June 10-14, 2016; New Orleans, LA; Full Report – Closing the Loop and Insulin Delivery – Draft

Executive Highlights

This document contains our coverage of closing the loop and insulin delivery at ADA 2016. Immediately below, we enclose our themes on the category, followed by detailed discussion and commentary. Talk titles highlighted in yellow were among our favorites from ADA 2016; those highlighted in blue are new full report additions from our daily coverage.

Themes

• The Automated Insulin Delivery (AID) field is now moving into the commercialization stage. Will the first hybrid closed loop system, Medtronic’s MiniMed 670G/Enlite 3, be on the US market by ADA 2017?! Medtronic reported strong 670G pivotal study results at ADA (see below), and an FDA submission is planned before the end of this month (June 2016). FDA’s Dr. Courtney Lias even expressed hope during ADA that the review will be quick, and the company’s pre-ADA Analyst Meeting maintained the US launch timing by April 2017. Usability will be key – one never knows exactly what the difference will be between using a technology “in the wild” (also known as real life) vs. in a very well managed clinical trial without a parallel control group. Hopes are high … key opinion leaders are also talking about 2017 commercialization for the 670G, meaning next year could be the year it all starts.
  
  • Meanwhile, commercial systems from 6+ companies are gearing up for pivotal studies in the next ~6-18 months. Six! Animas, Tandem, TypeZero/International Diabetes Closed Loop Consortium (IDCL), Beta Bionics, Bigfoot Biomedical, Insulet, and others have all announced pivotal study plans with commercial devices (roughly in pivotal trial order), which will make the next couple of years very competitive and exciting for this field. (See our competitive landscape here.) It’s been a fast turnaround from ADA 2015, where the 670G pivotal study was only beginning, and a transformative leap from ADA 2014, where the AID discussion focused nearly exclusively on academic systems.
• Medtronic headlined the AID discussion at ADA, presenting positive results from the first pivotal study of a commercial system, the MiniMed 670G/Enlite 3 hybrid closed loop. The single-arm, three-month study in 94 adults and 30 adolescents compared a two-week open-loop run-in period to 90 days on the 670G, showing a 0.5% reduction in A1c from a low study baseline of 7.4%; time <70 mg/dl declining 44% (from 6% to 3%); and time <50 mg/dl declining 40% (1% to 0.6%). It was impressive to see the impact on A1c, given the strong reduction in hypoglycemia and the very well-controlled population of pumpers. (We note that the pumpers were spending only 7% of the time below 70 mg/dl – that’s pretty great control to begin with.) Notably, those with a baseline A1c >7.5% actually saw a 1% reduction after three months on the 670G. Time-in-range (71-180 mg/dl) improved from 67% during the baseline run-in to 72% during the study, with time >180 declining moderately (27% to 25%). That improvement does not sound so significant, but it was clear from the glucose profiles that the MiniMed 670G made extreme highs (e.g., 300 mg/dl) more moderate (e.g., ~200 mg/dl) – a highly valuable improvement, but not showing up as a change to time in 70-180 mg/dl. The glucose profiles (see below) show how the 670G really tightened the range of glucose values throughout the entire day in both populations, with particularly strong efficacy overnight and in adolescents. The new Enlite 3 sensor is a definite improvement, with an overall MARD of 10.3% in this study and 10.5% in its separate pivotal accuracy study.
  
  • Overall, we see the 670G pivotal data as meaningful for the entire field of automated insulin delivery. This single-arm, uncontrolled study was clearly designed for speed and to prove safety to the FDA, so it’s hard to read too much into the efficacy results, where there were not pre-specified endpoints. Still, reducing A1c 0.5% in well-controlled patients at the same time hypoglycemia declining 44% is a huge win – and the results were consistent in adolescents and adults. It’s clear that automated insulin delivery can make a difference even with first-gen hybrid closed loop products, and even in those doing pretty well. Of course, the hope is it can really improve outcomes for those with much higher A1cs as well (for example, over >10%) or at high risk of severe hypoglycemia, groups that were both excluded from this study.
  • Stanford’s Dr. Trang Ly characterized the 670G results as “outstanding,” and said that the biggest challenge will be managing expectations: “This is not going to cure you,” she noted, as the first-gen 670G algorithm is still somewhat conservative, and hybrid systems like the 670G will still require patient effort – to be specific, carb counting, manual corrections, infusion set changes, sensor wear and calibrations, etc. The 670G algorithm itself has ten years of work behind it, but since it only modulates basal insulin delivery, it will be most efficacious overnight, and it will clearly not remove all patient burden during the day. We expect to hear more on the 670G’s usability and user-friendliness as it gets closer to market – some have said it requires a learning curve.
• Like so many other products, we expect those automating insulin delivery will improve with time, and the key will be in consistently improving the benefit-burden balance to expand the market. What will AID add onto patients’ and providers’ plates, and what will it remove? How will this balance vary based on what patients are already using – e.g., current pump-CGM users vs. MDI-SMBG users – and how will it change as insulins continue to improve, Bluetooth pens emerge, and standalone CGM sees better value with enhanced apps?
  
  • First-gen AID technology will likely be an upgrade for many current pump-CGM users, though the bar will be high for MDI-SMBG users – going from zero to two devices on the body will increase burden and cost, which means the corresponding improvement in outcomes needs to be big. For how many will that be worth it with gen one, gen two, etc.? What percentage of MDIs will ultimately wear AID systems?
  • What will most differentiate closed loop systems in a few years: outcomes/reimbursement, form factor, user interface, algorithms, other hormones, etc.? What will the adoption curve look like in five or ten years? What feature(s) will be the killer app(s) for AID that drive adoption – Full automation without meal announcement? Factory calibrated CGM? Smaller on-body devices? Integrated CGM/pump infusion sets? Will AID ever become standard of care in type 1? Will this technology appeal to MDIs that are struggling to manage blood glucose?
• The nuances of pivotal study design, standardizing outcome metrics, the reimbursement landscape, and the psychosocial impact are now top of mind in the field. Though these have always been in the discussion, all emerged as themes at JDRF’s annual Closed-Loop Research Meeting as well as in countless Closed Loop discussions, and all reminded us that the commercial side of closing the loop presents awesome design opportunities and nuanced challenges. Which system(s) and companies will nail these aspects in the coming years?
  
  • Pivotal Study Design: MGH’s Dr. Steven Russell summarized a soon-to-be-published Diabetes Care paper he co-authored with Dr. Roy Beck, sharing pivotal study design recommendations for AID systems – this will be fantastic to see! Drs. Beck and Russell recommend broad study inclusion criteria (MDIs included, wide age and A1c ranges), use of A1c and time <60 mg/dl as primary endpoints, a parallel group design (faster), 6-12 months in length, and compared to usual care. Pivotal AP studies have several goals besides regulatory approval – advantageous labeling, reimbursement, prescribing by practitioners, and adoption by patients – meaning study design decisions now will be critical down the road.
  • Reimbursement: Avalere Health’s Amanda Bartelme gave a fascinating overview of AID reimbursement, highlighting some of the key challenges: too much payer focus on A1c, hard-to-predict contracting negotiations, data needs that differ from FDA approval requirements, and more. Ms. Bartelme cautioned that “if a payer thinks every type 1 patient wants to go on this tomorrow, that’s huge dollar signs and huge panic.” It served as a reminder that nothing is a given with payers in this environment – even AID devices that reduce A1c, hypoglycemia, and patient burden will have to demonstrate return-on-investment, and ideally, short-term. Q&A sessions throughout the conference also expressed worry about AID reimbursement, as many clinicians are still battling payers to even cover CGM. From what we can tell, Medtronic will pursue the existing reimbursement channels for the MiniMed 670G (i.e., sensor-augmented pump reimbursed via DME), though this field seems ripe for a new business model (e.g., AID for $75 a month).
  • Standardizing Outcomes Metrics: Barbara Davis Center’s Dr. David Maahs (soon to be at Stanford) summarized another upcoming Diabetes Care paper focused on standardizing a short set of basic, easily interpreted outcomes in artificial pancreas studies. The paper has 24 authors, many of whom are considered leading thinkers in the field. The goal is for the entire field to report study outcomes the same way, easing interpretation, enabling basic comparison between studies, and accelerating adoption via regulators, HCPs, payers, and patients. We love this move!
  • Psychosocial impact: Stanford’s Dr. Korey Hood shared that a full set of validated questionnaires will be available by Fall 2016 to assess the psychosocial impact of automated insulin delivery. There are high expectations that AID will help manage glycemia and improve quality of life, but the field is still in need of tools to help regulatory approval bodies, payers, HCPs, and patients assess systems’ full benefit-risk balance. How do we measure better quality of life due to less diabetes-related stress or better sleep? As insulin delivery becomes automated, how do we protect against deskilling and human-machine interaction failures and confusion? What is the stigma associated with carrying extra devices in those not currently using a pump or CGM? Experiences will fall along a spectrum, of course, but these are critical questions as AID systems are on the cusp of commercialization. We hope these new questionnaires can help add context to the non-glycemic benefits of AID systems, and perhaps identify those that are optimal candidates for the technology. Our greatest hope is that AID massively improves outcomes and quality of life in those struggling on current therapies, whether they are using MDI, pump, SMBG, or CGM.
• Many called for closed-loop devices to include algorithms with customizable glucose targets. A patient panel at Diabetes Mine’s D-Data Exchange was clear that adjusting an algorithm’s aggressiveness is very key – some patients want more control, particularly in early-generation systems that will err on the conservative side. Indeed, Stanford’s Dr. Trang Ly pointed to this as an area for improvement in the Medtronic MiniMed 670G, which targets 120 mg/dl and does not allow the user to lower the target. The Bionic Pancreas team’s work comparing insulin-only to bihormonal control at different glycemic targets echoes the same point – an algorithm’s target does influence mean glucose and hypoglycemia, sometimes significantly so. Of course, there is a tough balance between customizability and simplicity – tweaking every parameter might be ideal for early adopters, but will add too much complexity that could hinder adoption. It’s a very delicate equilibrium, though targets will certainly go down over time as insulins get faster, sensors improve further, and the FDA and companies get more comfortable with these systems. It will be interesting to compare different commercial systems’ algorithms once they are available, as there may be meaningful differences in aggressiveness, meal announcement burden, initialization and training requirements, alarms, level of adaptation, and more.
• ADA 2016 shed the most light yet on OpenAPS, the DIY automated insulin delivery system created by Ben West, Dana Lewis, and Scott Leibrand. The community now has 80 users and has over 150,000 hours of AID use outside any clinical trial setting. An illuminating late-breaking poster presented fascinating data from 18 out of the first 40 OpenAPS users. Self-reported outcome measures showed median A1c dropped from 7.1% to 6.2%, an impressive 0.9% reduction in a well-controlled and motivated population. Self-reported median percent time-in-range (80-180 mg/dl) increased from 58% to 81%. Fourteen out of 15 respondents reported some improvement in sleep quality, and 56% reported a large improvement. Respondents were “extremely satisfied with the “life changing” improvements associated with using an APS,” even if they “require significant effort to build and maintain” and “cannot be considered a technological cure.” Though such “hacked together” DIY systems are often perceived as unsafe, the OpenAPS design considerations posted here show how it is designed for safety (e.g., only temporary basals, no automatic correction boluses, etc. – much like the 670G hybrid closed loop).
  
  • The “unapproved” OpenAPS concerned some clinicians at ADA. Overall, the small community left us with positive takeaways for the field: (i) automated insulin delivery can make a huge glycemic and quality of life difference, even for well-controlled patients; (ii) even though this DIY system is burdensome to set up and wear, patients would not do it and use it unless the benefits were worthwhile (hopefully a good sign for fully integrated commercial systems); (iii) lots of learning is occurring in the OpenAPS community that could be leveraged for commercial systems; (iv) OpenAPS could push the FDA and industry to move faster, and that is a good thing; and (v) the relative risks here seem low, given the involved burden of setup, the solid design for safety, and the real-world dangers of insulin therapy. 
• Some of the most compelling AID data came from the Cambridge team, who tested their system in inpatient type 2 diabetes – the first of this kind of closed-loop study ever done. The study shared striking improvements in efficacy and safety vs. the truly grim standard of care achieved with open-loop in the hospital. The parallel-arm study randomized 24 patients to receive either closed-loop therapy (n=12) or conventional subcutaneous insulin therapy per clinical guidelines with masked CGM (n=12) for a period of 72 hours. The data looked outstanding and terrifying at the same time – closed-loop control significantly improved time-in-target from 38% to 61% for the 100-180 mg/dl range (p<0.001). Mean glucose improved from 182 mg/dl to 161 mg/dl, just shy of statistical significance (p=0.065). The study used the Cambridge algorithm with unannounced meals, which made control much harder in the closed-loop arm. There was absolutely no difference in hypoglycemia (0% in both groups), and no severe hypoglycemia or adverse events were observed. We left the presentation reminded yet again of the very negative state of current inpatient glucose management. Indeed, we were downright disheartened by the standard of care overnight (mean glucose = 202 mg/dl), and the findings served as a striking reminder of: (i) the need for glucose management education in the hospital setting; and (ii) the great potential for inpatient technology to improve diabetes management and resulting outcomes. The tendency to accept hyperglycemia in inpatients is truly wrong, and we look forward to more studies of AID in this population.
• This ADA featured less insulin-only vs. bihormonal debate than at ATTD 2016 or last year’s Scientific Sessions. The need for a stabilized glucagon with chronic exposure data has pushed the Bionic Pancreas’ bihormonal timing to 2019-2020. Dr. Ed Damiano, CEO of Beta Bionics, revealed that Zealand’s liquid stable glucagon analog will be tested in clinical trials with the fully integrated iLet bionic pancreas in 2H16. The pivotal studies of the insulin-only iLet are still expected to start in 2Q17, with an FDA submission planned for the end of 2017. The bihormonal pivotal trial, which will begin after the start of the insulin-only pivotal trial, will require that a subset of the study cohort use the iLet for 12 months in order to gain chronic glucagon exposure for a new indication for use of glucagon in a bihormonal bionic pancreas. That puts the bihormonal FDA submission timing into ~early 2019, putting potential approval around late 2019 or 2020.
  
  • Timing aside, glucagon does allow more patients to reach a mean glucose <154 mg/dl without increasing hypoglycemia. The Bionic Pancreas team again shared their fascinating insulin-only vs. bihormonal glycemic target studies, first discussed at ATTD. The randomized, crossover study (n=20) compared usual care to insulin-only and bihormonal versions of the Bionic Pancreas at different glycemic targets (insulin-only: 130 and 145 mg/dl; bihormonal: 100, 115, 130 mg/dl) over three-day experiments. The insulin-only and bihormonal systems were actually very similar with a glycemic target of 130 mg/dl: a mean glucose of 161 vs. 156 mg/dl and time <60 mg/dl of 0.8% vs. 0.5%. As the bihormonal target dropped to 115 and 110 mg/dl, mean glucose improved to 146 and 136 mg/dl without increasing hypoglycemia. The team is now exploring an insulin-only target of 110 mg/dl, as the use of 130 mg/dl was intentionally conservative.

Detailed Discussion and Commentary

Posters

Pivotal Trial of a Hybrid Closed-Loop System in Type 1 Diabetes (99-LB)

R Bergenstal, B Buckingham, S Garg, S Weinzimer, R Brazg, J Ilany, B Bode, T Bailey, S Anderson, R Slover, J Shin, S Lee, F Kaufman

Medtronic presented a late-breaking poster on its single-arm, non-randomized, three-month pivotal study of the MiniMed 670G/Enlite 3 hybrid closed loop system in 124 participants (n=94 adults, 30 adolescents). The topline data compared to a two-week open-loop run-in period was excellent: (i) a solid 0.5% reduction in A1c from a low baseline (7.4%); time <70 mg/dl declined 44% (6% to 3%); and time <50 mg/dl declined 40% (1% to 0.6%). Nice! – it was impressive to see the impact on A1c, given the strong reduction in hypoglycemia and the very well-controlled population of pumpers. Notably, those with a baseline A1c >7.5% actually saw a 1% reduction after three months on the 670G. Time-in-range (71-180 mg/dl) improved from 67% during the baseline run-in to 72% during the study, with time >180 declining moderately (27% to 25%). While that improvement does not sound significant, it is clear from the compelling glucose profiles (see below) that the MiniMed 670G made extreme highs (e.g., 300 mg/dl) more moderate (e.g., 200 mg/dl), which didn’t show up as improvement in time in 70-180 mg/dl. The profiles are really worth looking at – they show how the MiniMed 670G tightened the range of glucose values throughout the entire day in both populations, with particularly strong efficacy overnight. Total daily dose increased ~7% from baseline (48 u to 51 u; p<0.001), though a smaller proportion of insulin was given as basal insulin (we unpack this below). Both adolescents and adults gained a bit of weight on the MiniMed 670G: +3 lbs (1.4 kg) in adults and +2 lbs (1.0 kg) in adolescents. While weight gain is never a positive, this does not strike us as big concern for the 670G or other hybrid closed loop systems, given potential for less hypoglycemia-induced weight gain. Hybrid closed loop was used for an impressive 87% of the three-month study period (median), with slightly higher usage in adults (88%) than adolescents (76%). The new Enlite 3 sensor showed a solid improvement, with an overall MARD of 10.3% vs. reference (i-STAT) over a 24-hour measurement period (part of the study’s six-day hotel phase). This was consistent with data shown at ATTD, though only 2% of points were in hypoglycemia.

• Overall, we see these pivotal data as very encouraging for the entire field of automated insulin delivery. This study was clearly designed for speed and to prove safety to the FDA, so it’s hard to read too much into the efficacy results, where were not pre-specified endpoints. Still, reducing A1c 0.5% at the same time hypoglycemia declined 44% is a huge win – and the results were consistent in adolescents and adults. This was also a very real-world study in current pump users, some of whom were on CGM prior to study start. It’s clear that automated insulin delivery will make a difference even with first-gen hybrid closed loop products, and even in well-controlled patients. Of course, the hope is it can really improve outcomes for those with an A1c >10% or at high risk of severe hypoglycemia, groups that were both excluded from this study.
• In talking to investigators, the patient reactions they’ve heard have been even more impactful than the data. Said Dr. Rich Bergenstal, “This is a definite step forward. Beyond what the numbers say, the true benefit is in the stories of sleeping at night, peace of mind. I’m happy to get a little help from technology.” With an FDA submission planned before the end of this month, and 80% of pivotal study patients using this device as part of the FDA’s continued access program, we assume many of the other pump companies are nervous about these results. Of course, that will push all of them to move faster and differentiate their offerings.

RESULTS

• The MiniMed 670G drove a strong 0.5% reduction in A1c from a low baseline (7.4%; p<0.001), including a 1% A1c reduction in patients with a baseline A1c >7.5%. Those with a baseline A1c of 7.0-7.5% saw a 0.3% reduction in A1c, while those with an A1c under 7% saw a 0.1% reduction in A1c. By the end of the study, 55% of trial participants had an A1c <7%, up from 31% at baseline.
  
  • The A1c improvement was consistent and highly significant across adults and adolescents: -0.5% in adults (baseline: 7.3%) and -0.6% in adolescents (baseline: 7.7%). Both were highly significant with p<0.001.
  • It is notable to see the impact on A1c, given the 44% reduction in hypoglycemia (see below) and the well-controlled population. We’re especially glad to see a strong 1% improvement in those with an A1c >7.5%, which brings hope this will be even more effective in those with an A1c >10% (who were excluded from the study).

	Baseline (open loop: pump+CGM)	Study Phase (hybrid closed loop)	Change	P-value
A1c – All	7.4%	6.9%	-0.5%	p<0.001
A1c - Adults	7.3%	6.8%	-0.5%	p<0.001
A1c - Adolescents	7.7%	7.1%	-0.6%	p<0.001

A1c Change if Baseline <7.0%	-0.1%
A1c Change if Baseline 7.0%-7.5%	-0.3%
A1c Change if Baseline >7.5%	-1.0%

• The MiniMed 670G tightened the range of glucose values throughout the entire day, particularly overnight. The modal day plots below show the median and interquartile range of sensor glucose values throughout the day and night in all patients (Panel A), adults (Panel B) and adolescents (Panel C). The gray band and dotted line shows the run-in phase, while the pink band and solid line show the study phase.
  
  • Consistent with pre-pivotal trials, the MiniMed 670G is most impactful overnight, driving patients down to ~140 mg/dl by early morning, tightening the range of overnight values, and eliminating hypoglycemia.
  • These modal plots really capture how much hybrid closed loop can make a difference at all times of day, even if it still requires manual boluses. Of course, systems are only going to get better with subsequent generations and faster insulin, and hybrid closed loop is a valuable first step.

Figure A: Modal Day – All Patients 

Figure B: Modal Day – Adults

• The 670G narrowed the range of glucose values in well-controlled adults, though was slightly more conservative than open-loop therapy during the day. Note in the below that the black average line at most daytime points is higher than the dotted average line. This makes sense given the well controlled, early adopter study population and first-gen hybrid closed loop. Future generation products with more aggressive algorithms and faster insulins should drive further improvement during the day.

Figure C: Adolescents

• The 670G showed excellent efficacy in the tough adolescent population, improving hyper glycemia at all times of day, particularly extreme highs after breakfast and moderate highs in the evening. Average glucose levels were fairly consistent, with perhaps a slight edge to the 670G.

• Time-in-range (% of SG in 71-180 mg/dl) improved from 67% during the baseline run-in to 72% during the study, with time <70 mg/dl nearly halved (6% to 3%), time <50 mg/dl declining 40% (1% to 0.6%), and time >180 improving moderately (27% to 25%). All had p<0.001, and the improvements were pretty consistent between adults and adolescents. [Exceptions: adults saw more benefit on time <50 mg/dl, while adolescents saw more improvement in time >180 mg/dl.]
  
  • While the improvement in time-in-range does not sound significant, there are several critical factors to consider: (i) the reduction in hypoglycemia is very meaningful; (ii) patients were doing quite well during the run-in (67% in range); and (iii) based on the glucose profiles below, the MiniMed 670G clearly made extreme highs (e.g., 300 mg/dl) more moderate (e.g., 200 mg/dl), which didn’t show an improvement in time in 70-180 mg/dl, but still improved overall glycemia (see profiles below). The latter also explains why A1c improved at the same time hypoglycemia declined (which should raise A1c) and time >180 mg/dl didn’t change significantly.

	Run-in (open loop: pump+CGM)	Study Phase (hybrid closed loop)	P-value
% of SG* <50 mg/dl	1.0%	0.6%	p<0.001
% of SG <70 mg/dl	5.9%	3.3%	p<0.001
% of SG 71-180 mg/dl	66.7%	72.2%	p<0.001
% of SG >180 mg/dl	27.4%	24.5%	p<0.001
Within Day SD of SG	50 mg/dl	47 mg/dl	p<0.001

* SG= Sensor Glucose

• The MiniMed 670G increased total daily dose ~7% from baseline (48 u to 51 u; p<0.001), though a smaller proportion of insulin was given as basal insulin. Since the MiniMed 670G only automates basal insulin delivery, the implication is patients were taking more manual boluses on hybrid closed loop (either in number or larger in size). Were they eating more or differently while on hybrid closed loop? If so, this could also explain the slight weight gain (see below)? The increase in insulin dose does not seem clinically significant, and many other studies have shown less or the same amount of insulin given.

	Baseline (open loop: pump+CGM)	Study Phase (Hybrid closed loop)	P-value
Total Daily Dose	47.5 u	50.9 u	p<0.001
% as Basal Insulin	53%	47%	p<0.001

• Both adolescents and adults gained a bit of weight on the MiniMed 670G: +3 lbs (1.4 kg) in adults and +2 lbs (1.0 kg) in adolescents. Oddly, these were both higher than the average weight gain for the overall study population reported on the poster (+1.5 lbs / 0.7 kg), so we’ve broken them out separately – we’re not sure what’s happening there but are following up with Medtronic.
  
  • While weight gain is never a positive, this does not strike us as a major concern for the 670G or other hybrid closed loop systems. We would not be surprised if further analyses reveal different eating behaviors while on hybrid closed loop. Keeping patients out of hypoglycemia could also help them lose weight, so the weight piece seems a bit difficult to interpret at this stage.

	Baseline (open loop: pump+CGM)	Study Phase (hybrid closed loop)	Change	P-value
Weight - Adults	79.9 kg	81.3 kg	+1.4 kg	p<0.001
Weight – Adolescents	67.4 kg	68.4 kg	+1.0 kg	p<0.001

• Hybrid closed loop was used for an impressive 87% of the three-month study period (median), with slightly higher usage in adults (88%) than adolescents (76%). This was not further specified, but we see this usage as encouraging in a well-controlled population. Plus, >80% of pivotal study participants have continued using the system through the FDA’s continued access program, a good sign the benefits are worth it from a patient perspective. 

ENLITE ACCURACY

• The new Enlite 3 sensor showed a solid improvement, demonstrating an overall MARD of 10.3% vs. reference (i-STAT) over a 24-hour measurement period (part of the six-day hotel phase). There were very few points in hypoglycemia (2%) and hyperglycemia (25%), making it hard to compare this accuracy with other sensors. Still, the results are pretty consistent with what we’ve seen for Enlite 3 in other studies, including 11% in the pre-pivotal study presented at ATTD. Larger and more robust studies of Enlite 3, presented in four poster abstracts here at ADA 2016 (879-P, 897-P, 901-P, 916-P), suggest a similar ~10% MARD.
  
  • Medtronic still recommends 3-4 daily calibrations for Enlite 3, though the poster did not specify how many occurred in this study. We’re also not sure on what day of the sensor life this 24-hour CGM-reference accuracy comparison occurred.

Reference Glucose Range	Mean Absolute Relative Difference (MARD)	Percentage of Points in Range
Overall	10.3%	-
>180 mg/dl	11%	25%
71-180 mg/dl	9.8%	73%
<70 mg/dl	12 mg/dl*	2%

STUDY STRENGTHS AND LIMITATIONS

Strengths	Limitations
Multicenter design to evaluate safety (10 sites) Large number of subjects, both adults and adolescents, using the system for 24 hours per day Three months of unsupervised home use of system Time in target confirmed by reference blood glucose measurements during hotel stay	Single-arm, non-randomized design with no pre-specified efficacy endpoints Data quantity imbalance between run-in (two weeks) and study phases (three months) Exclusion of subjects with A1c >10%, recent episodes of severe hypoglycemia or recent DKA

STUDY BACKGROUND AND POPULATION

• This single-arm, non-randomized study enrolled pump users (≥6 months), with or without current use of CGM. Participants had to have type 1 diabetes for ≥2 years, an A1c <10%, and age 14-21 years old (adolescents) or 22-75 years old (adults).
  
  • Adolescent participants (n=30) had a mean age of 17 years, mean A1c of 7.7%, a mean BMI of 24 kg/m², and a total daily dose of 0.8 u/kg/day. The study enrolled 16 adolescents females and 14 adolescent males.
  • Adult participants (n=94) had a mean age of 45 years, mean A1c of 7.3%, a mean BMI of 27 kg/m², and a total daily dose of 0.6 u/kg/day. The study enrolled 53 adult females and 41 adult males.
• The MiniMed 670G/Enlite 3 was used in open-loop mode (with CGM) during a two-week run-in phase (baseline), then in closed-loop mode in a three-month study phase (unsupervised). The three-month phase included a six-day, five-night hotel stay for supervised activity and frequent venous blood glucose measurements (during one 24 hour period) with a reference instrument (i-STAT). This study was conducted at 10 sites in the US and Israel. 
• The 670G hybrid closed loop algorithm (ePID) automatically increases or decreases basal insulin, but all boluses require user input and confirmation (i.e., meal and correction boluses). Though not noted on the poster, here’s what we know about the algorithm: it targets 120 mg/dl, which can be raised to 150 mg/dl during exercise; it has a max limit on insulin delivery per hour and uses basal modulation to keep blood glucose in range (i.e., a hybrid closed loop that still requires meal boluses, and it cannot just correct a 350 mg/dl in one big correction bolus); it gives a new dose every five minutes; the 670G uses open-loop parameters to initialize hybrid closed loop (total daily insulin, basal, insulin:carb, insulin sensitivity factor); at the start of hybrid closed loop, there is a sensor accuracy check, along with a glycemic target adjustment for a smooth transition to closed-loop; the algorithm can adapt over time as things change; the 670G will revert to open loop if the sensor is inaccurate; and it will switch to safe mode or the pre-programmed basal rate in cases like sensor failure.
  
  • It’s worth noting that a missed meal bolus on the 670G hybrid closed loop could still mean several hours above range, as the higher basal rate will take a while to bring blood glucose back in zone. Still, the system is clearly much better than most patients are doing right now, and the hybrid closed loop is a great way to go until insulin gets faster and algorithms get even smarter.

POST 670-G UPDATE

• Medtronic’s post-670G product will further close the loop by performing automatic correction boluses based on CGM values. Medtronic’s dinner during ADA suggested it will enter a clinical study this month (June) and be called the MiniMed 690G. This product will integrate the DreaMed MD-Logic algorithm to close the loop further – instead of only increasing basal insulin to gradually mitigate highs (670G) and bring blood glucose back to target, the next-gen algorithm will add automatic bolusing to correct highs. That should improve time-in-range much more and make the system more aggressive with missed meal boluses. The pre-ADA Analyst Meeting showed a picture of a smaller, touchscreen-looking, future-gen pump platform, but we’ve confirmed with Medtronic that is not the 690G.

Real-World Use of Open Source Artificial Pancreas Systems (104-LB)

D Lewis, S Leibrand, and the #OpenAPS Community

This illuminating poster presented self-reported outcomes from 18 out of the first 40 users of OpenAPS, the DIY automated insulin delivery system created by Ben West, Dana Lewis, and Scott Leibrand (now over 150,000 hours of AID use outside any clinical trial setting!). While using OpenAPS, self-reported outcome measures showed median A1c dropped from 7.1% to 6.2%, an impressive 0.9% reduction in a well-controlled and motivated population. Self-reported median percent time-in-range (80-180 mg/dl) increased from 58% to 81% - consistent with presentations of actual data we’ve seen recently from Mark Wilson (Day #1) and Chris Hanneman (D-Data last fall). Fourteen out of 15 respondents reported some improvement in sleep quality, and 56% reported a large improvement. Respondents were “extremely satisfied with the “life changing” improvements associated with using an APS,” even if they “require significant effort to build and maintain” and “cannot be considered a technological cure.” The poster notes that OpenAPS is designed to be, and has been, far safer than standard pump/CGM therapy, as measured by duration of hypoglycemia and hyperglycemia, with no reports of severe hypoglycemia or hyperglycemic events. The OpenAPS design considerations posted here are pretty instructive on the safety front (only temp basals, no automatic correction boluses, etc. – much like the 670G hybrid closed loop!). Our takeaways from this poster and inspiring community are: (i) automated insulin delivery can make a huge difference, even for well-controlled patients; (ii) even though the system is burdensome to set up and wear, patients would not do it and use it unless the benefits were worthwhile; (iii) lots of learning is occurring in the OpenAPS community that could be leveraged for commercial systems; (iv) OpenAPS could push the FDA and industry to move faster, and that is a good thing; and (v) the relative risks here seem low, given the setup burden, the solid design for safety, and real-world dangers of insulin therapy. 

• As an aside, and as would be expected, patient researchers – like any other researchers – buy badges to present posters. That is part of supporting ADA in bringing together so many researchers for discussion. However, patient-researchers do not have funding from work environments like manufacturers or universities. This creates a wonderful opportunity for a foundation or other organization to endow funds to create a pathway for patient researchers to not lose savings to present their data. We hope patient research will increasingly be supported by the existing healthcare community, since greater dialogue can be particularly beneficial. We are eager to see more patient learning make its way into professional organizations, such as Dr. Joyce Lee’s commentary from “Digital Health: Hope or Hype?” and we hope that can change. We salute Dana and Scott for submitting this poster and getting it accepted as a late-breaker and we look to the field to come up with creative solutions to support this work.
• OpenAPS now has over 80 users worldwide, though only 40 were using the system at the time of abstract submission, which makes this 18-person evaluation a near 50% response rate. Of course, as with any other research, the self-reported component of the outcomes may be interpreted cautiously.
• The poster has an instructive discussion section, noting that some healthcare providers are supportive of OpenAPS, and others showed a “lack of interest.” However, OpenAPS experiences “are instructive for what patients can expect from commercial APS when they become widely available, and can help HCPs be prepared to set patients’ expectations properly when discussing or recommending APS.”  We totally agree and hope to see more dialogue between the traditional healthcare community and the OpenAPS community. 
• The poster’s questions for HCPs to consider are also fascinating:
  
  • Artificial pancreas systems are already here. One of your patients may already be building one. Would you know it if they are? Do you discuss with your patients which tools they choose to use to help manage their diabetes?
  • APS are a powerful tool, but not a cure. Patients and HCPs will still need to do a lot of work to use them effectively to improve diabetes management.
  • Patients and HCPs must educate themselves and each other on how APS can be used effectively in daily life.
• Though OpenAPS has improved in wearability and form factor, it still requires carrying extra gear, as accessing pump commands remains difficult. The community has posted all the reference design, documentation, code, and community channels at www.openAPS.org, though this system requires a fair amount of effort and motivation to put together – hence why we don’t see it as high risk right now. In the process of building it, users must intimately understand how it works, and it is certainly not plug and play. We know smart people using the system now that spent hundreds of hours setting it up.

Patient User Experience Evaluation of Bolus Patch Insulin Delivery System (995-P)

V Zraick, D Dreon, R Naik, D Shearer, S Crawford, J Bradford, and B Levy

This poster presented solid data demonstrating that patients (n=44; 40 with type 2, four with type 1) were very pleased with J&J’s OneTouch Via (formerly Calibra’s Finesse), a “discreet, wearable, on-demand, mealtime insulin delivery solution.” Over 50% of the cohort completed product training within a half hour – a strong testament to the usability of the bolus-only, three-day wear device. Following the eight-week trial, 86% of users reported being extremely/very satisfied with the system, and 79% were extremely/very likely to request a prescription from their HCP. Similarly, 74% of patients said that they would incorporate the Via into their routine. There was an interesting learning curve that emerged in the data: After week one, patients reported that they dosed with the Via the same number of times that they typically would with their pen/syringe. By the midway and end points of the study, patients had adjusted to the device, and >50% reported injecting prandial and snack-time insulin more frequently than they had with their pens/syringes. Of course, because information about dosing frequency was self-reported, it’s hard to know how dosing actually changed. We will be curious to see the results of the ongoing OneTouch Via outcomes study (n=312), which has a primary completion date set for this December. J&J reported at its Medical Device Business Review last month that it will file OneTouch Via for approval in 2H16, and we’ve learned from the company that it will be commercially available in select market outside the US by late 4Q16, with US to follow soon thereafter in early 2017 – this is an update from the Medical Device Business review, which called for a launch within the next 12 months (by May 2017).

• HCPs were also big fans of the OneTouch Via! Every single one was satisfied at the end of the eight-week usage period and was likely to recommend. A large majority also viewed the Via favorably when compared with syringes and pens.
• This simple device has the potential to improve regimen adherence. With the Via, the majority of users in this study reported satisfaction with their abilities to (i) discreetly and easily administer a bolus in public without painful injections, (ii) worry less about the possibility of forgetting pens/syringes, and above all, (iii) lead less stressful lives. Thus the Via overcomes many barriers to usage associated with MDI.

Glucommander Outpatient, a Cloud-based Insulin Management Solution Adjusted Insulin Doses and Achieved 2.7% Drop in A1c Percentage Points (84-LB)

John G. Clarke, Bruce W. Bode

A Glytec poster (84-LB) showcased very impressive results from a 41-patient, uncontrolled, 3-month, outpatient study testing its Glucommander insulin dosing clinical decision support software – from a high baseline A1c of 10.3%, patients ended three months with an estimated average A1c of 7.6% (p<0.000001). The study enrolled 41 type 1 and type 2 patients (mean age: 38 years, BMI: 32 kg/m²) at Dr. Bruce Bode’s clinic in Atlanta, who were treated for 12 weeks with Glucommander Outpatient. The cloud-based software provided periodic insulin dose titration recommendations to a provider based on analysis of a patient’s SMBG glucose data, communicated wirelessly via the cellular-enabled Telcare meter. The provider then communicated the new insulin doses to patients via text message or email. The topline findings from this small study are very impressive – patients using Glucommander saw a 2.7% reduction in A1c (baseline: 10.3%) at three months, and only 1.6% of blood glucose values were <70 mg/dl. Strikingly, no values were <40 mg/dl and, on the human factors side, patients satisfaction results indicated that 96% of patients would recommend the service to family and friends. The poster hinted at Glytec’s strong long-term data as well, citing a smaller cohort of patients that have continued on Glucommander for six (n=14) and nine (n=5) months and have maintained this 2.7% reduction. Small cohorts, but still, this is a whopping improvement. The outcomes are encouraging given the challenges of titrating insulin and the potential for this software to scale expertise, though larger prospective randomized clinical trials are needed to confirm these positive early findings from an uncontrolled study. The company does plan to begin a larger study that includes cost-related metrics such as readmissions, emergency room visits, medication adherence, and healthcare provider productivity, and we’re hopeful that data will show this kind of clinical decision support is very warranted (a “no-brainer” many say). We’re not sure what the business model looks like going forward, but assume Glytec’s in-hospital experience will be very valuable as it thinks about going outpatient. As a reminder, Glucommander Outpatient is already FDA-cleared and is in the process of being deployed across the US. See our previous in-depth coverage here.

• How could the Glucommander software be scaled? Could it be packaged with existing devices or even drugs? We’ve long thought that insulin-dose titration is a missing piece in the diabetes data ecosystem, and this early data shows how much can be done (and parallels what Hygieia has shown in Europe). We wonder how this Clinical Decision Support software could be packaged with existing devices or even drugs on the market to enhance their effectiveness in the hands of providers. We also have to assume this product saved tremendous provider time, and we look forward to seeing larger studies showing cost-effectiveness. This is where we see digital health really driving better outcomes: collecting data seamlessly and making valuable recommendations that drive seriously better outcomes with less effort.

Do Type 1 Diabetes Patients Really Want An Artificial Pancreas? (1005-P)

S Franc, I Xhaard, L Orlando, M El Makni, M-H Petit, C Randazzo, and G Charpentier

This poster investigated real-world attitudes about closed-loop systems, asking 101 patients with type 1 diabetes to fill out an artificial pancreas questionnaire before and after a presentation about what such a system entails. The informational forum increased the number of patients expressing a desire to use automated insulin delivery – from 40% to 67% – though we were far more surprised by what the questionnaire revealed about the confusion around “artificial pancreas” terminology. Before the session, 42% of patients thought that an artificial pancreas involved an organ graft and 18% thought it was a smartphone app; these numbers changed to 16% and 68%, respectively, following the informational forum, though we’re still a bit shocked that many patients maintained these perspectives following the session [It does makes us wonder about how rigorous and engaging this training was, and how hard automated insulin delivery is to explain.] The session was more successful, however, at convincing patients that the artificial pancreas will be safe and beneficial: the percentage of patients reporting that they would be “extremely likely” to wear an artificial pancreas if it were available grew from 24% before the forum to 41% after it. Overall, it’s tough to read too far into the results considering how little is known about how these questions were asked, how the informational session was conducted, and who these patients were – however, the findings do hint at a general lack of knowledge about these devices and at the need for more rigorous educational efforts to counter misperceptions about what an artificial pancreas actually is.

• The poster reported that the recency of diagnosis (p=0.014), the existing use of pumps vs. MDI (p=0.058), and the willingness to use a smartphone to manage the device (p=0.038) were all positive predictors of desire to use an artificial pancreas. However, a number of other factors were not positively correlated: (i) dissatisfaction with current therapy; (ii) hope of improved HbA1c levels, decreased risk of hypoglycemia, or associated complications; (iii) hope of improved freedom or comfort; and (iv) the length of time needed to develop the device.
• Given that pump use positively predicts desire to use the artificial pancreas, we wonder whether this study may have overestimated the number of patients that would be accepting of such a device. Notably, 67% of patients who filled out the survey reported that they use an insulin pump. In a cohort that is more representative of the global MDI vs. pump distribution, we wonder whether fewer subjects would have expressed interest in an artificial pancreas.
• The artificial pancreas will make insulin therapy safer, increase time-in-range, and could improve burden, but patients certainly will not accept it if they do not understand it. We came away from this study reminded that for the product to make the impact we hope it will, the field needs a much more coordinated education effort to teach patients and providers how an artificial pancreas works and how it could benefit them.

Barriers to Device Uptake in Adults with Type 1 Diabetes (914-P)

M Tanenbaum, S Hanes, K Miller, D Naranjo, and K Hood

This study invited 1,503 adult patients (mean age=35 years) in the T1D Exchange to take a 30-minute web-based survey in order to understand barriers to insulin pump and CGM uptake. Coming into the survey, 32% of patients reported using both a pump and CGM, while 5% used just a CGM, 38% used just a pump, and 25% used neither. Unsurprisingly, a majority of patients cited financial burden as the biggest barrier to the use of either device – 60% of patients expressed concern about insurance coverage, the cost of the device, and the cost of supplies. Other popular barriers appeared far more modifiable: 35% did not like having diabetes devices on their bodies, 47% did not like the hassle of having to wear the device all of the time, 26% did not like how the diabetes devices looked on their bodies, and 20% were nervous that the device might not work. The survey also asked patients who had discontinued use of their devices for their rationale – patients reported abandoning CGM because of too many alarms, inaccurate data, a distaste for the device on their body, time requirements, or discomfort, while a majority of patients discontinued use of pumps because they didn’t like the device on their bodies or because the device was uncomfortable. We wonder how attrition breaks down by manufacturer. Younger adults (18-25 years old) were less likely to use devices than older adults, and this younger population had higher levels of diabetes distress and higher A1cs. Overall, findings suggest that cost remains the biggest barrier to address, but size on the body is not far behind. We wonder if many of the CGM quitters were on earlier systems that were less accurate (e.g., Seven Plus), and perhaps they would be less frustrated with the more accurate out now or coming soon. Of course, with self-reported data, there is always some question about the reliability of results, though the data echo much of what we hear about the real-world barriers to device uptake anecdotally.

• We’d note that CGM users in this study were five times more likely to be on a pump (38% used pump+CGM vs. 5% used MDI+CGM), echoing what Dexcom has long said – patients are more likely to be prescribed a CGM if they are already on a pump. We hope the positive results from Dexcom’s DIaMonD study can change that (see Drs. Howard Wolpert and Elena Toschi’s talks elsewhere in this report).
• The survey also compared the differences between CGM users and non-users. CGM users were, on average, five years older than non-users (38 years vs. 33 years; p <0.001), viewed technology more favorably, and had significantly lower A1cs (7.3% vs. 7.7%; p =0.003).

Closed-Loop Control Reduces Hypoglycemia during Extended Winter-Sport Exercise in Youth with T1D: The AP Ski Camp (103-LB)

Daniel R. Chernavvsky, Mark Deboer, Jessica Robic, Boris P. Kovatchev, Marc D. Breton

This poster shared findings from an RCT that investigated the efficacy and durability of UVA’s DiAs closed-loop system (n=8) vs. SAP therapy (n=8) in 16 adolescents with type 1 diabetes at a five-day ski camp (five hours of skiing/day!). The intense physical activities – compounded by altitude (~1,300 meters) and cold weather (-10 degrees Celsius) – brought a higher risk of hypoglycemia and a super challenging setting for testing closed loop. Overall, findings were consistent with the group’s previous impressive results and indicated that patients on closed-loop therapy experienced far tighter glycemic control than those on SAP – overall time spent <70 mg/dl decreased from 4.1% (SAP) to 1.6% (CL) (p=0.008) and nocturnal time spent <70 mg/dl decreased from 3.6% (SAP) to 1.6% (CL). Consistent with these findings, the incidence of overnight hypoglycemia treatments were cut roughly in half on closed-loop control (2.6 treatments/subject/day vs. 5.3 treatments/subject/day (p=0.04). There were no AP-related adverse events and patient evaluations of the system and study were reportedly overwhelmingly positive. Ultimately, the results confirm that UVA’s closed-loop system performs reliably and safely in cold temperatures, reducing hypoglycemia during and after intense prolonged exercise. The UVA’s system’s daytime target is 160 mg/dl, which makes it a bit conservative for tightly controlled patients, but excellent at mitigating lows. Bigger picture, we loved the very real-world nature of this study, and as systems get closer to commercialization, we’d love to see additional groups test their closed-loop systems in similarly rigorous environments in even larger populations. These “edge cases” will inform patients’ experience with closed-loop systems, and it’s crucial (for safety and peace of mind) that products are robust to the toughest real-world challenges: sensor offline, sensor inaccurate, incorrect fingerstick calibration, kinked or occluded infusion set, denatured insulin, water, high heat, etc.

Oral Presentations: Closing the Loop on Insulin Management – Are We There Yet?

Automated Artificial Pancreas System in Type 2 Diabetes in the General Ward: A Randomised, Controlled, Parallel-Design Study (84-OR)

Hood Thabit, MD (University of Cambridge, UK)

Dr. Hood Thabit presented brand new, exciting data from the Cambridge team’s automated insulin delivery system in inpatient type 2 diabetes – the first of this kind of closed-loop study ever done. The study shared striking improvements in efficacy and safety vs. the truly grim standard of care achieved with open-loop in the hospital. The parallel-arm study randomized 24 patients to receive either closed-loop therapy (n=12) or conventional subcutaneous insulin therapy per clinical guidelines with masked CGM (n=12) for a period of 72 hours. The data looked outstanding and terrifying at the same time – closed-loop control significantly improved time-in-target from 38% to 61% for the 100-180 mg/dl range (p<0.001). Mean glucose improved from 182 mg/dl to 161 mg/dl (p=0.065), though that signal was not significant. The study used unannounced meals, which made control much harder in the closed-loop arm. There was absolutely no difference in hypoglycemia in this type 2 population: time spent < 63 mg/dl = 0% [CL] vs. 0% [OL]. On safety, there were no severe hypoglycemia or adverse events associated with diabetes therapy and total daily insulin did not differ between the groups (63 units per day [CL] vs. 66 units per day [OL]). Ultimately, we were not sure what to expect coming into this oral, but the group’s impressive track record certainly delivered in another new setting. Ultimately, we left the presentation reminded yet again of the very negative state of current inpatient glucose management. Indeed, we were downright disheartened by the standard of care overnight (mean glucose = 202 mg/dl; see below), and the findings served as a striking reminder of: (i) the need for glucose management education in the hospital setting; and (ii) the great potential for inpatient technology to improve diabetes management and resulting outcomes. The tendency to accept hyperglycemia in inpatients is truly wrong.

• We were particularly impressed with data overnight Dr. Thabit shared – see Table 1 below. Closed-loop control doubled nocturnal time-in-target from 29% to 59% for the 100-180 mg/dl range (p<0.001), while mean glucose improved from 202 mg/dl to 161 mg/dl (p=0.065) with no hypoglycemia. There were zero episodes of glucose < 63 mg/dl on closed-loop vs. isolated incidents on open loop, making the hypoglycemia difference significant overnight too. We imagine nurses and other hospital-based HCPs would appreciate the chance to safely reduce blood glucose – many are very scared from what we know to have patients at “normal” blood glucose levels, given the risk (real or perceived) of hypoglycemia. Many inpatients likely run “high” against their wishes.
• This trial did NOT use meal announcements, making this the first fully automated closed-loop Cambridge study. All the type 1 studies have used meal announcements, but Dr. Thabit stressed that a fully automated system is particularly practical in the inpatient type 2 setting where patients and nurses are less well educated – more on this below.
• Dr. Thabit shared very positive user experience data from type 2 patients on the closed-loop system. According to Dr. Thabit, a majority of these patients had never been on a device previously, making these glowingly positive results all the more impressive.
  
  • Were you happy with your glucose levels in the hospital during the study?
    • Better than expected: 17 patients.
    • What you expected: 3 patients.
    • Worse than expected: 0 patients.
  • Were you happy to have your glucose levels controlled automatically by the system?
    • Better than expected: 18 patients.
    • What you expected: 1 patient.
    • Worse than expected: 1 patient.
  • If a friend or family member was in the hospital, would you recommend this system to them?
    • Yes: 19 patients.
    • No: 1 patient.
  • Would you be willing to wear a portable version of this system as part of your diabetes treatment at home?
    • Yes: 17 patients.
    • No: 3 patients.

Questions and Answers

Q: This is a very important study. Thank you. On the first day when sensors were not as accurate, did you wait any period of time before starting closed loop?

A: We did not wait. Once the sensor was put in and one hour warm-up period elapsed, closed loop went online.

Q: In the closed-loop group, how did you decide how much Lantus to use?

A: Our decision to give 20% Lantus was a pragmatic one after chatting with clinical colleagues abut what would avoid ketosis.

Dr. Hans DeVries (Academic Medical Center, Netherlands): I believe this is the first study from Cambridge when meals were not announced. Can you talk about the rationale behind that?

A: The decision to go with a fully closed-loop system was pragmatic because we expect this will be used this way in the real world. The last thing we want is for nurses or patients to miss a meal announcement. Our hypothesis was that in the type 2 diabetes population in the hospital, a fully closed-loop system is needed because patients and nurses are not as well educated.

Q: What was the patient perception like?

A: We got a very positive endorsement from patients. Many patients had never worn devices in their lives and live had a very good experience.

Similar Estimated A1c Results Reported between Patients with Diabetes Using CGM whether on Multiple Daily Injection (MDI) or Continuous Subcutaneous Insulin Infusion (CSII) (81-OR)

David Price, MD (Dexcom, San Diego, CA)

Dexcom’s Dr. David Price shared a retrospective database evaluation from six months of G4 Share users, showing no differences in mean glucose, estimated A1c, or glucose variability between pump (n=939) or MDI (n=648) users. The de-identified data on glucose values were supplemented by customer info (age, insulin delivery) that Dexcom collects. The pump and MDI groups had identical average glucose values and variability across every age group (from 2-6 year-olds all the way to 65+ year olds), with just a single small exception: glucose variability was statistically significantly lower in adolescents (13-18 years) using injections. As would be expected, adult CGM users had better average glucose values vs. pediatric CGM users by ~30 mg/dl (see tables below estimated from the charts shown). This analysis also aligns with results from the T1D Exchange showing that in every age group, the same pattern holds – similar A1cs for CGM users on MDI or a pump. Dr. Price noted that CGM use is increasing (now up to 16% in the T1D Exchange), but is overwhelmingly prescribed to pumpers: of all Exchange CGM users, 85% are on pumps vs. just 15% on MDI. Studies like this also underscore the inherent value in data streaming from devices to the cloud automatically – companies can use it to put data behind their arguments, drive study design, and inform marketing. Medtronic has written the book on this with CareLink, and we expect Dexcom will begin driving this too. 

All Pediatrics 2-18 years
	MDI (n=300)	Pump (n=301)	P-value
Mean CGM Glucose	~180 mg/dl	~180 mg/dl	0.92
Estimated A1c	~7.9%	~7.9%	--
Standard Deviation	~65 mg/dl	~60 mg/dl	0.39

~ Estimated from Bar Graphs

Adults >18 years
	MDI (n=403)	Pump (n=369)	P-value
Mean CGM Glucose	~158 mg/dl	~159 mg/dl	0.55
Estimated A1c	~7.1%	~7.1%	--
Standard Deviation	~57 mg/dl	~59 mg/dl	<0.23

~ Estimated from Bar Graphs

Home Use of a Bihormonal Bionic Pancreas vs. Conventional Insulin Pump Therapy in Adults with Type 1 Diabetes—A Multicenter, Randomized Clinical Trial (77-oR)

Ed Damiano, PhD (CEO, Beta Bionics / Boston University, Boston, MA)

Beta Bionics’ CEO Dr. Ed Damiano revealed that Zealand’s phase 2 liquid stable glucagon analog will be tested in the 4Q16 bridging study of the fully integrated iLet Bionic Pancreas. The pivotal studies are still expected to start in 2Q17, and the bihormonal pivotal will last 12 months to gain chronic glucagon exposure data. Zealand actually put out a press release announcing the collaboration with Beta Bionics, a big win for Dr. Damiano’s new public benefit corporation to commercialize the Bionic Pancreas. Xeris has not moved particularly fast on its stable glucagon, and Zealand’s deep experience in protein chemistry will be a major asset on this intractable problem. Dr. Steven Russell hinted at ATTD that Zealand’s phase 2 glucagon might be used, though today was the first confirmation it is the glucagon of choice going forward. Dr. Damiano did not comment on the submission timing, but we assume the plan is still an end of 2017 insulin-only PMA submission, with a potential PMA supplement to add glucagon in early 2019.

• Dr. Damiano’s oral presentation focused on the Bionic Pancreas 11-day multi-center home study (n=39), shared at several symposium presentations since GTCBio 2015 over a year ago. As a reminder, the study was the team’s first true home-use study, comparing 11 days of Bionic Pancreas to 11 days of conventional pump therapy. Mean CGM glucose improved from 162 mg/dl on usual care to 141 mg/dl on the Bionic Pancreas, projecting an A1c improvement of 0.8% (baseline: 7.3%). Time <60 dropped by two-thirds (from 1.9% to 0.6%), while time >180 declined from 34% to 20%. Dr. Damiano emphasized the device’s ability to dramatically reduce inter-subject variability, to not deliver excess insulin (0.62 u/kg/day vs. 0.66 u/kg/day), and to adapt to patients based on using just body weight at initialization.

Outpatient Glycemic Management in Type 1 Diabetes with Insulin-Only vs. Bihormonal Configurations of a Bionic Pancreas (79-OR)

Laya Ekhlaspour, MD (MGH, Boston, MA)

MGH’s Dr. Laya Ekhlaspour reiterated the insulin-only vs. bihormonal glycemic target studies first shared at ATTD. “Glucagon allows more subjects to have a mean glucose <154 mg/dl without increasing hypoglycemia.” We’ve reproduced the key findings below from the randomized, crossover study (n=20) comparing usual care to insulin-only and bihormonal versions of the Bionic Pancreas at different glycemic targets (insulin-only: 130 and 145 mg/dl; bihormonal: 100, 115, 130 mg/dl). The insulin-only and bihormonal systems were actually very similar with a glycemic target of 130 mg/dl: a mean glucose of 161 vs. 156 mg/dl and time <60 mg/dl of 0.8% vs. 0.5%. As the bihormonal target dropped to 115 and 110 mg/dl, mean glucose improved to 146 and 136 mg/dl without increasing hypoglycemia. The team is now exploring an insulin-only target of 110 mg/dl, as the use of 130 mg/dl was intentionally conservative.

• This study shows just how important glycemic target is to system performance, and echoes the takeaway from the D-Data Exchange that closed-loop system should have adjustable target set points. The big question for glucagon, of course, is how much mean glucose, hypoglycemia, and user experience improves when the hormone is added to an insulin-only system.

System Target BG	Control	Insulin Only 145 mg/dl	Insulin Only 130 mg/dl	Bihormonal 130 mg/dl	Bihormonal 115 mg/dl	Bihormonal 100 mg/dl
Mean	158 mg/dl	174 mg/dl	161 mg/dl	156 mg/dl	146 mg/dl	136 mg/dl
Time <60 mg/dl	1.4%	1.0%	0.8%	0.5%	0.9%	0.8%

Randomized Crossover Clinical Trial Comparing MPC and PID Control Algorithms for Artificial Pancreas (80-OR)

Frank Doyle, PhD (Harvard, Cambridge, MA)

Dr. Frank Doyle presented results from what he termed the “first, balanced, randomized study” comparing MPC and PID closed-loop control algorithms that came to two major conclusions: (i) that MPC outperformed PID on both the primary outcome and several secondary clinical metrics; and (ii) that both algorithms provided safe and effective glycemic control. This is the first time we’ve seen results from this crossover study, which randomized 20 patients to each algorithm for a supervised 27.5-hour session that incorporated both announced and unannounced meal challenges. Broadly, findings very strongly favored the MPC algorithm – patients using MPC saw significantly greater time in range (70-180 mg/dl) throughout the study vs. those on PID (74% vs. 64%, p=0.02). Patients on MPC also seriously reduced their mean glucose both during the entire trial (138 vs. 160 mg/dl, p=0.012) and during the five-hour period after the unannounced meal (181 vs. 220 mg/dl, p=0.019). Dr. Doyle noted that there were no statistically significant differences in hypoglycemia (<70 mg/dl) though he did acknowledge that the frequency of hypoglycemic episodes was higher in the MPC group. Ultimately, we would point out that these were both relatively basic iterations of both control systems and that the results do not preclude the use of PID in the closed-loop setting or suggest that all MPC systems are superior to those leveraging PID. Instead, we felt Dr. Doyle made the case that the core of the MPC algorithm may be better suited to managing artificial pancreas (insulin delays and pump constraints), in addition to that fact that MPC is a more flexible platform for adding other functionality.

• For context, Dr. Doyle shared that standard and matched versions of both the MPC and PID algorithms were developed using well-known model-based methods. The algorithms had identical set-point control objectives (110 mg/dl) and had similar built-in features to prevent insulin stacking to ensure equitable testing conditions. Indeed, Dr. Doyle presented data from an in silico modeling experiment confirming that both algorithms performed very comparable in this controlled setting.
• Study design: Dr. Doyle explained that both algorithms were compared in two supervised 27.5-hour closed-loop sessions. Challenges in the study were designed to mimic the use of an artificial pancreas in the real world and to stress the algorithms. These challenges included no prior optimization of insulin pump parameters to initialize the system, overnight control after a 65 g carbohydrate (CHO) dinner, response to a 50 g CHO breakfast (both bolused at mealtime), and an unannounced 65 g CHO lunch to evaluate a missed meal bolus scenario.

Questions and Answers

Q: This may be an artifact of the short study but I noticed that your two algorithms were pretty convergent for first day and where you saw separation was in evening. Do you have differences between these algorithms overnight? Because that’s when you really see that separation occurring.

A: There was insulin-on-board vs. insulin feedback built into the MPC and PID algorithms, respectively, so it could have to do with that. I will say that the number of occurrences of pump suspension was higher for PID, so there’s an indication that MPC was more effectively using insulin in the presence of pump constraints (including IOB).

Quantitative Evaluation of a Predictive Low-Glucose Management (PLGM) System (83-OR)

Bruce Buckingham, MD (Stanford University, Palo Alto, CA)

Dr. Bruce Buckingham shared results from an in-clinic, overnight study testing the MiniMed 640G predictive low glucose management (PLGM) algorithm’s ability to prevent hypoglycemia – this would have served as the 640G US pivotal study, though Medtronic is of course now leapfrogging the 640G to launch the 670G hybrid closed loop in the US. The 68-person trial induced hypoglycemia (via basal rate increase), and patients wearing the 640G/Enlite 3 were monitored with YSI over a nine-hour period to confirm hypoglycemia and sensor accuracy. The 640G’s predictive suspend algorithm successfully prevented hypoglycemia (two YSI values <65 mg/dl) in 41 of the 68 experiments, a solid 60% prevention rate. The mean duration of pump suspension was 105 minutes, and mean glucose was 83 mg/dl one hour after pump suspension, 129 mg/dl after three hours, and 152 mg/dl after six hours. It was good to see no significant hyperglycemic rebound, a testament to the algorithm’s design – it automatically resumes basal insulin delivery once glucose levels are recovering, a big step up over the 530G/Veo. This encouraging hypoglycemia prevention data parallels what we have previously seen on the 640G’s PLGM algorithm: the SportGuard trial from ATTD 2015 (41% reduction in hypoglycemia events <65 mg/dl) and the PILGRIM study presented at ADA 2013 (!) (80% hypoglycemia prevention). A deeper dive on the 640G can be found in our ATTD 2015 report, and our initial report on its January 2015 international launch.

• Will all pumps have PLGS (at minimum) in a few years? We’d note that Tandem is also developing a PLGS system (before its hypoglycemia-hyperglycemia minimizer), though all other companies are jumping straight to hybrid closed loop. See our AID landscape here.
• As we’ve previously noted, the MiniMed 640G suspends insulin delivery with a 30-minute prediction horizon and resumes insulin delivery once glucose levels recover. The suspend-before-low limit was set at 65 mg/dl in this study. For a “suspend before low” event to occur, both of these must happen: (i) the sensor glucose value is at or within 70 mg/dl above the low limit; and (ii) the sensor glucose value is predicted to reach or fall below a level that is 20 mg/dl above the low limit within approximately 30 minutes.
  
  • Following a mandatory 30-minute suspend time, basal insulin delivery will automatically resume if the following conditions are met: (i) the sensor glucose is at least 20 mg/dl above the low limit; and (ii) the sensor glucose is estimated to be more than 40 mg/dl above the low limit in 30 minutes. The pump will resume basal insulin delivery after a maximum two-hour suspension.

Questions and Answers

Q: Nice study, Bruce. This looks at one of the milder ways you can go hypoglycemic overnight, but another way is that you can overshoot the bolus dose before bed. What does this system do for these more immediate blood glucose decreases?

A: There was actually a study released this month from Australia – it looked at 28 patients given a morning bolus projected to bring them down to 55 mg/dl. 14% of patients became hypoglycemic in the control condition, 4% became hypoglycemic in PLGM. Also, overnight, everyone went hypoglycemic in the control condition, but only 13% did in PLGM.

Q: You were involved in both of these last two studies. There was 60% prevention in this one, 78% didn’t go below 60 mg/dl in the last one. Could you please compare the effectiveness of the two therapies?

A: This study was much bigger. We had 100% effectiveness at Stanford – other centers are not as experienced with these kinds of studies. It is hard to take one set of patients and compare them to another. Dr. Ly presented studies done at home…this was at a research center. I’m impressed with the similarities between the studies, given this fact. Bottom line, this system is fairly safe, it works, no hyperglycemia rebound, no DKA. But it’s not perfect, the amount of insulin on board really determines how effective these systems are.

Q: Has it been assessed in exercise induced hypoglycemia?

A: Not that I’m aware. [Editor’s Note: The PILGRIM study presented at ADA 2013 did test the PLGM algorithm under exercise-induced hypoglycemia; it was a small study in 22 patients, but showed 80% prevention.]

Overnight Closed-Loop (OCL) at Home Compared with Sensor-Augmented Pump with Low-Glucose Suspend (SAP-LGS) Improves Time in Target Range in Adults and Reduces Hypoglycemia in Adolescents (78-OR)

David O’Neal, MD (University of Melbourne, Melbourne, Australia)

Dr. David O’Neal presented data from a small crossover study showing that overnight closed-loop systems offer benefits for both adults and adolescents vs. sensor-augmented pumps. The study randomized 16 adults (baseline A1c = 7.3%) and 12 adolescents (baseline A1c = 7.8%) to either the MiniMed 530G overnight (control) or closed-loop control (intervention – Enlite 2 CGM, Medtronic pump, proprietary PID algorithm) for four nights at home. In adults, overnight closed-loop performance was strong – time in the tight range of 72-144 mg/dl increased from 45% to 58% vs. SAP (p=0.005) and time <72 mg/dl decreased from 1% to 0% (p=0.025) without a concurrent increase in mean glucose (145 mg/dl on closed loop vs. 152 mg/dl on open loop). There were fewer instances of symptomatic hypoglycemia on closed loop (13 events vs. 5 event; p=0.059), and all these benefits carried into the following day as time in range from 8 AM to 4 PM rose from 47% to 53%. In adolescents, the difference between the treatment arms was more nuanced – the main benefit was a reduction in overnight time spent <70 mg/dl from 2% to 0% vs. SAP along with a reduction in instances of symptomatic hypoglycemia (10 events vs. 1 event; p=0.007). Adolescents achieved quite impressive open-loop glucose control (time in range = 64%), leaving little room for time-in-range improvement with automation.

• There’s little question that automated insulin delivery will make a big difference overnight, even in very tightly controlled patients. In fact, we expect many patients doing pretty well on open loop therapy now to adopt early closed-loop systems, even if they only wear them overnight.
• We love too that overnight automation can actually live up to patients’ high expectations for AID systems. Overnight closed loop will really ensure patients wake up at a good glucose on most mornings, and patients’ open-loop control while ASLEEP can’t possibly be better than an automated system. First-gen hybrid closed loops are well suited for nighttime anyways, since they will modulate basal insulin delivery and won’t have to deal with meals or exercise.
• Aside from glucose control, the key overnight design principle is alarms – systems must take care of things in the background and let patients sleep overnight.

Questions and Answers

Q: What do you think was the driving factor for the differences between the populations you studied?

A: I suspect that part of what we’re seeing is the impact of parental oversight of adolescents. Because these patients were moving on sensor-augmented pump from less sophisticated systems, I think adolescents in particular were able to achieve very good control overnight given the accessibility of these new tools (e.g., 64% time in target range in the control group).

Q: The adults and adolescents used different CGMs, didn’t they? Does that impact our interpretation?

A: We found that the MARDs in adults and adolescents were very similar, so we do not think the differences in CGMs played a role.

Predictive Hyperglycemia and Hypoglycemia Minimization: In-Home Evaluation of Safety, Feasibility, and Efficacy in Type 1 Diabetes (82-OR)

Trang Ly, MD (Stanford University, Stanford, CA)

Dr. Trang Ly presented results from a randomized, crossover, at-home overnight study comparing a hypoglycemia-hyperglycemia minimizer algorithm to a predictive low glucose suspend (PLGS) algorithm (Enlite 2, Veo pump, bedside laptop with algorithm). The six-week trial accumulated 641 nights on PLGS and 648 nights with hypoglycemia/hyperglycemia minimization, recruiting 30 patients who were randomized every evening for six weeks to either system. As expected, results showed the hyperglycemia mitigation made an incremental difference – time in range (70-180 mg/dl) increased from 71% to 78% (p<0.001) and mean glucose decreased from 152 mg/dl to 143 mg/dl (p<0.001). The benefit came largely from less time spent >250 mg/dl, which dropped from 4.5% to 1.6% (p<0.001), while average time spent < 70 mg/dl was not significantly different between the groups (1.1% [HHM] vs. 1.0% [PLGS]). Dr. Ly stressed that the group has taken a conservative approach in designing the algorithm, opting for safety in this study with the intention of tweaking the system to get greater glycemic control in the future. Indeed, larger randomized trials and studies in pediatrics are planned for the near-term and we look forward to seeing more data on the added benefits of a hypo/hyper minimizer component vs. just PLGS. In addition to better glycemia, shaving off those extreme overnight highs is highly valuable from a sleep and quality of life perspective too! We’d note that most of the field is skipping straight to commercial products with hypoglycemia/hyperglycemia minimization, with the exception of Tandem, who is pursuing PLGS first

• J&J has previously used the name “Hyperglycemia and Hypoglycemia Minimization,” though this algorithm was not connected to Animas’ work. Still, it shows what can be done by adding even a conservative insulin modulation algorithm to shave off highs >250 mg/dl.

Questions and Answers

Q: Can you contrast this system with the UVA system?

A: We use a more conservative algorithm where the goal was not perfect control. This is very different from UVA’s overnight algorithm that targets a glucose of 120 mg/dl by morning. This algorithm was designed to be more conservative and to essentially provide a little bolus on top of the basal insulin people are getting. We’re still learning and it’s still evolving and we’re tweaking parameters. We’re going to be doing further studies in younger children. This was our first large trial using the full system.

Q: Have you considered a stricter target?

A: We started with a lower target of 120 mg/dl and we were seeing a bit of hypoglycemia so that’s why we raised it. We have to start somewhere, and it’s a constant balance between safety and efficacy.

Oral Presentations: ADA Presidents Oral Session

Closed-Loop Glucagon Administration for the Automated Prevention and Treatment of Hypoglycemia in Type 1 Diabetes (378-OR)

Courtney Balliro, RN, CDE (MGH, Boston, MA)

Ms. Courtney Balliro presented updated, full results from the Bionic Pancreas team’s placebo-controlled, double-blind, glucagon-only study in patients with type 1 diabetes. Time spent in hypoglycemia declined 78% overall and 93% overnight. The study enrolled 22 people with type 1 diabetes, who wore a Bionic Pancreas at home for 14 days that administered glucagon-only on seven of those days vs. placebo-only on seven of those days, interspersed in random order (insulin was self-regulated). Results were just as impressive as those first presented by Dr. Steven Russell at AADE 2015 – time spent in hypoglycemia (<60 mg/dl) was reduced from 5.8% with placebo to 1.3% with glucagon for the entire day (a 78% reduction). The overnight improvement was even more striking – 7.9% time spent <60 mg/dl with placebo vs. 0.5% with glucagon, a 93% improvement! Glucagon administration reduced hypoglycemia exposure (area over the curve and less than 60 mg/dl) by 77% (p<0.001) relative to placebo administration without affecting mean blood glucose (mean: ~154 mg/dl, in both arms). Performance was not at the expense of hyperglycemia, as time spent >180 mg/dl was not significantly different – 28% (glucagon) vs. 29% (placebo) (p=0.7) – nor due to greater insulin utilization – 39 units/day (glucagon) vs. 37 units /day (placebo) (p=0.12).

• Ms. Balliro stressed that the Bionic Pancreas did not deliver excessive glucagon (~0.48 mg/day – on par with previous studies) and noted that there were no differences in self-reported nausea. The team continues to show strong data on this front, countering those arguing that glucagon administration may prove problematic on nausea.
  
  • The other big question is the effects of chronic glucagon exposure, and it will take some time to get an answer. We learned from Dr. Ed Damiano on during ADA that the team will use Zealand’s liquid stable glucagon analog and plans to begin clinical trials with the iLet device in 2H16. As we noted on Day #2, the pivotal studies of the insulin-only iLet are still expected to start in 2Q17. The bihormonal pivotal trial, which will begin after the start of the insulin-only pivotal trial, will require that a subset of the study cohort use the iLet for 12 months in order to gain chronic glucagon exposure data for the FDA. Dr. Damiano has forecasted an end of 2017 insulin-only PMA submission, with a potential PMA supplement to add glucagon in early 2019.
• Ms. Balliro confirmed that the Bionic Pancreas team plans to test its glucagon-only Bionic Pancreas in patients with post-bariatric hypoglycemia and chronic hyperinsulinemia “in the next year or so.” The device could be particularly suited to the former group given that carbohydrate intake was reduced 35% in the glucagon arm. That said, Ms. Balliro stressed that the near-term focus is on bringing the dual-hormone device to the market – no surprise there – but it was terrific to hear that the team thinks a glucagon-only system does have commercial application.
• Glucagon-only study design: Patients wore a Bionic Pancreas at home for 14 days that administered glucagon-only on seven of those days vs. placebo-only on seven of those days, interspersed in random order (insulin was self-regulated). Patients were not restricted in their daily activities and were blinded to which days they were on glucagon vs. placebo, which Ms. Balliro noted was effective: patients were able to distinguish glucagon from placebo treatment on < 50% of days [i.e., less than chance].

Questions and Answers

Q: In the individuals that had partial hypoglycemia unawareness at baseline, did they have restoration of hypoglycemia awareness at the end of the study?

A: We did not look at that, but I’m going to say probably not because the amount of symptomatic hypoglycemia reported did not correlate with actual hypoglycemia.

Q: Can you talk about why you switched back and forth between glucagon and placebo?

A: We didn’t want patients to recognize patterns between hypoglycemia and other adverse effects they experienced. In other studies, they have days in a row so the reason we did this was to make sure it was blinded.

Q: Do you have plans to test the system in other groups?

A: We have plans to test the system in patients with post-bariatric hypoglycemia and chronic hyperinsulinemia in the next year or so. However, the hope is any patient that wants it will be able to get a bihormonal bionic pancreas first – so that’s our priority.

Oral Presentations: Management of Hyperglycemia in the Hospitalized Patient (with State-of-the-Art Lecture)

Sensor-Augmented Pumps vs. Multiple Daily Injections for Achieving Glycemic Goals in Hospitalized Patients with Type 2 Diabetes in China (18-OR)

John Shin, PhD (Medtronic, Northridge, CA)

Dr. John Shin presented results from a Medtronic RCT comparing use of sensor-augmented pump (SAP) therapy to MDI in people with type 2 diabetes hospitalized for high glucose levels. The study enrolled 81 patients who were randomly assigned to use either the Medtronic MiniMed 722 (n=40) or MDI (n=41) to manage their diabetes (baseline A1c = 10.0%). Patients were given general instructions to increase or decrease their insulin bolus and basal dose according to real-time CGM (intervention group) or seven daily fingerstick tests (control), with physicians overseeing the adjustments. Patients were discharged once their glucose levels were between 80-130 mg/dl pre-meal and 80-180 mg/dl post-meal. Patients using SAP had significantly shorter hospital stays (3.7 days vs. 6.3 days, p <0.001), with greater time spent between 70-180 mg/dl (77% vs. 64%, p<0.05) and less time spent < 50 mg/dl (0.04% vs. 0.32%, p<0.05). Apart from some mild bleeding around the sensor insertion site, there were no adverse safety events reported. We would have loved to see patients go home on the different arms and tracked over time – would the pump advantage last outside the inpatient setting?

• The proportion of time spent in relevant ranges almost unanimously improved with SAP. The one exception was time <70 mg/dl, which was not significantly different in the two arms

	SAP	MDI	p-value
≤ 50 mg/dl	0.04%	0.32%	p<0.05
<70 mg/dl	1.00%	0.64%	NS
70-180 mg/dl	77.44%	64.33%	p<0.05
≥ 180 mg/dl	21.56%	35.03%	p<0.05
≥ 250 mg/dl	4.13%	8.53%	p<0.05

Questions and Answers

Q: Why were the patients in the hospital?

A: Glucose management issues.

Q: People would not be hospitalized for this in the US. So this was not really a hospital study, but more of a randomized pump vs. MDI study.

A: Yes. We essentially looked at the current therapy patients were on and saw how we could decrease time spent in a Chinese hospital.

Q: Who was making the adjustments in the algorithm on a day-to-day basis?

A: It was done by a physician.

Q: Upon release, did the patients stay on the SAP?

A: No, they did not go home with the technology. Once they achieved the target, they left without the pump. We haven’t tested what happens after the hospital stay – whether they remain in range or if they have to come back.

Efficacy, Safety, and usability of A Clinical Decision Support System for Basal-Bolus Insulin Therapy in Hospital Routine Care (15-OR)

Katharina Neubauer, BSc, MSc (Medical University of Graz, Austria)

Ms. Katharina Neubauer presented data demonstrating the efficacy, safety, and usability of GlucoTab – a mobile decision support system that provides suggestions for insulin dosing – in non-critically ill type 2 patients (n=92). The software aids nurses by providing a starting basal dose based on a patient’s age, weight, renal function, and insulin sensitivity and recommending subsequent bolus dosing depending on the patient’s food consumption and blood glucose levels throughout the day. Results were solid, especially given the typically inadequate care inpatients with diabetes receive – patients on GlucoTab achieved a mean glucose of 159 mg/dl (baseline: ~220 mg/dl) with 69% of blood glucose measurements in-range (70-180 mg/dl). Notably, the software completely eliminated episodes in which blood glucose dropped below 40 mg/dl and only 2.3% measurements came in <70 mg/dl. Equally importantly, findings suggested that provider’s adherence with advised insulin doses exceeded 91%, implying that there was a relatively high level of trust in the automated system (this is always a key question when talking about handing over decision-making to technology). This confirms the huge potential of meaningful clinical decision support in the hospital setting, which could shorten hospital stays and make healthcare professionals’ jobs easier. This was not a randomized controlled study, so there could be some study effect, but it’s clear that digital insulin titration is a heck of a lot better than what’s happening in hospitals right now. We hope to see more work on hospital insulin management, particularly in automating insulin delivery with the technology that is already available.

• GlucoTab aids nurses by providing a starting basal dose based on a patient’s age, weight, renal function, and insulin sensitivity and recommending subsequent bolus dosing depending on the patient’s food consumption and blood glucose levels throughout the day. The process is repeated the following morning, with the new basal dose informed by the objective experience of the patient the previous day – i.e., if there were many hypoglycemic episodes, then the basal dose of insulin would be decreased. This study enrolled 92 hospitalized patients (40 female, mean age = 70 yrs), most of whom were already on insulin.
• We noticed one unexpected result – pre-lunch blood glucose values were elevated compared with other pre-meal blood glucose levels. We wonder whether the algorithm could be tweaked to address this concern or whether this was an artifact of the study design in some way.

Questions and Answers

Q: Does GlucoTab account for the number of carbohydrates that the patient eats?

A: No, it only considers if the patient is eating or not.

Q: Do you make adjustments for steroids?

A: Not many patients in our studies were on steroids. It was not in the exclusion criteria, but we have an ongoing clinical trial for patients treated with steroids.

JDRF/NIH Closed-Loop Research Meeting

Pivotal Study Design

Steven Russell, MD, PhD (MGH, Boston, MA)

Dr. Steven Russell summarized a paper he co-authored with Dr. Roy Beck, soon to be published in Diabetes Care. The table below highlights their thought on pivotal study design for artificial pancreas systems, building off the question we asked last year, “What is the appropriate control group for an artificial pancreas pivotal study?” Below the table is further commentary on this very tricky topic.

• Pivotal AP studies have several goals besides regulatory approval: advantageous labeling, reimbursement by payers, prescribing by practitioners, and adoption by patients. We loved hearing Dr. Russell emphasize the payer point upfront: “There’s approval, and then there’s approval. The design of the trial itself can make a very big difference in how persuasive the case is to payers.” This is where the study design considerations (see below) might really drive payer decisions about this technology. For instance, if a pivotal trial only enrolls sensor-augmented pump users, will payers only offer closed-loop to this group of patients?
• Sensor-augmented pump therapy is a scientifically valid control group (the best therapy available), but it is also a small slice of type 1 diabetes. “We want to democratize good glycemic control,” said Dr. Russell. Those already using a pump and CGM “are atypical, self-selected patients that may not be representative” of the broader population. Drs. Russell and Beck recommend that studies enroll those on MDI and pumps, with few exclusions (i.e., representative of the candidate population for artificial pancreas systems).
  
  • “Sensor-augmented pumps without automation will be obsolete soon.” This is another reason SAP may not make sense as the control group – what a good point! It will be interesting to see what the pump field looks like in five or ten years; will the vast majority of SAP users be on automated insulin delivery? Will it take much longer than expected for patients to move to automation? Will some SAP users never convert?
• “Time-in-range doesn’t tell you what’s happening with time outside range. We argue that A1c and time in hypoglycemia are the best outcome measures.” This is consistent with the co-primary outcomes the Bionic Pancreas team has always reported (mean glucose and time <60 mg/dl), and we wonder if other companies and investigators will pursue this going forward.
  
  • For comparison, Medtronic’s MiniMed 670G pivotal was a single-arm safety study comparing baseline control to three months, and it was not actually statistically powered to show a reduction in A1c. As we noted on Day #2, however, it did show a statistically significant 0.5% reduction. CGM endpoints were secondary: time-in-range (% of SG in 71-180 mg/dl) improved modestly from 67% during the baseline run-in to 72% during the study, with time <70 mg/dl nearly halved (6% to 3%), time <50 mg/dl declining 40% (1% to 0.6%), and time >180 improving moderately (27% to 25%).
• The FDA wants pivotal trials to include all ages, rather than limiting exposure to adults. “They know that this is likely to be prescribed off-label.” This echoes early June commentary from FDA’s Dr. Courtney Lias during her live Q&A webinar on the artificial pancreas.
• The FDA wants pivotal studies to be at least three month long, meaning a parallel group design makes sense to keep the study short.
• “You need range of A1c’s to make sure the system works well.” This has been a historic challenge in automated insulin delivery trials, and we hope investigators can really enroll a wide spectrum of patients going forward. Those most likely to benefit from these systems – high A1c patients, severe hypoglycemia – need to be included in studies!
• To encourage retention in the usual care group, the Bionic Pancreas bi-hormonal pivotal will include an “incentive study” – those randomized to usual care will get three months on the Bionic Pancreas at the end of the trial.

Pivotal Study Design Consideration	Recommendation	Alternative	Comment
RCT Type	Parallel	Crossover	Crossover design requires long washout
Study Population	Representative of population, patients who use MDI and CSII, with few exclusions. Range of A1c’s.	Adults-only, high-risk patients excluded	Given the potential for off-label use, the FDA may not approve if the device is not demonstrated to be safe in a broad population and payers may limit coverage to only the population that was studied
Randomization (artificial pancreas: Control)	2:1	1:1	2:1 randomization provides greater exposure to artificial pancreas, 1:1 randomization will require a smaller sample size or give greater power for same sample size if equal variance.
Control group	Usual care	SAP	Both scientifically valid, usual care has numerous pragmatic advantages
Superiority vs. non-inferiority	Superiority	Non-inferiority	Non-inferiority may be sufficient for approval but is not likely to drive reimbursement and adoption
Run-in period	Blinded CGM monitoring	Unblinded CGM (SAP training)	Unblinded run-in must be sufficient to achieve competency for SAP trial enrolling non-SAP users
Duration	6-12 months	3 months	3 months min for A1c; longer duration shows continuation of use and durability of effect
Primary outcome(s)	A1c, time <60 mg/dl	A1c only	A1c does not capture hypoglycemia; CGM more reliable and quantitative than participant recall

Outcome Measures for Artificial Pancreas Clinical Trials: A Consensus Report

David Maahs, MD (Barbara Davis Center, Aurora, CO)

Dr. David Maahs also summarized a soon-to-be-published Diabetes Care paper focused on standardizing a short set of basic, easily interpreted outcomes in artificial pancreas studies. The paper has 24 authors, all of whom are leading thinkers in the field. The goal is to facilitate interpretation and basic comparison between studies, and more importantly, to accelerate adoption of artificial pancreas technologies via regulators, HCPs, payers, and patients.

Outcome Metrics	Comment
A1c Mean CGM glucose	If intervention >3 months)
% time in 70-140 mg/dl % time in 70-180 mg/dl	All CGM measures should be reported for the overall 24-hour period, and also stratified by daytime and nighttime period. The time midnight to 6am is proposed as a nighttime definition.
% time <50, <60, <70 mg/dl
% time >180, >250, >300
Standard Deviation Coefficient of Variation	SD much more dependent on mean than CV
Severe hypoglycemia	As defined by ADA (adults) and ISPAD (children/adolescents)
DKA events	ADA definition
% of time closed loop active
TDD of insulin TDD of glucagon or other hormones
Other considerations # of symptomatic hypoglycemia events per week CGM calibration, MARD Study design and stratification into relevant subgroups ITT analysis Report medians (quartiles) instead of mean if not normally distributed

Driving Access to Artificial Pancreas Systems

Amanda Bartelme (Director, Avalere Health, Washington, DC)

Ms. Amanda Bartelme provided a reimbursement overview of automated insulin delivery, highlighting some of the key challenges: too much focus on A1c, hard-to-predict contracting negotiations, data needs that differ from FDA approval requirements, and more. JDRF is already talking to payers about AID reimbursement, though there was no commentary on what is being learned from those conversations. From what we can tell, Medtronic will pursue the existing reimbursement channels for the MiniMed 670G (i.e., sensor-augmented pump reimbursed via DME), though this field seems ripe for a new business model (e.g., AID for $75 a month). Ms. Bartelme cautioned that “if a payer thinks every type 1 patient wants to go on this tomorrow, that’s huge dollar signs and huge panic.” It served as a reminder that nothing is a given with payers in this environment – even devices that reduce A1c, hypoglycemia, and patient burden. How can industry show a positive short-term return-on-investment for AID?

• Diabetes therapies and technologies cannot be fully assessed by A1c alone.” Ms. Bartelme provided a value graphic showing the “current” and “desired” states, noting that payers and FDA (drug division) both have a ways to go before they accept outcomes beyond A1c. This ADA had much more discussion on the limitations of A1c, but it will take time for this to trickle up to decision makers. As a reminder, JDRF has an ongoing health policy initiative to define the key outcomes beyond A1c, and an FDA meeting will occur on August 29 on this very topic.

Current State	Desired State
Regulators and payers have a singular focus on A1c Regulators do not always allow other outcomes on product labeling Achieving payer coverage for diabetes technologies that improve other outcomes, but do not reduce A1c is challenging	Expanded set of diabetes outcome measures that reach beyond A1c, and that better reflect the impact of emerging therapies on people with diabetes Regulator and payer acceptance of a full range of diabetes outcomes

• Increasingly, payers require more and different data than are necessary to receive FDA approval. Most of all, payers look at whether something is worth paying for –is it safe and will it improve outcomes? What does the label look like? How was the study designed? Ms. Bartelme highlighted that the timing of data collection is important to ensure payer coverage post FDA approval – yet another reason why pivotal studies should be designed with reimbursement in mind.
• The payer contracting process brings uncertainty and can be a gating factor to access. Sometimes payers limit coverage to the FDA label, sometimes they expand beyond it, and sometimes they narrow it. Ms. Bartelme noted that “creative contracting” can support post-market data collection, patient affordability, and improved access. We wonder how that specifically played a role in the UHC/Medtronic partnership announced a few weeks prior to ADA.
  
  • “Fantastic data might make more challenging contract discussions. If a payer thinks every type 1 patient wants to go on this tomorrow, that’s huge dollar signs and huge panic.” Whoa – we hadn’t thought about that, but it’s a good point, and one we’re not sure companies can control. It serves as a reminder that nothing is a given with payers, and the key may be showing that short-term return-on-investment is very positive with AID.
  • Increasingly, payers are willing to entertain value-based contracts in addition to traditional fee for service arrangements. Will subscription approaches to pump therapy and automated insulin delivery become more common? If so, companies like Insulet and Bigfoot would seem to have an advantage, as their hardware has lower upfront costs. Would Medtronic be able to change its DME business model (~$5,000 upfront), or would their pump hardware need to be overhauled? What will Tandem, Animas, and Roche do?
• “FDA approval is necessary but not sufficient for payers to provide coverage. To be considered for coverage, a new therapy or tech must be reasonable and necessary, which is: determined by clinical evidence that shows an improvement in net health outcome; determined by proving that is non-inferior to or beneficial over existing alternatives; increasingly determined by showing improvement is attainable in a real-world setting.” That latter point is the trickiest in our view – clinical trials are almost always not real-world, particularly because the control group does so much better (and thus, underestimates the incremental impact of new technology).
  
  • A technology or therapy may also only be covered when certain other criteria are met: age, not achieving desired outcomes on traditional therapies. Both of these continue to make CGM reimbursement challenging, and we wonder if automated insulin delivery will make the process easier or harder.

Psychosocial Outcome Assessment

Korey Hood, PhD (Stanford University, Stanford, CA)

Stanford’s Dr. Korey Hood revealed that by fall 2016, a full set of validated questionnaires will be available to assess the psychosocial impact of automated insulin delivery. Fantastic news! The team has done 60 focus groups, 89 individual interviews, and engaged 400 participants ranging from no experience with closed loop technology to OpenAPS users. Some of the thematic areas include: burden, concerns, features, financial aspects, benefits, context, human vs. system, nighttime, quality of life, and relationships. Dr. Hood shared some example questions, such as, “I believe that using an AID system will... help me worry less about diabetes...” The assessment will also ask about benefits and tradeoffs to using systems – “If the AID system improves my A1c, I will put with ... meal announcement, carb counting, etc.” We love the holistic, behavioral view of these systems, which will be just as important as the impact on glucose. Enormous thanks to the Helmsley Charitable Trust for funding this major project!

Brief Summary of AP Highlights from the Past Year

Vincent Crabtree, PhD (JDRF, New York, NY); John Lum, MS (Jaeb Center for Health Research, Tampa, FL)

Dr. Vincent Crabtree and Mr. John Lum gave a broad overview of JDRF and NIH-funded artificial pancreas highlights from the past year, offering several updates we had not heard previously.

• Dr. Roman Hovorka’s NIH-funded DAN05 study will investigate closed loop therapy in 6-18 year-olds over 12 months. The study will randomize patients on pumps to either automated insulin delivery (n=65; MiniMed 640G/Enlite 3 + the Cambridge algorithm running on an Android phone) or standard pump therapy (n=65). It’s not clear when it will start. It’s outstanding to see this study getting off the ground after it received ~$6.4 million of the UC4 grant for major artificial pancreas trials (announced at DTM 2015). We’re not sure if Dr. Hovorka would apply for regulatory approval after this study, though it is notable that he is now using Medtronic hardware to do his research.
  
  • The planned CLOuD study will test automated insulin delivery in new onset type 1 diabetes over two years. The study will have 96 participants randomized to either closed loop or usual care, and the goal is to show an impact on C-peptide – wow! As a reminder, Dr. Bruce Buckingham and colleagues did a similar study, though the time on closed loop was just a brief period at diagnosis, followed by sensor-augmented pump therapy. Hopefully this study can show this technology makes a difference in preserving beta cell function if it is added right at diagnosis, which could be transformative for guidelines and treatment approaches. 
  • Dr. Hovorka also has three 24/7 home studies underway, continuing the team’s trailblazing work of longer-term, outpatient studies. We assume we’ll see some of this data at EASD 2016 or ATTD 2017.
• Dr. Crabtree is “hopeful” that the upcoming International Diabetes Closed Loop study (n=240) could be used to support regulatory approval. This is consistent with what TypeZero CEO Chad Rogers told us earlier this year. We’ve learned that the TypeZero control algorithm will reside on an Android phone, and the study intends to use insulin pumps from more than one manufacturer (Cellnovo has signed on thus far).
• Dr. Tim Jones will conduct an independent, six-month RCT of the MiniMed 670G hybrid closed loop (n=80), starting in 3Q16. The trial will randomize patients to standard therapy (both CSII and MDI) or use of the hybrid closed loop. An ongoing study has been examining the MiniMed 640G over six months of use, with completion expected in November. We love the idea of independent academic investigators validating commercial systems – this can grow the body of evidence around these systems, but do so impartially (helpful for payers).
• SGLT-2 inhibitors are now being studied in conjunction with closed loop at Yale (Dr. Jennifer Sherr) and Montreal (Dr. Ahmad Haidar). This could help cut the need for manual pre-meal boluses and make systems more fully automated.
• Plenty of work is happening to advance algorithms further, including fault detection (infusion set issues; CGM errors); coping with exercise; and making algorithms more adaptive. All clearly have a role in making this technology safer and enabling tighter control and lower glucose targets. 

Opening Remarks

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

Dr. Aaron Kowalski gave brief introductory remarks before handing it over to JDRF’s Dr. Vincent Crabtree and Jaeb’s Mr. John Lum. Dr. Kowalski highlighted the completed MiniMed 670G pivotal study, which reported earlier in the day with strong outcomes: a 0.5% reduction in A1c from a low baseline (7.4%); time <70 mg/dl declined 44% (6% to 3%); and time <50 mg/dl declined 40% (1% to 0.6%). “I cannot believe we are on the cusp; we are right around the corner from artificial pancreas. We are now talking about broader issues: special populations, reimbursement, etc. These are incredible times. I’m really grateful.” It’s been an amazing journey since JDRF founded its artificial pancreas program ten years ago, and we cannot wait to see how the 670G fares once it hits the market as the first commercial closed-loop product (FDA submission by the end of this month).

Panel Discussion

Dr. Aaron Kowalski: Steve [Dr. Russell] captured the best bullet I’ve seen in a long time. Sensor-augmented pump therapy without automation will soon be obsolete. It’s so cool to see that on paper.

Dr. Rich Bergenstal: I love the standardization for outcomes of artificial pancreas studies. My request is that it be branded not just metrics for artificial pancreas, but general metrics. To have one definition for artificial pancreas community, and then have “this long-acting insulin vs. this insulin” reported a different hypoglycemia is no good. These don’t sound like artificial pancreas metrics, this sounds like broader metrics. If I’m in clinic and wanting to know if I’m improving, I could use these too.

Dr. Maahs: We had that in the paper – studies of all type 1.

Dr. Roy Beck: You did not include LBGI or AUC. I actually like the ones you picked, but those seem like they might be better – they capture not just time but amount?

Dr. Maahs: We should consider this paper a first step. It was something we could agree on, a basic set of measures. Drs. Hovorka, Kovatchev, and Weinzimer wrote a paper with more details and more math. There is lots of room to do more.

Dr. Beck: We should be cautious about using the standard metrics to compare across studies, particularly those without a control group. The best way to compare across studies is to determine the within study intervention group versus control group difference and then to compare this difference among studies.  Even then there are potential problems since study designs vary, clinics are different, and eligibility criteria may be different. 

Dr. Peter Chase: We need some help with patient selection. We have these high tech families with low A1c’s. How do you go about getting a representative population?

Dr. Hood: Great question. In the pilot projects, we often don’t get a representative sample. The larger project is developing measures that can assess some of these aspects in CGM naïve, pump naïve, and system naïve users. The data that we have, data from participants is the first wave of information. We might have double and triple the barriers and problems when access is broadened. It’s likely that once you get beyond this early adopter population, we’re going to face  the same barriers to uptake. We are working to design interventions so that once we get beyond that first wave, we can onboard people.

Dr. Russell: It may be the only way to get more representative population is to achieve geographic diversity. We selected 16 sites with a geographical consortium, not just eastern and western seaboard. Clinics that had different populations and different socioeconomic data.

Dr. Beck: And higher A1c’s. Roman will use 25% above 8% or 8.5% to force that issue with a quota. Then, you fill the quota of under 8.5%. Boris will have half the patients with higher A1c’s.

Dr. Roman Hovorka: What happens at the end of the trial in those recruited from MDI to closed loop. Unless we are a big company, they cannot continue using an investigational device for treatment. What does the panel suggest?

Dr. Helen Murphy: In cancer trials, there is a cancer drugs fund that is made available to participants. Maybe Aaron and JDRF can start advocacy for artificial pancreas funding. [Editor’s Note: Dr. Murphy also gave a presentation on closed-loop in pregnancy, but it included confidential material, and we have not covered it in this report.]

Dr. Hovorka: I asked in the UK, and you cannot actually continue using an investigational device in the UK.

Dr. Maahs: In the MiniMed 670G trial, many adults were allowed to continue devices, though that is in the US.

Adam Brown (Close Concerns, San Francisco, CA): One of my biggest worries is we will undersell the benefit of this technology, because of the patients being enrolled in these studies. In the MiniMed 670G pivotal, they had a baseline A1c of 7.4% and spent 67% time-in-range during the run-in. How can we get a broader patient group? What about those with severe hypoglycemia? Those with really high A1c’s?

Ms. Bartelme: We can’t let perfect be the enemy of the good. Taking people in really poor control, and showing changes is not easy. There’s also the mental burden that is harder for people to wrap their head around. We have to educate payers on this front.

Dr. Beck: In the study today presented on CGM in MDI users, their mean time-in-range 70-180 at baseline was about 45%. I think that is pretty reflective of what the average probably is.

Dr. Beck: What is the ideal study design and potential to get payers to agree on what they might expect for the magnitude of effect?

Ms. Bartelme: We need a targeted approach: find a couple of high functioning payers that are more innovative. I think there is some value there. But you don’t want to have conversations with payers who say, “These are the 14 endpoints we want to see.” Typically six months to a year before FDA approval is a good time to start talking to payers. With more innovative payers, you can do progressive things and partner earlier on in the process. There is a desire among payers to do risk-based contracting and value-based payments. There is lots of talk about that in the drug world. Once you get people in the real world, the data is very difficult to collect. Fortunately, artificial pancreas technologies are collecting data with patients all the time.

Brandon Arbiter (Tidepool, San Francisco, CA): Across board, we hear that people would gladly take the device home; that was not my experience in the study. The user experience around alarms was so illogical and awful that I withdrew. But my question is that I realized i wasn’t in an end of study focus groups. Is there a process to collect feedback from those who drop out?

Dr. Hood: We can do the focus group later if you would like [Laughter]. You raise a great question. Tomorrow I will be talking about responders vs. non-responders to a treatment. There is an attribution, often false, that you didn’t respond to a treatment because of some set of reasons that didn’t have to do with the intervention. That means when an intervention works, it’s because of the intervention; when it fails, it’s because of the person. We don’t do enough about that. You offer a valuable set of experiences. Many times, when we get beyond two and four weeks and do an evaluation in six months, we got lots of good info on high vs. low users. We haven’t tapped into that enough.

Dr. Russell: We can learn from these reasons – we’ve only had one person voluntarily drop out of one of our studies. That person had spectacular blood glucose control with no hypoglycemia. But she felt very uncomfortable at that mean glucose. She was used to running very, very high, and actually felt low at a mean of 110 mg/dl. That’s one of the reasons why there is option to raise the target and get a higher mean glucose if you don’t want to achieve the lowest mean. Maybe over time, a person would become more comfortable. That wasn’t a problem I would have anticipated.

Joint ADA/JDRF Symposium: Optimizing Use of Technology and Therapeutics in Pediatric Diabetes

Automated Insulin Delivery and Bihormonal Artificial Pancreas in Pediatrics―Coming Soon!

Trang Ly, MD (Stanford University, Stanford, CA)

Stanford’s Dr. Trang Ly provided a whirlwind tour of automated insulin delivery systems in development, highlighting the now-complete MiniMed 670G pivotal study (“outstanding results,” particularly in adolescents and in well-controlled patients; see here); UVA’s DiAs system (pivotal trial starting later in 2016, but final device not decided on), Beta Bionics (works in insulin-only mode; big wins: adjustable target, qualitative meal boluses, few alarms; insulin-only pivotal starting in 2Q17, followed by the bihormonal pivotal), Tandem’s predictive low glucose suspend system (pivotal trial later this year), Animas’ hypoglycemia hyperglycemia minimizer (pivotal trial later this year), Bigfoot Biomedical (smartphone app to interface with system and bolus), Cambridge (“amazing results,” but no commercial partner), Diabeloop with Cellnovo’s pump and Dexcom CGM (pivotal trial later this year, apply for CE Mark in 2017), Insulet’s artificial pancreas system (first ever picture we’ve seen, human trials starting later in 2016), and Inreda’s bihormonal system in Europe (studies ongoing, CE Mark in 2017) WOW! We had not previously heard these timelines for Diabeloop or Inreda. Dr. Ly also highlighted that the next-gen Medtronic pump, the MiniMed 690G, is already in development and will include the DreaMed algorithm to perform automatic correction boluses based on CGM readings (entering a trial this month, according to Medtronic’s Saturday dinner).

At a higher level, Dr. Ly called for customizable glucose targets that allow patients to titrate algorithms’ aggressiveness. While this isn’t available in the first-gen 670G, it could be available in other systems (e.g., Beta Bionics’ iLet), and hopefully it will be widespread in next-gen products (of course, there is tough balance to strike here between complexity and customizability). Dr. Ly summarized that closed loop represents “a new paradigm in diabetes management,” and “these therapies will be transformative for pediatric diabetes care.” She emphasized that first-generation systems are still going to be conservative, still require maintenance, and will not be a cure – in short, “managing expectations” is critical, particularly with the first-to-market 670G. We couldn’t agree more on that point!

• Dr. Ly noted called the 670G pivotal trial data “outstanding,” and was particularly impressed with the adolescent results (see our coverage here). She noted that the adolescent patients were already doing well at baseline (A1c: 7.7%), and still saw an “incredible” 0.6% reduction in A1c AND less hypoglycemia.
  
  • Managing expectations will be the big challenge with the 670G. Dr. Ly noted that the glucose control “is not perfect,” and in this first-gen hybrid closed loop, the algorithm has to be conservative because it needs to be safe. She explained that the 670G will work great for patients who are already testing and doing pre-meal boluses, but it’s not going to do everything – in other words, it’s not a device you turn on, do nothing, and get an A1c of 7% with no hypoglycemia. Still, she called it “incredibly promising data” and is clearly psyched about its potential.
  • “Dr. Ly, thank you for giving me back the son I had before he was diagnosed with diabetes,” shared one mom of a 670G pivotal trial participant. This echoes what we have heard for a long time about automated insulin delivery – it frees up mind space and brings awesome quality of life benefits. We hope those are captured as the FDA weighs risk-benefit, and as payers weigh reimbursement. As a reminder, 80% of 670G pivotal study patients are still using this device as part of the FDA’s continued access program, a strong vote of confidence in the benefits vs. burden balance.
• Dr. Ly showed the first picture we’ve ever seen of Insulet’s artificial pancreas system, which included a slim handheld controller, an MPC algorithm built directly into the pod, and Dexcom CGM. Clinical studies will start in 2016, consistent with expectations, and we assume the plan to be in market in 2018 still stands. Insulet management had previously hoped the system might use a phone, but obviously a handheld is likely a faster and easier development path – we assume a phone app could still show data and perhaps add remote bolusing over time. We’re glad to see the handheld will be pretty slim and small, an improvement from the fairly clunky PDM that has not changed much since the OmniPod originally launched. As a reminder, the MPC algorithm is a commercial version of the UCSB algorithm licensed from Mode AGC in 4Q15.

Questions and Answers

Q: If you’re looking for A1c improvement, why not try this in A1c’s of 12-14%? 

A: That will be the next challenge. You have to think about why a patient has a 12-14% A1c. If they are doing it because they are not testing and not bolusing, closed loop is not going to fix that. Closed loop will ideally be the new normal like pump therapy should be the new normal. But we’re not going to fix a 14% A1c that doesn’t want to take insulin because he or she is trying to lose weight.

Q: People don’t want to push buttons. Won’t this make a huge difference now that patients don’t have to push the buttons?

A: In the first-gen you will have to push buttons. You need to calibrate the sensor, for instance, otherwise you won’t stay in closed loop. There is still a lot of work to be done on the patient side.

Q: What about pump failures, failure to insert, kinking, etc.?

A: It depends what you mean by pump failures. I think you mean infusion set failures. There are lots of different companies developing better infusion sets. From our experience with the 670G, it has better detection for that occlusion. I think that is still a work in progress. We’ve done some work at Stanford, UCSB, and RPI to develop closed loop algorithms to better detect infusion set failures. It’s tricky because it’s hard to tell. There is a balance between false alarms and true alarms. It’s really hard to develop those algorithms to better detect that, and we need to make them functionally better. Companies are working on that and studies are happening.

Q: In the first gen, it’s amazing that system can be way better than control group with a target range of 70-180. Can you imagine the next gen will be even better range, say a range of 70-130?

A: I think we need to get it out, and after the first-gen, allow patients to modify the set point, so it’s the right percentage of time-in-range for them. That’s what systems need to have. When patients first test it, they say it’s great and it works really well. But when we started to test it out for longer, say three or six months, patient say, “Well, actually, sometimes I want the algorithm to be more aggressive and I want to be a bit lower.” So it needs to be adjustable for patient needs.

Q: We have a lot of Medicaid and managed care patients; just getting CGM is difficult. Where do you see closed loop? Will it be harder or easier? Looking at the data, with a reduction of 0.6%, will insurance companies say that is not relevant?

A It’s going to take long conversations, lots of work, patient advocacy groups, and Aaron might be able to talk more about this. We need to get it out, and I think it will actually help improve things once it becomes standard of care, potentially all systems will need to have automated closed-loop control to remain competitive on the market. But someone is going to have to pay for it.

Q: My concern when CGM first came out is that all insurance companies were paying for it. But two years down the line, all the sudden they’re saying we’re not paying for it.

Dr. Aaron Kowalski (JDRF): This is a top priority at JDRF. We have started a large health policy initiative with the Helmsley Charitable Trust. It doesn’t do any good if we develop new technology or therapies that are not accessible. The payer feedback has been pretty positive, and we are engaging with the major players. We want to make sure these are accessible for anybody who would benefit.

Status of Insulin Pump and Continuous Glucose Monitoring Use in Pediatric Diabetes

Jenise Wong, MD, PhD (UCSF, San Francisco, CA)

Dr. Jenise Wong provided a comprehensive overview of device utilization in pediatric type 1 diabetes. Beginning with pumps, she noted that there are a host of studies documenting the glycemic and quality-of-life benefits in children and adolescent but noted that use remains very variable by country. Dr. Wong attributed the variability to a number of factors from reimbursement and guidelines around pump use to the availability of pump reps and supplies – she cited T1D Exchange data suggesting that pump penetration is as high as 60% in the US, more than three or four times the penetration in EU countries. She contrasted that level of penetration with CGM, noting that the penetration of this technology remains VERY sparse (~15% in the Exchange). She was very positive on the benefits of sensor technology, though noted that a significant discontinuation rate prevents serious uptake. We do think that part of this has to do with patients’ sky-high expectations for CGM technology and the reality that the devices – while improving in accuracy and usability – are still not perfect, particularly on day one. She noted that even among CGM users, few patients take full advantage of the technology since such a minority of patients routinely review device data (see below) – we think this is because device data does not offer nearly enough benefits relative to the hassle of obtaining it and making sense of it. Once powerful, automated pattern recognition and compelling insulin titration algorithms are in place, we think device downloading will improve, or perhaps move to a real-time notification model. Ultimately, she suggested that education on the importance of device data (both on a provider and patient level) is a key step toward greater penetration in the future. We agree with her, though believe that CGM reimbursement, factory calibration, on-body form factor, a BGM replacement claim, and clinical decision support are more important for expanding uptake.

• Dr. Wong presented valuable data on the lack of device data utilization from the T1D Exchange. Unsurprisingly, utilization is lowest among young adults with higher use among children and adolescents (where we assume parents are very involved with care). This does not surprise us one bit – CGM download software does not yet provide enough benefits to make the hassle of obtaining it and interpreting it worth it. Further, we believe the retrospective data mindset must change to a real-time model that gives patients more in-the-moment pattern recognition. What is more useful – a “morning high” pattern alert that is triggered every few months someone downloads, or a real-time notification at 10 am (“High pattern after breakfast observed for the past six days”)? In our view, personal CGM is a real-time technology, and patients stand to benefit the most from real-time, in-the-moment data analysis.

CGM Download Frequency	Never	1-3x per month	1x per week
Children	24%	37%	8%
Adolescents	36%	22%	12%
Young Adults	45%	16%	4%
Adults	42%	17%	5%

• Dr. Wong also shared broader data comparing data downloading frequency (BGMs + CGMs) among adult patients and caregivers. We assume these numbers are self-reported (meaning they are probably overestimates), and they still fall woefully short of where we would like them to be. This will unquestionably improve as devices become connected to the cloud and stream data automatically, such as Dexcom’s G5, Abbott’s LibreLink, Medtronic’s MiniMed Connect, and a host of cellular- and Bluetooth-enabled meters. However, it still begs the philosophical question – is the future of diabetes data in real-time decision support and automatic pattern recognition? Is it realistic to ever expect patients to download and review their historical data?

Device Download Frequency	Never	Sometimes	Routinely
Adults	69%	20%	12%
Caregivers	44%	40%	27%

Questions and Answers

Q: What is surprising to me is the lack of penetration of pumps in pediatric patients considering the low discontinuation rate. That must mean we’re not getting people onto pumps in the first place?

A: Yes, that’s the first barrier – getting people onto pumps. We talked about the 60% on a pump; what we didn’t talk about is what is keeping those other 40% from starting. There are social factors to be sure - racial disparities, socioeconomic disparities. There are very real ways we could change the system and address cultural barriers that would help. So you’re right that discontinuation isn’t as much of a factor in pump use. The factor is getting people to start.

Q: Is discomfort a limiting factor with CGM? I’ve heard with Libre in the EU that skin reactions stops some kids from using the device.

A: Skin reactions are a common reason for discontinuation for all CGMs. I don’t know if the solution is better troubleshooting from an educator and clinician standpoint or a device innovation that needs to happen. However, it is a common cause of discontinuation and people who are using it will still complain.

DiabetesMine D-Data Exchange

The standing-room-only DiabetesMine D-Data Exchange gathered some of the best minds in diabetes technology, headlined by: a vision for interoperable, component AP systems (FDA’s Dr. Courtney Lias); a standing ovation for OpenAPS developer Mark Wilson; a masterful overview of different closed-loop systems from Stanford’s Dr. Trang Ly; a strong desire for customizable closed loop algorithm glucose targets; and more need for industry to engage with and learn from the DIY community. See a couple themes immediately below, followed by full talk write-ups.

Themes

• There was consensus that closed-loop devices should: (i) include algorithms with customizable glucose targets; (ii) insulin-on board on the home screen; and (iii) have highly adjustable alarm settings. The patient panel was clear that customizing an algorithm’s aggressiveness is very key – some patients want more control, particularly because early-generation systems are more conservative. Indeed, Dr. Trang Ly pointed to this as an area for improvement in the Medtronic MiniMed 670G, which targets 120 mg/dl and does not allow the user to lower the target. Of course, there is a tough balance between customizability and simplicity – tweaking every parameter might be ideal for early adopters, but will add too much complexity that could hinder adoption. Dr. Ly felt strongly that systems should include insulin-on-board, a sentiment we agree with.
• The DIY community is dying to interact with industry, though companies are still resistant. OpenAPS co-creator Dana Lewis indicated that industry’s regulatory concerns are a “phantom worry.” Dr. Courtney Lias urged the DIY community to talk to the FDA. Ms. Lewis argued that companies can still talk to and learn from the OpenAPS community, even if it is not an “approved” product – engaging does not imply support or anything illegal. Dr. Lias said the FDA “comes from a place of understanding,” but OpenAPS technically falls under FDA jurisdiction. The agency has the option of using enforcement discretion and will enforce based on risk – FDA may not choose to spend resources on someone building a system for their own personal use. However, the Agency does want a level playing field: “There is no difference between a movement developing a platform, and a small company developing a platform. We are definitely open to talking about the community, and how we cover the responsibility piece and make sure adverse events are reported...I would push the patient community to start solving the responsibility piece, even if it’s not the current FDA path, we are open to talking about that. There are lots of ways to meet the FDA requirements...Vendors are skittish because of the area of responsibility....We can solve problems by talking together, not by assuming FDA will or won’t do something. I would invite people to talk with me about it, and encourage bigger discussion between this community and FDA – how do we get you what you want, and how do we get what we need?”

FDA on Interoperability & Artificial Pancreas Progress: Where Guidance Can Take Us

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

FDA’s Dr. Courtney Lias shared a strikingly optimistic goal to build an infrastructure for interoperable, modular, component artificial pancreas systems. “We don’t see a way artificial pancreas can be what it needs to be without this. I’m sharing our intention of solving this problem.” As she noted at last week’s AP Webinar, the current regulatory paradigms are easier for single companies submitting a combined pump/CGM/algorithm system (e.g., Medtronic’s MiniMed 670G). Showing a “FACE PALM” slide, she noted that this “system” framework could hamper innovation and product iteration. Dr. Lias highlighted Dexcom’s pump partnerships with Animas and Tandem as two examples, which were hampered by months of legal contracts and ironing who is responsible when things go wrong – a very inefficient process. Dr. Lias envisioned a day with standardized, interoperable devices, allowing patients to swap in different system components (pumps, CGMs, and algorithms; see picture below). She outlined a list of current challenges that FDA and the scientific community/industry must address (see table below), noting this might take ~10-15 years to totally solve though certain pieces could be accomplished in the short term. [Many in the audience thought it could be faster.] It was clear that device communication standards are mission critical in all of this, leading us to wonder how Dr. Joe Cafazzo’s and Melanie Yeung’s work in Toronto on diabetes device standards could be integrated into commercial products. Dr. Lias concluded that FDA wants to be the “lead penguin” on artificial pancreas component interoperability, which might even include forward-looking guidance to pave an easier path for interoperable products. Nice! We were highly impressed with her open and forward-looking perspective, and wonder if industry will consider more modular approaches to AP system design. Plug and play component AP systems could significantly enhance patient choice, spur innovation, and improve automated insulin delivery far more quickly than the current paradigm. This will be a further topic of discussion at the FDA/NIH Artificial Pancreas workshop in Washington, DC on July 6-7.

• “Mobile phones are a must. We’re very pro putting artificial pancreas systems on mobile phones.” The strikingly positive commentary was great to hear, since this has been an open question mark – what are the risks of patients dosing insulin from a phone? The devil is of course in the details here: will the algorithm sit on the phone and communicate to the pump and CGM separately, or will the phone simply act as a window to what’s happening in the fully integrated pump? We were definitely encouraged to hear optimism on this front, since it opens up so many advantages to developers: faster iteration (software updates), remote monitoring, better user experience, etc.

• Dr. Lias succinctly summarized the challenges around modular, component artificial pancreas systems, with a clear emphasis on the lack of device communication standards.  

Challenge	FDA	Industry/Scientific Community
Data format	Need assurance that devices compatible – standards will facilitate	Need to develop standards? Discuss needs
Secure/private communication protocols	Need assurance that devices secure – standards will facilitate	Standard under discussion, discuss needs
Component device modifications	Need assurance that device modifications will not have unintended impact on systems	Need to develop standards?
Post-market responsibility	Need to define who is responsible in case of component/system failure (investigations, complaints, etc.)	Should provide suggestions/input
Mobile phone capabilities (alarms, app priority)	FDA encourages use of mobile apps	Need to address current limitations and difference between operating systems
Operating system updates	FDA already has efficient pathway developed	Needs to identify technical challenges to allow multiple systems

Academic artificial pancreas update

Trang Ly, PhD (Stanford University, Palo Alto, CA)

Stanford’s Dr. Trang Ly provided an outstanding overview of artificial pancreas systems, calling the technology “transformative”; characterizing first-gen product algorithms as “conservative”; and summarizing systems’ strengths, areas for improvement, and algorithm nuances (see tables below). Dr. Ly argued that the biggest challenge Medtronic’s MiniMed 670G faces is managing expectations (“This is not going to cure you. That will be their biggest problem”). She also noted that MiniMed Connect doesn’t work with the 670G right now (and Bluetooth is not built in), and the lack of an adjustable glucose target needs to improve. On the bright side, she noted that the 670G is a very nicely integrated system and the pivotal trial is complete, putting it first in line to market. Dr. Ly said some of the next challenges for closed loop systems include user interface and human factors: “These will not last on the market if we don’t see improvement in user interface and how a person interacts with the system.” Totally agree there, and this is where a multitude of companies selling products is going to drive innovation and better all-around user experiences. Dr. Ly also highlighted room for improvement in fail-safe modes – “going back to pump mode with preset settings is not necessarily safer.”

System Strengths and Challenges

	Medtronic MiniMed 670G	UVA / TypeZero	Cambridge	Harvard	BU
Strengths	Integrated system 124 patients, three-month pivotal trial	Six-month data, n=13 Remote monitoring Software updates Pump agnostic	Three months, n=33 Lots of clinical data	Run-to-run optimization currently being tested	N=48 over 12 days Integrated, system Adjustable set-point (“really important”) Meal adaptation
Challenges	Managing expectations Need remote monitoring (won’t work with MiniMed Connect) Adjustable set point	Moving to commercial platform Connectivity Set-point may be too high during day	No commercial partner No remote monitoring System not integrated	No commercial partner (different algorithm from Insulet/Mode AGC) No remote monitoring Need longer studies	Glucagon – long-term effects Design being worked out – IOB Need longer studies

• Initialization Parameters: BU’s Bionic Pancreas has the clear advantage here, requiring only weight to start closed loop, and then it adapts over time. Initialization could play an important role in who these systems appeal to (current pumpers vs. MDIs) and how hard they are to train providers and patients on.

	Medtronic MiniMed 670G	UVA / TypeZero	Cambridge	Harvard	BU
Parameters to Start	TDD Basal I:C ISF Sensor for 48 Hours	TDD Basal I:C ISF Weight	TDD Basal I:C ISF Weight	TDD Basal I:C ISF	Weight

• Setpoint (Glucose Target): Cambridge, Harvard, and BU have the most aggressive targets; UVA/TypeZero is notably conservative during the day (160 mg/dl).

	Medtronic MiniMed 670G	UVA / TypeZero	Cambridge	Harvard	BU
Setpoint	120 mg/dl	Day: 160 mg/dl Night: 120 mg/dl	Treat-to-target: 104-131 mg/dl Optional individual setpoint	Day: 80-140 mg/dl Night: 90-140 mg/dl	Insulin and Glucagon setpoint is 100 mg/dl Adjustable by user up to 130 mg/dl
Exercise	Temp target: 150 mg/dl	Safety: no more than usual basal	Exercise-specific setpoint

• Automated Component: Most of the systems dose every five minutes, and most use MPC. It’s hard to know how meaningful these algorithm differences are before seeing them head-to-head in clinical trials.

	Medtronic MiniMed 670G	UVA / TypeZero	Cambridge	Harvard	BU
Algorithm	PID-IFB	Control to Range	MPC	MPC	MPC: Insulin PD: Glucagon
Dosing	Microbolus, 5 minutes	Basal: 5 mins Correction: 1 hr	Insulin infusion rate: 10 mins	5 mins	5 mins System-calculated meal dosing

#OpenApS and the Drive to DIY Diabetes Technology

Mark Wilson (OpenAPS, Freelance Software Developer, San Francisco, CA)

Mark Wilson received a standing ovation after his talk on the do-it-yourself OpenAPS, noting that open, interoperable diabetes devices will facilitate an ecosystem of diabetes innovation that manufacturers could never have imagined – and that will help drive adoption and dramatically improve products. Mr. Wilson predicted that the first pump to securely enable patients to access it (e.g., from a phone app) will see tremendous success. As an example of what devices with open APIs can offer, Mr. Wilson cited Fitbit’s wireless scale, Aria – the scale connects to Wi-Fi and sends the data to the cloud, allowing developers to connect to it and build novel apps that use the data. BeeMinder allows a user to connect their credit card and Fitbit scale seamlessly: “If I don’t weigh 160 lbs in three months, charge me $150. Fitbit won’t think of that, but it adds value to Fitbit. This is what we need from diabetes devices.” He characterized OpenAPS as a “platform” that enables different devices, closed-loop algorithms, and data displays to be tailored to each patient’s preferences. Indeed, he boluses from his Apple Watch, and in addition to selecting quantity of carbs, he denotes whether it is a lollipop, taco, or pizza bolus. Mr. Wilson lamented the difficulty of accessing current devices, citing Ben West’s ultimately successful six-year quest (!) to hack into the Medtronic pump. There is even a $13,000 reward being offered to crack the OmniPod’s communication protocol! Mr. Wilson characterized this as a waste of time, subtracting valuable resources from building compelling user interfaces and new ways of managing diabetes. His diabetes device-as-car analogy beautifully explained the rationale for OpenAPS: “Diabetes is not about the car; it’s about the drive. There is a perception that DIY people are car fanatics, and we just have to tinker with our devices – we want the fancy rims. The reality is we were sentenced to drive, and we were given a car – clearly by someone who has never been behind the wheel. We [OpenAPS] happen to open up the car and hotwire it, and it makes our drive so much better.” We fervently hope some of the learning from OpenAPS will make it into the hands of industry. There are now 81 people on the system globally, collecting over a thousands hours of closed-loop data every day.

Symposium: Inpatient Management of Diabetes and Hyperglycemia – Novel Insights and Effective Approaches

Automating Insulin Dosing Decisions – Benefits, Limitations, and Cost Effectiveness

Thomas Pieber, MD (Medical University of Graz, Austria)

Dr. Thomas Pieber shared results of a study that evaluated the clinical usefulness and cost-effectiveness of GlucoTab, a tablet-based workflow management software that provides suggestions for inpatient insulin dosing. The system provides glucose management advise directly at the point of care, by documenting blood glucose levels over time and recommending changes in insulin delivery based on a patient’s age, weight, renal function, insulin sensitivity, and food consumption. In a non-randomized study of the device in four clinical wards, results indicated that patients using GlucoTab experienced substantial improvements in glycemic management – the odds of a patient experiencing a hypoglycemic event was 3.1 times lower among patients managed with GlucoTab vs. those not managed with the software (95% CI: 1.4-6.8) while the odds of hyperglycemic events was 2.2 times lower (95% CI: 1.1-4.6). Notably, Dr. Pieber shared that these benefits could theoretically result in cost savings of as much as 26 million euros (~$30 million) per year in direct costs (excluding indirect costs and costs due to medication errors) if GlucoTab were implemented in all hospitals in his home country of Austria – this would more than cover the projected cost of implementing the program (six million euros). Dr. Pieber admitted that the software’s greatest shortcoming is that there is not yet RCT outcome data to support it. However, his team hopes to conduct a larger trial across Switzerland, Austria, and Germany in the near future that will strengthen the body of evidence supporting the software. He did not share any timing and it sounded like financial resources are the gating factor. We wonder if such a trial is even needed, given how grim and difficult some inpatient management is – software really can do this better.

Questions and Answers

Q: How accepting was the nursing staff to this degree of digitalization?

A: This is an important question because it’s subjective. We measured acceptance with a questionnaire and they liked it overall because it was organized. However, there is a threshold you have to overcome when introducing a new system, but they grew to like it.

Q: How many different medical or surgical departments and countries has this been evaluated in?

A: We want to get the funding to do a large randomized controlled trial in Switzerland, Austria, and Germany. Currently, our studies are in regional hospitals. Non-university hospitals are our next step.

Q: Was there one tablet per floor? Does the tablet provide other functions, or is the tablet solely dedicated to the practice of insulin management?

A: There would be one in a nursing group, which would take care of between 8-12 patients. The software is also web-based so they would not be limited by the single tablet.

Corporate Symposium: Partnering to Improve Patient Care: A Technological Evolution in Insulin Delivery (Sponsored by Johnson & Johnson)

Addressing the Unmet Needs in the Diabetes Space

John Wilson (Worldwide VP, Insulin Delivery, J&J Diabetes Care Companies, Wayne, PA)

Mr. John Wilson kicked off Johnson & Johnson’s Sunday night corporate symposium with an overview of the company’s contributions in the diabetes space. Indeed, J&J’s commitment to the field emerged as a theme throughout the evening, with Mr. Wilson stressing that the segment has a strong foundation and one that will be “enduring.” This commentary comes on the heels of similar optimistic remarks from J&J’s May Medical Device business review, which shared plans to launch the OneTouch Via (formerly Calibra Finesse) insulin delivery device by May 2017 and its long-awaited hypoglycemia-hyperglycemia minimizer with Dexcom CGM by November 2017 (within 18 months). [J&J has informed us that the OneTouch Via timing has now moved up slightly (launching in limited markets OUS in late 2016, with US launch to follow shortly thereafter), while the hypoglycemia-hyperglycemia minimizer is now more conservatively slated to launch within 18-24 months of May 2016 = November 2017-May 2018.]

Artificial Pancreas: What More Do We Need To Succeed?

Brian Levy, MD (Chief Medical Officer, J&J Diabetes Care Companies, Wayne, PA)

Dr. Brian Levy shared perspective on what it will take to succeed in the artificial pancreas race. Dr. Levy did not share specifics on the planned hypoglycemia-hyperglycemia minimizer with Dexcom CGM, but did confirm in Q&A that the pivotal study will start this fall (consistent with the May Medical Device business review, which called for a late 2016 pivotal trial). Commentary in Q&A suggested Animas is still working with the FDA on the study design, and though it will not be first to market, it does intend to be best in the market (particularly in pediatrics). Dr. Levy spent a good portion of his talk on the importance of incorporating quality pump systems, and shared confidence that the Animas pump platform is superior in terms of precision, accuracy, and bolus delivery. We’re not sure whether this difference is clinically meaningful from an outcomes perspective (as Lane Desborough has humorously said, “We precisely deliver inaccurate amounts of insulin”). Of course, small bolus increments are very important for children, and fast bolus delivery is important for those taking very large doses at one time (e.g., 10-15 units). Just as we heard at ATTD, J&J will have ancillary cloud-based software for its automated insulin delivery product – great to see and a definite must-have to stay competitive with Medtronic’s expanding data ecosystem.

• Dr. Levy spent most of his talk focusing on the pumps that will be used in closed-loop systems, highlighting Animas pumps’ accuracy and precision. He stressed that these improvements make Animas pumps more suited to insulin-sensitive populations, such as in children or the elderly. He also highlighted the water proofing as particularly useful in children, who “may be less gentle than desired.”
• J&J’s closed-loop ancillary software will be a cloud-based diabetes management ecosystem, “with the patient at the forefront.” Dr. Levy highlighted the ability to communicate data with friends and family, health professionals, and payers – consistent with several of Dr. Levy’s references to children, the slide presenting these ideas featured two young children at the center of the network.
  
  • Dr. Levy also highlighted Animas Academy, an educational website first discussed at ATTD, launched in the EU, and hopefully coming to the US soon. It offers practical pump therapy tips for patients, caregivers, and healthcare providers. Animas has done a good job of making the content holistic (everything from basal rates to exercise to sleepovers), layering the content within five broad chapters (Get started with your pump, pump therapy essentials, day to day life, for your kids, HCPs), and refining the content down to key bullet points. See the Animas Academy website.
• Dr. Levy noted that closed-loop developers still have a ways to go in terms of the psychosocial and practical elements of artificial pancreas design (e.g., minimizing burden of wear, user interface, human factors, etc.). We certainly agree that the questions of form factor and design will have critical implications for product development and market uptake of AP systems. Our biggest worry is that these systems will be designed for current pump and CMG users, which would be a mistake 
• Dr. Levy wrapped up by touching on the necessary improvements needed in CGM technology, noting that while accuracy has improved, he believes that MARDs in the single digits are needed for commercial AP systems to meet patient’s sky-high expectations – of course, this is where its partner, Dexcom stands with G5 (MARD: 9.0%). [For context, Medtronic’s much-improved Enlite 3 has an MARD of 10%, which is technically double digits, but certainly good enough to drive better outcomes. The 10% threshold is arguably arbitrary (it’s based on modeling work from Dr. Boris Kovatchev), but it does make for good memorable marketing.]

Pathways to the Artificial Pancreas

Satish Garg, MD (University of Colorado, Aurora, CO)

Dr. Satish Garg set the scene for the evening’s discussions with an overview of data on CGM/pump use and automated insulin delivery (AID), emphasizing the meaningful reduction in hypoglycemia offered by these technologies. He highlighted Medtronic’s 2013 ASPIRE study of the MiniMed 530G, showing a 38% reduction in nocturnal hypoglycemia with threshold suspend. Dr. Garg also referenced the recent Medtronic 670G trial (also presented at ADA 2016) showing a 44% reduction of both nocturnal and daytime hypoglycemia and a 0.5% reduction in A1c – “we are down to nearly zero hypoglycemia,” he stated, noting that the future for the artificial pancreas is bright, especially with the promise of sophisticated data analytics. In addition, Dr. Garg contextualized the “pathway to the artificial pancreas” by showing T1D exchange data (n=27,000) demonstrating that only ~33% of patients with type 1 diabetes are using insulin pumps, though a greater percentage of pediatric patients (65%) use the technology. Further, CGM usage has increased from 6% in 2012 to 15% in 2016; he noted, however, that uptake is still far below what was expected. In closing, Dr. Garg mentioned the July 21 FDA Advisory Committee meeting to discuss a replacement claim (non-adjunctive) for Dexcom CGM, suggesting that a positive outcome could increase uptake of the technology (especially in Medicare, of course). An insulin-dosing claim would be a key regulatory win for the field, and we will be curious to see how much it impacts penetration, if approved. This approval is not the only reason keeping people from adopting CGM (cost, perceived inaccuracy, device burden, inconvenience), though it does play a role in the value proposition and prescribing.

The Artificial Pancreas Roadmap: Finding the Ways to Keep Glucose in a Safe Zone

Eric Renard MD, PhD (Montpellier University, France)

Dr. Eric Renard provided an overview of several closed-loop control algorithms, including proportional-integral-derivative (PID), model predictive control (MPC), and fuzzy-logic (MD-logic), emphasizing the overall movement from reactive systems (PID) to predictive systems (MPC) that adjust insulin delivery according to expected glucose trends. However, Dr. Renard stated that MPC meal control remains a “weak point”, noting that current insulin kinetics and delivery route (e.g. subcutaneous injection) limit the ability for AID systems to mimic pancreas activity by delaying insulin action. To that end, he noted that elements such as meal announcement and insulin-on-board can help the system respond to blood glucose trends before they reach extreme highs or lows. Next, Dr. Renard walked the audience through the “roadmap to the AP in clinical practice”: from low-threshold insulin suspension to predictive low-glucose suspension to hypo-hyper minimization, and finally to multi-hormone systems (glucagon, amylin, etc.). According to Dr. Renard, these systems will progressively flatten glucose levels and keep them in a safer range. Of course, a hugely important part of new technology commercialization will be managing patient expectations as none of these technologies will be a perfect replacement pancreas, and all will require some degree of action by the user.

Partnering for a Patient-Centric Approach to Innovation

Krishna Venugopalan, PhD (Worldwide Director, Research & Development, Insulin Delivery, J&J Diabetes Care Companies, Wayne, PA)

Dr. Krishna Venugopalan shared his belief that improvements in closed-loop algorithms and the integration of sleep, activity, and meal-tracking data into closed-loop systems will prove to be the “most salient” advances as this field moves forward. He spoke specifically to the high expectations of closed-loop care – more automation, data analytics, personalized therapy – noting that researchers are going to have to integrate more inputs to achieve these goals. Dr. Venugopalan also noted the importance of improved infusion site management for improved reliability of insulin delivery, remarking that this goes hand-in-hand with still-needed extensions of pump/infusion set life (to reduce the burden of site and device changes). In conclusion, he suggested that the field is still some ways from addressing all patients concerns about the artificial pancreas (e.g., safety, discreteness, complexity, etc.) but acknowledged on a positive note that calls for patient-centered innovation have moved this conversation in the right direction of late.

Questions and Answers

Q: Will the first-ever closed loop Animas product be the HHM?

Dr. Levy: The simple answer is yes. That is the product we are working on to bring to market now. We are very excited to be starting a pivotal study that will hopefully meet FDA approval in the fall of this year.

Q: How far apart should the CGM sensor be from the pump needle to avoid inaccuracy in reading?

Dr. Renard: As a matter of fact, we have done some studies in the EU where we put sensors inside the insulin cannula – because we have a dream where patients will have to only make one insertion into their body – and we learned that this works perfectly well. So, there doesn’t need to be a large distance between the insulin infusion set and the sensing.

Dr. Garg: The amount of insulin going into the patient is so small that it doesn’t interfere with the sensing. In fact, companies are working to develop devices where everything – sensors and pumps – are inserted at once. While this is not available yet, we are hopeful this technology will be available soon.

Q: What work is being done to integrate CGM and pump data into electronic medical records such as Epic, Cerner, etc.?

Dr. Venugopalan: Data portability and interoperability are really important to us. Through OneTouch Reveal, we are creating standard APIs where data from our systems can be pushed into electronic medical records and displayed there. We have a very strong focus in the R&D team to drive full integration with EMR.

Dr. Garg: At the Barbara Davis Center, we have been working closely with Tidepool and Glooko. Through HL7 we are pretty close to being able to incorporate almost all the pump and sensor data into Epic in the next three to four months.

Q: What developments are planned between Animas’ pumps and Dexcom’s CGM?

Dr. Levy: We are partnered with Dexcom, and we have already released our Animas Vibe, a CGM-enabled pump. We are going to be using Dexcom’s technology moving forward with our artificial pancreas program, so that we can use the best CGM technology with our algorithms and pumps. The hypo- and hyperglycemia minimizer we mentioned will be using Dexcom’s CGM technology.

Q: For patient centric-designs, how much input comes from populations such as minorities, low socioeconomic status, etc.?

Dr. Venugopalan: Diversity of input is very important for us. Whenever we do market research or collect insight, we are very careful to ensure that there is broad inclusion of people from different demographics. It’s several thousands of data points so we can ensure that that broad coverage is there.

Q: Isn’t Animas late in terms of CGM integration since Dexcom has moved on to its Gen 5? Medtronic has also launched its 640G, which is more rigorous than the Animas pumps and is closer to the closed loop.

Dr. Garg: First, it is not a bad idea to be late. You can learn from others’ mistakes. I don’t think there’s anything “late” from Animas. Additionally, J&J is building in additional platforms and has started its first round of studies on Gen 6. They are working behind the scenes. I also want to assure you that it’s not just Medtronic’s 640G we are looking out for: Medtronic’s 670G will also soon be launched in the US.

Mr. Wilson: We are embarking on something ambitious, which takes time and is taking longer than anyone would like. However, we are not about being first-to-market; we are about being the best in the market. Those who are working to develop our product feel confident in the predictive nature and the accuracy and precision of our pumps. Those are all critical components of pumps in the market today, and it is even more important in the closed-loop solution. We also can’t have an artificial pancreas without CGM, and we are coupling our most accurate pump with the most accurate CGM. We are pushing as hard as we can to accelerate the product.

Dr. Venugopalan: Overall, we look at the opportunity significantly and seriously. For example, we are the only pump on the market that has been approved to market to children down to two years of age. We are coming out with automated solutions in hypo- and hyperglycemia reduction. We’re working with some of the best minds in the industry, and we are working with the FDA to craft our study designs. We are also expecting to have studies post-approval to focus on safety and to improve the public accessibility of our products.

-- by Adam Brown, John Erdman, Lauren Forbes, Samiul Haque, Varun Iyengar, Stephanie Kahn, Brian Levine, Hannah Martin, Ava Runge, Tony Thaweethai, and Kelly Close