ATTD (Advanced Technologies and Treatments in Diabetes) 2019

February 20-23, 2019; Berlin, Germany; Full Report – Draft

Executive Highlights

  • In automated insulin delivery (AID), ATTD was headlined by: Medtronic’s 670G pivotal data from the 2-6 year-old cohort and bold pipeline plans for the next year; a striking 84% severe hypoglycemia reduction in the six-month SMILE trial testing the MiniMed 640G; real-world data on Tandem’s Basal-IQ; FDA’s Dr. Courtney Lias on potential to create a lower-risk class II iController pathway for AID algorithms; JDRF’s interoperability vision; and Cambridge’s plans for AID commercialization with Dana. We also got to hold Roche’s Accu-Chek Solo pump for the first time in the exhibit hall, and we heard infusion set updates from Convatec (Unomedical) and Capillary Biomedical.

  • In CGM, Abbott shared big data on nearly 500,000 FreeStyle Libre users –  one of the most important abstracts of the year. It gives the largest global pulse on time-in-range yet, and reminds us how far the field must go with decision support, automation, education, and coaching (i.e., just giving someone a CGM is not enough – the data has to be used to drive changes). We also saw concerning baseline blinded CGM figures from the CITY study of CGM in adolescents and young adults (median 35% time-in-range), plus validation of professional CGM (Medtronic’s iPro 2) in the ADJUST study. And in Dexcom symposiums, we got a look at CGM data flowing directly into UVA’s Epic EMR (a big step forward for HCPs) and new G6 app updates.

  • The big highlights from decision support and connected devices were Hygieia’s Lancet publication, Sanofi’s announcement that it is developing connected disposable and prefilled pens, and a few more details on Onduo and DreaMed.

  • A group of diabetes thought leaders congregated in Berlin to arrive at a consensus on time-in-range targets – see our recap here. 70%+ time-in-range is the goal for type 1 and type 2 diabetes, with <4% below 70 mg/dl. The consensus targets will be officially presented in a late-breaking abstract on Friday at ADA.

This report includes our coverage of the 12th Annual ATTD conference. Immediately below, you’ll find our top five themes from the meeting, followed by highlights in the following categories:

  • Automated Insulin Delivery & Insulin Pump Therapy

  • CGM and SMBG

  • Digital Health

  • Beyond A1c

  • Diabetes Therapy

  • Exhibit Hall

  • ATTD Yearbook.

This report adds coverage of a number of posters, highlighted in blue at the end of each section.

ATTD 2020 will be held in Madrid from February 19-22 – we are looking so forward.

Table of Contents 


1. Automated Insulin Delivery: New Data (670G, 640G, Control-IQ, Basal-IQ), Medtronic’s Bold Pipeline Plans, Commercialization for Cambridge; Interoperability

  • ATTD shared important new hybrid closed loop data from Medtronic (2-6 year old pivotal, real-world data), plus a bold pipeline overview. The headline from the 2-6-year-old 670G pivotal (n=46) was that the system works similarly to the 7-13-year-old group: Participants saw a 0.5% A1c decline from a baseline of 8.0%, with a two-hour improvement in time-in-range (55%->64%) and no change in hypoglycemia. Medtronic intends to launch this indication within “1 year,” in tandem with a Bluetooth-enabled 670G and a remote monitoring app for caregivers. Also within “1 year,” the company aims to launch: (i) the next-gen MiniMed 780G with auto boluses, an optional set point at 100 mg/dl, and remote software updates; (ii) a 7-day wear infusion set; and (iii) non-adjunctive CGM with fewer fingersticks (Day 1 calibration). Can it deliver on these aggressive pipeline expectations within such a short time? Meanwhile, CareLink real-world data from commercial 670G users (n= 62,837) showed that the system is continuing to deliver results in the real-world similar to those seen in the pivotal studies: Type 1s on 670G who upload to CareLink are spending 69% in-range, while type 2s are up to 74% (with essentially the same amount of time spent in auto mode).

  • There were also two impressive predictive low glucose suspend (PLGS) data sets, one from Medtronic and one from Tandem. The SMILE study investigated Medtronic’s MiniMed 640G in a population at high risk of severe hypoglycemia. Results were striking: In the six-month RCT, individuals at high risk of severe hypoglycemia using the 640G system saw a 73% reduction in events ≤55 mg/dl and a whopping 84%(!) reduction in severe hypoglycemia events compared to those on the 640G pump alone (blinded CGM). This study shows the clear health-economic case for PLGS (and presumably hybrid closed loop) in this high-risk population, which was excluded from studies such as Tandem’s PROLOG (Basal-IQ) and ASPIRE. Tandem presented real-world Basal-IQ data, showing that 2,712 users spend just 17 minutes per day below 70 mg/dl, which is even 20 minutes less than seen in the PROLOG pivotal study!

  • On the commercial front, we learned that Tandem’s Control-IQ pivotal data will be presented at ADA, and a pediatric trial (ages 6-13) was expected to begin recruiting shortly. A related update came on UVA’s 10-month Project Nightlight study (n=60), which uses the Control-IQ algorithm and is set to complete in April; the algorithm continues to deliver solid outcomes, with 74% time-in-range and 1.5% <70 mg/dl. Also on the product front, Cambridge’s Dr. Roman Hovorka announced plans to commercialize the CamAPS FX closed loop system (Cambridge MPC algorithm, Dexcom G6, and Dana R/RS pumps). Dr. Hovorka has been doing phenomenal work in this space for such a long time, and we are glad to see a potential path to market.

  • A DIY automated insulin delivery session began with an excellent JDRF description of the vision behind component-based, mix-and-match AID. We also learned of the cleverly-dubbed DIWHY survey, which asked about DIY system use in 671 people from 27 countries. The main (self-reported) outcome suggested DIY systems improve A1c by 0.8%, with an impressive +20%-gain in time-in-range (64% to 84%) – that’s nearly five hours per day and two full months per year! Meanwhile, the Good News project is a multicenter, international study of the DIY AndroidAPS. The goal is to confirm the DIY system’s effectiveness, following promising pilot data comparing AndroidAPS vs. MiniMed 640G in a three-day ski camp study (DT&T 2018); time-in-range in the small study was equivalent between groups, with an 11 mg/dl advantage on mean glucose. An AndroidAPS poster (n=13) demonstrated the system’s benefit over SAP, with a time-in-tight-range (70-140 mg/dl) increase of 57% to 64% (+1.7 hours per day).

  • FDA’s Dr. Courtney Lias gave a talk expressing that the Agency should be able to create a class II iController pathway for automated insulin delivery. Tidepool is openly working with FDA on such a pathway for Tidepool Loop. Dr. Lias also reviewed the newly-created, class II ACE pump category for interoperable insulin pumps (currently only Tandem’s t:slim X2).

2. CGM Data: Illuminating (~500,000 Libre Users), Positive (iPro 2 CGM Study), and Concerning (CITY Baseline Data); New CGM Deployment Models (Direct-to-Consumer) Highlighted

  • On the CGM data front, we saw illuminating, positive, and concerning outcomes at this ATTD. The “illuminating” took the form of Abbott’s ~500,000-user real-world data set across 26 countries (the largest CGM dataset ever shared), providing a granular look at typical CGM outcomes for high-, median-, and low-frequency scanners. The median Libre user (10 scans/day) has a time-in-range of 56% (70-180 mg/dl), with 34 minutes/day (2%) spent <54 mg/dl and four hours/day (17%) at >240 mg/dl. The variability in this data set was also significant – as seen in the coverage below. A positive update came for Medtronic’s blinded iPro 2 professional CGM (n=102) in the ADJUST study, which drove a 1.3% A1c decrease (baseline: 9.4%). There was no control arm, but the study lends further credence to the use of four professional CGM applications per year. And finally, the concerning : Baseline blinded CGM data from the Helmsley-funded CITY study (n=153) of CGM in type 1s ages 14-25 years without recent CGM use showed median time-in-range of 35%, with a median of 60% spent >180 mg/dl. These individuals are now in the midst of six months of real-time wear, and results will read out at ADA.

  • A session devoted to new CGM support models provided plenty of food for thought, and promising paths forward to deriving more value from CGM. Jaeb’s Dr. Beck gave an in-depth look at the Helmsley-funded Jaeb and Cecelia Health study investigating direct-to-consumer CGM in type 1 adults (or type 2s on basal-bolus). The study is enrolling ~30 participants, mailing them a G6 or FreeStyle Libre, and then providing them with remote onboarding support and coaching/education for the duration of the study. We’re very eager to hear details on the coaching and patient sentiment on what it provides. Similarly, Onduo’s Dr. Ron Dixon described the CGM “hyperloops,” which entail direct shipment of Dexcom CGM, followed by telemedicine physician consults, and short real-time CGM sprints, supported by apps and coaches to identify trends and generate insights. We believe this sort of model could help to democratize CGM access, scale education, and help people get more insight out of using CGM and we look forward to seeing investment and results.

3. Win for Type 2 Diabetes Tech as Hygieia’s RCT Published in the Lancet; TypeZero Insulin Dosing Decision Works – If You Use it; DreaMed Pipeline Updates

  • An important diabetes technology visibility win came in the form of a Lancet publication of the six-month RCT of Hygieia’s d-Nav Insulin Guidance System (BGM with built-in insulin titration plus remote HCP support). In the study (n=181), the d-Nav system conferred a statistically significant A1c reduction of 1.0%, compared to 0.3% in the control group of only HCP support (p<0.0001). Could this publication help accelerate the thus-far slow US adoption of basal titration algorithms for type 2s? 

  • Also in basal insulin titration, Sanofi’s products (Diabeo, My Dose Coach) are being used by ~3,500 people in six countries, and a recent study in India showed that My Dose Coach resulted in a whopping 2.7% A1c reduction (baseline: 9.9%) in individuals initiating basal insulin therapy. Wow!

  • Interim outcomes from a study of TypeZero’s MDI dosing decision support (with Dexcom G5 CGM and Novo Nordisk smart pens) were more nuanced than we expected. The first-of-its-kind study had n=77 completed and n=2 ongoing at the time of ATTD, so final statistical outcomes were not shared. At a high level, however, decision support has not led to overall population benefits vs. a control group on CGM and MDI. Time-in-range (70-180) was a solid ~60% in both groups at baseline (with ~4.5% time <70 mg/dl), and both groups improved by a similar amount during the study. TypeZero’s MDI decision support – a smart CGM-based bolus calculator, basal and bolus titration, bedtime and exercise advice, and hypoglycemia warnings – has surprisingly not added much. Why? As with any technology, decision support only benefitted those who actually used it – and use varied quite a bit in this study. We’ll be interested to see more granular data when the study completes (perhaps at ADA), as this is a key area for diabetes technology to get right.

  • A DreaMed presentation confirmed that the company is working on a patient-facing bolus calculator (“Advisor Dose”), as well as insulin titration for SMBG/MDI users. The company announced a partnership with Biocorp for injection dose capture, which could play into both of these systems. We didn’t hear anything new on the DreaMed Advisor Pro, the very cool insulin pump settings adjustment feature that is piloting around the world.

4. All Three Insulin Manufacturers Now IN on Dose Capture, with Sanofi Announcements; Novo Nordisk (Limited) Pilot Data and Exhibit Hall Demo

  • While Lilly and Novo Nordisk had already made their connected dose capture strategies public, Sanofi had been quieter until a symposium at this ATTD: The company is indeed developing connected insulin pens in both the prefilled and durable form factors, for both basal and bolus insulins. This means all three major insulin manufacturers have connectivity strategies, and we’ll be excited to watch how their strategies, investment, and business models play out. Primarily, Lilly is developing its own closed loop and smart pen dosing decision support systems; Novo Nordisk is building connected dose capture devices and supplying the data to other companies in an ecosystem (e.g., Roche, Dexcom, Abbott, Glooko); and Sanofi is both building its own tools (pens + type 2 patch pump; dose titration apps) and rolling them out in coordinated care models such as Onduo. The big question is how the various cultures of these companies (and R&D, and businesses) adapt to the sometimes faster-moving worlds of diabetes technology.

  • We also saw a Novo Nordisk poster with outcomes from a small number of pilot connected pen users, as well as a demonstration of the NovoPen 6’s scanning in the exhibit hall. The poster reported outcomes from 31 children in the larger Swedish pilot (n=700 users); changes in glycemia were unremarkable, but both physicians’ and patients’ perceptions of the device improved over the course of the six-month study. The “Digital + Health” themed Novo Nordisk booth drew big crowds as reps demonstrated how to transfer injection history into the Diasend app simply by holding the back of the pen against the phone. We didn’t hear any updates on Lilly’s connected care products.

5. ATTD International Consensus Meeting Arrives at Target Times-in-Ranges for Different Groups of Patients (70%+ time-in 70-180)

  • A group of diabetes clinician and researcher luminaries descended on Berlin a day before ATTD to arrive at further consensus on CGM targets for different patient groups. The targets will be presented at ADA, with the headline goal of 70%+ time-in-range (70-180 mg/dl) for people with type 1 and type 2 diabetes. The researchers agreed on less than 4% below 70 mg/dl and less than 1% below 54 mg/dl. See our initial coverage of the meeting here, and note that the targets and subdivisions may be further refined in the ADA abstract.

Automated Insulin Delivery & Insulin Pump Highlights

Medtronic Closed-Loop Pipeline: “1 Year” = 670G with Bluetooth, 2-6 Year Indication; Auto Bolus 780G; 7-Day Wear set; “2+ Years” = Personalized Closed-Loop, Phone Control, Lower-Cost System

Medtronic Diabetes VP of R&D Ali Dianaty provided a clearer and more ambitious closed-loop pipeline than we’ve seen, headlined by specific plans for four product launches in “1 year”: (i) 670G with Bluetooth, mobile app, and a 2-6 year old indication; (ii) the 780G with auto boluses and remote software updating for in-warranty pumps (80% time-in-range, mean glucose of 135 mg/dl); (iii) a 7-day wear infusion set (!); and (iv) CGM with fewer fingerstick calibrations (just on day 1) and an insulin-dosing claim. The key slides below show the feature details, offering far more nuance and specificity than we heard at either JPM or in Medtronic’s 4Q18 call. Of note, the Bluetooth-enabled 670G with a 2-6 year-old indication will launch separately and before the MiniMed 780G with auto-boluses; as of JPM, we had thought Bluetooth wouldn’t come until the 780G. The timeline to launch a 7-day wear set in the next year was new and seems very aggressive. (In its booth, Unomedical told us this 7-day wear technology is not the coated Lantern, which is currently in a larger study at Stanford; see DTM). For context, the precursor Mio Advance has been FDA-cleared and available in Europe for nearly a year, but still has not launched in the US. (The companies are still building capacity, according to Unomedical.) This was the first time we heard the goal of a mean glucose of 135 mg/dl for the 780G with auto boluses, which would bring the average down by ~15 mg/dl from the 670G and represent another ~0.5% A1c reduction. (Technically, JPM and 4Q18 called for a 780G launch within the next 14 months – i.e., by April 2020.) Within the next year, Medtronic expects a 10x reduction in Auto Mode exits and half the alarms/alerts as on 670G – very big wins and terrific news to share with patients and parents and partners. Finally, it would be a huge gain if Medtronic could get to day 1 calibration for its CGM within “1 year,” bringing it closer to the no-cal G6 and Libre and obtaining Medicare coverage at the same time. (It’s unclear if Medtronic plans to go for iCGM at the same time, as was stated at JPM.)

  • In “2+ years,” Medtronic expects five product launches: (i) personalized closed loop (>85% time-in-range, mean glucose of <130 mg/dl, just received FDA breakthrough device designation); (ii) smartphone control (expanding on the remote monitoring app); (iii) a 50% smaller sensor with easier insertion; (iv) a lower-cost pump that is 50% smaller (new in the pipeline; it looks like half the size of an insulin pen); and (v) an all-in-one sensor/insulin infusion set with factory calibration (fingerstick replacement). The latter two are very ambitious and we’ll be curious to see how far beyond two years those are – Medtronic’s pipeline and timelines are often a moving target. Still, we’re glad to see a low-cost system is on the map now, as that is clearly where the field must go to reach better care for all patients globally – especially those in low-income and marginalized populations.

FDA’s Dr. Lias Optimistic Agency Can Create a Lower-Risk Class II iController Pathway for AID, Reviews Special Controls of New ACE Pump Designation

FDA’s Dr. Courtney Lias provided additional color on the special controls of the new class II ACE pump designation, and commented that “we’ll see, but it’s likely” the Agency will be able to create a lower-risk class II category for a standalone AID controller. Remarks on FDA’s evaluation of a possible class II “iController” were all of two sentences, but it is notable that Dr. Lias believes such a designation is possible. Currently, control algorithms are the critical class III components of AID systems, which determine how much insulin to give (or not) at set intervals. A recent Tidepool blog post notes that the nonprofit is working on the iController pathway with Tidepool Loop; we suspect Tandem and Dexcom might also try to go this route with Control-IQ (our speculation). We wonder what the special controls would be like for this category; obviously there’d need to be a tremendous degree of transparency on how the algorithm works, what it targets, and what kind of CGM accuracy it needs. We wonder how FDA would intend to handle incremental changes to an algorithm (if at all) – what would need a new submission and what would not? Perhaps these algorithms will be regulated similar to class II bolus calculators or basal insulin titration apps. In some respects, AID controllers are even less risky than those devices, since they only dictate the delivery of small increments of insulin at a given time, they rely on CGM data points trended over time, and they remove the human from the equation (this piece could also add risk, depending on the context). Regardless, an iController path would be the final domino, fully enabling a plug-and-play, interoperable ecosystem of component, 510(k)-regulated parts. We can’t believe how fast this is moving! Special kudos to FDA for its ambition for patients and the field – the field continues to be impressed by how quickly they are moving and how collaborative a partner they are to the field.

  • The ACE pump category was created just before ATTD when FDA authorized marketing of Tandem’s t:slim X2 as the first interoperable insulin pump; these pumps can be used as part of an automated insulin delivery (AID) system or as a standalone pump (with/without CGM). The primary advantage of the designation is that t:slim X2 and all future ACE pumps will be allowed to integrate into AID systems without having to perform additional clinical trials and submit new PMAs. Dr. Lias highlighted these clear benefits to sponsors, other component developers, and patients. In Q&A, she suggested that an ACE pump could theoretically also be used to send data to other connected devices/decision support apps so the pump company wouldn’t have to be involved in those app submissions. As far as regulatory is concerned, she emphasized that when a manufacturer designs an ACE pump, it must support that purpose across the board (e.g., in recall activities, software maintenance activities, and device modifications). The special controls themselves do not have a performance standard – unlike for iCGM – and Dr. Lias noted that there’s actually not a big difference in terms of required testing between a standard pump and an ACE pump. In both cases, basal and bolus accuracy, occlusion detection, drug compatibility, etc. must be demonstrated. However, ACE pumps must: (i) have procedures in place to ensure secure and reliable data transmission to/from digitally connected devices (including processes for authentic communications, necessary state levels such as battery and reservoir levels, risk mitigation design, and plans for how complaints are handled); (ii) incorporate failsafe design features (e.g., safe basal) and enable data logging to facilitate device failure investigations when multiple devices are involved; and (iii) provide more specific information on pump performance than what is normally given (e.g., so that AID algorithm manufacturers can see that the device is capable of delivering appropriate insulin rates and doses).

SMILE Study of 640G: Stunning 84% Reduction in Severe Hypoglycemia in High-Risk Population in Six-Month RCT

Dr. Pratik Choudhary (King’s College London) shared stunning outcomes from the SMILE study, testing Medtronic’s MiniMed 640G (predictive low glucose suspend) in a critical population – those at high-risk for severe hypoglycemia. The six-month trial randomized hypoglycemia-unaware patients to the 640G pump alone (n=77; blinded CGM) or the MiniMed 640G suspend-before-low with real-time CGM (n=76). The primary endpoint showed tremendous impact: a 73% reduction in events <55 mg/dl, 38% shorter events, and 79% less time spent <55 mg/dl with suspend-before-low. An even bigger headline came on the secondary endpoints: an 84% reduction in the rate of severe hypoglycemia – 52 events/100 patient-years with a pump alone vs. 8.5 events/100 patient-years with the 640G’s suspend before low. Wow! Time <70 mg/dl declined by a full one hour per day on 640G with suspend before low (8.1% to 2.8%) vs. no meaningful change on pump alone (9.4% to 9.1%). Talk about the power of predictive low glucose suspend! There was no significant between-group change in A1c levels, a sign that the 640G wasn’t causing rebound hyperglycemia, also a big deal. See the full outcomes below, including a plot of how the changes compare to other studies in hypoglycemia-unaware population. These data are a tremendous win for the field, especially because studies like PROLOG (Tandem Basal-IQ) and ASPIRE excluded this high-risk population.

  • As an aside, we loved hearing the new terminology: “technology monotherapy” (pump or CGM) or “technology dual therapy” (sensor-augmented pump). This just came out in a Diabetes UK position statement in Diabetic Medicine and co-authored by Drs. Pratik Choudhary, Partha Kar, and colleagues. We hadn’t heard it put this way before, but it makes a lot of sense – take a prescribing approach from the drug world and apply it to technology.

670G Pivotal in 2-6 Year Olds: 0.5% A1c Decline (Baseline 8%); +2 Hour/Day TIR Gain; No Change in Hypo; Consistent with 7-13-year data

Medtronic’s Dr. John Shin presented detailed results from the MiniMed 670G pivotal study in 2-6 year olds over three months of use (n=46, n=6,697 patient days). Results were nearly identical to the 7-13 year-old cohort: a 0.5% decline in A1c (baseline: 8.0%); an outstanding +2 hours/day spent in 70-180 mg/dl (from 55% to 64%); and no change in time <70 mg/dl (from 3.6% to 3.5%). Time in the tighter 70-140 mg/dl range also increased by two hours per day (34% to 42%) – a meaningful gain in a tough-to-manage age group. After three-months on the 670G, 52% of patients had an A1c <7.5%, up from 32% at run-in. Mean glucose declined from 173 mg/dl (run-in) to 161 mg/dl (670G), with a smaller change in standard deviation (65 mg/dl to 63 mg/dl); as a result, CV increased nominally from 38% to 39%. Median sensor wear was a strong 93%, with 87% of the time spent in Auto Mode – much higher than 81% in 7-13 years and 76% in 14-21 years. There were zero DKA or severe hypoglycemia events. As with the prior 670G pivotals in other ages, the design compared two weeks of run-in (CGM and pump in open loop) vs. three months of hybrid closed loop. The modal day profile shows the expected reduction in variability – especially overnight – though some is obviously due to the imbalance in use time (i.e., 14 days of CGM data from the run-in will have wider bars than 90 days of data). Still, the 670G’s overnight consistency in this age group probably saved parents a lot of sleep and worry. As of Medtronic’s pipeline plans (see above) the company plans to launch the 2-6 year old indication within one year, combined with Bluetooth and a remote monitoring app for caregivers.

  • Dr. Shin also showed some valuable comparisons to other age groups; the 670G has been remarkably consistent across all the key metrics. As shown below, all groups have seen a 0.4%-0.6% A1c reduction, a time-in-range gain of 1-2 hours per day (largest in the <14 year groups), and a similar ~3% time spent <70 mg/dl. Of note, the 2-6 and 7-13 year cohorts had the lowest time-in-range at run-in (~56%), and both groups improved by two hours per day while on the 670G (up to ~64%-65%). Auto Mode use was highest in the 2-6 year-old (87%) and 22+ year-old groups (88%), and lowest in 14-21 year olds (76%).

  • Participants in this 2-6 year-old group were all previously on a pump for >3 months, and 93% were previously on CGM. Would the outcomes have looked even better in those coming from MDI and/or fingersticks?



  • In a separate talk, Yale’s Dr. Jennifer Sherr also presented a topline look at the MiniMed 670G pivotal study in 2-6 year olds. Beyond the data discussed above, Dr. Sherr noted that this very young population is “hard” to manage, given all the variables – snacking, unpredictable eating, various caregivers, etc. The slide below nicely summarizes the glycemic impact of different technology studies, plotting A1c (Y-axis) vs. time-in-range (X-axis); the arrow indicates the change from baseline to study-end. Dr. Sherr noted that CGM clearly offers benefit on both dimensions (blue dots), though closed loop (green=670G; purple=other systems) really moves the needle further down to the right (i.e., lower eA1c, higher time-in-range). “Sensors provide a lot of help, but it’s not until we automate insulin delivery that we get into more targeted ranges.”

Real-World Tandem Basal-IQ Users (n=2,712) Spend 20 Mins Less Time <70 mg/dl Than Pivotal Trial Participants; N=1,000 Basal-IQ Study in the Works

Tandem’s Dr. Steph Habif presented solid real-world data (n=2,712) uploaded to t:connect from users with at least three consecutive weeks on Basal-IQ predictive low glucose suspend. Impressively, time-in-hypoglycemia was even lower in real-world users than that reported in the PROLOG pivotal trial across all measures. At three weeks, real-world users spent a remarkably low 1.2% time <70 mg/dl – 20 minutes less (and ~54% less) than that seen in the pivotal trial with Basal-IQ (2.6%). These improvements in time <70 mg/dl were maintained in real-world users both during the day and night. Real-world users spent a remarkably low 1.2% and 0.7% time <70 mg/dl during daytime and nighttime, respectively, while PROLOG participants spent 2.4% and 2.7% time <70 mg/dl. These results reflect differences of 17 minutes and 29 minutes in time <70 mg/dl during the day and night, respectively. A sub-cohort of 1,437 users with at least six weeks of Basal-IQ data uploaded to t:connect demonstrated that the reductions in hypoglycemia were sustained over time across all measures. Two caveats: (i) We don’t know baseline time <70 mg/dl in the real-world cohort (it was 4.5% on SAP in PROLOG); and (ii) G5 was used in PROLOG, while G6 is used in the real world. As an interesting benchmark for these data, the median FreeStyle Libre user (see highlight below) spends 2% of the day <54 mg/dl. While some believe FreeStyle Libre overreads hypoglycemia, Tandem’s Basal-IQ hypoglycemia metrics are over 80% lower.

  • Dr. Habif noted that Tandem is currently conducting a longitudinal patient reported outcomes (PROs) study, following 1,000 people on Basal-IQ during their first six months on the system. While data from the study are not yet available, she shared results from a dQ&A-conducted survey (n=417) of patients’ attitudes regarding the Basal-IQ vs. Medtronic’s MiniMed 670G. Participants rated Basal-IQ higher for every measurement presented, including satisfaction, trust, helping to feel more in control of their diabetes, helping to sleep better at night, and ease of use.

  • Split out by age, real-world data showed patients over 60 years-old (n=208) spend the lowest time in hypoglycemia, with just 0.92% <70 mg/dl, 0.22% <60 mg/dl, and 0.03% <50 mg/dl. Those in the 18-60-year-old age group had the highest time <70 mg/dl (1.29%) and <60 mg/dl (0.35%) – still, these are ridiculously low levels. No statistical analyses were provided, so it is unclear whether these differences are significant. These age-related trends held true during the day and night for time <70 mg/dl, with those over 60 years-old spending the least amount of time in hypoglycemia, followed by those <18 years-old and then those in the 18-60-year-old age group.

DIY AID: JDRF’s Vision of Interoperable Systems; Compelling OPEN Project and DIWHY Survey; Multicenter Study to Soon Test AndroidAPS

An exciting session on DIY Automated Insulin Delivery shared JDRF’s vision and new research efforts to study systems more rigorously. To start, JDRF’s Dr. Daniel Finan provided a great overview of the DIY Landscape (Loop, OpenAPS, AndroidAPS), highlighting the goal of the non-profit’s open protocol initiative – to marry DIY innovation with safety, transparency, and access. Dr. Finan beautifully explained the regulatory advantages of the iCGM and ACE pump paradigms, which move to a component-based, mix-and-match model rather than a system-based model of AID – see the slides below. The open protocol, interoperable configuration defines and regulates the components separately (see red boxes), allowing them to be connected afterwards and to be swapped in/out. The new paradigm should be faster for companies that don’t own all the components, avoid some of the complicated business relationships that have slowed the field down, simplify regulatory submissions, allow AID system components to keep up to date with new models, and allow users to mix-and-match components that fit their needs. Dr. Finan closed with a summary of outstanding issues: (i) what does an iController (algorithm) look like, what kind of data will be required to approve it, and what will be the special controls? (ii) Can we achieve true plug-and-play interoperability, or will there still need to be agreements between companies? (iii) what are the business models in an interoperable AID ecosystem? (iv) what about customer tech support – e.g., if something is not working, who should I call?; (v) Where should the algorithm live – the phone or the pump (there are pros/cons to each). Dr. Finan said manufacturers have been “very willing” to participate in the open protocol effort so far, and we hope to see more take part – Tandem, Dexcom, Insulet, and Senseonics are the next-most-likely candidates, in our view. This session also included updates on DIY research in Europe – the new Open Project and Good News Project are discussed below Dr. Finan’s slides.


  • On the research front, Dr. Katarina Braune (Charité – Universitätsmedizin Berlin) shared results from the DIWHY survey, asking about DIY system use in a remarkable n=671 people from 27 countries (81% Europe; median age of 42 years; 76% adults and 24% children). The main (self-reported) outcomes are shown below, suggesting DIY systems improve A1c by 0.8% with an impressive +20%-gain in time-in-range (64% to 84%). The “Why do you DIY” question was equally interesting (orange slide), showing the myriad reasons why individual choose to use these systems. Dr. Braune highlighted improved sleep – especially among parents – as the most strongly endorsed reason. The survey is still open until the end of the month at

    • She also discussed The Open Project – – an international collaboration studying DIY AID systems via prestigious EU Horizon 2020 funding (<1% acceptance rate!). Some of Europe’s leading research institutes are involved in the consortium, including Steno Diabetes Center. DIY systems (OpenAPS, AndroidAPS, Loop) will be studied on several fronts: clinical outcomes, quality of life, and barriers to scale. The consortium will even use machine learning to enhance systems’ predictive capacities.

  • Also on the research front, Dr. Lenka Petruzelkova (University Hospital Motol) mentioned the “Good News Project,” a soon-to-start multicenter international study of the DIY AndroidAPS. The goal is to confirm the DIY system’s effectiveness, following promising pilot data comparing AndroidAPS vs. MiniMed 640G in a three-day ski camp study (DT&T 2018). As the picture below shows, AndroidAPS was identical to the 640G on time-in-range, with a meaningful 11 mg/dl advantage on mean glucose.

Tandem/Dexcom Control-IQ Pivotal to Report at ADA – 95% Complete, not a Single Participant Has Dropped Out At Any Site (N=168)

Dr. Boris Kovatchev shared that Tandem/Dexcom’s Control-IQ/G6 pivotal study (14+ years) is 95% completed, will wrap up in ~mid-April, and results will be presented on Sunday of ADA 2019. The six-month study has randomized n=168 people over six months (2:1, closed-loop to SAP), and not a single study participant has dropped out at any site – a remarkable vote of confidence in the next-gen hybrid closed loop with automatic boluses and the no-calibration G6. Mean A1c is 7.6% (wide range of <7% to 10.5%), with 21% of participants on MDI and 30% previously not on CGM. Wow, is this data going to be exciting! Tandem’s 4Q18 call the week after ATTD targeted a US launch for Summer-to-End of Q3. (Can Tandem get the data submitted and approved quickly enough? We suspect so, as this algorithm has a lot of data and experience behind it.) Also in new news, a pivotal trial for Control-IQ in pediatrics (6-13 years) is expected to start recruiting soon – this will be important to get going as Medtronic moves ahead with the 2-6 year-old indication (see above). Also on Sunday of ADA 2019, we’ll hear data from the French FreeLife Kid artificial pancreas study – testing the same Tandem Control-IQ/G6 configuration at five sites in France (n=97 randomized). In the past year, >50,000 days of data have been logged on Control-IQ, tripling the amount at this time last year. Wow!

  • The Project Nightlight study (n=60) – which recently switched to Control-IQ – also shows consistently positive outcomes (74% time-in-range, 1.5% time <70), especially in the switch to the t:slim X2/G6 configuration. The ten-month (!) study will end this April, comparing evening/overnight closed loop vs. 24/7 closed loop vs. sensor-augmented pump. See interim CGM outcomes below – of note is the comparison between the middle column (Control-IQ on Android phone, Roche pump, Dexcom G4) vs. the right-most column (commercial-ready t:slim X2/G6 with embedded Control-IQ). Dr. Kovatchev noted the marked six-percentage-point improvement in time-in-range (+1 hour per day) just from switching the hardware, a reminder that form factor and reliable communication will help dramatically.

Dr. Bruce Bode presented Medtronic 670G data uploaded to CareLink from 60,394 type 1 patients and 2,443 type 2 patients in the US as of January 2019. Both groups saw increased time-in-range (70-180 mg/dl) after initiating auto mode, with type 1s increasing time-in-range from 61% to 69% (+2.1 hours) and type 2s increasing from 66% to 73% (+1.7 hours). Time spent in auto mode was comparable for type 1s (75%) and type 2s (76%) – both slightly lower than recommended by many KOLs and that seen in the pivotal (closer to 80%). After breaking out the data by age group, Dr. Bode noted that those ≥65 years-old (n=3,594) achieved the highest time-in-range (74%) and the greatest time spent in auto mode (80%). Indeed, time-in-range and time spent in auto mode increased across age groups, with the youngest age group (≤17 years-old) spending 61% of the day in-range and 64% of the day in auto mode.

  • Dr. Bode also demonstrated a CareLink feature capable of comparing national data to specific territories and even specific practices. In the example below, 670G patients (n=349) from Dr. Bode’s clinic in Atlanta achieved a slightly higher time-in-range of 73% when in auto mode as compared to the national average (n=69,373) of 72% time-in-range when in auto mode.

McGill Embarking on Two New Studies: JDRF-Funded, Free-Living Study of Pramlintide-Insulin System, and Inpatient Study of Triple Hormone AID (Fiasp + Pramlintide + Glucagon)

After presenting the dual hormone (pramlintide + insulin) closed loop data he initially detailed at ADA, McGill’s Dr. Ahmad Haidar shared that his group has two additional studies on tap:

  • A second JDRF-funded study (n=130) with the same dual hormone system, but this time for two-weeks at home and with no carb counting. This system will only have a qualitative meal announcement (“I’m eating”). The primary outcome of this trial, Dr. Haidar told us afterward, is quality of life.

  • A non-inferiority study of a fully-automated, triple hormone (Fiasp + pramlintide + glucagon) artificial pancreas. The control group in this trial is a single hormone AID system with Fiasp. The inpatients will have to wear three separate pumps with insulin, pramlintide, and glucagon, respectively, so Dr. Haidar emphasized that this is purely a probing, academic question at this point. Wow could this be amazing! Data from the first test is displayed below; red traces are glucose, blue traces are insulin, purple numbers are pramlintide (infused at fixed ratio to insulin), and green bars are glucagon.

Dr. Haidar reasoned that the insulin-and-pramlintide system might be “the next logical treatment in patients approaching, not yet achieving, A1c targets with the insulin system alone.” Literature shows that for a high A1c, the relative contribution of postprandial glucose (vs. fasting glucose) to A1c is 30%; for A1c near 7%, the relative contribution is 70%. Therefore, as A1c decreases, postprandial glucose contributes more than fasting glucose to A1c. This is a strong argument (with historical underpinnings from Prof. Monnier – see our EASD 2012 report) and opens up a potentially interesting commercial strategy, should a company take the idea and run with it, though the field is moving away from A1c, so we’d be curious to see a similar analysis with time-in-range (we imagine it’d turn out the same).

In the preliminary inpatient data presented at ADA, use of pramlintide (amylin analog) on top of rapid acting insulin during a 24-hour, in-clinic, JDRF-funded study boosted time in 70-180 mg/dl by nearly three hours/day vs. insulin alone. There were no increases in hypoglycemia or side effects. In a post-trial survey, 19/26 (73%) individuals agreed or strongly agreed that they would use a co-formulation product of rapid insulin + pramlintide if it were on the market. Only three (~12%) disagreed or strongly disagreed. This study used the Class AP algorithm that Lilly licensed for its in-development hybrid closed loop – we can’t help but wonder whether Lilly will pursue this dual-hormone approach (unsurprisingly, the company has said they are keeping an open mind with regards to closed loop system design).

Dr. Roman Hovorka’s Plans to Commercialize CamAPS FX Closed Loop System with Cambridge Algorithm, Dexcom G6 CGM, Dana R/RS Pumps; Closed Loop Data in 1-7-Year-Olds Find ~70% Time-in-Range

Although Dr. Roman Hovorka presented a slew of exciting closed loop data, the most unexpected update came when he briefly mentioned plans to commercialize a new AID system comprised of the Cambridge MPC algorithm, Dexcom G6, and Dana R/RS pumps equipped with data streaming to Glooko. While the specific path to market is unclear, Dr. Hovorka explained that the new system, called CamAPS FX, will be used in all of Cambridge’s upcoming studies slated to begin in Q1-Q2 of this year (listed below), including NIH-funded Dan05 (n=130 6-18 year-olds; six months), KidsAP02 (n=66 2-7 year-olds; four months), and a trial in pregnancy (n=90; time unspecified). We were particularly excited to see Dan05 on this list – as of ATTD last year, recruitment was placed on hold as investigators awaited manufacturing of the study device. At the time, the closed loop included a modified Medtronic 640G and Enlite 3 CGM; it appears the investigators decided to hold off in favor of the commercial-ready-looking CamAPS FX.

  • Dr. Hovorka presented results from a small randomized controlled crossover trial (n=23) investigating the benefits of diluting insulin in a closed loop system for very young children (1-7 years-old). The hypothesis, Dr. Hovorka explained, was that by diluting insulin, higher accuracy might be achieved. However, the three-week study found no additional benefit of diluted insulin, with both groups demonstrating a solid time-in-range of ~70% and “acceptable” time <70 mg/dl of 4.5%. Results from the study were published in the January edition of Diabetes Care. The data will inform the longer, four-month KidsAP02 (n=66) study investigating the CamAPS FX closed loop system in children 2-7-years old. Per, the study is not yet recruiting and is estimated to begin in March and end in June 2020.

  • Dr. Hovorka shared strong closed loop data from a 12-week randomized controlled parallel trial (n=86 type 1s) published in The Lancet in October. Participants were ≥6 years-old (50% >18 years-old) and sub-optimally managed (baseline A1c: 8.3%). The closed loop group (Enlite 3 sensor, modified Medtronic 640G pump, and Cambridge algorithm on Android phone) achieved 65% time-in-range (70-180 mg/dl), spending a significant~2.6 hours more in-range than the 54% in the control group (sensor-augmented pump therapy). Improvements in time-in-range were observed across all time-in-range values – generally a 10%-15% difference between the two groups across and a difference of nearly 20 percentage points among users with the highest time-in-range between groups. Closed loop drove a minor 12-minute difference in the reduction of time <70 mg/dl (-1.4% vs. -0.2%) and a significant 2.5-hour difference in the reduction of time >180 mg/dl (-13.2% vs. -7.5%). As seen in the modal day plot, most of the benefit came overnight. Those in the closed loop group also achieved a significant 0.4% difference in A1c reduction as compared to the control group: closed loop participants decreased A1c by 0.9% (baseline: 8.3%), and control participants decreased A1c by 0.5% (baseline: 8.2%). Improvements did not differ between children and adults.

  • With such a wealth of data collected across a diversity of age groups, Dr. Hovorka was able to demonstrate that young children exhibit the greatest variability in insulin requirements. Results are pending publication and show those ≤6 years-old to have a significantly greater coefficient of variation (CV) for insulin as compared to 13-17-year-olds and ≥18 year-olds during the day and night. Dr. Hovorka believes the data are a strong case for closed loop adoption in the very young. We agree that the pediatric population is where huge benefits could be had, particularly given the strong pediatric CGM uptake (51% penetration) in the new T1D Exchange data. Medtronic’s 670G is now approved for those ≥7 years-old (and we saw encouraging data in ages 2-6 here at ATTD; see above), Insulet is testing its Horizon closed loop system in 2-6 year-olds, Beta Bionics received FDA IDE approval to test its insulin-only Bionic Pancreas in those ≥6 years-old, Bigfoot enrolled children ≥7 years-old in its feasibility trial, and Diabeloop plans to begin pediatric clinical trials in children 6-12 years-old.

MiniMed 670G Workshop: Simplicity for HCPs vs. Patient Complexity? Crystal Clear Goals: >70% TIR, >80% Time-in-Auto Mode, Focus on Meal Boluses

A standing-room only, practical workshop on using the MiniMed 670G highlighted an interesting dichotomy: simplicity and clarity for HCPs vs. complexity for patients. The workshop covered the system’s features and onboarding process, highlighting that healthcare providers do not have many knobs to turn in the first-gen hybrid closed loop – they can tweak insulin-to-carb ratio (the big one) and active insulin time (minimal effect). In 670G case after case, the recommended thought process was the same: (i) look for >70% time-in-range (70-180 mg/dl); (ii) look for >80% time-in-Auto Mode; and (iii) troubleshoot if either metric is too low. To get higher time-in-range, the solution was always tightening the insulin-to-carb ratio and/or encouraging pre-meal bolusing. To get more time in Auto Mode, HCPs must identify common exit reasons (usually SMBG calibrations; first picture below) and coaching patients to avoid them. Medtronic framed the 670G management process as pretty straightforward for HCPs – two metrics to watch and a small basket of solutions to fix them. “It’s a different job for us” said Medtronic’s Dr. Ohad Cohen (Director of Medical Affairs, EMEA) – this is particularly true in light of less needed focus on hypoglycemia. This was a key message to deliver to providers in the first hands-on workshop since the system’s launch last fall in Europe. Dr. Cohen also shared more real-world 670G data, covering time-in-range outcomes in n=55,000 patients and up to October 2018. Results (second and third pictures below) have impressively mirrored the pivotal trial at ~73% in 70-180 mg/dl and 2.4% in hypoglycemia. Real-world data in MDI users and no prior CGM experience are similar, albeit in tiny subgroups.

  • 670G users, however, must still cope with a lot of fingersticks, jargon, mode juggling, icons, and alarms. The one that immediately stood out was fingerstick calibration – Dr. Cohen said today that when the 670G first launched in the US, average fingersticks were six per day! The 670G launching internationally and “switching” over in the US moves this down to ~four fingersticks per day, presumably via the transmitter software upgrade that reduces unnecessary blood glucose requests (launched as of December). Missed calibration and prolonged high blood sugars were tied for the most common reasons for Auto Mode exit (15% of the time; see below), followed by prolonged max delivery (10%), sensor algorithm under-reading (10%), and minimum delivery timeout (9%). Dr. Cohen framed most of these as “safety” constraints for a first-gen system, which is a very fair point from our view (first generation is always far harder – many remember – but many don’t! – Dexcom’s STS, as one example) and will certainly improve in next-gens like the MiniMed 780G. But today’s workshop reminded us that a no-calibration AID system driven by G6 or FreeStyle Libre is going to eliminate much of the complexity and frustration in 670G. We also hope to see systems move away from “modes” (Manual Mode, Auto Mode, Safe Basal, Suspend Before Low, Suspend on Low, etc.), which should help with training and user experience. We were also reminded of some of the odd design decisions in the 670G: an inability to see the basal insulin profile at any given time (it’s only shown as little pink dots on a secondary screen), exiting Auto Mode precisely at the times where it may be most needed (extreme highs and lows), burying vital information in nested menus (e.g., the important Auto Mode readiness checklist and ON button is three screens deep), correction bolus recommendations cannot be adjusted by the user, etc. Dr. Anna-Kaisa Tuomaala (Helsinki University Hospital) said three face-to-face visits are recommended for non-Medtronic pumpers to learn the 670G, and it takes some of her patients up to two months to learn how to use it. Obviously this is a first-generation product and many 670G experiences are still extremely positive. Plus, the system is performing well on population-level outcomes, which bodes well for next-gen less-burdensome options (including Medtronic’s own 780G, Tandem’s Control-IQ, Insulet’s Horizon, Bigfoot Autonomy, Tidepool Loop, Beta Bionics iLet).

Dr. Bode Enthusiastic About Potential of Ultra-Rapid Insulin in Closed Loop, Despite Lack of Available Data

Dr. Bruce Bode covered ultra-rapid insulin in closed loop systems, offering insight on the potential of Novo Nordisk’s Fiasp, Lilly’s URLi, and Adocia’s BioChaperone Lispro and concluding that in a closed loop, ultra-rapid insulins should lower A1c and improve time in range. Read on for the latest on the three most advanced ultra-rapid insulins:

  • Fiasp is currently under investigation in two closed loop studies, including one using Medtronic’s 670G (phase 4, n=20, expected completion May 2019) and another using Beta Bionics’ iLet system (phase 2, n=24, expected completion April 2019). The former comes from Stanford’s Dr. Bruce Buckingham, has a 2-week double-blind crossover design comparing NovoLog to Fiasp, and lists time in range and time <70 mg/dl as co-primary endpoints. To our understanding, ~20 patients have completed the trial, and Dr. Bode was optimistic that results would be presented as a late-breaker at ADA 2019 in San Francisco. And while Dr. Bode admitted that we have no robust dataset yet supporting a benefit with Fiasp in closed loop systems, he did present highly compelling data from a single patient of his: A 33-year old male with over 20 years of type 1 had an A1c on 7.3% and 70% time in range (70-180 mg/dl) on the 670G, but he was frustrated with postprandial spikes. Dr. Bode switched him from NovoLog to Fiasp with no other changes or intervention, and his A1c fell to 6.7% with a 4%-point gain in time in range (with a 1%-point gain in <70 mg/dl) in ~four months. He also noted a slight increase in bolus insulin and decrease in basal with this patient, arguing against the notion that HCPs need to increase basal insulin when switching to Fiasp – rather, he says, observed nighttime highs with Fiasp are a result of patients compensating for lower post-dinner values than they’re used to. When a patient has always gone to sleep at 150 mg/dl and is now seeing 120 mg/dl at bedtime, they often bring themselves up, leading to higher nighttime values: “It’s a behavioral issue, not an insulin issue.”

  • Regulatory submission for Lilly’s URLi (ultra-rapid lispro) is expected in 2019, and Dr. Bode highlighted that phase 3 PRONTO-T1D, PRONTO-T2D, and PRONTO-Pump results will all be presented at ADA 2019. Topline results from the first two were released October 2018, and Lilly has also – as expected – started a second, larger pump study (PRONTO-Pump 2) enrolling 526 type 1s. In terms of closed loop, he pointed to a phase 2 study comparing URLi to Humalog in 670G (n=50, expected completion September 2019), which he said two patients had entered as of last week and has a primary endpoint of time between 70 and 180 mg/dl. Lilly hasn’t yet released any phase 3 data on URLi (see phase 2 data in type 1 and type 2), and though PK/PD show improvements, it remains to be seen if this will translate to improved outcomes in 670G and other closed loop systems for type 1.

  • Finally, on Adocia’s BioChaperone Lispro, Dr. Bode commented that the company has gone back and forth on whether to do a study in closed loop. For context, Adocia recently decided to advance BC Lispro into phase 3, though the company is still open to partnerships for this highly promising candidate. Management has stated that the first phase 3 study will be in pumps, but, to our knowledge, no studies have been posted to Dr. Bode seemed particularly optimistic about BC Lispro, highlighting a 61% reduction in postprandial AUC in a phase 1/2 study and commenting that benefits are seen in both type 1 and 2. On the closed loop front, Beta Bionics is running a study comparing BC Lispro, Novolog, and Humalog in a single-hormone configuration on the iLet pump in up to 30 type 1s; according to Dr. Bode, the trial is already underway (though not listed as such on and we could see results “at ADA or even before.”

PEPPER Decision Support System Reduces Severe Hypo, Increases TIR in Proof-of-Concept Study Using Cellnovo Pump + Dexcom G5 CGM; Phase 3 Clinical Results Expected by Year-End

Imperial College London’s Dr. Parizad Avari reported phase 1 proof-of-concept data on the PEPPER (Patient Empowerment through Predictive Personalized Decision Support) AID system. This project, funded by the EU’s Horizon 2020 program, is using Cellnovo’s pump and Dexcom G5 CGM to investigate algorithms for diabetes decision support. PEPPER collects CGM measurements, insulin bolus doses, meal information, and activity details to produce a 30-minute glucose forecast. The tool can then send alerts/alarms (direct notification to a phone and/or a text message), suspend basal insulin, recommend carbohydrate intake, put a dynamic constraint on insulin bolus, and offer insulin pump commands. The phase 1 study enrolled six participants with type 1 diabetes. Between baseline (weeks 1-2) and study end (weeks 7-8), there was a significant drop in time spent <54 mg/dl (p=0.02) and time spent in-range (p=0.03), but no significant difference in severe hypoglycemia or time spent in hyperglycemia. There was a significant decrease in total number of alarms between the first and last two-week windows (p<0.05), and alerts to caregivers also declined in frequency with PEPPER (p<0.01). Dr. Avari noted insignificant effects on quality of life as assessed by PAID (the Problem Areas in Diabetes scale) and DQOL (Diabetes Quality of Life). She emphasized that these findings should be interpreted as proof-of-concept, and suggested a positive proof-of-concept at that, given significantly reduced time in hypo and significantly improved time-in-range. We won’t be too critical of the lack of significant effect on quality of life metrics or milder hypoglycemia, as this was a relatively short and super small trial (n=6). Rather, we’ll look for meaningful improvements on these outcomes in later-stage investigations – after all, one of the main objects of automation should be to minimize errors and ease diabetes burden, thereby improving quality of life. Phase 2 of this clinical program analyzes not only the safety of PEPPER, but also its adaptive insulin dose calculator with AI; according to Dr. Avari, this trial is complete. Phase 3 is an open-label study comparing PEPPER vs. standard therapy, and Dr. Avari announced that full results will read out by the end of 2019. Per, the phase 3 study is expected to complete by end of July. The primary endpoint is within-subject change in time-in-range between three months with PEPPER and three months without.

  • To our knowledge, Cellnovo is the only medical device company sponsoring the PEPPER project (see all the partners here), although Dexcom CGMs were apparently used in phase 1. Dr. Avari did not disclose which products are being used in phase 2 or phase 3 (nor is this information listed on

Harvard’s Dr. Frank Doyle Explains Approach to Long-Term Closed Loop Adaptation; Up to 93% TIR in UVA/Padova Simulation

Harvard’s Dr. Frank Doyle gave a masterful overview of the cutting edge of AID controller development, highlighting his group’s latest in silico data (up to 93% time-in-range) with a multivariate learning framework for long-term adaptation. The approach, known as Bayesian optimization, “underlies a lot of the more effective algorithms,” said Dr. Doyle, “it’s the engine underlying many machine learning approaches.” The framework is based on a normal controller algorithm with feedforward control (calculates meal and correction boluses) plus the variable basal rate; to this basic algorithm, multivariate learning framework adds long-term parameter adaptation. One loop of this module intends to automatically identify the correct parameter to adjust, and the other optimizes that parameter safely and efficiently. The first loop determines which parameter should be updated on a weekly basis in an intuitive manner: If the user has recurrent hypoglycemia/hyperglycemia overnight, it selects the basal rate to optimize; if the user has recurrent postprandial hyperglycemia/hypoglycemia, it selects the insulin:carb ratio (ICR); and if the user has overall hyperglycemia/hypoglycemia, it selects the controller algorithm. It ceases performing evaluations when glycemia has returned to normal, or if the data suggests there’s no way to better eliminate variability with further tuning. The second loop adapts the selected parameter through linear programming. Dr. Doyle et al. have challenged the algorithm with two scenarios in the UVA/Padova type 1 diabetes simulator: In the first scenario, the starting ICR is twice what would be optimal, and the basal rate is half optimal. In the second, both ICR and basal rate are doubled. As seen in the slides below, parameters in both scenarios self-adjust in predictable manners (given their erroneous starting points), causing time-in-range to climb dramatically while mean glucose decreases dramatically. Up next, this’ll be put to the test in clinical evaluation.

  • The Doyle group at Harvard maintains a database of artificial pancreas studies, mapping each study onto a plot of hypoglycemia vs. time-in-range. We enjoy seeing this figure transform each time Dr. Doyle presents, reflecting the latest trials – will the long-term adaptation approach begin to bring points in the red box at the top left?

Prof. Barnard Presents INSPIRE Instrument for AID PROs, How to Improve PROs in Diabetes

In a high-level talk on patient-reported outcomes (PROs), Professor Katharine Barnard plugged the INSPIRE measures as a way to evaluate a type 1 diabetes patient’s psychological experience with automated insulin delivery. A paper establishing the validity of INSPIRE was very recently published in Diabetic Medicine. The assessment involves a series of questionnaires (17-22 items each) gauging quality of life, fear of hypoglycemia, diabetes distress, and glucose monitoring satisfaction. Separate questionnaires are given to parents and partners of type 1s. This initial study found INSPIRE to be internally consistent, and Prof. Barnard suggested it could have important applications in the next generation of diabetes technology RCTs. A major challenge with PROs today, she explained, is that we have insufficient evidence correlating them to clinically-relevant outcomes in diabetes, such as change in A1c or time-in-range. She claimed that INSPIRE could show an association with A1c and other outcomes, and we hope to see more data on this going forward; at the very least, it may correlate with adherence, which may be in and of itself sufficiently compelling for payers. Prof. Barnard lamented the lack of standardization in how PROs are collected, analyzed, and reported (framing INSPIRE as one potential standard assessment), and she argued that we need more robust PROs than what the diabetes technology field is currently using. She discussed GOLD scores as one example: With a number between 1-7, we know with some certainty if a patient is aware or unaware of the onset of hypoglycemia, but we don’t capture how hypoglycemia is actually impacting their life, whether it’s exacerbating stress, fear, or other negative emotions. INSPIRE takes a more holistic approach, getting at “life impact” by evaluating someone’s diabetes distress right alongside their treatment satisfaction. Prof. Barnard shared enthusiastically that many clinical trials globally are starting to incorporate the INSPIRE measures, which will soon lead to data reported in a standardized fashion on individual experiences with diabetes technologies.

  • Moreover, Prof. Barnard advocated for more RCTs with PROs as a primary outcome. Another current challenge with diabetes PROs is that most studies are under-powered on these endpoints, she explained (sample size is too small, measurements are inaccurate, etc.). Where we have ample real-world experience with diabetes tech, say in the CGM category, Prof. Barnard argued that it’s high-time to focus clinical trials around user experience and psychosocial impact. See the Q&A below for her full take.

  • Prof. Barnard celebrated the fact that we’re now debating how to use PROs and not if. As she put it, “I’m thrilled to be giving this talk, because it means we’ve moved away from ‘should we assess PROs’ to ‘how should we best measure and report PROs.’” We’ve certainly seen substantial progress in how different stakeholders – including FDA/other regulators – look at PROs and factor them into decision-making. We’re strong proponents of more standardization across trials so that PROs can be useful in clinical decision-making as well; Prof. Barnard commented that it’s still impossible to reliably judge who will do best on a given technology, and she pointed in this direction as the path forward for diabetes PROs.

Questions and Answers

Dr. Korey Hood (Stanford University, CA): What happens when there’s a separation of clinical outcomes vs. psychosocial outcomes, e.g. when there’s an efficacious and safe clinical outcome but there’s a psychosocial side that’s falling behind, or maybe there’s a negative psychosocial effect? What should regulators and device companies do in those cases?

Prof. Barnard: This is perhaps an example of when to discontinue a device, and it shows that if we don’t understand the psychological impact of technology before we give it to a patient, it could cause harm – especially if there’s a prior expectation that the new device will be great. Take CGM, for example. We have lots of evidence now that CGMs improve time-in-range and biomedical outcomes, that they’re safe and efficacious. So maybe CGM trials should now be around usability and fitting into everyday living to meet individual user needs. There’s a point where you say: Insulin pumps are safe and beneficial; CGM is safe and beneficial. What we need to research now is psychological impact. Let’s step up and adequately assess this. Let’s power trials on psychosocial outcomes.

Q: I wanted to ask about the social determinants of health, and how that influences your interpretation of PROs?

Prof. Barnard: Social determinants (for example, socioeconomic status or gender) have a direct impact on people in terms of healthcare access, health literacy and numeracy, their ability to engage with care. A lot of devices are still very difficult to use if you have low numeracy or low literacy skills. So I think social determinants are very important, and we do actually look into some of these things in the INSPIRE measures. For now, this information usually comes from the demographic part of a study rather than the patient-reported outcomes part.

PROLOG Sub-Analysis Finds CGM Use Alone Does Not Confer Hypoglycemia Reductions Observed in CGM-Naïve Participants

In a Tandem symposium, we saw a new sub-analysis of the pivotal PROLOG trial for t:slim X2 with Basal-IQ predictive low glucose suspend showing outcomes split by prior pump and CGM use. Sansum Diabetes Research Institute’s Dr. Jordan Pinsker underscored that for those without previous CGM use, the addition of CGM alone was not responsible for the reductions in time <70 mg/dl achieved with Basal-IQ. As depicted below, non-CGM users (a small n=16) in the sensor-augmented pump (SAP) group still spent 6% time <70 mg/dl, which improve to 4% time <70 mg/dl in those on Basal-IQ – a difference of ~33 minutes. Those in the SAP group actually spent more time <70 mg/dl as compared to baseline (+14 minutes). Interestingly, the same cannot be said for the results broken out by pre-study insulin delivery. Prior MDI users saw 2.6% time <70 mg/dl on SAP and 2.3% time <70 mg/dl on Basal-IQ – an inconsequential difference between the two groups. Overall, these are very, very low rates of hypoglycemia – far lower than typically seen in real-world use.

MiniMed 780G Has Optional Set Points of 120 or 100 mg/dl; NIH-Funded FLAIR Study (670G vs. 780G) to Start Within Months

Dr. Rich Bergenstal shared that, beyond a design for improved user experience (“85%, 90% of time in auto mode”) and automatic correction boluses, the next-gen MiniMed 780G hybrid closed loop system will allow users to opt in to a blood glucose set point of 100 mg/dl (default will still be 120 mg/dl, as it is with 670G). This is the first time we’re hearing of the optional more aggressive target of 100 mg/dl, but it makes sense; the 670G’s glucose target is 120 mg/dl, but its mild auto-basal aggressiveness means it typically achieves a mean glucose of ~150 mg/dl in real-world use. The goal for the 780G is a mean glucose of 135 mg/dl. Dr. Bruce Bode earlier said that the NIH-funded FLAIR study of 670G vs. 780G would begin in May – in slightly more tempered fashion, Dr. Bergenstal stated that the study would begin “within months … we’re waiting for final approval.” His fingers are crossed that the system will be available by ATTD 2020 (Medtronic guided for within “1 year”; see above), despite a number of delays since the initial plan to commence FLAIR in 2017. As a reminder, FLAIR (n=112) is a crossover study (12 weeks on each system) with a primary outcome of superiority for time ≥180 mg/dl during the day (6 am-midnight). Dr. Bergenstal emphasized that the study is looking at “adolescents, those on MDI alone, the postprandial period, and those on 670G who are not at goal. We’re focusing on time-in-range, not really A1c. The baseline A1c will probably be higher than 7.2% or 7.4%.” It certainly seems that this system will be challenged in a way that 670G wasn’t (as a first-gen product in a safety study), which is a necessary part of extending beyond early adopters – we can’t wait to see how it fares!

NIH/NIDDK Roadmap to Implantable IP AID; JDRF Survey Finds Stakeholders Have High Bar For Adoption of IP Systems (A1c 6-6.5%, 90%-95% TIR, no hypo!); Dr. Eyal Dassau’s IP AID Proof of Concept

NIDDK’s Dr. Guillermo Arreaza-Rubin proposed several areas of improvement for developing a more efficient, user-friendly, and cost-effective new generation of implantable Intraperitoneal (IP) AID systems. Technical needs included: (i) a proper CGM partner that is either subcutaneous or IP sensing (although the final version should be IP); (ii) better insulins that are stable in pumps and do not aggregate or clog; (iii) smaller pumps with longer batteries, potentially lasting 10 years; (iv) enhanced catheter design to prevent occlusion; and (v) algorithms adaptable for IP delivery. To this end, he provided an IP closed loop product roadmap, identifying a minimum viable product and a final, fully integrated implanted AID system. Key differences include: (i) the transition between an iCGM to an integrated IP sensor; and (ii) moving from a six-week refill interval to no less than three-month interval. It’s difficult to tell just how far down the line even the minimum viable product may be. According to Dr. Arreaza-Rubin’s predictions, reaching this point alone will require the development of a small implantable pump, IP closed loop algorithm, and a reliable, intra-body Bluetooth system agnostic to CGM site.

  • JDRF’s Ms. Jackie Le Grand presented results from 66 interviews with development experts, physicians, payers, and patients to assess the market appetite for an implantable closed loop system. The biggest finding? When asked what outcomes would be necessary to convince patients with type 1 diabetes to adopt an implantable closed loop system, responders provided fairly rigorous standards: A1c of 6.0%-6.5%, time-in-range of 90%-95%, no time <70 mg/dl, and ~5% time >180 mg/dl. Concerns regarding the procedure and cost would also have to be addressed. (Such outcomes are already possible with low-carb eating and an existing system like DIY Loop.) Ms. Le Grand generally characterized physicians as viewing implantable closed loop systems “positively,” noting that the primary driving force was its potential to reduce hypoglycemia – that said, current subcutaneous closed loop already does this. General anesthesia was a major concern amongst physicians, as was durability in the event of a trauma. Durability was also a major concern for patients, in addition to undergoing surgery and the related risks. Patients also rated the system’s potential to reduce hypoglycemia and hyperglycemia as key clinical benefits and emphasized quality of life advantages, such as reduced burden (i.e., carrying fewer devices – we must say it’s hard to see this as a “major burden” in this world). On this latter point, Ms. Le Grand noted that “the thing we hear most from payers” is the need for “really strong evidence that drives results.” Payers are looking at “the bigger picture” and want to see peer-reviewed journal articles, improvement over the standard of care, and reductions in hospitalizations and emergency department visits. Ms. Le Grand believes Medicare and Medicaid are likely to eventually cover an implantable AP system, with type 1s getting clear preference over type 2s. We aren’t sure this kind of system can show dramatic improvement over traditional closed loop systems in development except for people challenged by extreme hypoglycemia – we aren’t sure what size of market this is but believe it’s small.

  • Picking up right where he left off at the JDRF/Helmsley Closed Loop Intraperitoneal (IP) workshop in 2017, Harvard University’s Dr. Eyal Dassau explained that following early feasibility evaluations of an implantable automated insulin delivery system in canines, a new PID controller was designed to enhance closed loop performance. Recently published data from a simulation show that the new PID controller developed from the canine outcomes resulted in a “sharp delivery similar to a bolus” following a large (90g carbs) meal. As compared to a PID controller from an earlier model, the canine-developed controller achieved significantly superior time-in-range (97% vs. 90%), tighter glycemic control (time 80-120 mg/dl was 73% vs. 55%) and time >180 mg/dl (3% vs. 10%). Next, Dr. Dassau compared the new PID controller between an IP insulin delivery-IP sensor system and an IP insulin delivery-subcutaneous system. In this simulation, both unannounced meals and exercise were included. The two systems produced similar time-in-range, but the IP-IP system generated “much tighter control,” with greater time spent between 80-120 mg/dl (77% vs. 67%). Both systems were superior as compared to a simulated IP-IP system using the previous algorithm. Dr. Dassau believes the data, published in Journal of Process Control, serve as a proof of concept for IP-IP closed loop, despite only being generated from a simulation.

Prof. John Pickup: Giving Pumps to All Type 1 Adults with A1c ≥8.5% Modeled to be Cost-Saving in UK After Five Years

In the heat of a very futuristic-sounding AID session, King’s College London’s Prof. John Pickup wondered aloud: “Can we afford the future?” While the audience held a collective breath, Prof. Pickup presented bits from a recently-published budget impact analysis showing that insulin pumps, at least, are affordable for the type 1 masses in the UK. In the paper, he and his colleagues model that bringing down the A1cs of all individuals who started with baseline ≥8.5% to ≤7% by any means would save the country £687 million (~$896 million) over the course of five years; that’s ~$7,300 per person! Those dollars, he said, are therefore equivalent to the money available or any effective, cost-neutral treatment strategy. [As a side note, this pan-population A1c reduction would reduce chronic complications by a factor of ~5.5 and DKA by a factor of 14.5.] Giving all of these individuals (type 1 adults with elevated A1c) pumps would thus become cost-saving after five years, according to the authors’ data. These results stemmed from a highly comprehensive budget impact model, where complications, severe hypoglycemia, and even lost productivity are accounted for. Sadly, this analysis didn’t examine CGM or AID, which we suspect would be even more cost-sparing.

Selected Questions and Answers

Q: Have you considered the number of new healthcare providers to handle the increase in new technologies?

Prof. Pickup: Yes, it would vary from one center to another. At least in the UK, we are forced to provide the same level of care with the same healthcare provider resources. I don’t think that’s actually a particular factor in the budget impact. Of course, it would depend on the diabetes technology and amount of extra education necessary. But for insulin pump therapy, we assumed it would be the exact same level of healthcare provider input as for MDI.

Dr. Anders Toft (Novo Nordisk CVP of Commercial Innovation): As we’re increasing the number of people wearing sensors, we’re increasingly aware of time-in-range. It’s a better measure of control than A1c … are NICE or other organizations in the UK, the home of cost-effectiveness calculations, looking at incorporating time-in-range in models?

Pickup: The answer to that is not yet. I think there’s probably, in NICE circles, more skepticism than there is in this audience. [Laughter] There’s more work to be done, as you are beginning to do, in linking time-in-range to risk of complications. One issue I think is the regulatory authorities…the other is the assertion that there’s some link between time-in-range and glycemic variability and I’m interested in what Dr. Bergenstal is saying about that. A lot of clarification needs to be done in that area, in assuming that time-in-range is as clinically meaningful in (a) complication risk and (b) what you do clinically about a poor time in range. There’s a lot of work to be done I think in clarifying that message for time-in-range. We’re not there yet.

Helmsley Funding Harvard Work on Glucose, Insulin, Lactate, Cortisol Sensing; iAPS System Overview; Six Women Enrolled in NIH-Funded Closed Loop Pregnancy Study

Dr. Dassau detailed a Helmsley-funded project to include additional sensors for biomarkers such as glucose, lactate, insulin, and cortisol in future closed loop systems. In collaboration with UCSD, Dr. Dassau is currently testing a proof of concept for an insulin sensor based on capillary glucose. We’re intrigued by the prospect of fine-tuning AID with additional inputs. Given the myriad factors (42!) that impact blood glucose, it seems logical that the next step is incorporating some of these components.

  • Dr. Dassau emphasized the need for human-centric design in closed loop therapy, reviewing the Interoperable Artificial Pancreas System (iAPS) smartphone app, capable of interfacing with CGMs, pumps, and algorithms while running on an unlocked smartphone. Results from a small in-clinic study (n=6 type 1 adults) published last month in DT&T found only “moderate improvement” in time-in-range, although Dr. Dassau was quick to note this was likely due to the extreme food and exercise challenges. Participants completed one week of sensor-augmented pump use followed by 48 hours of closed loop therapy with iAPS, Dexcom G5 CGM, and either a Tandem t:slim or Insulet Omnipod pump. Per the DT&T paper, patients can use iAPS to request an insulin bolus, log various activities, and view current glycemic status, including insulin-on-board. iAPS also provides alarms for system malfunctions and synchronizes data with a remote server for remote monitoring. Intriguingly, iAPS can interface with multiple algorithms, which could be used as party of a safety net for monitoring hypoglycemia, or updated throughout a study. Dr. Dassau mentioned that this proof of concept system for iAPS was then tested in a two-week outpatient randomized crossover trial, receiving “excellent feedback” from users.


  • As of January, six pregnant women with type 1 diabetes are enrolled in Dr. Dassau’s longitudinal, observation study of insulin requirements and sensor use during pregnancy. The cohort will be followed from before 16 weeks through six weeks postpartum. The NIH/NIDDK-funded study, first described at DTM 2018, will be led by Dr. Dassau and conducted by a consortium of investigators from Harvard University, Mayo Clinic, Icahn School of Medicine at Mount Sinai, and the Sansum Diabetes Research Institute. The aim is to develop and evaluate a pregnancy-specific artificial pancreas in a series of in-clinic and transitional environments followed by evaluation in an at-home clinical trial. We’re excited to see a focus on technology in pregnancy – despite very positive CONCEPTT results, CGM is still not considered standard of care during pregnancy, so closed loop therapy unfortunately seems a bit far down the line.

AID Poster Highlights


Important Details

AndroidAPS Hybrid Closed Loop in Home Setting is Safe and Leads to Better Metabolic Control

  • Type 1 children (n=13 males; average age 8.7 years) demonstrated significantly increased time-in-range (70-140 mg/dl) after six months on the OpenAPS DIY closed loop as compared to sensor augmented pump (SAP) therapy. Time-in-range increased from 57% to 64% (+1.7 hours).

  • Mean blood glucose decreased from 146 mg/dl to 133 mg/dl, and glycemic variability (SD) decreased from 56 mg/dl to 50 mg/dl.

  • Time in hypoglycemia (<70 mg/dl and <54 mg/dl) trended towards increasing, but the change was not significant. No episodes of severe hypoglycemia or DKA were observed. 

  • A1c was depicted as significantly decreasing in the chart, but specific values were not provided.

Closed-Loop Insulin Delivery to Manage Inpatient Nutrition Support: A Randomized Controlled Trial

  • Open-label, two-center randomized parallel study (n=43) investigating Cambridge University’s fully-automated closed-loop system using Fiasp vs. standard insulin therapy in non-critical care hospital patients receiving enteral/parenteral nutrition. Participants in the closed-loop group showed significantly higher time-in-range (100-180 mg/dl) than those in the control group (68% vs. 36%; +7.7 hours). This enormous increase once again shows the true value of closed-loop in the hospital setting – just like Cambridge’s previous work.

  • The closed-loop group spent less time in hyperglycemia (>180 mg/dl and >360 mg/dl) without increasing time in hypoglycemia. No severe hypoglycemia or hyperglycemia with ketonemia occurred in either group.

  • Average blood glucose was significantly lower in the closed-loop group as compared to the control: 153 mg/dl vs. a whopping 204 mg/dl. Glycemic variability (SD) was also lower in the closed-loop group: 41 mg/dl vs. 61 mg/dl.

  • The results suggest fully closed-loop therapy with Fiasp can significantly improve glucose management in patients receiving nutritional support in the hospital without increasing hypoglycemia risk.

CGM and SMBG Highlights

In the largest real-world CGM data set ever shared, an Abbott poster showed CGM metrics from nearly 500,000 FreeStyle Libre users across 26 countries (including the US). The data include a staggering 4.8 billion glucose readings collected from 470,463 readers (September 2014-May 2018). This gives an unprecedented global pulse on time-in-range and CGM outcomes in a sizeable fraction of FreeStyle Libre’s >1.3 million global users. Given that two-thirds of FreeStyle Libre users have type 1, it’s also a read on CGM outcomes to inform benchmarks – e.g., What are realistic goals? How do different technologies and studies compare to global time-in-range? We’ve created the table below to summarize the key stats, comparing CGM metrics for the median FreeStyle Libre user that scans 10 times per day vs. the lowest-scan group (4 scans/day) vs. the highest-scan group (40 scans/day). Our takeaways follow the table below, and pictures are below those.

N=470,643 readers

Median Libre User
10 Scans/Day

Lowest-Scan Users
4 Scans/Day

Highest-Scan Users
40 Scans/Day

Estimated A1c




(70-180 mg/dl)

13.5 hours/day

11.7 hours/day

16.9 hours/day

Time <54 mg/dl

34 minutes/day

34 minutes/day

24 minutes/day

Time >240

4 hours/day

6 hours/day

2.2 hours/day

  • What are CGM metrics in the median FreeStyle Libre user? A time-in-range of 56% (70-180), 34 minutes per day (2%) spent <54 mg/dl, and a staggering four hours per day (17%) spent >240 mg/dl. Even with a 10 scans per day, this shows a still-dangerous amount of time spent at extremely high and low glucose values – 4.5 hours per day spent <54 mg/dl or >240 mg/dl. And this is only the median – i.e., 50% of the FreeStyle Libre population has outcomes worse than these.

    • Implication: Continuous glucose data alone is not enough – better education on what to do with the data (food, mindset, exercise, sleep) and insulin automation are desperately needed to reduce/eliminate extreme glucose values.

  • Guidance on time-in-range goals? In this enormous global database, the lowest scanning group had a time-in-range of 48% vs. 70% in the highest scan group. Overall, 63% of FreeStyle Libre users had a time-in-range over 50% (12 hours per day). The poster does not allow for further extrapolation – e.g., what percentage of users have a time-in-range above 70% – the proposed goal? It does mention that in users with less than 10 scans/day, only half of them had >50% time-in-range. In those taking more than 10 scans/day, a majority (73%) had >50% time-in-range. (For those wondering about the stats here, the data have been grouped by scan frequency first, and then the mean is found for each scan-frequency group. This means a separate analysis will be needed to answer the “How many have time-in-range >70%?” question.)

    • Implication: A majority of FreeStyle Libre users have a time-in-range >50%, though a sizeable 37% are spending more than half the day outside of 70-180 mg/dl.  

  • Same technology, huge variability – data shows a major 1.5% eA1c difference between the lowest and highest scanning groups (8.2% vs. 6.7%), a 3.4 hour/day difference in time-in-range (56% vs. 70%), and nearly two fewer hours per day spent over 240 mg/dl (17% vs. 9%). Scanning frequency also has huge variability – a 10x daily difference between the lowest and highest groups in this data.

    • Implication: How can we shrink disparities in time-in-range? What else can be learned from Bright Spots – those with >65% time-in-range? Beyond scanning glucose more, what is the high-scanning group doing differently?

  • Even those with an on-target eA1c of 6.7% had a lot of extreme values – two full hours per day over 240 mg/dl and 2% of the day <54 mg/dl.

    • Implication: average glucose is wildly inadequate to characterize dangerous glucose extremes.

  • The difference in time <54 mg/dl (10 minutes/day) was small between the lowest and highest-scanning groups – will FreeStyle Libre 2 with alarms bring the time <54 closer to zero in both groups?

  • Higher glucose levels do not protect against hypoglycemia. Time <54 mg/dl was higher in groups with the highest estimated A1c and lowest in those with the lowest estimated A1c. In other words, running high all the time is not a strategy to avoid lows. 

  • For context, this expanded data set is 9x bigger than the one shared at ATTD 2017, and nearly twice as big as last year. It also shows Abbott’s leadership position on real-world Big Data in CGM – at least in terms of patient base.

    • We hope to see Abbott move to the hypoglycemia and hyperglycemia consensus definitions: “<54” and “>250,” rather than the <54 and >240. Fortunately, the critical 70-180 for time-in-range is being used!


Dexcom Pipeline Updates: “Hey Siri, What’s my Glucose?” First G6 Pro picture (fully disposable transmitter, real-time/blinded); G7 by end of 2020

Dexcom CTO Jake Leach provided a pipeline update, including five upcoming G6 app enhancements (e.g., “Hey Siri, What’s my glucose?”); the first pictures and product details on the G6 Professional (fully disposable, blinded or real-time); confirmation that G7 will launch by the end of next year in collaboration with Verily (full launch in 2021); a new “On the Bright Side” notification for Dexcom Clarity (Bright Spots coming to CGM!); and news that smart pens will send insulin data directly into the G6 app. Details and pictures are below!

  • Dexcom has five near-term enhancements planned for the G6 app: (i) a Siri Shortcuts feature – “Hey Siri, What’s my glucose?” “Your glucose is 110 mg/dl and flat” (nice convenience!); (ii) adding up to 10 followers on Dexcom Share; (iii) the ability to launch the Dexcom Clarity data management app from within the separate G6 app; (iv) a new G6 watch face complication for Apple Watch Series 4; (v) a 24-hour reminder before sensor expiration; and the ability for Android users to share data with Google Fit (similar to Apple Health). Direct-to-Apple Watch remains in development, and Mr. Leach admitted it has “taken a while” because Dexcom “really wants to get the user experience right.” No timing was shared, though CEO Kevin Sayer previously said to expect a launch this year.

  • Following FDA clearance in announced in 3Q18, we finally got to see the new “Dexcom G6 Pro” (professional version) – it includes a fully disposable G6 transmitter and 10-day sensor (factory calibrated), an ability to run in blinded or real-time mode, a reader to download the data in clinic after 10 days, and the ability for users to get real-time CGM data through the same G6 app. (Based on the Pro transmitter, the G6 will run in a simplified mode.) As of JPM, G6 Pro is expected to launch sometime this year. This is a huge upgrade over Dexcom’s current professional (G4) CGM, bringing it mostly on par with Abbott’s FreeStyle Libre Pro in terms of simplicity – a fully disposable transmitter and no calibration were the critical moves. G6 Pro is ahead of Libre Pro with the ability for real-time data, though behind on wear time (10 days vs. 14 days). We wonder if G6 Pro will allow for more “CGM sampling” – giving many Pro sensors to HCPs to encourage people to trial real-time CGM for 10 days. G6 Pro should also fit nicely with virtual clinics like Onduo that use CGM “sprints” (episodic wear) in type 2 diabetes. The sensor is the same as the current G6, a plus for manufacturing and accuracy.

  • Mr. Leach confirmed that G7 – the Dexcom/Verily Gen 2 product – will launch starting at the end of next year (2020), with a broader launch “throughout the world” in 2021. Fully disposable electronics, a “significant cost reduction,” a much smaller profile, no-calibration, real-time app transmission via Bluetooth, and extended wear (14 or 15 days, per JPM) are still the plan for G7. All of this is identical since last month and Dexcom’s December Investor Meeting – great to see consistency, since the goal posts on this product have moved. Gen 1 Verily was not mentioned, as expected following JPM. Mr. Leach specifically said G7 will be more suited for clinical impact in new markets (high performance, reliability, low cost, disposable): hospital prediabetes, gestational diabetes, type 2 diabetes on oral medications only (perfect for wearing for a period of time to titrate drugs, verify correct doses), obesity (behavior modification, how different foods impact the body).

  • Dexcom will soon launch new features in Clarity, including an “On the Bright Side” notification – automatically identifying days where users achieve their goals and receive a notification. We’re elated to see this Bright Spot pattern recognition coming to CGM, something we have advocated for over the past two years. Mr. Leach showed the same screens of the MDI decision support with TypeZero – including on-demand sleep and exercise dosing advice and a CGM-informed bolus calculator – but did not share timing. See Dr. Marc Breton’s presentation below for an update on that study.

  • For the first time, Mr. Leach said smart pens will send insulin data directly into the G6 app and onto Clarity – a huge simplicity win. We might have expected Dexcom to pull the insulin data from other apps (e.g., from the smart pen app itself), but this is excellent news for combining the two data sets on the phone in one app from the start. The talk showed both Lilly and Novo Nordisk as smart pen partners.

Dr. Bergenstal Adamant That We Need to Make CGM Access More Equitable if We Want to Flatten Population A1c Curve; Wishes for T2D Decision Support

IDC’s Dr. Rich Bergenstal wrote at this time last year, “CGM: Transforming diabetes management step by step!” The recent very concerning T1D Exchange data – in which CGM penetration increased but number of aggregate A1c statistics worsened –has led him to question whether the “!” at the end of his former statement should be amended to a “?”, and to rethink effective implementation of CGM. Indeed, he believes that there is more the system can do to support patients at every stage of CGM use: Starting CGM, wearing CGM, using CGM, and effectively using CGM. His strongest point came with regards to starting CGM – “It’s not just giving CGM to those with A1cs of 7.2% and getting them down to 6.9%. We won’t see a flattening of the curve if we only do that. We need to get to patients with A1cs of 9%, 10%, and 11% if we want to actually flatten the curve. We need to be more equitable, regardless of A1c, ability to pay, race, or ethnicity. It’ll take more work on our part but we’ve got to make it happen if we’re going to flatten that curve.” His following bullets were all the more important when taking these additional populations into account. On wearing CGM, he advocated for greater support for patients, down to the granularity of helping individuals determine which adhesive and sensor location works best for them (we believe the Cecilia Health-Jaeb Center-Helmsley pilot holds promise here). On using CGM, he encouraged clinicians to support patients in looking at their sensor data, including setting goals for time-in-range, hypoglycemia reduction, and (maybe) A1c. This bucket also includes having patients share data with family, if they are comfortable doing so. Finally, he hearkened back to the CGM-adaptation of Daniel Kahneman’s work (Thinking, Fast and Slow) that he unveiled at ENDO 2018. Read the ENDO report for more details, but in brief, “thinking fast” refers to real-time corrective action and use of trend arrows, and “thinking slow” refers to retrospective analysis, pattern control, and AGP – both are necessary for optimal CGM use, he argues.

  • Dr. Bergenstal outlined his vision for useful decision support in type 2 diabetes. In type 2 diabetes, he pointed to the recent ADA/EASD treatment algorithm, claiming that it’s useful in that it tells the clinician what to do, but doesn’t support them in following the recommendations. He raised an interesting paradox: The algorithm is truly patient-centered, offering granular guidance for all sorts of presentations of type 2 diabetes, but that quality also makes it inherently more complex and difficult to adhere to. “I would like for a decision support tool in my EMR to tell me that based on a bunch of criteria, my patient should be on a GLP-1, and tell me which GLP-1 is covered for that patient, and what would be a reasonable starting dose. That would help many PCPs try to follow an algorithm that sounds simple but is very complicated in the office.” Beyond automated decision support, he added, “It also would be bad to have a nurse who could train on injection and handle follow-up with a phone call to monitor side effects or effectiveness.

  • Dr. Atul Gawande’s (CEO of new Amazon-Berkshire Hathaway-JPM healthcare venture) recent New Yorker article, “Why Doctors Hate Their Computers,” resonated with Dr. Bergenstal (and countless other physicians). Pulling language from the article, he posited that we need decision support systems in diabetes that make the right care simpler and not more complicated while maintaining the human connection.

Median 35% TIR, 60% >180 mg/dl Seen in 153 T1s Ages 14-25 without Recent CGM Use (Baseline Data from Helmsley-Funded CITY Study); Full Results to Report at ADA 2019

Joslin’s Dr. Lori Laffel presented disheartening baseline glycemic data from the Helmsley-funded CITY study of CGM in young type 1s ages 14-25 years old (all either CGM-naïve or not active users in the past three months). The 153 participants are wearing blinded Dexcom G4 (with software 505) for two weeks, followed by six months of CGM. The sad story here is the baseline:

  • Mean A1c is 8.9%, and 25% of participants had A1cs above 9.6%.

  • Median time-in-range (70-180 mg/dl) is just 35% (8.5 hours/day). Half of the sample spends more than 15.5 hours per day out of range, and only 17% of participants spent ≥50% time-in-range.

  • Median time >180 mg/dl is 60% (14.5 hours/day)! Half of the group spends ≥7.9 hours per day above 250 mg/dl.

  • Only 21% of participants had “stable” glycemic variability as determined by CV ≤36%.

  • Median time <70 mg/dl (3%) and <54 mg/dl (1%) are in line with recommendations, but only because hyperglycemia is so pronounced.

Obviously, there is a lot of room for improvement in this cohort, adding fuel to the fire shared in the latest T1D Exchange Registry data. CITY will read out at ADA 2019, and our fingers are tightly crossed for a positive outcome – the study population is pretty much exactly the hump in the canonical T1D Exchange graph of A1c by age that needs the most help. But as the Registry shows, improving time-in-range and A1c is not as easy as just giving more people CGM; as Dr. Bergenstal explained above, patients have to be supported in effectively using CGM.

  • Dr. Laffel noted that the exclusion criteria of current/recent CGM use resulted in a very heterogeneous population where 38% are of minority race/ethnicity, and 41% are not privately insured. Interestingly, however, there were no demographic or clinical factors associated with baseline CGM metrics. Said Dr. Laffel, “This is a population with uncontrolled diabetes, and we’re not finding causal factors easily – there must be behavioral or psychosocial factors we’ve yet to measure. This work is so important on so many levels.

Prof. Lutz Heinemann on Diabetes Tech in 2025: Lower Cost CGM (<$2-$3/day), More CGM in Type 2, Telemedicine, Decision Support; ~3,000 Patients on DIY AID Systems Globally

Science Consulting Neuss CEO Dr. Lutz Heinemann kicked off ATTD 2019 in fantastic fashion, headlined by his predictions for diabetes tech in 2025:

  • For type 1, he anticipates: (i) the availability of cheap and reliable CGM with an <8% MARD and ~<$3/day; (ii) “affordable” AID systems at ~$11/day; (iii) “doc on demand” providers via telemedicine as part of patients’ standard care; and (iv) most therapeutic aspects of care facilitated by data download to the cloud and clinical decision support systems (CDSS). No surprises there, though we’d be interested to know the assumptions behind AID at $11 per day (~$330 per month, ~$4,000 per year) – if that estimate includes CGM and insulin, it’s a massive step down in costs relative to current systems.

  • For type 2, Dr. Heinemann expects: (i) SMBG to be widely replaced by CGM with <10% MARD and ~<$2/day (implying a different, lower-cost, slightly less accurate product than for type 1); (ii) increasing uptake of once-weekly basal insulins; and (iii) wide use of clinical decision support systems.

  • Supporting these claims, Dr. Heinemann gave several take-home trends:

    • Declining numbers of endocrinologists and diabetologists are at odds with the rising diabetes epidemic, necessitating the transfer of responsibility to machines and algorithms. 80% of diabetes-related procedures are routine tasks, which clinical decision support systems could primarily undertake, allowing physicians to add their own recommendations informed by their relationships with patients and patient circumstances. Notably, algorithms could also be used to detect the ~one-fifth of patients that require direct and fuller attention of a physician. As Dr. Heinemann pointed out, these algorithms aren’t necessarily intended to enable the treatment of more patients within the same period of time, but rather to empower providers with more time to talk to each patient. Given the trends in numbers of providers, this is still a concern – who is going to take care of everyone with diabetes?

    • While better devices have pushed major success in the ~$3 billion CGM market, Dr. Heinemann characterizes current CGM systems as still “miles off from ideal.” To this end, he labelled cost, sensor wear time, pain, and skin reactions as major areas of improvement, as well as the establishment of an accuracy standard such as that seen already with SMBG systems. (iCGM is sort of the de facto accuracy standard now, as we assume all companies will move to that over time given the regulatory and speed advantages. Due to arrows, point accuracy and data density leaves some room  for CGM to be less accurate than BGM.) Looking to the future, Dr. Heinemann was particularly excited about the Dexcom and Verily “G7,” slated to launch in late 2020, followed by a full rollout in 2021 (per JPM). However, he was skeptical that the project will stick to this timeline to launch it next year (he did not mention the broader rollout in 2021). At JPM, CEO Mr. Kevin Sayer shared that Dexcom expects to “finalize” G7 this year and provide “very clear timing” on the plan. With the amended Verily collaboration agreement and consistency between 3Q18 and the Investor Day, we think things are feeling clearer on this next major pipeline launch. Dexcom often gets products approved faster than expected, though given the massive organizational disruption for G6, there is a lot to get ready for G7. 

  • Dr. Heinemann asserted that ~3,000 people globally are currently using DIY insulin systems (Loop, OpenAPS, or Android APS). We’ve previously heard an estimate of ~2,000 Loop users, and this sounds in the ballpark taking into account all three systems. It also shows the great desire for more advanced and personalized AID; as Dr. Heinemann put it, “people with diabetes won’t wait for innovation.” That said, the DIY movement has also started to make regulatory waves, with Tidepool and Jaeb beginning an observational, virtual study of Loop last month to support FDA submission. (Adam is in this study and just mailed back the A1c kit – very cool!)

  • Dr. Heinemann also pointed to the environmental waste associated with diabetes products, showing a full trash bag from diabetes product waste accumulated in just two weeks. To this end, he highlighted that each Dexcom G6 inserter creates 80 g of plastic waste and each Abbott FreeStyle Libre sensor 71 g. For the latter, this equates to over 1 billion pounds of plastic created each year (!) – assuming 1.3 million users inserting two sensors per month. Further data from Dr. Heinemann on this topic are currently under review for the Journal of Diabetes Science and Technology.

  • What about exciting therapy innovation? Dr. Heinemann specifically pointed to glucose-responsive, or smart, insulin (GRI) – while he doesn’t believe a product will reach the market soon, he does anticipate that GRI may pose a threat to diabetes technologies such as a fully closed-loop. We think the prospects of GRI are a bit long-term (10+ years out) to make a judgment call at this point. After the first-ever clinical candidate was discontinued by Merck due to inadequate phase 1 results, GRI has experienced somewhat of a renaissance with investment from JDRF and Novo NordiskSanofi has also made some major investments. Still, development of an effective GRI does comes with wide-ranging technical challenges that are currently unsolved, as was highlighted in a 2016 JDRF/Helmsley Charitable Trust gathering on smart insulin. Perhaps most importantly, a molecule must be programmed sensitively toward a narrow and physiologically appropriate glucose range, which is no small feat; cost and insulin receptor kinetics are also key. We’d bet on no technology emerging for at least the next decade.

  • Digitalization will continue to change our world dramatically. According to Rock Health, investment in digital health surpassed $8 billion in 2018, and with major companies such as Google and Apple beginning to venture even further, diabetes technology is ripe for disruption (as it has been for many years). For these companies, according to Dr. Heinemann, digital health is a market of the future.

Real-World FreeStyle Libre Data: Published on ~50,000 users; Majority of Reductions in Hyperglycemia Achieved within First Two Months; Highest Libre Scanners Achieve 40% Greater Reduction in Time ≤54 mg/dl vs. Lowest Scanners

University of Leeds’ Dr. Ramzi Ajjan shared a slew of published and unpublished real-world data on FreeStyle Libre, continuing a trend (and much of the same data) from ATTD 2018 and ATTD 2017. It’s great to see Abbott churning out real-world data publications from the largest CGM user base in the world – all of it emphasizes that more frequent glucose data leads to better outcomes.

  • Real-world FreeStyle Libre data – first shown at ATTD two years ago – has finally been published in Diabetes Research and Clinical Practice from 50,831 readers (i.e., users) and 279,446 sensors, totaling 63.8 million glucose readings. This was notably smaller than the 250,000-user data shared at ATTD 2018 and the ~500,000-user data set shown above; still this is a big data set that is great to get in print. The average user scanned 16 times per day, with those in the highest scanning group scanning 48.1 times/day (three per waking hour!) vs. the lowest scanning group performing just 4.4 scans/day. Scanning frequency was significantly associated with lower time ≤54 mg/dl, lower estimated A1c, and higher time-in-range. Those in the highest scanning group achieved a 40% decrease in time ≤54 mg/dl (26 minutes vs. 43 minutes) and a 1.3 percentage point decrease in estimated A1c (6.7% vs. 8.0%) as compared to those in the lowest scanning group. Per the paper, time-in-range increased from 50% to 70% in the low vs. high scanning groups – a staggering 4.8-hour/day difference. These patterns were consistent across countries too.

  • Unpublished real-world FreeStyle Libre data showed that the majority of reductions in time in hyperglycemia (>240 mg/dl and >180 mg/dl) occur gradually over the first two months of sensor wear (first picture below). Sample size was not included. By contrast, the majority of reductions in hypoglycemia are achieved within the first two days of FreeStyle Libre wear (n=14,617 users; second picture an shown at ATTD 2018). This trend for hypoglycemia was consistent at glucose <70 mg/dl and <55 mg/dl, but especially at glucose <45 mg/dl, for which 74% of the total reduction was observed within the first two days. That said, users in this data set were still spending 82 minutes per day under 70 mg/dl; quite a lot and we wonder how that improves over longer duration wearing the sensor. Still – we’ve heard many KOLs suggest less than 4-5% time less than 70, which itself is still about an hour.

  • Unpublished FreeStyle Libre real-world data also indicated that scanning frequency sees an initial decline and then stabilizes around the second or third sensor. From the graph below, it appears that users scan roughly 15 times/day, nearly one per waking hour. Sample size was not included, so we’re not sure if it’s from the larger 250,000+ data set. As expected with FreeStyle Libre’s factory calibration, SMBG drops to <1 fingerstick/day and remains there – a good sign of confidence in the accuracy and reliability.

  • Real-world FreeStyle Libre data (n=4,793) over six months (12 sensors) demonstrated a similar reduction in time <70 mg/dl as compared to that observed during the IMPACT trial (n=119). It’s encouraging to see such strong correspondence between clinical trial and real-world data – something Medtronic has also shown with the MiniMed 670G. Again, we’d note that in either data set, users were still spending ~2 hours per day (~10%) below 70 mg/dl – far too much. This was also shown at ATTD 2018.

Year-Long Medtronic iPro 2 Professional CGM Study Results in Mean 1.3% A1c Reduction, Reduced Hyperglycemia and Variability with No Increase in Hypo; Interesting Therapy Adjustment Charts

Following the interim readout of eight-month data a year ago, Medtronic’s Dr. Bob Vigersky presented full 12-month results from the single-arm Portuguese ADJUST study of the iPro 2 professional (blinded) CGM in 102 people with type 2 diabetes. At 12 months, following quarterly CGM applications, each with a follow-up visit (in-person or by phone), A1cs had dropped by a mean of 1.3% (baseline: 9.4%). The greatest A1c reduction (-1.0%) unsurprisingly occurred after the first CGM application, and by study’s end, 81% had seen a drop of ≥0.5% (6% saw their A1cs increase). Mean glucose (185 -> 170 mg/dl), standard deviation (53 -> 49 mg/dl), and percent time >180 mg/dl (48% -> 37%) all improved without a concomitant increase in time <70 mg/dl or the number of hypoglycemia episodes (both low at baseline). Qualitatively, patients reported increased treatment satisfaction, less frequent detection of hyperglycemic symptoms, and no change in perception of hypoglycemia; there were also reports of improved communication between patient and healthcare team and a need for specific professional training on professional CGM protocols. Dr. Vigersky showed an interesting graphic overviewing the types of adjustments providers were making and at what times (below). The pie charts show how therapeutic changes became less common over time (top) as people’s profiles theoretically stabilized, and the types of recommendations providers made over time. Most participants saw tremendous benefit from participating in the study, and we hope to see more validations of the power of intermittent CGM in type 2s to drive positive therapeutic change (and reduce inertia!). 

  • These results were obtained in a diverse type 2 diabetes population with 53% unemployment (31% were retired, 11% unemployed) and in which ~27% hadn’t progressed in their studies past middle school. We’d be interested to see this population followed over time – following a year of intermittent blinded CGM-guided therapeutic and behavioral changes, what happens to A1c and self-management practices? Do they revert back to baseline? What is the optimal frequency of professional CGM moving forward for most people to maintain their ADJUST gains? Would results be stronger with real-time, intermittent CGM?

>3,000 Onduo Participants’ Mean Glucoses Before + After CGM (G5) in BCBS Pilot; Only Given to People Demonstrating ≥7 Fingersticks/Day

Onduo Medical Director Dr. Amit Majithia presented preliminary data from type 2 patients who received CGM as part of Onduo’s Blue Cross Blue Shield pilot conducted in three states. The plot below depicts the distribution of “well over 3,000” participants’ average blood glucose in the 14 days prior to CGM minus the 14 days after CGM. Negative numbers indicate that average blood glucose increased following CGM, while positive numbers reflect a decline. As demonstrated below, the majority of individuals hovered around a neutral effect, with the average suggesting some effect – though a sizeable portion saw an increase in mean blood glucose. This was an odd, obfuscating way to present the data, and it would have been far more instructive to see time-in-range, GMI, and time in extreme glucose ranges in this group. Dr. Majithia justified the surprising increase in that Onduo encouraged people to “go crazy” with their CGM initially, treating it as an “exploratory time.” (We assume this means testing the limits of what they can eat, etc.) Dr. Majithia added that for some participants, receiving CGM resulted in an A1c benefit equivalent to that achieved with a pharmacologic therapy – e.g., the trace below from one individual shows a 1.5% A1c reduction in five months (baseline A1c: 8.9%).

  • CGM use was shown to increase app activity. Dr. Majithia showed data reflecting a spike in activity (i.e., meal, medication, and BGM logging, messaging) during the first seven days of CGM. While activity declined a bit during the second week of CGM use, activity reverted to baseline when CGM was discontinued. It’s exciting to see that CGM data may drive further interest in diabetes management.

  • The funnel plot below depicts the drop-off rates for each stage in the CGM acquisition process. About half of participants who logged in had a CGM strategy initiated, consisting of coach education and evaluation of willingness to use. If participants completed the subsequent survey, “most” completed the remainder of the protocol, which includes a “provider visit,” CGM shipment, and CGM worn for at least three weeks. The provider visit could either be a face-to-face telemedicine video chat or an asynchronous review by a physician. One of the pilot states, Arkansas, required a face-to-face interaction prior to shipping the CGM. Interestingly, Dr. Majithia found that less people completed the survey if a face-to-face visit was required; however, these patients were more likely to wear the CGM than if they underwent an asynchronous review. Asynchronous review still ultimately drove ~10% more CGM wear, given the high survey drop-off rate in those who were required to undergo a face-to-face interaction. It’s possible that participants value CGM more after devoting the additional time to the acquisition process. In fact, Dr. Majithia was surprised that overall ~15% of participants who received CGM did not initiate the sensor after three weeks. He believes that because the participants received the device easily and at no cost, they lacked the same urgency than a patient who had to go to a physician’s office might experience. They might also look at the G5 inserter and be deterred, whereas the G6 might be more attractive.

  • Onduo participants were eligible to receive a CGM if they were able to demonstrate 7-14 fingersticks/day. While Dr. Majithia considered this requirement “pretty permissive,” we think this is actually too aggressive – it basically limits CGM use to those already pretty engaged, limiting the population and ability to show benefit. (For context, the median FreeStyle Libre user, as shown above, scans 10 times per day! This is a ludicrous number of fingersticks.) Still, as he pointed out, Onduo was using Dexcom G5, which requires two calibration fingersticks per day. We know it’s early, but if we were designing the intervention, we wouldn’t mandate fingersticks for enrollees to try CGM at least once – everyone can learn something from seeing their glucose values in real-time for a week, especially with a system like G6 or FreeStyle Libre that doesn’t require fingersticks.   

Real-World EU Eversense XL Time-In-Range and Sensor Wear Data by Country; Six-Month Outcomes in N=21 Italian Sample Show ~4 Hour/Day Increase in Time-In-Range

Dr. Concetta Irace (Magna Græcia University of Catanzaro) shared real-world Eversense XL data from the EU registry showing percent time-in-range (70-180 mg/dl) collected from patients across 11 countries as of September 2018 (sample size not provided). Percent time-in-range varied from 69% in Austria (~16.6 hours/day) down to 53% (~12.7 hours/day) in Switzerland. Dr. Irace also presented sensor wear time (percent of time spent wearing the transmitter) data ranging from 92% in Denmark to 89% in the Netherlands; both come out to about ~21-22 hours per day, which is strong but which we’d like to see even higher.

  • Dr. Irace also presented real-world data from a small sample (n=21) of Eversense XL users in Italy showing significant improvements in time-in-range, A1c, and glycemic variability after the first six months of wear. Time in a narrower range (70-160 mg/dl) increased from 43% to 58% (+3.6 hours) and A1c declined by 0.6% from a low baseline of 7.3%. Glucose standard deviation (SD) also improved by a solid 9 mg/dl, decreasing from 59 mg/dl to 50 mg/dl. Dr. Irace was particularly excited by the glycemic variability results, as they were not included in the pivotal trial.

  • A separate survey of 767 Italian Eversense XL users found high patient satisfaction. 87% of respondents reported feeling “very satisfied” or “satisfied” with the overall experience. An overwhelming 94% of respondents indicated that the procedure was “totally painless.”

Helmsley-Funded Jaeb+Cecelia Health Study Investigates Direct-To-Consumer CGM In Type 1 Adults; Participants Choose FreeStyle Libre or Dexcom G6 Plus CDE Support

Dr. Roy Beck shared that a Helmsley-funded study conducted by the Jaeb Center in collaboration with Cecelia Health (formerly Fit4D) and Wisconsin Research and Education Network (WREN – a clinical research network of primary care providers in Wisconsin) will investigate direct-to-consumer CGM uptake primarily in type 1 diabetes. The primary outcome is sustained use of CGM when initiated outside the clinic, with a goal of 6-7 days/week wear time. The study has started as of February and will also examine outcomes including: (i) occurrence of severe hypoglycemia, DKA, and hospitalizations; (ii) A1c measured at three months; and (iii) CGM metrics at three months. Participants must be adults (≥18 years-old) with type 1 diabetes or type 2 diabetes using basal-bolus injection therapy (pump or MDI) and cannot have used real-time CGM within the past two years. Participants can choose between the Dexcom G6 or Abbott FreeStyle Libre and are provided with information and guidance via a video conference with a CDE. Dr. Beck described an intentionally gradual, staged onboarding process to mitigate patients feeling overwhelmed. Participants are sent their CGMs in the mail and undergo an onboarding session with a CDE. After two weeks, CDEs introduce visualization tools and instruct participants on how to use CGM data for self-management. Two weeks later, CDEs provide troubleshooting tips and tricks. Throughout the whole study, participants will have access to CDEs. Dr. Beck estimates ~30 participants have already enrolled and the study population is fairly split in their decision to choose either the G6 or FreeStyle Libre. The study will also collect information on the factors that lead to participants’ selection. As the study expands, Dr. Beck expects to incorporate decision support tools and mental health support/coaching. We are very excited about this study, as it may pave a path towards non-prescription CGM. As Dr. Beck put it: “CGM is getting to the point where, like BGM, it shouldn’t require a prescription.” Dr. Beck anticipates results to be presented at ATTD 2020.

LifeScan Symposium: Physician Focus Groups on Future of SMBG; Bastian Hauck Delivers a Patient’s Perspective on SMBG; Prof. Barnard’s Sage Advice for Manufacturers

In a LifeScan-sponsored symposium, UCSD’s Dr. Steve Edelman, Prof. Katharine Barnard, and dedoc’s Mr. Bastian Hauck overviewed the future of BGM: In short, it’s not going anywhere for most patients, and the future of fingersticks has to be based in consumer-centric design, seamless connectivity, and decision support. (That sounds a lot like CGM!)

  • Dr. Edelman kicked off the presentations with a summary of findings from TCOYD-, CWD-, and ISPAD-sponsored focus groups that took part at ATTD and ADA 2018; participating clinicians (many with diabetes) were asked to ruminate on the future role of SMBG. Dr. Edelman hopes to publish these findings soon. Unsurprisingly, the clinicians expect that BGM will remain in high demand for individuals who can’t access CGM in the US, many type 2s who won’t need to rely on CGM 24/7, as well as the majority of the world – particularly countries like China and India – for many years to come. When asked what type of evidence they’d look for to justify the usage and reimbursement of connected BGM, respondents underscored a desire for real-world, pragmatic data, changes in patient behaviors/attitudes, as well as a slew of clinical/economic outcomes (A1c improvements reductions in hypoglycemia and/or ER visits/hospitalizations, and improvements in time-in-range). The clinicians emphasized a need for decision support tools, structured testing vs. “worthless testing,” highly usable technology for patients and healthcare providers, as well as adding minimal burden in terms of healthcare team time and staff. On this last point, Dr. Edelman commented, “A big mistake a lot of companies make is assuming physicians have a lot of time to learn new things.” Provider have less time than ever – alongside rising burnout – and the big hope is for clinical decision support tools to simplify visits. Alongside that is remote/virtual care “clinics” and coaches, which might be able to fill the gaps.

  • Mr. Hauck, speaking only from notes and without slides, rattled off a list of desires for BGM in the era of CGM.

    • “Make blood glucose meters sexy. I want one I really like to carry around…. If you were producing cellphones the way you produce BGMs right now, everyone in here would be out of business, because I wouldn’t buy it…design matters.” At the outset of his talk, he recalled a 2012 talk in which he warned manufacturers in attendance, “If you don’t watch out, if you don’t design your next BGM with consumers in mind, my next BGM will be my iPhone.”

    • “For me it’s really about open data; data that goes flawlessly, fluently, seamlessly from users to payers, to providers, and back. From hardware, software, to the cloud and back … I have an appointment with my diabetologist every three months here. That’s not because I need to see her every three months – maybe I do in two months, or not for next six months. I only go to get insulin. It’d be much smarter to have a pattern recognition algorithm that calls me when an algorithm detects something is not quite right with my blood glucoses.”

    • He wondered why we don’t have automatic bolus calculators for MDI, whereas we do for pumps. Of course, Companion Medical’s InPen is available in the US and CE marked, with a planned 2019 launch, but only ~2,000 people are using InPen in the US. It’s still early days for MDI dose capture, though Lilly’s and Novo Nordisk’s entries in the near-term could change that in a hurry.

    • “EHR; I don’t understand why it’s not working. All of this data should be in the EHR, along with all kinds of other data.”

    • “The interface of tomorrow will not be an app, something you click on, look at and push buttons. It’ll be something between cloud and messenger, Siri and Alexa…maybe a  reminder on the fridge, on the TV when I sit down to watch a movie, or maybe on my alarm clock if it’s something I need to do when going to bed. The great thing about open data is it’s kind of everywhere. The idea of a ‘Homespital’ – a mix of being in the home and in hospital all the time, 24/7. That’s the perfect integrated solution for diabetes management.”

    • “I don’t see why we couldn’t do something like trends/curves/alarms/reminders in something as simple as BGM. It won’t be as good as CGM, but it could be better than what we have now.”

  • Prof. Barnard reiterated many of Dr. Edelman and Mr. Hauck’s points, and added a few particularly poignant musings. For one, she noted the absurdity in the fact that when we have a new tool that promises to reduce burden for patients, “we make them jump through all kinds of hoops to get it…it’s a bit weird. We shouldn’t be making people jump through hoops to take care of their healthcare needs.” Prof. Barnard also advised manufacturers that “one size fits nobody very well, but a single device with a broad range of functionality fits many, many people.” (FreeStyle Libre is probably the best recent example of the latter.)

  • Head of Portfolio Strategy Mr. David DeJonghe and Clinical Affairs Director Dr. Mike Grady doubled down on their presentations from the EASD symposium, highlighting the CE-marked One Touch Verio Reflect (with the Blood Sugar Mentor; launched in France and Germany) and recent data supporting One Touch strip accuracy and outcomes data. To his talk, Mr. DeJonghe added that the One Touch Reveal app is approaching two million downloads. The Mentor meter’s messages are basic pattern recognition, but they are far better than most people get (since most don’t download or look at data). For example, if the meter finds that a user has been high each of the last four days in the evening, it will notify her and her healthcare provider, and ask her if anything has changed (i.e., eating high-carb lunches, eating late lunches, large/unnecessary mid-afternoon snacks, less activity than before, etc.). If it detects hyperglycemia, it might suggest a walk, and if it detects a low or near-low, it will recommend juice and offer to set a reminder to re-check blood glucose in 15 minutes.

  • According to Dr. Grady, LifeScan makes five billion strips per year – ~33% of the world’s annual production!

Dr. Roy Beck Urges Inclusion of Intermittent CGM in Future CVOTs

Acknowledging that it would probably be unethical to repeat the DCCT as a randomized controlled trial with CGM, Jaeb’s Dr. Roy Beck urged pharmaceutical manufacturers to use CGM in future CVOTs: “It would’ve been great if CVOTs had had CGM, but that was not done. It’s never too late! Any of you embarking on large CVOTs, there’d be incredible value in having blinded CGM in them at intervals.” Though he didn’t expressly volunteer, we imagine he and his Jaeb colleagues would be more than happy to assist (or at minimum, advise) with the administration and data aspects of such an ambitious venture. We point out that most CVOTs started a long time ago, when current CGMs were not possible to use. He pointed to the PERL RCT of allopurinol to reduce the progression of kidney disease, which uses CGM and is expected to complete this summer. In that study, investigators are deploying blinded CGM (Libre Pro) “every three or six” months over the course of the study, in order to determine how well CGM metrics, namely time-in-range, predicts progression of kidney disease. This will be great to see. With the upcoming 2019 launch of G6 Pro, there is another disposable device well-suited for clinical trial monitoring. We wonder if there’s a role for other funding bodies – NIH? Helmsley? – to provide CGM for more of these studies – and the brainpower to analyze and synthesize the data?

  • Through the majority of his talk, Dr. Beck reviewed evidence that: (i) time-in-range is largely a measure of hyperglycemia; (ii) ~14 days of CGM are needed in order for derived metrics to approximate those compiled out to 90 days; (iii) 30%, 50%, and 70% times-in-range roughly equate to A1cs of 9%, 8%, and 7%; (iv) for a given percent increase in time-in-range, the decline in A1c is greater when baseline A1c is higher; and (v) higher time-in-range (as measured by quarterly 7-point SMBG profiles) in the DCCT correlated with lower microvascular complication burden (Diabetes Care 2018). One astute attendee from Sweden asked if Dr. Beck had performed the DCCT time-in-range analysis while holding A1c constant (i.e., at a given A1c, did greater time-in-range predict fewer complications?) and vice versa. The answer was no, but Dr. Beck said it was a good idea. 

Selected Question and Answer

Prof. Stephanie Amiel: Would FDA accept the DCCT post-hoc analysis as adequate support for time-in-range as a useful metric in diabetes?

Dr. Beck: That was the impetus for trying to do this analysis. So yes, I hope so. We’ve been coordinating with JDRF in terms of communication with the FDA, because we want a concerted effort from the community in dealing with FDA on this issue. It’s very different talking hypo with the device side vs. the drug side of FDA, because the device folks have been dealing with CGM metrics for a while. It’s the drug side that we really have to convince. But hopefully we’ll present our analyses and make inroads there. That’s the hope

ISPAD 2018 Clinical Practice Consensus Guidelines Include Chapter on Diabetes Technology for First Time; Stronger Stance than ADA on Pumps, CGM, and AID

In line with ADA’s 2019 Standards of Care, ISPAD’s 2018 Clinical Practice Consensus Guidelines included a chapter on diabetes technology for the first time. The chapter consists of recommendations for insulin pumps, CGM, sensor-augmented pump therapy (SAP), closed loop systems, digital health (i.e., apps, decision support, bolus calculators), downloading technologies (e.g., Glooko, Tidepool), telehealth, and impact on quality of life. In today’s session, Yale’s Dr. Jennifer Sherr presented on insulin pumps, University of Otago’s Dr. Martin de Bock on CGM, and University of Cambridge’s Dr. Martin Tauschmann on SAP and closed loop. See below for the top takeaways.

  • On insulin pumps: ISPAD took a stronger stance than the ADA in favor of pump therapy in pediatrics, claiming that pump therapy “can be used safely and effectively in youth with type 1 diabetes to assist with achieving targeted glycemic control (B)” and “can assist with reducing episodes of hypoglycemia (B).” Moreover, pump therapy was found to be “appropriate for youth with diabetes, regardless of age (B)” and to “reduce chronic complications of T1D in youth, even when compared to those with similar A1c levels on MDI therapy (B).” To compare, ADA guidelines were more conservative, noting that pump therapy “may be considered as an option for all children and adolescents, especially in children under seven years of age. C

  • On CGM: We were pleased to see a strong statement (A-level evidence) in favor of CGM to “effectively” lower A1c, reach target A1c, reduce glucose variability (for pumpers and MDI-treated patients), and increase time-in-range in the pediatric population. Once again, ISPAD was stronger than the ADA, which recommended that CGM “should be considered in children and adolescents. B” ISPAD guidelines also included the possibility for CGM to evaluate “clinically meaningful outcomes beyond HbA1c (E),” including time-in-range (70-180 mg/dl), time in hypoglycemia (Level 1: <70-54 mg/dl; Level 2: <54 mg/dl) and time in hyperglycemia (Level 1: >180 mg/dl; Level 2: >250 mg/dl). As Dr. de Bock noted, it’s a bit strange this recommendation was graded with only E-level evidence, given that it seems obvious CGM would be necessary to evaluate CGM metrics like time-in-range. Similar to the ADA guidelines, FreeStyle Libre received a separate recommendation: “Use of intermittently scanned/viewed CGM (isCGM), also known as flash glucose monitoring, in the pediatric population is safe (C).” Other notable recommendations are included below:

    • “Real-time CGM can be used effectively for reducing mild to moderate hypoglycemia and shortening the time spent in hypoglycemia in the pediatric population with T1D (B).”

    • “The effectiveness of CGM in children and adolescents with T1D is significantly related to the amount of sensor use (A).”

  • On SAP and AID: ISPAD’s strong recommendation for SAP therapy in children and adolescents was similar to that included in the ADA guidelines. ISPAD advises: “Sensor augmented pump (SAP) therapy is superior in children and adolescents over MDI with self-monitoring of blood glucose (SMBG) in reduction of HbA1c without an increase in hypoglycemia or severe hypoglycemia (A). However, this benefit is mediated by adherence to sensor therapy, with at least 60% use being associated with these findings.” ISPAD’s guidance on AID systems was much stronger than the ADA’s. ISPAD recommends: “Automated insulin delivery (closed loop) systems improve TIR, including minimizing hypoglycemia and hyperglycemia (A).” ADA’s recommendations advised: “Automated insulin delivery systems may be considered in children (>7 years) and adults with type 1 diabetes to improve glycemic control. B” See below for more notable ISPAD recommendations:

    • “Automated insulin delivery systems have proven to be especially beneficial in attaining targeted control in the overnight period (A).”

    • “Predictive low glucose suspend (PLGS) systems can prevent episodes of hypoglycemia and have been shown to reduce hypoglycemia exposure (B).”

    • “Low glucose suspend (LGS) systems reduce the severity and duration of hypoglycemia while not leading to deterioration of glycemic control, as measured by HbA1c (A).”

CGM Posters


Important Details

Multi-National Performance Assessment of the WaveForm Cascade CGM System

  • A multi-center accuracy study of the WaveForm Cascade CGM (1 calibration/day) conducted in Slovenia, Croatia, and Serbia (n=60 type 1s and 2s) found 14-day mean MARD vs. YSI to be 12.3%. Day 1 MARD vs. YSI was 12.3%, dropping as low as 11.5% on day 7 and climbing to 13.6% on day 14.

  • The results of the study will be included in Waveform’s CE Mark submission. The company plans to launch the Cascade CGM in Europe in 2019, in line with DTM expectations.

  • Consensus error grid analysis found 88% of CGM/YSI pairs to fall in Zone A and 99% to fall within Zones A and B.

  • During home use, 14-day mean MARD vs. once-daily BGM fingerstick was 13.6%. Days 2-3 mean MARD was 12.5%, dropping to 11.8% across days 5-6 and 8-9, and then climbing to 15.0% during days 11-13.

Home Use Accuracy of the Eversense CGM System When Worn on Abdomen and Upper Arm

  • A Senseonics study investigating accuracy of the 90-day Eversense CGM during home use (n=18 type 1s) found overall MARD vs. SMBG to be 14.9%. Sensors worn on the arm (MARD: 14.0%) were more accurate than those worn on the abdomen (MARD: 16.3%).

  • 78% of the readings were found to be within 20 mg/dl or 20% of the reference. Accuracy for readings <70 mg/dl was strong, showing that 73% of readings were within 20 mg/dl or 20% of the reference.

  • A1c improved by 0.31% (baseline: 7.1%) at 180 days. No device-related adverse events were reported during the trial. Sensor survival rate was 100% with a mean wear time of 23.4 hours/day.

Hypoglycemia Avoidance After Adoption of a Next-Gen CGM System Including a Predictive Low Glucose Alert

  • A retrospective, real-world Dexcom study in ~1,400 G6 users found the introduction of a predictive low glucose alert significantly reduced hypoglycemia. The Dexcom G6 Urgent Low Soon (ULS) alert alarms when an estimated glucose value ≤55 mg/dl is predicted within the next 20 minutes. ULS was provided as an option with the G6 and was left enabled among >97% of users.

  • Users were categorized based on their low threshold alert setting (70 mg/dl or 80 mg/dl). The transition to G6 from G5 was associated with significantly reduced time <70 mg/dl and <55 mg/dl regardless of threshold setting. G6 use was also associated with greater time spent in-range as compared to G5 data, with a significant increase shown among those with a hypoglycemia threshold set to 70 mg/dl. Importantly, the reduced hypoglycemia did not come at the expense of increased hyperglycemia; rather, time spent >250 mg/dl declined significantly after the transition to G6 for both low threshold alert settings.

  • Those with a threshold set to 70 mg/dl spent more time in hypoglycemia as compared to those with a threshold of 80 mg/dl. However, those with a 70 mg/dl threshold spent less time in hyperglycemia.

Cost Effectiveness of Real-Time Continuous Glucose Monitoring (rt-CGM) Compared with Self-Monitoring of Blood Glucose (SMBG) for Type 1 Diabetes Patients in the United Kingdom

  • A cost-effectiveness study of Dexcom G6 vs. SMBG in type 1 diabetes patients using MDI found the G6 to be highly cost-effective with an incremental cost-effectiveness ratio (ICER) of £4,161/QALY (~$5,400/QALY). This ICER was determined for a patient population according to Diabetes UK’s most recent technology consensus guidelines recommending CGM for patients with A1c ≥8.5% and impaired hypoglycemia awareness. Analyses were conducted on 5,000 hypothetical patients and 1,000 microsimulations.

  • Sensitivity analyses were performed by varying diabetes complication costs ±20%, altering time horizons, increasing hypoglycemia rates by 25% for CGM, reducing hypoglycemia disutility rates by 50%, and diminishing disutility of non-severe hypoglycemia events over time. All analyses resulted in ICERs below the NICE threshold of £20,000/QALY, except when researchers diminished the disutility of non-severe hypoglycemia events over time, which resulted in an ICER <£25,000/QALY.

  • Treatment effects and baseline characteristics were based on DIaMonD and HypoDE trials, while all other model assumptions were based on other published literature. Cost and clinical outcomes were discounted at 3.5% annually by NICE guidelines.

  • This important study highlights the cost-effectiveness of CGM in a high-A1c, hypoglycemia-unaware population. These studies are useful in showcasing the impact of identifying the patients who stand to benefit the most from a technology.

Digital Health Highlights

Hygieia d-Nav RCT Published in The Lancet! 1.0% A1c Reduction in 6-months vs. 0.3% in Control Group; Dr. Bergenstal Underscores Benefit of Remote Care and Automated Insulin Titration in T2D

Dr. Rich Bergenstal announced that a six-month RCT (n=181) of Hygieia’s d-Nav Insulin Guidance System (BGM with built-in insulin titration plus remote HCP support) has been published in The Lancet. See Hygieia’s press release on the publication. The study found a statistically significant A1c reduction of 1.0% with the d-Nav system, compared to 0.3% in the control group of only HCP support (p<0.0001); baseline A1cs were 8.7% and 8.5%, respectively. It is worth noting that, the intervention arm had significantly higher mean glucose (185 vs. 166 mg/dl) and mean fasting glucose (185 vs. 157 mg/dl) at enrollment, and mean glucoses were not significantly different at study’s end (though the d-Nav group certainly saw a bigger decline). Frequency of SMBG-measured hypoglycemia (<54 mg/dl) was low and similar in both groups (~0.3/month), and there were three and two severe hypoglycemia events in the d-Nav and control groups, respectively. Achieving A1c <7% without a severe hypo was ~7x more common in the d-Nav group; achieving A1c <8% without a severe hypo was ~2x as common. Total daily insulin dose increased by 0.48 units/kg in the d-Nav group (0.77 -> 1.24 units/kg) and only 0.05 units/kg in the control group (0.71 -> 0.76 units/kg), with the average d-Nav participant making 1.1 adjustments per week, 20% of which were down-titrations (compare to the majority response of 1 insulin dose adjustment per year reported in a baseline questionnaire). As expected with increased insulin dosing, average weight gain was higher in the d-Nav group than the intervention group – 2.3 kg (~5 lbs) vs. 0.7 kg (~1.5 lbs), respectively – though Dr. Bergenstal noted that a few pounds (~3) is a small trade-off for a 1.0% A1c decline. As he put it: “This is what it takes… You can’t just call once a week and ask ‘how are you doing?’ then adjust… You have to go up, then down, then up, then down.” Indeed, his “key slide” – a case study of dynamics in insulin needs displaying the immense variability between all doses of insulin (see below) – underscored that there is no one “right” dose of insulin, even for each person. These titrations, while necessary, are extremely difficult to accomplish on one’s own but much more manageable with an algorithm and care team behind you. People in the study seemed to trust the recommendations: 70% reported that they were “comfortable” or “very comfortable” with d-Nav adjusting doses, while only 10% were “not comfortable” – we’d be curious to see outcomes split along this variable. This manuscript is probably the highest-profile publication for the remote diabetes care/insulin dosing titration landscape ever – we sincerely hope that it drives greater consideration of these technologies in the eyes of patients, providers, and payers. Regardless, Dr. Bergenstal seemed sure that conversation of technology in type 2 will be escalated sooner rather than later, quoting Dr. Mark Evans’ accompanying editorial for the paper: “Faced with increasing pressures and demands on primary care, a substantial niche seems likely for technology to help in type 2 diabetes.” To be sure, simple algorithms like those on the d-Nav device carry tremendous potential to improve the outcomes of countless people who are not using insulin optimally/safely and for those who are not yet using insulin due to fear on their or a caregiver’s/HCP's part.

  • Hygieia received FDA 510(k) clearance for a type 2 insulin titration app capable of supporting “all types of insulin regimens” just three days prior to this publication. The d-Nav device with built-in titration that was used in this study, however, has still yet to be cleared (to our knowledge).

  • Despite guidelines suggesting GLP-1 agonists as first-line injectables and insulin’s resulting demotion within these algorithms, the real-world effectiveness of insulin is undoubtable. Based on NHANES database analyses, ~50% of people with type 2 taking insulin achieve an A1c <7% in clinical trials, while only ~30% reach that same goal in the real-world, conferring a real-world-to-clinical-trial effectiveness ratio of 60% (30/50). For GLP-1 agonists, that ratio is 42%, based on an average A1c change of 1.25% in clinical trials vs. 0.52% in the real world (0.52/1.25). Dr. Bergenstal’s point was not to compare the effectiveness of these two agents – which would be impossible given that the endpoints are different – but rather to assert that demotion of insulin within a treatment algorithm does not preempt its efficacy or preclude its use. His point was that we must find a way to use insulin more effectively. To this end, he provided his criteria for effective type 2 insulin management trials and devices, which must: (i) provide effective and safe insulin management; (ii) provide timely insulin adjustments; (iii) get the patient involved in their own management; and (iv) allow HCPs to play a critical but supportive role. As examples, Dr. Bergenstal highlighted both Hygieia’s d-Nav and CeQur’s bolus patch (formerly Calibra’s Finesse bolus patch); for the latter, he presented a trio of posters originally seen at ADA 2018 demonstrating patient and provider preference for the patch over standard insulin pen delivery.

Sanofi Out of the Digital Woodwork: Launching Durable, Prefilled Connected Pens and “All-in-One” Connected Patch Pump in “Very Next Few Years”; 8,000+ Using Digital Solutions, Including ~3,500 on Titration

In an early morning symposium, Sanofi Head of Integrated Care Mr. Gilles Litman provided a long-awaited updated on the company’s digital health strategy. Of note: (i) The company is internally developing connected insulin pens in both the prefilled and durable form factors; (ii) work on the prefilled, connected type 2 patch pump with Verily and Sensile continues; and (iii) the company’s insulin titration products (Diabeo, My Dose Coach) are being used by ~3,500 people in six countries, and a recent study in India showed that My Dose Coach resulted in a 2.7% A1c reduction in individuals initiating basal insulin therapy. Overall, there are 8,000+ people using Sanofi’s digital products globally – early days to be sure! With these new details, Sanofi’s strategy is becoming more clear; relative to Lilly and Novo Nordisk, the Paris-based insulin manufacturer had leaned toward fewer details. Read on for more and pictures.

  • Sanofi is developing both prefilled and durable pens, both with connectivity built into the body of the pen (i.e., not a cap or other attachment); in a separate conversation, Mr. Litman told us they will launch “in the very next few years.” In his talk, he added: “We want people to engage in a different way by bringing connectivity to our devices to capture data in a seamless way. We’re developing disposable and reusable pens. They will enable better dialogue, data and decision-support, and solutions for self-titration. And this is basal and basal-bolus patients.” Sanofi seems well behind both Novo Nordisk (launching first connected pens in EU in early 2019 following a Swedish pilot in 2018) and Lilly (filing first dose capture technology with the FDA this year). Sanofi has also invested in and is working with Common Sensing’s Gocap (currently 510(k) exempt) in a pilot with Innovation Health (Aetna) and One Drop, and the company didn’t explicitly rule out further collaborations with the startup in our conversation. Sanofi has made the decision to install connectivity directly into its devices, but the cap form factor has the advantage of working with disposable pens; will it pivot to use Common Sensing? Since Sanofi currently manufactures 300 million pens per year, connectivity is no small investment – further underscoring the consensus view that better monitoring and decision support is the way forward in insulin injection therapy.


  • Confirming our suspicion of a deliberate roll-out, Sanofi’s insulin titration app portfolio (Diabeo/My Dose Coach/Voluntis Insulia) are used by just 3,500 people in six countries (including India, France, Germany, Canada, and Mexico). There are currently 10 pilot programs and four commercial launches. A 12-week study in India (n=137 type 2s initiating 100 U insulin glargine) reported toward the end of last year a 2.7% A1c reduction from a high baseline of 9.9% with no incidents of symptomatic low blood glucose or hypoglycemia hospitalizations. Average insulin dose increased by seven units per day, and 38% of participants achieved <7% A1c. “Very promising outcomes,” Mr. Litman commented. There was no control group initiating basal insulin without My Dose Coach so the marginal impact of the app is an unknown, but a 2.7% A1c reduction is obviously quite meaningful. A Sanofi Integrated Care rep estimated that commencing basal insulin generally results in a ~1.5% A1c drop without My Dose Coach – loosely implying an effect size of -1.2% – though she added that “that’s not in people with baseline A1cs >10%, like they are here.” In France, basal-bolus titration/telemedicine through Diabeo are reimbursed by the government, and we are also glad to see the company’s focus on implementation of decision support in low-resource countries with health systems of varying means and composition. Arguably, these self-titration apps can have the biggest public health impact in these settings. It is taking longer than expected for titration to take off in the US, where there are five FDA-cleared apps, including Sanofi’s My Dose Coach and Voluntis’ Insulia (a Sanofi commercial partner) and the just-cleared Hygieia d-Nav. For his part, Mr. Litman said Sanofi is learning from experiences, “How do people use it? How is it reimbursed? How do physicians use it? Testing and learning with stakeholders. What we do with titration is different in different countries. Cultural, psychological aspects are different. This is valuable for us.”

    • “EGIDE” in the slide below is an independent group of “former health ministers, payers, clinicians, academics” assembled by Sanofi with a goal of accelerating connected care acceptance, reimbursement, and development.


  • Sanofi boasts at least five collaborations in the diabetes vertical, each laid out on the slide below. Sanofi and Evidation continue to leverage the latter’s Real Life Study Platform to better understand the impact of behaviors on health and economic outcomes at scale and what is the best intervention for a given individual. The company already has a framework for what works best for which person, which it is currently testing with Evidation. Science37’s virtual clinical study platform is presumably being used to validate various aspects of Sanofi’s digital portfolio, though Mr. Litman didn’t specify.


Dr. Lias Leaves Door Open on Regulatory Path for CGM-Based Bolus Calculators (Depends on Impact of Point Inaccuracy); Overviews Nuances of Decision Support Regulation – Many Questions Remain!

FDA’s Dr. Courtney Lias acknowledged that there are many questions about the regulatory path for diabetes decision support tools in the US (both inside and outside the Agency), part of which was apparent in her musings on the use of CGM to inform bolus calculators: “What if we use CGM instead of BGM to put values into blood glucose calculators? That’d be very useful, now that CGMs are used more often. The question is what is the outcome of algorithm? There is some imprecision with SMBG point values, but there’s a different bias for CGM – a larger bias – so a point glucose value is not as accurate as a single point value from BGM. So we’d be looking at information to understand whether CGM-based calculators should be looked at the same as BGM – what is the impact of the inaccuracy of CGM on the output insulin doses? We’re talking to a lot of people about this, and we’ll hopefully have information soon. If it’s low risk, then maybe we can apply standard blood glucose calculator regulations to CGM – if the recommendations are not always great, then new calculators will probably be developed to incorporate trend and other CGM information to enable that sort of feature.” We guess class II is very possible here, at least for any iCGM with a non-adjunctive indication. Plus, the FDA cleared DreaMed Advisor Pro clinical decision support software for optimizing pump settings based on retrospective CGM data (and now SMBG data) created a new class II medical device category (“Insulin Therapy Adjustment Device”). One of our questions at the time was whether real-time, patient-facing, CGM-based decision support would fall into this new category; it sounds like this question remains to be fully addressed. We also wonder how the iCGM designation will play into this – might some bolus calculators (and possibly basal titration apps) be cleared for explicit use with iCGMs that have demonstrated consistently high performance? Our guess is that CGM-based bolus calculators will take advantage of the wealth of trend and historical data offered by continuous glucose tracking; doing so might require greater algorithmic complexity, though it could also be kept simple based on the dose adjustment trend guidelines published for both Dexcom and FreeStyle Libre. Between Abbott, Bigfoot, Companion Medical, DreaMed, Lilly, Glytec, Dexcom (TypeZero), and Novo Nordisk, we can think of quite a few players who might be willing to work with the FDA to push the envelope here.

  • Does a 510(k) decision support submission necessitate the presentation of clinical data? “Depends on what the product does.” She explained that an app that simply gives nutrition and exercise advice is probably not regulated, while an app with standard insulin-dosing bolus calculator might entail data on human factors, and a new algorithm (i.e., DreaMed’s Advisor Pro) “might need clinical data to show safety.”

  • Further exploring other nuances of decision support regulation, Dr. Lias touched on intended use (what does the app do?), transparency, and intended user. Software that provides more general recommendations, is more transparent in its decision-making processes, and is intended for expert users (HCPs, nurses) might be subjected to less FDA scrutiny; on the other hand, software that provides highly specific advice (e.g., insulin dose), has more opacity in how it arrived at its recommendation, and is patient-facing may be scrutinized more thoroughly, but also has potential to make a bigger impact on the lives of people with diabetes.

  • We were glad to hear Dr. Lias voice her strong support for integrated devices and decision support tools. “Integrated devices are key. Diabetes devices that are designed to work together, including for decision support, will invariably lead to better outcomes. The more integrated a bolus calculator is with a pump, the better outcomes there’ll be. For example, you’ll get better estimates from a bolus calculator that knows the insulin on board than you would from one that doesn’t.” Likewise, she began her talk by empathizing with the nurses who don’t really understand insulin dosing, but are asked to support inpatients, including those who do not have diabetes and so can’t tell if a dose is way off.

CGM-Based Decision Support for MDI Patients: It Only Works if You Use It – And Not Everyone Is in the TypeZero/Dexcom/NN Study

UVA’s Dr. Marc Breton summarized mixed top-line outcomes from the ongoing study testing TypeZero dosing decision support, Dexcom G5 CGM, and Novo Nordisk NFC-enabled smart pens in MDI users vs. CGM and manual insulin dosing. The first-of-its-kind study has n=77 completed and n=2 ongoing, so final statistical outcomes were not shared. At a high level, decision support did not lead to big benefits in the overall population. Time-in-range (70-180) was a solid ~60% in both groups at baseline (with ~4.5% time <70 mg/dl), and both groups improved by a similar amount during the study. According to Dr. Breton, this was due to the introduction of CGM and insulin degludec titration, which occurred in both groups. TypeZero’s MDI decision support did not drive further benefit – a smart CGM-based bolus calculator, basal and bolus titration, bedtime and exercise advice, and hypoglycemia warnings. Why? As with any technology, decision support only benefitted those who actually used it – and use varied quite a bit in this study. A sizeable 25% of the decision support group used the bolus calculator less than twice per day, and the bedtime advice was barely used at all – 29% never used it and 48% used it less than once every three weeks. The same was true of the exercise advice. On the plus side, those who used decision support consistently had less time-in-hypoglycemia (<70 mg/dl) – based on the graph (picture #3 below), time <70 went from ~6% at baseline to ~3% in the group defined as “users.” “Our system provided the expected benefits, but you had to use it a minimum number of times per day.” Dr. Breton noted that one app update during the study meaningfully changed satisfaction, a reminder that user experience (UX) is going to be critical in this domain. (In fact, UX might be more critical, since MDI decision support requires opening an app and following the advice vs. an AID system that just runs in the background.) Full data will presumably come at ADA.

  • Dr. Breton also emphasized that this study was very hard to recruit for – to achieve 77 completers, the study team had to contact 751 people! Apparently, many MDI users were not interested in having to carb count and do intensive insulin therapy – and these were people who agreed to be contacted. We see huge potential for CGM-based MDI decision support, but learning is clearly still in the early stages. (To roughly approximate it, we’d say this field is where AID was three to five years ago.) As with most studies, we’d note the baseline management was better – mean A1c of ~7.4%-7.5%; time-in-range of 60%; CV of 35% - relative to the average person with type 1. The study also used NFC-enabled pens that had to be manually scanned before meals – more onerous than seamless Bluetooth-enabled options.

DreaMed Pipeline: “Advisor Dose” Patient-Facing Bolus Calculator, Pump<->Transition Support (w/ JDRF), Insulin Titration for SMBG/MDI (w/ Harvard), Glucose Forecasting (?); PWD Survey on Decision Support

DreaMed’s Dr. Revital Nimri covered the company’s decision support future (including the patient-facing Advisor Dose) and presented survey data from Schneider Children’s Hospital on patients’ needs and desires when it comes to decision support.

  • Dr. Nimri displayed a few screengrabs of an in-development real-time, patient-facing decision support system with actionable alarms and a “smart bolus” called “Advisor Dose.” CSO Prof. Moshe Philip has hinted at a patient-facing version of the FDA-cleared Advisor Pro pump settings decision support (retrospective) system a number of times in the past, but we’re not sure that’s what this is (at least not in its current form). Rather, this app appears to integrate CGM data, track last insulin dose – possibly based on injection dose capture with Biocorp’s Mallya cap, though the “Add Bolus” button implies manual entry. There also appears to be manual qualitative food and meal entry. Dr. Nimri didn’t say whether this app was limited to MDI or if it worked with pumps as well, nor how far along it is in development. Another concern is regulatory, since the bolus calculator seems to be based off CGM data – as FDA’s Dr. Lias said (see above), CGM’s lower point accuracy may require the consideration of trend and historical data in addition to the current glucose number. We could see Advisor Dose becoming the framework for future patient-facing DreaMed decision support innovations.

  • For the first time since it was announced in May 2017, we heard a public update on the research collaboration between DreaMed, Schneider Children’s, and Harvard to develop a novel dosing support for type 1s on MDI and fingersticks. There’s still no commercial timeline attached to this project, but as seen in the slide below, researchers are tackling some of the thornier problems, such as offering MDI support with minimal fingerstick data. In this particular slide, an in silico simulation of a patient who is only performing three fingersticks a day, according to Dr. Nimri. We’re not entirely clear about what the slide is showing beyond that; we’d guess it’s either (i) a representation of how the researchers are going about imputing “missing” blood glucose data or (ii) a simulation of a patient who is only taking three fingersticks per day but receiving dosing support advice.

  • As shown in the slide below, JDRF has funded DreaMed’s work on supporting patients as they navigate the insulin challenges of transitioning between MDI and pump (apparently in collaboration with Dr. Cobelli et al. at the University of Padova). We’ve also heard this potential use case from companies developing closed loop and MDI dosing support, such as Lilly and Bigfoot.

  • DreaMed is also working on glucose prediction one hour into the future (likely with type 1s, given that focus in the group’s other work). In the trace below, the 21-day profiles appear to track very well (with a 13% error rate), though there is room for improvement. The true positive rate for hypoglycemia prediction is just 56%, meaning 34% of hypoglycemic events are not predicted by the algorithm. Plus, there is one false positive hypoglycemia alarm every 1.5 days. Hyperglycemia prediction is more robust since the data set is bigger (i.e., it is more common than hypoglycemia). On the high end, the true positive rate is 91%, meaning fewer than 1 in 10 hyperglycemic events are missed. However, the false positive rate is higher than it is for hypoglycemia, at one per day. We’re not sure if this module would be used simply to inform insulin dose recommendations or actually be presented to patients to aid their decision-making, but we suspect 1.66 false alerts per day – 0.66 false hypoglycemia and 1 false hyperglycemia – is too high. Medtronic and One Drop have both already launched patient-facing blood glucose prediction software; the real-world accuracy and clinical impact of these efforts remains a big question.

  • Dr. Nimri also presented some interesting data obtained from surveying 47 type 1 pumpers (+68% “constant” CGM users) ages 6-30 at Schneider Children’s. Some of the most notable findings:

    • 54% of patients never change their insulin pump settings between visits. Of those who do, basal rate is the most commonly-adjusted parameter, a fact which surprised Dr. Nimri.

    • 29% never/almost never amend insulin boluses according to the sensor glucose trend arrows; 29% do it all the time. For those who do not adjust their insulin dose based on arrows, the most frequently-cited reasons are lack of confidence (37.5%) and the high difficulty of doing so (25%). This is a small, highly-selected sample, but there was a significant correlation between adjusting insulin dose according to glucose trends and the level of A1c. This is a reminder of the desperate need for trend-adjusted, CGM-based bolus calculators – preventing a high or low blood sugar with a smarter bolus is better than ignoring the arrows and dealing with a subsequent prolonged high or low.

    • See the following table for attitudes on various decision support products/modalities. These are very encouraging numbers for adoption, though this is a young, tech-friendly population. We’d be super interested to see this kind of work expanded in larger, more representative surveys.


Willing to try


Would you be willing to use an algorithm that adjusts your pump setting after downloading your devices at home?


Would you be willing to use an algorithm that suggests insulin-dosing in real-time?


Would you be willing to get text messages each time you need insulin adjustment?


Would you trust dosing recommendations given by automatic algorithm?


Do you think that automated algorithms for insulin dosing will release some of the burden of managing diabetes?


Questions and Answers

Q: Whose responsibility is it when an adverse event like severe hypoglycemia happens with this software? Is it the patient or is it the healthcare provider or is it manufacturer who is going to take responsibility for the damage?

Dr. Nimri: It’s a complicated question. Good question. I don’t know all the answers for this question. I believe that if we give patients our Advisor, it will be the same, like closed loop. This needs to be examined in the future. This was the first question I had for closed loop. Now we have closed loop out there – so who’s responsible if severe hypoglycemia? It’s the same situation.

Prof. Eric Renard: The difficult part is to validate your algorithms. You have so many pieces of advice – if you have ten doctors, who is right?

Dr. Nimri: The outcome!

Prof. Renard: But you can only make one decision, so you can’t live the outcome.

Screenshots of Onduo App Tracking, Remote CGM Onboarding (“CGM Hyperloop”); RCT Coming Down the Pike; Management “Agnostic” on Integrating Other Data Sources

Onduo Head of Clinical Affairs Dr. Ronald Dixon presented screenshots of the tracking and CGM capabilities of the “virtual diabetes clinic” (connected BGM/CGM + remote coaching + app), which is currently in the midst of a 2019 national launch with payers, employers, and health systems. Log-able events in the app include meals (photos), snacks, physical activity, and even mood, representing a strong start to comprehensive education on blood glucose affecters (see below). The app looks pretty much the same as when it launched one year ago in three states.


  • We also got a breakdown of the onboarding process for CGM within Onduo – reportedly the first such 100% remote system of its kind (see photos below). The slide refers to the process as a “CGM hyperloop.” After establishing a willingness to use CGM with a coach, participants have a telemedicine visit with a physician (asynchronous or synchronous), followed by shipment, insertion, pairing, and onboarding of CGM (either assisted by coach or following videos in the Onduo app). The participant is then set to use CGM in Onduo, gaining insights from coaches and the app (see screenshot below with time-in-range [80-200 mg/dl], steps, and a “dinner challenge”) without ever physically entering the clinic. The app sets a goal for the user to spend 75% of time in the wide 80-200 mg/dl range, and tracks peak glucose post-meal to help users identify spike-prone foods. As a reminder, CGM-derived glucose values in the app are three hours delayed.

  • Prompted by a salient question from Dr. Satish Garg on the proportion of Onduo benefit attributable to CGM, Dr. Dixon revealed that the company plans to conduct a study in the next year to tease out the variables most responsible for Onduo outcomes and behavior change. The company is currently conducting a single-arm study (size, methods, outcomes not shared), that will inform an RCT on this subject. We’re eager for more details (will different arms be “knock-outs” of the various parts of the Onduo platform?) and expect, as was implied by Dr. Garg, that access to real-time CGM will be one of the principal factors motivating behavior change.

    • Dr. Dixon presented a compelling case study in which CGM drove significant time-in-range improvements for one Onduo member who was already “well-controlled” (A1c=6.4% at baseline). In the first seven days of CGM, the participant’s time-in-range hovered around 50% with an average blood glucose of ~180 mg/dl. Three months later, time-in-range was often greater than 90% and average blood glucose had dropped to 144 mg/dl. Importantly, adjustments were made only to the patients’ diet, based on a heightened understanding of how eating patterns affected their glucose levels through Onduo; no medication changes or increases in steps were needed.

  • After refinement of current processes, Onduo may allow data integration from other sources. As Dr. Dixon put it, “We are agnostic on what data sources [Onduo] could take, as long as they are going to help people with diabetes.” The app does pull activity data from Apple Health/Google Fit.

Dexcom Clarity in EMRs – Demo of UVA Integration into Epic (PDF upload); Dexcom EMR program to expand this year; more UVA phases

Following a mention at last year’s ATTD, a Dexcom Clarity workshop shared the workflow of Dexcom CGM data flowing directly into the EPIC electronic medical record. Dexcom is using a software company called Redox, which pulls CGM data in a PDF from Clarity and uploads the PDF to the “media” tab of a patient’s EMR. The program has been live within UVA’s Epic system since January, and we got a great look at it today. During a visit, an HCP adds an “order” for the CGM data within Epic, and the report is delivered in less than a minute directly into the EMR – no need to log in separately to Clarity. The process only requires a one-time patient authorization, which can be done on a patient’s phone. (The patient logs in to his/her Dexcom Clarity account and “authorizes” the connection, similar to authorizing other data partner apps like Glooko.) The order is linked to the CPT billing code 95251, allowing the HCP to bill for CGM data interpretation. UVA uses this integration for both telephone and in-person visit encounters. Dexcom plans to expand the program this year, and it also works with Cerner and Athena Health EMRs – nice! Said Director of Data Partnerships Dr. Nate Heintzman, “Scalability and interoperability are baked into this approach.” Ohio State’s Dr. Kittie Wyne, who was in the audience, eagerly asked in Q&A how she could get it immediately in OSU’s version of Epic. Though some work will be required locally – e.g., getting the health system’s IT department to agree – having the technical aspects figured out is a big step.

  • Getting a PDF from Dexcom Clarity automatically into the EMR is an important step, but not the end goal. UVA’s Dr. Ananda Basu shared that there are two more phases for UVA’s CGM data integration into the EMR, which he hopes to complete in the “next 1-2 years.” Phase 2 aims to add time-series CGM data (i.e., not a static PDF), insulin data, heart rate/activity data, and deploy analytics and custom reporting. Phase 3 will expand the automated data collection and storage, including more advanced analytics, sending “mini reports” into the Epic record, alerting healthcare providers, and real-time remote monitoring. Excellent!

  • There are now 13 launched apps that work with Dexcom’s retrospective data API, including new additions Welldoc Bluestar, Social Diabetes, Fitabase, and Validic. The Developer.Dexcom app gallery will be updated soon. The other nine include One Drop, Glucose Buddy, Tidepool, Achievement by Evidation, Ensa, Rimidi, Glooko, Center Health, and 1Bios. Over 1,000 third parties have registered as Dexcom Developers ( and “hundreds of prototype apps” have been created thus far. Dexcom took the lead in September 2017 when it launched this public API for third-party apps to leverage retrospective CGM data; to date, it remains the only program of its kind in CGM.

  • In line with recent Dexcom updates, Dr. Heintzman noted ongoing work on a (i) CGM-based insulin dose calculator (the acquisition of TypeZero should help a lot here); (ii) adding more context around CGM data (diet, activity, insulin – “The more context, the better the insights and outcomes,” said Dr. Heintzman); (iii) population-level data; and (iv) standardized APIs for expanded access and more data partnerships.

Dr. Kovatchev’s VIP (Virtual Image of the Patient) Approach Fleshed Out: CGM Daily Profile Clusters -> Transition Probabilities -> Optimizing Treatments –> Mapping Individuals onto Virtual Representations

UVA’s Dr. Boris Kovatchev provided the most detailed explanation of his VIP (Virtual Image of the Patient) concept to date, clarifying how a therapeutic suggestion could be made based on in silico representations. The process begins by generating clusters of daily CGM profiles. He showed how the ~9,000 CGM daily profiles generated from the 126 IDCL Protocol 1 participants during the course of the study fell nicely into three general clusters: 1) tight control/intensive treatment (77% time-in-range, 5.6% time <70 mg/dl); 2) glucose volatility/hyperglycemia (27% time-in-range, 1.2% time <70 mg/dl); and 3) intermediate/average control (55% time-in-range, 1.8% time <70 mg/dl). [It’s worth noting that the clusters were roughly the same for the control group (SAP) and the closed loop group, though SAP participants had 4% less time-in-range and 3.2% more time below 70mg/dL while in cluster 1.] Next, each participant’s clusters were linked in a chain – i.e., an individual’s day one profile fell into cluster 1, day two fell into cluster 3, etc. – and their transition probabilities were calculated. Transition probabilities reflect the likelihood that an individual will fall in each of the three clusters the day after her profile matches a given profile. For example, one person in the closed loop group has a 67% chance, following a day in the ~77% in-range bin, of repeating this same sort of day; meanwhile, this probability for one individual in the SAP group was 0%. The next step is to map an individual’s CGM profile to a representative virtual image of the cluster they currently fall in, and then to use in silico experiments to optimize treatment for each daily cluster (“for a cluster 1 day, do XX”). This procedure can be automated, applied daily and adapted for either MDI support or closed loop, and then applied to individuals. The overall objective is to maximize the probability that individuals will transition to cluster 1.

Verily’s Dr. Zisser: “Information Has a Tremendous Power Once We Share It”; Nodes and Connected Networks, Lack of Endos, Tribute to Dr. Lois Jovanovic

Verily’s Dr. Howard Zisser delivered a dynamic talk on connectivity, best summarized up by his closing line: “Information has a tremendous power once we share it.” In the course of his talk, we found most notable his discussion and examples of nodes and connected networks (specifically connecting a node to itself, and connecting a node to others), a visualization of the sheer lack of diabetes specialists in Arkansas, and a touching tribute to the late, great Dr. Lois Jovanovic.

  • “The first thing we can do with connectivity is connect patients with themselves. The first patients who put the Dexcom on, it was like ‘I was blind, now I can see.’” He showed a before and after trace of a type 2 who used CGM. At baseline, A1c was 8.1% and nearly every strand of the spaghetti plot danced above 180 mg/dl; just eight weeks later, A1c had fallen to 6.4%, 20 pounds of weight had been shed, and nearly all of the spaghetti was flat in the range of 80-180 mg/dl. “They quickly change behavior…just by connecting the patient to themselves, you’ll see remarkable things.” Dr. Amit Majithia presented a pressure test of this idea earlier in the meeting (see above), showing the change in mean blood glucose of participants 14 days before vs. 14 days after CGM in Onduo – most people saw minimal change, but Dr. Majithia noted that Onduo coaches encouraged people to “go crazy” with their CGM initially, treating it as an “exploratory time” (presumably test the limit of what they can eat, etc.). The next step beyond self-to-self connectivity is, of course, self-to-other; i.e., sending information to the healthcare team and back to the patient, and to family members. Dr. Zisser reminisced back to his days in a Sansum lab with Dr. Eyal Dassau, where the two found a way to get Dexcom information out of a patient’s Dexcom and onto a laptop, thus marking the start of Dexcom Share.  

  • The picture below of people with type 2 in Arkansas vs. the endocrinologists in Arkansas says 1,000 words about the rationale for connectivity and remote care.

  • Dr. Zisser concluded with a touching tribute to one of his mentors, the late Dr. Lois Jovanovic: “I’ve never known anyone to love a patient so much.” He recalled how in 1985 she tried to make a “pocket doc,” to “get Lois into a RadioShack TRS-80 PC-1 program for type 1s on pumps.” “Most of these sessions [at ATTD] can trace directly back to Lois.” He also shared that Dr. Jovanovic’s grandmother was one of Dr. Frederick Banting’s first patients. See warm thoughts on Dr. Jovanovic from a handful of the myriad people in and with diabetes who were so lucky to have known her.

Dr. Riddell Hints at New Exercise Support App for Type 1s – Undisclosed Industry Partner, Beta Version to Launch in ~Six Months

Dr. Michael Riddell gave a rundown of available exercise apps for type 1s, and announced a new app – in development with an undisclosed industry partner – that will recommend basal, bolus, and carb intake based on activity; he expects a beta version to launch within six months. Needless to say, we’re intrigued. Dr. Riddell didn’t share many concrete details, beyond that the backend of the app will follow the logic of the decision tree from the 2017 consensus statement on exercise management in T1D. We’ll be looking to see how this app distinguishes itself from others already in the App Store, because as Dr. Riddell himself stated, there are quite a few. This is not to say that the field is saturated with exercise management support for type 1s – on the contrary, we see significant unmet need, and Dr. Riddell displayed data to this end. A small minority of people with type 1 diabetes who exercise regularly use apps today, but they say they would rely on an app if they found the right one, which suggests growing demand. From the same survey, Dr. Riddell highlighted that patients want more exercise-related support from peers with diabetes. He gave a nod to Team Novo Nordisk, and recommended incorporating social media communities into these apps; we wonder if this will be a component of the mystery up-and-coming app. Moreover, different platforms bring different offerings. Dr. Riddell reviewed that Engine1 and DiaBits feature active coaching, while Glucose Buddy and Bant give personalized feedback and reminders, but no coaching. Most of the apps he discussed have a free intro version, with the exception of Engine1. Since patient preferences and needs are particularly unique around exercise, we’d love to see this new app fill gaps in the market.

Barriers, Pitfalls, and Must-Haves of Diabetes Digital Health from Prof. Barnard

University of Bournemouth’s Prof. Katharine Barnard enumerated the pitfalls of digital health apps that have led to poor uptake to-date: (i) Health-tech is not as whiz-bang or attractive as consumer tech, which does not have regulatory hurdles slowing it down; (ii) poor previous experience make people unwilling experimenters (we hear this most often re: CGM); (iii) poor functionality – if it’s not intuitive and easy-to-use, people won’t use it; (iv) lack of infrastructure supporting the app, causing results to fall short of potential (i.e., integration into EHRs, availability to HCPs); and (v) lack of access – not just to smartphones but also inadequate health literacy, numeracy, or physical barriers such as colorblindness. On smartphone access, Prof. Barnard mentioned that 77% of US adults in 2018 had access to a smartphone regardless of socioeconomic status or ethnicity, and 36% of the world’s population (2.5 billion people) will have access to a smartphone in 2019, creating a fantastic opportunity to provide direct, personalized healthcare at a low cost to many people. However, digital health tools must hit some baseline characteristics: (i) Respect of privacy and security, including user safety; (ii) a clinically-meaningful benefit (for both user and HCP) relevant to clinical guidelines; (iii) an economic benefit for payers; and (iv) for users, timely information (e.g., medical, dietary, physical, practical, peer support, health services access, etc.) that is relevant, easy to understand, encouraging, and supportive. (See Adam’s ADA 2018 presentation for a different framework for diabetes apps.) In short, people with diabetes want an integrated care experience that reduces management burden and facilitates easier control; superfluous, complex apps with little value will simply not be used.

  • In Q&A, Prof. Barnard struck us with a particularly salient quote on assumptions of technological capability (or lack thereof) across populations: “I think onboarding is crucial for all of these tools. My own word of advice when it comes to technology is to never think that someone can’t or won’t use it. Offer the opportunity, because you never know what their ability or willingness is. Never exclude somebody based on your own filter of who will benefit, because invariably you will be wrong.”

DreaMed Advisor Pro Multicenter RCT (Advice4U) to Report Sometime After ADA; Fully Enrolled, ~50% Completed; Dr. Forlenza’s Take on Time-Saving

Drs. Lori Laffel (Joslin) and Greg Forlenza (Barbara Davis), investigators in the multicenter, Helmsley-funded Advice4U study of DreaMed’s Advisor Pro clinical decision support software, shared a status update on the trial and a clinician’s perspective of using the software, respectively. Unfortunately, though we were hoping for an interim readout of three-month data, DreaMed now tells us that we won’t see any results until the full six-month read out at a post-ADA meeting (perhaps ISPAD?). At this stage, the trial is fully enrolled (n=112) and ~50% have already completed the study. In the study, there have been ~600 in-person or telemedicine visits between the two groups, 83% of which have resulted in pump settings change recommendations. The remaining 17% of visits were skipped or did not have recommendations due to a host of reasons (insufficient data, technical difficulties, time change snags, and device malfunctions). But, Dr. Laffel emphasized, success with the system has improved as the study has progressed; while the percent of visits with no data generation approached 25% at visit three, this number had fallen to ~10% by visit nine. Dr. Laffel conveyed her excitement for this particular study and the idea of remote monitoring/decision support in general: “In the DCCT, they had monthly visits and weekly telephone contacts. Who has that much time? We wonder why we haven’t seen those results translated into clinical practice today – we’re lacking the personnel, expertise, and resources. This highlights to me why remote, automated, decision support could be helpful.”

  • Dr. Forlenza explained in no uncertain terms where the value in Advisor Pro lies: “In a typical visit, we would spend ~1/3 of our time analyzing that profile, playing technician and tuning the data. But with a system like this, we could actually practice the art of medicine, figuring out how to incorporate better behaviors into patients’ lives instead of just turning knobs and flipping switches.” His ~1/3 estimate rings true with recent findings that it takes providers 18 minutes to analyze pump data, and Drs. Bruce Buckingham and Desmond Schatz estimated during a recent CWD panel that they spend a remarkable 45-60 minutes in each patient visit – most not even looking at the data. Of course, as many have pointed out, this is far more than most people get with their endocrinologist or (most likely) PCP. 

Digital Health Posters


Important Details

A Personalized Meal Grading System Using Professional CGM (iPro 2) with the FoodPrint Report by Nutrino

  • A Medtronic study showed the associations between Nutrino’s FoodPrint grading system and glycemic profiles from six days of blinded CGM data. 3,514 meals were recorded by 329 individuals and scored by FoodPrint from “A” to “F” in terms of glycemic impact.

  • The FoodPrint grade correlates well with other glycemic metrics like time-in-range. Time-in-range was shown to increase from 54% to 77% in correlation with grades F to A.  Presumably, this reflects greater time spent in range following an A meal than an F meal (across individuals), though it may also reflect the time-in-range achieved by individuals who eat foods predominately of a given grade.

Glycemic Control and Satisfaction with Analyses Using an Insulin Pen with Memory and Downloading Function

  • A study investigating glycemic control and satisfaction in 31 children (9-18 years) using the NovoPen 5 Plus during the Swedish pilot (n=700 users) found time-in-range (72-180 mg/dl) to improve statistically, though with small clinical significance, at six months (49% to 51%; +29 mins/day). No significant changes in hypoglycemia frequency were observed

  • Both physicians’ and patients’ perception of the difficulty of the device improved at six months. Interestingly, physicians appeared to be more daunted by the technology than patients at the start of the study: ~60% of physicians found the pen to be “quite difficult,” whereas only ~10% of patients reported feeling the same. This makes sense, as the NFC-enabled pens don’t change the patient use case, but does require extra steps from the provider in-clinic (e.g., downloading, interpreting). The improvement in perception of difficulty is encouraging.

A K-Nearest-Neighbors Approach to The Design of an MDI Decision Support System in Type 1 Diabetes

  • An OHSU study funded by Helmsley and Dexcom trained and validated a K-nearest-neighbors (KNN) algorithm to develop a decision support system (KNN-DSS) for people with type 1 diabetes treated with MDI. The algorithm generates insulin regimen recommendations including: (i) basal insulin increase/decrease; (ii) AM/PM/Overnight insulin:carb ratio adjustments; and (iii) sensitivity factor adjustments.

  • The KNN-DSS was trained in a 15-week in silico experiment in 70 virtual patients to reduce hypoglycemia (<70 mg/dl) and improve time-in-range (70-180 mg/dl). The KNN-DSS was validated in a 12-week in silico experiment in 29 new virtual patients. The validation study included a range of insulin dosing errors.

  • The in silico validation study showed time-in-range to significantly increase from 63% to 78% (+3.4 hours) and time <70 mg/dl to significantly decrease from 4% to 1% (-41.8 minutes).

  • The KNN-DSS was validated in vivo using data collected in a four-week, at-home clinical study (n=12) of patients with type 1 diabetes. Participants wore a Dexcom G6 CGM and used Bluetooth-enabled insulin capture devices (Companion Medical’s InPen or Common Sensing’s GoCap). Meals and exercise were recorded using a third-party app. Data were reviewed weekly by endocrinologists and inputted to KNN-DSS to generate recommendations and determine agreement between the algorithm and endocrinologists.

  • Acceptable agreement between KNN-DSS and endocrinologists was defined as a perfect match with physician recommendations or titrating insulin in the same direction. 69% of recommendations were found to be acceptable and just 10% were found to disagree (i.e., opposite to the physician, or titration insulin in opposite directions).

  • A pilot study is planned to evaluate the KNN-DSS for MDI-treated patients. Decision support for MDI users is low-hanging fruit, and we’re glad to see these strong results.

Beyond A1c Highlights

ATTD Time-in-Range Consensus Targets Preliminarily Set for People with T1D/T2D, Frail People with T1D/T2D, Pregnant Type 1s, and Pregnant Type 2s/GDM; Audience’s Reactions

Prof. Tadej Battelino reported on the discussions and outcomes of the ATTD Consensus meeting on time-in-range, which took place this past Tuesday. Read our report on the consensus points and other takeaways from the day’s proceedings. A fascinating Q&A followed, during which attendees raised a number of concerns – all of which were voiced at the meeting earlier this week. Below, we’ve bolded the audience comments, followed by a summary of the panel’s response.

  • A couple of people wondered what the guidelines meant for (i) people not on CGM, and (ii) resource-poor nations. Prof. Battelino said this was a very important question but didn’t necessarily know the answer; he threw out the idea of making targets for percent of 7-pt SMBG profiles in range. At the consensus gathering, we frequently heard the idea that in setting these goals, the experts may elevate the number of patients using CGM. But ultimately, Prof. Battelino hopes competition will eventually drive cost down or incentivize lower-cost CGM manufacturing.

  • Dr. Des Schatz: “I’d love to see a majority of my patients getting to 70% time-in-range, but I’m seeing adverse effects on academic performance. What’s happening is, kids are looking at this and are disappointed in themselves, and parents are wondering whether it’s really the right thing to do. What is the effect of setting a high expectation?” Dr. Beck recalled hearing this concern from many pediatric endocrinologists at Tuesday’s gathering; the group decided that it should set an ambitious target, but also encourage patients to make meaningful 5% improvements (one hour per day) in their time-in-range numbers. (Prof. Simon Heller spoke up at the consensus meeting and on the panel, voicing concern that those non-experienced endocrinologists or PCPs caring for people with diabetes might end up “beating their patients up” with the consensus targets.) The lofty target was established partially due to ethical concerns – how could a target be relaxed to a level that is known to be harmful to people? This is a tough argument – all goals have downsides. Read Adam’s diaTribe take on time-in-range goals and why he focuses more on “process goals.”  Also see diaTribe’s 2017 piece on expert opinions on CGM, which didn’t change all that much by the Consensus meeting – and numbers will only improve with more wearing AID.

    • Dr. Schatz elaborated on his position after the meeting in an email: “What I am concerned about is the (lack of) importance of small measures of success that will continue to help patients ultimately realize goals. This is real time – not three-month HbA1c. So, asking for 5%-point improvement every 3 months would be reasonable e.g., 40%-45%. I have found that for treating overweight and obese patients - we set long term goal but define short term measures of success. Behavior is hard to change, but make it realistic and it is more likely to become long lasting.”

  • As was raised at the Consensus meeting on Tuesday, an attendee wondered how the consensus intended to handle varied performances/ accuracies of different CGMs. The consensus does not address this issue at all. Prof. Battelino hopes that there will be harmonization among manufacturers in the future, and Dr. Beck commented that (i) average CGM metrics tend to be “right on” even if individual points may not have the same accuracy as BGMs; and (ii) for the purposes of a clinical trial, the control and intervention groups use the same sensor so bias is theoretically controlled for. 

Dr. Hirsh Surveys HCPs: Patients & Endos Ready to Embrace Time-in-Range, but PCPs Still on the Fence

UW’s Dr. Irl Hirsch shared results from an email survey he conducted (n=62, 66% response rate) asking whether patients, endos, and PCPs are ready to change from A1c to time-in-range: Though he conceded that this is “not the type of work the Jaeb Center would do,” the answer leaned in favor of “yes” for patients and diabetes specialists, but was a resounding “no” for primary care doctors. As shown on the slide below, 65% of all respondents felt that patients were ready to jump ship (from A1c), while 59% felt that endocrinologists were ready. (We hope that it’s not actually “jumping ship” – we’ve always thought about time in range supplementing A1c rather than replacing it.) Zero respondents – “not a single person in the survey” – thought non-endos were ready to adopt time-in-range over A1c though our bet is that some of the difference is a matter of exposure. The vast majority of type 2s and a significant percentage of type 1s in the US (estimated at 70% but unknown) see a PCP for their diabetes care, so this perceived resistance from the primary care community may be a barrier for the time-in-range and broader beyond-A1c movement. Again, we doubt it’s true resistance but more an absence of familiarity. And indeed, Dr. Hirsch was cautiously optimistic. He described a long road ahead for time-in-range to gain traction, but highlighted a recurring pattern wherein endocrinologists adopt a new innovation in diabetes and PCPs later follow suit. This is happening slowly but surely for CGM; Dr. Hirsch mentioned that many general internists have been asking him of late how to order a CGM, because their patients are asking for it (he attributed growing patient awareness to direct-to-consumer advertising in the US – no doubt some is also due to word-of-mouth news among patients). Indeed, continued patient-facing education on the utility of time-in-range as a diabetes metric should have spillover effects on PCPs. We look very forward to the day when all patients/providers involved in diabetes care are comfortable with time-in-range; obviously getting CGM to those groups will be the first step. During Q&A, one audience member asked what can be done to accelerate the transition from A1c to TIR, and Dr. Hirsch emphasized that this movement must be all-hands-on-deck. “It took us a long time – years, maybe decades – to teach A1c and what that number is. This is a bit more complicated, but the good news is, it’s doable. We need to start in the medical schools, teach this in CME, and it’s just got to be an all-out effort.”


  • Dr. Hirsh sent out 94 surveys and garnered 62 individual responses. Of these respondents, 53% were adult endocrinologists, 8% were pediatric endocrinologists, 10% were PCPs, and 29% were nurse practitioners and/or CDEs. Notably, everyone answered all three questions about the TIR-readiness of (i) patients; (ii) endos; and (iii) non-endo HCPs. Across the board, 89% of providers felt that non-endos aren’t ready for a change to time-in-range, and 11% said “maybe” (even though maybe wasn’t listed as an option – people wrote it in). We wonder the degree to which the endo crowd was a bit dismissive of PCPs.

  • Dr. Hirsch also received unsolicited comments from survey respondents, and he displayed several slides-worth (see below for an example). The theme of patient-related comments seemed to be that newly-diagnosed individuals (especially children) should learn TIR from the get-go, although it may be hard for longtime diabetes vets to make the transition (this also seems like speculation). One internist commented “internists are too lazy,” which sparked laughter from the audience, though this gets at an important point – PCPs are busy enough as it is that simplicity and minimal burden are critical. Another comment implied that time-in-range will be much more confusing than A1c, but Dr. Hirsch disagreed and so do we. Time-in-range is out of 100% - higher is better; A1c is on a more confusing scale. Time-in-range also affects patients’ quality of life every day, is more intuitive than a weighted average calculated every three months based on hemoglobin glycation. A major point, as well, of CGM, is how it can help you do better by identifying the right therapy – the numbers are clearly not an end in and of themselves.


  • Dr. Hirsh called attention to his just-published editorial in Diabetes Care advocating for time-in-range. He and co-authors Drs. Jennifer Sherr (Yale) and Korey Hood (Stanford) highlight Dr. Roy Beck’s DCCT analyses correlating TIR (calculated by seven-point SMBG) with microvascular outcomes (ADA coverage). Specifically, a 10% decrease in time-in-range was associated with a 64% increase in risk for retinopathy and a 40% increase in risk for microalbuminuria. The editorial suggests that these findings would have been “even more striking” had CGM been used in the DCCT, and while it doesn’t make practical sense to repeat a DCCT-like study with CGM, this technology should absolutely be incorporated into other diabetes clinical trials.

Researchers Describe Detrimental Effects of Hyperglycemia on Cognitive Function; Update on DirecNet from Yale’s Dr. Weinzimer

In back-to-back talks, Dr. Pratik Choudhary (King’s College London) and Dr. Jasna Šuput Omladič (University of Ljubljana, Slovenia) highlighted the negative cognitive effects of acute and chronic hyperglycemia. Dr. Omladič presented preliminary results from her ongoing study of youth with type 1 diabetes (n=40). Participants (20 adolescents age 11-19 with T1D + 20 matched controls) underwent one structural MRI scan and two functional MRI scans – one during a euglycemic clamp and a second, 30 minutes later, during a hyperglycemic clamp. While in the fMRI machine, participants completed the Tower of London task (assesses planning), the Flanker task (assesses attention), and the Visual Spatial Working Memory task (VSWM). Dr. Omladič focused on the last of these, showing how type 1s perform similarly to controls on the VSWM with in-range blood sugar (remembering ~1.7 out of 4 positions on the first try), but perform significantly worse on this test with high blood sugar (remembering ~2.6/4 positions vs. >3 positions, on average). Both groups do better during the second condition because they’ve seen the task once before, but the key takeaway is that a significant difference in working memory emerges between adolescents with type 1 diabetes and those without (no p-value reported). Over a lifetime with type 1 diabetes, acute hyperglycemia accumulates and can have a pronounced impact on cognitive ability. Dr. Choudhary summarized the relevant findings from the DCCT/EDIC: Recurrent episodes of severe hypoglycemia had no significant effect on cognition across several domains (problem-solving, learning, memory/recall, attention, motor speed, etc.), but higher A1c was associated with reduced psychomotor efficiency and motor speed. He concluded that while acute hypoglycemia does impair cognitive function, the much greater consequence on population-level cognition comes from chronic hyperglycemia. This is worth restating – persistent hyperglycemia is more dangerous for brain function than are isolated instances of hypoglycemia – especially in countering the fear of hypoglycemia that keeps patients/HCPs from targeting a lower A1c, or optimizing insulin titration. We don’t want to minimize the importance of avoiding hypo (not at all!), but with the ACP recommending an A1c goal of 8% for type 2s, and with still-prevalent fear of hypo affecting the initiation and titration of insulin in clinical practice, it’s crucial that the field doesn’t forget why it treats high blood glucose in the first place – to minimize complications, including microvascular disease, macrovascular events, and deterioration of cognitive ability.

  • In the same session, Yale’s Dr. Stuart Weinzimer provided an update on the DirecNet program of type 1s diagnosed at an early age (4-10 years-old). Corroborating his fellow faculty, he explained how long-term hyperglycemia was inversely correlated with IQ as well as learning and memory scores. DirecNet researchers have continued glycemic, cognitive, and neuroanatomical assessments on the study population; two- and four-year follow-up data are forthcoming. Late last year, new results published in Diabetes Care showed how a single episode of DKA can negatively affect multiple aspects of cognition – Dr. Weinzimer listed full-scale IQ, executive function, learning, and memory. Lastly, he announced a pilot study investigating how hybrid closed loop affects the trajectory of neuroanatomical and cognitive development in children with type 1 diabetes, which will hopefully report data within the year.

Prof. Amiel on Key Challenges to Studying Hypoglycemia in Clinical Trials

London’s Dr. Stephanie Amiel discussed challenges to studying hypoglycemia in clinical trials, but emphasized that it’s a critical area of research nonetheless. Because severe hypo is relatively rare, RCTs have to enrich their study population with participants at high-risk for hypoglycemia (much like conventional diabetes CVOTs enrolled type 2s with established CV disease). Dr. Amiel presented a dilemma between open vs. anonymous self-report of hypoglycemia: Incidence of hypoglycemia magically drops by 70% when you shift protocol from anonymous to open reporting, but anonymous reporting introduces its own issues, namely that a patient’s understanding of severe hypo may not align with the official definition. Open reporting with a physician or diabetes educator might help patients identify their own hypos more precisely, but then the problem circles back to under-reporting. Dr. Amiel further underscored that patients may have inaccurate recall of their hypoglycemia over several months or a year; this is definitely the case when investigating hypoglycemia unawareness. On a positive note, she shared several examples of successful hypoglycemia RCTs, including HypoDE presented at ATTD 2018 and her own ongoing study of HARPdoc, a new intervention that aims to lower hypoglycemia incidence by addressing patient misconceptions – e.g. “I have to be low all the time because I must not be high.” Dr. Amiel presented very preliminary results (she’d only seen them ~48 hours prior), sparking our curiosity for more detailed data.

Diabetes Therapy Highlights

Experts Sound Off on Type 1 Adjuncts: Drs. Mathieu and Danne Raise Safety and Efficacy Questions for 2.5 mg Empagliflozin; Dr. Garg Highlights ~15% of Type 1s Using Off-Label Meds

A triumvirate of experts on type 1 diabetes adjunct therapy – comprised of Prof. Thomas Danne, Dr. Satish Garg, and Dr. Chantal Mathieu – delivered a tour-de-force series of presentations on the topic. Of particular note, both Prof. Danne and Dr. Mathieu expressed substantial concern over the quality of data supporting the 2.5 mg empagliflozin dose tested in Lilly/BI’s phase 3 program in type 1. Dr. Mathieu first addressed the EASE program’s efficacy data (see EASD 2018 – 2.5 mg was tested in EASE-3 but not EASE-2), pointing to the gradual loss of efficacy on A1c lowering and less impressive weight loss with the 2.5 mg dose, particularly when compared to the 10 and 25 mg doses that drive more glucosuria. Very notable was her caution over the DKA data for 2.5 mg. While there was no observed increased risk of DKA vs. placebo with the 2.5 mg dose (only 2 DKA events occurred in this arm), Dr. Mathieu offered two strong words of caution: “very few events and only six months.” Indeed, while EASE-2, which tested only 10 mg and 25 mg, was extended out to 52 weeks, EASE-3 was continued only to week 26 – a shorter duration that stands in strong contrast to full 52-week programs from both AZ (DEPICT, for Farxiga) and Sanofi/Lexicon (inTandem, for sotagliflozin). Both speakers also made the highly-compelling point that when six-month data for DEPICT-1 were presented at EASD 2017, there was no DKA imbalance between dapagliflozin and placebo, either – but one did emerge in the 52-week (ADA 2018). In Dr. Mathieu’s assessment, this calls into question whether an imbalance would have emerged had EASE-3 been carried out to a year. Prof. Danne echoed these sentiments and further questioned the efficacy of 2.5 mg empagliflozin on the endpoint that seems to be most important to patients – time in range. As he put it, “What patients are reporting is that their glycemic variability is going down and they feel glucose swings less…But with the low dose we don’t have a very significant improvement in time in range.” In EASE-3, patients saw a mean 6% daily gain in time between 70 and 180 mg/dl, compared to +12% with 10 mg, +8% with 25 mg, and +2% with placebo – translating to a placebo-adjusted gain of just-under 1 hour with 2.5 mg. While any time-in-range gain is a positive, Prof. Danne understandably characterized this effect as underwhelming relative to the consistent ~2-3 hour gains conferred by the higher doses, as well as dapagliflozin and sotagliflozin in DEPICT and inTandem. On balance, both experts admitted that, while 2.5 mg empagliflozin may indeed offer efficacy with a lower or no risk of DKA, the data that exist so far should be taken as no more than an indication that this might be the case.

  • We appreciated Dr. Mathieu’s reminder not to forget about heart and renal outcomes for people with type 1 diabetes and the role adjunct therapy could play in lowering residual CV/renal risk. She emphasized that there’s “still a lot of morbidity and excess mortality” among type 1s, reasoning that the biggest remaining question on this front is whether the CV and renal benefits well-established with SGLT-2 inhibitors in type 2 diabetes will translate to type 1. On the first day of ATTD 2019, Dr. Per-Henrik Groop presented preliminary, mechanism-based evidence in favor of a renal benefit in type 1, while cautioning that bona fide clinical data is still on it way. To be sure, the rise of SGLTs in type 1 has led to increasing and insistent calls for a full CVOT in type 1 diabetes – which, doubtless, we’d love to see.

  • Prof. Danne also presented the recently-published international consensus on DKA risk management with SGLTs in type 1, contextualizing the document as a starting point to be further developed and bolstered by additional investigation. Certainly, there’s a fine line to walk between (i) the need to make effective and actionable recommendations applicable to the current and increasingly common use of SGLTs for type 1 and (ii) the immense lack of data on whether any given risk management and DKA treatment strategy is actually effective, particularly in the real world. In our observation, the consensus committee is doing an excellent job of walking this line, and we hope to see a continued push forward on testing and optimizing DKA risk minimization strategies. This is all the more important in light of a point Prof. Danne highlighted: In the type 1 programs, both placebo and treatment arm DKA rates came in below those reported in general practice; while there is an increase in DKA risk with SGLTs, there is also an enormous opportunity to approach DKA minimization more effectively on a general and global scale. How can the field get smarter about DKA prevention for all type 1s?

  • Dr. Garg reviewed the history of and current utilization of adjunct type 1 treatments, painting a picture of immense unmet need – and high demand – for solutions in addition to insulin. As he reviewed clinical data on pramlintide, metformin, colesevelam, DPP-4 inhibitors, and GLP-1 agonists in type 1 diabetes, a narrative emerged that most strategies tested to date have resulted in less-than-enthusing effects on glycemic control and safety (both hypoglycemia and DKA). At the same time, there is undeniable demand for more options: Most recent T1D Exchange data (n=22,697) indicate that ~15% of people with type 1 in the US are using off-label diabetes medications, including metformin (811 patients, 4%), GLP-1s (300, 1%), SGLT-2s (232, 1%), pramlintide (131, 1%), DPP-4s (9, <1%), and “other” drugs (28, <1%). In particular, we find it notable that metformin remains so highly used as an adjunct despite what Dr. Garg described as “hardly any change in A1c, more GI side effects, and more severe hypoglycemia,” based on a study conducted by Jaeb. To be sure, many type 1s do garner meaningful benefit from off-label adjuncts, but on average, no breakthrough therapy that works for all or even most patients has been found – a trend that was highly evident in Novo Nordisk’s two ADJUNCT studies of Victoza in type 1 (see Dr. Paresh Dandona’s study at ADA 2018). Our sense is that, among adjuncts, inter-patient variability in response to SGLT-2 inhibitors is much lower than with others, leading to more consistent effects across study enrollments (though this is our own interpretation) – Dr. Garg labeled the class the “ideal adjunctive prescription,” particularly due to its insulin-independent mode of action.

  • According to Dr. Garg, AZ’s submission for a type 1 indication for Forxiga in Japan just received a “favorable review.” As a reminder, AZ submitted Forxiga for type 1 to Japan’s Pharmaceutical and Medical Devices Agency (PMDA) in May 2018, well ahead of the late 2018 submission to FDA and a couple of months after EMA began reviewing the EU submission. While this news does not appear to have crossed the wire, a favorable review (or even approval) in Japan comes with little surprise – the country approved Astellas’ Suglat (ipragliflozin) for type 1 diabetes in December 2018, with apparently little concern over DKA. To our understanding, Sanofi/Lexicon have not submitted sotagliflozin in Japan. We’ve previously estimated Japan’s type 1 population at ~17,000 people (~13.5 cases/100,000 persons) – a relatively small market compared to >1 million type 1s in the US. On the other hand, we continue to understand that reimbursement and pricing (and thus, profitability) are favorable in Japan, and approval here could represent a gateway into the rest of Asia.

Selected Questions and Answers

Q: Is there any correlation of DKA risk with weight?

Prof. Danne: I’m happy with the 27 kg/m2 recommendation – basically, I think it tells us this drug not for everyone. I think the statistics for that are weak but it’s a sign that this isn’t out there for everyone. So, as an evidence-based person I think the evidence is rather weak but it is a starting point. There is some protection and those people have an opportunity perhaps for more benefit because weight loss is important; however, it isn’t a sure thing that above 27 kg/m2 is safe against DKA – that’s certainly not the case.

Q: Are there guidelines for restarting the SGLT-2 inhibitor if DKA is secondary to other causes, like pneumonia?

Prof. Danne: If the patient understands the need for doing ketone monitoring the next time he feels sick and for stopping the SGLT-2 inhibitor so it doesn’t get that far – you can start again on the lowest effective dose and make sure he has access to ketone monitoring and you as a medical expert.

Q: Should SGLTs be discontinued on hospital admission regardless of whether a surgical procedure is planned?

Prof. Danne: Any patient getting surgery should stop: The risk of DKA is far too high in my mind, even if it’s just a tooth extraction. I think a hospital state is high stress. If the patient does good ketone monitoring, they could continue, but I’d always be on the safe side. Why put the patient at risk? It’s easy to top and restart. We all know how little knowledge there is about diabetes in so many hospitals – Adam Brown just wrote a crazy article about what happened when his appendix ruptured.

Q: Would you recommend SGLT inhibitors for people with an A1c below 7.5%?

Prof. Danne: I would – the main point is reducing glycemic variability and improving time in range. Even a low A1c is sometimes caused by a lot of hypoglycemia, and glucose variability is a factor promoting hypoglycemia. If you have a patient with near-normal A1c and high glucose variability, I wouldn’t hesitate to start it.

Prof. Mathieu Sees Promise in SGLTs for Type 1, But Advocates for Conservative Safety Measures: Prioritize Patient Education; Only Prescribe to Type 1s Looking to Lose Weight

In reviewing the available phase 3 data on SGLT inhibitors in type 1, Professor Chantal Mathieu took a conservative stance, framing these agents as “promising adjunct therapies” while cautioning that “we have to strike the right balance between benefits and side-effects.” She walked through key findings from DEPICT (for AZ’s SGLT-2 dapagliflozin), inTandem (for Sanofi/Lexicon’s SGLT-1/2 dual inhibitor sotagliflozin), and EASE (for Lilly/BI’s SGLT-2 empagliflozin). As a lead investigator for the DEPICT program, Prof. Mathieu described how all trial participants received extensive education around DKA – certainly more than the average type 1 patient receives in the real world. For evidence that this education paid off, she pointed to exceptionally low rates of DKA in the placebo arm of DEPICT 1 (0% after 24 weeks and 2% after 52 weeks), which came in well below the ~8% DKA incidence seen in the T1D Exchange. There were 27 total episodes of DKA in DEPICT 1 (22 on dapa vs. 5 on placebo), and although several of these were attributed to pump failure or a missed insulin dose, Prof. Mathieu emphasized that many events were unexplained. She seemed to suggest that comprehensive DKA education is critical before any patient with type 1 diabetes starts SGLT treatment, and the question remains, how can regulators enforce this?

  • Prof. Mathieu underscored that SGLT inhibitors should be used very carefully in type 1s where weight loss is undesirable, though in many patients weight loss will be a positive outcomes. She presented this as a safety consideration, as preventing undesired weight loss could also decrease risk for DKA (ostensibly by maintaining insulin requirements) and unpredictable glucose fluctuations. Early in February, CHMP recommended AZ’s dapagliflozin (Forxiga) for EMA approval in type 1s, noting that “this treatment should only be initiated and supervised by specialist doctors” and that patients “should be educated about risk factors for DKA and how to recognize its signs and symptoms.” Moreover, CHMP endorsed an indication only in type 1s with overweight/obesity (BMI ≥27 kg/m2), likely because AZ enforced a lower-level BMI cutoff in enrolling the DEPICT trials (we suspect that higher baseline body weight comes with greater insulin requirements, and maintaining sufficient insulin doses while avoiding hypoglycemia is an important protective factor against DKA). The inclusion/exclusion criteria for inTandem1 make no mention of BMI, though ultimately, baseline BMI was ~28 kg/m2 on average (indicating overweight) in both clinical programs. We were somewhat surprised to learn of CHMP’s BMI-based decision, since there was no consideration of a weight-based indication at the FDA Advisory Committee for sotagliflozin just a couple weeks prior, though we’re now wondering whether the BMI criterion was something explicitly put forth in AZ’s regulatory application but not Sanofi/Lexicon’s. Overall we think it’s a very smart place to start, because it means that the fear of “opening the floodgates” as expressed by some wont’ happen. We’re awaiting FDA’s regulatory ruling on sotagliflozin by March 22, 2019. Using dapagliflozin exclusively in type 1s with overweight/obesity is a more conservative approach than approving for all BMIs, and as the new indication is rolled out in Europe, we’ll be looking to see if this has a positive impact on population-level DKA rate. While some might’ve been disappointed by a more conservative regulatory approval for this new class of adjunct type 1 therapy (given the tremendous unmet need), we’ve become increasingly aware of the population-level threat of DKA (a major theme at the sotagliflozin Ad Comm) as have most and we have only characterized the approach as a very smart one (since it’s also modifiable). It’s smart to proceed with caution, and to this end, we appreciated Prof. Mathieu’s suggestions for exercising the utmost safety in prescribing SGLT inhibitors to people with type 1 diabetes as a starting point. As Dr. John Buse put it at ADA, “I’d rather that some people just not get SGLT inhibitors based on our recommendations, rather than see an epidemic of DKA after these drugs are approved. I’d rather the 20% of patients who could really benefit get this drug.” We do think many will want it who do not “need” to lose weight so this may be an ongoing check.

  • During her discussion of EASE 3, Prof. Mathieu shared measured enthusiasm for the 2.5 mg dose of empagliflozin. This very low dose of SGLT-2 inhibitor was not associated with heightened DKA risk (2 events in the 2.5 mg empa arm vs. 3 events in the placebo arm), but Prof. Mathieu warned that these numbers were collected over only six months. She reminded the audience that there was no significant DKA signal in six-month DEPICT 1 data either, but the risk appeared by the one-year mark. She highlighted the persistent DKA signal with 10 mg and 25 mg empagliflozin. Perhaps most importantly, Prof. Mathieu explained how lower risk also means fewer benefits: 2.5 mg empagliflozin was “at the edge of showing enough of an effect on glycemic control and body weight” (i.e., A1c drop and weight loss were more muted with 2.5 mg empa vs. the higher doses of Jardiance). This comment captured Prof. Mathieu’s overarching theme of needing to balance benefits and side-effects. After all, if a low dose of empagliflozin doesn’t bring about clinically-meaningful changes in A1c, time-in-range, or weight, is it worth any chance of elevated DKA? This is still very much an open question in our view (especially because the data certainly did not show no impact at a low dose), and even as SGLT inhibitors for type 1 approach the market, we see big opportunities for further research ahead. We keep coming back to, for example, cardiovascular outcomes trials in type 1 and how terrific it would be to see this approach take hold.

Lilly Symposium Covers Unmet Needs in Hypoglycemia Rescue Therapy; CRASH Results on Response to Severe Hypo in the Real World

Dr. Pratik Choudhary discussed results from the cross-sectional CRASH study (also presented on poster 145) during a Lilly-sponsored symposium on severe hypoglycemia. People with type 1 (n=110) and type 2 diabetes (n=98) were surveyed on their experiences of severe hypo. A majority of these events happened at home (88% for type 1s and 77% for type 2s), and a majority of patients only ate or drank sugars to treat the low (76% of type 1s and 79% of type 2s). In contrast, only 10% of type 1s and 5% of type 2s injected glucagon to recover from severe hypoglycemia, and shockingly, only 43% of type 1s and 33% of type 2s mentioned the episode to a healthcare professional within one week. The reason for this became clear during Ms. Cajsa Lindberg’s talk, which offered the patient perspective on severe hypoglycemia. Ms. Lindberg, who was diagnosed with T1D 17 years ago, described how challenging it is to discuss severe lows with an HCP; she told the story of a recent conversation she had with a new diabetes nurse, who upon hearing about Cajsa’s hypos, said only “okay, I’ll keep tabs on that because it may affect your driver’s license.” Fear of losing a driver’s license is not at all trivial for people with diabetes, and Dr. Choudhary cited data showing that when a rule to revoke licenses from those at-risk for hypoglycemia is implemented, reports of severe hypo went down by half almost immediately in a Danish study. He concluded his talk with a call-to-action, charging HCPs to advise patients/caregivers about severe hypoglycemia and glucagon without exacerbating stress and fear. This is a fine like to walk, and it’s no doubt easier said done, but it’s an imperative.

  • In CRASH, nearly a third of type 1s and nearly half of type 2s who experienced severe hypo did not possess a glucagon prescription. After their event, only 3% of type 1s and 1% of type 2s obtained a glucagon prescription. A major factor contributing to under-utilization of glucagon is the complexity of current options, which require a cumbersome mixing process prior to injection, and we’ve also heard from US patients that glucagon often represents an additional, pricey diabetes supply that they don’t anticipate needing to use – or feel others won’t be able to use, given the complexity. We imagine CRASH is one way for Lilly to set the stage for its next-gen nasal glucagon, submitted to FDA in 2Q18 with a decision expected in 2Q or 3Q19. An audience member inquired about the status of nasal glucagon during Q&A, and although the faculty couldn’t discuss the drug openly (since it’s not yet approved), they agreed with her that it will be a more patient-friendly option in hypoglycemia rescue therapy (compared to glucagon reconstitution kits). As Prof. Battelino put it, “there are seven steps to prepare glucagon before injection; I can’t even list a couple.” The next-gen glucagon competitive landscape also includes autoinjectors from Xeris (filed with FDA in 3Q18, decision anticipated by 3Q19) and Zealand (on track for 4Q19 FDA submission). We’re eager to see these advanced products on the market, as glucagon is one of the areas of diabetes therapy most in need of improvement, from our view.

  • Global incidence of severe hypoglycemia is only climbing as diabetes prevalence rises. In this same session, Professor Tadej Battelino showed that the number of severe hypoglycemic events requiring emergency medical services nearly tripled between 1997-1998 and 2011-2012 (which reflects the tremendous economic toll of severe hypo). He displayed WHO data on all-cause mortality related to severe hypo, which peaked in 2010 but has remained high since then (the odds ratio for death is ~1.5 comparing individuals who have experienced severe hypoglycemia vs. those who have not); Prof. Battelino attributed this to the growing type 2 diabetes epidemic. Dr. Choudhary commented that the number of type 2s on insulin now exceeds the total number of people on the planet with type 1 diabetes, reinforcing that severe hypoglycemia is a pressing problem in T1D and T2D alike, one that must be explicitly addressed by the medical community.

  • Ms. Lindberg told a moving story about her journey with chronic disease. She was diagnosed with type 1 diabetes at age 13 and was later diagnosed with brain cancer, which multiplied her severe hypos 10-fold. She described the intense fear she felt one night after a severe nocturnal low, sharing that she was “scared to go back to sleep and never wake up again.” Ms. Lindberg emphasized the need to disentangle hypoglycemia from feelings of failure. Prevention is key, and advanced technology can lessen the burden of hypoglycemia, but low blood glucose will happen anyway, she explained, echoing Dr. Choudhary’s charge that HCPs increase focus on glucagon training. “If I could pick one thing to be without,” Ms. Lindberg surmised, “it would be the hypos. They have a big impact on me physically, emotionally, and socially. They’re probably the scariest thing about type 1 diabetes.” Ms. Lindberg is President of the Swedish Diabetes Association and serves as a global advocate in non-communicable diseases.

Dr. Skyler Sweeps Over Beta Cell Replacement Landscape: Enthusiastic about Industry Commitment, Immune-Evasive Stem Cells

Wrapping up ATTD 2019, the venerable Dr. Jay Skyler gave an optimistic, sweeping overview of beta cell replacement landscape. Primarily driving his optimism was the increased industry interest and investment in the field in recent years, including most of the major pharmaceutical companies in diabetes (see below) – a key theme we identified in 2018. In his words, “When industry gets this involved, success can be predicted to be coming.”

  • In allogeneic stem cell research (cells are from a different host), Dr. Skyler highlighted the work of Semma Therapeutics and ViaCyte – two companies near the top of our competitive landscape. The former has garnered considerable attention for its “conformal coating” encapsulation device, which they hope will eliminate the need for immunosuppression and surrounds each naked islet cell to allow for better implantation and functionality (but remains preclinical). ViaCyte has brought two programs into the clinic, and the lead candidate right now is PEC-Direct, a device that allows for direct vascularization of graft cells to avoid the negative impact of foreign body response but requires immunosuppression. Proof-of-efficacy data from a phase 1/2 trial are expected as early as mid-2019. PEC-Encap – fully-enclosed beta cells in an implantable, semi-permeable membrane – is set to return to the clinic as early as mid-2019, with a revamped encapsulation device to reduce foreign body response developed in collaboration with W.L. Gore. The “last” generation, PEC-QT – immune-evasive stem-cells in a vascularized device – is being developed in partnership with gene-editing giant CRISPR Therapeutics and is currently preclinical. Dr. Skyler did, however, seem particularly bullish on the potential for hypoimmunogenic stem cells in the near future, pointing to a paper published in Nature Biotechnology only four days earlier suggesting that hypoimmunogenic cell grafts can be engineered for universal transplantation – wow! Moreover, another paper published just three weeks earlier in Stem Cell Reports proposed a new method for preprogramming cell death into cancerous cells forming during stem cell induction and selecting for pancreatic beta cells, potentially allowing for cleaner, safer cell differentiation. Both of these papers chip away at what the Diabetes Research Institute’s Dr. Cherie Stabler called the greatest hurdle in beta cell replacement therapy at the 2019 JDRF Mission Summit: producing a fully-functional stem-cell derived beta cell.

    • As Dr. Skyler explained, another major hurdle for beta cell replacement is producing the appropriate microenvironment mimicking a functional islet – neural cells, endothelial cells, pericytes, and many other types of cells – not just the beta cell – all interact to create a functioning islet. Session chair Dr. Des Schatz was noticeably nodding along with this assessment. On this note, we point to recent research from UCSF’s Hebrock Lab on the creation of “enriched beta clusters,” a more advanced cell type than other insulin-secreting stem cell-derived products – while these aren’t fully functional islets, researchers do seem to be making strides toward better stem-cell derived products, as well as on the best environment to house them.

  • Orgenesis was the only company mentioned conducting autologous stem cell research (cells coming from the same host) – a multi-step process that requires the modification of multiple cellular processes. The idea here is to revert liver cells to a suggestible stem-cell state, induce pancreatic differentiation, then fine tune the signaling pathway to allow generation of cells which secrete insulin – certainly no small feat. We haven’t heard much from the company in recent years and the project remains preclinical.

MannKind Poster Demonstrates Comparable Effects of BMI, A1c, Diabetes Duration, and Afrezza Use on Pulmonary Function

A poster from MannKind found that BMI, A1c, and diabetes duration had comparable effects on pulmonary function to inhalable insulin Afrezza. Over a two-year study period, Afrezza use was associated with the greatest decline in FEV1 score (a test of lung capacity, healthy values range from ~2-5 L depending on age and height) of any variable presented for that time period: 0.056 L in type 1 (p=0.019) and 0.041 L in type 2 (p=0.011), both vs. usual care. For comparison, each additional five years of type 1 diabetes at baseline was associated with 0.065 L decline (p=0.0028), and, in type 2, each additional A1c or BMI point was associated with a 0.047 L (p=0.0014) or 0.025 L (p<0.001) decline, respectively – see below. Therefore, although Afrezza was associated with the greatest transient effect on FEV1 – which has also been shown to be reversible after Afrezza discontinuation – the reduction in lung capacity is comparable to the effects produced by greater BMI, A1c, and diabetes duration. Moreover, although all of these effects were statistically significant, the poster’s results section stated that none were clinically significant. Doubtless, this poster was meant to put to bed any remaining concerns about the pulmonary effects of inhaling Technosphere insulin, and the lack of a clinically significant FEV1 decline could palliate HCP concerns.

As has been reported, the most common pulmonary reaction with Afrezza is cough, which occurs in about 1 in 4 individuals.  The product label reports a discontinuation rate of 2.8% based on the Afrezza clinical program and cough is reported to diminish over time for most individuals.  For additional perspective, our team recently spoke to Afrezza expert Mr. Mark Harmel – the CDE responsible for onboarding participants into the STAT trial – who stated that inhaling more slowly or sipping water prior to inhalation usually mitigates cough. While few would question the impressive glucose-lowering efficacy of Afrezza as an alternative to injectable mealtime insulin, lung-related risks have affected Afrezza specifically and inhaled insulins broadly since the market entry (and exit) of Pfizer’s Exubera. However, FDA recently removed Afrezza’s REMS program a year earlier than scheduled.  The REMS for Afrezza required the company to explain the risk of acute bronchospasm for patients with chronic lung disease taking the inhaled insulin to healthcare providers.

Dr. Ritzel Discusses a Hypothesis for Titration-Period Hypoglycemia Differences in Head-t0-Head BRIGHT Study (Toujeo vs. Tresiba), Building a Case for Clinical Relevance of Results

During a Sanofi-sponsored symposium, Munich’s Dr. Robert Ritzel discussed a hypothesis for the difference in hypoglycemia rates observed during the titration period of BRIGHT (Sanofi’s head-to-head study of Toujeo vs. Novo Nordisk’s Tresiba). As a reminder, BRIGHT data (n=929) presented at ADA 2018 showed that, during the 24-week study’s 12-week titration period, Toujeo was associated with a 26% lower risk of hypoglycemia <70 mg/dl (OR=0.74, 95% CI: 0.57-0.97, p=0.03) and a 37% lower risk of hypoglycemia <54 mg/dl (OR=0.63, 95% CI: 0.40-0.99, p=0.004) compared to Tresiba (see the Diabetes Care publication). However, there was no significant difference in hypoglycemia rates during the following 12-week “maintenance period,” and the somewhat high p-values and fairly wide confidence intervals during the titration period have been pointed out to us. With a transient benefit and no difference in either fasting glucose values or A1c, the field has been less-than-convinced of any clinically meaningful advantage with Toujeo over Tresiba. Still, Dr. Ritzel offered a possible explanation that highlights the importance of individual characteristics of each insulin: Given that hourly data from the titration period show the hypoglycemia differences occurred between ~4 am and ~12 pm, Dr. Ritzel hypothesized in conjunction with the published PK/PD data that Tresiba may have a slightly stronger effect on fasting glucose. In support of this hypothesis, he pointed out that both insulins were injected in the evening and have only small differences in their PD profiles, reasoning that the “highly dynamic” period of titration may allow each insulin to exhibit some different, inherent characteristics more readily.

  • The hypoglycemia results may seem more meaningful within the context Dr. Ritzel established in the first half of his presentation: A recent observational, retrospective analysis (n=40,627 type 2s) shows that hypoglycemia in the first 12 weeks of basal insulin use significantly predicts risk of hypoglycemia over 24 months (OR=5.71, 95% CI: 4.67-6.99). That said, to our understanding, this analysis did not attempt to control for hypoglycemia risk factors; despite the common aphorism that “hypoglycemia begets hypoglycemia,” this data alone is not convincing evidence of a causative relationship between early and later hypoglycemia – rather, hypoglycemia in the first 3 months is likely a marker of general hypoglycemia risk. To this point, we appreciated Dr. Ritzel’s closing remarks: “I continue to document hypoglycemia data and the initial response when I change or initiate insulin, and I try to incorporate that into the treatment strategy to achieve the patient’s goals long-term.” This could be a key lesson to take away from BRIGHT: A patient’s early experience on insulin is likely indicative of how things will go in the future, and HCPs can adjust treatment strategies accordingly.

Close Concerns' Thoughts

  • We continue to feel the most important aspect of next-gen basals Tresiba and Toujeo is that both are light-years ahead of first-gen basal insulins on both hypoglycemia risk and the stability/duration of action they offer many patients. On BRIGHT specifically, perhaps Dr. Alice Cheng put it best at EASD 2018 when she said, “I don’t think the big takeaway from BRIGHT is that it demonstrated non-inferiority in terms of A1c reduction between the two next-gen insulins. The big takeaway is that both insulins were able to lower A1c from an average above 8.5% to 7.0% in just 24 weeks.” There’s no doubt in our mind that more patients could benefit from access to either next-gen basal given that astounding data, and we hope comparative studies will offer insight on when, to whom, and how HCPs can best prescribe either agent to improve outcomes and experiences for people with diabetes.

  • Still, BRIGHT did raise the stakes for Novo Nordisk’s own study comparing Tresiba to Toujeo, and we expect to learn far more about how these next-gen agents stack up once those results are released in 2Q19. Overall, from a patient perspective, we don’t really care too much about the differences since patients won’t have choice anyway in most geographies – we just care that patients who experience hypoglycemia using Lantus, Levemir, NPH, etc. are offered one Of the two next-gen insulins. Standing in contrast to BRIGHT, hypoglycemia is the primary endpoint in Novo Nordisk’s study. From what we’ve seen – particularly today – Sanofi is striving to leverage BRIGHT’s hypoglycemia findings into a strong case for the use of Toujeo over Tresiba, but our sense is that most thought leaders either see them as neutral (see Dr. Cheng’s quote above) or possibly still favor Tresiba ever-so-slightly over Toujeo but believe it doesn’t matter too much since choice is out of their control. Indeed, if one is going to draw comparisons, it’s important to consider BRIGHT’s findings in the context of both agents’ broader profiles – including the fact that Tresiba’s longer duration of action allows for recommended dosing “once daily at any time of day,” compared to at the same time every day for Toujeo, which can add valuable flexibility to a patient’s daily insulin management (as we understand it, for some patients Toujeo can actually work better if taken closer to every 18 hours than 24 hours…).

A-One Study Finds 0.52% Absolute A1c Improvement with One Drop Premium + Afrezza vs. One Drop Premium + Injected Insulin

Results from the Mannkind-funded A-One trial (n=119 type 2s) found One Drop Premium (unlimited test strips and 24/7 in-app support from CDEs) + MannKind’s inhalable mealtime insulin Afrezza (technosphere insulin) to confer a 0.52% absolute A1c improvement as compared to One Drop Premium + current injected insulin over three months (0.93% vs. 0.41%). Importantly, the A1c reduction observed with One Drop + Afrezza is also 0.39% greater than Afrezza without One Drop (-0.55% A1c reduction), demonstrating additive benefit.

Notably, A1c stability or reduction was seen at all baseline A1c’s in the dual intervention arm in both the intent-to-treat and per-protocol analyses, while A1c was stable or rose with One Drop and injected insulin for all baseline A1c’s <8.5 in both analyses (see below). This could be related to less hypoglycemia risk with Afrezza, allowing for tighter glycemic control in this group relative to those receiving injectable insulin. One Drop announced the poster via simultaneous press release.



  • The study was originally set to randomize 400 people with type 2, and we’re wondering what accounted for the drastic (70%) size reduction. The randomization flow diagram states that 265 people were originally randomized, with major dropout reasons including no longer interested (n=21), lost contact with subject (n=49), and unable to receive prescription (n=67).

  • We see these results as beneficial for both parties: One Drop garners evidence of compatibility across the spectrum of mealtime insulins while MannKind has another study in its pocket demonstrating benefit over injectable competitors (the first being the head-to-head STAT study vs. NovoLog). Also, exclusion criteria specified that participants could not have had previous experience with either intervention, making a case that switching to either alone – or even better, both – could be beneficial.

Type 1 Biomarker Update from Dr. Des Schatz: Low Neutrophil Count (Innate Immunity Defects), Pancreas Volume, IGFs As Emerging Pieces of Early Detection and Prevention Puzzle

University of Florida’s Dr. Desmond Schatz gave a fantastically comprehensive overview of the known and emerging biomarkers for type 1 development and progression, highlighting recent progress on the roles of innate immunity and pancreas size. On the former, previous research has identified significantly lower neutrophil (an immune cell) counts in autoantibody-positive relatives of people with type 1 or new onset type 1 diabetes, but more recent work has found that the transcriptomic signature for neutrophil count might also be altered in first-degree-relative but autoantibody-negative subjects (age 5-18), suggesting it could serve as a biomarker alongside genetic risk even prior to autoantibody formation. With respect to pancreas volume, a great presentation at ADA 2018 demonstrated a clear downward trend from age-matched controls, to autoantibody positive first-degree relatives of type 1s, to those with type 1 diabetes. Earlier in 2019, this spectrum was expanded to include first-degree relatives without autoantibodies, first degree-relatives with one antibody, and first degree relatives with >2 antibodies – demonstrating a semi-linear relationship between decreasing pancreas size and progression through type 1 risk, though with an apparent levelling-off after one autoantibodies is present (see below). Interestingly, after type 1 onset, there is no correlation between duration of diabetes and relative pancreas volume; the same goes for serum trypsinogen – a biomarker for pancreas volume also shown to decrease in autoantibody positive subjects. Doubtless, further research into these preceding events and a critical mass of broad type 1 biomarker research are both needed to help complete the puzzle (likely also with the use of “big data”) of onset and prevention. To this end, Dr. Schatz briefly highlighted some emerging, unpublished data from his lab suggesting lowered insulin-like-growth-factor-1 (IGF-1) in at-risk, new onset, and established type 1s, as well as decreased IGF-2 levels in only at-risk and new onset individuals (see below). Only adding to the puzzle, Dr. Schatz concluded his talk by reminding the audience that type 1 may progress through different, age-related mechanisms, based on autoantibody incidence. This fascinating Diabetologia paper from the TEDDY study captures the concept well; incidence of different autoantibodies (GADA and IAA, specifically) and their chronology varies by time, with IAA spiking earlier, suggesting multiple pathways to type 1 development.

DEPICT Post-Hoc Examines Impact of Baseline MAGE on Farxiga-Driven Improvements in A1c, Glucose Variability

Prof. Phillip Moshe presented a subgroup analysis of pooled DEPICT data (AZ’s Farxiga in type 1, n=1,591), examining the impact of baseline MAGE. As shown below, when splitting the cohort at the 33rd, 50th, and 66th percentiles, there was no consistent association between baseline MAGE and adjusted mean change at 24 weeks (first figure below). However, the figure supplied in the online abstract (also below) displays the full spectrum of baseline MAGE values and corresponding changes, with consistent patterns between DEPICT-1 and DEPICT-2: These data indicate that patients with progressively higher baseline MAGE saw greater drops in variability – a finding that makes sense, given patients with higher variability at baseline simply have a higher base from which to fall. However, there was no correlation between baseline MAGE and change in A1c at 24 weeks. The sum of these findings is encouraging: The glucose variability (and, ostensibly, time in range) benefits of dapagliflozin aren’t limited to any given subgroup of baseline variability. We would have been interested to see a subgroup analysis examining baseline MAGE and DKA risk, though our understanding from Prof. Moshe was that there is no detectable association. As regulatory approval for SGLT inhibitors (particularly Farxiga) in type 1 marches slowly onward, the field is eager for data that can inform patient selection for both greater benefit and lower DKA risk; and while general consensus guidelines for patient selection have been published, we imagine many HCPs could benefit from more precise guidance.


Renal Protection in Type 1? Dr. Groop Cautiously Points to Renal Hemodynamic Changes – Also Observed in Type 2 – as Early but Promising Suggestion of SGLT-2 Mediated Nephroprotection

University of Helsinki’s Dr. Per-Henrik Groop offered mechanistic data indicating that the renal protection of SGLT-2 inhibitors well-established in type 2 diabetes could quite possibly translate to both type 1 diabetes and non-diabetic CKD, with the caution that clinical outcomes data are still to come. Most importantly on the type 1 front, Lilly/BI’s EMPA-KIDNEY renal outcomes trial (n=~5,000) is now enrolling all comers with kidney disease, including type 1s, type 2s, and those without diabetes who meet the CKD criteria (expected completion is June 2022). While it remains to be seen how large the study’s type 1 cohort will be, the expansive inclusion criteria are undeniably exciting, will offer the best evidence to date on long-term outcomes with SGLTs in type 1, and stand in contrast to both (i) J&J’s completed CREDENCE renal outcomes trial (type 2 diabetes only) and (ii) AZ’s ongoing Dapa-CKD trial (CKD with or without diabetes, but no type 1) – see below for a rundown of renal outcomes trial designs, including eGFR criteria. Kudos to Lilly for figuring out how to have type 1s in these long-term outcomes trials.

Despite his assertion that “we actually don’t know” what SGLT-2s will do to renal function in type 1s, Dr. Groop reviewed preliminary but compelling clinical data offering strong support for Lilly/BI’s decision to include this population in EMPA-KIDNEY. Namely, in type 1s, eight weeks of empagliflozin treatment has been shown to both (i) improve renal hyperfiltration (i.e., reduce glomerular filtration rate); and (ii) reduce interglomerular pressure by ~6-8 mmHg, in patients with existing hyperfiltration during both euglycemia and hyperglycemia (p<0.01 and p<0.0001 vs. baseline, respectively). No negative renal effects have been observed in type 1s with normal renal function. Moreover, Dr. Groop continued, it seems the reduction in renal hyperfiltration is driven by a significant reduction in renal blood flow and an increase in renal vascular resistance (both p<0.001 vs. baseline at 8 weeks). These last two effects are key, he showed, because they’re “consistent with afferent arteriole vasoconstriction” – the key mechanism of the “Tubular Hypothesis” proposed by Drs. Hiddo Heerspink, David Cherney, and others in a key 2016 paper in Circulation – see the full cascade of effects outlined in the second figure below. As such, it seems that SGLT-2s have a similar effect on the renal physiology of people with type 1 diabetes as they do on those with type 2, lending promise to the use of SGLT-2s to slow the progression of renal disease in type 1s.

  • For the most recent data on SGLT-2s for renal protection in type 2 diabetes, see this meta-analysis of all three completed CVOTs (EMPA-REG OUTCOME, CANVAS, and DECLARE) in The Lancet, plus this compelling commentary from Drs. Subodh Verma and Javed Butler at AHA 2018. Additionally, we point you to recent post-hoc analysis of pooled DEPICT data (dapagliflozin in type 1), which demonstrated a dose-dependent reduction in urinary albumin creatinine ratio (UACR) in patients with albuminuria (defined as UACR ≥30 mg/g) at baseline (n=251) – preliminary and short term data, but promising nonetheless.

Dr. Tadej Battelino Criticizes EMA Indication for Dapa in Type 1s Only if BMI ≥27 kg/m2; Explores Potential CV Benefits to SGLT Therapy in T1D

Conference co-chair Professor Tadej Battelino was critical of the EMA’s potential new indication (if accepted) for SGLT-2 inhibitor dapagliflozin (Forxiga), which authorizes its use in type 1s only if they have a BMI above 27 kg/m2. He argued that the entire type 1 diabetes population could benefit from an adjunct therapy that offers additional A1c-lowering, improved time-in-range, and weight loss – not to mention possible cardioprotection. We would not disagree but just might point out that to start, a smaller group is advised to get the therapy. To illustrate the substantial unmet need for CV risk reduction among type 1s, Dr. Battelino displayed data (image below) on greater all-cause death, CV death, and hospitalization for CV events compared to matched controls. Prof. Battelino established the relationship between glucose variability (and especially postprandial glucose variability), oxidative stress and endothelial dysfunction, and elevated CV morbidity/mortality. The jury’s still out on the precise cardioprotective mechanisms of SGLT-2 inhibitors (in type 2 or type 1 diabetes), but Prof. Battelino outlined one hypothesis in which these drugs reduce glucose variability, thereby alleviating oxidative stress and endothelial dysfunction, which brings down CV event rate, though other well-regarded hypotheses focus on other mechanisms – though, surely, these are not mutually exclusive. He added, explicitly, that glucose variability is not correlated with obesity, so limiting dapagliflozin’s use to type 1s with BMI ≥27 kg/m2 is an oversight of the most meaningful benefit an SGLT-2 inhibitor could have for type 1s – CV risk reduction. Prof. Battelino acknowledged the uncertainty around mechanism of CV benefit, but focused attention on what is certain: “If you ask about the SGLT’s cardiovascular mechanism, the honest answer is ‘we don’t know,’ but we do know that people with type 1 diabetes are only getting worse with metabolic control and consequently CV outcomes.” While many say the exact reasoning behind the committee’s decision isn’t clear to us, it seems likely that it was motivated by safety considerations rather than efficacy considerations – we believe it is a very smart starting point and presumably could broaden if all goes well. We’ve heard from thought leaders that insulin sensitivity declines at higher BMI, which could protect against DKA. Insulin requirements are also greater at high BMI, and taking sufficient insulin alongside an SGLT-2 inhibitor (while avoiding hypoglycemia) is certainly important in preventing DKA.


  • Prof. Battelino touched on several other hypotheses for the mechanism underlying an SGLT-2 inhibitor’s CV benefit as well: shifting energy metabolism to increase circulating ketones; reducing cardiac afterload; relieving inflammation; decreasing blood pressure; and providing nephro-protection, which is closely tied to CV health. We are very excited to hear more interest in outcomes trials and type 1.

  • He also pushed for a pediatric indication for SGLT inhibitors as adjuncts to insulin, citing evidence to show that adolescents and children are “a very vulnerable group in type 1 diabetes.” As far as we’re aware, no large-scale clinical trials have been launched yet to investigate these agents in type 1s under 18 years-old. We imagine manufacturers are waiting to see how the drugs perform on the market for adults, especially considering the safety concerns surrounding DKA.

Dr. Choudhary Highlights Potential for Population-Level Impact with Next-Gen Basals While Acknowledging Adaptability of AID for Nuanced Adjustments

King’s College’s Dr. Pratik Choudhary affirmed the potential of automated insulin delivery for nuanced basal insulin adjustments while maintaining that improvements in MDI via next-gen basal insulins will have a more profound impact on population health. As he put it, ~30%-40% of people with type 1s are using pumps and/or CGMs and another few percent DIY closed-loop systems, leaving the majority relying on MDI regimens – and using basal insulin (not to mention type 2s on insulin, for which an even smaller fraction are using tech). To this end, the more stable PK/PD profiles and associated lower incidence of severe and nocturnal hypoglycemia of next-gen insulins (compared to first-gen basals, such as Lantus) represent major steps forward on a broad scale. Specifically, he reviewed outcomes from the EDITION (non-inferior A1c reduction with less nocturnal and severe hypoglycemia with Toujeo vs. Lantus), OPTIMIZE (improved A1c and treatment satisfaction with no hypoglycemia increase in type 1), and SPARTA (0.4% greater A1c reduction with Toujeo after switching from twice-daily insulin dosing) studies to underscore improvements of Toujeo over Lantus for both type 1s and type 2s. We would add to these DEVOTE, Novo Nordisk’s CVOT for Tresiba, which established a robust and impressive 40% risk reduction vs. Lantus on severe hypoglycemia and garnered Tresiba a first-ever hypoglycemia label update in the US; though this data is limited to type 2, and though it is not an indication, we still view it as quite meaningful. When you scale the hypoglycemia risk reduction next-gen basals offer to the millions of patients worldwide using basal insulin, the impact on safety, patient quality of life, and healthcare expenditures could be monumental. Dr. Choudhary emphasized that even as AID grows, a significant proportion of people with diabetes will remain who either: (i) cannot access the technology due to cost or other barriers; or (ii) do not wish to use these systems. We disagree slightly with the second – while this is of course true today, as it was with first-gen pumps and CGM, we believe far better diabetes management will make AID the killer app to use pumps and CGM for those that don’t. In other words, although early-stage AID is not for everyone, we believe that later-stage AID will be for a very sizable percentage of people with diabetes that can access it. Access issues will undoubtedly remain but we also believe this will improve over time. Ultimately, both scenarios render further improvements in MDI and basal insulins not only of great importance, but necessary for better population-level outcomes. All this said, he acknowledged that minute differences in basal adjustment, such as those made due to variable parameters like number of steps, are better captured (at the moment) by algorithms rather than insulin innovation, rightfully extolling the potential of AID – reminding us that there is no one-size-fits-all solution for any diabetes issue at the present moment.

Drs. Bailey & Heise Look to the Future of Diabetes Therapy, Highlighting Potential and Progress in Multi-Receptor Incretin Therapies and Glucose-Responsive Insulin

Dr. Cliff Bailey gave a master class in multi-target therapies, focusing on incretins in a rundown of noteworthy molecules – see below for a tremendously useful table of the different metabolic effects of nine (!) different incretin receptor agonists he covered. He opened with an emphasis on structure, drawing the distinction between (i) combinations/mixtures that add together two peptides with no physical linkage; (ii) chimeric peptides take epitopes of given proteins and mixing sections together (these seem to be growing increasingly common); and (iii) hybrid/fusion peptides that physically link two peptides together. And while, he says, there are apparently endless opportunities for forming and delivering peptides, there is a limit to how far you can push a receptor: “If you give a receptor a bit too much work to do, it might turn out to be sensitized, and you may stimulate the post-receptor pathways as a result.” All in all, Dr. Bailey’s remarks painted an enthusiastic picture of the potential dual- and multi-agonists hold in diabetes, but with the caution that novel molecules bring unknowns – a dynamic playing out already in the highly-efficacious but possibly hard-to-tolerate molecules we’ve seen from Lilly (tirzepatide, now in phase 3), Sanofi (SAR425899, just discontinued), AZ (MEDI0382, finishing phase 2), and others.

Additionally, Profil’s Dr. Tim Heise reviewed strategies developed to-date in the quest to create a glucose-responsive insulin: “The good news is some of these really seem to work, the bad is – only in mice [so far].” He outlined four primary scientific strategies:

  • Chemically modified insulins sensitive to glucose: These molecules take existing insulin analogs and add glucose-sensitive components. For example, insulin detemir plus a PBA molecule binds to albumin (inactive state) at low glucose concentration, but releases from albumin at high glucose concentrations.

  • Glucose-oxidase (GOx) based systems: Utilizing the same enzyme as glucose sensors, GOx systems require the removal of hydrogen peroxide byproducts through a catalase reaction. The movement of glucose into a microgel causes swelling (via electrostatic repulsion), and swelling enables insulin release.

  • Phenylboronic acid (PBA) systems: PBA forms reversible esters with the diols in glucose, and PBA-linked insulin in a “smart gel” can be delivered via a subcutaneous catheter. When glucose binds to the supply, hydration causes insulin release, but when the depot is dehydrated (hypoglycemic conditions), a skin layer forms over the gel and insulin cannot escape. This system has been tested, with success, in mice, but Dr. Heise cautioned that it may not be highly convenient for patients as it has been tested in subcutaneous catheters. Another approach with injectable PBA-based nanoparticles combined with sodium alginate showed promising results in mice with STZ-induced type 1 diabetes, but data in humans are still outstanding.

  • Glucose-binding protein based systems: Dr. Heise closed with the first-ever approach described, a glycosylated system based on competitive binding with glucose – specifically, the binding of glucose to lectin allows for the release of insulin. This became Merck’s first clinical GRI, MK-2640, which was discontinued in phase 1 due to insufficient efficacy and variable patient requirements (see published results).

He closed on a more imminently hopeful note: While the field has to wait for a “smart insulin,” smart pens are here now and stand to make diabetes management and insulin dosing seriously less cumbersome. Game on!

Prof. Danne Throws Weight Behind Blood Over Urine Ketone Testing, Highlights Need for Outcomes Data Supporting CGM-Based Metrics; Casts Potential of Closed Loop + Adjunct Therapy in Positive Light

Professor Thomas Danne made a case for the adjunct use of SGLT inhibitors for type 1 in the presence of appropriate technology, also giving a nod to the future consolidation of adjunct therapies with closed-loop insulin delivery. In the wake of the publication of the landmark consensus on DKA risk mitigation with SGLT inhibitors in type 1 diabetes (see Diabetes Care), Prof. Danne – who led the consensus effort with ATTD – was tasked with addressing the intersection of SGLT-2s and technology, a talk he opened by asserting that the field can make adjunct therapy work through technology. Two threads of this argument emerged during Prof. Danne’s presentation:

  • Continuous glucose monitoring can be used to demonstrate the full benefit of adjunct SGLT use, and it may even drive the use of adjunct therapy. With the rise of CGM, patients have 24-hour data on their blood glucose levels and are increasingly looking for ways to avoid postprandial spikes and hypoglycemia. The reductions in glucose variability offered by SGLTs are a huge draw for patients, and Prof. Danne emphasized the potential of adjunct therapy to help – “if we can do it wisely.” To be sure, there’s a strong relationship between the outcomes beyond A1c movement and the push for adjunct SGLT use in type 1, and Prof. Danne posited that considering just two values – time in range and time below range – could “to some degree … explain the whole glycemic profile.”

    • On balance, Prof. Danne was far less optimistic about the “uncharted ground” the field stands on with regulatory agencies and payors. Despite high-profile analyses (specifically, Drs. Roy Beck, Rich Bergenstal, et al.’s recent DCCT analysis) that have linked derived measures of time-in-range to long-term microvascular complication risk, he said, the field still needs to make a concerted effort to ensure CGM-derived metrics considered most valuable by the field are shown to be valuable at the level required by other stakeholders. As he put it, “Today we don’t have the data, and there’s a desperate need for the scientific community to prove this.” He didn’t elaborate on this point, but we imagine Prof. Danne envisions a DCCT-like, outcomes based study using CGM.

  • Ketone monitoring technology, with a strong preference for blood ketone monitors, can make the use of SGLTs in type 1 safer. While some preference for blood over urine ketone monitoring has emerged over the past year, Prof. Danne’s weighing of the pro/con profile of each (see image below) laid bare the superiority of blood ketone testing. Still, he admitted, there is no scientific data showing that one can’t safely use an SGLT inhibitor with urine ketone testing alone (just as so many questions in this arena haven’t been answered in a data-driven way). However, the greater precision of blood testing translates to much clearer direction on what to do at any given ketone level. If cost is not a barrier, blood ketone monitoring should always be chosen not only for on-treatment monitoring, but also for establishing a baseline ketone level and ensuring the patient knows how to measure ketones. Ultimately, he said, “We have to use the technology and move away from urine ketones.” While we do understand this view, the access aspect can be challenging; one patient we know recently paid $600 for 100 blood ketone strips.

  • On a final note, Prof. Danne looked toward the future of type 1 care, “We can’t be here at ATTD and not talk about closed loop,” referencing the single-day DAPA-DREAM pilot study originally presented at IDF 2017 (also see this EASD 2018 poster). As background, the crossover study combined DreaMed’s fully closed loop with twice-daily dapagliflozin or placebo, testing unannounced mixed meals in 15 young adults (ages 18-20). The data show an additional very compelling 2.8 hours per day in range (70-180 mg/dl), 100% time in range overnight, and a 40% reduction in bolus insulin with dapagliflozin, though in an admittedly very small and short study (see full results just below). Larger studies will be needed to determine how closed loop use might impact DKA risk (pump use is a well-recognized risk factor from phase 3 programs of SGLTs in type 1) and whether new, unique algorithms would be beneficial in this use case – but it’s certainly encouraging to see that SGLTs can likely further improve the advances poised to come with closed loop.

Exhibit Hall


We confirmed with Abbott that the recently announced integration of insulin dosing data from Novo Nordisk’s connected pens directly into the FreeStyle LibreLink app and LibreView diabetes management systems will become available between late 2019 to early 2020, providing more specific timing than that originally noted (“as soon as possible”). The booth representative confirmed that the FreeStyle Libre 2 has only launched in Germany (in line with October’s CE Mark), implying that the “gradual European rollout” may not have commenced. We last heard during Abbott’s 1Q19 call that FreeStyle Libre 2 has been submitted to the FDA as an iCGM. No updates on this front were shared.


Arecor brought a small, three-person booth to showcase preclinical results for its ultra-concentrated (U1000) rapid-acting insulin and ultra-rapid-acting prandial insulin. Of note, the ultra-rapid-acting candidate is set to enter a European phase 1 trial in “the next few months.” This is a delay from Arecor’s original timeline estimating 2018 human clinical trials, and it has been updated on Arecor’s product page. The U1000 candidate is on a slightly longer timeline and will likely enter the clinic in “late 2019 or early 2020”; we note that Arecor also originally slated the candidate for 2018 clinical development, so this is a substantial delay. Should it meet this new timeline, the candidate would likely be the first-ever U1000 mealtime insulin to advance into clinical development. Worth noting, representatives mentioned that they were meeting with major insulin manufacturers Sanofi, Lilly, and Novo Nordisk at ATTD to talk about micro-pump concepts for the U1000 insulin, expressing stronger enthusiasm for this candidate than the ultra-rapid molecule. Representatives did not mention Arecor’s liquid-stable glucagon or preclinical combination products for diabetes referenced in a recent, ~$8 million round of fundraising.


BD’s “Diabetes Personal Assistant” app, Briight, released at ADA, now has “over 30,000” users, reflecting a steady increase from 20,000 users as of AADE in August. BD’s Digital Diabetes Product Leader Mr. Ed Liebowitz shared that the minimal viable product for the app has now been validated, demonstrating that optimized structured education can increase engagement. Bright currently has 4.1/5 stars on Google Play (30 ratings) and 4.3/5 stars on the App Store (12 ratings). The Digital Health team is working on version 2.0 of Briight, which he expects to be live this calendar year. Mr. Liebowitz noted that Briight is still intended to support both BD’s injection and infusion patients, alluding to the type 2 patch pump. Last we heard on BD’s 4Q18 call earlier this month, the pump was submitted for US and EU regulatory approval, with an initial “early product introduction” still slated for “late in calendar year 2019.”


This ATTD, Biocorp unveiled new branding for its two-piece dose capture pen attachment (“Mallya”; formerly “Easylog”) and announced its first two diabetes partnerships (to our knowledge) with AgaMatrix and DreaMed. AgaMatrix will be the non-exclusive distributor of Mallya in the US, EU, and UK, and the two companies also signed a co-development agreement to develop “a breakthrough innovation.” Given AgaMatrix’s work in closed loop (WaveForm), we wouldn’t be surprised if Mallya is used to inform open loop insulin delivery. On the commercialization front, reps told us that they expect AgaMatrix to begin distributing the CE-marked pen attachment in Europe – starting with UK, France, and Germany – beginning in 1H19. As for the DreaMed collaboration, Biocorp will provide insulin dosing data to inform the MDI/basal-only versions of DreaMed’s Advisor Pro clinical decision support, with clinical trials set to begin this year.

Cam Med

Cam Med – developer of the ultra-thin, flexible, bandage-like Evopump capable of delivering multiple drugs at minimum cost – showcased its interim 1 ml prototype device (see below) and gave timing updates at its one-person booth. CEO Mr. Larry Alberts shared that the 2 ml beta prototype of the device, will be done in March, then tested in a preclinical model for two months. Provided positive results are achieved, the product will progress through human factors studies and device evolution to support an IDE application by the end of 2020. Boasting accuracy within ±5% at individual NovoLog doses of 5 mL and 1 mL and built-in pressure and temperature measurements to ensure proper delivery under varying environmental conditions, Cam Med has inspired investor confidence to-date. When we were first introduced to the company at ATTD 2017, it had won several competitions and secured $440,000 in non-dilutive funding; now, it has received ~$1.6 million in prizes and grants and $1.5 million in seed equity funding. Notably, the project is currently being funded in part by JDRF, with potential continuation of the partnership after preclinical studies. Of note, the eight distinct reservoirs within the patch enable potential multi-drug delivery (e.g., insulin + Symlin, GLP-1, and/or glucagon), and Mr. Alberts noted that he had struck up conversations on this topic before and during ATTD with developers of these formulations. JDRF’s funding announcement also explicitly stated artificial pancreas development as an end-goal. On cost, he asserted his hope for the Evopump to compete with pens. Currently, he said, each Insulet Omnipod costs roughly $12 to make; Evopump can currently be manufactured for less than $2, with the hope to get down to $1 through volume – wow! This is in part due to the electronic controller – which will have a simple UI containing a light, bolus-button and vibration in addition to having a Bluetooth module enabling communication with a handheld device – being reusable between patches. There are many barriers to entry in patch pumps, but among the biggest is extremely high-quality manufacturing and scale. Can Cam Med make it happen, where others have struggled?


Capillary Biomedical

We learned at Capillary Biomedical’s booth that the company no longer anticipates launching the wire-reinforced, multi-ported (three side holes) SteadiSet infusion set in 2019. On the other hand, while the 2019 launch was going to be for three-day wear, Capillary now ambitiously expects to leapfrog straight to a 2020 launch of seven-day wear. See our August coverage for a deep dive on Capillary Biomedical and the SteadiSet. Also new at this booth was a picture of the sleek, one-handed inserter (see below), which has a similar feel to FreeStyle Libre, Mio Advance, and G6. Nice!


A Cellnovo booth representative confirmed that the Gen 3 system was submitted to the FDA “early this year,” aligning with timing shared on the 3Q18 call in October. The Cellnovo booth touted its once-refillable cartridge, which launched last month. The booth representative hopes this feature will make Cellnovo’s pump more attractive to type 2 patients. Currently Cellnovo is integrated with two Bluetooth-enabled BGMs (Roche AccuChek Guide and Fora). CGM integration is in the pipeline, and Cellnovo is in discussions with Dexcom “and others.” However, users can view their CGM data via Apple Health in Cellnovo’s newly launched Cellnovo Monitoring app. We also received updates on two out of the three AID studies in which Cellnovo is involved. Per the booth representative, TypeZero will “propose a product” in 2021. Last we heard at EASD in October, Cellnovo was “getting ready for a trial” with TypeZero. Cellnovo and Diabeloop are currently working together on the business details of including Cellnovo in Diabeloop’s commercial system. Although the first arm of the Diabeloop study was completed with the Cellnovo pump, the second arm was completed with the Kaleido pump. As of CES in January, Diabeloop plans to launch its CE Marked adult hybrid closed loop system (DBLG1) in early 2019 in France ahead of a broader European scaling in 2Q19, using Dexcom G6 CGM and Kaleido’s patch pump. (Following ATTD, Cellnovo ran into financial trouble; more details here.)

Convatec (Unomedical)

Unomedical showed a number of brand-new infusion set concepts in its booth, including sets for dual-hormone delivery, a patch pump base plate design leveraging the all-in-one Mio Advance inserter, and a combined set-CGM sensor design. Pictures are included below. A poster also showed encouraging data from the ongoing Stanford study (n=24) testing up to 10-day wear with the coated Lantern set that includes slits along the sides. Interim outcomes (n=10) look encouraging, as 100% of sets have lasted seven days or more (mean: 8.9 days) and half of sets have gone for 10 days. There is still unexplained hyperglycemia (see table below), and clearly quite a bit of variability between different people. A follow-on, larger crossover study (masked, randomized) will compare the coated Lantern catheter against commercially available infusion sets. Based on these interim outcomes, we’d guess the company will go for seven-day wear and not ten. Unomedical is also working with Medtronic on a separate seven-day wear set that uses a different technology; a launch is ambitiously expected in one year, according to Medtronic’s presentation. We also asked about a Mio Advance launch in the US – the excellent inserter has been available OUS since last ATTD and FDA cleared for nearly a year – and the team told us it is still building capacity to support a US launch. This is prudent, as Mio Advance is such a tremendous upgrade and we know demand will be sky high. Medtronic does have an exclusive deal on Mio Advance for an unspecified amount of time, a downside for Tandem.

  • Coated Lantern study at Stanfordinterim data from n=10 patients, with two patients excluded in red:

  • Unomedical also debuted some never-before-seen infusion set concepts: (i) dual hormone delivery (Beta Bionics?); (ii) a dual CGM sensor/insulin catheter set with a side-by-side design and Mio-Advance-like insertion (2+ years away, per Medtronic’s pipeline); (iii) a patch pump base plate that uses the Mio Advance insertion device (looks ideal for Roche’s Solo); (iv) an all-in-one inserter like Mio Advance, but with the tubing bundled into the lid; and (v) an all-in-one inserter like Mio Advance, but with an angled insertion.



Dexcom’s booth advertised G6 at its first ATTD, along with the Clarity mobile app that launched internationally last fall. The Clarity team in the booth shared excitement for the upcoming launch of “On the Bright Side” notifications (read more details here), allowing users to set time-in-range goals and be notified (via automated push notification) when the goal is achieved – a beautiful way to bring a Bright Spots focus to CGM data! (Historically, CGM has been focused on the opposite: notifying users when they’ve done something wrong or made a mistake.) The rep explained the tough balance between notifications and nuisance, and the team spent a lot of time deciding what time window to look back on for the “Best Day” notification – seven days was the right balance. See the slide below, taken from Dexcom’s symposium:


ATTD start-up grant winner Dianovator promoted its proprietary insulin algorithm (patent pending), which not only doses insulin but also predicts the PK/PD profile of different insulins within the body. CEO Dr. Fredrik Ståhl founded the company based on his dissertation work on insulin algorithms, which is now incorporated into a beta system with a CE mark application coming in the next few months. Most notably, the ability to predict the PK/PD profile of insulins has the potential to enable earlier hypoglycemia alarms than current systems. A retrospective analysis to test this hypothesis is currently underway, with an RCT planned if this first step i successful.


According to reps at the DreaMed booth, Advisor Pro hasn’t actually started rolling out in the US or Europe through Glooko’s Population Tracker. Why? It’s mostly due to IT departments at clinics, who have a lot of say regarding software and are apparently proving to be a “big obstacle.” That said, there are a dozen or so clinics who are interested, and DreaMed is speaking with them one-by-one to see if they can clear all necessary hurdles and get their clinical decision support tool in front of providers. We’re hoping they’ll share their negotiation best practices with the rest of the field so that each new company doesn’t have to reinvent the wheel in terms of pitches to IT crews and figuring out the right business model. On that end, it sounds like DreaMed is looking to offer clinics a three-month trial, followed by a three-year discounted rate. For clinics that use Glooko, that probably means a premium monthly subscription fee on top of that paid for Glooko (we’ve even heard of a possible Glooko Gold, Glooko Silver, and Glooko Bronze setup); through Tidepool, which is free to clinics, DreaMed would still likely charge a set monthly fee depending on the number of patients (i.e., one fee for clinics with 0-200 patients, a higher fee for clinics with 201-500 patients, and so on). As with almost all adoption curves, this one sounds like it’s going to take time, but we’re glad DreaMed is on the front lines making headway.


Glooko’s 2018 Annual Diabetes Report (single-page infographic) was the primary focus of its exhibit hall booth. The report analyzes one of the largest bodies of combined diabetes data in the world: nearly 15 billion data points collected from over 2 million people with diabetes in 17 countries – up ~6.5 billion points, half a million people, and two countries from 2017! Once again, Belgium had the best glycemic metrics across the board, with an average blood glucose of 156 mg/dl; the highest was 198 mg/dl, in New Zealand, once again displaying great disparity in glycemic control across geographies. Belgium also had the lowest rates of hypoglycemia (5.8%) and hyperglycemia (35.4%) – both lower than last year – while Spain (21.8%) and New Zealand (77.4%) occupied the top spots for each, respectively. The most common times for hypoglycemia and hyperglycemia were Thursday at 2 am (8% of readings <70 mg/dl) and Saturday at 11 pm (48% of readings >200 mg/dl), respectively. The most common time to check blood glucose was 7 am, and the least frequent was 2 am.



As a reminder of just how little patients rely on data to manage their diabetes, fingerstick frequency ranged from 2.7 readings/day in South Africa to 4.4 readings/day in the Netherlands – the exact same as last year. (This could partially reflect growing CGM adoption, especially in Glooko users.) Yet, perhaps unsurprisingly, the number of blood glucose tests and average blood glucose during the week correlated inversely, as the two days of the week with the lowest number of blood glucose tests (Saturday and Sunday) were the two that looked to have the highest average blood glucose.


Christmas Day leap-frogged Valentine’s Day and Halloween as the holiday with the highest average blood glucose values. However, average blood glucose on all of the holidays seen in the figure below rose considerably in 2018; left to right, 2017 values were 168 mg/dl, 158 mg/dl, 140 mg/dl, and 131 mg/dl. What accounted for these changes? More stress and less vigilance on these days than last year? Shifts in the population studied that reflect more or fewer users celebrating these holidays? Likely the latter. Intriguingly, the report notes that the best and worst days (highest and lowest percentage of readings between 70-144 mg/dl) were September 28 and New Year’s.


The two countries with the highest average blood glucose – Australia and New Zealand – were also the two countries with the least number of sleep interruptions (1.4% and 3.2%, respectively; measured by having at least one glucose BGM measurement between midnight and 6 am). France came in third but also had the second lowest average blood glucose, suggesting that the higher blood glucose values sign in Oceania might be more related to geographical or cultural differences than sleep patterns.

Altogether, we cannot wait to see more analyses like these, and we salute all companies working to publish real-world data. In the future, we’d love to see an even further expanded report including details of device distribution over geography and time, demographic breakdowns, and situational context for results (e.g., what are clinics/providers/countries with good outcomes doing to manage patients on a population level?)


This was Insulet’s third major European exhibit hall appearance – following ATTD 2018 and EASD 2018 – and the second after taking European distribution over from Ypsomed last July. The new Omnipod Dash PDM was on display in a corner of the booth, though the focus was clearly on the current Omnipod – no surprise, given recent remarks delaying a Dash international launch to late 2019/early 2020. In fact, the reps were not even aware the Dash has a CE Mark!


LifeScan’s exhibit touted the relatively new OneTouch Reflect BGM, featuring the Blood Sugar Mentor that provides trend analysis and tips on both the meter and phone app. According to reps at the booth, the meter has launched in both Germany and France, Italy is on deck, and Belgium and Austria will see the product soon thereafter. As of October, OneTouch Reflect is with both FDA and Health Canada, with FDA clearance potentially expected by this fall. We learned about some more of the insights on Monday at a LifeScan symposium.


Medios Technologies housed a small, corner booth to showcase its AI retinopathy detection device and app; an iPhone 6 attachment (looks like a nail gun, see below; both are shipped together – small modifications have been made to the phone’s camera to promote better fundus photography) allows photos of the retina to be taken without the need for pupil dilation. The whole process takes about 15 seconds – ~10 to place the device to the eye and align it properly, ~5 to analyze the image. The app assesses both image quality (prompting for retake if need be) as well as probability of having retinopathy – it was 96% sure that one of our esteemed associates had healthy eyes. However, if she had been diagnosed, a detailed report would have highlighted areas of concern in the fundus photo. Altogether, we found the process quite quick and pleasant, at least compared to the traditional ophthalmoscope which requires dilation, a severely bright light, and uncomfortable chin placement. When asked about timing, representatives told us that the device part of the system is approved in the US, but the algorithm is currently being investigated with “a few more months” to go. That said, the display boards of the booth touted the algorithms specificity – 93% sensitivity, 92% specificity in a 900-patient sample from Diacon hospital (data currently under submission), although to our understanding, the diagnoses made aren’t prescription level. The idea instead would be to have the device in PCP offices as a way to quickly identify those who need referral to an ophthalmologist, similar to the way Verily/Google are proceeding in India and Thailand. Doubtless, a shortage of specialists and lack of referrals – issues multiple companies are currently working to remedy through AI-based systems that can be utilized by PCPs – play a role in eye exam uptake, and we are thrilled to see another company producing such a quick and simple device.



Medtronic’s expansive booth advertised the MiniMed 670G at its first ATTD following the international launch at EASD. Unlike the US, the outstanding Unomedical Mio Advance infusion set inserter was advertised alongside the 670G – we are highly anticipating this to be available stateside following under-the-radar FDA clearance nearly a year ago. In the iPro2 professional CGM section, a new Medtronic brochure advertised results from the very cool Adjust study (see above): after quarterly blinded CGM applications over one year, each with a follow-up visit (in-person or by phone), A1cs dropped by a mean of 1.3% (baseline: 9.4%). In real-time CGM, the Guardian Connect app was on display for Apple iOS, and we confirmed with Medtronic that it has had select availability on some Android phones internationally. The Google Play listing for Guardian Connect, however, has tragically low reviews – 1.3/5 stars, 268 reviews, 5,000+ downloads – criticizing poor performance and compatibility with the latest Android OS and models. Medtronic told us (via email) an upcoming update will expand availability. Guardian Connect Android will be key for Medtronic’s international CGM expansion, as Android is more dominant OUS and competitors all have Android compatibility (Abbott FreeStyle LibreLink, Dexcom G5/G6, Senseonics Eversense). As of JPM, Guardian Connect on Android is expected to launch between April 2019-April 2020.


Shanghai-based Medtrum boldly advertised three integrated patch pump and CGM products, including a predictive low-glucose-suspend system reportedly available in six European countries and with ~200 users so far (Germany, Denmark, Sweden, UK, Spain, Italy). The current A6 system includes a semi-disposable tubeless patch pump, a handheld controller, a seven-day CGM (two calibrations/day), and predictive low glucose suspend. The A7 system adds a secondary display mobile app and improves the CGM to 14-day wear and one calibration/day, with an expected launch this year in Europe and “2020” in the US (seems highly ambitious). The P7 system, currently in R&D, adds direct smartphone control (no handheld) and a no-calibration CGM. Like Roche’s Solo, Medtrum’s tubeless patch pump is semi-disposable – in this case, the reusable part (colored) reportedly lasts four years, while the disposable part is replaced every three days. The company also offers its Bluetooth-enabled S7 CGM as a standalone product (countries unclear). This was Medtrum’s largest booth following smaller and less-confidence-inspiring EASD exhibit hall appearances (e.g., 20182015). We’re not convinced of the CGM’s accuracy nor the ability to reliably scale the patch pump’s manufacturing. Still, it is notable that Medtrum owns its own tubeless patch pump and CGM – a combined offering that no other company has at this point.

Mellitus Health

FDA-cleared and CE-marked clinician-facing dose titration software Insulin Insights was the focus of Mellitus Health’s one-person display. The software, which showed highly robust reductions in A1c at three (-1.9%) and six months (-2.4%) from a high baseline (10%) in six month data presented at ADA 2018, is based on the ADA standards of care and equipped to process “all approved insulins.” Connected BGM readings with an associated mobile app are analyzed with respect to eight different insulin regimens, generating a recommended insulin adjustment. Most recently, the company announced a partnership to integrate Insulin Insights with Smart Meter’s cellular-enabled iGlucose BGM, though this partnership was not mentioned in our discussion with representatives.

Novo Nordisk

Novo Nordisk’s ATTD booth, emblazoned with bold blue text, “Digital + Health,” was all about the CE marked, NFC-enabled NovoPen 6 and NovoPen Echo. Crowds gathered around a small table to watch reps perform a sham insulin injection with a NovoPen 6, and then hold the pen against the surface of a Sony smartphone, thereby transferring the injection history directly to the Diasend app. The app tells the user when the upload is complete (just a few seconds), and then gives a menu of insulin options for the user to indicate what was just injected. We imagine future pen-app pairings will be able to distinguish between types of insulin doses; this feature will be especially an especially useful risk mitigation consideration, particularly when these apps adjust/recommend insulin doses based on past glucose and insulin data. On day #1 of ATTD, Abbott and Novo Nordisk both announced a partnership to integrated connected insulin injection device data into FreeStyle Libre software “as soon as possible”; Novo Nordisk already had similar partnerships with Dexcom, Roche, and Glooko.


Percusense – headed by CEO Mr. Brian Kannard and Dr. Rajiv Shah – is developing a multi-analyte sensor platform for diabetes: glucose, ketones, oxygen (at the infusion catheter to assess infusion set viability), and lactate and oxygen (to detect worsening comorbidities). From a CGM perspective, the company intends for a 14-day wear, factory-calibrated device. A poster in the ATTD booth also gave more details on a three-day integrated sensor and infusion set (picture below). The poster touted: (i) Low-cost sensor manufacturing through high volume microelectronics processes; (ii) a 30-minute warmup period; and (iii) innate interference rejection (insulin, glucagon, acetaminophen), precluding the need to rely on costly membranes.


The Roche booth displayed its highly-anticipated Accu-Chek Solo patch pump, which has launched to ~200 people in Austria, Switzerland, Poland, and the UK. Roche expects the pilot launch will stay fairly limited and “definitely” will not exceed 10,000 patients in 2019. A booth representative shared that a pre-submission of Solo to the FDA is “running.” The rep added that Solo has received “excellent feedback” from patients so far. We were impressed by the very sleek on-body form factor, but the demo confirmed our view that the insertion process is more burdensome and complicated than Omnipod (see multiple components below). Solo consists of: (i) a plastic baseplate with adhesive; (ii) a 200-unit disposable reservoir – the representative emphasized that patients do not have to fill up the entire 200U so as to mitigate insulin waste; (iii) the reusable pump itself, consisting of the electronics, which last 120 days; and (v) a touchscreen, Bluetooth-enabled handheld controller with an integrated BGM. Solo is recommended for 3-4 days of use, although the representative underscored that this is not due to adhesive wear or battery life, but instead related to insulin absorption and utilization. The inserter is reusable (it looks like a computer mouse) and lasts for two years, and the pump holder and cannula (available in 6 mL and 9 mL) come together as a sterile package. Users are guided through the insertion process on the handheld controller. The handheld also includes a camera to scan a unique code located on the pump base for pump activation and pairing. The booth representative expects that the “last phase” of Solo will include pump control via a smartphone app, which will also integrate CGM data. We wonder if Roche may wait for this final version to launch in the US – the current controller doesn’t come close to Insulet’s excellent user experience on Omnipod Dash. The booth representative also anticipates “several cloud-based platforms,” including Glooko and Roche’s “in-house solution,” to be compatible with Solo.

  • In addition to the section on Solo, the Roche booth also had portions dedicated to mySugr and Senseonics’ 180-day Eversense XL CGM. Notably, “Roche” and “mySugr” were printed at equal size beneath “Accu-Chek” on the booth banner, reflecting Roche’s commitment to its digital ecosystem. At the Eversense XL section of the booth, representatives performed sensor insertion/removal demonstrations. We learned that Eversense XL just received reimbursement from three major insurance companies in Germany – a big win, which the booth representative believes will help to increase adoption. As expected, Senseonics recently renewed its distribution agreement for Eversense XL with Roche, adding 17 new countries.


Photo-based meal-logging app developer Snaq demoed its beta version at a small booth with fake food. Based on a photo, the app can identify food, count nutrients, and estimate weight, with the last being the product’s differentiating selling point. For example, in our demo, Snaq correctly identified 37 g of fake risotto and allowed the user to add the mushrooms and tomatoes present in the concoction that the camera couldn’t pick up. Given how difficult carb counting can be, especially with meals not prepared by the person themselves, this ability to estimate weight could significantly improve the accuracy of carb counting and insulin dosing, and the company is currently looking for partnerships to investigate titration and bolus calculation in the long-term future. Eventually, representatives are hopeful that insulin dosing recommendations could be made based on an iterative learning process – how a particular user responded to a specific food and insulin dose previously informs their next dose – an exciting concept for sure!


At the SOOIL booth, we learned that the company is no longer planning on bringing its smartphone-controlled Dana RS insulin pump to the FDA with a version of the open source OpenAPS algorithm as previously announced at ADA. Instead, SOOIL plans to submit as an ACE insulin pump (referencing FDA’s newly created pathway for interoperable pumps) and develop its own proprietary algorithm. It’s not clear if SOOIL still plans to submit with an open communication protocol in addition to registering as an ACE pump; given that Loop will be submitted to the FDA, it may be a bit redundant, although we would have liked to see a pump designed specifically around the needs of the DIY community. SOOIL has a six-month clinical trial planned in the US and ambitiously expects to launch by the end of 2019. The booth representative was uncertain as to whether the pump will be submitted as a standalone device or with SOOIL’s proprietary algorithm. Following ADA 2019, SOOIL plans to register its own algorithm and CGM within four months. This is the first we are hearing that SOOIL plans to develop its own CGM and it’s possible we could have misunderstood the representative’s comments – no further details were provided.


Tandem appeared in its first ATTD exhibit hall following the first international sales in 3Q18. Reps told us the t:slim X2 is now in 10 countries (currently with Dexcom G5 integration OUS), far more than we realized. The booth advertised Tandem’s Basal-IQ with Dexcom’s G6, which is expected to launch outside the US throughout 2019; timing will vary by country, and we expect to hear more in Tandem’s 4Q18 call today. Reps told us the PROLOG pivotal trial for Basal-IQ (presented at ATTD last year) has “really changed the conversation in the field” – no surprise, given the strong 31% hypoglycemia reduction, excellent user experience, and no fingersticks. We also asked about the upcoming Control-IQ hybrid closed loop with automatic correction boluses – reps were very excited about the fully enrolled US pivotal study (n=168), which has not had a single dropout and has seen very high time in closed-loop (see the update on Day #2).


The Ypsomed booth featured a map of planned YpsoPump expansions, including US and Canada in its anticipated launches for 2019. At EASD in October, a booth representative confirmed that the YpsoPump is still under FDA review following submission in May. A Canadian YpsoPump launch was initially expected in October 2018, now reflecting a notable delay. YpsoPump’s Canadian launch will come firmly behind Tandem’s t:slim X2, which was expected to begin Canadian shipments by end of 2018. On the pipeline front, the booth representative shared that pump control will be on the Ypsomed mylife app “in the near future.”

ATTD Yearbook Highlights

Download the ATTD Yearbook here.

Self-Monitoring of Blood Glucose

  • Reporting on his and Dr. Irl Hirsch’s last ATTD Yearbook SMBG chapter, Dr. Satish Garg underscored the importance of this technology due to the mass unaffordability of CGM. He highlighted a number of abstracts, this year focusing on more on the efficacy of various strategies around BGM, not the technology itself. For example, he cited studies showing: (i) the importance of behavioral self-monitoring implementation in veterans; (ii) financial incentives had no improvements in A1c in teens; and (iii) blood glucose testing resulted in cost savings. Ultimately, he emphasized that decision support is where the field needs to go; “Whether with CGM and/or SMBG, only time will tell depending on cost. There’s no doubt in my mind that clinical decision support systems are effective in people with diabetes. However, it is time now to consider that mobile apps need to be standardized – there are many zillions of apps, and people get lost. There are significant reductions in medication errors when people use these apps. And they’ve been shown to reduce hypoglycemia.” Read the full chapter here.

Continuous Glucose Monitoring in 2018

  • Dr. Tadej Battellino reviewed three key studies highlighted in the CGM chapter, which has seen 18,000 downloads and is the fifth most downloaded chapter. He featured Dexcom’s DIaMonD type 2 MDI results, showing a benefit of CGM at 24 weeks. We were also particularly excited to see CONCEPTT and RESCUE included in his selection. In the 12-month RESCUE study, national Belgian reimbursement of CGM for 515 insulin pump users drove a striking 75% decline in hypoglycemia/DKA hospitalization. In total, the intervention resulted in an estimated nationwide cost reduction of €345,509. Patients achieved a significant 0.3% reduction in A1c (baseline: 7.6%) and improvements in time <70 mg/dl (11% to 5%; -1.4 hours) as well as quality of life metrics. RESCUE was published in The Journal of Clinical Endocrinology & Metabolism last March. Dr. Battellino characterized CONCEPTT as “completely landmark,” emphasizing that participants achieved time-in-range (68%) very close to that proposed as the consensus goal in pregnancy (70%) – of course, the consensus recommendations were made in part based on the CONCEPTT Data. He also noted the very encouraging newborn health outcomes and pointed out that the NNT were “rather low” to achieve the strong newborn reductions in large for gestational age (LGA), hypoglycemia requiring dextrose, and NICU admission >24 hours. CONCEPTT was published in The Lancet in November 2017. Read the full chapter here.

Insulin Pumps

  • Dr. John Pickup highlighted six themes in the insulin pump chapter: (i) acute diabetes complications in CSII vs. MDI; (ii) infusion set and site problems; (iii) pumps in practice; (iv) sensor-augmented pump therapy; (v) pump therapy for type 2 diabetes; and (vi) use of fast-acting insulin in pumps. Providing just a taste of what we’re sure is quite a comprehensive chapter, Dr. Pickup shared two real-world studies. A large study (n>30,000) using registry data from type 1 patients >20 years across 446 centers in Germany, Austria, Luxembourg, and Switzerland examined outcomes with pump therapy vs. MDI. A1c (8.04% vs. 8.22%; p<0.001), insulin (0.84 U/kg vs. 0.98 u/kg; p<0.001), and severe hypoglycemia (9.55 events/100 patient-years vs. 13.97 events/100 patient-years; p<0.001) were significantly improved on pump therapy. DKA events were also significantly lower in pumpers compared to patients on MDI (3.64 episodes/100 patient-years vs. 4.26 episodes/100 patient-years; p=0.04), which may sound surprising due to the prevalence of occlusions/kinking in infusion sets, though Dr. Pickup added that this trend is surfacing in more and more registries. We suspect this might be a correlation – people who are more likely to go on pumps are less likely to go into DKA at baseline. In The second highlighted article was a retrospective study in type 1 adults (n=503) comparing long-term A1c changes over 10 years with different pump types (Medtronic, Animas, Roche, Omnipod). A1c benefits were maintained over 10 years and there was no difference in A1c changes between pumps, including patch vs. tubed pumps. Read the full chapter here.

Decision Support Systems and Closed Loop

  • DreaMed’s Dr. Revital Nimri kicked off her review of the decision support and closed loop chapter with a dose of perspective: While there were only two closed loop studies evaluated in the very first ATTD Yearbook in 2009, this number has climbed to 37 in a decade. She first outlined three themes for closed loop in 2019: (i) “Expanding the loop” – bringing closed loop to non-traditional populations, such as type 2s hospitalized for non-critical care (Dr. Roman Hovorka’s Cambridge group) and adults with hypoglycemia unawareness; (ii) addressing physical activity; and (iii) addressing postprandial hyperglycemia (including automated meal detection). Next, three themes (verticals) in decision support: (i) retrospective decision support (automated pump settings adjustment; DreaMed Advisor Pro); (ii) real-time insulin dosing advisors (shown to reduce hypoglycemia and glycemic variability for both pumps and MDI); and (iii) and predictive glucose advisor (enables participants to modify their initial dosing decision). Finally, she overviewed a couple of studies that leverage closed loop with additional inputs and/or periodic adaptations, positing that these methods will “probably be the next-generation closed loop…as long as (the former) will not add to the patient burden.” The studies she cited were from Dr. Jessica Castle et al. (glucagon, exercise detection) and Eyal Dassau et al. (zone-based MPC algorithm with adaptation). Read the full chapter here.

New Insulins, Biosimilars, and Insulin Therapy

  • Dr. Lutz Heinemann focused on biosimilars in his presentation on new insulins – last year’s most-downloaded chapter – which, despite a relative lack of interesting and novel publications, he identified as a newly-matured area of research. Indeed, he chose to focus more on the commercial/regulatory landscape than the science, simultaneously expressing enthusiasm over how many companies are currently working to develop biosimilars (he cited >20) and questioning how many candidates will actually come to market. Specifically, he cited Merck’s decision to pull Lusduna (biosimilar insulin glargine) commercialization (October 2018), a decision Dr. Heinemann attributed to Sanofi’s litigation against the candidate (Merck cited an unfavorable commercial environment). This is balanced against other wins: 2018 saw the entry of the first-ever biosimilar mealtime in with Sanofi’s Admelog (lispro). Dr. Heinemann also raised the issue of insulin quality – a conversation Dr. Alan Carter’s and his 2017 DTM presentation and subsequent publication has driven. As more insulins hit the market, this will only become more important, and he criticized the fact that companies are not required to report data on insulin batch quality.  In the coming years, Dr. Heinemann anticipates “drastic changes” in the insulin market, particularly with regard to cost, and he predicted “5 to 10 insulin glargines on the market in a number of years.” Read the full chapter here.

New Medications for the Treatment of Diabetes

  • Dr. Satish Garg chose biosimilar insulins and SGLT inhibitors for type 1 as the two therapy classes that made the biggest splash in the past year. There’s no denying that SGLTs for type 1 have stepped squarely into the spotlight. Dr. Garg summarized key findings from inTandem, DEPICT, and EASE (focusing only on EASE-2 and skipping any mention of EASE-3 or the 2.5 mg dose of empagliflozin). He gave a nod to the positive CHMP opinion on dapagliflozin for type 1 in Europe, and announced that this indication just received a positive opinion in Japan following AZ’s submission in May 2018, though we haven’t found this news or a press release online. FDA hosted an Advisory Committee on Sanofi/Lexicon’s sotagliflozin for type 1 earlier this year, and a decision on that novel molecule is expected in late March. Dr. Garg acknowledged that SGLTs increase relative risk for DKA in type 1 patients, but emphasized the power of education to counteract this. He explained that DKA was less common in the placebo arm of many of these trials than it is in the real-world T1D population (~5% incidence in the T1D Exchange), which suggests that careful patient education can “reduce the background risk and the risk associated with SGLT inhibitors.”

  • Dr. Garg highlighted Sanofi’s Admelog as the first-to-market biosimilar mealtime insulin. This biosimilar formulation of insulin lispro showed comparable efficacy/safety vs. Lilly’s Humalog in the SORELLA studies. Dr. Garg lamented the fact that in the basal category, biosimilar insulin glargine (Lilly/BI’s Basaglar) hasn’t been dramatically discounted from its predecessor (Sanofi’s Lantus). He shared his hope that rapid-acting biosimilars prove significantly cheaper for patients. Read the full Yearbook chapter on new medications for diabetes here.

Using Digital Health Technology to Prevent and Treat Diabetes

  • Schneider Children’s Dr. Tal Oron, stepping in to at the last second to present, reviewed the digital health technology chapter of the yearbook (first-authored by Canary Health’s Dr. Neal Kaufman). The chapter breaks “digital therapeutics” into three verticals: (i) digital services – aim to modify patient behaviors to improve health outcomes; (ii) adjunctive digital therapeutics – support the use of traditional therapies such as medications, devices, and monitors; and (iii) digital drug replacements – seek to provide a clinical benefit as a replacement for a traditional treatment. We’re not sure we’d parse the landscape in this manner, since almost every digital health tool does both (i) and (iii), though any taxonomy in this nascent field is useful. The authors reviewed 19 articles from the year, including those from/about Canary Health, Welldoc, Glooko, Livongo, One Drop, and Omada. A separate paper that Dr. Oron highlighted found that a simple self-management website was cost effective at ~$25,000 per QALY gained – under pretty much any cost-effective threshold – and the number needed to treat to make the technology cost-saving relative to usual care was just 363. We’re hoping for far more reimbursement, uptake, and economic analyses of these apps/services, which have the potential to address the problem of scale in diabetes and general wellness. Read the full chapter here.

Immune Intervention in Type 1 Diabetes

  • On immune interventions for type 1, Dr. Desmond Schatz focused on oral insulin and methyldopa as advances in diabetes prevention, rather than on cure efforts. In his words, “until we can prevent, there will never be a cure. Even if we implant cells or replace islets, the disease is likely to come back because of its autoimmunity aspect.” To this end, he started by reviewing the TrialNet Oral Insulin trial, which found no overall effect with oral insulin on type 1 prevention in autoantibody-positive relatives of people with type 1. However, there was a benefit in the secondary stratum of participants with mIAA and ICA or both GAD65 and ICA512 antibodies, in which type 1 onset was delayed by an average of 31 months – in his assessment, highly meaningful. As such, Dr. Schatz believes oral insulin could be part of a cocktail or combinatorial agent in the future, for a certain endotype (that is, multiple prevention/treatment strategies will likely be needed). To this end, three studies are currently ongoing for oral insulin: (i) another TrialNet trial (immune effects and safety in stage 1/2 t1d); (ii) the Pre-POInT-Early (dose-ranging study in genetically at-risk children following beneficial immune effects in a pilot study); and (iii) GPPAD’s POInT trial (primary prevention of autoimmunity).
  • On methyldopa, Dr. Schatz pointed to a 2018 JCI study showing that the common blood pressure drug reduces inflammatory T cell response to insulin in people with type 1 diabetes and blocks the DQ8 antigen, a known source of disease risk. Based on these results, a study of methyldopa in autoantibody-positive individuals (also conducted by TrialNet) has been approved and was recently posted to As we’ve come to appreciate (and expect) from Dr. Schatz, we left his talk more inspired and optimistic than we entered. Oral insulin and methyldopa are only two of a plethora of agents under investigation for type 1 prevention, and his enthusiasm gives us hope over the many irons in the fire – particularly as combination therapies become more of a reality. To be sure, our competitive landscape signals increasing interest, even if widespread prevention and cure remain a long way off. Read the full chapter here.

Technology and Pregnancy

  • Out of an impressive 2,300 papers on diabetes technology and pregnancy, Dr. Helen Murphy chose 10 to include in the ATTD Yearbook chapter. She provided a sobering reminder of the need to focus on improving outcomes in pregnancy: the rate of stillbirths in type 1 diabetes and type 2 diabetes remains significantly higher than that in the background population. The rate of large for gestational age (LGA) has also increased; in fact, the proportion of LGA babies born to type 1 mothers has increased over the last decade by almost 10%. In addition to showing the strong CONCEPTT results evaluating CGM in pregnancy, Dr. Murphy also shared a cost-analysis of CGM in pregnancy presented as a poster in this meeting. She was particularly excited by the significant reduction in duration of NICU stay with CGM vs. SMBG (6.6 days vs. 9.1 days). NICU admission is quite expensive, and Dr. Murphy hopes the data will be compelling for payers. In total, Dr. Murphy estimates real-time CGM used by all type 1 pregnant women in the UK could save £10 million/year as compared to no CGM use. Read the full chapter here.

Advances in Exercise, Physical Activity, and Diabetes Mellitus

  • Dr. Michael Riddell picked out four intriguing, emerging findings/research areas in exercise and diabetes management: (i) significant, pancreatic fat content reduction after just 2 weeks of exercise; (ii) exercise volume, intensity, frequency, and duration are all associated with reduced premature mortality regardless of kidney function; (iii) mini-dose glucagon for hypoglycemia prevention in type 1 during exercise; and (iv) single- and dual-hormone artificial pancreas systems in exercise. On the third, Dr. Riddell noted that necessary snacks prior to exercise to avoid hypoglycemia often preclude net calorie expenditure during exercise, making glucagon a much more attractive option not only for improved glycemic control but also for achieving weight loss. Regarding AID systems, which he posited may have begun to “crack the nut” of exercise in type 1, Dr. Riddell extolled their adaptability in difficult situations (e.g., on the ski slope or in unannounced exercise) as well as in returning agency and improving quality-of-life. Read the full chapter here.

Diabetes Technology and Therapy in the Pediatric Age Group

  • Dr. David Maahs shared several papers examining use of diabetes technology (pumps, sensor-augmented pump therapy, and hybrid closed loop) and diabetes therapies (ACE inhibitors, statins, and metformin) in pediatrics. A large DPV registry study (n=30,579) comparing pump vs. MDI in children found pumps to confer lower rates of severe hypoglycemia, DKA, and lower A1c (8.04% vs. 8.22%). A six-month RCT (n=154) comparing the 640G with sensor-augmented pump therapy found the 640G to reduce time <70 mg/dl from 2.6% to 1.5% (-16 minutes). Looking at hybrid closed loop therapy with the 670G, Dr. Maahs presented a sub-analysis from the three-month 670G pivotal looking at just 14-26-year-olds (n=31). Time-in-range increased an “impressive” 14% and A1c decreased by 0.75%, and a significant linear relationship was observed between time-in-range and time in auto mode. On the therapy side, Dr. Maahs described a paper published in NEJM (n=443) finding use of ACE inhibitors and statins in adolescents with type 1 diabetes to result in no difference in the change in ACR (urine microalbumin). ACE inhibitors were associated with lower incidence of microalbuminuria (HR=0.57) but were not considered significant in the context of the primary outcome. Dr. Maahs pointed to a 12-month RCT (n=90) investigating metformin (1 g twice daily) in children with type 1 diabetes in which vascular function was found to improve independent of A1c and insulin dose decreased. Difference in adjusted A1c between groups over the 12-month study period was a significant 1.0%. We were pleased to hear Dr. Maahs end his presentation by honing in on the main goal – helping patients to stop worrying about highs and lows so that “kids can get on with what they’re trying to do.” Read the full chapter here.

Diabetes Technology and the Human Factor

  • Schneider Children’s Dr. Alon Liberman summarized the diabetes tech and human factor chapter, which he co-authored with Prof. Katharine Barnard. He expounded upon two of the articles that were included in the chapter, each with a focus on parenting and pediatric diabetes: One UVA paper detailed the perspectives of patients and physicians on how to improve the safety and functionality of an artificial pancreas system for younger children. Both parents and pediatricians emphasized that there are several risks associated with the application of system’s developed for adults/adolescents in children; to mitigate risk of injury, they recommend, among other things: (i) simple instructions/pictorials accompanying alarms to make the correct response easier for young patients and their care providers; (ii) a modified/flexible system to help children assume more responsibility of their diabetes management; and (iii) a messaging feature in artificial pancreas systems. The second article, a paper from Yale, looked at CGM and remote monitoring practices for children in school. Dr. Liberman emphasized the tremendous reported utility of remote monitoring, which the study found to ease parental concerns about overnight hypoglycemia and school-day safety. “One clear area for improvement,” said Dr. Liberman based on the data, is in augmenting the amount of retrospective glucose review to make insulin dose adjustments (i.e., Dr. Bergenstal’s “Thinking, Fast and Slow); 62% of respondents upload their data quarterly or less. Read the full chapter here.

Practical Implementation of Diabetes Technology: It is Time

  • In her Yearbook debut, Barbara Davis Center’s Ms. Laurel Messer took on the ambitious task of summarizing the current state of the “practical implementation” of diabetes technology, emphasizing that “it’s time we start looking at these devices…in the real world.” Her chapter, which concerned itself with aspects such as clinical utility, patient education, real-world use, economics/access, and benefits/barriers, zeroes in on three themes: (i) Use of technology in the real world (e.g., DIY, remote monitoring in school, and results of expanding access); (ii) practical challenges and solutions with diabetes technology (e.g., skin issues, insulin set failures, and targeted pump education); and (iii) access and cost-effectiveness of technology. She highlighted two articles from the year: The Belgian RESCUE study of CGM, in which the national health authority offered to reimburse CGM use in a subset of pump users (n=515) to see what the economics looked like in the real-world setting. We covered the highly impressive results in a poster at EASD 2017: People in the study saw A1cs decline alongside impressive reductions in hospitalizations and absenteeism. “I applaud them for finding a unique way to use payers to test a technology in the real-world. It’s not just about A1c or time-in-range, it’s also about the individual and societal benefits of technology.” On the opposite end of the spectrum, Ms. Messer highlighted a small Danish study (n=143 children/adolescents from two clinical sites) showing tremendous prevalence of skin reactions to pumps and CGMs. At the very moment of the study, 63% of pump users and 46% of CGM users were currently experiencing a skin reaction, and 90% of pump and 80% of CGM users reported having ever experienced a skin reaction. Ms. Messer concluded with three calls to action. We need: (i) better surveillance and tracking of skin reactions needed; (ii) more disclosure from industry about adhesive agents; and (iii) more clinical practice guidelines recommending skin assessment for chronic device use, like there is for statins/niacin, foot care, and peripheral neuropathy. Read the full chapter here.


-- by Adam Brown, Ann Carracher, Brian Levine, Payal Marathe, Peter Rentzepis, Maeve Serino, and Kelly Close