June 9-13, 2017; San Diego, CA; Full Report – Closing the Loop, Pumps, and Insulin Delivery – Draft

Executive Highlights

This document contains our coverage of closing the loop, pumps, and insulin delivery at ADA 2017. Immediately below, we enclose relevant themes from this category, followed by detailed discussion and commentary. Talk titles highlighted in yellow were among our favorites from ADA 2017; those highlighted in blue are new full report additions from our daily highlights coverage.

Note that there is definitely some subjective overlap with digital health (insulin dose titration, apps) and glucose monitoring (CGM), and some talks may appear in only one or multiple reports.

Themes

Automated Insulin Delivery Field (Finally) Gets Commercial: Medtronic MiniMed 670G, Tandem PLGS, Insulet Horizon, Diabeloop

• The automated insulin delivery (AID) field took on a more commercial feel this year, led by the broader launch of Medtronic’s MiniMed 670G hybrid closed loop. The rollout to 20,000+ priority access program participants was announced at the start of FDA, and followed pivotal trial data shared at ADA one year ago and the earlier-than-expected FDA approval in September. At this ADA, we saw encouraging data from the 670G “customer training phase” (n=730), highlighting sustained glycemic improvements (from 63% -> 74% time in range, a 9-mg/dl improvement in mean glucose), time in Auto Mode (92%), and sensor wear time (95%) that resembled the pivotal study. Meanwhile, the MiniMed 670G product theater drew a standing-room-only crowd (five deep in the back!) to hear Drs. Rich Bergenstal and Jennifer Sherr share HCP tips, expectations setting, case studies, and CareLink reports. In other words, the discussion moved to, “How do we roll this product out in clinical practice?”, rather than, “Is this product coming?” On the marketing front, Medtronic blanketed the city with 670G ads, and the company’s during-ADA Analyst Briefing shared strong confidence in the device and patient enthusiasm so far. Medtronic believes it is “2-3 years ahead” of the competition (we think it’s less if companies hit their timelines), and a series of incremental improvements to the 670G are planned near-term (no specifics provided, but this should include pediatric approval and Bluetooth-connectivity, among others). The MiniMed 670G Priority Access Program launch is expected to run into the fall, and in parallel, Medtronic is taking orders for non-priority-access-program individuals that want to get on the device.  
• Also on the commercial side of AID, we saw very encouraging new data from other products in development: Tandem’s PLGS device, Insulet’s Horizon hybrid closed loop, and Diabeloop’s system. All showed strong outcomes on the algorithm front, with launches planned in the 2018-2019 time frame. We’re very glad to see the makings of (hopefully) a thriving commercial market where many products can compete – not just on algorithms and glycemic outcomes, but on the increasingly important metrics of device form factor, appeal to MDIs, user experience/interface, business model, connectivity/remote upgrades, ease of prescribing, and far beyond. We’ve been talking about the “academic research-commercial product chasm” in this field for years, and finally, it felt like this ADA had more tangible industry movement to get actual products to market. (See our competitive landscape here.) That said, the insulin pump field is also more challenging commercially – e.g., no updates from J&J, Roche was not even in the exhibit hall, Tandem’s stock price took a bit of a hit – and many players like Bigfoot are expanding to focus on MDI dose titration too. The latter is great to see, and it will be fascinating to see how patients segment between MDI+CGM+decision support and automated insulin delivery (pump + CGM).
• The closed-loop academic research community is looking ahead, with a focus on meals and exercise, algorithm personalization, and new populations (e.g., hypoglycemia unawareness, in hospital). With a pipeline of commercial products coming, we’ll be fascinated to see how the academic research community’s focus evolves – where can it push the field forward? How should dollars be allocated to closed-loop research to benefit the most people with diabetes? Our own Adam Brown raised this as a topic of panel discussion at the annual JDRF Closed Loop Research dinner: what will drive the field’s growth in the coming years, and how should the portfolio of research align? We were glad to see the Cambridge team again present highly compelling data on closed-loop in type 2s in the hospital (a major area of need and potential to save lots of nursing time), while the UVA group showed the clear value of closed-loop in those with hypoglycemia unawareness (significant benefits on time <70 mg/dl). For now, the leading academics continue to focus on optimizing algorithms, since obviously tighter control with less user hassle will be key for making the products as great as possible. Key areas of research on the radar include: making meal control more automatic (faster insulins, adaptive algorithms, other hormones like glucagon or amylin), adding sensors that can inform systems of activity (jury is still out on this one, since metrics like heart rate can be a false signal), and perhaps even implantable systems (faster meal response and less hypoglycemia, though obviously a different scalability question from current subcutaneous systems). The insulin vs. bihormonal debate was a bit quieter and more data-based this year, which is fantastic in our view – let head-to-head data drive what patients, providers, and payers prefer.

Companies are Getting the Message: User Experience, Patient Feedback, and Integration with Consumer Electronics Really Matters

• More than ever, device companies are hearing the message: user experience, patient feedback, and consumer electronics integration are critical. Dexcom secured Android G5 approval just before ADA and launched on the Google Play store shortly thereafter – the app already has between 500-1,000 downloads and received an update just after ADA on June 18. Also on the eve of ADA, Apple’s Worldwide Developer conference talked about Dexcom’s G5 on stage, which will communicate directly from transmitter-to-watch in the next version of WatchOS – talk about a victory for the patient experience and a major sign of Apple’s commitment to this partnership! Dexcom was a pioneer on mobile medical apps and it shows in the product’s continued uptake, particularly in MDI. The Clarity mobile app also got a terrific facelift, giving users expanded retrospective data reporting right on the phone. Insulet was a standout in this area too, headlined by a first look at the upcoming Omnipod Dash platform. The locked down Android phone has substantially upgraded the PDM user experience, while the next-gen Horizon system (putting automated insulin delivery on Dash) has undergone six usability studies already (31 unique participants to date, including some 670G users). Said Program Director of Advanced Technologies Jason O’Connor, “It’s so critical to not rush a product to market, just because it has functionality. It has to have the right user experience to deliver a product that is going to integrate into people’s lives. We may not be the first to market, but it does mean we’ll deliver an exceptional product.” Insulet has even solicited feedback from OpenAPS community members, which would have been unheard of a few years ago. Tandem’s compelling media day also reminded us that the Device Updater – allowing t:slim X2 pump users to upgrade their software from home, including to add G5 CGM integration and automated insulin delivery – will be a key differentiator going forward. We expect the most successful diabetes devices/apps will be driven by continuous software updating and learning from users, just as the most successful consumer technology is. That said, software is a whole different ballgame than hardware, and continuously building/maintaining apps for a growing number of devices can be burdensome – which companies will nail this?

The DIY Community Going Strong – Autotune, Loop, and Pushing Industry to Move Faster

• The do-it-yourself (DIY) community is still pushing the envelope of diabetes technology. Patient innovators Dana Lewis and Scott Leibrand shared a late-breaking poster on Autotune, an algorithm that automatically recommends changes in pump settings based on CGM data. Even in an engaged group of users, most found the recommendation helpful and have reportedly changed their pump settings accordingly. We also heard plenty of hallway chatter on Loop/RileyLink, the DIY system that runs hybrid closed loop off an iPhone app. Those we spoke to praised the overnight time-in-range, the algorithm’s customizability (e.g., set point, aggressiveness), the ability to bolus from an iPhone (with TouchID) or Watch, the on-phone user experience (showing exactly what the algorithm is doing), and seeing projected glucose after carbs are entered. At the Diabetes Mine D-Data Exchange, major industry players (Medtronic’s Dr. Fran Kaufman, Insulet’s Dr. Trang Ly,  Bigfoot’s Bryan Mazlish, Tandem’s John Sheridan) showed encouraging openness to engaging with the DIY and broader patient communities. We hope that moving forward, companies can give these innovators a “sandbox’ to play in, allowing lead users to help drive innovation. Dexcom is taking a lead on this with its developer APIs, which will launch later this year and allow third parties to access retrospective data (three-hour-delay), create and manage pre-commercial (prototype) apps, play with simulated (sandbox) data, learn how to become a Dexcom data partner, and even submit an app for commercial approval. Hopefully the time window will shrink, driving an ecosystem of useful real-time decision support. We’ll be fascinated to see how different companies harness the brilliance of this community – there is obviously a fine line to tread here, but one with significant upside in our view. 

New Devices in Exhibit Hall, but Clear Changes in Device Industry – Startups had Booths, While Roche Did Not Appear

• We were glad to see several notable new tech products in an ADA Scientific Sessions’ exhibit hall for the first time: Medtronic’s MiniMed 670G hybrid closed loop and Guardian Connect CGM for MDIs, Insulet’s new OmniPod Dash PDM, J&J’s OneTouch Via bolus-only patch insulin delivery device, Dexcom’s Android G5 and upcoming touchscreen receiver, and many others.
  
  • Medtronic’s newly launched MiniMed 670G was probably the most talked-about device at ADA 2017, drawing sizeable crowds to the booth and dedicated product theater. The on-device experience was what we expected – a definite pump screen and display improvement over older Medtronic pumps, but still definite room to improve on simplicity, a consumer-grade experience, and connectivity/remote software updates. The company is managing expectations well, though we’ll have to see what early reviews are like once it really rolls out.
  • Insulet’s new Omnipod Dash PDM won our award for most improved device, bringing marked advantages in user experience on the locked-down Android phone (FDA submission in 2H17). Insulet’s emphasis on leveraging user feedback really shined in its product theater and in-booth demonstrations. We think this device will be quite well received from the loyal Omnipod user base, and it clearly gives Insulet a logical path to add Horizon automation and concentrated insulin.
  • J&J’s OneTouch Via bolus-only insulin delivery patch device drew crowds to see the device in the flesh for the first time in an exhibit hall. Reps disclosed that the product’s updated manufacturing process (submitted in November) has received FDA 510(k) clearance, but did not give any specifics on the launch timing front. It was great to finally get to hold the device, and it was clear that booth-goers were intrigued – we overheard one who was amazed by the patch’s small form factor. 
  • Dexcom’s just-approved Android G5 and new touchscreen G5 receiver were on display for the first time following recent FDA approvals. We would describe the new receiver as “built like a tank” – it’s bigger than the current option, and the touchscreen is more resistive than one found on a consumer-grade phone. This thing looks like it could be thrown at the wall and not break, which brings two deliberate advantages: (i) fixing the reliability issues that have challenged Dexcom’s current receiver; and (ii) addressing the Medicare reimbursement requirement that the durable G5 receiver last for three years. (This latter point we only heard in hallway chatter.) We assume once this new receiver is out, more non-Medicare users will switch to displaying data on a smartphone – it definitely loses some cool factor relative to the current iPod Nano-like receiver.
  • Abbott’s FreeStyle Libre Pro and Tandem’s t:slim X2 were not new to US exhibit halls, but did draw notable crowds interested in the next-gen devices with meaningful improvements – the former to professional CGM and the latter to Bluetooth connectivity/remote software updating.
• One of our first observations upon entering the massive exhibit hall was: “Where’s Roche?” In fact, the major BGM player did not have a booth at this year’s conference to show off its new Accu-Chek Guide BGM or soft-launched Insight CGM system in Europe. This was not highly surprising from one perspective – it’s a tough time for Roche’s US BGM business and it has discontinued new US pump sales – but the juxtaposition with other players was noted. Glooko, Companion Medical, One Drop, and many other smaller device companies were indeed present in the massive ADA hall. This was perhaps a sign that: (i) Roche is being more choosey with its marketing resources (do exhibit hall booths have good ROI relative to other investments?); (ii) reflective of Roche’s smaller geographic sales base in the US; (iii) fewer new Roche products to showcase at this ADA; and/or (iv) something else. While bigger device companies like Medtronic, Abbott, BD, etc. were of course present in a big way in the hall, it felt like a greater number of small, innovative players were present this year and the average age is definitely younger! It will be fascinating to see how large AND small players drive innovation in the years ahead, and whether the current trends of tech partnerships will continue/increase at the next few ADAs.

Detailed Discussion and Commentary

Oral Presentations: Pumps and Loops

Inpatient Safety and Feasibility of the Tandem Predictive Low Glucose Suspend (PLGS) Insulin Pump System

Gregory P. Forlenza, MD (Barbara David Center for Diabetes, Aurora, CO)

BDC’s Dr. Greg Forlenza presented very positive, first-ever data on Tandem’s Predictive Low Glucose Suspend (PLGS) device, showing strong results in a small (n=10) overnight inpatient feasibility study: 100% suspension and resumption of insulin delivery as intended (per algorithm); a median reference blood glucose of 88 mg/dl at suspension and 83 mg/dl at resumption; and a median CGM nadir at 71 mg/dl and a peak at 91 mg/dl after suspension. Notably, there were no reference glucose values <60 mg/dl, a clear sign that the system is really preventing hypoglycemia in the overnight setting. The PLGS algorithm suspended insulin a median of three times per patient per night, with a total median suspension time per night of 117 minutes (patients typically had one long suspend session every night). The study used a modified commercial version of the t:slim G4 pump, and Tandem has very wisely chosen a well-studied PLGS algorithm used in several academic studies (Buckingham, Maahs, et al.) – said Dr. Forlenza, “It’s basically just tenth grade math!” (more details below on the mechanics). Consistent with what Tandem has said, the PLGS pivotal will start this summer (IDE approved in May) and update hardware to the Dexcom G5 and t:slim X2. Dr. Forlenza revealed the first study details we’ve heard: n=90; crossover design comparing PLGS to sensor-augmented pump therapy, with three weeks in each arm; patients 6+ years old; and a primary outcome of reduced time <70 mg/dl (thank you, FDA!). Tandem is clearly benefitting from being second-to-market and using a well-studied algorithm, as this regulatory program is far less onerous than what Medtronic went through with the 530G threshold suspend studies (ASPIRE) – of course, PLGS is also safer and more effective than threshold suspend. Assuming the pivotal runs on time and the submission gets in quickly later this year, Tandem’s estimate for an early 2018 launch seems very realistic to us.

• The Tandem PLGS algorithm has been well studied in academia and is easy to understand: it uses a linear regression of the last four glucose values to project a glucose value 30 minutes into the future. If that projected glucose value is expected to be less than 80 mg/dl, or if the current CGM value is less than 70 mg/dl, insulin delivery is suspended. Delivery resumes when CGM values begin to rise, preventing rebound highs.
  
  • The PLGS algorithm is similar to that used in several large in-home studies. One is often quoted by Dr. Aaron Kowalski (showing the tremendous burden/danger of overnight hypoglycemia in youth with diabetes): “Predictive Low-Glucose Insulin Suspension Reduces Duration of Nocturnal Hypoglycemia in Children Without Increasing Ketosis” (Buckingham et al., Diabetes Care 2015 and Maahs et al., Diabetes Care 2014). The Tandem pivotal trial will be a three-week crossover study. The previous studies were 42 nights, but the system was randomized to be active or inactive each night. In the Tandem pivotal study, it will be active for three weeks or inactive for three weeks, and the participants will know if it is active.
  • Importantly, there are no active alarms for when Tandem’s PLGS turns off or turns on the basal insulin delivery, unless the patient chooses to activate the alarms. This is a tremendous win for the user experience and correct a major hassle in the MiniMed 530G threshold suspend device.
  • Said Dr. Forlenza, “The bottom line from the research is that PLGS is safe and it works. In adults, PLGS typically reduces hypoglycemia by 50-80%. In children, it’s a >50% hypoglycemia reduction.”
• The Dexcom G4 CGM was placed 1-5 days prior to admission, though we’re glad to see the investigators allowed day 1 CGM use (when all CGMs are less accurate). Patients were admitted to the CRC at 7 pm after eating small dinner, at about 6 pm. The study pump was placed and patients were given a correction bolus with a target of 100 mg/dl. As is typical for PLGS studies, basal insulin was manually increased until a system generated suspension occurred.
• “The study will collect data that will be used for planning a pivotal study, and the study data are intended to be used to support a Premarket Approval (PMA) application.” - ClinicalTrials.gov page

Closed-Loop Systems Compared with Insulin Pump Therapy for Outpatient Glucose Control in Type 1 Diabetes: A Meta-analysis

Alanna Weisman (University of Toronto, Toronto, Ontario, Canada)

An in-depth, just-published meta-analysis of 27 closed-loop studies (585 subjects) presented by University of Toronto’s Dr. Alanna Weisman determined that closed loop therapy confers a mean increase in time in range of 12.6 percentage-points vs. conventional pump therapy (over the course of a day, that translates to more than two additional hours!). The paper goes on to break down efficacy measures by time (daytime vs. overnight), algorithm, single vs. dual, etc. Overnight, the time in range improvement over pump therapy is 3 percentage points greater than the improvement seen in the daytime. The publication describes time in range improvements from MPC and Fuzzy Logic algorithms in the mid-teens, while PID algorithms only boosted time in range by 7%. Dual hormone systems increased time in range by 19.5 percentage points, while single hormone systems had a slightly lower benefit at 11 percentage points. In Q&A, Cambridge luminary Dr. Roman Hovorka was hesitant to accept conclusions that MPC is better than PID or that dual hormone is better than single hormone – the nature of a meta-analysis allows it to pull out trends for further investigation in proper head to head studies, but differences in study design and population make it impossible to definitively claim superiority by lumping all studies together. In secondary outcomes, time <70 mg/dl and <50 mg/dl were 2.5 and 0.5 percentage points less, respectively, with closed loop systems than with standard pump therapy, though significance evaporates when studies without remote monitoring are included – Dr. Weisman hypothesized that trial administrators were reaching out to intervene and prevent episodes of prolonged or severe hypoglycemia. This work confirms what is already known - closed loop algorithms increase time-in-range and reduce hypoglycemia – though obviously head-to-head effectiveness and cost of different systems are impossible to answer now.

Artificial Pancreas in an Acute General Hospital: A Randomised Controlled Study

Hood Thabit, MD (University of Cambridge, UK)

For the second straight ADA (see 2016), Cambridge’s Dr. Hood Thabit presented very positive closed-loop data in type 2 diabetes patients in an acute general hospital setting. The study randomized insulin-using patients to conventional care (n=22) or fully automated subcutaneous closed-loop therapy (n=21) with the Florence system for up to 15 days (Navigator II CGM, Dana R pump, tablet running Cambridge’s MPC algorithm). As expected, glycemic outcomes were strongly in favor of closed-loop, including far more time in 100-180 mg/dl (59% vs. 37%; p<0.001), a lower mean glucose (168 mg/dl vs. 187 mg/dl; p=0.22); far less time >180 mg/dl (31% vs. 49%; p=0.01); and a lower between-day CV of glucose (15% vs. 26%; p=0.02). Other endpoints trended in the right direction, but were not statistically significant (presumably from the small sample size). Similar to ambulatory studies, closed-loop drove far superior overnight glucose profiles and statistics, including 66% vs. 40% time-in-range (p<0.001) and a mean glucose of 147 mg/dl vs. 171 mg/dl (p=0.07). Daytime in 70-180 mg/dl almost doubled on closed loop: 52% vs. 33% (p=0.006). There was no severe hypoglycemia or DKA during the study. Notably, this study population was also quite representative: ~69 years, BMI of ~32 kg/m², A1c of ~8.3%, ~19 years with diabetes, and ~12 years on insulin therapy. Dr. Thabit concluded that subcutaneous closed loop without meal announcement is feasible and safe in a heterogeneous population of insulin-treated patients in the medical/surgical wards. Dr. Thabit estimated that nurses spend ~2 hrs/day on inpatient glucose control, meaning closed loop can make an enormous difference on HCP burden, in addition to improving glycemia (and hopefully other outcomes). Longer studies will be needed to look at infection rates and length of stay, but we have little doubt that closed loop has enormous potential in the hospital setting.

• Dr. Thabit praised the FreeStyle Navigator II sensor, which is resistant to interference issues in the hospital (e.g., paracetamol, low oxygen). We see this as a natural place for Abbott to take CGM, though the company has never commented publicly on bringing CGM into the hospital. As a reminder, Dexcom had a longstanding relationship with Edwards to bring CGM into the hospital, but the partnership was dissolved years ago. Dexcom has more recently commented that the G6 sensor could be suited for the inpatient setting, though it’s unclear how near-term this is on Dexcom’s radar (presumably in the ~5-year time-frame). Medtronic also did a little work on in-hospital CGM with the Sentrino sensor, but there has been no update on this project since the CE Mark in 2012.

Diabeloop Closed-Loop Achieves Better Blood Glucose Control than Sensor-Augmented Pump over Three Days Involving Intensive Physical Exercises, Gastronomic Dinners, or Rest in T1D Patients

Sylvia Franc, MD (CERITD, Evry, France)

Dr. Sylvia Franc presented pooled data on Diabeloop’s closed loop from three inpatient, overnight, crossover studies lasting three days each: gastronomic meals (n=10), exercise (n=14), and steady-state use (n=14; initial results shared at ATTD in February). Combined, use of the Diabeloop MPC algorithm, Dexcom CGM, and a Cellnovo pump vs. open loop in 38 people with type 1 diabetes drove very strong improvements in overnight time-in-range (70-180 mg/dl) –67% in open loop vs. 87% in closed loop (p<0.0001). Meanwhile, overnight time >180 mg/dl declined from 28% to just 11% (p=0.0003), and time <70 mg/dl was more than halved from 5% to 2% (p=0.067). The improvement was particularly striking following the gastronomic dinners (Japanese, French, and Italian), where even “expert patients” were unable to avoid nighttime hyperglycemia from high fat/protein meals – time in 70-180 mg/dl increased from 54% to 81%, with a substantial shrinkage in variability. The algorithm’s performance also improved over the course of the study – time-in-range on day three (82%) eclipsed performance on day one (75%; p=0.01). Patients also had greater treatment satisfaction on closed loop (DTSQ). The next step for Diabeloop is a crossover home study in 12 centers in France, comparing three months on Diabeloop’s system to three months on open-loop therapy (pump + CGM). The study is still listed as “not yet recruiting” on ClinicalTrials.gov, though according to Cellnovo’s April update, results are expected by the end of 2017 ahead of an expected 2018 filing/launch in Europe.

Do Patients under Pump Therapy Benefit from Using the Predictive Low Glucose Management System for Prevention of Hypoglycemia and Improved Glycemic Control in Type 1 Diabetes Mellitus?

Petros Thomakos, MD (Athens University Medical School, Athens, Greece)

Dr. Petros Thomakos presented data from a head-to-head study demonstrating that predictive low glucose suspend (MiniMed 640G; n=30) is more effective at preventing hypoglycemia, without a concomitant increase in hyperglycemia, than low glucose suspend (MiniMed Veo; n=30). Patients using PLGS had ~half as many episodes of hypoglycemia, defined as ≥20 minutes at ≤54 mg/dl, per week (1.9 vs. 3.6 episodes; p=0.0004) – further analysis revealed that the effect remained even after adjusting for hypoglycemia unawareness. The area under the curve ≤54 mg/dl also reached statistical significance in favor of the PLGS group. Notably, the decrease in hypoglycemia can be attributed to increased time in target, because area under the curve ≥180 mg/dl trended toward statistical significance in favor of the PLGS system. There was only one episode of severe hypoglycemia in the three-month trial, in the Veo group, and no DKA. In both groups, the events commonly preceding hypoglycemia were, in order of frequency: overcorrection of hyperglycemia, basal rate increase, miscalculation of carbohydrates, and bolus wizard override. The only difference was that PLGS, as one would expect, was more protective against hypoglycemia, and without rebound hyperglycemia. At Monday’s JDRF/NIH closed loop meeting, Jaeb’s Mr. John Lum showed a slide with a picture of Dr. Aaron Kowalski and a speech bubble reading: “Why the &*%^@ doesn’t every pump do this?” During his presentation of Tandem’s successful PLGS feasibility study (highlight #1), Barbara Davis Center’s Dr. Gregory Forlenza added that the Tandem algorithm is really simple, calling it “basically 10th grade math!” We agree with Dr. Kowalski – every single pump should have this feature, since it is so much safer than standard therapy.

Single and Dual-Hormone Closed-Loop Glucose Control with Automated Exercise Detection to Prevent Hypoglycemia in Type 1 Diabetes

Peter Jacobs, PhD (Oregon Health and Science University, Portland, Oregon)

OHSU’s Dr. Peter Jacobs presented initial results from a four-day, four-way randomized crossover trial (n=21; 12-17 completed, depending on experiment) comparing single hormone (SH) artificial pancreas (Tandem pump, Dexcom sensor), dual hormone (DH) artificial pancreas (two Tandem pumps, Dexcom sensor), predictive low glucose suspend (PLGS), and current care (CC). Both of the closed loop systems also included wearable Zephyr heart rate and accelerometer sensors to automatically detect exercise – the algorithm shuts off insulin delivery for 30 minutes at the start of exercise, then limits infusion rate to 50% maximum for the next hour, while delivering glucagon as needed for dual hormone. The below table summarizes glycemia across all four groups across all four days. At a high level, both closed-loop systems were better than PLGS and standard care on mean glucose and time-in-range. Surprisingly, dual-hormone in this trial was not better than single hormone on mean glucose, time-in-range, or time in hyperglycemia – its advantage came on hypoglycemia, which was cut by 55% with the addition of glucagon (1.3% vs. 2.9%). The tradeoff, of course, was experiencing more hyperglycemia with this particular dual hormone approach (27% vs. 22%). In a sub-analysis of the exercise period, the addition of glucagon significantly decreased hypoglycemia (threshold not specified) during and after exercise from 7.3% and 6.3% in PLGS and single hormone, respectively, to just 1% (p<0.05). There was no significant difference in time in euglycemia during these post-exercise periods, implying that the decrease in hypoglycemia with glucagon translated to a small increase in hyperglycemia. Based on an AGP-like trace of glucose values, Dr. Jacobs noticed that the addition of glucagon helped prevent early hypoglycemia ~25 minutes into exercise. Conversely, blood glucose trended low from 25 minutes until the end of the trace (~250 minutes) with the insulin-only system, and was lower than 70 mg/dl from 25-60 minutes. This is consistent with previous work showing that glucagon has a pretty quick action profile. In the dual-hormone arm, there was an expected trend toward more glucagon delivery on high activity days (>500 ug/day), compared to days of low activity (>300 ug/day) (p=0.08). Meanwhile, insulin delivery did not differ between high and low activity days, hovering around 43 units/day regardless of activity level. We love these sorts of head-to-head-to-head-to-head studies, since they are so rich with learning on the relative incremental benefits of different approaches. From this data, it seems like OHSU’s dual-hormone system does not add much to its insulin-only approach, with the exception of hypoglycemia during exercise.

Q: A frequent question is when should patients reduce or stop their basal insulin with reference to when they start exercise – do you have any data on that? If you decide you’re going to exercise at 12 noon and running at one unit per hour, when would the effect of insulin stop so you would have no hypoglycemia during intense exercise?

A: We responded to data in real time, so it’s not informative. Recommendations from the PEAK project and Dr. Michael Riddell are good – I’m not a physician, so don’t take my prescription, but generally people reduce basal rate one hour or two hours before, and have a carbohydrate before the start of exercise.

Dr. Bruce Perkins (University of Toronto, Ontario): There’s still going to be doubt about the value-add of glucagon, and my question is whether there can be blinding in a study. Is the insulin-only system similar enough so you can mask the use of glucagon – have a placebo?

A: They are identical. It would be interesting to look at a blinded case, but talking to subjects, they love the system, particularly the single hormone – feedback like “these were the best four days of my life” – but subjects like dual hormone as well, especially that feeling of safety. Blinding that would be interesting.

Dr. Joseph El Youssef (OHSU, Portland, Oregon): We have done blinding, and subjects picked up pretty quickly that they had glucagon.

Dr. Nicholas Argento (Maryland Endocrine and Diabetes Center, Laurel, MD): Are you considering announcing exercise and using glucagon that way?

A: Yeah, certainly the more information a system can get from a patient, the more accurate it can be – especially for the single hormone case.  

Oral Presentations: New Insights into Prevention and Treatment of Hypoglycemia

Restoration of Hypoglycemia Awareness with Closed-Loop Therapy

Stacey Anderson, MD (UVA, Charlottesville, VA)

UVA’s Dr. Stacey Anderson presented data from an important test of closed-loop control in patients with hypoglycemia unawareness or a high risk of hypoglycemia (n=44) – a critical group typically excluded from studies. Compared to sensor-augmented pump therapy (Dexcom Share CGM + Roche Combo), use of the DiAs closed loop system over four weeks significantly improved glycemia, including time <70 mg/dl (a remarkable 72% decline with closed loop vs. a 15% decline on open loop; p=0.005), time in 70-180 mg/dl (an 11% improvement with closed loop vs. an 8% decline on open loop; p=0.009), and LBGI (a 66% improvement on closed loop vs. an 11% improvement on open loop; p=0.003). Unfortunately, counter-regulatory response (epinephrine) did not improve over four weeks with use of closed loop, though the Clarke questionnaire trended towards improvement (not statistically significant) – perhaps with a longer study awareness and counter-regulatory response could be restored. Dr. Anderson noted in Q&A that 70% of these patients were on CGM at baseline (mean A1c: 7.4%), possibly leaving less room for improving awareness relative to those naïve to CGM. On the other hand, it’s still striking that patients were spending nearly two hours per day at baseline <70 mg/dl – and they were mostly on CGM! Obviously, this is a group that stands to benefit substantially from automated insulin delivery, particularly in the dangerous overnight period.

Effect of Ethanol Intoxication on the Antihypoglycemic Action of Glucagon

Laya Ekhlaspour, MD (Massachussets General Hospital, Boston, MA)

Dr. Laya Ekhlaspour, who at last year’s ADA discussed the merits of a bihormonal bionic pancreas, delved deeper into the practicality of micro-dosing with glucagon, presenting findings suggesting that ethanol intoxication does not affect glucagon activity. In a random order crossover trial (n=15) using simultaneous hyperinsulinemic-normoglycemic and ethanol clamps, volunteers with type 1 diabetes were given two identical 50g injections of glucagon while sober and while intoxicated with a blood alcohol content of 0.1%, achieved by delivering the equivalent of 4 drinks in 20 minutes (not your average closed-loop study!). Blood glucose rose following glucagon injection and then stabilized at 90 mg/dl for the duration of the study. Trends were similar in both arms of the study, indicating that ethanol intoxication does not seem to alter glucagon activity and that micro-dosing with glucagon to reverse hypoglycemia is still a safe and effective option when an individual is intoxicated. However, Dr. Ekhlaspour did concede that liver glycogen levels were not recorded during the study, which, as one audience member pointed out, most likely contributed to the observed effect. In the case of people with alcoholism (who might be poorly nourished), reduced glycogen stores could interfere with the efficacy of glucagon to raise blood sugar. These results are encouraging on the risk mitigation front for a bihormonal bionic pancreas in everyday use.

Oral Presentations: Translating Therapeutics to the Real World

A Prospective, Pragmatic Clinical Trial to Compare the Real-World Use of V-Go in Type 2 Diabetes Patients

Mark Cziraky (HealthCore, Inc, Wilmington, DE)

Valeritas’ Mr. Mark Cziraky shared results from a nicely-done real-world study comparing V-Go to standard care that demonstrated improved A1c with decreased insulin usage in type 2 patients initiating the device in a community setting. The trial employed cluster randomization (where study sites rather than individual patients were randomized to V-Go [n=169] vs. standard therapy [n=246] for up to four months) and – in a very smart appeal to insurers – was designed such that all treatments, medications and supplies in both groups were obtained only via patient insurance and co-pays. Consistent with previous V-Go data, primary outcomes showed significant A1c reductions from baseline with V-Go (-0.95%, p<0.001) vs. standard therapy (-0.46%, p<0.001). V-Go was statistically superior to standard therapy (p<0.002), though the group did have a significantly higher mean baseline A1c (9.9% vs. 9.3%; p<0.001). Mr. Cziraky proposed that this discrepancy may have indicated a selection bias of more advanced diabetes patients at sites initiating V-Go (since sites were randomized a priori). Mean total daily insulin dose was significantly reduced from a similar baseline only in the V-Go group (-17.6 vs. -0.4 units, p=0.001) while patients on V-Go also received fewer concomitant diabetes medications compared to control patients. Impressively, these differences summed to a nice cost-effectiveness benefit in favor of V-Go – per 1% A1c drop, mean per patient per day cost of diabetes treatment was $24.48 with V-Go vs. $39.95 (control). We are particularly keen to hear how payers perceive this kind of financial analysis. It’s great to see a prospective, randomized study of V-Go, which adds to its long-documented retrospective outcomes.

• Study Design: The trial used a cluster randomization, where study sites rather than individual patients were randomized to V-Go (n=169) or control therapy (n=246) for four months. Patients initializing V-Go stopped other insulin therapy, while diabetes treatments were continued in control group patients. All treatments, medications and supplies were obtained using patient insurance and co-pays. As a consequence, Mr. Cziraky noted that the largest reason for V-Go discontinuation (n=12) was associated with access, insurance coverage, and cost. In this way, even though the study was prospective and randomized in nature, it still had a very real-world feel, taking into account issues pertaining to access in addition to clinical utility.
• Consistent with previous data, V-Go appears to have been particularly effective in the toughest of patients (baseline A1c >9%). The below table summarizes the proportion of patients that were in each A1c bracket at the beginning and end of the study.

• Mr. Cziraky also shared positive data from the user experience assessment of V-Go: (i) 85% of patients confirmed their willingness to keep using V-Go after the end of the study; and (ii) 94% of patients reported using V-Go as directed for the duration of the study. There was no background on how this question was asked, though we assume patients were given binary choices (e.g., Yes/No). The data do point to why V-Go uptake could be strong with more funding, especially in those that have avoided current pumps due to comfort/wearability.

Posters

Maintaining Glucose Control at One+ Year of MiniMed 670G System Home Use: Single Center Experience (126-LB)

S Garg, R Slover, D Giordano, G Forlenza, S Lee, J Shin, and F Kaufman

A Medtronic poster detailed very positive glycemic data from the 19 Barbara Davis patients who have partaken in the MiniMed 670G Continued Access Program for over a year. We saw most of this data at ATTD, and had three main takeaways: (i) The study phase improvements in A1c, time-in-range, hypo/hyperglycemia, and glycemic variability have been mostly maintained out to one year in BDC patients (with an exception for A1c in adolescents); (ii) Time in hybrid closed loop (auto mode) and sensor utilization have dropped off slightly in both BDC groups, particularly in adolescents (~2 less hours per day spent in Auto Mode at one year vs. the pivotal study phase); and (iii) BDC adults did better on the system than adolescents at one year. See the key metrics below. There has still been no severe hypoglycemia, nor DKA, in any of the patients on the system, though there have been nine instances of severe hyperglycemia (blood glucose > 300 mg/dl with ketones >0.6 mmol/l or symptoms of nausea, vomiting, or abdominal pain) in the one year the patients have been on the system – on an annualized basis, this is still far less than the one event seen in the two-week baseline run-in period (~26 per year expected). Over the one year of use, total daily insulin dose did not change significantly in adults or adolescents, but basal percentage decreased in adults (56% at baseline to 46% at one year). [At ENDO 2017, Dr. Rich Bergenstal suggested that the ideal basal percentage may be more in the neighborhood of 25% – cover meals better than you used to, and the basal will just shut off if it needs to. Patients will of course vary; on hybrid closed loop using a DIY system, Adam’s basal is 75% of his total daily dose.] These results are from just one center, but they suggest largely sustained improvements following the single-arm, three-month pivotal trial presented at ADA 2016 – good news for Medtronic! One area to watch going forward is time in auto mode, though even adolescents are spending nearly 15 hours a day in closed loop.

• Thanks to a slide in Medtronic’s ADA Analyst Briefing, we’ve also seen clinical results from the real-world customer training phase (n=730): 670G users spent 74% time-in-range in Auto Mode (vs. 72% in the pivotal), only 2% of the time <70 mg/dl (vs. 3% in the pivotal), 23% of the time >180 mg/dl (vs. 25% in the pivotal), and had a mean glucose of 151 mg/dl (vs. 150 mg/dl in the pivotal) – nice alignment with the above results! In these two-to-three months of Customer Training Phase observation, the 730 patients have been in auto mode for over 22 hours every day – we wonder how this will change over time, or perhaps Medtronic has made improvements to keep patients in Auto Mode for a longer period.

Automatic Estimation of Basals, ISF, and Carb Ratio for Sensor-Augmented Pump and Hybrid Closed-Loop Therapy (127-LB)

D Lewis and S Leibrand

In this fascinating poster, #OpenAPS patient innovators Dana Lewis and Scott Leibrand share observations from a cool new DIY tool: Autotune. The algorithm uses CGM and insulin dose data (from pumps, open- or closed-loop) to automatically tune underlying basal rates, ISF, and carb ratios. The tool was created in January, and this poster shared self-reported feedback from 16 users. Overall, 75% of surveyed users made changes to their insulin pump settings after running Autotune. Notably, 100% of people felt the basal suggestions were accurate (83% actually changed their basal rates), and 88% of people felt the ISF suggestions were accurate (67% actually changed their ISF). Users were less sure of carb ratio estimations: only 69% felt the estimates were accurate (58% actually changed their carb ratios). On average in the surveyed population, Autotune estimated a needed 10% change in hourly basal rates, a 29% increase in carb ratios, and a 19% increase needed in ISF. We’d point out that this is a very engaged group that is pretty dialed in already, so the “normal” population of users would probably need much bigger changes. Indeed, “Patients felt strongly that using data to assess changes to pump settings should be the norm rather than relying on the current methods of guessing or weight-based estimations.” Hear, hear! The poster concludes that patients and HCPs could benefit from using this type of tool to help make ongoing changes to settings – we wholeheartedly agree and expect to see many commercial dose titration products launch in the coming years (see our competitive landscape). Interestingly, most Autotune users planned to (or did) discuss results with their HCP, and providers “were interested and supportive in people using this tool.” The poster notes that “(n=1)*300+ people worldwide” are self-building various types of DIY hybrid closed loop artificial pancreas systems – a number we assume is much higher, given all those that are presumably not tracked.

• How Autotune works: Insulin dosing and carb data, glucose data from CGM, and pump profile settings are used to calculate expected blood glucose impact (BGI) for each glucose value. Each glucose value is then categorized as being most attributable to basal, ISF, or carb sensitivity factor (CSF = ISF / carb ratio), and used to calculate adjustments to basals, ISF, and CSF. For each hour, total BGI deviations, and necessary adjustment in basal to bring deviations to 0, are calculated; 20% is applied to the previous 3 hours’ basals. Median deviation for an entire day’s ISF-attributed data and necessary adjustment in ISF to bring the median deviation to 0 is also calculated; 10% is applied. Last, total BGI deviations during observed carb absorption are calculated and compared to total carb intake to calculate new CSF; 10% is applied to the carb ratio.
• Autotune currently works with a single ISF and carb ratio. There is interest from the community to adapt the tool to support use for multiple ISF and carb ratios.  There is also interest in using data from Autotune to confirm whether multiple ISF and carb ratios are needed, or if they are proxies for mistimed insulin dosing at other times of day.

Safety and Feasibility of Omnipod Hybrid Closed-Loop in Children Aged 6-12 Years with Type 1 Diabetes Using a Personalized Model Predictive Control Algorithm (132-LB)

B Buckingham, G Forlenza, J Schneider, T Peyser, E Dassau, J Bok Lee, J O’Connor, J Layne, and T Ly

This single-arm, multi-center, inpatient, 36-hour feasibility study tested the Omnipod Horizon Automated Glucose Control algorithm in a very young pediatric population (6-12 years). The investigational hybrid closed loop system used a modified Omnipod and PDM (Bluetooth->RF relay), a Dexcom G4, and a portable tablet running the control algorithm. The study included 12 patients (mean age: 10 years) who wore the system for one overnight period and five meals (30-90 grams of carbs). Overall mean glucose was 157 mg/dl, with 70% time in range (70-180 mg/dl), 2% time <70 mg/dl, and 28% time >180 mg/dl. As expected, glycemic outcomes were better overnight, improving to a mean of 149 mg/dl, 87% time in 70-180 mg/dl, and just 0.1% <70 mg/dl – wow is this going to make a difference for parents! Overall time in the tighter target of 70-140 mg/dl was lower at 39%, though this is not a population to be aggressive and the system does not issue automatic correction boluses. Two example patients (see below) showed that the algorithm worked as expected, automating basal insulin between meals to curb highs and lows – for some five-minute intervals, it delivered more than five times the pre-programmed basal rate, while at other times, no insulin was delivered at all. We love seeing this transparency, since it really illustrates how the algorithm works. The combined glucose profile plots showed much tighter variability at all times of day, with the exception of breakfast (where postprandial highs looked slightly more common on hybrid closed loop). In a subtle nod at Medtronic, the poster concludes, “…it is imperative to include the pediatric population in clinical trials for commercial launch” (the 670G was only initially approved for 14+ years). This study affirms that Insulet is indeed very committed to a pediatric launch (key for its patient base) and now moving quickly to get a competitive product in 2019 (pre-pivotals this year and pivotals in 2018) – both were not the case at ADA 2016!

Omnipod Hybrid Closed Loop Performance Over 36 Hours for Two Representative Pediatric Patients:

Impact of Center for Medicare Services (CMS) Insulin Pump Policies on Patients with T1D (1036-P)

NB Argento and AL Peters

New data from a T1D Exchange myGlu.org survey (n=124) indicates that the Center for Medicare Services’ (CMS) mandated quarterly visits for pump users may be quite detrimental to patient safety. In 43% of surveyed respondents, this policy led to potentially adverse pump practices: nearly one in three respondents said they have extended pump site duration past three days, 11% had to obtain emergency supplies, 7% said they had reused supplies, and 4% said they had switched to injections. The authors note that there is no supporting evidence for this quarterly visit policy. Ugh! The result, as participant comments indicate, is that once a patient goes on Medicare, pump companies (Medtronic was specifically mentioned) will no longer send supplies without prompting and instead rely on doctor visits as a marker for when to replenish materials. In the case of patients with remote access, the requirement poses a significant inconvenience, since HCP visits might require traveling hundreds of miles. Likewise, amongst those who are well-controlled, these visits may not even be medically necessary, unduly increasing cost and administrative burden on the system. Data from the T1D Exchange cohort examining mean A1c levels vs. frequency of visits in patients 65 and older did not even yield a significant correlation, further adding to the lack of evidence for this policy. Given these findings, it may be time for CMS to reconsider the quarterly visit policy for pump users. With the lack of government response to the crisis caused by competitive bidding – a poster presented at ADA estimates that the program has left 90,000+ insulin-dependent patients with partial or no SMBG – we don’t expect a rapid response without patient uproar. We’re glad to see Drs. Argento and Peters bringing yet another antiquated CMS policy to the forefront.

The Efficacy of Single-Hormone Artificial Pancreas at Controlling Glucose Levels in Insulin-Treated Patients with Type 2 Diabetes: A Randomized Crossover TriaL (1050-P)

N Taleb, A Carpentier, J Rene, V Messier, J Morais, A Haidar, and R Rabasa-Lhoret

This poster from the Montreal artificial pancreas group suggested that single hormone closed loop in type 2 patients (n=10; 90% male; ≥3 insulin injections per day; mean A1c=8.1%) has the potential to reduce total daily insulin dose and hyperglycemia without a concomitant increase in hypoglycemia in the in-patient setting. Patients spent just over 24 hours in a clinic either on closed loop or MDI, and were then crossed over – the visit included standardized meals and 15 minute walks. The most striking finding was the dramatic ~33% reduction in daily insulin dose between the AP and MDI conditions (0.66 units/kg vs. 0.98 units/kg; p=0.009). Throughout the experiment, there was no significant difference in percentage of time <72 mg/dl (both ~0%). There was, however, a slight trend toward lower percentage of time >145 mg/dl (23% vs 36%; p=0.18), and toward a notably lower mean glucose (117 mg/dl vs. 130 mg/dl; p=0.16) in the AP condition. Overnight, percentage of time >145 mg/dl was completely eliminated in the AP group vs. 17% in the MDI group (p=0.05), and mean glucose was borderline significantly lower in the AP group (101 mg/dl vs. 122 mg/dl; p=0.06). Overnight plasma glucose levels, collected every 20 minutes, were more frequently in target in the AP group (72-180 mg/dl post-meal; 72-144 otherwise) (p=0.02). There are very few studies on closed loop performance in type 2 diabetes, but the benefits seem just as applicable to this group as in type 1. We hope a thriving commercial environment ultimately develops in this area, either leveraging lower cost pump/CGM technology or applying closed-loop algorithms to better inform open loop on MDI.

PAQ Insulin Delivery Device Significantly Improves Glycemic Control Without Increase in Hypoglycemia (1051-P)

JK Mader, L Lilly, F Aberer, C Lanz, T Poettler, S Becvar, M Trautmann, J Warner, and T Pieber

Medical University of Graz’s Dr. Julia Mader et al. presented further evidence that the CeQur PAQ basal-bolus patch delivery device can help overcome hurdles associated with MDI. 24 adults were placed on PAQ for three months, five of which also wore CGM. Compared to a one-week baseline period, at three months, A1c had fallen an impressive 1.5% (baseline: 8.6%; p<0.0001) and time in range (70-180 mg/dl) for those on CGM had increased from 51% to 70% (p<0.001) – that translates to four more hours per day in range! These improvements, along with decreased glycemic variability and much-improved 7-point SMBG profiles, came without a concomitant increase in hypoglycemia. Patient satisfaction, as measured by DTSQ score, was significantly boosted, and patients reported PAQ to be more flexible and convenient than their daily injections. To this end, PAQ seems to facilitate adherence, as meal time dosing increased by 0.9 times/day, significantly improving both pre- and post-prandial blood glucose levels. Though small and observational, this study confirms that the three-day PAQ insulin delivery device could be effective in overcoming the hurdles to insulin therapy – we hope it moves along to commercialization and manufacturing scale. We reported last April that PAQ was slated for a 2017 US launch, though Senior VP Mr. Jay Warner confirmed in late September that the company had not yet filed with the FDA (it was CE marked in 2012). AS we understand it, the company is working on the manufacturing piece. There was no concrete update on the device at ADA, though we are hopeful – boy has it been in development for a while. CeQur raised an impressive $100 million in Series C financing in September 2015 and announced a significant facility expansion last April.

Role of Appropriate Premeal Insulin Bolus on the Hybrid Closed-Loop System 670G (1053-P)

A Roy, B Grosman, R Bergenstal, B Bode, B Buckingham, S Garg, S Lee, and F Kaufman

This poster attempted to parse out factors that differentiate patients who saw a rise, drop, or no change in A1c over the three-month MiniMed 670G pivotal trial (presented at ADA 2016). To minimize the effect of baseline A1c, 51 (of 124) subjects with initial A1c between 6.5% and 7.5% were divided into rise, drop, and no change groups. Amongst the groups, there was no difference in total daily insulin dose throughout the trial, nor meal announcing/carb entering patterns. However, the percentage of daily insulin from boluses was significantly higher in the A1c-drop group: 50% vs. 46% for no-A1c-change and 42% for A1c-rise ( p=0.0082). This difference can be largely attributed to a lower (more aggressive) carb:insulin ratio (p=0.08) in the group that experienced a reduced A1c. As a control consideration, the authors looked at overnight mean glucose, finding no difference between the three groups, indicating that the system works the same across all three groups in the absence of mealtime behavioral interventions – no surprise there. In line with Dr. Rich Bergenstal’s recent presentations, these results indicate that a more aggressive carb:insulin ratio could be beneficial for 670G users, at least from an A1c perspective – the automated basal rate will simply drop or turn off if a bolus was too large. In other words, this user-adjustable setting is a way to overcome the 670G’s conservatism during the day, especially because it cannot issue automatic correction boluses. We would be interested in seeing granular time in range data – does a more aggressive meal bolus strategy have any impact on time <70 mg/dl, 70-180 mg/dl, or >180 mg/dl? These results, assuming the observed A1c drop isn’t accompanied by a tendency toward more hypoglycemia, have implications for training patients and for providers setting up/maintaining the MiniMed 670G.

• The sub-analysis also showed that individuals who entered the study with a higher baseline A1c experienced the largest A1c drops (R=-0.75) – a very encouraging finding. This inverse correlation is a common pattern in device and drug intervention studies – those with the highest A1cs have the most runway to come down! However, it is clearly not a given with automated insulin delivery, since those with higher A1c’s are (arguably) spending less time focused on their diabetes and engaged with devices. We see this as a highly positive finding for the field: those who need the most help may be the most likely to benefit from automation.

Safety and Feasibility of Omnipod Hybrid Closed-Loop in Adolescents with Type 1 Diabetes Using a Personalized Model Predictive Control Algorithm (1055-P)

B Buckingham, G Forlenza, J Schneider, T Peyser, E Dassau, J Bok Lee, J O’Connor, J Layne, and T Ly

This single-arm, multi-center, inpatient, 36-hour feasibility study tested the Omnipod Horizon Automated Glucose Control algorithm in 12 adolescents (mean age: 15 years). The investigational hybrid closed loop system used a modified Omnipod and PDM (Bluetooth->RF relay), a Dexcom G4, and a portable tablet running the control algorithm. Adolescents wore the system for one overnight period and five meals (30-90 grams of carbs), with limited physical activity. Overall mean glucose was 153 mg/dl, with 73% time in range (70-180 mg/dl), 2% time <70 mg/dl, and 25% time >180 mg/dl – almost identical to the similar pediatric study described above. Glycemic outcomes were better overnight, improving to a mean of 149 mg/dl, 85% time in 70-180 mg/dl, and just 0.2% <70 mg/dl. Relative to the 48-hour run-in period, time <70 mg/dl improved 60% during hybrid closed loop (2.0% vs. 5.2%; p=0.014). Similar to the pediatric data, the system curbed highs and lows at most times of day, though median glucose did approach ~190 mg/dl in some in the late post-breakfast and post-lunch phases (see profile below). The poster notes that another recently-completed study tested the algorithm in adults consuming high-fat meals and moderate intensity exercise – we’re glad to see Insulet putting this device in a variety of settings and populations before embarking on larger ambulatory studies.

Omnipod Hybrid Closed Loop Performance Over 36 Hours for Two Representative Adolescent Patients:

Effect of a Lower Glucose Target on Glycemic Outcomes with an Insulin-Only Bionic Pancreas (1062-P)

CA Balliro, L Ekhlaspour, A Esmaeili, FH El-Khatib, D Mondesir, R Selagamsetty, MA Hillard, M Maheno, RZ Jafri, ER Damiano, and SJ Russell

This important poster from the MGH/BU group looking into the impact of different algorithm targets for the insulin-only bionic pancreas. The investigators noted that, with the glucose target set to more conservative 130 and 145 mg/dl values, mean glucose was 161 mg/dl and 174 mg/dl, respectively, with minimal time below 60 mg/dl. But what is the impact of more aggressive targets of 110 mg/dl and 120 mg/dl? Over the course of three days, the lower 110 mg/dl and 120 mg/dl targets yielded comparable mean CGM glucose values – 153 mg/dl and 156 mg/dl –  with comparable time spent <60 mg/dl (1.3% and 1.1%) and between 70-180 mg/dl (70.4% and 69.5%). In addition to lowering mean glucose levels, the lower targets resulted in higher time in range relative to the higher targets (as we would expect). Further, the lower targets cut glucose variation of usual care in half (SD=15-17 mg/dl vs 32 mg/dl), pointing to the added benefits of a closed-loop system. (We’d note that an SD of 15-17 mg/dl approaches SD in people without diabetes.) These findings suggest that it is safe to adjust the glucose target of the single-hormone bionic pancreas, at least down to 110 mg/dl. As the threshold drops lower, we assume hypoglycemia would be the clear tradeoff. This poster is yet another reminder that patients and HCPs should be able to set their own closed-loop glucose target, which is not possible in the first-gen MiniMed 670G (it targets 120 mg/dl). A pivotal trial of the insulin-only bionic pancreas was most recently expected to start at the end of 2017/early 2018, with a potential PMA submission by mid-2018 and subsequent late 2018 launch. We’re not sure if the target glucose value will be adjustable in this trial or in the initial commercial product.

Comparing the Patient Experience of an Insulin-Only and Bihormonal Bionic Pancreas (876-P)

CA Balliro, MA Hillard, D Mondesir, SJ Russel, and ME Larkin

A trial comparing eight different configurations of the bionic pancreas, including six different glucose targets in either the bihormonal or insulin-only model, found that patient satisfaction (n=24) was significantly greater for the bihormonal version regardless of the glucose target (p<0.001). In fact, 71% of participants indicated a preference for the bihormonal bionic pancreas, while the remaining respondents expressed no preference – that is to say, that none preferred the insulin-only version. A 23-item survey was administered, prompting participants to rate the extent to which they agreed with various statements such as “I spent much less time thinking about my diabetes” and “I found it hard to trust that the bionic pancreas could control my blood sugars” on a scale of one to five. Responses were collected following each of the eight arms of the study (see below) and averaged separately to create a total satisfaction score for the individual configurations. Overall, the bihormonal model set to a glucose target of 110 mg/dl achieved the highest average satisfaction of 4.35, and the insulin-only model at a target of 145 mg/dl fielded the lowest score of 3.37. While the bihormonal model beat out insulin-only for every glucose target, the least popular target for the bihormonal configuration in overall satisfaction was 115 mg/dl. The investigators provided analysis of one question in greater detail (“It helped to prevent low blood sugars from happening”) and again saw significantly higher rates of agreement for the bihormonal configuration as compared to the insulin-only (4.20 vs 2.84; p<0.001). Interestingly, while the bihormonal model set to a target of 110 mg/dl still took first place for this question, the target of 115 mg/dl was a close second – we might’ve expected that the bihormonal bionic pancreas set to a higher 130 mg/dl target would have won here. We wonder how responses might change if the patients were using the devices over longer periods of time, considering the cost of the system, changing two hormone cartridges. It will be great to better understand (hopefully) in the pivotal trial. Although glucagon adds system complexity and a higher regulatory burden, these gains in patient satisfaction make an encouraging case for a dual-hormone system. As of last week’s Zealand glucagon results, Beta Bionics’ dual hormone iLet Bionic Pancreas device was expected to begin an NIH-funded bihormonal pivotal trial in “2H18” (pending phase 2b results). Assuming it runs into 2019 and the FDA submission takes about a year (for the new glucagon), this is likely a 2020-2021 launch at the earliest.

Joint ADA/JDRF Symposium – Progress towards an Artificial Pancreas

Hybrid Closed-Loop Insulin Delivery Systems—U.S. Perspective

Eda Cengiz, MD (Yale University, New Haven, CT)

In her review of hybrid closed loop systems, Yale’s Dr. Eda Cengiz shared glycemic outcomes from the MiniMed 670G Customer Training Phase, spanning March-May 2017 in 730 people (N=24,000+ patient days). She emphasized strong real-world alignment with the pivotal data, and we’ve enclosed the outcomes below using Medtronic’s own slide from yesterday’s ADA 2017 Analyst Briefing slide deck. (Dr. Cengiz’s numbers today were slightly different from what Medtronic shared yesterday, presumably due to rounding, mean vs. median, or different endpoints.) In the Customer Training Phase, 670G users spent 74% time-in-range in Auto Mode (vs. 72% in the pivotal), only 2% of the time <70 mg/dl (vs. 3% in the pivotal), 23% of the time >180 mg/dl (vs. 25% in the pivotal), and had a mean glucose of 151 mg/dl (vs. 150 mg/dl in the pivotal) – nice to see such alignment. Relative to manual mode, time-in-range on Auto Mode improved quite a bit more in the Customer Training Phase than in the pivotal – an 11-percentage point gain in real-world use (63%->74%) vs. a five-percentage-point gain in the pivotal (67%->72%). Mean glucose also improved by a nice 9 mg/dl in the Customer Training Phase (160->151 mg/dl) vs. no change in the pivotal. Customer Training Phase users spent a median 92% of the time in Auto Mode, also a rise from 87% in the pivotal. Guardian Sensor 3 wear remained strong at 95%, a positive sign for real-world use of Medtronic’s new sensor so far. Though these results are still in a very early adopter and enthusiastic population, they do inspire confidence as the device starts rolling out to 20,000+ priority access program participants (see coverage here, which only provided some of the data below).

• Beyond the MiniMed 670G, Dr. Cengiz shared that she is working on a “personalized medicine” project testing a closed-loop module specifically for females with type 1 diabetes. We look forward to a future where a slew of algorithms are available to meet the needs of many different patients! On the horizon, Dr. Cengiz said the field still needs to improve on system training, more advanced technology with less burden, faster insulins, and more personalized management to reach a broad group of people with diabetes.

Developing the Next Set of Clinical Decision Support Tools for the Artificial Pancreas

Richard Bergenstal, MD (International Diabetes Center, St. Louis Park, MN)

In a most valuable talk on clinical decision support tools, IDC’s Dr. Rich Bergenstal shared positive views on CGM standardization and the MiniMed 670G, highlighted the balance between population health and personal care (fascinating!), and commented on the amount of CGM data needed to make a clinical decision (two weeks is sufficient). We really enjoyed his views on the various needs clinical decision support tools must address – the balance between population health (A1c) vs. personal care (CGM); CGM “metrics” vs. CGM “profiles”; risk – “something needs to be done” vs. action – “here is what needs to be done”; and research & regulatory needs (publications, PIs, indications) vs. clinical management (therapy adjustments). Diving into CGM reporting, Dr. Bergenstal reminded attendees of Friday’s well-received consensus session, which honed in on the key CGM thresholds for reporting: <54, <70, 70-180, >180, and >250 mg/dl. Dr. Bergenstal noted that while the numbers are hopefully nailed down, the “terminology” is still open for debate – e.g., “Level 2 hypoglycemia” vs. “serious” vs. “clinically significant” vs. “take immediate action.” He noted that this will be discussed further at The diaTribe Foundation-organized meeting, “Glycemic Outcomes Beyond A1c: Standardization & Implementation,” in Bethesda, MD on July 21. Dr. Bergenstal also plugged an upcoming Diabetes Care publication from Dr. Roy Beck and colleagues, “The Fallacy of Average: How Using A1c Alone to Assess Glycemic Control Can Be Misleading” (see our take on this from last summer at diaTribe.org/BeyondA1c). Turning to AID, he said that ~70%-75% time-in-range (70-180 mg/dl) is a good goal for hybrid closed loop devices, given all the studies done to date. Dr. Bergenstal also showed new CareLink reports for the MiniMed 670G, which do a “nice job” of showing glucose profiles in Manual Mode vs. Auto Mode (similar to the pivotal trial plots, the profile shrinks in Auto Mode). Additionally, he covered the excellent one-page AGP report and noted the increasing commercial momentum, including in Abbott’s FreeStyle Libre, Dexcom’s Clarity (launched this week), Diasend/Glooko, and Roche. On Dr. Bergenstal’s recommended reading list? The Undoing Project by Michael Lewis, Reclaiming Conversation by Sherry Turkle, and the just-published Bright Spots & Landmines by our own Adam Brown.

• “Two weeks of CGM data, most of the time, is representative of a longer period of time.” Dr. Bergenstal previewed poster 115-LB, which uses Senseonics 90-day pivotal trial data to answer an important question – how much CGM data is needed for a clinician? In comparing two weeks, one-month, and three months of CGM data, glucose patterns were very similar in all three cases. In our view, this is an especially good sign for clinicians using professional CGM.
• The MiniMed 670G, Dr. Bergenstal joked, should be called “The Sleeping and Crying Machine. These parents are finally sleeping, and by and large, the fathers of these 13-17 year-olds come in – and while we’re talking about the numbers – they’re crying. ‘I finally have hope for my child.’”
• Dr. Bergenstal characterized Dr. Bob’s Vigerksy’s Glucose Pentagon, combining five glycemic metrics into one graphic, “an amazing tool.” Poster 1049-P uses this glucose pentagon to summarize MiniMed 670G pivotal data – the pentagon area shrinks in comparing baseline to Auto Mode, indicating overall glycemic improvement. Dr. Bergenstal likes that the pentagon can compare pre-post treatment and also allows for comparison to people without diabetes. We like the cool factor of it, though it obviously doesn’t drive clinical decision making (how to change therapy) as much as the AGP.
• www.agpreport.org is a nice website, showing all the AGP reports, including a new one for automated insulin delivery!

Beyond Hybrid Closed-Loop Insulin Delivery Systems

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

Yale’s Dr. Stu Weinzimer reviewed some of the latest literature in closed loop systems beyond current hybrid models (from Harvard IP delivery, BU bihormonal, and OpenAPS) and introduced Barbara Davis’ Ms. Laurel Messer’s CARE plan for the application of automated insulin delivery to clinical care.

• Like most physicians, Dr. Weinzimer sees the “Wild West of [DIY] automated insulin delivery” as a double-edged sword – on the one hand, it worries him that people are using unregulated technology to dose a potentially lethal drug, but on the other hand, the “positive disruption” is not lost on him. While there is without a doubt a risk to these systems, they have generated positive outcomes, largely anecdotally, and more recently in peer-reviewed literature: OpenAPS’ Ms. Dana Lewis et al. recently published a piece in JDST (a poster at ADA 2016) showing glycemic metrics for >100 OpenAPS users, comprising >250,000 total closed loop hours. For those of you keeping track, that’s at least 125,000 more hours than the Cambridge system has seen in patients. And the real-world data is encouraging: Relative to baseline, users of the system included in the study have seen a self-reported median 0.9% A1c drop (from a low baseline of 7.1%), and time in range (70-180 mg/dl) leapt from 58% to 81%. DIY systems do have risks, but so does dosing insulin every single day. Patients can iterate quickly and develop products that fit their needs outside of the traditional clinical trial and regulatory cycles.
• BDC’s Ms. Laurel Messer has a publication in press (Pediatric Diabetes) explaining her “CARE” framework for the application of automated insulin delivery to clinical care. CARE is an acronym for Calculate (how does the system calculate insulin delivery?), Adjust (what are the adjustable components?), Revert (when should control be returned to the user?), and Educate (where does the user go for help?). Ms. Messer has over three years of experience with the Medtronic 670G hybrid closed loop system (read about some of her fantastic insights here). We’re eager to read more about the CARE model, but at a high level, we like that it emphasizes the human-machine interaction and the importance of graceful handoffs between closed loop and open loop. Dr. Weinzimer commented that it’s on industry to reach out to providers to facilitate smooth uptake for patients. 
• Early in his talk, Dr. Weinzimer called attention to Dr. Eyal Dassau et al.’s recent Diabetes, Obesity, and Metabolism publication showing much improved glycemic control with IP vs. subcutaneous insulin delivery in the context of fully-automated artificial pancreas. Time in the tight 80–140 mg/dl range was 40% in the IP group vs. 26% in the subcutaneous group (p=0.03). Other outcomes, including mean glucose, time in a broader 70-180 mg/dl range, and time in hyperglycemia were all superior in the IP group – numbers aside, we were most struck by the blunting of postprandial spikes in the accompanying glucose profile graphs. Opinion leaders argue that IP delivery is more physiologic than subcutaneous delivery (most recently at a JDRF/HCT workshop), but what will it take to design a product that is affordable and desired by patients, particularly as improved subcutaneous options become more cost-effective, less invasive, and improve in patient/prescribing burden?
• For insulin+glucagon bihormonal closed loop, Dr. Weinzimer displayed data from the latest BU/MGH publication showing home use of a bihormonal bionic pancreas decreases average blood glucose and time in hypoglycemia over the team’s 11-day multicenter study. Participants were randomly assigned to bionic pancreas (n=20) or usual care (n=19, conventional or sensor-augmented pump). Over the duration of the trial, the mean blood glucose was ~140 mg/dl in the bionic pancreas group, and ~162 mg/dl in the comparator arm. Time in hypoglycemia was also significantly lower in the bionic pancreas arm (0.6% vs. 1.9% time <60 mg/dl). The bionic pancreas has conferred very strong glycemic outcomes in studies (mean glucose + hypoglycemia), and the qualitative meal announcement (no carb counting) and easy initialization (based solely on patient weight) make it an appealing option for the user. Skeptics point out that the incremental value of glucagon vs. its cost is an unknown – though this is often the case with new innovations. As a reminder, an NIDDK-funded pivotal trial of the bihormonal bionic pancreas is slated for mid-2018, while the insulin-only version is expected to enter its pivotal trial in late 2017/early 2o18.

Hybrid Closed-Loop Insulin Delivery Systems – European Perspective

Roman Hovorka, PhD (University of Cambridge, UK)

The highlight of Dr. Hovorka’s presentation was an update on several closed loop studies set to begin in the near future. Included in their ranks are several notable studies: KidsAP testing Cambridge’s closed loop over 12 months in patients 1-7 years old; CLOuD, a 24-month  study in newly diagnosed 10-17 years olds (C-peptide is an outcome), and two Australian studies of the MiniMed 670G. Dr. Hovorka also shared that his Cambridge group’s trials, which have overwhelmingly shown increased time in range and decreased hypoglycemia, have cumulatively enrolled 190 patients and studied 125,949 hours of closed loop operation. Whoa! By our calculation, the NIH-funded, 12-month DAN05 trial of Cambridge’s system (6-18 year olds) should multiply that number of closed loop hours studied by 10 – n=130 over 12 months equates to over one million hours! Thanks to NIH, JDRF, and other funding bodies, our understanding of these systems is going to grow exponentially in the next three years! Otherwise, Dr. Hovorka mostly reviewed the team’s previously impressive data in a variety of settings (adults, children, pregnancy) and again highlighted the variability in overnight insulin delivery typical with closed loop (no two nights are the same, and closed loop can cope with that variability).

Upcoming Planned Large Closed-Loop Efficacy Studies

Symposium: Digital Data – Clinical Liability and Patient Safety

Guidance From The Food and Drug Administration

Courtney Lias, PhD (FDA, Silver Spring, MD)

In an encore of previous presentations (NIH Workshop, AADE), the FDA’s Dr. Courtney Lias shared her long-term hopes for an ecosystem of compatible and interchangeable diabetes devices (pumps, CGMs, apps) that patients can swap in and out. She reiterated her goal of device communication standardization, where automated insulin delivery components can exchange data seamlessly without separate approvals for every conceivable system combination (i.e. sensor A + algorithm A vs. sensor A + algorithm B). We continue to love this vision and look forward to tangible steps that make it a reality. While much of this vision we have heard from Dr. Lias before, we are glad to see her continuing to share it publicly – it reinforces the agency’s commitment to making it happen. We also noticed that Dr. Lias included in her message today that type 2s can benefit from this ecosystem too, suggesting that FDA is aware of the importance of enabling this population (i.e., not just technology for type 1s). Indeed, there is tons that type 2s – especially those on insulin – would gain from the CGM, pump, and algorithm developments in type 1, and we hope with greater innovation and population expansion costs will come down. We were thrilled to hear Dr. Lias’ broader message of prioritizing patient choice. Below, we include a few of the standout quotes that highlighted her patient- and innovation-friendly presentation:

• “What we need to work toward is communication that is SO CLEARLY worked out that FDA won’t even have to worry about [next-gen devices] being interoperable with existing platforms. They just will be. We need more standardization.”
• “Diabetes devices do not meet the expectations of consumer devices. For example, I just bought a Sleep Number bed, which is yet another example of a connected device. It lets me optimize my sleep and – unsurprisingly – it comes with an app. And guess what? If I wanted to, I could even set up an Apple Watch with a linked app. And for ALL of this, I would NOT have to look at a manual. I’m just able to use it. That’s where we need to get in diabetes.”
• “You may say that the downside of these connected, consumer devices being used incorrectly is not as high as the risk of an insulin pump being used incorrectly. That’s true. So security is important. But, I do my banking online. So as a consumer, I have the expectation that these things are possible.”
• “For patients with diabetes, the numbers can make all the difference. But if you can’t get to data in a format you need or when you need it, then it’s no use at all … Ultimately, our current solutions don’t help patients as much as they could.”

Symposium: Cognitive Functioning and Decision-Making in Diabetes

The Right Devices at the Right Time

Katharine Barnard, PhD (Bournemouth University, UK)

In Dr. Katharine Barnard’s view, technology has massive potential to change the diabetes management landscape, but psychosocial considerations can’t be ignored: With great technology, there needs to come effective on-boarding, support, and expectation management. Like her comments at the JDRF-Helmsley Charitable Trust Closed Loop Intra-Peritoneal Infusion Workshop, she emphasized keeping patients' individual needs and lifestyles in mind when choosing which devices can best be utilized. She stressed that health is a state of complete physical, social, and mental wellbeing, not just the absence of disease – such a great point, and shades of former Surgeon General Dr. Vivek Murthy. While tech can contribute enormously to this end, simply giving someone a device is not sufficient.

• Healthcare providers express significant anxiety regarding implementation of closed loop systems. Dr. Barnard reported that, while 91% of providers believe closed loop systems will optimize diabetes control for patients with type 1 diabetes and 66% said any insulin-dependent patients could benefit from the approach, 63% of providers believe closed loop devices will require more professional time than previous approaches, and 95% believe more educators will be necessary. These figures demonstrate an underlying need to support clinicians just as much as supporting patients – technology moves fast, clinical practice changes slow, and devices have historically added burden to providers (e.g., EHRs, reimbursement and prior auths, cable downloading, etc.). We expect the prescribing burden of closed-loop systems will follow a similar adoption curve to other innovation – early adopters, early majority, etc.
• Dr. Barnard: “I have not met someone who does not genuinely believe they are trying hard to manage their diabetes.” And yet, outcomes are still not where we need to be. As Dr. Howard Wolpert noted in a panel discussion the next day, every patient wants to have good glycemic control – “they don’t need a sermon about that.” Dr. Barnard stressed that knowing when to and when not to prescribe a device is critical. She called for the development and use of device-specific measures to assess psychosocial impact for people with diabetes and partners. In addition, she expressed a need for ensuring robust, replicable, and consistent psychosocial assessment in clinical trials to better understand patient experience, despite the common criticism that this is a difficult metric to record. Lastly, health care providers must be supported in their use of appropriate psychosocial measures and be given tools to aid in decision making.
• Dr. Barnard reminded attendees that people with diabetes think about their condition A LOT (~600 times a day, according to one estimate). Therefore, devices have to hit a high bar: free up some of this headspace and improve quality of life! For some patients, this may mean not implementing a pump if cost or discretion is critical to the individual; for others, the number of alarms associated with CGM may be a significant turnoff. The answer, Dr. Barnard argued, lies with the healthcare provider, who is responsible for facilitating communication around patient needs and barriers.

Symposium: Technology and Automation for Inpatient Insulin Management

Advances in Glucose Monitoring in the ICU

Stanley Nasraway, MD (Tufts Medical Center, Boston, MA)

Dr. Stanley Nasraway characterized the future of the inpatient glucose management field as “chilling,” in part due to ambiguous FDA metric targets and the NICE sugar trial results, which put a major damper on the field and greatly reduced investment. According to Dr. Nasraway, there are just three companies pursuing inpatient glucose management devices – OptiScan, Glysure, and Admetsys – and the OptiScanner is the only continuous glucose monitoring system to have completed a pivotal study in the US. The OptiScanner 5000 is currently under FDA review and has been shown to be incredibly accurate in critically ill patients: A multi-center, accuracy trial (n=200) demonstrated a MARD of 7.6% with over 95% of the data collected located in Zone A of the Clarke Error Grid. Furthermore, population coefficient of variation (PCV) was reported to be an impressive 9.8% (PCV commonly falls in the 20%-30% range for similar systems). The device connects to a venous line, extracts just 0.1 ml of blood, centrifuges the sample, and performs midinfrared spectroscopy to directly measure plasma blood glucose. Spectroscopy requires zero calibration, and the system provides automated readings, generally set to record every 15 minutes. We especially like the automatic and accurate readings, which translate to huge reductions in burdens for nurses, as well as improved information on trends and time in range, particularly at night. We do expect the subcutaneous CGM companies to move into this area, though it will obviously take them time as they continue to focus on the outpatient markets. Cost-effective and accurate inpatient glucose monitoring is critical, as patients are often unable to self-administer their intensive insulin therapy and may have extraneous physiological factors altering their glycemia. We hope this field sees a resurgence as the technology improves and studies show the clear ROI of saving nursing time and (hopefully) improving patient outcomes.

• In an already incredibly limited market, the statuses of the remaining two companies, GlySure and Admetsys, are currently unknown. The GlySure system operates via a fiber optic sensor, which sits in the blood stream, requiring an initial three calibrations followed by once-daily calibrations for the remainder of use. A somewhat promising MARD of 9.9% was reported; however, Dr. Nasraway claims GlySure is now targeting the outpatient market, so progress for the inpatient population is unknown. Admetsys is a closed-loop system that mixes dextrose and insulin through a central venous catheter with another line for a glucose oxidase sensor. The system aims to prevent hypoglycemia and keep glucose levels within a defined range. Whether this will be implemented in the ICU is unclear and the company has no plans to test in the inpatient setting until at least 2018.
• Just days before ADA began, Admetsys received an investment (of undisclosed size) from T1D Exchange. The funding was part of T1D Exchange’s “multi-million dollar initiative” to support the development of automated insulin delivery technologies. As a reminder, Bigfoot Biomedical won the first investment from this initiative back in February.

Mini-Symposium: The Future Face of Diabetes Care—Beta-Cell Replacement vs. Technology (Supported by a grant from The Leona M. and Harry B. Helmsley Charitable Trust)

The Near-Term Future of Diabetes Management Is Closed-Loop Pump Technology

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

Upon taking to the podium, JDRF Chief Mission Officer Dr. Aaron Kowalski immediately reframed the cell replacement therapy vs. closed loop debate by changing the title of his talk to “The NEAR TERM future of diabetes management is closed-loop pump technology.” We agree that an ultimate win might be cell and immune therapies, but with the state of diabetes care today and the time horizon for advanced biological therapies, we need better near-term solutions for patients. “Mechanical solutions are much easier than biological solutions,” he explained, “It’s hard to replicate evolution – hummingbirds flap their wings very fast. Try to have a man fly – it’s a really, really hard problem to solve. With technology, we can do not exactly the same thing (i.e., a helicopter), but amazing things. The goal to create a machine that functions like an islet is a lofty goal. To fly like a hummingbird is hard, but we’re taking incremental steps – low threshold suspend, and now the first hybrid closed loop.” He expressed great enthusiasm for the closed loop landscape, first overviewing the “really exciting” Medtronic MiniMed 670G pivotal data (shared at ADA last year), and then alluding to other companies looking to commercialize similar products: Animas, Bigfoot, Tandem, Insulet, Roche, and Beta Bionics (and also DIYers). Still, he acknowledged that there are many challenges relating to hardware that keep it from behaving like an islet: (i) Insulin kinetics – subcutaneous insulin is not physiologic. To Dr. Kowalski, IP insulin delivery is a slam dunk and a “better way to do it,” but the challenge lies in creating a device that people will want to adopt en masse (this topic was debated extensively at the recent JDRF/Helmsley Charitable Trust IP AP Workshop); (ii) Burden of having to wear a device – “ nobody in the history of man that I’m aware of wears an insulin pump just for kicks”; (iii) Biologic variability on a patient-to-patient basis; and (iv) Exercise. While a cell therapy would presumably address all of these issues, JDRF is still investing resources in next-generation closed-loop technologies (smaller form factor, implantable, longer wear, more automation, additional inputs, adjunct therapies) – see this slide for a list of new JDRF projects in 2016 and projects in the pipeline for 2017.

Diabetes Mine D-Data Exchange

Highlights: Insulet Shows Cool Horizon Screens; FDA’s Strong Call for Interoperability Between Companies; Dexcom API This Year

At DiabetesMine’s D-Data Exchange, we enjoyed seeing the DIY community and industry in what we felt was the most productive discussion yet. Highlights included Insulet’s Dr. Trang Ly on the OmniPod Horizon system (first screenshots from near-final design – major investment here), FDA’s Dr. Stayce Beck on why device interoperability really matters for automated insulin delivery progress (FDA may even have a workshop), Dr. Nate Heintzman gave an update on Dexcom’s APIs for retrospective data at developer.dexcom.com (launching later this year), and an artificial pancreas progress panel included views from Medtronic’s Dr. Fran Kaufman, Bigfoot’s Bryan Mazlish, Insulet’s Dr. Ly, Tandem’s John Sheridan, and our own Adam Brown. MIT innovation professor Dr. Eric Von Hippel gave an outstanding keynote, emphasizing that the lead-user (DIY) community is a critical driver of innovation in all fields – diabetes is no exception. The HOW of this for industry is still an interesting question – how should/can/will industry create open sandboxes to enable innovation for the DIY community and build it into commercial products? We felt the conversation dynamic shifted this year and was more patient-friendly, “how-to,” and pro-DIY than ever before.

• Insulet’s Dr. Trang Ly (VP & Medical Director) gave a first glimpse of the Horizon hybrid closed loop user interface on the Dash PDM – wow is Insulet investing heavily in user experience! The screenshots below are “close to the finish line” of the design process, with a prominent display of the CGM value and trend, IOB, and bold use of colors. Other information has been kept to a minimum, a good move in our view. In addition to the clinical trials testing the algorithm, Insulet has already performed six usability studies (31 unique participants to date, including some 670G users) to get feedback on the Horizon’s user interface and data display. Dr. Ly also highlighted Insulet’s cool use of Lightning Labs – condensed user experience design processes that occur in a short period of time with cross functional teams. The group designs and iterates quickly based on user feedback. Insulet even invited six members of the OpenAPS community and spent hours hearing about their experiences – yes! She noted that user interface’s use of colors is “controversial” (some like it, some don’t), and the company is also debating how to display the system’s insulin dosing graphically (microboluses vs. basal rate can be confusing). Dr. Ly shared sincere commitment to getting lots of patients’ feedback (including from MDIs), and Insulet is clearly investing deeply in this area. She reiterated the expectation for a pivotal trial of the Horizon product next year. As previously described, the algorithm will be embedded on the pod and talk directly to the Dexcom G6 CGM, meaning users will remain in closed loop even when the PDM is out of range.
  
  • To date, Insulet has completed studies of the algorithm in 82 patients, collecting ~3,500 hours of closed-loop data: “The algorithm is doing what it should be doing – reducing mean glucose and increasing time-in-range without causing more hypoglycemia. We’re well on our way to creating a commercial artificial pancreas product.”

• FDA’s Dr. Stayce Beck hoped for an interoperable future where AID systems have multiple interchangeable components made by different manufacturers (“plug-and-play”). The Agency “hopes” to have a workshop on this topic and seems highly committed to continuing this discussion. Nice! Noted Dr. Beck, “The anticipated pace of AID innovation challenges the current regulatory framework for medical devices. Every time one of the components is modified, a company has to come in to FDA with a new submission. Users can’t mix and match systems that meet their needs. We’re at a point where technical solutions can be implemented.” We’ve heard this in theory before from FDA’s Dr. Courtney Lias, but it sounded like things have progressed internally at FDA, and Dr. Beck said the solution is “under construction” –  the entire community now must come together to help the FDA figure out how to do this. For instance, the field must nail down the interoperability, connectivity, and data approaches (the HOW), enabling devices from different companies to seamlessly and safely talk to one another. We see enormous innovation potential and patient choice enabled by plug-and-play systems – using one company’s pump, another company’s sensor, a third company’s algorithm, and being able to swap components in and out as desired. We hope a workshop does indeed happen and it drives progress and standardization in the field. Dr. Beck also commented on dosing insulin from a smartphone, sharing that the agency is open to it and very platform/device agnostic – companies must simply demonstrate the device works as intended and is robust to different failure scenarios (e.g., the mobile medical app will be prioritized when battery is low, a game is being played, etc.). We loved her opening comment, “We don’t see mHealth as a ‘challenge,’ but as an ‘opportunity.’”
• Dexcom’s Director of Data Partnerships Dr. Nate Heintzman shared that the company’s open API for retrospective data will launch “later this year” at developer.dexcom.com. As a reminder, this will allow third parties to access Dexcom’s APIs (retrospective data, three-hour-delay), create and manage pre-commercial (prototype) apps, play with simulated (sandbox) data, learn how to become a Dexcom data partner, and even submit an app for commercial approval. This important effort was first announced at the D-Data Exchange last fall, but clearly there are a lot to details to work out – the “early 2017” planned launch has slipped a bit, but things sounded very close to launch at this point. Dr. Heintzman shared a “top 10 things to know list,” noting that 500+ people have already signed up for email updates on developer.dexcom.com. He reminded attendees that developers will have access to quite a bit of information to develop novel retrospective data apps, including glucose data, statistics, device info, and calibrations. Like Dexcom Clarity, this platform is a class I medical device (retrospective CGM data), and Dexcom will support data partners and share best practices through the developer portal. This is clearly a huge internal investment from Dexcom and we hope it drives an ecosystem of innovation and novel ways to use retrospective CGM data.
• A lively panel discussion featuring closed loop industry members (Insulet’s Dr. Trang Ly, Medtronic’s Dr. Fran Kaufman, Bigfoot’s Mr. Bryan Mazlish, and Tandem’s Mr. John Sheridan) and moderated by our own Mr. Adam Brown showed encouraging industry openness to engaging with the DIY and broader patient communities. From the audience, Dr. Eric von Hippel called for data to be readily available to patients, providing what Adam called “a sandbox” for innovators to play in and access devices. “It’s crazy when 90% of the effort of the user community has to go to hacking into devices, just to get access to the data. Life is slightly unsafe. If you ensure safety, you also ensure slow progress. There must be a way – in every other field, users can take risk on themselves. If I want to build a rocket-powered car, I can do it, even though it’s unsafe. There must be a way to sign rights that I want access to this damn thing and it’s no longer your [industry’s] problem.” Another audience member emphasized that remote monitoring on a pump is absolutely not an optional feature – kids with diabetes are sent to school, where there are no or few trained professionals, and parents have the right and need to see how their child is doing. Still others asked for companies to re-think and improve infusion sets, the traditional four-year pump warranty cycle, and algorithm transparency. To our delight, none of the panelists once responded “No, we can’t do that.” Of course there is work to be done on the best vehicle for the DIY community to interact with industry productively, but panelists were very open to hearing the audience – Dr. Kaufman even asked attendees to “go on a date” with Medtronic so they can get as much input as possible. We find this level of progress quite remarkable, since this dialogue didn’t exist even two years ago – we owe a hearty thanks to these industry leaders and the patient innovators who have pushed the field ahead.
• MIT Sloan School of Management’s very smart Dr. Eric von Hippel, speaking to a room full of DIYers and lead users, encouraged some collaboration with industry…but not too much! In collaborating with companies too closely, non-industry innovators may end up with “indicia of commerciality,” which could result in a product that FDA can regulate (under the Commerce Clause, FDA cannot regulate non-commercial medical innovations). According to Dr. von Hippel, there’s more to lose than, for example, the OpenAPS community’s right to use their DIY closed-loop systems. He explained that truly patient-led, grassroots innovation systems interact quite robustly with the producer innovation paradigm to the profit of both parties: During a producer’s R&D phase, it will offer innovation support to free innovators, who in turn churn out innovation designs upon which manufacturers can base products. It’s a difficult line to toe, however, as too close an interaction – a blended system – could lead to FDA involvement, and the scaffold could crumble.
  
  • According a study from Dr. von Hippels’ group, the number of medical patient innovators outnumber producers by over 100:1. Holy moly! Innovation has traditionally been seen as something that just producers do, and this is perpetuated by the fact that innovation is only accounted for in government statistics until a product is formally introduced onto the market – the user innovator is typically invisible. In the US alone, there are an estimated ~384,000 individuals working on medical consumer innovations.
• According to Companion Medical CEO Sean Saint, Apple added insulin doses into Health Kit as of the Worldwide Developer Conference earlier during ADA. It will be available starting in iOS 11 this fall, and the beta is out now – see the iOS 11 developer page here. This has been long awaited and could be a nice data ecosystem enabler – especially as connected pens and connected pumps come out.

Tandem Media Day

Highlights

Tandem held a compelling, confidence-inspiring media day at its headquarters today, sharing updates on the pipeline and a tour of its manufacturing. From a patient perspective, we left the day feeling very positive about Tandem’s innovation path and priorities ahead. Highlights included:

• Tandem’s new t:lock infusion set connector is still expected to launch in 3Q17. We learned that some of Tandem’s distributors actually buy Luer Lock sets in bulk from Animas (!) at a lower price, resulting in “tens of millions of dollars” in lost revenue over the past four years. With the new t:lock connector, distributors will now have to buy infusion sets directly from Tandem, allowing it re-capture a meaningful amount of lost revenue. Unomedical will still manufacture the sets and all current models will still be available – just with a different cartridge-set connector. Tandem’s customers also win too, as moving away from the Luer Lock connector (invented in 1896 (!) and not designed for insulin delivery) and to the t:lock will save wasted insulin and reduce priming time by 30 seconds, correcting major desired customer improvements. (Both have been on customers’ top 5 wish lists for the past two years straight.) We saw the connector in person and the difference is hardly noticeable:

• We were blown away by the Tandem Device Updater demo, which took just a few minutes to update a t:slim X2 to add Dexcom G5 CGM integration. Wow! This is a very compelling feature and one we imagine all pumps will move to. Once the t:slim X5 G5 is approved (under FDA review and still expected this summer), customers will be able to update their t:slim X2 to add G5 integration before the new pumps even start shipping – talk about rolling out new innovation quickly via software!  The update process has clearly had a ton of human factors work put into it, with a series of checklists and a simple step-by-step process spelled out via an email and a user-specific update code. (Cool subject line: “Dexcom G5 Update Now Available for your t:slim X2.”) The software works on Windows and Mac and longer-term, over-the-air updates delivered to the pump via a phone and Bluetooth might be possible. We remain enormously excited about this update feature for pumpers, as it removes the hardware and cost hassles of upgrading. It will only become more compelling as Tandem moves to automation.
• We saw the first screenshots of Tandem’s first-gen secondary display mobile app. For the first time, management shared it hopes to launch by the end of the year. The app will talk to the t:slim X2 via Bluetooth, wirelessly upload pump data to t:connect, and mirror key metrics from the pump display. Bluetooth BGM integration is also a goal (retrospective data only), and phone users will also receive notifications of pump alerts. Future generations (no timing given) hope to add more ambitious features, like remote bolusing, decision support, training and education, share notifications, and automated pump setup.
• Tandem is “preparing” for pivotal study enrollment for its predictive low glucose suspend device with Dexcom’s G5. A launch is still planned for early 2018. This is right in line with the 1Q17 update, which expected the pivotal to wrap up in 3Q17.
• The second-gen TypeZero hybrid closed loop is still expected to launch by the end of 2018. We confirmed that the main study phase of the pivotal trial (IDCL) still hasn’t started yet. Management implied the hardware changes to go from the research platform on a smartphone to an integrated t:slim X2 with Dexcom’s G6 may be part of the reason for the delay. This isn’t surprising, given the magnitude of this trial and learning from the initial site training phase. Dexcom’s G6 sensor wasn’t finished with its pivotal study as of a month ago, so perhaps it isn’t ready quite yet.
• Tandem’s t:sport patch pump is still in development (no timing shared), and we got a first hands-on look at the small on-body component (roughly the size of Insulet’s OmniPod). Similar to Cellnovo, the plan is to have a 4-inch tubing that connects the pump to the infusion site. It’s still unclear whether smartphone app control will be possible, or whether a handheld will be the only way of interacting with the system. Tandem has obviously de-prioritized this project relative to the nearer-term automation pipeline.

• “Tandem is doing everything with international in mind.” No further details were presented, but obviously this represents another potential growth opportunity beyond the US – although one with potentially high costs. No timing has ever been given on this front.
• Management was honest that Tandem’s stock price is quite low right now (~$0.80; a “financing overhang” from a recent fundraise), and the company will need to raise additional capital. We believe the financial community continues to undervalue Tandem’s pipeline and overvalue the competitive impact of the MiniMed 670G. Management again characterized it as a pausing of the market and looked forward to the G5 integration launching. The power of Bluetooth connectivity and Tandem’s Device Updater should enable a steady stream of innovation, and Tandem’s marketing and attention to patient needs remains outstanding. Certainly, sales will need to improve as the year goes on, but this should happen assuming G5 integration launches as expected.

JDRF/NIH Closed-Loop Research Meeting Research Meeting

The Next Five Years

Frank Doyle, MD (Harvard University, Cambridge, MA)

Harvard’s Dr. Frank Doyle peered into his crystal ball and predicted that fully closed loop (no meal announcement), embedded algorithms, adaptive control, faster insulin, and multiple sensors will all be a reality in the next five years. None of the prophecies were earth-shattering by any stretch – we too believe they are within reach – but Dr. Doyle offered some interesting commentary on the near-term future of closed loop. He sees “glimmers of hope” in the algorithms arena, chiefly from the room for improvement in processing power embedded on a pump (“the [computer power] footprint on a pump pales in comparison to that on a cellphone”). Even so, he pointed out, existing capabilities can be extended by energy-aware algorithms or smart glucose monitoring. The algorithm will also have to take into account the PK/PD of faster insulins, if and when they do come to market (FIasp is already available outside the US and was resubmitted to FDA recently, with a decision expected in 3Q17). This shouldn’t be a problem ­– Dr. Doyle’s colleague Dr. Eyal Dassau has previously said that faster onset insulins would only make algorithms more effective. Lastly, he called for the integration of additional “sensors” into closed loop systems. The systems he is referring to, however, are not advanced biometric analyzers, but simple, everyday applications and consumer devices such as calendar events, health apps, fitness trackers, and GPS – these all produce rich and valuable information that should not be neglected for algorithm performance. Dr. Doyle reminded the audience that technology predictions are always a fraught area, and “the best way to predict the future is to create it.”

• Our own Adam Brown brought up a critical question in Q&A: What features(s) will significantly drive closed-loop adoption to a broad population, and how should research dollars be allocated accordingly? For closed-loop technology to reach its full potential – improving lives at a massive public health scale – what features are needed? What must improve or change from current devices? Adam argued that cost will be the number one driver of this field’s expansion, given what we hear from patients (even in the US), studies of device barriers (e.g., Hood et al., Diabetes Care 2016), and the success of products like FreeStyle Libre. While burden and system performance are certainly critical, cost is the big driver of diabetes technology adoption in our view – the best closed-loop system is meaningless if patients/payers cannot afford it or see value in it. The closed-loop research community continues to focus a bit more narrowly on algorithm optimization, which is important for increasing the value of system, but is unlikely (in Adam’s view) to be the major driver of this field. We believe significantly less expensive closed loop systems would be a very high ROI area of research – and something fit for the community (NIH, JDRF, academia) to further address, as JDRF is doing in its T1D Outcomes Program, as we understand it. From our view, of course, showing the value is undoubtedly critical, since that lead to better reimbursement. We also like the point of getting all the technology right first. If cost is, by anyone’s definition, a critical metric for this field’s commercial viability, how do research priorities change? Costs might also influence study design: “what trials will most influence payers to cover closed loop” vs. “what trials would answer scientific/algorithm questions?” In a world where R&D/payer spending per patient has been shrinking, but where we’d like to see it increase, we see value in this area to improve public health on a broader scale. We think it would be valuable to look to the technology community, where over time, costs have dropped precipitously for various products – products, of course, that do not have costs like regulatory, admittedly. How could the field get to 80% of the benefits of closed loop systems with 20% of the costs? Would it ever be possible that the field make current pump+CGM therapy 10x less expensive – or, could value be much better demonstrated so that payers will be poised to deliver technology to a broader population?

Overview

John Lum (Jaeb Center for Health Research, Tampa, FL)

Jaeb’s very smart and humble Mr. John Lum kicked off the annual meeting by briefly identifying the year’s three major themes in closed loop since ADA 2016: eagerly-awaited approvals; academic-industry translation and partnerships; and much larger, longer-duration closed loop studies. For this community, the MiniMed 670G tops the list of eagerly-awaited approvals and launches. This was indeed a tremendous milestone for the field – a first commercial hybrid closed loop product – but Mr. Lum also included Abbott’s FreeStyle Libre Pro and a replacement claim/Medicare reimbursement for Dexcom’s G5 CGM on his list. Next up, academic-industry translation: Mr. Lum applauded Insulet for licensing the UCSB/Harvard/Mode AGC algorithm (which was developed with funding from JDRF and NIDDK), and Tandem for licensing the Cameron/Bequette/Buckingham/Chase PLGS algorithm. We agree that it’s fantastic to see the fruits of academic researchers’ labors included in future commercial products. Bigfoot Biomedical, Mr. Lum added tongue-in-cheek, didn’t quite make the academia to industry jump, but Jeffrey Brewer, Lane Desborough, and Bryan Mazlish are all making the leap from “parent” to industry look easy. Last, Mr. Lum excitedly discussed the four large NIH-funded pivotal AP trials starting in 2017-2018: iDCL trial, Cambridge pediatric study (DAN05 – first subject enrolled June 2017), FLAIR study (670G vs. 690G with Drs. Rich Bergenstal and Moshe Phillip), and the BU-MGH bihormonal study. Following this field every day, it’s easy to forget to take a step back and consider the tremendous progress that has been made toward an artificial pancreas, not only in this past year, but in the past decade – closed loop seemed like a research project not very long ago:

• Mr. Lum reminded attendees that the first JDRF closed loop meeting at ADA took place almost exactly a decade ago – and it looked a lot different. The meeting was driven, of course, by Dr. Aaron Kowalski, and there were only 12 invitees. At that time, discussion was algorithm-centric, CGMs were not nearly good enough, and there were no products on the horizon. Needless to say, we’ve come a heck of a long way, and the launch of the MiniMed 670G has ignited waves of nostalgia – at last July’s NIH AP Workshop, Dr. Kowalski recalled how the artificial pancreas was just a concept in 2005, with incredibly scarce data. Dr. Bruce Buckingham followed by expressing amazement that in just 11 years, talk has shifted from conceptual feasibility to quality of life in real-world use. Nice!

“Right Now!” What’s Being Done with Closed-Loop Systems Today?

Richard Bergenstal, MD (International Diabetes Center, Minneapolis, MN)

Running with the theme of “how far we’ve come,” Dr. Rich Bergenstal reviewed the steps the closed-loop field has taken in the past decade to get to where we are today, to the point where “this is going to become the standard of care.” “It all began with a roadmap,” he began, as he pointed to Dr. Aaron Kowalski’s publication several years ago. “We have Dr. Kowalski to thank for that, though he’d probably give the credit to JDRF, and JDRF would probably give the credit to the larger community.” This roadmap has been followed to the T, and guided various researchers and clinicians through in silico modeling, inpatient trials, short-term supervised trials (hotel studies; or in Dr. Bruce Buckingham’s case, B&B studies), longer at-home studies, review articles and editorials, and finally hybrid closed loop meta-analyses and more editorials. The most intriguing part of his talk was a 17-year-old’s CareLink report, shown in the (admittedly blurry) image below. The pink trace and time-in-range bars on the right represent 670G manual mode; the blue trace and time-in-range bars on the left represent 670G auto mode. The difference, not only overnight, but all day, is striking –a lower mean glucose at all times, far less variability, and time in range nearly tripled with just a two-percentage-point increase in hypoglycemia. In manual mode, this user was spending ~nine hours per day between 250 and 400 mg/dl! Very compelling and a testament to this technology nice applicability in those running at very high glucoses. How do we get more patients on these devices? According to Dr. Bergenstal, “it’s up to us” to address issues of cost, access, and equity. Hear, hear!

Mini-Symposium: Exercise for Type 1 Diabetes (Supported by a Grant from The Leona M. and Harry B. Helmsley Charitable Trust)

Mechanisms Regulating Glucose Metabolism During and After Exercise in Type 1 Diabetes

Iñigo San Millán, PhD (University of Colorado School of Medicine, Denver, CO)

Dr. Iñigo San Millán described the glucose-altering physiological events that occur as a result of exercise and their implications for exercise type, cool-down, and closed loop. The muscle has two pathways through which it uptakes glucose, one insulin-dependent and one insulin-independent. The insulin-dependent pathway operates like that of any other tissue – insulin receptors activate the transport of GLUT4 receptors to the cell surface, enabling glucose uptake. The insulin-independent pathway, also mediated through GLUT4 transport, is triggered as a result of muscle contraction and is largely regulated by calcium release, which increases during physical activity. During exercise, both pathways are recruited, ultimately leading to a drastic increase in the number of GLUT4 transporters at the cell surface, facilitating uptake of glucose by myocytes and subsequent muscular contraction. The combined efforts of these two mechanisms serves to explain the occurrence of post-exercise hypoglycemia experienced by those with diabetes, as more glucose is taken up by the muscle during activity than at rest when only the insulin-dependent pathway is at work. However, hyperglycemia can also occur, especially immediately after high-intensity exercise, possibly due to high catecholamine activity (especially adrenaline), less insulin-independent glucose uptake due to sudden cessation of activity, and/or high blood lactate levels leading to increased rates of gluconeogenesis in the liver. Cool down exercises have been shown to mitigate hyperglycemic events following intense physical activity, potentially by stimulating enough insulin-independent glucose uptake to counteract heightened gluconeogenesis and the body’s response to catecholamines. Dr. Millán advised that these physiological pathways, as well as the individual metabolic changes that accompany improved fitness, be considered when designing the artificial pancreas for exercise.

Exercise and the Development of the Artificial Pancreas

Michael Riddell, PhD (York University, Toronto, Canada)

While the advent of an artificial pancreas ushers in some truly incredible and exciting benefits, Dr. Michael Riddell cautioned that the power, practicality, and patient burden of this new technology must be carefully considered. In particular, Dr. Riddell detailed six hurdles he anticipates regarding the implementation of automated insulin delivery during exercise, including: (i) variability in exercise type and intensity; (ii) methods by which an exercise event may be registered by the system; (iii) potential deterioration of CGM sensor accuracy; and (iv) hormone responsiveness. For now, hybrid closed loop systems improve time in range and reduce hypoglycemia (particularly overnight) very effectively, but other scenarios such as postprandial control, exercise, and general day-to-day variability (e.g. in sickness or pregnancy) will require innovations in insulin kinetics, insulin delivery, sensor lag, additional hormones/adjunct therapies, and/or additional inputs. 

• Regulation of glucose homeostasis is complex and varies with exercise type and intensity, posing unique issues for the use of closed loop during physical activity. While steady state aerobic activity like long-distance running tends to result in a decline in blood glucose levels, intense anaerobic activity such as powerlifting can induce an adrenaline response that raises glucose levels, increasing insulin requirements. Dr. Riddell noted that mixed activity often has a moderating effect on glucose and is thus the safest form of activity for initial experiments. To complicate matters further, individual responses to these exercise types also vary (and response can be different on different days), raising important questions as to whether the algorithms can be tailored to an individual’s specific reactions. Dr. Riddell proposed that physiological input measures such as heart rate and body temperature be considered to adjust for exercise intensity. An ADA oral out of OHSU showed that automatic exercise detection via heart rate and accelerometer sensors, combined with glucagon, markedly decreased exercise-induced hypoglycemia.
• While there are several fitness trackers on the market and some closed-loop feasibility studies have added them, we haven’t seen any compelling data suggesting these additional signals meaningfully improve systems. A study conducted at Yale found that under normal settings, the artificial pancreas is not sufficient to protect against hypoglycemia during exercise, thus supporting the need for anticipation of an exercise event (in the absence of an announcement ~1 hour or more prior). Dr. Riddell provided several suggestions for possible triggering metrics, including heart rate, body temperature, and lactate levels. Fitness trackers are a potential solution, but engineering efforts will be required to advance data access and facilitate the interface – this area is still quite early in the algorithm world. Of course, by the time an activity tracker detects exercise, halting insulin delivery will still take about an hour to impact glucose, meaning faster on-off insulins may really be what drives this field.
• Adding glucagon improves systems’ ability to mitigate exercise-related hypoglycemia relative to insulin-only systems. In fact, a recent trial from the Montreal closed loop group found that participants engaged in physical activity spent on average 24% of the time in hypoglycemia (<60 mg/dl with symptoms or <54 mg/dl with or without symptoms) when using a single hormone system as opposed to 12% of the time when using a dual hormone system. The real test will be in comparing these systems in daily use, the user experience pros/cons of each (glucagon allows for more spontaneity, but is also more supplies to manage), and the cost-benefit.

Corporate Symposium: Continuous Glucose Monitoring as Standard of Care: Clinical Outcomes, Improved Access and Therapeutic Use (Sponsored by Dexcom)

Continuous Glucose Monitoring in Patients Treated With MDI

Richard Bergenstal, MD (International Diabetes Center, St. Louis Park, MN)

Dr. Richard Bergenstal provided a thorough overview of the utility of CGM in patients on MDI (both type 1 and type 2), focusing on the proven benefit of CGM independent of pumps. He pulled from a host of recent literature – including the JAMA-published DiaMonD and GOLD studies – to make a number of important points: (i) CGM indisputably improves glucose control when used consistently; (ii) CGM improves time-in-range without increasing hypoglycemia; (iii) CGM is beneficial regardless of education, age, numeracy, and a number of other demographic factors; and (iv) CGM reduces diabetes distress. He emphasized, too, that access to CGM readings has been shown to improve glucose control WITHOUT changing daily insulin dosage, suggesting that there is a significant effect on lifestyle (that, in his view, is often underappreciated). Dr. Bergenstal closed by applauding the FDA’s non-adjunctive label approval for Dexcom’s G5 (“a HUGE step forward”) and sharing big hopes for the benefit of CGM in the Medicare population (once the details of administering coverage are figured out).

Clinical Review Of Patients

Steve Edelman, MD (University of California San Diego, CA)

In a discussion of the clinical benefits of CGM, the outspoken Dr. Steve Edelman left no doubt about his views on the value of the technology, stating “CGM is the single most important advance for type 1 diabetes since the discovery of insulin,” and going on to note even more emphatically that, “not using CGM in type 1s is bordering on malpractice.” He highlighted the value of trend arrows (“if you don’t look at those, you’re missing one of the most important parts of your CGM”), the challenge of adjusting insulin around glucose variability (“every day is different for a person with type 1 … despite following the same rules”), the importance of education around alerts and alarms, and the even brighter future for CGM (more accurate, less interference, etc.). Of course – as he has done before – he also did not pass on the opportunity to share with providers his “strong” view against blinded CGM: “If you can engage a patient, then that’s the key. Engagement is the KEY with CGM. You want patients to see in real time, ‘Gee, look what happens when I forget my medication!’ Do we blind home glucose monitoring? No, of course not!” (We’d note that for those not willing to wear real-time CGM or unable to afford it, blinded CGM is better than very limited/no glucose data at all.)

iDCL Update

Boris Kovatchev, PhD (UVA, Charlottesville, VA)

Dr. Boris Kovatchev’s broad overview of his team’s contribution to automated insulin delivery research was highlighted by new preliminary data from the now “complete” training phase (n=20) of the NIH-Funded International Diabetes Closed-Loop (iDCL) Trial that continues to show promising time-in-range and hypoglycemia numbers in a small group of patients. Dr. Kovatchev shared data from 11 individuals in which the training protocol achieved strong glucose control consistent with past results shared at ATTD: overall mean glucose of 145 mg/dl (150 mg/dl in the “first half of the night”; 125 mg/dl in the “second half of the night”), just 1% of the time <70 mg/dl, and 77% of the time in 70-180 mg/dl (75% in the “first half of the night”; 91% in the “second half of the night”). We have not seen this cut of the overnight iDCL data before, though the improvement as patients get further into sleep is consistent with this algorithm and awesome for the user experience and waking up around 120 mg/dl to start every day. Notably, Dr. Kovatchev also highlighted the very notable win that patients saw no readings <60 mg/dl or >300 mg/dl, evidence that the system is mitigating the most dangerous of highs and lows, particularly with automatic correction boluses to stem highs. We hope this data will also be borne out by the main phase of the trial (CT.gov posting here), which Dr. Kovatchev confirmed will begin “soon” (no specific timing update, though Tandem said the same yesterday). As a reminder, the main study will serve as Tandem/TypeZero’s hybrid closed loop pivotal (still not yet open for recruitment, according to ClinicalTrials.gov) and will randomize 240 patients in a 2:1 ratio of closed-loop control vs. sensor-augmented pump therapy over six months.

• Dr. Kovatchev included a critique of the Bionic Pancreas in his discussion, noting that some of the results with insulin-only systems have actually seen less hypoglycemia than those of the dual hormonal system – “Interesting to note considering that glucagon is supposed to reduce this.” Of course, we’d emphasize that it’s impossible to compare across trials given the entirely different study designs (sedentary vs. exercise/meals), study populations (pediatrics vs. adults), meal bolus strategies (announcement or not), study lengths, and study settings (super challenging diabetes camp settings vs. less challenging daily ambulatory life settings vs. less challenging inpatient studies). In our view, this debate will only be settled when head-to-head data AND user preference information is available. The benefits vs. costs of different systems will obviously vary greatly between users and even within users over time.

Product Theaters

What's Next for the Omnipod System? Innovation at Work (Presented by Insulet)

Trang Ly, MD, PhD (Insulet Corporation, Bedford, MA)

In a packed product theater, Insulet provided an on-stage demo of the upcoming Bluetooth-enabled Omnipod Dash PDM (FDA submission in 2H17) and a first-ever deeper look at the user experience for the Horizon Automated Glucose Control PDM (pivotal trial next year, five-day, three-center hotel study starting next month). We snapped many never-before-seen pictures, posted here (notable shots below). Insulet CEO Patrick Sullivan actually opened the product theater, emphasizing the main topic: “meaningful innovation.” We first saw the new Bluetooth-enabled Dash PDM in the exhibit hall on Saturday, and this live on-stage demo left us equally impressed with the Android device and much-upgrade user experience. The Horizon demo was also strong, and it was actually done on an iPhone (perhaps for demo purposes only, or a sign that direct pod-to-smartphone transmission is a possibility at launch). It surprised us to see how far along the Horizon user interface is for the product – it looked quite polished, and even the Dexcom transmitter-pairing process was integrated (pictures here). The company is still wrestling with what to call the different device modes (e.g., manual vs. auto, closed loop vs. open loop, etc.), but the feedback-gathering process has been impressive: the talk paused twice and asked attendees to take a few minutes and fill out an on-seat survey to share thoughts on Dash and Horizon. (The last question on each related to willingness to prescribe the product in MDIs – Insulet’s clear target market.) Said Program Director of Advanced Technologies Jason O’Connor, “It’s so critical to not rush a product to market, just because it has functionality. It has to have the right user experience to deliver a product that is going to integrate into people’s lives. We may not be the first to market, but it does mean we’ll deliver an exceptional product.” [He then showed a picture of a lame, earlier-to-market smartwatch from Suunto vs. the far improved Apple Watch] Based on what we’ve seen, the company’s 110,000+ patients worldwide (65,000 in the US) will see very meaningful upgrades with both products.

• Insulet has now tested its algorithm in 82 patients (3,384 hours of closed loop, 70 nights). VP & Medical Director Dr. Trang Ly summarized the cumulative data in a useful table, including results from two new posters at ADA in 6-12 year olds (132-LB) and adolescents (1055-P). She noted that mean glucose is coming down and time-in-range is increasing as the studies go on, the latter now up to 85%. Dr. Ly said the algorithm is doing a “fantastic job” of doing what a good closed loop controller should do: improve time-in-range and lower mean glucose without causing more hypoglycemia. She did note that following the licensing of UCSB’s algorithm, Insulet made “some substantial changes” to ensure safety.

• Insulet is about to launch into a longer duration, five-day hotel study next month at three centers. Boy is the company lucky to have Dr. Trang Ly, who has done this rodeo many times with other closed loop studies in the academic setting. Since really committing to this product last year, the company has rapidly completed a number of studies!
• Horizon will offer some user control and customizability with the algorithm, including the ability to change the target glucose. The correction factor and basal rates are also inputs to the algorithm, unlike the MiniMed 670G – which automated insulin delivery unrelated to the preprogrammed basal rates. (The only two things users can adjust in the 670G to affect basal automation are the insulin:carb ratio and the insulin action time.) Interestingly, 35% of Insulet’s Glooko-using patients use extended wave boluses and temp basals, features Insulet will now plan to incorporate into its closed loop controller – this shows the power of user data to drive important product design choices!
• Said Dr. Ly: “The pediatric population is very important to us. We’re not going to market without having done pediatric studies...What people talk about is the incredible improvement in quality of life. That’s why people use Loop and OpenAPS – it makes them feel a lot of safer at night.”
• The brilliant Dash platform allows Insulet to update the PDM’s software and add automation and concentrated insulin, but keep the same pod. We see this as compelling and quite differentiated, an approach Tandem has also taken with its t:slim X2 platform.
• Dr. Ly mentioned Insulet’s “Incredible collaboration with Glooko, which now reaches 44,000 patients in the US – they have a mean age of ~30 years, an average glucose of 185 mg/dl (eA1c: 8.1%), take 4.3 fingersticks per day, and bolus about 4.9 times per day. Presumably Insulet can start to stratify this data and learn what different centers or patients do differently – what behaviors separate those doing really well from those struggling?

MiniMed 670G System with SmartGuard HCL Technology: Driving Patient Outcomes through Automation (Presented by Medtronic)

Richard M. Bergenstal, MD (International Diabetes Center, St. Louis Park, MN)

Medtronic’s MiniMed 670G product theater was standing-room-only (five-deep!), offering clinicians best practices for setting expectations, tweaking parameters, glimpses of CareLink reports, comments on Guardian Sensor 3, and two impressive before-after case studies. IDC’s Dr. Rich Bergenstal ran through the rationale for the hybrid closed loop device, emphasizing that “Current therapies cannot adjust for intra-patient variability. How can we expect one basal rate? There is no such thing as A basal rate. It’s changing every five minutes!” He joked that while he often thinks he is “clever” with complicated basal schemes, the reality is basal needs change every night and throughout the night – sometimes, the 670G is delivering much less than pre-programmed, and sometimes it delivers much more. This will be a change for HCPs!

• A slide headlined, “It is vital to set realistic expectations with patients,” shared the following advice for HCPs: Glucose levels will not be perfect – they will improve over the first few weeks as the system learns the patient and the patient learns to use the device. Trust the system and let it work as designed. The goal is to maximize time in Auto Mode. Focus on time-in-range (70-180 mg/dl) using CareLink reports and sensor glucose review screen on the pump for immediate feedback on progress. A higher degree of engagement is needed at first (i.e., respond to alerts and take action as guided) and there is a learning curve for patients and HCPs. CareLink data review is essential. Encourage patients to simply read pump screen and follow instructions.” (This last one may be paternalistic for type-A patients, though for the vast majority it will almost certainly improve outcomes.)
• Dr. Bergenstal reminded attendees that the preset Manual Mode basal rates have nothing to do with Auto Mode. Only two settings that relate to Auto Mode can actually be fine-tuned by an HCP or patient: the insulin-carb ratio and the insulin action time. As he did at ATTD, Dr. Bergenstal said that tightening the insulin:carb ratio can help with postprandial highs on the 670G. He mentioned in Q&A that the insulin action time doesn’t “play a huge role,” but it can increase the size of the manual bolus correction doses. At his clinic, the team usually uses a three-hour insulin action time, though sometimes they have moved down to 2.5 hours. 
• Drs. Bergenstal and Sherr covered two fascinating MiniMed 670G case studies in young patients with high A1c’s, showing quite dramatic improvements in time-in-range. In a 14-year-old teenager previously on Dexcom CGM, the 670G drove estimated A1c from 8.5% (Manual Mode) to 7.2% (Auto Mode), with time-in-range (70-180 mg/dl) improving from 44% to 73%. “The only time you get a teenager at 93% wearing a device is if they believe the investment is worth it,” said Dr. Sherr. The second case, a 23 year-old with a baseline A1c of 8.8%, spent just 27% time-in-range in manual mode. After switching to Auto Mode, time-in-range improved markedly to 75%. While these were obviously cherrypicked case studies, it will be interesting to see the spectrum of improvement on this product – who will benefit the most? Who will not?
• Dr. Sherr praised the much-improved Guardian Sensor 3, though Medtronic is definitely emphasizing “3-4 calibrations per day.” As a reminder, two calibrations per day are required at minimum, though the MARD improves to 9.6% on 3-4 per day – this also allows Guardian to cross the “10%” MARD threshold. The paired Ascensia Contour Next Link 2.4’s strong accuracy was also a big selling point. Interestingly, Dr. Sherr said that the transmitter’s on-board chip that monitors sensor health “puts the onus back on technology” and seems to relieve her patients of self-blame when the sensor isn’t working: “Oh, I didn’t do something wrong.” This feature also reduces outliers, which are one of the most frustrating parts of wearing CGM.
  
  • Dr. Sherr mentioned that Medtronic’s new “oval tape is like gold” and helps the sensor stay on for seven days. This is the first we can recall hearing of this tape change. Medtronic has of course retained the underwhelming clamshell transmitter, though we expect this to be phased out eventually, per the FreeStyle Libre-like transmitter pictures from the 2016 Analyst Day.
• Dr. Bergenstal showed several CareLink reports (see below), which highlight before-after glucose-profiles in Manual (red) vs. Auto Mode (blue). Those he showed confirmed data from the pivotal trial – less variability, especially overnight. We like that these reports place a big focus on time-in-range, depicted with large colored bars on the bottom left side. The before-after plots should also convince skeptical HCPs that the device is making a difference.
• Medtronic is positioning overnight use of the 670G in Auto Mode as “auto pilot,” while daytime use is “co-pilot.” We like this analogy and find it much more intuitive than “Hybrid Closed Loop.”

-- by Adam Brown, Brian Levine, Maeve Serino, Varun Iyengar, Melissa An, John Erdman, and Kelly Close