DTM 2018 (Diabetes Technology Meeting)

November 8-10, 2018; Rockville, MD; Day #3 Highlights - Draft

Executive Highlights

  • In the fan-favorite annual CGM updates session, we (finally!) saw Dexcom G6’s accuracy stacked up against the specific iCGM special controls (lower 95% CI bound included), first G6 accuracy data for 14-day wear and 1-hour warmup, longevity data on a new and improved adhesive, and plans for diminished lag time. Abbott shared day-by-day 14-day wear accuracy and discussed new manufacturing protocol through which each individual sensor is factory calibrated. Medtronic detailed electrochemical impedance spectroscopy as an algorithmic means to boost accuracy while lowering user calibrations (“self-calibrations”). Lastly, WaveForm (AgaMatrix) shared new photos and details on its 14-day wear, one cal/day Cascade CGM, now slated for a 3Q19 EU launch.

  • Day #3 was a big day for decision support, as Prof. Moshe Phillip shared that the DreaMed Advisor Pro has received an updated indication for adjusting pump settings based on SMBG data alone, expanding on the initial indication for CGM. Interestingly, provider/algorithm consensus on pump settings adjustment is stronger when given SMBG profiles than with CGM traces. Elsewhere, Dr. Boris Kovatchev explained how an in silico digital twin could drive care and described ongoing efforts to bring Dexcom summary data into the UVA EMR. NIH’s Dr. Andrew Bremer made it clear that clinical decision support – without the information overload – is a growing area of interest at the NIH. There were also multiple posters from Glytec (73% reduction in ICU severe hypo!) and mySugr.

  • In an industry triple-header, we heard infusion set updates from Convatec (Unomedical) on the Lantern catheter (soon-to-start Stanford study testing 10-day wear!); Capillary Biomedical on potential for 7+ day wear (much lower inflammation in pig studies); and Pacific Diabetes Technologies single-port CGM/insulin catheter (2Q19 feasibility study). We also saw promising early-stage data out of the University of Missouri on a light-responsive insulin delivery system.

  • Big picture, an important poster co-authored by Dr. George Grunberger surveyed 102 primary care physicians and 100 endocrinologists on their use of diabetes technology in patients with type 2 diabetes. Moving away from diabetes technology entirely, Dr. Simon Heller made a case for screening individuals with diabetes for hypoglycemia-induced cardiac arrhythmias.

Greetings from Rockville, Maryland, where we have the final installment of DTM 2018 highlights. Don’t forget to read our previous days if you missed them:

Day #1 – 670G UX Data from Mayo/ASU; 3M’s thoughts on adhesive for >14 day wear; implantable CGM/pumps roundup; Dr. Jane Seley and Lilly on smart pens; Ascensia Contour app real-world data; the “holy grail” for CGM in the hospital is…

Day #2 – Dr. Roy Beck on time <70, <54 in DCCT; Dr. Courtney Lias on BeyondA1c outcomes; Insulet Horizon update; CGM-based decision support; new AID study data from IDCL; FDA on iCGM

Table of Contents 

Top CGM Highlights

1. Revealed: Dexcom G6 Accuracy vs. iCGM Special Controls (high bar indeed); G6 Enhancements for 14-Day Wear, 1-hr Warmup, New Adhesive, Less Lag; Vagueness on Verily; TypeZero Next-Gen Algorithm Work

Dexcom VP Mr. Peter Simpson shared a number of never-before-seen slides, including G6’s accuracy stacked up against the iCGM special controls and data on next-gen enhancements to G6: 14-day wear and 1-hour warmup (still sub-10% MARD, >90% of points within 20/20), a new adhesive that propels ~90% of sensors lasting 14 days (up from ~75% now), and reduced time lag down to 2.5 minutes vs. YSI (currently ~4 minutes). Vagueness on the fully disposable Verily sensor continues, with just a side profile picture shown this time around without any scale comparator – less detail than we got two years ago at DTM, and it remains unclear if two Verily generations will launch. We also heard about new TypeZero/Dexcom work on a next-gen closed-loop algorithm, which will enable simplified/no meal announcement, a more aggressive and personalized algorithm to optimize time-in-range, and a simplified system startup. Dexcom and TypeZero are also joining forces on MDI decision support, including a trend-based bolus calculator and more dosing guidance. Many slides and details are enclosed below!  

  • For the first time since G6 launched, we got to see the sensor’s accuracy stacked up against the iCGM special controls – 95% lower confidence bound finally included for all glucose bins. The G6 label does not actually share the confidence intervals around the point estimates, so this was a long-needed slide. The takeaway remains the same, as Mr. Simpson noted: “FDA has set a very high bar for accuracy.” However, it’s now even clearer just how high that bar is, since G6 barely crosses the thresholds in some areas. (One could more easily argue now that these controls were designed around G6, and like any line drawn in the sand, there is some arbitrariness here – e.g., why “87%” and not “86%” for within 20%?) This slide also reminds us that using the lower bound of the 95% confidence interval raises the bar even higher, since it requires: (i) a point estimate that exceeds the threshold by a meaningful margin; or (ii) a study large enough that the confidence interval is closely aligned to the point accuracy. All in all, it’s clear that Abbott and Medtronic are not likely to obtain iCGM accuracy standards unless they can make leaps forward in sensor accuracy.  Senseonics may meet the bar with Eversense, though it is prioritizing 180-day wear first.

  • Mr. Simpson also shared brand-new G6 platform enhancement data, starting with an n=78 trial of 1-hour startup and 14-day wear. The data look quite good, with MARD still under 10% – including between hour 1-2 – and 91.9%-93.7% of points within 20% of 20 mg/dl. Only data from hours 1-2, day 10, and day 14 were presented, so we’re not sure what the aggregate performance looks like. Adam asked in Q&A about whether this data meets the iCGM standard, and Mr. Simpson clarified that the study was not powered to adequately answer that question; indeed, far more patients/paired points would be needed to get tight enough confidence intervals, especially with the above in mind. Dexcom’s 3Q18 call had not confirmed this trial’s completion, which clearly occurred well within the original plans for “2H18.” CEO Kevin Sayer said Dexcom will first get its new adhesive (see below) ready before moving to 14-day wear, so this enhancement sounds like a mid/late-2019 launch at the earliest – assuming a larger trial confirms 14-day/1-hour G6 still meets the iCGM accuracy benchmarks. For context, G6 currently has an overall MARD of 9.0% and overall 94% of points within 20/20, with a day 1 MARD of 9.3% and 91.1% within 20/20; the below looks right on par.

  • In our view, this is among the most important enhancements to G6, since it would match FreeStyle Libre’s updated wear/warmup time. (Libre 14-day launched in late October in the US following FDA approval in July.) Plus, a 14-day wear G6 would give Dexcom more pricing cushion and better margins, reducing sensors from three down to two per month.

  • Mr. Simpson also showed new data (n=24) from a “high reliability patch” for G6, which prompts an ~20% relative increase in 14-day sensor survival vs. the current adhesive – per the graph below, ~75% of current G6 patches survive to 14 days, which increases to ~90% with the next-gen adhesive. No further details or timing were shared, though this seems like an excellent and necessary upgrade. On the 3Q18 call, CEO Kevin Sayer shared that this new G6 adhesive is coming out in the “not-too-distant future.” We’d guess sometime in 2019 at this stage, though it was not specified. The size of the patch skirt looks unchanged, so presumably this changes the stickiness or materials. We’re not sure if this patch will reduce skin reactions, which can prevent some from using CGM. On Thursday at DTM, 3M noted the hard tradeoffs that come with adhesive, since improving on one dimension (stickiness) often comes at the expense of others (irritation).

  • Dexcom is also working on further reducing G6’s sensor time lag vs. YSI, which is ~4 minutes now and ~2.5 minutes with the next-gen version. The data was marked as unpublished and For context, the G4 platinum had a 6.5 minute lag, G4 with software 505 brought it down to 5.7 minutes, and G5 brought it down to 5 minutes. In Q&A, a comment noted that even if the current G6 lag was zero, it could still be up to five minutes (maximum) since readings only display in five-minute intervals; reducing updates to every minute would help the user experience on the lag front. Of course, that would also demand more battery life to relay Bluetooth reading, so the tradeoff is not quite that simple!

  • Regarding the next-gen disposable CGM with Verily, Dexcom is “not quite ready to show the final design,” but Mr. Simpson did show a profile of the very thin wearable. It looks quite slim, as expected, but there was no size comparator, making it hard to know just how thin it is (e.g., is it the originally planned penny-sized thickness?). This was the only Verily device shown, and it was unclear if this was the first- or second-gen version; we assume the latter, which has the low-cost advantage, the next-gen sensor, and is still under development. As of Dexcom’s 3Q18 call, the company was “considering options” for the gen one Verily device (using current G6), and the smaller, low-cost, second-gen device was slated for “late 2020/early 2021.” Dexcom’s upcoming investor day on December 4 should hopefully give more specifics on the Verily plans.

  • Citing the acquisition of TypeZero, Mr. Simpson noted the team is already working on a next-gen closed loop algorithm with “full automation” (no meal announcement!), simplified startup and meal announcements, optimization to increase time-in-range, and personalized delivery based on a patient’s data signature. Seeing both “meal announcement” and “no meal announcement” implied to us that the system will integrate manual meal announcement when a user does so, though it will be able to cope without it – perhaps via more aggressive automated correction boluses. The first-gen version of this algorithm is in its pivotal trial within Tandem’s Control-IQ system, slated for a summer 2019 launch.

  • TypeZero/Dexcom continue to work on better tools for MDI users, including a trend-adjusted bolus calculator, actionable insights, smart pen integration, and both real-time and retrospective therapy guidance. No timing was shared in this talk. The 3Q18 call said pilots of this technology are ongoing, with a broader commercial rollout “certainly by 2020.” Dr. Marc Breton’s talk on Day #2 gave a great overview of the technology, which is leaps better than the guidance MDI users have now.

2. 14-Day FreeStyle Libre Accuracy Results: 10.3 mg/dl MAD for Glucose ≤60 mg/dl, Day 1 MARD vs. YSI: 10.8%

Abbott’s Dr. Marc Taub presented data from the 14-day FreeStyle Libre accuracy study (n=95). The four-center study in 95 adults with diabetes (n=80 type 1s; n=15 type 2s on insulin) found 90.7% of sensor results within ±20 mg/dl / 20% of the reference value. Specifically, for glucose <80 mg/dl, a strong 89.9% of sensor readings were within ±20 mg/dl, and for glucose ≥80 mg/dl, 90.7% of sensor readings were within ±20% of the reference value. 89.5% of sensor results were in Zone A of the Consensus Error Grid, and 100% of sensor results were within Zones A and B. The 14-day version is actually a bit more accurate than the 10-day version, with an overall MARD vs. YSI of 9.4% vs. the 9.7% in the 10-day studies. The similar version in Europe had a higher MARD vs. BGM of 11.4% in the pivotal CE Mark trial. In this study, while accuracy remained relatively stable and consistent over the 14-day period, MARD was slightly higher on day 1 (10.8%) before dropping sharply to 8.5% on day 6. We were interested to see accuracy metrics by glucose level, as accuracy in hypoglycemia has been an area of concern for FreeStyle Libre. Mean Absolute Difference (MAD) was a strong 10.3 mg/dl for glucose ≤60 mg/dl (n=84) and 10.0 mg/dl for glucose 61-80 mg/dl (n=354). As Dr. Nicholas Argento pointed out during Q&A, participants wore two FreeStyle Libre sensors in the study, and Abbott used the longest lasting sensor to analyze the data, whereas the FDA used the primary sensor (the first applied sensor for each patient). The interpretation distinction led to a discrepancy between the MARD reported in the FDA’s SSED (10.1%) and Abbott’s press release (9.4%). In response, Dr. Taub claimed that the accuracy study was not designed to determine duration of wear and that Abbott’s analysis was pre-specified. Last we heard from Abbott in August, the company is “currently working” with the FDA to address the identified discrepancies.

  • Dr. Taub also described the process by which FreeStyle Libre sensors undergo individual calibration adjustment. He explained that over 80 parameters of manufacturing are automatically inspected and used along with their correlation to the sensor sensitivity to calculate individual adjustments for each sensor in a given lot. For example, in vivo measurements have shown the sensor signal to be inversely correlated to membrane thickness and directly correlated to the enzyme layer area. While these adjustments are fairly minimal, Dr. Taub noted that they “very subtly boost performance.” It is remarkable that Abbott does individual sensor calibration in the factory; to support 1 million patients globally, the company is likely making over 20 million sensors each year.

3. Medtronic Leveraging EIS to “Self-Calibrate” CGMs, Reducing Calibration to 2-3x/Week While Reducing MARD by ~0.5%

Medtronic’s Dr. Andrea Varsavsky detailed efforts to leverage a glucose-independent diagnostic already being used in Guardian Sensor 3 – electrochemical impedance spectroscopy (EIS) – to “self-calibrate” the system via an algorithm, potentially reducing manual calibration burden. EIS is currently used in Medtronic Guardian Sensor 3 to monitor the current status of the sensor and the tissue in which it is embedded; it’s also already used to eliminate outliers by detecting sharp changes in sensor accuracy, and it therefore serves to improve accuracy of the system. But, Dr. Varsavsky said, there’s more EIS potential to be reaped in the form of real-time self-calibration. In an in vitro demonstration, Medtronic researchers modulated manufacturing parameters to generate an array of sensors with various glucose sensitivity levels; the team saw a strong correlation between impedance and sensor sensitivity, leaving potential for sensor algorithms to self-calibrate a CGM (i.e., correct for baseline sensitivity discrepancies in real-time without needing factory calibration). The correlation exists, albeit in slightly messier form, in vivo, as demonstrated by a proof-of-concept where researchers compared no-calibration models of in vivo Guardian 3 data with and without EIS signals. Models with EIS consistently outperformed models without EIS by 2%-5%-points on MARD, establishing that EIS is “a useful signal to help with self-calibration.” In another demonstration, Medtronic trained an algorithm on a massive data set comprised of 1,799 subjects, 5,505 sensors, and 131,138 paired CGM-reference points. Applying this algorithm to a separate set of ~600 sensors led to an ~80% reduction in calibrations (from 2-3/day -> 2-3 calibrations/week) along with a 0.5%-point lower MARD relative to Guardian Sensor 3. One huge advantage to this work is that it doesn’t entail any changes to sensor design or manufacturing, just leveraging existing technology in Guardian Sensor 3. Dr. Varsavsky implied that future sensors will be designed specifically to leverage EIS in a bigger way, potentially pushing calibration and accuracy improvements even further. Absent from the talk were any mentions of commercial plans – updates to next-gen Guardian Connect or Guardian Sensor 3? EIS-based self-calibration baked into the next-gen Project Harmony sensor (see ADA poster) or the Project Unity sensor? Regardless, this seems to be a very cool development for Medtronic, who seems significantly behind Abbott and Dexcom on factory calibration.

4. Waveform Cascade 1 CGM Slated for EU Launch in 3Q19, US Pivotal in September 2019; Gives Reading Every Minute, Rechargeable Transmitter and Reusable Inserter (Needle-Free!) Both With 2-Year Life; Mean MARD 12.8% So Far, Further Studies Underway

WaveForm’s (AgaMatrix) Dr. Misha Rebec provided some new details and data on the Cascade 1 CGM, which is now slated for a potential EU launch in 3Q19 and an FDA filing in December 2019. The company had previously hoped for a CE Mark in 2018. The sensor has a decent feature set, though is not approaching the accuracy of its factory calibrated competitors. We’ve heard rumors of a potential lower-cost profile, though moving lower that FreeStyle Libre seems hard to believe. WaveForm has pushed back its timelines meaningfully over the last year – will it meet its expectations moving forward? Can it reasonable compete in Europe with a one-cal/day sensor when two factory-calibrated sensors are already available (FreeStyle Libre and G6)?

  • Cascade CGM system details (see pictures of the sensor, inserter, and app): We had previously heard that the sensor would be 14-day wear, average one calibration per day (it will launch with one per day in Europe), no receiver (BLE smartphone communication), limited interferences (e.g., no acetaminophen issues), one-hour warmup time or less (45 minutes in ongoing studies), and “painless” insertion. Dr. Rebec added that the sensor gives a reading every minute (vs. every five minutes with Dexcom and Medtronic) and the transmitter is rechargeable and reusable for up to two years. The inserter itself is also reusable for up to two years, and remarkably, there is no needle involved in the insertion – the 280 um-diameter filament itself pierces the skin in a “virtually painless” process. The sensor is first loaded into the inserter, three buttons are pressed, then the inserter is lifted, leaving the inserted sensor behind. Dr. Rebec also showed a number of pictures of the app.

  • Timing Updates: WaveForm anticipates a “potential” CE Mark and EU product launch in 3Q19. A US pivotal study (100 participants at three sites) is expected from September to November 2019 ahead of a December 2019 FDA filing. In the interim, the company expects to: complete EU safety and effectiveness study in January; submit technical clinical files in EU in 2Q19; submit an IDE in the US for “15-day study” in March; conduct user interface studies in the US/UK to augment EU performance data in February/March; and submit pivotal study protocol to FDA in June.

  • Study Progress and Accuracy Data: WaveForm conducted seven studies in 2017-2018, including six over the last 18 months. Mean MARD across those studies (n=87 total) was 12.8% (median: 9.3%). On the Consensus Error Grid, 85.2% of readings fell within in Zone A and 13.6% fell within Zone B (98.8% together). There were 87 participants in total, and 95% of sensors made it the full study period – only two of the studies investigated 14-day wear, where sensor survival was an impressive 96%. Dr. Rebec also referred to recently completed studies: one comparing accuracy to Dexcom G5 (11.0% MARD for Cascade vs. 12.2% for G5; 10-day wear), one comparing accuracy to FreeStyle Libre (11.9% MARD for Cascade vs. 14.5% for Abbott; 14-day wear), and an EU pre-pivotal study with 20 subjects and 14-day use (12.6% MARD). The company is now in the midst of an n=60, three-site safety and effectiveness study, which will comprise part of the CE mark submission package. Each subject wears two Cascade sensors for 14 days and has five in-clinic days during which YSI is used to calibrate the sensor ­– obviously this does not resemble real-world use and we find it hard to believe that regulators would accept a protocol in which the comparator method is also used for calibration. (In Q&A, Adam asked about this, and Dr. Rebec said that the protocol is “in agreement with what our discussion with FDA has been on this topic.” We find that very hard to believe.) The first site for this study in Slovenia (n=20) completed in mid-October. At this site, 85% of sensors survived for the full 14 days and mean sensor MARD was 11.2%: 11.7% between 80-180 mg/dl, 9.3% between 180-240 mg/dl, and 11.9% in >240 mg/dl (<80 mg/dl wasn’t shown). MARD was 10.8% on day 1, 11.5% on day 4, 9.3% on day 7, 12.0% on day 10, and drifted up to 13.3% by day 14.

Top Decision Support Highlights

1. DreaMed Advisor Pro Indication Updated to Support Pump Setting Adjustments Based on SMBG Alone; Helmsley-Funded Pivotal Study Underway and Nearly Fully Enrolled (n=112)

Schneider Children’s Prof. Moshe Phillip shared that the DreaMed Advisor Pro indication has been updated to support insulin pump settings adjustment for individuals on SMBG alone (Advisor Pro “Version B”), complementing the initially-FDA cleared CGM version (“Version A”). We’re not sure whether Version B is yet available to clinics who are already using Advisor Pro through Glooko, nor if it has received a CE mark. Prof. Phillip presented unpublished data supporting the SMBG indication, the SMBG-EXPERT study, which is analogous in nature to the CGM-EXPERT study (formally published in June). Whereas the CGM-EXPERT study presented expert diabetes clinicians with CGM traces and assessed level of agreement on pump settings adjustment between providers vs. the DreaMed Advisor, the SMBG-EXPERT study did the same with 15 patients’ SMBG profiles (average of 6 fingersticks/day). Interestingly, there was greater inter-HCP and HCP-Advisor agreement when presented with SMBG, suggesting that access to more data (with CGM) leads to a wider variety of treatment approaches that the algorithm can more finely assess. In SMBG-EXPERT, physicians agreed with each other on the direction of basal rate adjustment 50% of the time and with the Advisor 51% of the time; providers agreed with each other on direction of carb ratio adjustment 54% of the time and with the Advisor 55% of the time; providers agreed with each other on direction of correction factor adjustment 50% of the time and with the advisor 47% of the time. In CGM-EXPERT, these levels of directional agreement were all in the 40s%. (It is striking how inconsistently providers agree with each other!) Prof. Phillip et al. performed a retrospective analysis (n=434 cases) to compare the level of agreement between CGM-based recommendations and SMBG-based recommendations when both data streams were available. There was a linear relationship between average number of SMBG per day and agreement in direction of insulin dose adjustment: At 3 SMBG/day, agreement between Advisor Version A and Version B was 55%, while at 9 SMBG/day, agreement rose to ~65%. Prof. Phillip commented that at times, Advisor Version B will decline to offer a recommendation if there is insufficient fingerstick data. We’re still amazed that the probability that two experts’ recommendations for directionality of insulin adjustment alone is essentially the same as a coin-flip; this makes a strong case for a software-based algorithms like Advisor Pro that can take some bias out of the process (not to mention time!), add highly objective data analysis, and ideally learn over time to optimize recommendations.

  • Prof. Phillip brilliantly laid out the potentially disparate goals of providers and patients in a clinic visit (table below) and quantified time constraints facing diabetologists. A 2017 paper found that it took providers 18 minutes to analyze pump data alone, 13 minutes for MDI data, 15 minutes for bolus calculators, and 11 minutes for meter data. This is obviously impossible to squeeze into even a 30-minute appointment, let alone focusing on the aspects of diabetes that truly matter to people with diabetes! Prof. Phillip also calculated that if his practice made weekly phone calls to titrate insulin, it would require 11 extra providers (15 minutes per phone call x 1,700 patients -> 425 hours/week -> 53 work days/week -> 11 extra providers)! To put a face on the statistics, two nights prior to his talk, a diabetologist told Prof. Phillip that he takes patient charts home every night because he can’t finish the work in practice. He told him that “visits are more complicated now that we have data from CGM, pumps, meters, food diabetes…we have to cope with all that data, and also listen to the patient, refer them to an expert for other conditions, deal with their other diseases and medications…” For anyone who’s read Dr. Atul Gawande’s recent long-form piece in The New Yorker (“Why Doctors Hate Their Computers”), the comments from Prof. Phillip’s friend should ring a bell.

Patient expectations from the clinic visit

Clinician expectations from the clinic visit

  • “Get rid of my diabetes”

  • Show interest, sympathy, understanding

  • Praise me on achievements

  • Be tolerable to my under-achievements

  • Remember what I told you during previous meetings

  • Adjust perfectly my blood glucose levels

  • Make sure we prevent complications

  • Take “them” off my back

  • Cure my diabetes

  • Perfect A1c

  • No episodes of DKA or severe hypoglycemia

  • Negative results on routine periodical tests

  • No evidence of complications

  • Great spirit

  • No social issues

  • No special needs

  • Reasonable time in the office

  • Prof. Phillip noted that the Helmsley-funded multi-site pivotal study of Advisor Pro is underway and “3-4 patients short” of full enrollment (n=112). The update came in Q&A in response to an endocrinologist’s question about who is “correct” – expert HCPs or the Advisor? As of June, DreaMed hoped to report results from the first three months of the trial by the end of 2018 or into 1Q19 – the latter seems more likely at this point, perhaps at ATTD.

  • As at a meeting in August, Prof. Phillip said that “once we show it to clinicians and they use it and like it and trust it, there’s not a reason to not give it directly to patients.” He has never publicly attached a timeframe to a patient-facing, non-provider-gated pump settings adjustment app. At AACE 2017, Dr. Phillip said that Advisor Pro support for MDI is also in development. We haven’t heard updates on the MDI front since, though DreaMed also announced a collaboration with Schneider Children’s Medical Center and Harvard School of Engineering and Applied Sciences to develop products to optimize insulin delivery for type 1 patients on MDI and fingersticks.

2. Dr. Kovatchev Explains Vision for How “Digital Twin” Can Guide Care in Future; UVA/Dexcom Work on Integrating into EMR

UVA’s Dr. Boris Kovatchev explained his vision for how VIP (Virtual Image of the Patient) therapeutics – generating a “digital twin” of each patient – will drive diabetes care in the future. The basic idea is to use available signals to map each individual to one or more virtual subjects in an in silico (virtual) population, run algorithms to see how treatment parameters can be optimized for those simulated subjects, and then apply them back to the real-life patient. For example, the system will intake medical record and CGM data; Dr. Kovatchev provided the example of a 62-year-old male with type 1 diabetes and an 11.7% A1c who, over a six-hour period, experienced prolonged hypoglycemia pre-meal, a large glucose excursion following the meal, and hypoglycemia again following the meal. The system knows that the individual has a basal rate of 2.0 unit/hour and ate 62 grams of carbs for the meal, bolusing six units to cover it. A digital twin, whose glucose trace was nearly identical to the individual’s when given the same amounts and timing of insulin and carbohydrate, was identified within the in silico population. After running simulations seeking to minimize hypoglycemia and hyperglycemia, the algorithm suggested that the individual, if faced with the same scenario, reduce basal rate by 10% and reduce premeal bolus by 2 units (and ideally give it earlier). Had the individual followed this strategy in the first place, the simulation suggests his mean blood glucose would have been 105 mg/dl instead of 79 mg/dl, hypoglycemia would’ve been avoided, and he’d have had a mean amplitude of glucose excursion (MAGE) of 111 mg/dl vs. 121 mg/dl. It’s worth pointing out that the VIP approach as described is only descriptive of what happened and not yet prescriptive, but if the digital twin can be faithfully identified or engineered in silico, it’s not difficult to imagine how Dr. Kovatchev and team would be able to give tailored advice for therapy adjustment, especially insulin dosing around meals and exercise.

  • Over the past year, UVA’s Diabetes Technology Clinical Program and Dexcom’s Clarity team have implemented a program that allows patients to consent to summary CGM reports getting pulled right into their providers’ EMR; the initiative is “in the final stages of testing.” The patient can consent in the very first meeting with the provider, at which point a report is generated and presented to the doctor. At follow-up appointments, since the patient has already consented, the provider can simply pull up a summary CGM report. On a similar note at ATTD, Dexcom’s Dr. Nate Heintzman showed a screenshot of an EHR-integrated Clarity experience, which was already live at Children’s Hospital LA. As we understand it, Dexcom’s latest partnership with Validic could accelerate and begin to democratize provider access to CGM data within current workflows.

  • Remember that first snowboard jump on G6 and Tandem’s Control-IQ that Dr. Kovatchev displayed at ATTD? It turns out that 12-year-old boy was also equipped with an activity tracker, and his glucose and activity data were sent directly to the cloud and relayed to the Diabetes Technology Center so a report could be generated in PDF form. The ski study in February was a successful pilot test of the technology and the team is “moving on with that.” 

3. Glytec’s Glucommander Reduces Adjusted Length of Stay by 18%, Cuts Severe Hypo by 73% in Critical Care Unit; Safe and Effective for IV Insulin in Labor and Delivery

Glytec presented two retrospective posters at DTM demonstrating strong outcomes in the hospital setting for its insulin titration software.

  • In a retrospective study (n=382) from Riverside Medical Center, Glytec’s Glucommander software reduced length of stay, blood glucose, hypoglycemia, and bounce-backs to critical care unit after transition to the general ward. 174 of the patients were dosed insulin off of Glucommander recommendations, while 208 were dosed through standard (paper) protocols. Albeit retrospective, and the fact that the Glucommander group had significantly lower starting A1c (8.1% vs. 8.7%) and higher starting blood glucose (300 mg/dl vs. 264 mg/dl), the clinical outcomes were encouraging. With Glucommander, there was 73% less incidence of severe hypoglycemia (0.11% vs. 0.41%), 39% less mild/moderate hypoglycemia (2.34% vs. 3.85%), 20.7 mg/dl lower mean glucose (166.6 mg/dl vs. 187.3 mg/dl), and 31.2 mg/dl lower final critical care unit blood glucose (154.7 mg/dl vs. 186 mg/dl). Length of stay index (actual length of stay vs. expected length of stay) was 1.12 days with Glucommander vs. 1.37 days with standard protocols, suggesting an 18% reduction in length of critical care unit stay (adjusted for expected stay length). Further, there were no cases of bounce-back to the critical care unit following transition to general ward in the Glucommander group, while 1-2 patients had to revert per month in the non-Glucommander arm. Cost-effectiveness data were not presented, but we have to imagine they would be favorable from the reductions in length of stay and severe hypoglycemia alone. Our back-of-the-envelope calculations: Assuming the average length of stay in the critical care unit is three days at a price of $20,000 per day, then the hospital could save $5,000 per patient by using Glucommander. We know from Adam’s recent hospital stay (along with many reader stories) that diabetes management is miles away from optimal in the hospital setting – software like Glucommander could make a huge difference. We wonder how the outcomes from Glytec would compare to those from closed loop; the Cambridge group has published a couple of high profile studies of fully closed loop in non-critical care settings (ADA 2018; ADA 2016).

  • A second poster showed that Glucommander was safe and effective in 204 young women during labor and delivery in a Honolulu facility from November 2015-June 2018. Median time on the system was seven hours and time-to-target was six hours, with average initial blood glucose of 152 mg/dl dropping to 113 mg/dl by the end of the period. Mean first day blood glucose was 128 mg/dl vs. mean last day blood glucose of 117 mg/dl. There was no comparator arm here, but the poster showed that 90+% of readings across subjects were in 70-180 mg/dl; 7.2% were >180 mg/dl, and 0.53% were <54 mg/dl.

4. mySugr Excellent Real-World Data from US mySugr Bundle Users: -16 mg/dl mean glucose, TIR +2 hrs/day; AGP-Style Data Visualization for SMBG

We noticed two mySugr posters of interest, one on using an evidence-based continuous probability estimation to generate an AGP-style data visualization from SMBG measurements, and the other showing real-world data from US mySugr Bundle users. See below for the details.

  • To determine the real-world changes in glycemic control following use of the mySugr Bundle (mySugr app + unlimited strips + CDE coaching), data from 52 US participants (56% type 1, 37% type 2, 6% LADA) were evaluated over four months. Mean blood glucose dropped significantly by 16 mg/dl (baseline: 154 mg/dl) and time-in-range improved significantly from 64.5% to 73.0% (+2 hours/day). Significant declines in time-in-hyperglycemia (-8.9%) and eA1c (-0.43%; baseline: 6.7%) were also observed, and SMBG checking frequency increased by 17.5% from 5.8 fingersticks/day to 6.2 fingersticks/day. A clinically relevant change in eA1c (defined by EMA guidelines as ≥0.3%) was achieved in 31% of the population. A subgroup analysis found that for those with baseline eA1c ≥6.7% (n=26), mean blood glucose significantly declined by 40 mg/dl (baseline: 186 mg/dl) and glycemic standard deviation decreased by 14 mg/dl (baseline: 71 mg/dl). For the subgroup with baseline eA1c <6.7%, no significant changes were observed, suggesting that the positive effects of the mySugr Bundle are enhanced in those who are less well-controlled. Positive trend data taken two months after bundle registration but before Bundle initiation indicate that the initiation of CDE coaching may positively impact the sustainability of glycemic improvements, as well as contribute to longer-term benefits. Importantly, the participant population was broad: 77% were using insulin, only 19% were using insulin pumps, and 23% were on non-insulin therapies. The results are very encouraging, and we’d be interested in seeing outcomes across a larger population.

  • In an attempt to develop an AGP-style glycemic data visualization tool for SMBG patients, the mySugr team applied a kernel density estimation (KDE) to obtain a continuous probability for glycemic events from SMBG measurements. A kernel essentially determines the probability of a glycemic event (i.e., hypoglycemia, hyperglycemia, in-range) being present before and after the time of measurement. By summing all kernels, the likelihood of recorded events can be shown over a day. The poster notes a limitation for KDE in that it uses non-periodic data, resulting in a non-continuity of the estimated probabilities. By switching to a Van-Mises Kernel, the probabilities can be translated into a continuous transition. To evaluate the accuracy of the model, 14-day CGM datasets were randomly subsampled to generate artificial SMBG datasets. The CGM and SMBG datasets were plugged into the model and compared for fit, revealing that increasing SMBG measurements from once daily to three/day provided the greatest error reduction, while a plateau was reached at ~4 fingersticks/day. The investigators therefore chose four measurements/day over a duration of 14 days as the minimum requirement for the model. Given the immense utility of AGP both for patients and providers, we’re excited to see efforts aimed at expanding its benefits to SMBG users. It also seems to be a first step at eventual glucose prediction for SMBG users. One Drop has a version of this for type 2 non-insulin users, though obviously doing so in those with more glucose fluctuation is exceedingly more difficult.

5. NIH’s Dr. Bremer Touts NIH’s Interest in Visual Analytics, Machine Learning to Prevent Info Overload in Clinical Decision Support; Emphasizes Applicability to Disparate Types of Diabetes

While he purposely didn’t include any RFAs in his talk, NIH’s Dr. Andrew Bremer spoke expectantly about the possibilities that arise when data science, technology, informatics, and phenotyping are incorporated into clinical decision support. He noted that the quantity of data continues to grow, along with (our understanding of) the heterogeneity in diabetes. Accordingly, he applauded efforts from the likes of Dr. Kovatchev and Prof. Phillip, but asked how “we can increase applicability in the field, to help the child with type 2, or the woman with gestational diabetes.” One area that NIH has invested in is visual analytics, consolidating analytical information in interactive, easy-to-understand visual interfaces in order to prevent information overload for providers using clinical decision support tools. He gave the example of Glucolyzer, “an interactive tool featuring hierarchical clustering and a heatmap visualization to help registered dietitians identify associative patterns between blood glucose levels and per-meal macronutrient composition for individuals with type 2 diabetes.” By hovering over a bar on the screen, users see a prediction of their likely glucose spike if they were to eat a given meal. (The idea is cool though the user interface looks very confusing and overwhelming to us.) Dr. Bremer also said that other advances in data science, such as mechanistic machine learning, are also promising. A fun role for him at the NIH, he concluded, is to play matchmaker – to recognize individuals pushing technology, maybe not in diabetes at the outset, but recognizing where technology is going and how it can be applicable in the field of diabetes.”

Top Insulin Delivery Highlights

1. Unomedical Extended 10-Day Wear Infusion Set Study Starting Soon at Stanford, Large Multicenter Study in 2019; Set+Sensor Designs In Development

Convatec’s (Unomedical) Dr. Matthias Heschel discussed the company’s infusion set pipeline, sharing that an upcoming Stanford study will test the next-gen “coated” Lantern catheter for extended wear over 10 days. Following mixed seven-day data shared in an ADA poster, Unomedical is ambitiously extending wear further in an upcoming outpatient study at Stanford (not yet listed on ClinicalTrials.gov). As a reminder, the Lantern catheter includes several slits along the side to allow insulin to flow out of multiple places (e.g., for occlusion or kinking), and this next-gen version has a “coating” to suppress the body’s foreign body response over extended wear. The ADA data on coated Lantern in n=16 type 1s showed a trend towards reduced insulin action and more hyperglycemia over time, though “safe” wear out to seven days. We’ll be interested to see if the hyperglycemia/insulin trends are seen in this upcoming outpatient study, particularly with ambitious ten-day wear. Dr. Heschel said a “large multicenter study” of coated Lantern will start in 2019, presumably using lessons from this smaller Stanford study. Notably, the set is expected to remain class II (510(k)), even with the anti-inflammatory coating. Dr. Heschel also briefly covered the gen one version of Lantern (non-coated) as an example of improved catheter reliability. The approach indeed seems like a better version of BD’s FlowSmart, though it’s still unclear when Lantern gen one will launch; perhaps we’ll hear an update at ATTD. Dr. Heschel’s first pipeline slide reviewed the company’s all-in-one novel serter – marketed as Medtronic’s Mio Advance – which is available outside the US and received under-the-radar FDA clearance in March. No US launch timing has been shared on this excellent product, though as of ATTD, “other countries” were expected to launch “later in 2018.” Medtronic told us at EASD that it is seeing major demand for the set outside the US, and the lag time between clearance and US launch reflects the need to build sufficient supply. Will we see a launch this year?

  • Unomedical also has set+sensor and dual-hormone delivery concepts in development. The former includes double-port designs with a two-needle insertion (i.e., like a snake bite) and also ambitious single-port configurations that integrate the CGM sensor within the insulin catheter (i.e., one needle). Unomedical plans to exhibit “selected concepts” at upcoming events. On the dual-hormone front (second picture below), we believe Beta Bionics is building its own dual-hormone set; however, this suggests potential to use a Unomedical version.



  • For context, Medtronic’s 2018 Analyst Meeting expected a “combination set with extended wear” to launch “Beyond” April 2020 (2+ years away). We assume this is in partnership with Unomedical – called “Project Duo” back in June – though it was not specified at the time.

2. Capillary Biomedical Soon to Start Human Study on Extended Wear SteadiSet with Sprinkler-Like, No-Kink Catheter (3 Side Holes)

TJU’s Dr. Jeffrey Joseph shared an update on Capillary Biomedical’s plans for an extended wear, no-kink (wire reinforced), multi-ported (3 side holes) insulin catheter that provides “consistent absorption of insulin from day to day and dose to dose.” The company has NIH funding to soon begin a first-in-human trial of SteadiSet, gathering reliability, safety, and PK/PD data for extended set wear out to 7 days and potentially beyond. Our coverage of Capillary in August called for a “late 2019” launch of a three-day wear version of SteadiSet, though Dr. Joseph focused on the next-gen extended wear potential in this presentation (7-14 days). He presented data from Capillary’s cool DTM poster (download it here), which compared its set to a standard teflon catheter (Unomedical) over 14 days of wear (!) in pigs. The tissue histology data showed significantly less inflammation with Capillary’s set, both in terms of thickness of the inflammatory layer and total surface area of inflammation (graphs enclosed below). The company believes its soft, flexible, SteadiFlow cannula produces less tissue inflammation due to: (i) less motion-induced tissue trauma (the catheter bends and flexes); and (ii) greater reliability of insulin delivery due to the wire-reinforced cannula that resists kinking and multiple holes (four in total) that produce redundancy. We like how data driven this company is and look forward to seeing if the promising extended wear data in pigs is confirmed in humans.

  • See our August coverage for a deep dive on Capillary Biomedical and the SteadiSet. At the time, the company had raised $2.9 million in seed funding and planned to launch gen one (three-day wear) in “late 2019” through pump partners. The three-day wear version is expected first (easier path to market), followed by an extended wear version that will obviously require more clinical testing.

3. Pacific Diabetes’ Integrated CGM/Insulin Catheter: MARD of ~12-14% in Small OHSU Study

Pacific Diabetes’ Dr. Bob Cargill shared new accuracy data on the company’s single-port, integrated insulin/CGM sensing catheter. In a small, 10-hour human feasibility study at OHSU (n=8 completers), the sensor demonstrated an MARD of ~12%-14% using a “retrospective” two-point calibration, with 70% of points in Zone A and 26% in Zone B. The company has reportedly seen lower MARDs “down around” 10% in pig studies, which have tested the sensor/catheter out to seven days. Since DTM 2017, Pacific has improved the sensor chemistry/stability and transitioned to a flexible cannula with a 60% smaller cross section. The on-body form factor has improved to a sleeker wearable (see below), targeted to enter a feasibility study in 2Q19; an FDA IDE submission is expected soon. The company has funding from HCT, NIDDK, and JDRF, and is looking for industry partnerships, academic collaborations, and investment. The goal here is ambitious – sensing and infusion in one site with one needle – and we the commercial viability rests on extending insulin infusion length. We cannot imagine moving back to a three-day CGM sensor, which not only puts tremendous pressure on price/margins (10 sensors/month instead of 2-3), but puts a full one-third of the wear-time on day one of insertion. Can Pacific get infusion set length out to 7-14 days? We did not hear anything about that in this talk.

  • As we noted at DTM 2017, one of the first challenges Pacific faced was designing a sensor unbiased by the presence of insulin preservatives at the same site. When exposed to insulin, conventional, platinum-based sensors exhibit a rapid spike resembling hyperglycemia followed by eventual poisoning and loss of sensitivity. By turning to redox-mediated chemistry, Pacific Diabetes developed a sensor capable of functioning normally, showing no significant glucose spike when insulin is infused at the sensing site. Pacific uses a sensor array and different chemistry, which mostly eliminates the artifact with only a brief artificial spike (lasting ~5-15 minutes) following a large bolus.

4. Light-Induced Insulin Delivery Controlled by CGM; Still Early Stage but Shows Substantial Promise in Rats; Potential for Bihormonal System

University of Missouri’s Dr. Simon Friedman described his team’s fascinating efforts to develop a light-induced insulin delivery system. An insoluble polymer, which when reacted with a specific wavelength of light releases pure insulin, is injected below the skin and stimulated non-invasively through a LED light source controlled by CGM. Dr. Friedman was quick to address the common concern regarding the potential for ambient light to affect the system; according to Dr. Friedman, the light source is 50-100-fold more intense than ambient light, which is already adequately restricted by covering the injection site with the light source. Early results published in 2013 demonstrated that pulses of insulin are released in a predictable fashion upon photolysis of the polymer. In rats, insulin was shown to be “immediately” released after two minutes of radiation, with a very fast peak and decline – in fact, Dr. Friedman highlighted that 30%-40% of the peak insulin level is reached within five minutes. The rats exhibited a glucose reduction in response to the light irradiation, but Dr. Friedman characterized the reduction as “not robust enough.” His team achieved a “more substantial” reduction in glucose by applying a second pulse of light. While Dr. Friedman described this first-generation material as “fine,” his team recently developed a second-generation material that provides five-fold peak insulin while using one-fourth of the light relative to the first-generation version, yielding a 20-fold improvement on a per light basis. Moreover, by attaching a hydrophobic tag to the material, the material is rendered soluble in a mildly acidic injectable formulation, but precipitates in a neutral pH environment (i.e., the skin) – an important consideration for the commercial and clinical feasibility of the therapy. Upon photolysis, native insulin is released, leaving behind a much smaller, more easily absorbed molecule to clear the body than the original polymer. Next steps, according to Dr. Friedman, include investigating photolysis via a longer light wavelength, performing multiple-day trials, and developing a similar system for glucagon delivery. Excitingly, Dr. Friedman’s team has already established light-responsive glucagon release, and Dr. Friedman ultimately sees potential in leveraging the system for bihormonal non-invasive glycemic control. In such a system, insulin and glucagon could be individually controlled by using two distinct wavelengths of light. There was substantial excitement in the room during Dr. Freidman’s talk – we can see why. While obviously in its early stages, this system could radically change insulin delivery for people with diabetes.

Top Big Picture Highlights

1. Dr. George Grunberger’s Poster on PCP and Endocrinologist Perspectives Surrounding Diabetes Technology Use in Type 2 Patients

A BD-funded poster listing Dr. George Grunberger as first author evaluated primary care physician (n=102 PCPs) and endocrinologist (n= 100 ENDOs) perspectives on the use of diabetes technology to optimize intensive insulin therapy in patients with type 2 diabetes. Among other inclusion criteria, PCPs must have personally treated ≥20 patients with type 2 diabetes each month, while ENDOs must have personally treated ≥80 patients with type 2 diabetes each month. At least 25% and 50% of these patients, respectively, must be prescribed insulin therapy. Not surprisingly, the results indicated that ENDOs are more likely to prescribe use of diabetes technology than PCPs. However, both ENDOs and PCPs reported willingness to utilize diabetes technology in patients with type 2 diabetes if A1c targets are not met using basal-bolus therapy. CGM is the most commonly prescribed technology for ENDOs (87%) while traditional tubed insulin pumps are the most commonly prescribed technology for PCPs (74%). ENDOs report substantially more comfort initiating pump therapy than PCPs: nearly all ENDOs were “comfortable” or “very comfortable” with initiating tubed insulin pumps (96%) and wearable tube-free patch devices (93%) whereas more PCPs felt “comfortable” or “very comfortable” initiating a wearable tube-free patch (74%) than a tubed insulin pump (52%). The latter point is interesting, as despite the preference for wearable patch delivery devices, more PCPs report prescribing tubed pumps (74%) far more often than patch devices (23%); definitely an opportunity for Insulet, Valeritas, BD, CeQur, etc. When asked to rank the top four features that deter/prevent participants from using diabetes technology, both groups included device complexity and patient acceptance (ENDOs ranked complexity as #4 while PCPs ranked it as the #1 deterrent). We were somewhat surprised to see cost/insurance coverage listed as the #1 deterrent for ENDOs but not included at all for PCPs. Given that the vast majority of type 2 diabetes patients receive their diabetes care from PCPs, many leaders in the space have advocated for increased efforts to facilitate technology uptake among general practitioners. From the results of the survey, it would appear that this might best be achieved by decreasing device complexity and cutting down on the extra time and training required by the device. Abbott’s FreeStyle Libre has made great strides in checking these boxes – it remains to be seen if insulin delivery devices can get to that point of simplicity.

2. Dr. Simon Heller Argues for Hypoglycemia-Induced Arrhythmia Screening in Diabetes; Demonstrated Cardiac Effects of Hypoglycemia in Type 1s and 2s

University of Sheffield’s Dr. Simon Heller presented compelling evidence in favor of screening patients with diabetes for hypoglycemia-induced arrhythmias. In the largest observational study examining the effects of symptomatic hypoglycemia on cardiac arrhythmias, heart rate variability (HRV), and cardiac repolarization in young adults with type 1 diabetes, 37 patients with type 1 diabetes <50 years-old underwent 96-hours of simultaneous ECG and CGM monitoring while continuing their normal activities and diabetes management. The results, published in 2017 in Diabetes Care, found several differences in cardiac responses to hypoglycemia between the day and night. Bradycardia (abnormally slow heart rate) was significantly more common during nocturnal hypoglycemia vs. nocturnal euglycemia (IRR: 6.44; 95% CI: 6.26-6.63; p<0.001) but occurred significantly less frequently during daytime hypoglycemia vs. daytime euglycemia (IRR: 0.02; 95% CI: 0.002-0.26; p=0.002). The frequency of atrial ectopics (premature heart beat) was also significantly higher during daytime hypoglycemia as compared to daytime euglycemia (IRR: 2.29; 95% CI: 1.19-4.39; p=0.013). The investigators also detected cardio-acceleration during daytime hypoglycemia but not during nocturnal hypoglycemia, and confirmed a proarrhythmogenic of hypoglycemia by showing significant extension of the QTc interval and TpTend interval plus a change toward an abnormal T wave shape during the night and day. After reviewing the results further, Dr. Heller realized that the bradycardia had actually been observed in just one individual. Dr. Heller interpreted this finding as indicative of bradycardia as “probably very specific to a few individuals.” Still, he does not find the occurrence to be an anomaly unworthy of investigation; rather, he remains interested in determining which factors led this individual to be more susceptible to bradycardia. Given the evidence above coupled with a demonstrated increased risk of overnight death in young people with type 1 diabetes as compared to those without diabetes, Dr. Heller concluded that there is a case for screening patients with diabetes for hypoglycemia-induced arrhythmias.

  • We’d be very interested to see a cost analysis study for an arrhythmia screening program, possibly coupled with screening for broader CV risk/mortality. At Diabetes Canada, University of Toronto’s Dr. Lawrence Leiter provided a useful review of the literature on the relationships between glycemic variability, hypoglycemia, and CV risk/mortality. As he pointed out, establishing causation may remain a point of debate, as ethical concerns preclude an RCT where people are randomized to have more severe hypoglycemia or greater glycemia variability. Still, the association alone suggests that the issue should be further addressed.

  • A smaller study (n=23) published in Diabetes in 2017 observed cardiac function during experimental hypoglycemia in patients with type 2 diabetes. Participants with type 2 diabetes demonstrated greater heterogeneity of repolarization during hypoglycemia, as demonstrated by T-wave symmetry, despite comparable epinephrine levels as compared to matched controls without diabetes. The investigators concluded that these mechanisms could feasibly contribute to arrhythmias during clinical hypoglycemia.

 

-- Adam Brown, Brian Levine, Maeve Serino, and Kelly Close