dr. Klonoff and his team created a finer gradation through implementation of 0.5 risk units, providing 15 points of division (e.g., splitting the “none” risk zone into “slight risk for hypoglycemia,” “lower,” and “none”). Although the FDA advocates for use of this 15 division system (and the paper states, “We expect that for regulatory processes the 15-zone distribution will be used.”), Dr. Klonoff highlighted that the 15-zone distribution could be condensed into eight zones by disregarding directionality of risk, and the original nine zones could be condensed to five zones by similar methods. Additionally, Dr. Klonoff commented that a “pass-fail” model could also be used. For example, a certain distribution percentage points of data need to be above or below a cutoff score.
  • Dr. Klonoff outlined several advantages of SEG over PEG and CEG, including that the latter grids didn’t account for DCCT trial results that came out in 1993, analog insulins that emerged in 1996, new information about hypoglycemia, and raised accuracy standards for meters. For example, the CEG has a 20% error separating the A and B blocks. Additionally, it is unclear who the clinicians were that were surveyed in the development of the PEG; PEG is based off of 100 people surveyed at ADA 1994. He also reviewed the development of CEG and PEG, remarking that while CEG focused only on treatments and PEG only focused on outcomes, SEG blends both treatment and outcomes.
  • Dr. Klonoff briefly touched on the fact that this study supports BGM use in patients with type 2 diabetes not on insulin, since HCPs ranked risks across blood glucose levels the same in this population as in other groups. This is significant given the recent move in some areas (such as Oregon) to restrict strip access to patients on Medicare and Medicaid. We hope that such clinical data continues to be circulated among HCPs, payers, and CMS to demonstrate the HCPs do see the risk in not testing patients with type 2 diabetes.
  • The Surveillance Error Grid” was published on the Journal of Diabetes Science and Technology today. Along with individual authors, Dr. Klonoff also called attention to organizations pivotal in the development of the error grid, including the Diabetes Technology Society, the FDA, the ADA, the Endocrine Society, and the American Association for Advancement of Medical Instrumentation. Additionally, he acknowledged the hard work of error grid panel members, including 25 people in academia and industry such as Medtronic, Dexcom, LifeScan, Abbott, Bayer, Roche, and Sanofi – a strong circle of the leaders in Diabetes Care technology.
  • Notably, Dr. Klonoff added that the FDA has already begun using the SEG as a model to assess other measuring devices. We think that this bodes well for the FDA actually beginning to use the SEG as a post-market surveillance tool for BGM. Currently, the FDA does not conduct post-market surveillance, but assessing and enforcing meter accuracy remains a concern for both patients and providers, particularly in ICU settings where BGM use is still off-label.
  • The SEG was developed by surveying 206 clinicians and 28 non-clinicians asking what actions would be taken for each blood glucose value between 0 mg/dl and 600 mg/dl. The survey takers were then asked to assign risk (from “none” to “extreme”) if a reference value is misread either high or low. From there, all responses were averaged, allowing the SEG developers to impart more granularity on the grid by taking 0.5 risk units (Dr. Klonoff explained it similar to grades – although a student may only be able to get a 3 or 4 in a given class, if their grades are averaged across all classes, then they would be able to have a GPA of 3.5). See our coverage of methodology of developing the SEG in Day #1 of the Diabetes Technology Meeting in October 2013.

Questions and Answers

Q: Could you speak to how looking at the consensus between people responding to the survey translates into figuring out what level of granularity is appropriate for drawing boundaries between risk levels?

A: We had a large number of respondents, so we with mean; each person was his or her own control. We had an extreme idea of what the extreme scenarios would be. Additionally, often people would call blood glucose levels clinical significant when they were not in the “green” zone.

Comment: If you have a mean of 3.5; are most people saying 3 and 4 rather than 2 and 5?

A: Each value had a specific definition associated with it, and participants had to accept those definitions. Once defined, it is easier to chop values into gradations – it is like chopping up grades. It is very mathematically oriented.

Self-Monitoring of Blood Glucose in Non-Insulin Users – What is the Evidence?

Richard Grant, MD, MPH (Kaiser Permanente Northern California, Oakland, CA)

Dr. Richard Grant brought a primary care physician’s perspective to the discussion of self-monitoring of blood glucose. He argued that SMBG in non-insulin-using type 2 diabetes patients improves glycemic control only when prescribed in the context of a larger educational effort and as a tool to effect change in self-care or medication. In a review of 12 randomized controlled trials of patients with type 2 diabetes for at least one year, SMBG reduced mean A1c by just 0.26% compared to control treatment (lifestyle and oral), with mixed A1c results in individual studies. Additionally, Dr. Grant provided sobering results from the DISTANCE survey, in which 15% of patients reported that their SMBG results were not used by anyone to make adjustments to diet, exercise or medicine. Notably in Q&A, a PCP from Oregon criticized the DISTANCE study, commenting that the data were cited by the state of Oregon to restrict test strips for people with diabetes not on insulin. Dr. Grant was quick to clarify that he “would never have come to the conclusion that test strips should be restricted for all patients with type 2 diabetes not on insulin.” Rather, he would focus on individualizing care and on prescribing SMBG to patients who will benefit from it. With regard to the Oregon legislation, Dr. Grant even commented, “Using population-based prescriptions to restrict strips doesn’t make any sense... I do not agree with it at all.”

  • Dr. Grant also cited the well-known STeP study by Dr. William Polonsky et al. that highlighted the benefit a structured testing protocol: 1.2% A1c reduction compared to 0.9% reduction in the control group (Diabetes Care 2011). Patients with type 2 diabetes (n=256) were assigned to a 7-point testing schedule to be completed on the three consecutive days prior to study visit. The seven points included fasting, pre-prandial/2 hr postprandial at each meal, and bedtime tests. Unsurprisingly, this structured SMBG protocol required extensive education and diabetes care team support. The control group received quarterly clinic visits that focused specifically on diabetes-management and were given free blood glucose meters and strips as well as access to an office point-of-care A1c capability (n=227). Dr. Grant noted that though the control patients received good diabetes care, the structured testing still showed benefit.
  • In the DISTANCE Survey of the Kaiser Permanente Northern California diabetes registry, among patients who said that they used SMBG, 15% reported that their SMBG data were not used by anyone to make adjustments to diet, exercise, or medicine. Breaking down results into components, 37% of patients reported that both they and their and provider used data to change care, 34% reported that only they themselves used the data, and 14% reported that only their provider used the data. For providers not to use SMBG data is a “worst-case scenario” in Dr. Grant’s opinion. We found these data demoralizing, especially since they were used by the state of Oregon to justify restrictions on test strips in patients who are not treated with insulin.
  • Dr. Grant noted that most patients with type 2 diabetes not on insulin are being treated in primary care settings where PCPs have an endless list of competing priorities for the 15-minute visits. Primary care physicians have a typical patient panel of 1,500-2,000 patients, and type 2 diabetes prevalence makes up 10-25% of these patients (~200 patients with type 2 diabetes). Given the urgency of behavioral interventions on diet, exercise, smoking, medication adherence, etc., interpreting SMBG may rank at the bottom of PCPs’ priorities. Another challenge is that 80% of patients with type 2 diabetes have concurrent chronic conditions like COPD, heart failure, and obesity. However, Dr. Grant also noted that SMBG data could be used to leverage lifestyle counseling and optimize medication management.
  • Dr. Grant recommended that SMBG prescriptions should be made in the context of a shared-decision making framework to individualize care and ensure SMBG is the most time and cost-effective strategy.

Questions and Answers

Q: I was surprised that you didn’t consider in your review the PRISMA study in Diabetes Care in 2013 that is the largest comparison structure as SMBG in type 2 diabetes patients with more than 1,000 patients and up to one year follow up. The results showed significant reduction of A1c. This was thanks to a higher frequency in changes of medication exactly as you noted. I would emphasize that I wouldn’t consider SMBG useless in these categories of patients. Maybe we could discuss if it is cost-effective, but clinical usefulness in my opinion is clearly demonstrated.

A: I only included those in the original Cochrane comparison. As you said, A1c went down because of medication. I would argue that SMBG isn’t necessary to change medication. Also, I wouldn’t argue that SMBG is useless, but that if it is used, it should be used correctly. You can have excellent A1c control without SMBG.

Comment: I also disagree with your statement because if you have a patient on a sulfonylurea, hypoglycemia is a real danger and SMBG can help prevent this danger as well as improve quality of life.

A: I agree that hypoglycemia is important. There are a great number of patients not at risk for hypoglycemia, though. In these patients SMBG may not be as useful. Part of this discussion is not that it’s a bad thing, but that in larger context there are patients who don’t need it.

Q: I’ve also seen this particular research being used against us. In Oregon, the legislature cited this [DISTANCE study] to restrict strips for Medicare and Medicaid patients not on insulin. Have you looked at broad orals and tiered out non-hypoglycemia agents? I agree in part, but not in whole. We still see sulfonylureas as the number two prescribed medication in Oregon. Your data is currently being leveraged against us, but the data is not one group of people and I think you’d agree. Yes as SMBG may not be effective in primary care but this legislation is restricting SMBG and instead pushing the use of agents like sulfonylureas.

A: In the study I presented, we were really trying to predict the worst-case scenario. It doesn’t matter what they’re on. If you prescribe SMBG, someone should be looking at that data. Back to Oregon, I wouldn’t conclude that we should restrict test strips. We should use strips for certain patients. Using population-based prescriptions to restrict strips doesn’t make any sense. I do not agree with it at all. Some patients would be tremendously motivated. Equally, some patients wouldn’t benefit at all. That is the whole theme of the ADA/EASD recommendation; we need to individualize care to move levers.


Hypoglycemia Prediction Using SMBG Data and Patient Medication Information (397-P)

B Sudharsan, M Shomali

This poster presented the latest update to WellDoc’s exciting type 2 diabetes hypoglycemia prediction model, which was first unveiled at DTM 2013. The original model accurately predicted hypoglycemia risk (90% of the time) on the following day based on seven prior days of infrequent SMBG data (e.g., ~1 test per day) – this poster explored the additional benefit of adding patient medication information (drug dosing and class: short-acting insulin, long-acting insulin, pre-mix insulin, orals). Notably, the enhanced model was also constructed to predict the hour of the occurrence of hypoglycemia on the following day, a big step over the previous model’s aim to predict whether hypoglycemia would occur in the next 24 hours. Adding medication information significantly boosted the model’s specificity for accurately predicting hypoglycemia – 92% in the enhanced model vs. 70% in the previous SMBG-only model. The model’s sensitivity for predicting hypoglycemia remained high at 89%, comparable to the prior model’s 92% sensitivity. The study concluded that real-world SMBG frequency (~1 test per day) and medication information can provide adequate data to predict hypoglycemia in type 2 diabetes. The plan is to eventually incorporate this prediction module into BlueStar, WellDoc’s FDA-approved mobile prescription therapy for type 2 diabetes We continue to be impressed by the company’s approach, which centers on using data, algorithms, and real-time feedback to help patients better manage diabetes with minimal provider burden.

  • As we understand it, the WellDoc clinical and behavioral R&D team intends to optimize the patient education and coaching around predicted hypoglycemia, and then incorporate the hypoglycemia prediction model into BlueStar. Once incorporated, BlueStar’s automated, real-time coaching will educate patients about how to best manage and avoid hypoglycemia. From a patient perspective, this system would be an incredible asset to managing diabetes, particularly in those who don’t test very often or are at high risk of severe hypoglycemia.
  • The researchers used de-identified self-monitored blood glucose (SMBG) data and medication information from a randomized controlled trial (Quinn et al., 2011) to train a probabilistic model. For each data sample, 11 SMBG data points were used in the seven days prior to a hypoglycemic event (defined as SMBG <70 mg/dl). Control samples used for training contained no hypoglycemia on the eighth day. The model was constructed to predict the hour of the occurrence of hypoglycemia. In order to validate the model after training, 2,099 samples not used for training the model were presented to the model without the SMBG data from the eighth day. Sensitivity and specificity for predicting the hour of hypoglycemia or no hypoglycemia on day eight were then calculated. Further validation was performed with another distinct data set of 524 samples.
  • The model is grounded in a key assumption: most type 2s are not CGM users or high frequency testers. As a result, this model was designed to work based on a very real-world testing frequency observed in type 2 patients. Indeed, we think a model based on one test per day is pretty magical from a clinical and commercial relevancy standpoint. The hypoglycemia prediction is especially relevant in type 2s, where there are more patients on hypoglycemia-causing agents than there are type 1s in total.
  • We’d note that WellDoc has been pretty quiet following January’s $20 million Series A round of financing (led by Merck’s prestigious Global Health Innovation Fund) – the investment was expected to fund a dedicated sales force to regionally rollout BlueStar.

Oral Presentations: Diabetes Self-Management Education – Making a Difference

Comparison of Different Models of Structured Self-Monitoring of Blood Glucose in Type 2 Diabetes (14-OR)

Yi Sun Yang, MD (Chung Shan Medical University, Taichung, Taiwan)

Dr. Yi Sun Yang presented data on a comparison study (n=96) of three different models of structured self-monitoring of blood glucose in non-insulin using patients with type 2 diabetes: i) six-paired tests/week (48 tests/month), ii) three-pair tests/week (24 tests/month), and iii) seven-point profiles/week (28 tests/month) – described below. All three SMBG testing models met the primary endpoint, reduction in A1c, at three and six months, but the six-pair and seven-point testing models provided slightly greater reductions in A1c after six months (-1.7% and -1.8%; baselines of 8.8% and 8.9%, respectively) compared to the three-pair testing model (-1.1% from a lower baseline of 8.5%). The secondary endpoints of hypoglycemia occurrence and treatment change were similar among the three groups, but based on the PDSMS questionnaire, patients in the six-pair group reported more negative attitudes about their diabetes self-care – not surprising considering the higher volume of tests. The results are strong at face value, but are hard to interpret since there was no control group and all patients received diabetes education. In addition, 25% of patients did not complete structured testing, and it’s not clear how these patients were counted in the results. Still, the data are encouraging and in line with other studies (e.g., Polonsky et al., Diabetes Care 2011) supporting the value of structured SMBG in patients not on insulin.

  • 106 patients with type 2 diabetes not on insulin were randomized to one of three structured SMBG models: six-pair testing/week (n=37), three-pair testing/week (n=36), or seven-point testing per week (n=33). Diabetes education and self-care goal and regimen suggestions were also provided. The primary endpoint was change in A1c from baseline to 24 weeks and secondary endpoints were change of treatment, lifestyle modification (defined as any change in diet, exercise, or other self-care behavior), and questionnaires on patient-reported outcomes [WHO-5 general well-being scale, Perceived Diabetes Self-Management Scale (PDSMS), Short form - Problem Areas in Diabetes - Chinese version (S-PAID-C), and Center for Epidemiological Scale – Depression (CES-D)].
    • Model 1 (n=31): Six-pair testing per week for a total of 48 tests per month. For example, a patient on this structured model would test pre- and post-breakfast on Monday and Tuesday, then test pre- and post-lunch on Wednesday and Thursday, and lastly test pre- and post-supper on Friday and Saturday.
    • Model 2 (n=31): Three-pair testing per week for a total of 24 times per month. For example, a patient on this structured model might include pre- and post-breakfast tests for Monday, Tuesday, and Wednesday. Dr. Yang noted that this model provided three pairs of consecutive data to encourage immediate change diet and activity if necessary.
    • Model 3 (n=34): Seven-point testing per week for a total of 28 times per month. For example, a patient on this structured model would on one given day test pre- and post-meals for a total of three pairs and once at bedtime.
  • After six months, A1c declined in all three groups: -1.7% in the six-pair testing group (baseline 8.8%), -1.8% in the seven-point testing group (baseline: 8.9%), and -1.1% in the three-pair testing group (baseline: 8.5%). Additionally, fasting plasma glucose and postprandial glucose levels were significantly reduced in each of the three models. The proportion that did not complete structured SMBG were similar (25%). No severe hypoglycemic events were reported.
    • The secondary endpoints of hypoglycemia occurrence and treatment change were similar among the three groups, but patients in the 6-pair group reported more negative attitudes about their diabetes self-care based on the PDSMS questionnaire. As this model required approximately double the number of tests (48 per months) compared to the other two models (24 and 28 per month), the attitudes of increased burden are not surprising. 
  • The study had some important limitations: the sample size in each group was relatively small, and 25% of participants did not complete structured SMBG. We also note that there was no control group and 10 of the initially allocated patients were excluded from analysis either due to discontinued intervention or dropout during the follow-up period.
  • Baseline characteristics were similar across the three intervention groups.


Six-pair glucose testing

Three-pair glucose testing

7-point glucose testing

Age (years)




Duration of diabetes (years)




A1c (%)




BMI (kg/m2)




Fasting Plasma Glucose (mg/dl)




Postprandial Blood Glucose (mg/dl)




Questions and Answers

Q: How often were medical therapy adjustments made during the course of six months?

A: Patients visited at 12 and 24 weeks, so we had adjustments at the 12-week checkpoint.

Q: I noticed that you presented the data but not by intention-to-treat. If you did intent-to-treat analysis, what did you find?

A: We didn’t analyze that statistic yet, but we may calculate that later.

Insulin Delivery


Efficacy and Safety of Insulin Pump Therapy in Type 2 Diabetes: The Opt2mise Study (102-LB)

Y Reznik, O Cohen, I Conget, R Aronson, S Runzis, J Castaneda, S De Portu, SW Lee, Opt2mise Study Group

This poster presented the long-awaited results from the randomized, six-month Opt2mise trial, comparing insulin pump therapy (n=168) to MDI (n=163) in type 2 patients in poor control (mean A1c: 9.0%). Following a run-in phase, patients were 1:1 randomized to either use a pump or MDI. From a baseline of 9.0%, A1c declined by 1.1% in those on an insulin pump compared to 0.4% in the MDI group (p<0.001) after 27 weeks; 55% of the pump group achieved an A1c <8% vs. 28% of the MDI group. CGM data (baseline vs. six months) revealed no significant increase in hypoglycemia. Meanwhile, the group on pumps used 20% less insulin than those on MDI (p<0.001). HDL cholesterol improved by 8% in the pump group and declined by 7% in the MDI group (p=0.01). One episode of severe hypoglycemia occurred in the MDI group, while none occurred in the pump group. It was valuable to see this positive data from a randomized, controlled, multi-center study of pumps in type 2 diabetes – most importantly, we like that the investigators enrolled a population that could most use easier and more convenient approaches to insulin delivery. Given the high starting A1c of 9.0%, the magnitude of reduction (-1.1%) was perhaps not quite as high as some would have expected although patients may have been very hard to manage. We wonder if insulin titration could have been better, if a simpler device with on-body bolusing (e.g., Valeritas’ V-Go or CeQur’s PaQ) could have helped drive patients even lower, or if this simply underscores what a challenging population this is to manage. Results from this trial were published in The Lancet (Reznik et al.) shortly after ADA on July 3. 

  • Following a three-visit run-in phase to optimize MDI therapy, 331 patients were randomized to six months of either pump therapy (n=168) or MDI (n=163). The objective of the run-in phase was to optimize MDI therapy. All oral medications were replaced by metformin, and insulin therapy was intensified to >0.7 units/kg/day. During the study phase, the pump group initially used the same total daily insulin dose as before; patients randomized to MDI continued titration to target range. After six months, the MDI arm crossed over and switched to the pump. Both groups then spent months six through 12 on the pump during the study’s continuation phase. 
  • Patients had a mean age of 56 years, a mean 15 year duration of diabetes, a mean A1c of 9.0%, a mean BMI of 33 kg/m2, a mean total daily dose of ~109 units per day. The study had a high completion rate – 90% in the pump group vs. 96% in the MDI group.
  • From a baseline of 9.0%, A1c declined by 1.1% in those on an insulin pump compared to 0.4% in the MDI group (p<0.001) after 27 weeks; 55% of the pump group achieved an A1c <8% vs. 28% of the MDI group. As would be expected, patients in the highest tertile of baseline A1c realized the largest improvement in A1c after six months of pump use.

Baseline A1c Tertile




Difference in A1c Change (MDI-Pump)


-0.5% (p=0.01)

-1.1% (p<0.001)

  • Despite the improved A1c, the group on pumps used 20% less insulin vs. those on MDI (p<0.001) at the end of six months. The MDI group saw total daily insulin dose steadily increase from 106 units per day to ~120 units per day. Meanwhile, the pump group saw total daily insulin dose decline from 112 units to ~100 units per day.
  • CGM data (baseline vs. six months) revealed a significant improvement in 24-hour mean glucose, a significant reduction in hyperglycemia, and no significant increase in hypoglycemia. CGM data was collected over six days at baseline and at the end of the study. We assume the iPro2 was used, though it was not specified.




Change in 24-hour Mean Glucose

-23 mg/dl*

-6 mg/dl*

Change in time spent >180 mg/dl

-226 minutes per day**

-57 minutes per day

Change in time spent <70 mg/dl

+9 minutes per day

+ 5 minutes per day

*p<0.01; **p<0.001

  • One episode of severe hypoglycemia occurred in the MDI group, while none occurred in the pump group. There no episodes of DKA in either group. Four device-related serious adverse events occurred in the MDI group: two hyperglycemic hospitalizations (not DKA), one episode cellulitis, and one abscess.

Oral Presentations – Prandial Insulin Therapy

Effective Use of U-500 Insulin via Insulin Pump in Severely Insulin Resistant Patients

Anand Velusamy, MRCP (King's College Hospital NHS Foundation Trust, London, UK)

Dr. Anand Velusamy presented an uncontrolled study examining the impact of U500 insulin delivered via pump in very poorly controlled (baseline A1c: 10.4%), highly insulin resistant patients (mean daily dose: 306 units) with type 1 (n=3) and type 2 diabetes (n=11). A1c declined an impressive 1.9% at six months (n=14), 2.3% at 12 months (n=11), and was maintained out to 36 months (n=6). At the same time, total daily insulin requirements declined from 306 units at baseline to 244 units at 12 months and 250 units at three years. Weight did not change significantly, with patients gaining an average of 2 kg at 12 months and three years (baseline: 108 kg). Perhaps most notable were the cost implications – using U500 in the pump vs. U100 insulin was estimated to save ~2,200 pounds per patient per year (~$4,000). Though the study was uncontrolled and patients did receive nursing support, we thought these were very strong clinical results in a highly challenging population. Of course, this gels with efforts from companies like Insulet, Tandem, and Medtronic, who are all now actively pursuing type 2 focused products (Insulet’s U500 OmniPod with Lilly; Tandem’s 480-unit reservoir t:slim; Medtronic’s new type 2 business unit). The need is incredibly great in the severely insulin resistant population, and it’s great to see movement from industry that corresponds to the positive clinical data that continues to accrue on this front.

  • Patients had a mean age of 56 years, a mean of 16 years of diabetes, a mean A1c of 10.4%, a mean total daily dose of 306 units, and a mean weight of 108 kgs. Prior to the study, eight patients were on basal/bolus insulin regimens, one was on a mixed regimen, three were on U500, and two were on U100 in pumps (i.e., they changed their reservoirs every day).
  • The protocol for U500 initiation included a 30% reduction in total daily dose. Half of this reduction was used as flat basal replacement (we assume this means that 15% of the baseline total daily dose was pre-programmed as the initial basal). Patients used fixed dose boluses to start, though learned carb counting over time. A “bolus advisor” (we assume the pump’s bolus calculator) was used to provide corrective doses. Patients had downloads and telephone support to titrate insulin doses – this was likely a contributing factor to the robust reductions in A1c. We wish the study had been controlled to compare. The model of pump was not specified.
  • Most patients were either lost to death or bariatric surgery over the three-year study period. From a baseline of 14 patients, there were 11 patients at one year (one had bariatric surgery, two were “intolerant”), and six patients left at three years (two patients died, one had bariatric surgery).

Questions and Answers

Q: This is a difficult group of patients. Some were already on U500, some were on a pump. What about use of GLP-1 therapy in these patients?

A: That’s a very valid question. These patients were on very high doses of insulin. They had long-duration diabetes. Three patients had type 1 diabetes and two had renal impairment, which precluded use of GLP-1 analogs. A couple patients were working towards bariatric surgery, and we were trying to optimize control quickly. GLP-1 could be used in this scenario.

Q: I live in the diabetes/obesity belt in America. We use lots of U500. Regarding insulin dosing with meals, they were on a fixed dose with meal? They were not carb counting?

A: Totally daily dose was reduced by 30%. They were on a fixed basal rate. As a starting dose, they did fixed boluses, but had help from bolus advisors. They did not carb count initially, but eventually they did.

Q: How much time before the meal was the bolus given? Was there hypoglycemia data?

A: There was no increase in hypoglycemia in terms of rates. They did have a couple of hypos, but none needing any third party assistance compared to the previous regimens.

Symposium: Closed-Loop Insulin Delivery – One Step at a Time (Supported by a grant from The Leona M. and Harry B. Helmsley Charitable Trust)

Present State of Insulin Delivery

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

The renowned Dr. Buckingham presented a compelling and thorough overview of the history, present status, and future of insulin delivery systems. With respect to pens, Dr. Buckingham emphasized that the future of the market rides on the integration of insulin delivery with glucose information, noting that it doesn’t take much to transfer the technology given to pump users – bolus calculators and correction factors with insulin on board – to a display on a pen. Notably, Dr. Buckingham also argued that insulin sets are presently the “weak link” in insulin delivery, as sets remain relatively faulty and become occluded or result in unexplained hyperglycemia in many users. Dr. Buckingham noted that scarring and hyperpigmentation remain big problems at infusion sites as well. Finally, Dr. Buckingham focused on pump development, quickly reviewing the history of the market (even discussing the early DANA and APS systems!) before reviewing the Debiotech Jewel and Medtronic MiniMed 640G pumps, among others, that represent the future of the industry. He concluded by suggesting that the development of pumps with Bluetooth connectivity, integrated continuous glucose monitoring systems, and web connectivity for data sharing is a must-have as the industry looks to move toward the development of practical closed loop systems.

  • Referring to the development of pens, Dr. Buckingham noted that it “doesn’t take much” to integrate insulin delivery with more glucose information. He would like to see the technology given to pump users – bolus calculators and correction factors with insulin on board – displayed on a small screen on a pen. He referenced one product, approved in Europe, that is moving in this direction.
  • Infusion sets are the “weak link” in insulin delivery. Dr. Buckingham noted that we have developed decent technology for insulin delivery itself, but the issue is that patients are not necessarily getting the insulin. In particular, he cited data that 66% of patients using infusion sets suffer from occlusions or unexplained hyperglycemia. Given the disparity in approved duration of use for infusion sets (two to three days) and continuous glucose sensors (six to seven days), he also noted the limited period of an infusion set undermines the use of combination sets or patch pumps and sensors on a single platform.
  • Scarring and hyperpigmentation remain big problems at infusion sites. Dr. Buckingham reported that young patients often resort to using a sensor until failure, rather than obeying approval guidelines, in order to avoid the hassle of removing and replacing infusion sets.
    • In a small study performed by Dr. Buckingham examining infusion site failure, 64% of sites failed before seven days of wear and 30% of subjects suffered from hyperglycemia. This trial enrolled 20 subjects, ages 13-47, who all used Silhouette infusion sets filled with either Humalog or Novolog. They were asked to use the sets continuously for one week or until failure (identified via hyperglycemia). Based on these data, Dr. Buckingham emphasized the need for infusion sets approved for longer-term wear and noted that treatment of infusion site areas with hyaluronidase might offer a solution to improve insulin uptake.
  • Of all the pumps he reviewed, Dr. Buckingham spent the most time discussing the Medtronic MiniMed 640G. This system features a new user interface, pump design, transmitter, an and the Enlite 3 sensor. Dr. Buckingham was impressed by the system’s predictive low glucose management algorithm, which anticipates blood glucose 30 minutes into the future and can suspend the delivery based upon that forecast. As of Medtronic’s last update, launch in the EU was expected by April 2015, while a US trial was just getting started ( Identifier: NCT02130284).
  • Dr. Buckingham drew attention to the integrated use of CGM, Bluetooth connectivity, and web connectivity as three important elements that must be incorporated into closed loop technology going forward. He also mentioned that it would be helpful, for the consumer and researcher, if common communication standards were established between devices.

Questions and Answers

Q: I was wondering whether you could speak to the using of plastic vs. glass reservoirs?

A: I don’t think that’s a big issue. We tested insulin coming out for a week, and we didn’t see any increased fibrillation. I don’t see any glass tubing coming out of a body in the future. That seems fragile to me. Alternatively, Ed Damiano was trying to use DMSO is his bi-hormonal system, but that’s toxic to a lot of chemicals in reservoirs.

Q: Could you speak about any studies on site failure? On subcutaneous failure and scaring related to that?

A: I’ve got an ongoing study regarding this issue, so I can’t speak to the results.

Comment: I have trouble with patients not inputting all their lows into their system. It would be helpful if we had pumps that mandate that data goes into the system.

A: The logic here escapes me. The FDA does not want us to have devices that can wirelessly transmit all the data to the cloud. You’re right that numbers sometimes get transposed. Patients don’t do it immediately. There are a lot of issues. Electronic meter transmission is important and should be done.

Q: Regarding site failure, it sounds like body fluids are coming back up into the catheter. Could we get a one-way valve?

A: You’re always on the innovative side.

Q: Are there any other ideas for improving infusion sets beside hyaluronidase?

A: You could co-infuse an anti-inflammatory. You could coat the infusion set with an anti-inflammatory. There are other ways to reduce inflammation, too. And also, the question is: How much of the inflammation is the material? We saw no difference in using a steel versus a Teflon catheter in our testing. And the inflammation may not be the insulin itself. It may be stabilizing agents that are in insulin. So I’m not sure where it’s coming from.

Dr. Irl Hirsch (University of Washington, Seattle, WA): Clearly, we have not done well with infusion sets. However, in some adults, we don’t see inflammation, even after 20-25 years of pump therapy, but we do see that absorption is all over the map. In these cases, we’ve taken pump holidays, taking patients off pump therapy for three to four months. So we’re getting these people who are getting such poor absorption, but not because of occlusions, but because of scarring. It really goes to show that we don’t know anything about what’s happening under the skin. And what’s concerning is that now we’re putting children on pump. If they develop scarring, they’re not going to be able to continue on the pump in 30 years, or even worse, begin closed loop therapy.

A: I hope you’re wrong.

Q: Are implanted pumps going away?

A: The person who could best answer that question is Eric Renard. Eric?

Dr. Eric Renard (Montpellier University, Montpellier, France): There is nothing happening on this front in the US. But we have programs in France that should expand in Europe, especially to Germany. It is a clear answer for the problem for inflammation with infusion sites. In this case, using the peritoneal route is very effective.

Q: Is insulin precipitation due to the loss of cresol?

A: I don’t know.

Meet-the-Expert Sessions

Insulin Pump Therapy

Irl Hirsch, MD (University of Washington, Seattle, WA)

Dr. Irl Hirsch provided a concise overview of insulin pump therapy, covering bolus calculators, pump download best practices, and several patient case studies. His talk emphasized the importance of proper bolus calculator settings (his clinic defaults to five hours), focusing on the nighttime basal first, relying on certain statistics (standard deviation times two or three should be less than the average blood glucose), and coaching patients on optimal insulin dose timing. He walked attendees through his approach to several typical cases, showing all Medtronic downloads (“over 70% pumps used in this country are Medtronic”). His presentation did a good job of tying overwhelming pump download data to specific patient behaviors, which translated very clearly into clinical recommendations. Said Dr. Hirsch, “I know this is a session only on pumps. But it’s getting harder and harder to separate pumps from sensors.” As a testament to the forward-thinking, tech-savvy ADA attendee base, ~75% of audience members had patients on CGM.

Data, Digital Health, and Connected Devices

Special Meeting: DiabetesMine D-Data Exchange

App Demos

  • NightScout/CGM in the Cloud is a remote monitoring platform for people wearing CGM. The system consists of a Dexcom G4 Platinum CGM receiver wired via USB cable to an Android phone. That links up with a database, a cloud server, and an app running on a glanceable display. The goal is safety and peace of mind. See the online instructions here for putting the system together. The entire project was crowdsourced and used open source software development. About two months ago, a tipping point was hit and a Facebook group was created. This “caused an avalanche” and was cited as a “shining example of #WeAreNotWaiting.” The Facebook group has 1,270 users and gets 50-100 new people every day.
  • Tidepool showed off the latest version of Blip, which is now in a clinical trial at UCSF. As a reminder, this diabetes data platform is intended to display device data together in a very sleek and highly usable web interface – Mr. Look told us that one endocrinologist was ecstatic when he learned that Blip can seamlessly integrate Medtronic pump and CGM data. Mr. Look ran us through several fascinating examples where he communicated with his daughter’s endocrinologist and other members of the Tidepool team to troubleshoot out of range blood sugar numbers. We cannot wait until this rolls out.
  • We got a look at the latest innovation in the sleek mySugr Diabetes Companion app. This app has a beautiful design and encourages consumers to “tame their diabetes monster” by logging glucose values, insulin, exercise, mood, etc. The newest innovation uses the smartphone camera and image recognition to scan glucose meter values into the app – it’s cable free and very cool! As we understand, this was a massive coding undertaking and required rewriting much of the standard optical recognition technology.
  • Do-It-Yourself-Pancreas (#DIYPS) is an impressive cloud-based CGM alarm system/remote monitor with predictive analytics developed by Dana Lewis and Scott Leibrand.
  • Joslin HypoMap powered by Glookoread our report from earlier this week on this hypoglycemia unawareness survey and web-based module, spearheaded by the very smart Dr. Howard Wolpert.
  • LabStyle Innovations’ Dario is an all-in-one smartphone BGM that plugs into the headphone jack of smartphones. The meter has been soft-launched in Europe and is under FDA review in the US. Read our previous report on LabStyle Innovations.
  • Galileo Cosmos, a project of Anna McCollister-Slipp, is focused on data visualization.
  • Ben West’s “Let’s chat with an insulin pump” (hacking a Medtronic pump).

“Unconference” Interactive Group Discussions

This session featured breakout groups with discussion centered around several topics. We’ve detailed the group leader summaries below.

  • Engaging with Regulatory Bodies – Type 1 dad Mr. Lane Desborough (Medtronic Diabetes, Northridge, CA) summarized what sounded like an excellent small group discussion that included the FDA’s Dr. Stayce Beck.  He noted that the FDA really does want to engage and better understand what #WeAreNotWaiting is. He explained that the FDA’s job was significantly easier 15 years ago, as only a handful of companies could actually make a medical device. Now, one particularly motivated person can create a medical device. We thought this was an incredibly astute point, especially considering the agency’s limited resources. That said, the #WeAreNotWaiting movement is about individuals taking things into their own hands, which creates a situation with “multiple shades of gray” on what constitutes a medical device. Certainly, an artificial pancreas is clearly a medical device, and a Fitbit is clearly not a medical device – but what about a secondary display of CGM data on a smartphone or tablet? And in the case of Nightscout/CGM in the Cloud, does putting open source code on a website count as “distributing” a medical device? (In speaking with the FDA’s Dr. Stayce Beck after the session, she told us that Nightscout/CGM in the Cloud is indeed a regulated medical device, though the regulation is “tricky.” Technically, the software developer is responsible for pursuing regulatory approval, though in the case of open source development, it’s less clear. Still, Dr. Beck seemed excited and encouraged by the development, and noted that CGM in the Cloud is exactly the kind of device that the FDA wants to see in the marketplace.) Mr. Desborough emphasized that all of these shades of gray can be resolved by following the FDA’s pre-submission process – it’s free, easy, and there are timelines by law with how fast the FDA must respond. This process is as simple as giving the FDA a two-page document and a list of unanswered questions.
    • A questioner wondered about regulation linked to personal development of a secondary CGM display, and when such a product crosses the line and becomes regulated. The audience seemed to conclude that building such a device for personal use is okay, but once it is shared with one person, it drifts into regulation. Still, it was noted that the terms “share” and “distribute” fall squarely into the shades of gray area.
  • Barriers to Device and App Adoption – On the device side, cited barriers included: cost/insurance; “it reminds me of my diabetes (e.g., extra alarms that you don’t have control over); cost-benefit analysis/short-term psychological focus (i.e., benefits of better control accrue over long term); something on my body; devices only give negative feedback and fail to give positive feedback; people get into established routines over time and are resistant to change from what works; cool, consumer-friendly design; prescribers not prescribing these devices, in part due to data management; and basic awareness that these devices are even available. On the app side, cited barriers to adoption included usefulness of data and manual vs. passive data collection. Attendees pointed to the need to take consumer-friendly design into account, the importance of collecting data mindlessly (vs. manually logging), including a social component (kudos, commenting, comparisons vs. other people), allowing goal setting, integrating challenges, and tracking personal bests (i.e., providing positive feedback and feelings of success). The idea of comparing one’s diabetes data to others was pointed to as both a motivating or demotivating factor, depending on the patient.
  • Data visualization – Audience members agreed that there are different ways of representing data, including use of log scales, acute vs. long-term care, and systems like Tidepool. Some pointed to the use of heat maps to really understand long-term care processes. All agreed that the bar is incredibly low right now, since any visualization is better than a logbook, and rates of downloading are so incredibly low. One group member pointed out that simply liberating the data is the first big challenge before optimally visualizing it.
  • Meeting Device Makers – This discussion centered on the challenge of adopting universal standards for diabetes devices. Group members expressed frustration that standards are not a priority amongst CEOs, VPs, people in charge of operations, and those in marketing (“It has to get into their top three. It’s not even in their top 10”). Some argued that transferring to universal standards does have a return on investment – products can potentially get to market faster and might even have an expedited regulatory review (though this remains to be seen with the FDA). The automobile and World Wide Web were two examples of the consumer market driving standards. A representative from PCHA/Continua highlighted that universal diabetes device standards are not that far away – the CGM standards document “is technically sound and stable and ready” to be in the Continua guidelines within a year. Universal standards on insulin pump data read outs are expected in 2015, and insulin pump control and command standards are expected in 2016.
  • Open source development and device hacking – This group allowed the makers of  Nightscout, Tidepool, and the Do-It-Yourself Pancreas to demo their devices.

Call to Action: #WeAreNotWaiting Pledge & Goals for Fall

Howard Look (President/CEO, Tidepool, Palo Alto, CA)

Tidepool’s Mr. Howard Look wrapped up the day with an inspiring talk and call to action. He shared unanswered questions surrounding data and made a case for device makers to open their thinking, data, and device protocols.

  • “#WeAreNotWaiting – this brilliant hash-tag means so many things to so many people” – For peace of mind that our children with type 1 diabetes are safe. To allow people with diabetes to have a choice in how they see their own diabetes data. To bring together the best and brightest minds from around the world to help make things better for people with diabetes.
  • “We’ve been talking to device makers a lot. We’re making a slow progression here, and it’s a stepwise forward progression in helping them become comfortable.” Mr. Look described the progression of data as a ladder – lower tiers entail the release of less controversial data, while higher tiers are often more challenging for companies to part with:
    • Does the patient own his or her own health data?
    • Can patients donate and repurpose their data? Mr. Look argued that these first two are an obvious pass/fail test – “We must all agree on this.”
    • Cloud services – machine accessible APIs? Devices – documented protocols?
    • Devices – allow identification?
    • Data format/protocols complete and unambiguous?
    • Provide safety/efficacy diagnostic data.
    • Source code available for inspection?
  • Mr. Look wondered, “What if there were an open and transparent scorecard?” He explained that Tidepool has not done this yet, but device makers must decide where on the seven-step ladder they feel comfortable. “You don’t have to have them all,” he said, “but what do you want to be able to say about your device?”
  • Who owns the data? At first, it’s an easy answer – “It’s my disease, it’s my data.” However, Mr. Look’s deeper dive revealed how much more complicated it is – personal health data, contextual data, device identification data, diagnostics/safety /efficacy/proprietary data. The balance between patients’ owning this data and device makers owning this data is an important and critical question confronting the field.
    • Personal health data: blood glucose data, basal rate settings, IOB, ICR, ISF, basal rate change events, boluses. It’s fairly uncontroversial that patients own this data.
    • Contextual data – location, activity tracker, meal information, calendar events.
    • Device identification data – “where it gets interesting.” Device identification, brand, model, and revision.
    • Diagnostics/safety/efficacy/proprietary data – ISIG values, pump occlusion pressure, internal temperature, battery recharge cycles, internal error logs. “As a device maker, you might not want your competitor to know this.”
  • “I’ve removed the name of the manufacturer, but here is an example of a troubling end user license agreement (EULA)” – “Any data submitted through [the service] shall be property of [the service] and you hereby waive all right, title, and interest to the submitted data.”
    • While Tidepool has not decided on its EULA, Mr. Look hopes it looks something like this – “Any data submitted by you through our service is owned by you. We are stewards of your data.” “If you like to make your data accessible to someone else, or to different software, just let us know.” “If you like to donate your data to anonymous research, just let us know.”

New Clinical Collaborations for D-Data Innovations

David Kerr, MD (Director of Diabetes Research and Innovation, Sansum Diabetes Research Institute, Santa Barbara, CA)

Dr. David Kerr (newly transitioned to Sansum from the UK) shared his view on creating a “Smart Diabetes Society” – one with devices that are open (interoperable); based on a cloud architecture; adaptable (to physiology and through learning); are social (big data), effectiveness-based (evidence, trust), incentives focused (stickiness), and use data semantics that both machines and humans can understand. He argued for adaptive diabetes systems that save patients and clinicians time, not simply software that collects and spits data back out. Notably, Sansum is focusing efforts on diabetes and exercise through a big data collection project hosted at

  • Dr. Kerr noted some key areas where diabetes technology can really improve.
    • Unattractive – “You would not choose to wear some of those devices. They are plain ugly.” We need more consumer electronic-like devices.
    • Impersonal technology – “make it mine”
    • Inaccessible technology – visual, functional, cognitive
    • One-size-fits-all technology – reservoirs, tubing, strips
    • Unconnected technology – it should sync with phone, records, and be social
    • Unintelligent technology – we need education and learning
  • “Clinicians want to do less; not more. Doctors are tearing their hair out and saying, ‘I don’t want all this data.’” Dr. Kerr believes it is more important that the individual has the data and the machine/algorithms support the learning. He emphasized the need to create “adaptive diabetes systems.”
  • According to the 2014 Diabetes App Market Report, mobile diabetes apps are currently used by only 1.2% of the target group. The analysis also revealed that 14 diabetes app publishers have 65% market share of the app market. 

Continua’s New Personal Health Alliance – Applications to Diabetes Care

Horst Merkle (Vice Chair, Personal Connected Health Alliance [PCHA]; Director, Diabetes Management Solutions, Roche Diagnostics)

Mr. Horst Merkle discussed the work of Continua/PCHA to drive towards interoperability and universal device standards for diabetes devices. He noted that the FDA acknowledges IEEE 11073 interoperability standards, and Continua developed the device profiles of the seven mentioned devices, including the glucose meter profile (10417). Continua/PCHA are looking to expand the guidelines for diabetes soon – insulin pump data read out (expected in 2015), insulin pump control and command (expected in 2016), and CGM (the slide said “2016,” though a latter comment suggested this could come in 2015). Mr. Merkle emphasized that these standards are incredibly important steps as the field moves towards the artificial pancreas.

-- by Melissa An, Adam Brown, Hannah Deming, Varun Iyengar, Hannah Martin, Emily Regier, Joseph Shivers, Jenny Tan, Sanjay Trehan, John and Kelly Close