February 14-17, 2018; Vienna, Austria; Day #2 Highlights – Draft

Day #2 of ATTD 2018 is a wrap, and we’ve got a packed highlights report for you. See below for key takeaways from the day, organized into technology and therapy sections.

In case you missed it, here’s our day #1 report again, with six highlights covering Unomedical/Medtronic’s new MiniMed Mio Advance insertion set, DreaMed’s CE marked/under-FDA-review Advisor Pro clinical decision support software for pump settings, and Dr. Roy Beck’s Beyond A1c 2.0 talk at the diaTribe Foundation’s post-ArtWalk dinner.

We have another two days here in beautiful, wintery Vienna! The ATTD agenda certainly isn’t slowing down, and neither are we … see our full conference preview for a sneak peek at days #3-4.

Diabetes Technology Highlights

1. Tandem PLGS pivotal: 31% drop in <70 mg/dl (-19 mins/day), no increase in 180 mg/dl, no nuisance alarms; well-controlled users; FDA submission this quarter (1Q18), summer launch

Dr. Bruce Buckingham presented positive results from the pivotal trial (PROLOG) of Tandem’s predictive low glucose suspend device with Dexcom’s G5 – the system drove a 31% reduction in mean time <70 mg/dl, meeting its primary endpoint. Dr Buckingham praised how the suspensions occur without alarms, operating in the background – yes! Tandem expects an FDA submission this quarter (1Q18) and launch is still slated for summer 2018, pending approval – with these results, this timeline feels very reasonable, given the product’s low risk, obvious benefit, and how quickly Tandem has moved this program ahead (the feasibility study was shared at ADA 2017, and this study only started last August). The algorithm, now branded “Basal-IQ” and embedded in the t:slim X2 pump, was tested in a randomized crossover, home pivotal study that included two weeks of baseline data collection, followed by three weeks of PLGS vs. three weeks of sensor-augmented pump (SAP) therapy with t:slim X2/G5 (or vice versa). The primary outcome was time <70 mg/dl, and the study enrolled 103 participants in a wide age range (6-72 years; mean age: 24 years) and very engaged in their diabetes (baseline A1c: 7.3%, 83% pump user, 84% CGM user). Unlike other PLGS studies, this study was not enriched for hypoglycemia, nor did it include an in-clinic hypoglycemia induction.

• Mean time <70 mg/dl declined from 4.5% during SAP to 3.1% with PLGS, a 31% reduction (-19 minutes/day; p<0.001). Though the absolute reduction was small, we’d point out these rates of hypoglycemia were quite low, and the comparator arm was using Dexcom G5/t:slim X2, already experienced with pump and CGM, and came with baseline time-in-range at 64%. The median reduction in time <70 mg/dl was smaller: from 3.2% during SAP to 2.6% during PLGS (-9 minutes/day). Interestingly, more benefit from PLGS came during the day: -14-minutes vs. -9 minutes at night.
• For those experiencing >5% time <70 mg/dl at baseline, the improvement with PLGS was bigger – a 40% reduction in hypoglycemia from 6.7% during SAP to 4% with PLGS (-39 minutes/day). This group was just under 30% of the study population, explaining the smaller overall effect size. For those experiencing <5% in hypoglycemia at baseline, PLGS took hypoglycemia from 2.7% to 1.9% (-12 minutes/day). Other subgroup analyses by high/low baseline A1c and adult/pediatrics revealed the system was equally effective.
• Time-in-range improved very slightly from 64% on SAP to 66% on PLGS (+29 minutes/day; p<0.001) – again reinforcing how well this study population was doing at baseline. Mean glucose did not change – 159 mg/dl on both SAP and PLGS.
• Suspensions of insulin delivery occurred on 91% of days, with a mean of 5.7 suspensions per day. The mean duration of suspension was only 18 minutes, but that still meant a cumulative total suspension time of 104 minutes per day. For a group with more hypoglycemia, we’d imagine the suspension data and time <70 results would be more dramatic. The study had one severe hypoglycemia event in SAP and none on PLGS. 
• The study had a 99% completion rate, ~95% CGM usage, and strong system usability ratings: >88% of participants agreed/strongly agreed that they would like to use the system frequently; thought the system was easy to use; felt very confident using the system; did not need the support of a technical person to use the system; and did not need to learn a lot before getting going with the system.

• As described at ADA 2017 (when the first feasibility data was shared), the Tandem Basal-IQ (PLGS) algorithm is very straightforward: it uses the last four sensor glucose values to predict the sensor glucose 30 minutes into the future. Insulin delivery is suspended if the predicted glucose is <80 mg/dl OR the observed sensor glucose falls below 70 mg/dl. Insulin delivery resumes when the sensor glucose is rising OR if suspension exceeds 120 minutes (to prevent rebound hyperglycemia).

2. MiniMed 670G pivotal in 7-13 years: A1c drops 0.4% (baseline: 7.9%), +2 hours/day in-range (56%->65%); submitted to FDA last month

Dr Bruce Buckingham presented long-awaited, positive pediatric (7-13 year-old) data on the MiniMed 670G/Guardian Sensor 3 hybrid closed loop. Medtronic announced today it was submitted to FDA last month, which would expand the 670G indication from 14+ years to 7+ years. The single-arm, three-month, 105-participant peds study showed similar outcomes to the adult/adolescent study presented 20 months ago – compared to a two-week open-loop run-in period, 670G drove a 0.4% reduction in A1c in the pediatric population (baseline: 7.9%); time-in-range (71-180 mg/dl) improved from 56% to 65% (a strong +2 hours/day); time <70 mg/dl declined from 4.7% to 3.0% (-24 mins/day); and time >180 mg/dl improved from 39% to 32% (-1.7 hours/day). The A1c endpoint and all CGM metrics were highly statistically significant in favor of the 670G (p<0.001) – we only wish Medtronic had reported with the consensus definitions for time-in-range and hypoglycemia! (Perhaps the study was already underway…) The percentage of patients with an A1c <7.5% improved from 36% at run in to 51% at study end; the higher the baseline A1c, the more pediatric patients benefitted from the 670G. In line with the A1c improvement, mean glucose improved from 169 to 162 mg/dl, with a slightly bigger benefit overnight –166 to 155 mg/dl (p<0.001). Coefficient of variation improved slightly, from 40% to 39% (p=0.01). The modal day plots we saw in the adult/adolescent study were nicely replicated here, showing a tightening of glucose levels at all times of day on the 670G, and especially overnight. Notably, patients gained ~5% of their body weight during the three-month study (+5 lbs from a baseline of 94 lbs), along with adding 2.8 units/day to their daily dose on 670G (baseline: 36 units) – we wonder how kids’ eating patterns changes on the device! There was no DKA or severe hypoglycemia, and patients were in Auto Mode 80% of the time – with the average age at 11 years, we’d guess parents were likely managing a lot of the 670G interaction. Overall accuracy for Guardian Sensor 3 was higher than in prior studies: MARD of 11.9% (n=3,271 paired points), and only 1% of the paired points were <70 mg/dl; the number of daily fingersticks was not reported, but was presumably at least four/day and perhaps more. Dr Buckingham concluded that “the judge of these systems is whether you continue to use it” – notably, 97% of pediatric participants enrolled in the 670G continued access phase, notably higher than the “>80%” that continued in the adult/adolescent study. There is an ongoing study in 2-6 years, which would be a big win for many parents experiencing many sleepless nights.

• Dr. Buckingham was quite excited about these pediatric results, and his comparison to the adult and adolescent pivotal data (shared at ADA 2016, published in JAMA and DT&T) was one of our favorite slides in the talk. As the table below shows, Auto Mode Use and time-in-range varied a bit between the age groups, while A1c reductions were very much in line (0.4%-0.6%). Overall, it’s very positive to see the consistent improvements across ages, especially the more challenging pediatric and adolescent groups.

• The single-arm, three-month, multicenter study occurred at nine centers (8 US, 1 Israel), mimicking the adult/adolescent study design – including home, in-clinic, and hotel portions. Children had to be 7-13 years, A1c <10%, pump therapy for >6 months (with or without CGM), and a minimum total daily dose of >8 units per day.

Questions and Answers

Q: How were the gain (aggressiveness) settings adjusted?

Dr. Buckingham: They are adjusted each night at midnight based on total daily dose and glucose values. The adjustment is automatic and takes into account data from the previous six days. Once on the system, it continues to update at midnight. Some of these kids started out on the system several years ago, and now they’ve gone through puberty – they’ve had dramatic changes in insulin requirements, and the settings and the system cope with that.

Dr. Roman Hovorka (Cambridge): It’s great to see data moving into younger populations. I might be difficult here... I love these technologies to work, but this was a single arm study – before/after. We know very well if people enroll in a study, their glucose control will improve. How much of the benefit was due to the 670G? Can we really reliably say these improvement can be attributed to the device?

Dr Buckingham: I’d make two points: the data Dr. Fran Kaufman presented for the commercial use of the system in a non-study environment showed similar improvements, which continued over time. There is also a randomized trial being conducted, but it will take a while to get results completed.

Dr. Hovorka: But can we say the 670G reduces A1c and increases time-in-range? Can we reliably say so based on this study?

Dr Buckingham: We can only say it reduced from baseline.

Dr Hovorka: I agree. 

3. Dexcom G6 Update: No calibration, 2-hr warmup expected, but will accept a cal if a user chooses; Updated app: urgent low “soon” alert, alarm profiles; Direct-to-Apple Watch is NOT under FDA review

In one of Dexcom’s strongest symposiums ever, SVP Jake Leach shared the most comprehensive G6 feature update we’ve seen – including the no-calibration plans, accuracy data from recent clinical studies, sensor and app updates, and new branding. The product continues to look outstanding and will update every aspect of Dexcom CGM – some very meaningfully, like the new sensor and inserter. We detail all the new learning below, especially news that G6 will indeed be no-cal, but will have the ability to accept calibrations if a user chooses. Mr. Leach mentioned that a two-hour warmup is expected for the no-cal version – if approved, that would be a huge improvement over Abbott’s 12-hour no-cal warmup in the US. The G6 accuracy data shown today was from “a combination of recent clinical studies” (not clear if this included the never-before-seen pivotal data), and it continues to be on par/improved vs. G5 in terms of all accuracy metrics – overall MARD 9.0%, 94% of points within 20/20, a Day 1 MARD of 11.2%, and 94% within 20 mg/dl in hypoglycemia. Mr. Leach also confirmed in Q&A that direct-to-Apple-Watch connectivity is not under FDA review, contrary to what the New York Times reported in December – this is not too surprising to us, as Dexcom has been vague in recent months about the timing here. Moving to the Watch as a standalone primary display device (Class III) is a big deal and will require a lot of development effort. Combined with Dr. Boris Kovatchev’s compelling talk on AID (see above), along with Dr. Nick Oliver on the iHART CGM study (Libre vs. G5; see our coverage), we thought this was one of Dexcom’s most exciting  conference symposia in some time. As of JPM, G6 is under FDA review and Dexcom expects a US launch in “2H18,” and if all goes well with the FDA, it could launch by “mid-year.” Dexcom positioned this as a global 2018 G6 launch at JPM, but OUS plans were not mentioned today.

• Like JPM, today’s talk made it clear that G6 will be a no-calibration-required system. But in new news, Mr. Leach mentioned G6 will accept calibrations if a user would like to enter one. A two-hour warmup was also mentioned, which had not previously been confirmed for the no-cal version. Mr. Leach made it clear that G6 is “truly designed to be a no-cal system,” and Dexcom will recommend patients do not calibrate it with fingersticks – each sensor will be calibrated in the factory. However, optional calibration will be advantageous in the “very rare” case where a sensor is off – quite smart to add, as some key opinion leaders have criticized FreeStyle Libre on this front (i.e., when Libre is reading inaccurately, there is no way to correct it). The two-hour warmup is presumably still under discussion with the FDA, but if Dexcom can get that through, it would be a huge advantage over FreeStyle Libre’s current 12-hour warm-up in the US.

• Mr. Leach showed new G6 accuracy data today combining data from recent clinical studies – Dexcom has maintained/improved on G5’s accuracy, even with the move  away from calibrations. It was not clear how many patients/paired points were included in the slide below, and if this dataset included the main G6 pivotal (on which we’ve never seen data). Regardless, the accuracy looks very much on par with G5, but eliminating calibrations. Mr. Leach later said that G6 features a 40% improvement in sensor variability (less manufacturing variability) and a 10x improvement in signal:noise ratio (more glucose signal, less background signal) –which “significantly drives performance,” especially in the low glucose range.

• G6 will have a slightly updated app, maintaining the same general look as the G5 app, but moving the menus around and using up more of the screen. The events and settings now appear below the trend graph, possibly to make entering “food” or “exercise” events and accessing alarms more top of mind. As a reminder, the current user experience puts Events, Share, and Meter Calibration as icons across the top, with settings accessed in a side menu. Presumably the new structure will also de-emphasize calibration, given comments above.

• We finally got to see the finalized G6 predictive “urgent low soon” alerts, a feature Dexcom has long talked about adding. This more intelligent low alert notices when glucose is dropping rapidly and is expected to cross the low threshold – e.g., “55 mg/dl within 20 minutes.” Mr. Leach emphasized it does not issue increase the nuisance factor, given the 93% detection rate. We’d guess it can be toggled on/off, similar to the rise/fall rate alarms.

• In a long-needed update, G6 will also add alert schedules within the app, allowing patients to set glucose alarm settings for different times of day. Like basal profiles, this will allow different glucose thresholds to be set at night vs. during the day, for instance. The G6 app will also add more alarm/sound customization options.
• The new G6 applicator continues to get rave reviews in Dexcom’s tests: 98% of pediatric subjects rated the new applicator as “very easy” or “somewhat easy” vs. 59% for G5; 76% rated the new applicator as “didn’t feel anything” vs. 30% for G5. This is going to be a major upgrade to the current syringe-like inserter.
• Other notable features that were new and/or confirmed:
  
  • Indicated for 2+ years old, well ahead of Abbott (18+ years) and Medtronic (14+ years) in the US.
  • Wear time of 10 days, with 14 days coming with the first-gen Verily disposable.
  • A 20-feet Bluetooth transmission range, plus a 50% improvement in the stability of the sensor signal. Dr Kovatchev later raved about the consistent G6 connectivity with the Tandem t:slim X2 pump. We’ll be interested to see if this has improved over G5 in real-world use, since Bluetooth dropout is still an issue even at very short distances.
  • The new ~30% smaller G6 transmitter will maintain the three-month life. The key change here will come with the disposable Verily gen one sensor. (Launch expected in “late 2018/early 2019,” per JPM.) Reordering transmitters remains a weak point of the current G5 system, so this will be a welcome upgrade.
  • The sensor geometry minimizes the wound healing response. The insertion needle is the same size as G5 at 26 gauge. Combined with the new applicator data, however, the pain seems to be smaller.
  • Acetaminophen blocking, which we’ve long known about. We’ll be interested to see if this allow G6 to enter the hospital.
• Dexcom’s partners slide now lists just two pump companies (Insulet, Tandem), one smart pen player (Lilly), and six AID partners (including University of Cambridge and Imperial College London) – we cannot recall seeing the latter two before, though perhaps they were shown because of the EU location of ATTD.

• Direct to watch connection remains “under development,” but Dexcom did not provide specific timing. Both Apple Watch and Fitbit Ionic were shown on this front, confirming the Fitbit partnership is also striving to do go direct-to-watch. We continue to hear from parents that this will be very welcome, as use with the Series 3 Apple Watch would mean their kids can avoid carrying around a smartphone.  
• Dexcom aims to bring the public API and Clarity mobile app outside the US soon, but no specific timing was shared. These have been warmly received in the US, especially the notifications in the Clarity mobile app.

4. Dr. Boris Kovatchev on Major AID Trials and Excitement over Dexcom G6/Tandem X2/TypeZero; Dramatic New Ski Camp Results

Dr. Boris Kovatchev shared updates on a number of UVA/TypeZero studies using Dexcom’s G6 and Tandem’s t:slim X2 with the embedded Control-IQ algorithm. The clear headline was a ski camp study from last month, showing a 6-hour/day improvement in time-in-range. More broadly, we were struck by (i) Dr Kovatchev’s excitement over the Dexcom G6/Tandem t:slim X2/TypeZero combination, which seems to be quite reliable on the connectivity and durability fronts; and (ii) how big of a win this is for Tandem, who is now in some very major closed loop studies outside of the iDCL pivotal. Remarkably, the UVA team has tested variants of the TypeZero algorithm over 14,500 days in over 450 clinical trial participants (30 clinical trials) at 15 research sites on four continents. Dr. Kovatchev emphasized throughout that the algorithm automates correction boluses, and tightens basal modulation as the night goes on – separating it from other systems.

• A new AID ski study wrapped up last month, testing the integrated Tandem X2/Dexcom G6/TypeZero system in 13-18 year olds (n=25) – over 48 hours of skiing, time-in-range (70-180) increased from 50% on SAP vs. a remarkable 73% on Closed Loop, translating to 6 more hours per day in range (p=0.01)! Average glucose declined 35 mg/dl (baseline: 179 mg/dl; p=0.03). Time <70 mg/dl increased from 0.8% to 3.1%, but it was not statistically significant (p=0.7). The G6 CGM was connected 97% of the time, and closed loop was active for 95% of the time (99% excluding warmup) – very strong connectivity in extreme conditions. Dr Kovatchev showed the impressive picture below of someone doing a snowboard jump while on the integrated system – nice! The next phase of the trial will test the system in skiing 6-12 year olds.

• Several long-term studies of closed loop control are going on now or starting soon, including:
  
  • Project Nightlight (ongoing), testing 24/7 closed loop control vs. overnight only in 110 participants over 10 months. The trial has been using Dexcom G4 and a Roche pump, but is now transitioning at the midpoint to the Tandem t:slim X2 and Dexcom G6 system with Control-IQ. Outcomes include A1c, risk for hypoglycemia, and preference for overnight vs. 24/7 closed loop control. 84 participants have been recruited out of 110. The data looked so strong from the initial tests that it helped give FDA confidence in moving to the pivotal iDCL trial.
  • The international closed loop (IDCL) pivotal trial testing the t:slim X2 with Control-IQ and Dexcom G6. 150 participants will be randomized 2:1 to closed loop vs. SAP, with main efficacy outcomes as time-in-range (70-180 mg/dl). Per last month’s update, a two-week, at-home study at seven sites was expected to start enrollment this quarter (1Q18), followed by the pivotal trial this year in preparation for a 1H19 launch.
  • The FreeLife Kid AP Study in France. The trial will begin this summer in France and use the Tandem/G6/Control IQ system. It will enroll 120 children over six year, first testing overnight only and then moving to 24/7 closed loop. “We believe this will be a landmark pediatric study of a size and duration that has not been done before.”
  • The Virginia hypoglycemia safety project, which will test closed loop in young children 5-10 years old and older adults 65+ years! The goal is to prevent extreme blood glucose events. Usual care for three weeks will be compared to three weeks of closed loop (Tandem/Dexcom/TypeZero) in a crossover design. Notably, the study will include lots of neurocognitive testing.
• Dr. Kovatchev also briefly covered CGM-based MDI decision support with the inControl Advice app paired with Dexcom CGM. The system includes six advisory modules design to work in concert: exercise advice, sleep advice, smart bolus calculation, hypoglycemia prediction, estimated A1c, in silico therapy simulation. An ongoing multi-center trial (n=132) at UVA, Stanford, and Mount Sinai will report three-month pilot results tomorrow. As we heard at DTM, Novo Nordisk is supplying the NFC-enabled NovoPen Echo and a connected Tresiba pen for this ongoing study.

5. Dr. Fran Kaufman Real-World 670G Data From ~14k Patients Broken Down by Age; Medtronic’s “Best Practices” for Glycemic Goals, Time in Auto Mode, and Time Wearing Sensor

Medtronic Diabetes CMO Dr. Fran Kaufman presented the latest set of real-world 670G data uploaded voluntarily to CareLink by 13,906 US patients (of a possible “over 25,000” now using the system). The user base in the US has been augmented by ~5,000 in the past ~three months, up from 20,000+ in November. As we’ve come to expect, the system’s real-world performance continues to largely resemble that in the pivotal, particularly in adults. Across all ages (7+; presumably the 479-patient 7-13 age group represents off-label use), auto mode time in 70-180 mg/dl was ~71%, time <70 mg/dl was ~2%, time <54 mg/dl was 0.49%, and time >180 mg/dl was ~27%. The largest divergence from the pivotal study was the ~96 minutes fewer per day spent in auto mode in the real world (from 87% in the three-month pivotal to 80% in real-world use). Unsurprisingly, the pediatric and especially the adolescent groups had lower time in range than the other age groups, but still fared significantly better on hybrid closed loop – 7-13 year olds and 14-21 year olds saw time in range increase from 52% to 65% (+3.1 hours/day) and from 53% to 63% (+2.4 hours/day), respectively. In this cohort, people ≥60 years old stayed in auto mode for the longest, at ~20.4 hours per day, and also spent the most time in range, at ~75% (up from baseline of 68%). These data suggest that 670G consistently improves glycemia across all age ranges, and supplements the pivotal data in 7-13 year olds presented by Dr. Bruce Buckingham on the same day (see above). As for the 2-6 year-old cohort, Dr. Kaufman said that a seven-site, n=50, three-month study is in the process of wrapping up. We wonder if we might see this data later this year, perhaps at ADA?

• Based on observations from this real-world data set and the pivotal studies, Medtronic came up with “best practices” for times in ranges, in auto mode, and wearing the sensor. The picture below details the following recommendations: Time between 70-180 mg/dl at >70%, time 50-70 mg/dl at ≤3%, and time <50 mg/dl at ≤1%. In order to achieve these, Medtronic’s analytics team recommends staying in auto mode >80% of the time and wearing the sensor >85% of the time. We thought this relatively simple analysis was helpful, as it gives providers and patients some specific goals to aim for. The difficulty with targets, of course, is that they need to be individualized (e.g., should they be different for adolescents?), and they could lead to feelings of failure if they are not achieved. On the other hand, having an idea of where the “typical” patient falls and how that correlates with outcomes and behaviors (including something as simple as sensor wear time) is valuable and needed to help guide therapy and identify problem areas. We wonder how glycemic goals should/would change for type 1s using different device combinations and type 2s on various device and drug combos.

6. Glooko’s MIDS Basal Insulin Titration System Receives FDA 510(k) Clearance; 18 Months Collaboration with Experts, $10s of Millions Invested to Deliver MIDS

Glooko CEO Mr. Rick Altinger announced that the company’s Mobile Insulin Dosing System (MIDS; basal insulin titration) has been cleared by the FDA after a ≥nine-month review. See below for the second and most up to date screenshots of MIDS (see the first from ATTD 2016) – the interface appears to be very intuitive, as we’ve come to expect from Glooko. This news comes after substantial back-and-forth with the Agency, and makes Glooko the seventh (by our count) basal titration system to have received FDA clearance – see the others here. We’re excited to see how quickly the number of products have increased, and now it’s about go-to-market, business models, and getting providers on board to prescribe these product. According to the press release announcing FDA clearance, Glooko engaged in 18 months of collaboration with experts around the world to design and develop MIDS. Mr. Altinger noted in his presentation that Glooko has invested “tens of millions of dollars” in the system – far more than we would have guessed. User experience design reviews, human factors studies, pre-market programs, and feedback sessions all contributed to the ultimate delivery of MIDS. At DTM 2017, we saw encouraging data from Glooko’s 14-day MIDS feasibility study (n=14 type 2s): average blood glucose dropped 18 mg/dl (baseline 164 mg/dl), the proportion of in-range readings (80-180 mg/dl) increased by nine percentage points (baseline 64%), and the proportion of hyperglycemic readings (>250 mg/dl) decreased by 11 percentage points (baseline 14%). In today’s presentation, Mr. Altinger did not mention the larger clinical study currently underway investigating MIDS in 240 type 2 patients with Novo Nordisk’s Tresiba. The 16-week study is still listed as recruiting on CT.gov, with an expected completion date of November 2018. As a reminder, MIDS analyzes patients’ fasting blood glucose levels and recommends insulin dose adjustments based on the provider’s pre-configured treatment plan.

• Glooko is currently used by 7,000 providers and over 1.5 million people with diabetes, resulting in over 8.5 billion data points – wow! Glooko is compatible with 180 devices and is now available in 24 countries (up from the most recent update of 23 countries) and in 15 languages.

7. Guardian Sensor 3 Pediatrics (2-18) Accuracy Trial: 2-Cal MARD 10.9%, 3-4-Cal MARD=10.1%

Medtronic CMO Dr. Fran Kaufman’s talk was so rich with new data that we had to split it into two highlights: In this edition, she briefly presented a pediatric Guardian Sensor 3 accuracy trial (initially shared by Dr. Jennifer Sherr at ISPAD) that found twice-daily calibration MARD to be 10.9% (YSI/SMBG unclear). The trial enrolled 145 patients ages 2-18 who underwent a six-hour frequent sample test on day one, three, or seven. This will be key to get a pediatric label in the US, given how fast peds CGM penetration is rising in the US! With 3,102 paired points, two calibration MARD was 10.9%, but with larger-than-expected error bars (10.7%). 2,890 paired points for 3-4 calibrations (reflecting the current recommendation for adults) yielded a slightly improved MARD of 10.1%, plus/minus 9.3% on either end. In the 670G pivotal trial for 7-13 year olds, the sensor and I-STAT instruments showed fairly high agreement, with 68% of sensor readings and ~66% of I-STAT readings falling between 70-180 mg/dl. As noted elsewhere in this report, the percentage of points in hypoglycemia was quite low at 1% in this trial. Guardian Sensor 3 is undoubtedly a major accuracy improvement for Medtronic, but still has some catchup to play on calibration and form factor to Dexcom’s G5 (MARD 9.0%, two cals/day), Abbott’s FreeStyle Libre (US MARD: 9.7%, no-cal), and the upcoming no-calibration G6 (MARD of ~9.0%, see above; expected in the US in 2H18). Notably, we learned today that the Dexcom G6 will be indicated for 2+ years old, ahead of Abbott’s US FreeStyle Libre (18+ years; currently recruiting for US pediatric trial) and Medtronic’s Guardian Sensor 3 (currently 14+ years but likely to change with this pediatric submission and that for the 670G).

8. Harvard’s Dr. Frank Doyle III on the Next Frontier in AID Algorithms: Adaptation Based on Multiple Zones and “Trust Indices”

Harvard’s Dr. Frank Doyle III tested the audience’s technical knowledge in his review of what’s to come in automated insulin delivery algorithms: (i) novel adaptation algorithms using multiple zones and (ii) different modeling strategies depending on degree of confidence in the current state (“trust index”). Taking multiple zones allows the system to go beyond a one-size-fits-all algorithm with a gain multiplier depending on distance from the target value, employing a different glucose-normalizing strategy if a patient is euglycemic, slightly elevated, very high, etc. His group has already begun testing multi-zone adaptations based on glucose and rate of change in the UVA/Padova simulator, and shown improvements in time in range, mean glucose, hyperglycemia, and hypoglycemia vs. SAP This data, which “further raises the bar of what has already been a very robust system,” will be presented at the American Control Conference this Spring. To explain the philosophy behind the “trust index,” an estimate of the quality of the system’s predictions based on changes in physiology, external challenges, sensor accuracy, etc., Dr. Doyle drew an analogy to the stock market: At the end of December, when the market’s volatility index was low, an analyst made a bold 15-year prediction for the S&P. Now that the volatility index is astronomical, the same analyst probably wouldn’t be as comfortable making such a projection, or at least the error bars would be significantly greater. When the model seems to be accurate, it can be more aggressive; when lower confidence, aggressiveness should be dialed back. Running an algorithm adjusted to modulate aggressiveness and strategy based on trust in legacy data pushed time in range all the way up to 91% in one analysis! The trust index work has even made it into the clinic (with a Dexcom G4 and Omnipod at Sansum in a 48-hour study), and a manuscript is currently under review. In the study, time in range (70-180 mg/dl) reached 88%, and time <70 mg/dl was just 1.5%. These data are early and in the clinic, but Dr. Doyle claimed that this level of time in range may set a “new high water mark” for his own work. He wrapped up his talk with the simple conclusion: “Adaptation and our ability to not just stick with a fixed algorithm has to happen. We have to be responsive to changes in patient, sensor, and perhaps the demands that patient and doctor will put on themselves.” The potential for even smarter AID algorithms has not been discussed in the past few years as much as faster insulins and additional sensor inputs, but we’re very optimistic about the potential here – given 42 factors that affect blood sugar, there is a lot of variability that algorithms might be able to model!

9. Real-World Medtronic Guardian Connect Data Shows Predictive Alerts to Reduce Glycemic Excursions; Integration with Android “a Number One Priority”

Medtronic Diabetes Medical Affairs Director Dr. Ohad Cohen presented real-world data uploaded to CareLink by 2,541 patients on the Guardian Connect standalone mobile CGM. Notably, predictive alerts reduced glycemic excursions for hyperglycemia and hypoglycemia by 39% and 60%, respectively. Hyperglycemic and hypoglycemic excursions were defined as ≥three consecutive sensor values (≥15 minutes) beyond the high (210 mg/dl) and low (70 mg/dl) thresholds. The window of evaluation for excursion start times was 60 minutes following the alert, as predictive glucose alerts can be set to notify users 10-60 minutes prior to a predicted low or high glucose excursion. Not only did predictive alerts help users to avoid excursions, they also significantly reduced the number of excursions lasting more than 60 minutes. This study shows the power of predictive, prospective alarms and analytics in diabetes – in addition to preventing dangerous and uncomfortable glycemic events, it provides a learning opportunity without the classical “punishment.” Data were collected from users with ≥five days of sensor data from January 2, 2017 – January 2-2018. We’re not sure how much of a read on the user base this data set is – Medtronic has never shared how many Guardian Connect users there are worldwide.

• In today’s presentation, Dr. Cohen noted in response to an audience member’s question that integration with Android devices is a “number one priority.” As a reminder, Guardian Connect is available on iPhone in the EU and Australia, and it is currently slated for a US launch by April 2018 along with the paired Sugar.IQ app. We’re assuming Guardian Connect is still under FDA review, as it has been for well over a year, but did not receive a specific update today, on Medtronic’s 3Q17 call, or at JPM. How will Medtronic’s iPhone app compare to Dexcom’s G5 app? When will Abbott be in the US with FreeStyle LibreLink on iPhone and Android? How far away is Medtronic’s Android in the US and globally?
• Dr. Cohen shared separate real-world data from 258 patients using Guardian Connect for at least nine months. Strong time-in-range (70-180 mg/dl) was apparent immediately (in the first month) and persisted throughout the period – as Dr. Cohen highlighted, not seeing a decline here is promising, though obviously consistent with the benefits of real-time CGM. It’s unclear what the baseline time-in-range was (pre-Guardian CGM), which prevents conclusions about relative efficacy. On average, users spent 61% of the time in range, 35% of the time in hyperglycemia (>180 mg/dl) and 4% of the time in hypoglycemia (<70 mg/dl).

• Dr. Cohen detailed real-world data from 3,263 patients collected over January 1, 2017-January 15, 2018, showing the high sensor glucose alert to be by far the most commonly utilized alarm – on average, this alarm sounded three times/day. Low sensor glucose and low predictive glucose alerts tied for second, averaging twice/day. Following them were notifications for high predictive glucose (once/day), falling glucose (once/day), and rising glucose (less than once/day). Dr. Ohad emphasized that these alerts are highly customizable, as users not only can select specific alarms and choose distinct thresholds, but also can adjust the alarms’ individual timing, ranging from 10-60 minutes prior to an event. Like Dexcom Share, CareLink Connect users for remote monitoring can also customize alarms; one audience member gave an example of a parent using predictive and sensor glucose alarms, while the child uses only sensor glucose alarms.

10. Medtronic’s Diabeter Update: ~2,500 Patients and Counting; $75,000 est, Savings per Patient Over Lifetime; Cool Ther@pyMail Feature

Diabeter’s visionary founder Dr. Henk Veeze provided an update on the Medtronic-owned chain of Dutch clinics with an advanced integrated care model that delivers value-based medicine. During the course of the rapid-fire talk, he shared news of an expanded user base, detailed Diabeter’s superb outcomes, cost per-patient, and move to bundle-based payments, introduced the access-anytime Ther@pyMail service, and offered a number of quotes about Diabeter’s value proposition. Spoiler, our favorite is: “If you have $1,000 to spend, you better invest in diabetes care, because you get far more than on best day in stock market.”

• Diabeter now takes care of nearly 2,500 patients, up from just 200 a decade ago, and is beginning to take care of adults (past the age of 25) who don’t want to age out of the Diabeter model. Dr. Veeze’s clinic has now hired three internists, and the group is seeing solid growth and distribution across the Netherlands. We believe patient advocacy – people who were a part of Diabeter but refuse to leave upon turning 25 will not only drive growth, but also indirectly make the rest of the health system want a piece of Diabeter’s model.
• As depicted in the image below, Diabeter’s patients have lower A1cs compared to other clinics in the Netherlands, and even relative to the T1D Exchange. These improved outcomes, according to Dr. Veeze’s model of complications and costs, could save a system with 30,000 type 1s (A1c=9%) $2.2 billion over a child’s lifetime ($75,000 per patient). ~$41,600 of these savings come from complication reduction, while ~$33,000 come from improved productivity. Regarding bundled payments, Diabeter teams up with Diabstore to provide medical devices and hospital care, but is also responsible for 100% of all costs, including other healthcare providers, other suppliers, and pharmaceuticals; any profits in the case of a same or lowered bundled costs are shared.

• Ther@pyMail is an on-demand email report that patients can receive after uploading their data. Within five minutes, they receive the report, which includes a color-coded heat map of their recent blood glucose values, insulin dosing advice, bolus recommendations, comparisons to “patients like me,” and more. WOW! We still have a long way to go on this front across the world, even in tech-savvy clinics. We love this feature, and according to the below graph below, so would payers – upon introduction of the email service, the percentage of patients with an A1c <7.5% went through the roof, though it soon came back down considerably. Perhaps it drove early engagement, which tailed off once the novelty wore off.

• Quotable Quotes
  
  • “If you have $1,000 to spend, you better invest in diabetes care, because you get far more than on best day in stock market. We are a reverse pension fund, you pay something now and save money later.”
  • “With a hospital, you talk with payers about budgets. With me, you talk about patients.”
  • “I have to deal with seven insurance companies every year. I show them data every year. But I never discuss prices until at the very end…they agree quickly at the end, because the Diabeter model is cheaper. They ask us to do more because it’s to their advantage. Total cost of patient comes in our bundle, if save on cost, they hand over the money.”

11. Dr. Bergenstal Calls for CGM in CVOTs – “We Would Learn Amazing Correlations Between Glucose Variability and Outcomes”

IDC’s Dr. Rich Bergenstal advocated for all diabetes CVOTs to use CGM, which he argued is the most realistic path to correlate glucose variability with outcomes. In order for a measure of glucose variability to be “meaningful,” it has to show significant association with microvascular risk, Dr. Bergenstal explained. It’s unrealistic (but not impossible) that the field will sponsor another DCCT-like trial with coefficient of variation (CV) as the independent variable, but CVOTs are ~10,000-person datasets ripe for the picking. “Where else do we have 10,000 patients followed for five years with adjudicated outcomes? Why not just throw in a CGM?” We’re love this idea, as having CGM tracings as a component of CVOT results would provide hard evidence on the relationship between glucose variability and microvascular outcomes, while also giving researchers heaps more information about the drug’s action profile. As Dr. Roy Beck mentioned at the diaTribe Foundation’s “Beyond A1c 2.0” dinner last night, the NIH-sponsored PERL study is using CGM longitudinally to investigate how time-in-range affects kidney disease; the study was initiated in October 2017 (thanks to Dr. Irl Hirsch and others), and is expected to complete in August 2019 per ClinicalTrials.gov. Dr. Bergenstal echoed Dr. Beck’s commentary today, suggesting CGM in CVOTs as a way to demonstrate the promising applications of beyond-A1c glycemic metrics, proving their utility in the clinical trial setting (comparing drugs/devices) as well as the real-world clinical setting (identifying an optimal treatment plan for the individual with diabetes). “We could get some amazing data. We would learn amazing correlations between variability and outcomes,” Dr. Bergenstal emphasized. “We should work hard to make this happen.”

• Notably, post-hoc analyses of DEVOTE (Novo Nordisk’s CVOT for basal insulin Tresiba) have started this effort of correlating beyond-A1c glycemic metrics with outcomes. Across both arms of the study, Tresiba or Lantus, day-to-day glycemic variability was significantly associated with all-cause mortality (HR=1.58, 95% CI: 1.23-2.03, p=0.0004). Severe hypoglycemia also significantly increased risk for death in DEVOTE – in the first 15 days post-severe hypo event, the hazard ratio for all-cause mortality was 4.20 (95% CI: 1.35-13.09), which decreased to 2.78 (95% CI: 1.92-4.04) in the one year following a severe hypo episode. If the quality of life impact of glucose variability wasn’t compelling enough, DEVOTE has now shown RCT evidence linking variability to death. But DEVOTE didn’t use CGM – imagine the level of detail and analysis that would be possible with this technology.
• We hope to see more manufacturers invest in CGM for diabetes clinical trials, including CVOTs, although it may take a movement from thought leaders, patients, and even FDA and NIH, as Dr. Bergenstal suggested. His second research proposal around glucose variability was for NIH to create a “CGM toolbox” like there is for quantifying cognition or emotion. This could be a valuable resource to HCPs and researchers, alike, with the need-to-know on standardized CGM metrics, hypoglycemia definitions, time-in-range targets, and patient assessment tools.
• Interestingly, Peyser et al. and Garcia et al. published accompanying papers in the January 2018 edition of Diabetes Technology & Therapeutics introducing a new metric for variability, the glycemic variability percentage (GVP), which takes both the period and amplitude of the glycemic tracing into consideration. Calculating GVP is made possible by the richness of CGM data and was shown to capture differences in glycemic variability undetected by traditional metrics like CV or standard deviation (SD). The field seems to have reached a consensus on CV for now since it captures variation without being biased by the mean and is rather easy to understand, but is not without its problems, as it weights hypoglycemia and hyperglycemia equally – clinically, straying 70 mg/dl above the mean should probably not be viewed the same as straying 70 mg/dl below the mean.

12. Simple Basal Insulin Titration Algorithm Results in Superior A1c Reduction vs. Physician-Managed Titration in Type 2 Patients

Dr. David Russell-Jones (Royal Surrey County Hospital, UK) presented data (n=602) from a 24-week, multicenter randomized parallel-group study showing a simple patient-managed insulin glargine U300 titration algorithm results in slightly, but significantly greater A1c reductions than physician-managed titration. Type 2 patients not at goal on their previous regimen were randomized to either patient-managed or physician-managed groups. For the patient-managed group, dose adjustments were made by participants every time the median fasting self-monitored plasma glucose (FSMPG) differed from the target for three consecutive days – if the FSMPG was >130 mg/dl, the recommended dose adjustment was +3 U/day, while if the FSMBG was <80 mg/dl, the recommended dose adjustment was -3 U/day. For the physician-managed group, adjustments were recommended by the treating physician at each study visit or telephone call depending on glucose values. At 24-weeks, those in the patient-managed group (n=300) achieved an A1c decrease of 1.0% (baseline: 8.4%) while those in the physician-managed group (n=302) achieved an A1c decrease of 0.9% points (baseline 8.4%). The difference in reductions between groups were found to be superior. Though the 0.1% improvement can hardly be viewed as clinically significant in isolation, given that it was patient-driven makes it highly so. 67% of the patient-managed group and 58% of the physician-managed group achieved an FSMPG of 80-130 mg/dl without confirmed (<54 mg/dl) or severe hypoglycemia. Targets were achieved in a greater proportion of those in the patient-managed group regardless of prior insulin status. There were no differences in hypoglycemia events between groups. Given that very few providers have the time to provide tailored insulin titration, non-inferiority alone would be a huge win for the field – demonstrating superiority in A1c reductions is icing on the cake.

Diabetes Therapy Highlights

1. Novo Nordisk’s Fiasp Shows Postprandial Efficacy vs. NovoLog in 472 Pump Users with Type 1; Equivalent A1c-Lowering and Hypoglycemia in 16-Week Data from Onset 5

Dr. David Klonoff presented results from Novo Nordisk’s Onset 5 study of Fiasp (faster-acting insulin aspart) vs. NovoLog (insulin aspart) in pump users. After a four-week run-in period, adults with type 1 diabetes on a pump (n=472) were randomized to Fiasp or to NovoLog for 16 weeks. A1c-lowering was non-inferior for Fiasp compared to NovoLog, dropping from a baseline 7.5% to 7.44% with the next-gen and from the same baseline to 7.35% with the first-gen (p<0.001 for non-inferiority). Dr. Klonoff reported an estimated treatment difference of 0.09% for A1c (95% CI: 0.01%-0.17%). Fiasp did show significant benefit over NovoLog on a one-hour post-meal test, with an estimated treatment difference of 16.39 mg/dl for postprandial increment at week 16 (95% CI: -25.73 mg/dl to -7.06 mg/dl, p=0.001). The improvement in postprandial glucose with Fiasp vs. NovoLog at one hour post-meal was supported by significantly lower post-meal plasma glucose levels at 30 and 120 minutes, and by 0-1 hour and 0-2 hour ISF glucose increments as measured by CGM and by postprandial SMBG testing. Hypoglycemia data was resoundingly neutral across groups, with an estimated treatment ratio of 1.00 (95% CI: 0.85-1.16). Dr. Klonoff explained how this figure encompassed severe hypoglycemia as well as blood glucose-confirmed symptomatic hypoglycemia. Looking only at severe glucose-confirmed events, hypoglycemia appears to be more frequent with Fiasp vs. NovoLog (21 vs. 7 episodes), but excluding three participants who all experienced severe hypoglycemia during the run-in (and who all happened to be randomized to the faster-acting aspart arm), these values fell to 11 vs. 7 events. Dr. Klonoff reviewed nearly equivalent adverse event rates for Fiasp-treated and NovoLog-treated patients in the study. He concluded that Fiasp is safe and effective for CSII, an important finding, as Novo Nordisk’s advanced prandial insulin is currently approved for use in pumps in Europe but not in the US or Canada. Fiasp was just launched in US pharmacies earlier this month, at parity pricing to NovoLog, which is a huge win for patients.

2. One-Year Onset 1 Results Confirm Fiasp’s Safety/Efficacy in Type 1 Patients on MDI; Modest A1c Benefit + Enhanced Postprandial Control One Hour Post-Meal

In the same session of orals, Dr. Bruce Bode continued the focus on faster-acting Fiasp with one-year data from Onset 1 (n=1,143 type 1 patients on MDI). These trial extension findings largely paralleled the 26-week results, showing modest but statistically significant A1c improvement and superior performance on a one-hour post-meal test with Fiasp vs. NovoLog. After 52 weeks, A1c declined from a baseline 7.6% to 7.5% for Fiasp-treated patients, staying at 7.6% for patients on NovoLog (estimated treatment difference -0.1%, p=0.042). A1c did climb slightly from the 26-week point, when Fiasp patients were down to a mean 7.3% and NovoLog patients were down to a mean 7.4%, for an estimated treatment difference of -0.15% (p=0.0003). Fiasp demonstrated a 16.5 mg/dl benefit on postprandial glucose control one hour following a meal (p=0.0002), and showed a nonsignificant 7.6 mg/dl treatment difference on a two-hour post-meal test after 52 weeks. Dr. Bode also pointed to a 4.14 mg/dl benefit with Fiasp vs. NovoLog on 9-point SMPG at the one year mark (p=0.047). He shared detailed slides on hypoglycemia and treatment-related adverse events, neither of which differed significantly between groups over the full year. Severe hypoglycemia occurred in 37 of 386 Fiasp patients (9.6%) and in 46 of 380 NovoLog patients (12.1%), which seems like a numerical discrepancy, though Dr. Bode emphasized that “severe hypoglycemia was not an issue” in Onset 1. Blood-glucose confirmed hypoglycemia occurred in 361 (93.5%) of Fiasp patients vs. 369 (97.1%) of NovoLog patients. All other safety data was also well-balanced between the two mealtime insulins, and according to Dr. Bode, showing convincing, longer-term safety of Fiasp was the main reason for the Onset 1 trial extension. These findings are consistent with the overall Onset program, and could provide reassurance to patients/HCPs looking to start on Fiasp for boluses. The product was launched earlier this month in the US, where it’s only indicated for MDI (and not pumps – see above for more on this). We do think highlighting safety will be of help in promoting the new drug and generating early uptake, especially since Fiasp is priced on par to NovoLog, so people could be motivated to make the switch for smaller postprandial excursions and more mealtime flexibility. Although some consider the glycemic benefits to be “marginal” rather than truly significant compared to earlier prandial insulin options, we maintain that current rapid-acting mealtime insulin is just not good enough for many patients, so we’ll take any improvement in postprandial excursions, A1c, or even the peace of mind that comes with a shorter tail, less residual insulin in the bloodstream, and possibly less fear of hypoglycemia.

• During Q&A, Dr. Bode did address the issue of why A1c doesn’t decline as substantially with Fiasp as might be expected given consistent improvements in postprandial control. He hypothesized that patients on the advanced insulin engage in protective eating to avoid going to bed low/nocturnal hypoglycemia, which leads to greater glucose variability overnight. Dr. Bode underscored the relatively long mean duration of diabetes in Onset 1, between 19-21 years, and alluded to the difficulty of changing habits. “If you’re used to going to bed at 150 mg/dl, you want to keep doing that, even if your post-dinner blood sugar is down to 100 mg/dl,” he explained. “You can’t change people’s behavior after 20 years.” We’re not sure that’s entirely true, and we’d be interested in behavioral interventions that can support patients making the switch to Fiasp or another faster-acting insulin, but we also appreciated Dr. Bode’s suggestion that putting faster-acting aspart in pumps or in hybrid closed loop may be the ideal application. Again, we direct you to Onset 5 results on Fiasp in pump users, presented in the same session by Dr. David Klonoff (see above).

Questions and Answers

Dr. Bruce Bode (Atlanta Diabetes Association, GA): I’ll start with a question to myself. Why do you think there was an A1c difference in Onset 5, and the post-meal was always better in Onset 1, but there was no significant drop in A1c?

A: We’ve been using Fiasp in pumps, and one thing my patients are feeding back to me is the greater variability they’re feeling overnight. I wonder if that’s at play.

Dr. David Klonoff (Mills-Peninsula Health Services, San Mateo, CA): That’s a good question, Bruce. What we saw was that glucose levels in faster-acting aspart people rose quite a bit after dinner, and just stayed higher all-night long. So, people taking Fiasp in a pump need to be carefully treated, and they may need a higher basal rate over time than they’re used to.

Q: Were you surprised there was no difference in A1c? And how would you design a study to show a difference?

Dr. Bode: The only problem with increasing basal overnight is you increase the risk for hypoglycemia. I think it’s protective eating, is what it is. If you’re used to going to bed at 150 mg/dl, you want to keep doing that, even if your post-dinner blood sugar is down to 100 mg/dl. So, these patients might eat, and that would then cause variability overnight. Note that we’ve seen this in every single trial of rapid-acting insulin.

Q: So, how do you design a study to prevent this nighttime variability?

Dr. Bode: You put Fiasp in a hybrid closed loop. You can’t change people’s behavior after 20 years. If you’re new to diabetes, maybe.

Comment: When you test new insulins, you need new tools. So, maybe it’s not hybrid closed loop, but you at least need CGM with Fiasp. Patients don’t feel their glucose all the time.

Dr. Bode: Yes, and there’s a sub-analysis of Onset 5 with CGM.

Dr. Klonoff: Someone else did that analysis. That’s where the pattern was higher overnight; that’s where the A1c was lost. You had better control after meals, and if you lost it somewhere, it was overnight.

3. Dr. Robert Ritzel Illuminates New Data from BRIGHT (Head-to-Head RCT of Toujeo vs. Tresiba), But More Questions Remain; Full Results Presentation Expected at ADA 2018

Munich’s Dr. Robert Ritzel shared more granularity on BRIGHT study results comparing Sanofi’s Toujeo (insulin glargine U300) vs. Novo Nordisk’s Tresiba (insulin degludec), although we’ll likely have to wait until ADA 2018 for hypoglycemia findings. This Sanofi-sponsored open-label trial (n=929 adults with type 2) is the first head-to-head RCT of next-generation basal insulins, so the results could certainly have commercial impact. Sanofi issued a topline release in December 2017, and also discussed the very preliminary data during a corporate symposium at IDF. Dr. Ritzel announced that ~50% of participants in each arm achieved target A1c <7% after 24 weeks (mean baseline A1c was 8.7% in the Toujeo group, 8.6% in the Tresiba group). He showed balanced treatment-emergent adverse events: 44% of Toujeo-treated patients (202 of 462 individuals) and 48% of Tresiba-treated patients (221 of 462 individuals) experienced any treatment-related AE. Serious adverse events occurred in 5% and 4% of the Toujeo and Tresiba groups, respectively, while 4% and 7% of each group discontinued from BRIGHT early. We’re eager for even more specifics on this data: What was the average A1c decline with Toujeo and Tresiba? Sanofi’s topline release reported that Toujeo met its primary endpoint of non-inferiority vs. Tresiba for A1c reduction over 24 weeks, but no mean values of p-values were disclosed. How did treatment satisfaction compare across the two advanced basals? We were pleased to see in the topline release that PROs, as measured by the Diabetes Treatment Satisfaction Questionnaire (DTSQ), are a secondary endpoint in BRIGHT. And of course, as one audience member brought up, we’re very much looking forward to hypoglycemia results. In fact, hypoglycemia risk reduction could be one of the most important advantages to next-gen basals; Tresiba has already shown significant risk reduction vs. Lantus in SWITCH and DEVOTE, and is awaiting an FDA decision on a possible label claim reflecting this benefit. Even more than seeing how Toujeo stacks up to Tresiba on hypoglycemia frequency, we’d love to gain a better quantitative sense of how Toujeo offers hypo benefit over earlier basal insulin options. While these head-to-head results will be intriguing, our view is that both these products are strides above what came before, offering meaningful improvements to patient care. We’d love to see greater commercial uptake of Toujeo and Tresiba alike – sales are strong, but could be so much stronger (the next-gen basal insulin class grew 57% YOY to $2.1 billion in 2017, and accounted for 22% of the $2.5 billion basal insulin market in 4Q17). Sanofi management announced on the company’s 4Q17 earnings call that full BRIGHT results will be presented at ADA 2018 in Orlando, while Novo Nordisk has launched its own head-to-head RCT, with data anticipated in 4Q18. Notably, the primary endpoint of the Novo Nordisk-sponsored RCT is hypoglycemia rather than A1c decline. We take this as another sign that the companies believe in the hypo benefits their advanced products offer.

Questions and Answers

Q: What was the hypo data?

A: That is going to be presented at another conference. The data was not yet available when we submitted for this conference.

4. Dr. Bode on Lilly’s Ultra-Rapid Insulin Lispro – New Phase 3 PRONTO-Pump Study to Begin Next Week; Describes Speed as More Physiologic; Emphasizes Positive Patient Experiences on Afrezza

Dr. Bruce Bode highlighted promising ultra-rapid-acting insulin candidates in the pipeline, including Lilly’s phase 3 ultra-rapid lispro. He shared that a new phase 3 study in people with type 1 on an insulin pump (n=48) will begin in the next few days. According to ClinicalTrials.gov, PRONTO-Pump is expected to complete in September of this year. Interestingly, the primary endpoint is rate of infusion set failures with ultra-rapid insulin lispro vs. Lilly’s Humalog (insulin lispro); secondary endpoints include percent of participants with ≥1 infusion set failure, rate of premature infusion set changes, and time interval until infusion set change. This suggests to us that Lilly is investing deliberately in developing its ultra-rapid candidate for pump use, which is fitting, since speed of bolus insulin has significant implications for pumps as well as closed loop – Dr. Bode spoke to these applications of faster-acting insulins, both in this morning plenary and in an afternoon oral presentation of Onset 1 (see above). PRONTO-Pump joins PRONTO-T1D and PRONTO-T2D in the phase 3 program for ultra-rapid lispro; these pivotal studies in type 1 and type 2 diabetes are expected to complete in September and February 2019, respectively. Dr. Bode additionally discussed Adocia’s BioChaperone Lispro and Diasome’s HDV insulin lispro; both candidates are in phase 2, though BC Lispro is phase 3-ready (Adocia continues to seek a new development partner following termination of the Lilly partnership in January 2017). As for ultra-rapid-acting insulins already available to patients, Dr. Bode touted the benefits of Novo Nordisk’s Fiasp and MannKind’s inhalable Afrezza. He underscored that the philosophy behind these products is to more closely mimic the kinetics of endogenous insulin secretion, which has both a fast onset and offset of action (resulting in tighter postprandial glucose control and lower hypoglycemia risk). We do see mealtime insulin as one of the areas of diabetes care most in need of improvement, which is why we’re excited by the prospect of Fiasp (just launched in the US) and Afrezza; on Afrezza, uptake has been sluggish thus far, but the product received a label update from FDA last October to reflect an ultra-fast PK/PD profile vs. Humalog. We’d love to see these more physiological options in even more patient hands, though we note that access remains a barrier.

Questions and Answers

Q: On Afrezza, what’s the efficacy of using it post-meal or splitting the dose? How often do we have to give a second dose after the meal?

A: If you look at the PK/PD profile of any of these insulins, it’ll tell you what will happen. If you’re having a high-fat meal that increases insulin resistance, you need to check glucose two hours after you eat and maybe take a correction. More than half of patients will eat a high-fat meal, so at least half will need a correction dose. If you’re eating simple sugar, you’re fine with one inhalation. If you’re eating steak and potatoes and a dessert, you’re going to need a second inhalation. Dr. Garg said early on to put this in the protocol, and we did put it in AFFINITY 1, but only 20% ever used it. In STAT, we encouraged everyone to always check blood sugar and if you’re >160 mg/dl, take a second inhalation.

Q: Is there any concern with not having more flexibility of dosing with Afrezza?

A: Patients love this product. The masses aren’t on it, because many endos think it causes lung cancer, but it has no action in the lung – it’s just a way to get through. Most people need to use a sensor, whether it’s flash (Abbott) or Dexcom, to decide when to take an extra inhalation.

5. Dr. Garg Optimistic for Tresiba Hypo Label Update, Regrets Lilly’s Discontinuation of PEGylated Insulin Peglispro

In a review of concentrated insulin products on the market, Dr. Satish Garg expressed optimism that FDA could approve a hypoglycemia claim for Novo Nordisk’s Tresiba (insulin degludec), and he lamented Lilly’s discontinuation of insulin peglispro (in December 2015). According to Dr. Garg, FDA was planning an Advisory Committee to discuss the requested label change for Tresiba, but the meeting was canceled. Does this imply that FDA is leaning toward an approval? Dr. Garg shared a positive outlook (we’re keeping our fingers crossed). Novo Nordisk management expects an FDA decision by end of 1Q18. To be sure, there’s no precedent at FDA to reflect hypoglycemia benefit on an insulin label – Dr. Garg explained that all labels come with hypoglycemia warnings, and none compare one insulin to another – so this claim for Tresiba would be revolutionary. In 3Q17, EMA approved a similar Tresiba label update, based on positive data from SWITCH 1, SWITCH 2, and DEVOTE (40% risk reduction for severe hypoglycemia and 53% risk reduction for severe hypo overnight vs. standard of care Lantus, p<0.001 for both comparisons). To further illustrate the value of insulin degludec, Dr. Garg presented a case study of a 44-year-old man switching from Sanofi’s Lantus to Tresiba: The patient’s A1c dropped from 7.8% to 6.5%, his total daily insulin dose dropped 18%, and his CGM tracings showed remarkably less glucose variability. Dr. Garg alluded to the importance of time-in-range for patient quality of life and treatment satisfaction, and in this regard, Tresiba is definitely superior to earlier basal insulin options. He touched upon Sanofi’s next-gen basal Toujeo (insulin glargine U300) as well, which demonstrated a significant reduction in nocturnal hypoglycemia vs. Lantus in pivotal studies. Sanofi management recently suggested that Toujeo could show a similar hypoglycemia benefit to Tresiba, meaning both next-gen basals are a leap above first-gen options like Lantus and Novo Nordisk’s Levemir (insulin detemir). We certainly agree with this latter point, and the former could be determined by head-to-head studies of Tresiba vs. Toujeo. The Sanofi-sponsored BRIGHT trial (an RCT) will report full results at ADA 2018, while Novo Nordisk’s head-to-head study is expected to complete and read out in 4Q18. Dr. Garg also pointed out the important consideration that Toujeo increases an individual’s insulin dose by a mean ~20% vs. Lantus, which is included on the product’s US label.

• In Dr. Garg’s view, “insulin peglispro was the best concentrated insulin” in development, prior to discontinuation in early December 2015. Lilly made this decision based on liver toxicity seen in phase 3, or what Dr. Garg described as issues related to liver enzyme levels and liver fat. These safety concerns aside, he touted the PEGylated insulin candidate for its “hepatic specificity,” which more closely resembles endogenous insulin action. He highlighted the continued weight loss seen with Lilly’s peglispro, which is certainly meaningful, given that most insulins are associated with weight gain – eliminating this side effect could improve adherence, treatment satisfaction, and overall outcomes for people with diabetes. Dr. Garg shared hope that “someone else will take up some other way of doing PEGylation” for the promise of an insulin therapy with fewer side-effects and durable efficacy (peglispro showed significant A1c reductions and a decline in fasting plasma glucose over 78 weeks, in addition to being the only insulin that lowered body weight). Rezolute (formerly AntriaBio) has a PEGylated basal insulin candidate in phase 1, with a study currently underway at Prosciento in San Diego.
• Dr. Garg concisely outlined why we need concentrated insulin: For one, the field is still lacking a basal or prandial insulin that mimics healthy beta cells. Rising BMI in the population introduces a need for higher insulin doses at lower volume, since increasing volume alone would lead to unpredictable absorption, local discomfort, medication complexity, poor glucose control, and heightened hypoglycemia risk. Moreover, Dr. Garg has calculated that there are at least ~100 million people in the world requiring insulin therapy (~30 million type 1s, ~80-100 million type 2s, including those who are antibody-positive), and fewer than one million of them (1%-2%) are on a pump. “Let’s not forget about the 99%,” he said, referring to the majority of diabetes patients who aren’t on an insulin pump due to access issues, knowledge gaps, and/or implementation challenges. The size and scope of this population further underscores the unmet need for better concentrated insulins.
• Dr. Garg concluded his talk with a shoutout to Dr. Irl Hirsch’s annual rants in Diabetes Technology & Therapeutics (read the 2018 edition here), which highlight cost as a persistent issue. In any discussion of insulin, and especially insulin innovation, we can’t ignore the rising cost of this essential medicine, he emphasized.

6. Dr. Irl Hirsch Rants: How Do Positive CVOTs Figure into Cost-Effectiveness Calculations; Are Insulin Analogues Not Cost-Effective for T2D?

For the second consecutive ATTD, UW’s Dr. Irl Hirsch delivered a fascinating talk on cost economics, this time exploring the (complex) impact of positive CVOTs on the calculus of cost-effectiveness research. Before certain anti-hyperglycemic drugs were shown to be cardioprotective, cost-effectiveness was easier from the payer point of view: Cost per 1% A1c drop would be a logical calculation, and Dr. Hirsch found it to clearly point to generic drugs (metformin, SU glyburide, and TZD pioglitazone; ~$3-$8/1% A1c drop) as the obvious choice compared to DPP-4 sitagliptin ($693/1% A1c drop), SGLT-2 canagliflozin ($493/1% A1c drop), and GLP-1 liraglutide ($736/1% A1c drop). This analysis largely excludes acute outcomes such as DKA and severe hypoglycemia, but shows a vast gap that, according to the DCCT and basic economics, points to the generics. But, as Dr. Hirsch pointed out, CV disease and PVD (peripheral vascular disease) contributed 57% of diabetes healthcare costs in 2012; this on top of positive CVOT findings requires a recalibration on both the payer’s and the provider’s parts. Dr. Hirsch proceeded to review cost-effectiveness analyses of empagliflozin (Lilly/BI’s Jardiance) in Turkey, Italy, Greece, and the UK (the official EMPA-REG cost analysis). In all four, researchers deemed empagliflozin cost-effective, with assessments ranging from “reasonable and bearable for payers” in Greece to “extremely cost-effective” (Dr. Hirsch’s words) in Turkey. ICER (incremental cost-effectiveness ratio; a measure of the difference in cost of two approaches divided by difference in their effects) ranged from $2,183/QALY (quality-adjusted life-year) in Turkey to 4,811 euros/QALY in Italy, where the rule of thumb threshold for coverage is 30,000-50,000 euros/QALY. Notably absent from this set of analyses (“the elephant in the room”) was the US…how would the models fare there, where both healthcare as a percentage of GDP and per-patient costs are shooting up, wondered Dr. Hirsch. The cost half of the equation was a definite cause for alarm: Dr. Hirsch’s research showed that whereas one month of empagliflozin costs $54, $119, and 55 euros in the UK, Canada, and Spain, respectively, the retail cost per month in the US is closer to between $449 (GoodRx.com) and $516 (average wholesale price). Barring the scenario in which empagliflozin is significantly more effective in the US, the cost-effectiveness likely takes up to a 10x hit in the US (still a good deal for payers, per the rule of thumb). We too would very much like to see this analysis in the US, and Dr. Hirsch said that an abstract for empagliflozin cost-effectiveness analysis has actually been submitted! We loved this talk because it reminded us just how complex health economics are, barely touching the tip of the iceberg of complexity; Dr. Hirsch pointed out that analyses must also take into account acute savings and costs from glucose lowering (acute hyperglycemia/hypoglycemia), chronic savings and costs from glucose lowering, additional time to work instead of disability before death, and the fact that the costs of everything from complications to drugs are highly dependent on the country. We’d add to that list that (i) cost-effectiveness is a great population-level metric, but there are certainly some individuals for whom a certain treatment would have a sky-high ICER, and others for whom ICER would be $0.50/QALY and (ii) ICER is not the only metric of interest – if a patient doesn’t like a certain drug because of its mode of administration, side-effect profile, or something else, she won’t take it and receive the projected benefit. The shift toward a value-based healthcare system relies on not only determining which drugs are “most” cost-effective across a whole lifespan and population, but also on curating the right treatments for the right person at the right time, and on a continually-learning ecosystem – we see registries and nations with unified EHRs as crucial pieces in solving this puzzle.

• BD’s Dr. Larry Hirsch astutely asked if the fact that most CVOTs have been conducted in people with existing CV disease, thus far, exaggerates the cost-effectiveness of cardioprotective drugs. In other words, how would the health economics play out in the general population? The data to answer this question doesn’t yet exist, which is why Dr. Hirsch framed the talk at the outset by showing two patients – one without CV disease and another with cardiomyopathy – and asking what would be the most cost-effective medications for both. It’s too early to say at this point whether cardioprotective drugs should be administered in people without established CV disease. Without cost as a factor, the answer should be a resounding yes, but cost is obviously a (if not the) major dictator of the medications and tools available to patients today. Notably, SGLT-2s may be the first class to show cost-effectiveness in a broader type 2 population, including those without prior stroke/MI – there have been hints in CVOT data that these agents may confer heart failure benefit in a primary prevention population. A post-hoc of CANVAS, for example, found significant risk reduction for heart failure hospitalization in both cohorts, people with and without baseline CV disease. AZ’s DECLARE CVOT for dapagliflozin includes heart failure in a co-primary endpoint and also enrolls a larger primary prevention cohort (~50% vs. 33% in CANVAS and 0% in EMPA-REG), so this could also move the needle on SGLT-2s as a more cost-effective diabetes therapy (DECLARE results are expected in 2H18).
• On a discouraging note, a systematic review published last year in PharmacoEconomics found that the cost-effectiveness of insulin analogs in type 1s is $27,051/QALY gained, but this number was $425,913/QALY gained when it came to type 2s. In Dr. Hirsch’s words, insulin analogs in type 2 diabetes are “about as cost-effective as a pediatric heart transplant. I thought that was a joke until I looked it up.” The authors concluded that “current evidence suggests that insulin analogs are cost-effective in T1D; however, the evidence for their use in T2D is not convincing.” Dr. Hirsch said he’ll still use insulin analogs in his type 2 patients (which we were happy to hear since we would advocate for more, not fewer people to get on insulin); this story is no more than a cautionary tale about the importance of taking cost-effectiveness analyses with a grain of salt.
• Last year, Dr. Hirsch’s talk revolved around the economic benefit of intermittent CGM ($8,893/QALY). This cost seems to be significantly higher than the cost-effectiveness of empagliflozin, but recall that ICER depends critically on the comparator approach/treatment, so they are not directly comparable. Still, $8,893 is a great deal for a payer, according to most economists, said Dr. Hirsch.

7. Dr. Sherr on Benefits to Nasal Glucagon – Speed, Accuracy, Ease of Use; Lilly Plans to File in US/Europe in 2018, Expected Approvals by 2019; Company Working Hard on Manufacturing for Hopeful Swift Launch

As Lilly approaches regulatory filing of nasal glucagon (planned for 2018), Yale’s Dr. Jennifer Sherr provided a timely and comprehensive review of RCTs and real-world trials of the candidate completed to-date. Pivotal studies in adults and pediatrics showed a 3 mg dose to be safe and effective across the entire age spectrum for type 1 diabetes. Since these phase 3 trials relied on HCPs to administer the nasal glucagon, Dr. Sherr also summarized findings from a simulated usability study published in 2017. Participants included primary caregivers of patients with diabetes (n=16) as well as acquaintances who weren’t trained (n=15). In a mock scenario, participants encountered a mannequin representing someone with type 1 or type 2 diabetes unconscious due to severe hypoglycemia; a backpack with glucagon was nearby on the floor, and distracting sounds/stressors were deployed to model the urgency and stress of a real-world emergency. The results were impressive: 94% of instructed caregivers and 93% of untrained acquaintances successfully delivered a full dose of nasal glucagon to the mannequin, while only 13% of caregivers and no acquaintances delivered a full dose of intramuscular glucagon (p<0.001 for both comparisons). Among primary caregivers, the mean time to administer glucagon was only 16 seconds for nasal vs. 113 seconds for intramuscular; these values were 26 seconds vs. 144 seconds for untrained acquaintances. This parallels real-world data presented by Dr. Elizabeth Seaquist at ADA 2017 (n=129), showing how 70% of caregivers were able to administer nasal glucagon in <30 seconds, while 98% were able to do so in less than two minutes. Speed is of the essence for rescue therapy, and Dr. Sherr spoke to the challenges of current glucagon reconstitution kits, which require a lengthy mixing process prone to error in a high-stress hypoglycemia situation. In a small study conducted at a diabetes camp, 70% of parents experienced difficulty with a glucagon reconstitution kit, five of them injected only air/diluent (and not glucagon), and the average time to administration was 2.5 minutes (150 seconds). Dr. Sherr also presented real-world data on nasal glucagon for pediatric patients, showing that 94% of caregivers found the product easy-to-use, and 94% were relatively satisfied, satisfied, or very satisfied after a hypoglycemia episode. Moreover, in all severe hypoglycemia events (33 for 14 participants), the individual returned to normal glycemic status within 30 minutes, which was a similar finding to the real-world study in adults. The last trial she discussed, one published recently in Diabetes, Obesity, and Metabolism, investigated the effects of the common cold or nasal congestion on the efficacy of Lilly’s next-gen glucagon in healthy volunteers. There were no significant differences in PK/PD performance or safety, and Dr. Sherr concluded that nasal glucagon could be used effectively even if a patient is congested.

Questions and Answers

Q: Does the nasal glucagon product need to be refrigerated?

A: No.

Q: What’s the shelf life?

A: I’ll defer to my colleagues at Lilly on that, but my understanding is that it’s up to a two-year shelf life. [Editor’s note: In a call with Lilly management prior to ADA 2017, we learned that shelf life for nasal glucagon will likely be comparable to that of current glucagon reconstitution kits, 1.5-3 years depending on refrigeration.]

Q: When will it be available?

A: Nasal glucagon is still in development worldwide, but there are plans to submit packages both in the US and in Europe in 2018.

Q: Is it going to be more expensive than current glucagon? Do you have any sense about the production difficulties?

A: I can’t speak to cost, but what I can say is that Lilly’s working very hard to make sure manufacturing is all set, so that when the product is approved, it will be ready for distribution right away.

8. Fiasp in Pumps: Dr. Eric Zijlstra Indicates that Fiasp May Be Even Faster Via CSII; Dr. David Russell-Jones Reviews Evidence for Fiasp in Pumps

At a Novo Nordisk corporate symposium, Profil’s Dr. Eric Zijlstra presented data indicating that Fiasp might actually perform better in a pump than via injection, based on available data. Compared to NovoLog, pumping Fiasp results in onset of exposure 12 minutes earlier, compared to 10 minutes earlier for injection, and insulin exposure is three-fold higher for Fiasp vs. NovoLog with pump compared to two-fold higher with injection. Importantly, this trend holds true for offset of action as well. In pumps, Fiasp’s mean offset time is 35 minutes earlier than NovoLog’s, and it’s only 12 minutes earlier with injection. While these are striking results, Dr. Zijlstra emphasized that they are indirect comparisons from different studies (total n=48 for pump and n=218 for injection), and shouldn’t be used to draw hard conclusions. That said, we certainly think further investigation is warranted. To be sure, Fiasp is an improvement over other rapid-acting insulin analogs regardless of how it’s taken, offering more flexibility/less uncertainty around meals. This could correlate to less fear of hypoglycemia, which gets in the way of optimal glucose control for so many patients with diabetes (79% of type 1 patients and 58% of type 2 patients lower their insulin doses following a severe hypoglycemia episode). While Fiasp is approved for pump use in the EU, it’s currently only available in pens or vials in the US/Canada. But we’ll be interested to see if this changes, especially if the next-gen, faster-acting insulin continues to show even more impressive efficacy in a pump setting. Dr. Zijlstra emphasized that an ideal insulin for pumps would have even faster onset (even Fiasp can’t compete with endogenous insulin secretion in speed) and offset (to avoid postprandial hypoglycemia), with a particular eye toward application in a fully-automated closed loop – many believe subcutaneous insulin delivery kinetics are the ultimate limiting factor, and IP administration needs to be explored more in-depth if full closed loop will ever be reached.

• University of Surrey’s Dr. David Russell-Jones reviewed data from the phase 3 Onset program, supporting the use of Fiasp in type 1 diabetes and in pumps. Onset 1 (n=1,143) demonstrated modest but significant A1c improvements with Novo Nordisk’s Fiasp vs. NovoLog in adults with type 1 diabetes (0.15% greater with Fiasp at 26 weeks). See above for the 52-week data, presented by Dr. Bruce Bode as an oral. Additionally, Dr. Russell-Jones discussed Onset 4, a small, six-week compatibility trial of Fiasp in pumps for type 1 diabetes, which randomized patients to pumps with Fiasp (n=25) or NovoLog (n=12). On a primary endpoint of number of occlusions over six weeks, no confirmed cases occurred in either group, a positive lead-in to the proper large-scale Onset 5 trial of Fiasp in pumps – see above for those results as well (presented by Dr. David Klonoff), demonstrating Fiasp’s edge on postprandial glucose.

Select Questions & Answers

Q: Does Fiasp make a clinical difference for patients?

Dr. Russell-Jones: In the UK, Freestyle Libre has just been approved for coverage for people with type 1 diabetes. Anyone who puts on a CGM for the first time runs back to us when they find these huge post-meal excursions in blood glucose. Now we can say, “try this,” and see if they do improve with Fiasp.

Dr. Eda Cengiz: From a clinical care perspective, it is a huge difference if something improves post-meal blood glucose. Another thing is, 10 minutes faster onset is big if you’re sitting at the dinner table and everyone is waiting for you. I know Fiasp is not approved yet for children, but often our kids actually eat and then parents bolus afterwards for them. These “incremental” changes are important. In the future, we might see big improvements in artificial pancreas systems.

9. Dr. Stephanie Amiel Sheds Light on the “Unlearning” of Hypoglycemia Unawareness, Points to Specific Cortical Regions of the Brain

To a packed room (despite the 8 AM hour!), King's College London’s renowned Dr. Stephanie Amiel explained the neurological signatures of impaired hypoglycemia awareness, and she described her research into how awareness can potentially be restored. Educational interventions and the use of technology such as CGM or insulin pumps can improve hypoglycemia awareness, but it’s unclear exactly how this occurs. Dr. Amiel’s team performed fMRI scans on 12 people with hypoglycemia unawareness, both before and after an educational intervention. Although the data are still being analyzed, preliminary findings suggest a possible difference in anterior cingulate cortex (ACC) signaling before and after the intervention. This brain region is responsible for decision-making, conflict-resolution, evaluating actions, and detecting events that require a behavioral response – all of which is very consistent with a role for mediating the restoration of hypoglycemia awareness. Interestingly, the ACC is not one of the brain regions known to be involved in the development of hypoglycemia unawareness – among these, according to Dr. Amiel, are the globus pallidus (responsible for the aversive sensation that decreases with hypo unawareness), the pre- and post-central gyri (responsible for the symptomatic responses to hypoglycemia), and the temporal gyrus (responsible for consolidating the memory of a hypoglycemia experience). In our view, this raises the intriguing possibility that rather than simply reversing the neurological changes that mediate hypoglycemia unawareness, the awareness-restoring behavioral intervention activated an additional compensatory process mediated by the ACC. We eagerly await the full rundown of study results from Dr. Amiel. In the meantime, we’re fascinated by this potential mechanistic explanation for the “unlearning” of hypoglycemia unawareness.

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