American Diabetes Association 78th Scientific Sessions

June 22-26, 2018; Orlando, FL; Day #1 Highlights – Draft

Executive Highlights

  • Diabetes therapy: Experts at the invite-only ATTD Consensus meeting on managing DKA for type 1s on an SGLT inhibitor covered a lot in four hours, with seemingly complete consensus on the need to identify the right patient types for SGLTs (if not exactly which ones) and the fact that appropriate ketone testing would be required (what sort of testing wasn’t agreed on). There was also agreement on the massive need for patient and physician/HCP education and treatment on symptoms and treatment of DKA and the need to and importance of optimizing insulin therapy prior to starting SGLTs. There was debate on exactly how to monitor ketones (via urine or blood?), how often to check, whether to have A1c cutoffs, and how to adjust insulin dose (basal or bolus?). Dr. John Buse captured the importance of this endeavor, to come up with standard best practices for DKA risk management: “I’d rather that some people just not get SGLT inhibitors based on our recommendations, rather than see an epidemic of DKA after these drugs are approved. I’d rather the 20% of patients who could really benefit get this drug.” To be sure, SGLTs won’t be a magic pill in type 1 diabetes, and understanding the safety component is obviously critical. But these 30 thought leaders gathered because they believe in the immense value of SGLT inhibitors as adjunct therapy in type 1 (as do we), and we’re excited to see the final consensus manuscript. Prof. Thomas Danne, who organized the meeting, is shooting for a publication by year-end. Other therapy highlights include DELIVER-2 data supporting Toujeo’s cost-effectiveness (>$1,400 saved per patient per year), a session all about CVOTs featuring Drs. Steve Marso and Darren McGuire, and new insights on type 2 diabetes remission from DiRECT.

  • Diabetes technology: A pre-conference symposium on CGM brought an all-star lineup to share opinions on the state of the field: Drs. Irl Hirsch, Anne Peters, Rich Bergenstal, and Ms. Davida Kruger. Meanwhile, Diabetes Mine’s off-site D-Data event was a goldmine for highlights, including One Drop’s new 12-hour glucose prediction for type 2s not on insulin (very cool), plans for the first submission of an Open Protocol insulin pump (SOOIL’s Dana), FDA’s Dr. Courtney Lias on iCGM, a moving speech from Dr. Anne Peters recalling her grueling fight to obtain CGM coverage for a homeless man, and many cool product demos. On the new data front, we saw outstanding results from a new approach to FreeStyle Libre Pro in India (one clinic visit halfway through) drove a nine-hour per day improvement in time-in-range! Medtronic’s Dr. Huzefa Neemuchwala presented the latest on Sugar.IQ’s limited learning launch (+36 minutes/day in-range – that’s 220 hours a year!) plus a look at engagement metrics, the pipeline, and the addition of Fitbit tracking. Senseonics’ 180-day XL showed a strong MARD of 9.4% in adolescents/adults. We also captured promising data from a pilot of UVA’s MDI decision support system, were excited to hear about the OHSU group’s MDI decision support app and Helmsley-funded, Jaeb-coordinated pilot study (T1-DEXI), plus Harvard’s in-development photo meal recognition bolus calculator!

Here we go! ADA 2018 is underway, and will continue through the weekend, until Tuesday afternoon. This report covers 21 highlights from day #1 – 16 in diabetes technology and five in diabetes therapy.

We’ll have lots more coverage of this meeting coming your way over the next few days, and we’ll post every highlight we write at (our Resource Hub!). Also at this link, you’ll find our category documents organizing abstracts by topic, plus our day-by-day conference preview for a glimpse of what’s to come on Saturday, Sunday, Monday, and Tuesday.

Back in a flash … happy reading!

Table of Contents 

Diabetes Therapy Highlights

1. How to Manage DKA Risk in Type 1s on an SGLT Inhibitor? Important Questions Emerging

Before official ADA programming began, 30 thought leaders gathered in a small Hyatt conference room to start drafting a consensus paper on managing DKA risk in type 1s taking an SGLT inhibitor. After four hours, the group felt very aligned on multiple big-picture fronts and is still working on specifics. Everyone wholeheartedly agreed that the benefits of this drug class in people with type 1 diabetes far exceed the risks; Drs. Satish Garg and Helena Rodbard described dramatic weight loss and how this makes patients want to continue their SGLT regimen, while Dr. John Buse and others emphasized the decreased glucose variability/increased time-in-range. Prof. Thomas Danne, who organized this meeting on behalf of ATTD, mentioned that every single person invited accepted his invitation – wow! “I think that goes to show how people feel about these agents, and wanting to get them distributed to patients with type 1 diabetes safely,” he remarked. We were certainly overwhelmed by the passion in the room, as we’re acutely aware of the unmet need for adjunct type 1 diabetes therapies, and it’s inspiring to know that so many of the most highly-esteemed diabetes docs are committed to this, too. At this stage, there are more questions than answers when it comes to the very specifics but there is a long way to go before standardizing best practices for DKA risk mitigation. Ultimately, we’re optimistic that this consensus group will help make the practice of SGLTs for type 1 safer in the real world. Sanofi/Lexicon have filed an NDA for SGLT-1/2 dual inhibitor sotagliflozin (intended brand name Zynquista), and AZ is targeting 2H18 for FDA submission of SGLT-2 inhibitor dapagliflozin for type 1. Both sota and dapa have been submitted to EMA for a type 1 diabetes indication, and Lilly/BI will present topline phase 3 data on empagliflozin in type 1 (the EASE program) on Tuesday. We’re anticipating an FDA Advisory Committee in January, and Prof. Danne set the goal of having a manuscript published by end of 2018, “to be looked at before regulatory meetings” for sotagliflozin and dapagliflozin. 

There was some disagreement on where this paper should be submitted – Diabetes Care, or a journal more widely-read by PCPs, like JAMA? Dr. Jay Skyler pointed out that primary care providers and hospitalists are an important audience; DKA is less familiar within these healthcare communities, and ER staff might need the most education on how to treat an episode of DKA, since they’re the providers on the front lines. Dr. William Tamborlane captured a key theme of the consensus when he suggested an aim toward recommendations that “educate the uneducated HCPs without tying our hands.” That is to say, diabetologists well-versed in prescribing SGLTs to type 1s may be able to maximize benefit and maintain safety even in those at high-risk for DKA (e.g. more frequent intense exercise, higher starting A1c, etc.). To this end, the group settled on issuing clear and specific guidelines rather than mandatory requirements. Below, we summarize some of the major areas of agreement that emerged from this gathering, as well as some areas of continued debate. We follow that with future directions for research and quotable quotes. We look forward to reading this consensus paper once finalized.

Representatives from Sanofi, Lexicon, AZ, and Lilly/BI lined the back wall of the conference room, but were not active participants in the discussion. Similarly, two representatives from FDA’s Division of Metabolism and Endocrinology Products listened to the conversation but did not comment; they were Drs. Mahtab Niyyati and Mitra Rauschecker. Prof. Danne invited EMA to send representatives but did not hear back. Our two cents is that many people with type 1 diabetes are already taking SGLT-2 inhibitors off-label, and regulatory approval will make this practice much safer overall. The thought leaders acknowledged several times today that we have very limited evidence on SGLT-related DKA in type 1s, because across RCTs there have been <30 events – this group is in a unique position of writing a consensus document that is less data-driven, and is more so built from clinical expertise. Approval of these first oral adjunct treatments would also offer an opportunity for real-world evidence collection on safety as well as efficacy.

  • To be sure, there was strong agreement in the room on some points, most notably that these therapies should be approved for type 1 diabetes given the immense clinical benefits of weight loss and increased time-in-range. Dr. Jeremy Pettus suggested starting the consensus manuscript with an explicit emphasis on these benefits, so that HCPs don’t shy away from SGLTs due to a small (but serious) risk of DKA. Others echoed this sentiment, underscoring that DKA risk should be manageable in the real world. The question remains how best to manage it.

    • To maximize efficacy and ensure safety, thought leaders agreed that education will be key. Dr. Anne Peter spoke from her experience prescribing canagliflozin (J&J’s SGLT-2 Invokana) to patients with type 1 diabetes in J&J’s phase 2 trial, when she saw three episodes of DKA. One year later, she was prescribing SGLT-2s to type 1s without running into any DKA. Dr. Peters emphasized the value of teaching providers how to teach their patients (“I learned how to teach DKA risk management,” she said).

    • There was consensus on how to treat DKA when it appears, via the STICH paradigm: STop the SGLT inhibitor + Insulin + Carbs + Hydrate. Thought leaders discussed eating 30g of carbs, injecting insulin, and re-checking ketones every three-four hours, then seeking emergency medical care if necessary. Prof. Danne acknowledged that it’s counterintuitive for a patient to both administer insulin and eat carbs when blood glucose is in-range (the room nodded in agreement); getting this message across to patients, HCPs, and ER staff is thus critical. There were also mentions of checking for pump failure. Importantly, there’s a distinction between ketosis and DKA, and Dr. Chantal Mathieu stressed that DKA is reversible if certain steps are taken swiftly.

    • Not all type 1s are ideal candidates for an SGLT inhibitor, and these drugs shouldn’t be taken in certain scenarios. The experts agreed that patients on a low carb diet probably shouldn’t add an SGLT to their medication regimen. Dr. Peters shared that she sometimes frames this as a choice for her patients: Do you want an SGLT inhibitor or do you want to stick to a low carb diet? She added that the risk/benefit ratio of SGLT inhibitors may be skewed toward risk in patients with lower engagement/adherence (i.e. skipping insulin doses) and in those with limited access to emergency care. In her experience, SGLTs are riskier in her East LA patients vs. her West LA/Beverley Hills patients, and the sad reality is that individuals of lower socioeconomic status, who are more vulnerable to diabetes and its complications, who stand to benefit immensely from an adjunct therapy that improves their time-in-range, body weight, and insulin regimen, these people also face the greatest DKA risk. Of course, access issues are inescapable for any diabetes drug class. Still, it’s upsetting to think that access to care is so closely linked to safety outcomes in the case of SGLT inhibitors as adjunct treatments in type 1. When should patients be told to hold their SGLT inhibitor? The group listed surgery, fasting, and intense physical activity (Dr. Helena Rodbard told the story of one of her patients climbing Machu Picchu in Peru, and leaving her SGLT-2 at home).

    • Wallet cards emerged as a promising safety strategy for anyone with type 1 diabetes taking an SGLT. Dr. Mathieu recommended dedicating one side of the card to risk minimization (e.g. be wary of high alcohol intake) and the other side to treatment (e.g. STICH). To us, this seems like a fairly simple, feasible, and impactful idea. Should these cards originate from the manufacturer and be dispensed at the pharmacy, or be given to patients by their prescribing clinician? How might FDA regulate the language on these cards?

  • The morning featured more debate than agreement, however. There was continued discussion on ketone monitoring (how often, and with urine or blood?), insulin adjustments (reduction by how much, and should this come from basal or bolus?), and A1c cutoffs (what threshold, and should insulin optimization always come first?).

    • Ketone monitoring frequency/methodology: As Prof. Danne put it, “if we had a continuous ketone monitor, there wouldn’t be a controversy here.” But the question stands as to how often patients should check ketones. Daily (in the morning)? A few times per week (largely for education purposes)? Only when one feels ill? There was limited agreement on which of these options is best or safest for the majority of patients. Prof. Simon Heller asserted early in the discussion that he would preferably ask all patients for daily ketone checking, because it’s precautionary and because people are more likely to follow a daily routine. Dr. Peters’ personal protocol stipulates daily checking unless the patient feels “100%” (for some that may be much?), and Dr. Paresh Dandona endorsed checking once or twice each week, mostly so that the patient is informed about DKA. Dr. Peters also aptly pointed out that we ask people with diabetes to check their blood glucose every morning, so the concept isn’t unheard of. Notably, Dr. Mathieu was very opposed to the requirement that patients should check ketones every morning. She explained that all the cases of DKA she has witnessed have happened during the day, absent of morning ketones. Moreover, the burden of checking daily for an unlikely event isn’t sustainable; patients will do it for a while and then the practice will fade away (Drs. Mathieu, Pettus, and Tamborlane all agreed on this point). As such, perhaps patients should be taught to check ketones during illness or prolonged activity, and most importantly, when feeling unwell. There were audible murmurs of disagreement when the conversation veered toward “checking ketones only when ill is OK.” Increasingly, we wonder if this will differ drastically by individual patient, making it challenging to write one recommendation for the majority. We did very much appreciate Dr. Garg’s emphasis on the high cost of blood ketone monitoring. He suggested that this will frustrate patients if they aren’t actually detecting a problem. He further underscored that there isn’t concrete evidence that daily checking prevents episodes of DKA. This was not a protocol used in clinical trials.

    • Dr. Ele Ferrannini argued that assessing ketone trends is more important than a single measurement in time: “You can’t find a random occurrence with a random strategy.” If ketones rise in response to insulin limitation, that’s a signal for danger. He also described how DKA is always linked to a recognizable cause. There’s always a trigger, which means there’s always a way to prevent DK from occurring. Dr. George Grunberger echoed this view that DKA is distinctly preventable, referencing AACE’s consensus conference in asserting that DKA can always be traced back to a trigger.

    • Dr. Peters asks patients looking to start on an SGLT-2 off-label to check ketones daily for one-two weeks before initiating the drug. She uses this to understand adherence and to educate patients on how to monitor for ketones; it also allows her to establish a personal baseline for that particular patient, to confirm that he/she isn’t in ketosis on a daily basis without the drug. Dr. Pettus added that these one-two weeks of lead-up also help ensure that the patient has access to ketone testing supplies. Dr. Mathieu pointed out that it can indicate whether a patient is under-insulinized, with the caveat that high morning ketones can be a consequence of overnight hypoglycemia; Dr. Skyler made a plug for CGM to discern this.

    • At what precise ketone level should patients/providers be concerned? Suggested cutoffs were 0.5-1.0 mmol/L (risk of ketosis; repeat check in 30-60 minutes), 1-3 mmol/L (confirmed ketosis of varying severity; take fluids, eat carbs, and bolus insulin), and >3 mmol/L (high risk of DKA; consider going to ER if persistent vomiting and signs of dehydration). Many seemed concerned that the lower end of these recommendations was actually too conservative. Dr. Mathieu explained that the 0.6 mmol/L cutoff used in phase 3 trials was based on insulin pump studies; she described it as a level at which everything is not yet lost and where giving carbs and insulin can solve the problem. Dr. Pettus expressed worry that many patients would sporadically have levels around 0.6 mmol/L, causing undue stress; in his clinic, people with type 1 are taught that anything over 1.0 and certainly 1.5 mmol/L is cause for concern. Dr. Tamborlane stated that for anything <1.0 mmol/L, he’d tell patients to eat, take insulin, and check ketones again at lunch. Dr. Buse added that he doesn’t think the drug even needs to be held at 0.6 mmol/L, especially if the patient knows how to handle it. Once again, we’re moving toward this notion that DKA risk management will have to be personalized.

    • The issue of blood vs. urine ketone monitoring was another point of debate, though there did seem to be some agreement that “urine is fine, but blood is better” (in Prof. Danne’s words). Indeed, the only hurdle to absolute endorsement of blood testing seems to be the prohibitive cost, at ~$1-2 per strip plus another meter (not trivial, especially with daily checking). As Drs. Pettus and Mathieu emphasized, people hate peeing on strips for reasons of both convenience and perceived hygiene. Dr. Mathieu further suggested that checking via urine won’t catch DKA early enough. What’s more, blood checking is faster and far more precise – at ketone levels between ~0.5-0.6 mmol/L, it can be really hard to tell if there’s a problem from urine testing. Moreover, Dr. Ferrannini explained that ketones aren’t always filtered into the urine, depending on hydration level. All this said, urine is better than nothing, and it could make sense for patients to regularly monitor with urine strips and follow up with a blood meter if they sense a problem. We were glad to hear the room call on industry to work with payers and ketone meter/strip manufacturers to make these supplies more affordable and more frequently reimbursed. Dr. Jake Kushner even positioned this as a “business opportunity” for SGLT manufacturers. The idea that ketone strips could be included in the box with an SGLT inhibitor was met with great enthusiasm.

    • Insulin dose reduction: Should basal or bolus insulin be reduced, and by how much? Should this differ based on the molecule used? Dr. Buse advocated that the group create a table of insulin adjustment recommendations by agent (he alluded to differences between sotagliflozin, dapagliflozin, canagliflozin, and empagliflozin). Dr. Garg argued fiercely that basal insulin should be changed minimally, because as he put it, DKA happens when basal insulin gets interrupted. This argument is in line with what Drs. Garg and Dandona discussed at AACE 2018, when they speculated that the lack of a large basal depot in pump users was responsible for the seemingly higher DKA risk in this patient population. This idea that basal insulin offers a baseline level of protection against ketogenesis makes sense to us. On the other hand, Dr. Peters – who has prescribed SGLT-2s off-label to >100 type 1s, with minimal DKA – recommended reducing basal insulin by 10% if the patient is at target A1c or has a history of nocturnal hypoglycemia. Dr. FJ Ampudia-Blasco was another proponent of reducing basal insulin, and he shared that in his practice, they reduce basal by 20% in those close to target and ≤10% in those with higher A1c. Dr. Pettus offered that, per this late-breaker to be presented on Sunday, ~70%-80% of the insulin reduction in inTandem1 and 2 came from bolus insulin doses; we aren’t aware of any data on this from DEPICT, but Dr. Mathieu seemed to suggest that basal was more commonly reduced in that program. As such, both the type and magnitude of recommended insulin reduction seems to remain a point of contention, and again there will be need for personalization of care. Dr. Pettus called for more data on this front, also raising the possibility that a higher SGLT dose might equal greater reduction in insulin dose. We agree that this issue warrants further investigation, and we wouldn’t be surprised if Dr. Peters’ protocol of reducing basal works so well because she’s only been able to use SGLT-2 inhibitors thus far: Is it possible that basal insulin reduction is better for SGLT-2 inhibitors, while bolus reduction is better for SGLT-1/2 inhibitors? This could be sorted out in Dr. Buse’s table. For reference, the inTandem protocol suggested a 30% reduction in first meal bolus, otherwise maximum tolerated standard of care insulin therapy as determined by the HCP. The DEPICT program capped insulin reductions at 20% of the total daily dose; in DEPICT 1, average reductions were 9% for 5 mg and 13% for 10 mg dapagliflozin vs. placebo.

    • Patient selection: Should there be an A1c cutoff or other recommended restrictions on SGLT inhibitor prescription – and if so, what? Much of the disagreement on patient selection centered around whether it’s safe to give an SGLT to a patient who has an A1c around or above 10%-11%. Some felt that the safest way to use these agents would be to help patients already in fairly good control, already highly engaged in their diabetes self-management. Dr. Skyler argued for insulin optimization before starting an SGLT therapy in any type 1 patient. “Even if you’re around 9%, you should be increasing insulin first, not just adding an SGLT-2 inhibitor. Get patients in good control and then smooth things out with the SGLT.” He explained that high A1c is often a sign of under-insulinization, and Dr. Pettus noted that there may be a tendency to prescribe a pill that’s seemingly “easier” without due consideration of how to maximize insulin benefits. At the same time, Dr. Skyler acknowledged that this practice isn’t always possible: “15% of the teenagers in my practice have an A1c >11%, and they’ve been ‘optimizing insulin’ for five years.” Dr. Helena Rodbard look a slightly more liberal view and was unwilling to outright exclude people with an A1c of ~9%-10%, explaining that she just wouldn’t cut back on insulin when adding an SGLT inhibitor in this population.

    • Dr. Ferrannini considered insulin optimization something to recommend but not mandate, and he also cautioned against thinking of SGLT inhibitors as a cure-all for poor glucose control. Others also shied away from the idea of putting hard and fast rules in place, evoking Dr. Tamborlane’s philosophy of “educating the uneducated HCPs without tying our hands.” Dr. Peters added an illuminating dimension when she recounted two sets of patients, the first being older people with type 1, some of whom have CV disease and want an SGLT inhibitor for the cardioprotection, and the second being women with slightly high A1cs who shy away from more insulin because of fear of weight gain. In her words, “I’m more nervous with a higher A1c, but if they aren’t ketotic and they’re willing to work with me, I’ll do it.” Dr. Buse put things aptly when he offered the example that being on a pump is objectively a higher-risk situation, but those patients aren’t going to be excluded. “All of these things are not an absolute…Why even name a number? It’s not like the risk starts at one point.” He described how a hard cutoff just hurts the people hovering around that cutoff.

    • Dr. Garg described another unfortunate reality that people with high A1c, most likely to benefit from an SGLT inhibitor (glucose-lowering, weight loss), might also face higher risk for DKA. That said, while it’s well-established through T1D Exchange data that risk for DKA is generally increased among patients with higher A1cs, it’s unclear how SGLT-related DKA occurs – and it may not be correct to assume that patients with a higher A1c automatically carry risk factors for DKA, or that these risk factors are actually relevant in terms of SGLT-related DKA. There simply isn’t enough data to determine whether patients with a higher A1c see more DKA with SGLT inhibitors – an industry member in the back of the room actually shared the limited evidence they have may even suggest the opposite. Ultimately, though, they and other members of industry sitting in the back of the room confirmed that the number of DKA events in trials conducted to-date is too small to discern real trends.

  • The meeting concluded with a rundown of lingering questions and knowledge gaps. Most of this four-hour conversation identified uncertainties to explore further, but there were a few interesting questions that came up only toward the end:

    • Do the CV/renal protective effects of SGLT inhibitors carry over into type 1 diabetes? Dr. Peters noted that many of her older patients with type 1 who also have CV disease ask to start an SGLT-2 inhibitor off-label; they’ve heard of the CV benefits in type 2 diabetes and want to experience the same. JDRF’s Dr. Sanjoy Dutta shared that the organization has been surveying pharmaceutical industry reps as well as KOLs, and has found that there’s some interest in conducting a type 1 CVOT down the road (though the first priority is definitely investigating these agents for glucose control). Dr. Garg was skeptical about the feasibility of this trial, but Dr. Buse defended the plausibility as long as the CVOT enrolls type 1s with established CV disease. We’d certainly be keen to see this CVOT, and it’s good to hear early interest from JDRF. We’ll be curious to hear from Sanofi/Lexicon, AZ, and Lilly/BI on their interest in funding such a CVOT. Notably, a Lilly/BI representative brought everyone’s attention to EMPEROR HF and EMPA-KIDNEY, as both programs will enroll subgroups with type 1 diabetes. These studies of SGLT-2 inhibitor empagliflozin in heart failure and CKD will offer a glimpse at CV/renal protection in type 1. According to, people with type 1 diabetes are excluded from AZ’s Dapa-HF and Dapa-CKD.

    • If we have relatively limited data on SGLT use in adult type 1s, use in pediatrics and geriatrics is really an “evidence-free zone.” We expect to see clinical trials enrolling adolescents in the mid-term future, provided sotagliflozin and dapagliflozin are approved for adults with type 1 diabetes. As for the older type 1 patient population, Japan’s Dr. Koutaro Yokote emphasized that it’s not about age, but about self-management capacity: If a patient has cognitive dysfunction, he will definitely face higher risk of DKA, but otherwise SGLTs could have tremendous benefits in the elderly (we loved this sentiment, and couldn’t agree more).

    • Dr. Ele Ferrannini pointed to SGLT inhibitors/physical activity as another “evidence-free zone,” highlighting the possibility for future research. His comments elicited nods from almost everyone around the table.

    • Prof. Simon Heller advised that after this consensus paper is finalized, a prospective study be conducted to rigorously evaluate the proposed protocols for DKA risk management. We don’t yet know what works, and like most things in clinical science, this question demands gold standard RCT evidence. Prof. Heller elaborated, “these opinions are not based on a huge amount of evidence, and to that end, I hope we’re going to actually study whether our protocol is effective. We owe it to patients.”

  • There were so many incredible insights shared at this gathering, some too good to paraphrase, and so we bring you quotable quotes:

    • “I had two sisters both randomized in a sotagliflozin trial, one to the study drug, another to placebo. Soon after starting therapy, one loved me and the other hated me, and that speaks to the benefits of these adjunct treatments in type 1 diabetes.” – Dr. George Grunberger

    • “We’re writing this manuscript because we think this drug class is worth it. – Dr. Jeremy Pettus

    • “I’d rather that some people just not get SGLT inhibitors based on our recommendations, rather than see an epidemic of DKA after these drugs are approved. I’d rather the 20% of patients who could really benefit get this drug.” – Dr. John Buse

    • “My patients have said to me, ‘doc, I’m melting away, I want to keep taking this drug.’ Patients really care about the weight loss with these agents.” – Dr. Satish Garg

    • “People like the weight loss, and what really makes them feel better is that their blood sugars are less variable.” – Dr. John Buse

    • “If we had a continuous ketone monitor (CKM), there wouldn’t be a controversy here. – Prof. Thomas Danne

    • AACE hosted a conference on this issue a few years ago, and we found that in every single case, DKA could have been prevented. It was never unexpected; you could always trace it back to a trigger. The conclusion of that conference was that this is a real risk, and let’s make sure that people are educated on safety strategies. You’d hate this class to be damned because people don’t know how to use it.” – Dr. George Grunberger

    • “Let’s scream it loud – DKA can happen at normal glycemia when you’re on SGLT inhibitor therapy.” – Prof. Moshe Phillip

    • “Let’s keep in mind that these recommendations are not for the Anne Peters of the world. We need consensus advice for the people who aren’t as experienced as everyone around this table, but who will encounter type 1 patients who want an SGLT inhibitor, and who may encounter SGLT-related DKA.” – Dr. Chantal Mathieu

2. DELIVER-2 Cost-Effectiveness Analysis Finds >$1,400 Healthcare Savings Per Patient Per Year After Switching to Sanofi’s Toujeo

Dr. Tim Bailey headlined a symposium on real-world evidence with new cost-effectiveness analysis from Sanofi’s DELIVER-2 study. To complement the original DELIVER-2 findings that switching to Toujeo (insulin glargine) from another basal insulin reduced the risk of hypoglycemia in a real-world setting, Dr. Bailey unveiled that this switch additionally conferred an estimated $1,439 of healthcare savings per patient per year. He elaborated that this was driven by lower hypoglycemia-related healthcare utilization, specifically: Fewer hospitalizations (2.8% of people on Toujeo vs. 4.3% of people on other insulins, p=0.037); a smaller proportion of people visiting the ER (3.1% vs. 5.1%, p=0.007); and fewer patients requiring outpatient care (12.5% vs. 15.4%, p=0.011). We certainly hope these compelling numbers resonate with payers, as we continue to believe that next-gen basal insulins should be viewed not as a cost, but as an investment in greater health and safety. This data further supports findings from RCTs indicating that Toujeo – and next-generation basal insulins overall – are a major step above first-generation analogs, particularly with regard to their hypoglycemia benefits and reduced glucose fluctuations. 

  • UW health economist Dr. Sean Sullivan contextualized that real-world studies like LIGHTNING and DELIVER-2 help fill the gaps in the RCT landscape. Beyond supplementing RCT data, Dr. Sullivan argued that real-world studies are critical in their own right to provide insight on a broader swath of the patient population. Leicester’s Dr. Kamlesh Khunti echoed this point, noting that only 16% of people with type 2 diabetes would have been eligible for the EMPA-REG OUTCOME CVOT (similarly small fractions of the actual patient population would have qualified for ACCORD [11%], ADVANCE [36%], VADT [9%], and PROactive [3.5%]). Dr. Sullivan further noted that RCTs provide a limited picture of the cost-effectiveness of therapies by addressing only “can it work” under perfect circumstances, with no insight into “does it work” in the wild. 

    • We were interested and excited to hear from Dr. Sullivan that major regulatory bodies know this and are slowly beginning to emphasize real-world evidence. FDA has characterized RWE as “an important step toward a fundamentally better understanding of states of disease and health,” he said. In a past editorial, former Senior Medical Officer of the EMA Dr. Hans-Georg Eichler proposed a new regulatory assessment protocol that places safety/efficacy evidence from RCTs and relative effectiveness and cost vs. health benefit evidence from real-world studies on equal footing.

  • Western University’s Dr. Stewart Harris discussed the “efficacy gap” between tightly controlled, high-touch RCTs and real-world studies, where patients receive much less intensive follow-up. Work from Drs. Steve Edelman and Bill Polonsky has estimated that up to 75% of this efficacy gap is attributable to lower adherence in the real world vs. a more engaged clinical trial setting. We appreciated a sentiment from the audience during Q&A: “My takeaway from the Edelman and Polonsky data is that if we had free medications and frequent check-ins with an all-star clinical team, real-world evidence would look just like RCT data.” We couldn’t agree more, and are reminded of commentary from Dr. Mikhail Kosiborod at AACE 2018 that even the placebo arm of a large RCT represents an improvement over current standard of care.

3. 10 Years Later: Drs. Marso & McGuire Reflect on FDA’s 2008 CVOT Guidance, Chart Next Steps

Drs. Steven Marso and Darren McGuire discussed all things CVOTs in an extremely timely session, considering this year marks the 10th anniversary of FDA’s CVOT guidance. Although this was set up as a debate, the two cardiologists agreed that there are opportunities to innovate in CVOT design, to make these large (and expensive) studies more fruitful. Incorporating heart failure as a primary endpoint, especially in SGLT trials, seems like a no-brainer. Probing for cardioprotection in primary prevention also seems like a logical next step to EMPA-REG (100% secondary prevention), LEADER (81% secondary prevention), and CANVAS (66% secondary prevention). Indeed, the DECLARE trial for AZ’s Farxiga (dapagliflozin) includes a co-primary endpoint of hospitalization for heart failure/CV death alongside the classic three-point MACE (non-fatal MI, non-fatal stroke, CV death), and features a larger primary prevention cohort (60%) than ever seen before in a diabetes CVOT. We look very forward to DECLARE results, expected in the second half of this year. Dr. Marso listed “no diabetes” and “obesity” as additional patient populations to investigate in “gen-2 and gen-3” CVOTs, and to this end, we’re particularly excited about Novo Nordisk’s planned SELECT study of semaglutide in obesity. We agree with the conclusion from both these presentations – that CVOTs are here to stay, and it’s worth figuring out how to maximize on industry’s investment by refining trial design. Our one caveat is this: It’s already challenging to compare between CVOTs given differences in study population, study design, study protocol, etc. We implore that in re-envisioning the diabetes CVOT, the field pushes toward more standardization, because this further enhances insights gained and could support treatment decisions. For example, Dr. Marso alluded to greater heart failure benefit with SGLT-2s vs. greater impact on atherosclerosis with GLP-1s; both therapy classes have shown compelling cardioprotection, and now we should help providers identify the patients most likely to benefit from each so that we can improve real-world outcomes.

  • Dr. Marso suggested that it may be worthwhile to revisit the hypothesis that tight glucose control leads to better CV outcomes. “I think LEADER and EMPA-REG put two nails in the coffin for the glycemic hypothesis, and SUSTAIN 6 loosened one of them in my mind,” he explained. Diabetes CVOTs are designed with glycemic equipoise, in order to isolate the CV effects of the molecule. In LEADER and EMPA-REG, change in A1c was similar between the active treatment arm and the placebo arm, but there was a wider A1c gap in SUSTAIN 6 – from a baseline 8.7%, A1c dropped a mean ~1.3% with semaglutide vs. only 0.4% with placebo. Dr. Marso pointed out that the relative risk reduction for three-point MACE was substantially greater in SUSTAIN 6 (26%) vs. LEADER (13%) or EMPA-REG (14%), and he postulated that perhaps improved glucose control contributed to this. To be sure, semaglutide is an incredibly potent glucose-lowering agent, and at ESC 2017, Dr. Esteban Jodar shared a similar hypothesis about the extra benefit of tight A1c control amplifying the GLP-1’s demonstrated cardioprotection.

  • Drs. Marso and McGuire shared opposing views on heart failure with DPP-4 inhibitors. Dr. McGuire reiterated some of his ESC comments in arguing that heart failure is not a class effect of these molecules (only SAVOR-TIMI for AZ’s Onglyza found a significant risk signal, and TECOS for Merck’s Januvia showed a resoundingly neutral hazard ratio of 1.00 for this endpoint). Dr. Marso pointed to the VIVIDD study (n=254) in which Novartis’ Galvus (vildagliptin) was associated with increased left ventricular chamber size and numerically higher heart failure events (though this did not reach statistical significance). Even though this was a small trial, Dr. Marso argued that VIVIDD + SAVOR-TIMI + EXAMINE (for Takeda’s Nesina, again there was a numerical imbalance in heart failure hospitalizations but this wasn’t significant) provide reason enough to conduct follow-up CVOTs with DPP-4 inhibitors. Our sense is that most thought leaders side with Dr. McGuire on this one, and aren’t concerned about heart failure as a DPP-4 class effect; FDA did add heart failure warnings to all DPP-4 product labels (including Januvia) last year, in what we consider an excessively conservative move.

4. Dr. Taylor Presents New DiRECT Analysis: Beta Cell Recovery Separates Responders vs. Non-Responders; Points to Importance of Early Intervention for Diabetes Remission

Presenting new data from DiRECT, Newcastle’s Prof. Roy Taylor showed how success of the intervention (whether or not a patient enters type 2 diabetes remission) depended on recovery of beta cell function. The only significant difference between responders (n=40) and non-responders (n=18) in DiRECT was diabetes duration (mean 2.7 years vs. 3.8 years, p<0.05 for comparison). Prof. Taylor previously hinted at IDF that duration of diabetes predicted responders vs. non-responders, so we were glad to get our eyes on this data. That said, Prof. Taylor underscored that those with longstanding diabetes can still achieve remission with substantial weight loss. Pathophysiologically, responders and non-responders saw many of the same effects from dramatic weight loss: At five months, both groups had below-normal liver fat (the non-responder group drifted slightly above normal at one year, while the responder group did not) and saw similar decreases in pancreatic fat as well. Weight change itself was not significantly different between the groups, though the non-responders did see a slightly greater (though not significantly) regain over the first year. Prof. Taylor takes this as an indication that in both groups, intra-organ fat was reduced and lipid metabolism was normalized – so what was the difference? At both five and 12 months, responders saw a highly significant increase over non-responders in first phase insulin response, and by 12 months, they also had superior maximal insulin secretion. Dr. Taylor explained that there is something inherently different about the ability of the beta cell to recover function between the groups. This was reflected in the study’s A1c results: Responders saw a rapid and significant drop in A1c sustained from five to 12 months, but A1c of non-responders was essentially constant from day one to day 365. Prof. Taylor concluded that reduction of liver and pancreas fat is necessary but not sufficient for remission of type 2 diabetes, and that remission is possible in the majority of people with a disease duration less than six years, dependent on the residual ability of the beta cell.

  • DiRECT suggests that there’s a certain window in which type 2 diabetes can much more easily be sent into remission. While Prof. Taylor and his team of researchers will continue to look for biomarkers that might better predict whether someone will respond to the intensive weight loss intervention, we take these results as a clear sign that earlier, more aggressive intervention in cardiometabolic disease is an important key to better diabetes outcomes and, now, even remission. We note that there were smaller, non-significant baseline differences between responders and non-responders as well, most notably on A1c (7.4% vs. 7.9%) and fasting glucose (150 mg/dl vs. 168 mg/dl). These pathophysiological studies were not conducted on the whole sample, but rather on those who lived close enough to the imaging center in Newcastle (which may have introduced experimental bias). Responders were defined as those who achieved an A1c and fasting plasma glucose below the threshold for a diabetes diagnosis.

  • As a reminder, the DiRECT trial (n=298) – first presented at IDF 2017 – evaluated an intensive weight loss intervention (~800 calories/day for three-five months with stepped food reintroduction and structured long-term weight maintenance support) for type 2 diabetes reversal. Severe caloric restriction led to a mean weight loss of ~30 lbs. While 24% of participants on the intervention achieved the target ≥15 kg (~33 lbs) weight loss, 46% of this group (68 individuals) achieved diabetes remission, vs. only 4% in the placebo group (six individuals). Among people who reached the ≥15 kg weight loss goal, 86% achieved remission. We await longer-term (two- and three-year) results, but this trial has certainly made waves in the diabetes community already. Perhaps most notably, it was implemented by primary care nurses and dieticians through the UK’s NHS, after they received only a day’s worth of training, which is hugely promising for the economic and logistical scalability of this intervention.

University of Utah’s Dr. Simon Fisher presented hot-off-the-press data from his lab on ways to lower mortality following severe hypoglycemia. He explained that while preventing hypoglycemia altogether remains elusive, it may be possible to reduce the mortality of severe hypoglycemia with pharmacotherapy. According to Dr. Fisher, arrhythmia (and the ensuing acute cardiorespiratory arrest) is the main culprit for sudden death with severe hypoglycemia; seizures, which can lead to respiratory arrest, also play a part. In the study he discussed, blocking both of these mechanisms prevented hypoglycemia from becoming fatal. In rodents, combination therapy with the beta blocker atenolol (to dampen the sympathoadrenal response that generates arrhythmias) + levetiracetam (an anti-seizure drug) completely prevented death from insulin-induced severe hypoglycemia. With placebo, approximately 40% of the rodents died. Mortality rate was also around 40% with beta blocker monotherapy (all deaths from seizure, since arrhythmia was prevented) and levetiracetam monotherapy (all deaths from arrhythmia and cardiac arrest, since seizures were prevented). We remain sharply focused on ways to prevent hypoglycemia – through patient education, advanced therapies (GLP-1s, SGLT-2s, next-gen basal insulins), and CGM – but it’s nonetheless encouraging to imagine a treatment that reduces the mortality associated with severe hypoglycemia when it does occur. We are looking forward to further updates from Dr. Fisher’s lab, and our interest is piqued for whether this combination regimen could enter the clinic for people at particularly high risk of severe hypoglycemia. To be sure, this is extremely early-stage research right now.

  • Although the sulfonylurea class is a notoriously bad actor when it comes to hypoglycemia risk, Dr. Fisher’s work suggests that SUs actually reduce the risk of fatal arrhythmias during severe hypo. In other words, while SUs cause more hypoglycemia, they simultaneously reduce the chance that an event will be fatal. Dr. Fischer described how sulfonylureas also attenuate arrhythmias, thereby lowering hypoglycemia-related mortality.

Diabetes Technology Highlights

Role of Continuous Glucose Monitoring in Diabetes Treatment (ADA Pre-Conference Event)

1. Drs. Irl Hirsch and Anne Peters on Fascinating CGM Cases: “Very Helpful” in T2D for Lifestyle Change, But Hard to Show in Trials; Q’s on Libre’s Accuracy; 670G Hacks

In back-to-back talks, Drs. Anne Peters and Irl Hirsch shared fascinating CGM case studies, with a greater-than-usual emphasis on the benefits of CGM in type 2 diabetes. Noted Dr. Hirsch, “CGM can be very helpful in type 2 diabetes.” Two striking cases stood out: professional CGM helping to identify an inaccurate Medicare-provided BGM (!) in one of Dr. Hirsch’s 85-year-old type 2s, and CGM-driven food choice changes resulting in multi-point A1c reductions. Many of Dr. Hirsch’s patients with type 2 diabetes on CGM “have made major changes in lifestyle,” one of the “most interesting things” he’s seen so far. Dr. Peters added, “I put CGM on anybody I can – prediabetes, type 2 diabetes, type 1 diabetes – both professional and real-time CGM. It helps me manage my patients.” Dr. Peters noted that CGM’s benefits are “harder to show in trials,” especially in type 2s with less risk of lows. (Indeed, Abbott’s REPLACE study in type 2 insulin users with a high A1c did not drive as much efficacy as many had hoped.) Dr. Peters added that there is little data on use of CGM in type 2s overall – and especially in those not on insulin – which drives less robust clinical guidelines recommendations. (Of course, this will improve as the products are used and designed in smarter ways for type 2, as software helps drive lifestyle change (e.g., meal photos + CGM), and more studies happen.) On the product front, both offered quite balanced views on what’s available, including mentions of Dexcom’s G6 (“quite accurate” in Dr. Peters’ early experience; read diaTribe’s test drive), Senseonics’ Eversense (approved yesterday), MiniMed 670G in 7-13 year olds (yesterday’s approval was “a big deal, a real big deal,” in Dr. Hirsch’s view), and Tidepool (Dr. Peters: “I love tidepool for looking at data, it makes my life so easy”). Abbott sponsored the session with an unrestricted educational grant, and though many of the cases focused on FreeStyle Libre, it was balanced on studies and honest product commentary. See below for highlights on some pretty unique cases – 670G in down syndrome, Medicare G5, recurrent DKA from daily marijuana use to treat neuropathy, death from severe hypoglycemia while on CGM – and more.

  • Several audience members mentioned FreeStyle Libre’s accuracy in Q&A, questioning the 9.7% MARD and noting over-reading hypoglycemia and sensor errors. Most of the questioners seemed to be from outside the US, which makes us wonder about the one-hour warmup outside the US vs. 12 hours in the US – it’s likely the US 9.7% MARD with a 12-hour warmup meaningfully improves real-world accuracy. Drs. Hirsch and Peters also rightly pointed out that a meter is a not a gold standard, making “true” accuracy impossible to measure in real-world use. We completely agree!

  • Professional CGM identifies an inaccurate offshore meter: An 85 year old man, with type 2 diabetes with an average glucose of 185 mg/dl, checking with fingersticks 2.5 times/day, and a lab-measured A1c of 11.8%. When he was put on CGM, the A1c was consistent with the CGM-measured average, far higher than the meter average. Dr. Hirsch discovered the Medicare-provided meter was reading consistently low!

  • Use of real-time FreeStyle Libre in type 2 diabetes. A 68 year-old man (retired surgeon) with type 2 diabetes for 20 years; on basal-bolus insulin, metformin, and empagliflozin; an A1c around 9% for 10 years; and coronary artery disease, hypertension, and dyslipidemia. After three months of FreeStyle Libre, his A1c dropped 1.6% (baseline: 9%) to an average glucose of 153 mg/dl and time-in-range of 72%. Dr. Hirsch: “I see this over and over. What did he do? He changed his food. He wasn’t eating bagels and drinking juice and cereals. Nobody is more excited about new medications than me. But as we’ve moved to new diabetes medications, we’ve lost sight of fact that diet and exercise has such a huge role” (This has also proved true anecdotally as we’ve shared Adam’s book, Bright Spots & Landmines, with many type 2s and people with prediabetes.)

  • MiniMed 670G use in a type 1 with down syndrome. Dr. Hirsch showed a case of using hybrid closed loop in a type 1 with down syndrome. Time in Auto Mode on the 670G was only 67% of the time; “I want him over 80% of the time, and would really like over 90% of the time.” Dr. Hirsch noted the 46% bolus, 54% basal split, offering the ability to strengthen the insulin:carb ratio (more aggressive). In his clinic’s experience, 670G users need only 40-45% as basal insulin. Dr. Hirsch also shortened the active insulin time, as we’ve heard others do.

  • Medicare-covered senior moves to Dexcom G5 once covered: This individual was on CGM and always had A1c’s in the 7%-range, but aged out when he went on Medicare – driving an A1c rise to 8.5%. On the trace from fingerstick data, he ran high overnight, given fear of lows without CGM. Dr. Peters got him on the Dexcom G5 Medicare as soon as it was available. He is now down to an average glucose of 146 mg/dl, with a much higher percentage in range and a higher basal rate overnight (more safety). Most importantly, he says, “I feel better. I feel like I’m in more control.”

  • Recurrent DKA from daily marijuana use to treat neuropathy (in hospital at least once a month). Dr. Peters shared the puzzling case of someone experiencing DKA recurrently, with no clear cause. She put the patient on CGM and ultimately concluded it wasn’t directly diabetes-driven DKA; rather, DKA episodes came from vomiting after daily use of marijuana to treat peripheral neuropathy. Once the patient stopped smoking, the vomiting stopped, the DKA ceased, and A1c came down to 6.8% This served as a cautionary tale for the audience: “It is helpful to look at old records and listen to patients.” What a puzzle!

  • Severe hypoglycemia death while on CGM. Dr. Peters shared this tragic case of a type 1 with “unbelievable variability…no matter how much we tried, we could not get her to give insulin in a way that was rational.” Her husband, who also has type 1 – (they met each other at diabetes camp), took on a lot of her management. The woman had 3-4 severe lows every week, and when her husband to leave for an overnight trip, the woman passed away from severe hypoglycemia while sleeping. Dr. Peters showed the CGM trace – a long period low, followed by a massive spike over 400 mg/dl (presumably a last-ditch counter-regulatory response), followed by death and a period with flat-line low glucose. Reflected Dr. Peters, “Obviously we cannot prevent all severe hypoglycemia and death Probably 1-2 times per year I have a patient that dies from severe hypoglycemia. This isn’t trivial – over-dosing on insulin. I cannot stop them. Even this woman, whose husband had taken over, could not leave her for long.”

  • Libre vs. Dexcom – how do Drs. Peters and Hirsch decide what to prescribe? For somebody who has clear hypoglycemia unawareness, Dr. Hirsch prefers Dexcom for the alerts/alarms and sharing (especially at night). He noted that “Nobody has ever compared hypoglycemia outcomes between the two,” though we’d note Dr. Nick Oliver’s iHART CGM study did do this – with Dexcom coming out on top. Dr. Peters does like alarms for those with lows, but lets patients decide – they are the ones wearing the device. She added that part of it is coverage – Libre is much less expensive for those paying out of pocket, but in California, it is “less well covered.” If patients are paying out of pocket, she uses Libre because it is less expensive.

2. Davida Kruger on Professional CGM’s Profitable Economics/Workflow ($750K annually); IDC's 9 Steps for Reviewing AGP with Patient

In a morning CGM session, Henry Ford Health System’s Ms. Davida Kruger and IDC’s Dr. Rich Bergenstal discussed the economics, infrastructure, and process of implementing CGM in a clinic. Ms. Kruger shared that her clinic grosses $750,000 in revenue from >1,400 patients using 50 FreeStyle Libre Pro systems and 50 Dexcom G4 Platinum systems. She also discussed some of the measures that her clinic takes to enable such a workload. Dr. Bergenstal gave a nine-step run-down of his approach for navigating an AGP

  • Ms. Kruger’s clinic is a professional CGM powerhouse: In 2017, it put >1,400 patients (~30 average starts per week) on professional CGM, grossing $750,000 in revenue (and with a positive profit). Wow! These stats really put a face to Dr. Etie Moghissi’s claim at AACE that CGM is “not a money-losing proposition” – Ms. Kruger even pointed out that endocrinologists can get paid more for CGM than for thyroid procedures, which is why it is important to stay on top of billing. Ms. Kruger herself sees 15-18 complicated patients every day, each on MDI or a pump – if they’re not, and just taking metformin, for example, then Ms. Kruger simply doesn’t have bandwidth to see them. This sort of complexity and volume hearkens to the importance of infrastructure for professional CGM workflow, which Ms. Kruger discussed in detail: Her practice has 50 Dexcom G4 Platinum and 50 FreeStyle Libre Pro systems, assigned dedicated resources of people and (organized) space, sign-out books for devices, secretarial support, a dedicated cleaning staff, EMR templates/smart sets for both Libre Pro and Dexcom G4 Platinum initiations, and set procedures to maximize efficiency (e.g., patient starts device in the clinic, drops off/mails the device 7-14 days later, medical assistant downloads the data, NPs do most of the interpretation, and then calls the patients). We would love to see experienced teams from clinics with established professional CGMs like Henry Ford’s do rounds at clinics looking to begin their own program – establishing this workflow is non-trivial and certainly went through a great deal of trial-and-error. Ms. Kruger also had a fair number of notable quotes in her talk; see the following bullets.

    • “For my money, if I put professional CGM on someone, I want them to see the data. It’s not my diabetes; give the patient more data!” She later specified that she lets the patient decide and will often recommend real-time CGM for people engaged in their diabetes and will benefit from seeing the data (also because the only real-time professional option available, G4 Platinum, requires fingerstick calibrations). At the end of the day, she pointed out that Libre Pro is great for some patients, while G4 Platinum is great for others: “When Libre Pro became available, we thought our Dexcom needs would go down, but it just doubled our number of total systems. The people we didn’t think Dexcom was right for could go on Libre Pro.”

    • “In 2018, reimbursement should not be the thing holding you back from wearing CGM.” This was great to hear from someone on the front lines! She cited Medicare coverage of therapeutic CGM, and hopes for Medicaid coverage in her native Michigan soon. The latter is quite spotty and something Dr. Anne Peters brought up today – California still doesn’t have statewide Medicaid coverage for CGM.

      • Regarding Medicare coverage, she recommended documenting requirements for CGM coverage – MDI/pump, four SMBG/day, etc. – in the charts of Medicare-age patients, even before requesting CGM. As a follow-up, she strongly advised against requesting CGM from Medicare until the patient has met the coverage criteria, because there’s a chance that after a first denial, coverage won’ be granted in the future.

    • “In my clinic, we move people from professional to personal – we want them to know what they’re getting so we do professional first.”  We love this approach, and wonder how many other clinics are reliably using it! It also suggests professional CGM will be a funnel for personal CGM, a big advantage for Abbott right now. Will Dexcom launch a professional version of the current G6 or wait for the disposable gen one with Verily?

    • “Though not approved by FDA, we use CGM in pregnancy or when women are trying to become pregnant.” Given that one in six experience hyperglycemia during pregnancy (IDF Diabetes Atlas), we firmly believe that more women should be on CGM to detect and more aggressively treat hyperglycemia as soon as possible in pregnancy.

  • Dr. Bergenstal outlined IDC's nine steps for interpreting CGM data side-by-side with a patient (see bullets below). He also overviewed a number of CGM rules of thumb, consensus guidelines, and hot topics: (i) At least 14 days of data is desirable for making treatment decision; (ii) the goal is to get glucose profiles “narrow, flat, and in-range”; (iii) GMI (glucose management indicator) has been proposed to FDA and the diabetes community as a replacement for eA1c – “stay tuned to see if it ultimately flies”; (iv) MiniMed 670G pivotal data – 72% time-in-range, 6.9% A1c, 3% <70 mg/dl, 25% >180 mg/dl, is a good starting point for targets; (v) There is a ~0.4% A1c drop for every 10% increase in time-in-range – “you need to know about that correlation, because patients will ask for it”; and (vi) Aim for coefficient of variation (CV) ≤36% (Monnier et al.) or standard deviation (SD) < mean glucose divided by three. See the following bullets for IDC's process of navigating an AGP report with a patient, and a sample marked-up AGP below:

    • 1: Check for adequate data (as close to ~14 days as possible). “Don’t spend a lot of time if there’s no data.”

    • 2: Mark up the AGP, noting factors that may affect the management plan. “Just print it out and write some stuff on it. Like to know a little bit – how old is person, do they have type 1 or type 2 diabetes, for how long, do they have known CVD, how much do they weigh (that helps with dosing insulin), and their eGFR. Include these little things right in front of me.” Dr. Bergenstal also asks the patient when they wake up, eat breakfast, lunch, dinner, snack, and put medications and other contextual information under the profile.

    • 3: Ask the patient, “What do you see?” Listen. Notably, ever speaker in the session underscored this approach. Said Dr. Bergenstal, “Usually you walk in and they’re already talking about it – they know why they went high at lunchtime.”

    • 4: Look for patterns of low glucose levels. Dr. Bergenstal usually “treats the cloud,” referring to the AGP’s lines denoting the lower 10% decile and the upper 90% decile of glucose. In 15 minute visits, he typically bases therapeutic adjustments on this AGP picture (especially if the 10/90 lines dip low), and will only verify with the daily snapshot reports if he has time.

    • 5: Look for patterns of high glucose levels.

    • 6: Look for areas of wide glucose variability. Dr. Bergenstal boiled variability down to timing and amount. Timing refers to insulin/meals, weekend/weekday, snacks, exercise, and stress; Amount refers to insulin, the insulin:carb ratio, and exercise intensity. “Anything to reduce the variability makes the other therapeutic changes more effective.”

    • 7: Compare to past AGP and reinforce successful strategies. “We never go back enough – we can show that we changed something and it worked. Maybe it also brought up a new issue, but we’re treating that now.”

    • 8: Get an action plan.

    • 9: Copy the AGP for the patient and the EMR.

3. Davida Kruger Speculation: “The Dexcom Professional CGM Will Change to No Fingersticks” (No Timeline or Hardware Info Shared)

In the midst of a Q&A, Henry Ford Health System’s Ms. David Kruger relayed speculation that Dexcom professional CGM will move to no fingersticks. Though we are not surprised by this news – given G6 – Dexcom has never discussed its professional CGM plans publicly (and, in fact, hardly ever talks about its professional CGM portfolio or pipeline). Ms. Kruger didn’t say more on when this iteration could launch, nor on the hardware configuration. It is possible that she could be referring to a professional CGM based on G6 hardware – with a reusable transmitter and disposable sensor – though we find it more probable that Dexcom would wait for the Verily gen one sensor (fully-disposable, potentially 14-day wear, with the G6 sensing platform). According to Dexcom’s 1Q18 call in May, the gen one Verily CGM is in “validation and verification.” A clinical trial of 14-day G6 wear in 2H18 could open the door for both an expanded indication for G6 (from 10- to 14-day wear), and for the gen one Verily sensor. There hasn’t been a firm timing update on Verily gen one in quite a while – it was last slated for a late 2018/early 2019 launch as of JPM in January. However, The more recent February call said timing depended on G6’s FDA path – G6 has now received clearance from FDA, but the push a 14-day G6 study will need to be conducted first. We look forward to hearing public updates from Dexcom on their next professional CGM play, since this side of their portfolio hasn’t been updated for the better half of a decade (in stark contrast to the personal CGM side).

  • A factory-calibrated, 14-day, fully-disposable Dexcom professional CGM would have a similar profile to and could be very competitive with Abbott’s FreeStyle Libre Pro. Dexcom’s future professional product might still be toggle-able between real-time and blinded and might still have alarms, which would differentiate it from Libre Pro. At the same time, Abbott is working on a real-time CGM with alarms to incorporate into Bigfoot’s Loop automated insulin delivery system, and we could envision them building out a personal and professional platform off of this base. The biggest opportunity for Dexcom in advancing their professional CGM pipeline is to decrease cost, which many attending CDEs/NPs cited during Q&A as a reason for putting their patients on Libre Pro. Ms. Kruger noted that her clinic receives transmitters for free because of the bulk of their purchases (and can often stretch them for up to a year), but G4 sensors still cost ~$20 more than Libre Pro sensors – the lower-cost Verily platform could push Dexcom closer to parity with Abbott’s pricing and change prescribing dynamics drastically.

    • An OUS launch of Medtronic’s Envision Pro (rebranded from “iPro 3”) is expected by April 2020 per the company’s recent analyst meeting. The 2016 analyst meeting expected this professional CGM to be single-use, blinded, clamshell transmitter, and with a MARD of ~11%; we’re not sure if that is still the case.

Diabetes Mine D-Data Exchange

1. One Drop to Launch 12-Hour Glucose Prediction for T2 Non-Insulin Users In 3Q18 (Very Cool) 860,000 users globally

One Drop shared bold new plans to launch  “Automated Decision Support in 3Q18 for type 2 users not on insulin: a forward-looking 12-hour glucose prediction. Wow! The interface looks very slick and focuses on whether predicted glucose is in line with a user’s goals. As the plot below shows, the app will give a 12-hour, forward-looking, colorful glucose trace based on recent history – the predictions are based on fingerstick data, food logging, exercise data, and more. One Drop is leveraging machine learning models and metabolic simulations. Notably, the presenter was One Drop’s Head of Data Science Dr. Dan Goldner, who previously worked at NASA creating simulation models for multi-year strategic problems! We’ll see a poster on the model’s accuracy tomorrow (46-LB).We love this concept of data-driven decision support for a population that does not have much data to learn from – assuming it is accurate, of course. This could also be highly motivating and behavior changing, in the same way Loop’s predicted glucose is for DIY closed loop users. (“Oh wow, I’m going to be really high in the next few hours. Maybe I should eat fewer carbs at dinner.”) One Drop’s model needs just a single fingerstick to get started and continually adapts its 12-hour forecast based on the inputted data. This is certainly a big run at Medtronic/IBM Watson’s Sugar.IQ, which is only designed for Medtronic CGM users and does not yet give this level of granularity around glucose prediction. One Drop also gives paired lifestyle-focused tips – e.g., walk after meals – which the company says excludes it from FDA regulation at this time. At first the insights and recommendations will come as notifications with frequency controlled by user (second picture below). Over time they will get smarter, as One Drop collects more data about what users like. The feature “is coming for people on insulin,” though no timing was shared; we assume this will require some regulatory effort. We’d note Medtronic’s plan to launch forward-looking glucose prediction with Sugar.IQ is more than two years away (per its Analyst Meeting), meaning One Drop could be quite ahead on this, especially if it can accelerate the incorporation of insulin and CGM.

  • One Drop now has >800,000 users, >50 million blood glucose readings, and >1 billion data points. The user base continues to rise from 650,000 last November and 750,000 in March.

2. SOOIL to Submit to FDA with Open Protocol Dana Pump Designed Around Needs of DIY Community; Potential OpenAPS Inclusion?

SOOIL Korea’s Justin Walker shared bold plans to submit the company’s smartphone-controlled Dana RS insulin pump to the FDA/CE Mark with an open communication protocol – an interoperable pump to meet the wishes of DIY users. He also mentioned plans to register a version of the open source OpenAPS algorithm (of Dana Lewis/Scott Leibrand fame!), implying in Q&A it would sit in a phone app and communicate with the Dana pump and an iCGM. SOOIL actually submitted interest to JDRF’s Open Protocol Initiative (which Roche is also part of), and this bold plan could be a big asset in the US market if the company can figure out the regulatory path – an open pump that patients can buy brand-new for use in systems like Loop and OpenAPS could be compelling! The timing on FDA submission was a bit unclear, though SOOIL hopes to release the open protocol pump “shortly” – it had a few minor design updates, but looked mostly similar to the already-available Dana RS. As we’ve previously noted, Dana RS is the only pump to our knowledge with approved smartphone bolus apps for iOS and Android. The pump is no-frills, but has become a popular choice in the DIY community outside the US, as the direct smartphone control means no communication relay device or “rig” is needed (e.g., AndroidAPS users simply run the closed loop app on the phone and wear the Dexcom and Dana RS pump). Dana’s RS pump is notably CE Marked and Korean FDA approved and available in both Europe and Asia. Interestingly, SOOIL has been building pumps since 1979 and its move to smartphone control was driven by the DIY community, including Mr. Walker’s own experience.

3. FDA’s Dr. Courtney Lias on iCGM pathway, AID integration

FDA’s Dr. Courtney Lias reviewed the exciting integrated CGM (iCGM) 510(k) pathway that came with Dexcom’s G6 clearance, including clarification on how a pump company will integrate multiple iCGMs – e.g., Tandem. Her comments echoed JDRF/HCT’s AID Interoperability Meeting, but they took on additional relevance after yesterday’s approval of Tandem’s t:slim X2/Basal-IQ as the first pump with iCGM compatibility (only Dexcom’s G6 for now). We’ve been wondering: if Senseonics obtains iCGM designation for Eversense, what is the process for Tandem integrating that sensor into Basal-IQ? Dr. Lias clarified today that when an iCGM is cleared in the future, Tandem would evaluate if they want to claim use with it, perform whatever communication protocol/development work is needed, ensure its AID algorithm will work with the iCGM, and add the new iCGM to their labeling. Notably, the FDA will NOT have to get a new submission for this iCGM addition – Tandem can add new iCGM sensors to their pump without regulatory submission at all! In other words, it won’t quite be plug-and-play, as the onus is on the pump company to integrate the iCGM device and get the labeling – and in this case, Tandem would have to do some work, since G6 is a no-calibration system, Eversense requires two per day, and the sensor wear lengths are obviously 10 days vs. 90 days. (Presumably Senseonics and Tandem would have to work together in some capacity, but would not need a complicated contract and no coordinated PMA submission.) Dr. Lias emphasized that the iCGM path does not require factory calibration, and it does not mandate that a company connect with other systems – e.g., if an iCGM company wants to remain closed and doesn’t want to connect to another company, it doesn’t have to.

  • Dr. Lias concisely summarized the benefits of the iCGM pathway, which aims to accelerate innovation, reduce regulatory hassle, and enable patient choice. Separating the iCGM from the AID or other compatible device system will make upgrades/modifications more efficient and reduce the need for duplicative regulatory submissions. It will also allow connected systems to be updated more quickly and with a predictable process. Notably, it works for many different business models, as companies can choose to be open or closed. Last, it will reduce the contract hassle that stalled many pump/CGM regulatory efforts in the past. For more complete comments, see our previous piece on iCGM here and Dr. Lias’ JDRF/HCT talk here.

  • Related to the FDA De Novo clearance of DreaMed’s Advisor Pro software for optimizing pump settings, FDA has posted the Special Controls for the new category of “Insulin Therapy Adjustment Device”read them here. They are much less specific than the iCGM special controls, mainly focused on properly documenting inputs and outputs, software validation and verification, etc.  

4. Product Demo Highlights: Ascensia Challenge winner Whisk (AI nutrition platform); Spike CGM app floors audience; Many More

As always, Diabetes Mine’s D-Data event included many product demos; we include brief highlights below.

  • Ascensia announced its Diabetes Innovation Challenge winner Whisk – a cool UK-based company with an AI-driven nutrition platform and “Culinary Coach.” As winner of the challenge (cash prize: €100,0000), Whisk plans to create a personalized food experience for people with type 2 diabetes that will learn from their blood glucose readings and make tailored diabetes food recommendations – e.g., what meals spike blood sugar more than others. Nice! Whisk is “the leading food AI player globally” and its technology is used by many companies – powering ~100 million recipes on various sites and half a million shopping lists every month at grocery retailers including Walmart, Tesco and Amazon Fresh.  The company has a “food genome” that can break down foods’ nutritional content and flavor, as well as retailers’ store items and prices. Whisk will build the food capabilities into the Contour Diabetes app, aiming to take friction out of planning, buying, and finding food (focus on type 2). Users will be able to share their diet preferences and receive tailored recipe recommendations, a simple shopping list to make those recipes, and even add the items to an online grocery shopping cart to buy them.

  • The open-source, free Spike CGM app wowed the DIY audience and patients in the room, but also alarmed the regulators. The free app aims to help users “Get the Most Out of Your CGM Transmitter.” It connects directly to the Dexcom G4/G5 and add-on devices for FreeStyle Libre (e.g., Ambrosia BluCon transmitter), bringing some pretty bold features: no seven-day sensor shutdown for Dexcom, no 112 days hard G5 transmitter shutdown, no two-hour warmup periods without data, no ??? sensor output, and it can display raw uncorrected glucose values. For FreeStyle Libre, it can extend the sensor life and can display values even from warm-up periods. Users can also input carbs and insulin and calculate insulin on board, carbs on board, and get absorption curves. The app can also speak glucose readings, an advantage for the visually impaired and athletes; one surgeon even uses it in the US while he is doing surgery! Spike was released to the open source community recently (>6,000 users). It is available via Test Flight in beta (only an email address needed to download app), though is most definitely not FDA approved and most definitely counts as a medical device. (Dr. Lias’ seemed perturbed on seeing this demo and spoke to the gentleman during the break.) That said, this app could be a good model for industry of where CGM apps could go in terms of data collection, data display, decision support, and more. The visual design is a bit clunky, but see the features here, which were widely praised by patients in the room.

  • D-Data had a few other demos, including AADE Dana platform (to inform educators about diabetes tech; launching in August), Novo Nordisk/Glooko’s Cornerstones4Care app (lightly criticized by Cherise Shockley and Dr Jeremy Pettus as to what it adds over existing apps), Metronom’s Health’s CGM (second round of human trials end of this year, regulatory submission in early 2020 – delayed at least six months from the ATTD timing), Senseonics’ Eversense (approved yesterday, and the one update today was a confirmation it is approved on both Android and iOS), Mytonomy (microlearning via video; presented by Deb Greenwood), and LoopDocs (the wiki-how on setting up the DIY system, Loop).

5. Standing Ovation for Dr. Anne Peters Access Work at D-Data Exchange

Dr. Anne Peters received a standing ovation for her tireless, two-year crusade against the State of California to obtain a CGM for a homeless type 1 with diabetes complications and significant hypoglycemia. The story will be included in our full report (truly remarkable), reflecting the tireless, relentless champion that Dr. Peters is. Notably, we learned that Dr. Peters and colleagues have produced guides to help low-literacy individuals learn about diabetes technology. At the end of her talk, she dramatically held up a huge shopping bag full of thumb drives with low-literacy manuals for diabetes tech; we’re excited to dig into them and will report back in our full report with more.

Oral Presentations

1. Compelling FreeStyle Libre Pro interim intervention: Increases Time-In-Range by ~9 Hours/Day (42% to 80%) in Type 2s in India

Dr. Akshay Jain (LMC Diabetes & Endocrinology) detailed the successful implementation of an interim intervention technique (IIT) using Abbott’s FreeStyle Libre Pro in India. Given limited resources, the investigators sought to determine if one intervention during a single 14-day blinded professional CGM session could drive patient outcomes – think of this approach as a cheaper alternative to running one professional CGM, contemplating, then running another. 105 type 2 adults with A1c >7% – on oral agents and/or insulin – and no previous history of CGM use were recruited. During the first visit, participants were evaluated by an endocrinologist and initiated with FreeStyle Libre Pro. Participants returned within one week for a data download (i.e., halfway through the 14-day sensor wear). Individualized dietary and pharmacotherapy modifications were discussed, and patients were shown how specific food choices correlated with their glycemic excursions. Less than a week later, participants returned for a final evaluation once the 14-day wear was up. Incredibly, in just 14 days, all outcomes improved significantly: Daily average glucose dropped from 191 mg/dl t0 137 mg/dl; time-in-range increased from 42% to 80% (+9 hours/day); time <70 mg/dl decreased from 6% to 1% (-1 hour); and time >180 mg/dl decreased from 52% to 18% (-8 hours). Importantly, ITT seems to be particularly effective in reducing hypoglycemia. A sub-analysis of 27 patients identified to have recurrent episodes of hypoglycemia pre-IIT revealed the intervention to result in a significant increase in time-in-range (66% to 87%; +5 hours) and a dramatic decrease in time <70 mg/dl (21% to 2%; -4.5 hours), without any significant increase in hyperglycemia. These results are downright spectacular, and show the incredible potential of low-cost CGM used in clever ways to change the lives of people with diabetes in resource-poor (and rich) locales. Could this sort of intervention be deployed at the level of primary? We wonder how durable the improvements were, or if patients would require follow-up CGM sessions to regain sound glucose management habits. Dr. Jain did mention that many participants managed to maintain their dietary recommendations because of the visual impact from the CGM tracings.

  • How many of the patients were on MDI? How many were just on basal insulin? This question won’t exit our minds, as we look at the incredible 80% and 87% time-in-range figures. Our instinct is to say that most of the patients were on orals, but then again, some were experiencing five hours of hypoglycemia every day (or is this Libre Pro overreporting?). One possible explanation is that many of the individuals were over-prescribed insulin or sulfonylureas at first and were down-titrated by their physicians.

  • Dr. Jain acknowledged that many patients were prevented from signing on for a repeated trial due to financial constraints – in India a doctor’s visit alone ranges from $5-$15, and the FreeStyle Libre Pro costs $40 (we’re not sure if the patient is responsible for the entire cost). Based on the huge improvement in glycemia, we see this as an approach ripe for philanthropic investment (after confirmatory trials). What if, for example, every pregnant woman had a single Libre Pro and clinic visit early on in pregnancy to set her on a path to avoid hyperglycemia?

2. Sugar.IQ Improves Time-In-Range by 36 Mins/Day, Time >180 by 30 Mins/Day, Time <70 by 6 Mins/Day; Sugar.IQ+Fitbits; Impressive Glucose Prediction Feature Data; Logic Behind Sugar.IQ Insights

In a highly-anticipated oral, Medtronic Global Head of AI and Digital Health Dr. Huzefa Neemuchwala shared the latest batch of Sugar.IQ data from 256 530G/Enlite + MiniMed Connect users between April-August 2017. Relative to baseline metrics, Sugar.IQ conferred a 36-minute/day improvement in time-in-range (+2.5%-points), a 30-minute/day decrease in time >180 mg/dl, and a 6-minute/day decrease in time <70 mg/dl (all statistically significant). Though these are fairly modest by closed loop standards, the Medtronic press release notes the impact: 36 minutes per day translates to over nine additional days per year in-range! Participants in the study also experienced 1.22 fewer high episodes (>180 mg/dl for >120 minutes) per month and 0.95 fewer low episodes (<70 mg/dl for >20 minutes) per month. Dr. Neemuchwala noted that all of the users benefitted from 530G’s low glucose suspend feature through the baseline and experimental arms of the study, likely masking what could’ve been a more marked improvement at the low end of the glucose range. The implication is that MDI users who will be using Sugar.IQ in the commercial launch (with Guardian Connect) may see greater reductions in hypoglycemia; we agree. During the course of the 31+ patient-years of use, Sugar.IQ generated 655 insights related to hypoglycemia and 699 related to hyperglycemia. Notably, 231 of the 256 (~90%) users recorded at least two weeks of data, demonstrating a solid pattern of engagement, though we wonder what the engagement curve looks like after that point (at least ~30% of all engagement with the app came within those first two weeks). For context, the time-in-range data is almost identical to the 33-minute-improvement in an n=136 subset of users, presented at DTM – time in hypoglycemia and hyperglycemia have never been presented before, and episodes of hypoglycemia/hyperglycemia have previously been reported as percent reductions, so direct comparisons aren’t possible. For the limited learning launch of a novel app, we consider these data to be very encouraging, and we see big potential for Sugar.IQ to drive Medtronic’s Guardian Connect CGM marketing and outcomes in the real world. More details and screenshots from the presentation – including findings from giving some of the 256 Sugar.IQ users Fitbits, plus the glucose prediction feature in the pipeline – below! Medtronic and IBM Watson posted a press release immediately following the presentation, announcing the commercial availability of Sugar.IQ with the Bluetooth-enabled standalone Guardian Connect CGM; see our coverage of the launch from last week here.

  • Medtronic randomly gave 134 Sugar.IQ users Fitbits during the course of the study, and discovered that glucose responses to meals and activity vary greatly. Each participant wore CGM for 50 days, wore a Fitbit for 47 days, and manually logged food for 12 days. Not surprisingly, investigators found that when people had a recent meal and then exercised for a median 45 minutes beginning with glucose in-range, there was an assortment of glycemic responses: (i) In 799 cases, people remained flat; (ii) In 371 cases, glucose rose slightly (mean ~140 mg/dl) toward the end of the exercise session; (iii) In 237 cases, glucose went above range (mean ~200 mg/dl) at four hours; and (iv) In 163 cases, glucose went above range (mean ~220 mg/l) at two hours. This data demonstrates the importance of personalization, as glucose responses vary greatly. We’re curious if a given individual is likely to stick to the same response to exercise or jump between the four “bins” depending on internal and external variables, and how big an impact the type of exercise has on glucose response (it could be that those who go above range are doing intense anaerobic workouts or under bolused).

  • Dr. Neemuchwala showed that users “like” insights more over time, a testament to Sugar.IQ’s ability to tailor them. In the graph below, the first insights are liked by ~80% of users, while by the 18th insight, ~100% of users like them. We found the examples of helpful vs. not-helpful tips (as “liked” or “disliked” in the app) to be particularly interesting. Users found “I see that on days when your sensor glucose is [between 80-100 mg/dl] at dawn, you tend to stay in target more all day. I like this pattern!” and “I see that you tend to spend more than 80% time in target [between 3-6 pm]” helpful. On the other hand, “I notice that [between 3PM-6PM], you tend to stay in target longer when you’ve spent [more than 75%] time in target over the previous three hours. Way to keep up a positive trend!” and “I notice that when you take [1-2] bolus(es) [between 6PM-9PM], you tend to spend more time in range. That’s what I like to see.” At first glance, there is not a drastic difference between the two sets of insights, particularly because they are all pats-on-the-back, but the differing responses could be chalked up to: (i) the most useful insights don’t require user action, while the least useful insights ask the user to keep up a good behavior or (ii) the least useful insights are more specific and less simple. It could also be that the n=256 users comprise too small a data set, and therefore the different ratings are an artifact. Regardless, it’s awesome to see this be a big focus of research and thought. Dr. Neemuchwala stated, “We don’t want to overload the individual with a lot of insights. It’s not hard to come up with them, they’re basically just correlations. The hard part is to figure out which part to deliver at what time. If I’m driving and then I go low, telling me that is more relevant than telling me what happened three days ago.” He noted that the logic behind who gets what insight and when is “fairly complex and rule-based”; it depends on glycemic acuity (how urgent is the insight?), starts simple and then works up to more complex insights, tailors to user preference and feedback, avoids repeating insights, and is internally consistent with its insights (i.e., not contradicting itself).

  • Dr. Neemuchwala gave two awesome case studies of Sugar.IQ in action. In the first, a woman discovered a “problem” meal – “natural peanut butter spread low sodium” – tracked the meal using the glycemic assist feature, and learned how to manage it (image below). In the second case study, a 63 year old male who had had type 1 for 41 years found out through Sugar.IQ that his meals were hyper-loaded with carbs on Fridays (40-55 grams per meal), which was causing him to spend 71% of the four hours following a bolus above range. After receiving the insight, the man began logging lower carb meals on Fridays (20-30 grams per meal) and reduced his time spent above range following a bolus to 39%. These cool anecdotes show the power of Sugar.IQ, but crucially depend on users manually entering food data – how will patients engage with the app in the real-world, beyond the limited learning launch? We’re excited to see – we could imagine many coming back to the app repeatedly to see what’s new!

  • Dr. Neemuchwala showed new mock-up screenshots illustrating how the Hypoglycemia Prediction feature (within four hours) could look in-app. We love the sample alert, which reads: “Now 110 mg/dl. You are EXPECTED to experience a low within the next 4 hours.” As for prediction accuracy, prediction accuracy is impressive with 90% for two hours in advance, and 84% for four hours in advance, on average (based on CGM data from 10,000 type 1 users; measured as ROC AUC). These numbers seem like they would align with Medtronic’s prior claim of ~90% hypoglycemia prediction accuracy. Per the recent analyst meeting, Sugar.IQ with Hypoglycemia Prediction is set to launch within the next year (by April 2019).

  • During an analyst meeting earlier this month, Medtronic announced that it would launch forward-looking glucose prediction for MDI users “beyond” April 2020 – Dr. Neemuchwala detailed “Glucose prediction research,” which we assume to be the same feature, today. The machine-learning model was trained on CGM data from 60 type 1s, tested on CGM data from 20 type 1s, and validated on CGM data from another 20 type 1s. Every five minutes, the glucose prediction algorithm predicts glucose four hours into the future. In the first set of data, researchers pitted the algorithm’s glucose predictions for the next four hours (from a set moment in time) against the actual readings from modern CGMs – on the Clarke Error Grid, the algorithm was able to lace 95.57% of all estimations in zones A and B, indicating good predictive power – though we wish Zone A had been broken out, since it likely wouldn’t meet where Zone A of current CGMs are at. Of course this feature won’t look too foreign to the DIY community, who already uses algorithms to prognosticate future glucose levels, but we’re excited to see Medtronic investing in bringing this game-changing technology to the masses. One Drop separately announced a 12-hour forward-looking glucose prediction for fingerstick users with type 2 not on insulin; see below for more on that.

3. 180-Day Senseonics Eversense XL MARD of 9.4% vs. YSI in Primarily Pediatric Population (n=30 Adolescents, 6 Adults); 83% Values Within 15/15% of Reference

Dr. Ronnie Aronson (LMC Diabetes & Endocrinology) presented full results from a study of Senseonics’ 180-day Eversense XL CGM conducted in a primarily pediatric population (n=30 adolescents, 6 adults), headlined by a very strong MARD of 9.4% vs. YSI. The mean age in the study was 17 years overall; the average adolescent age was 14 years, and the average adult age was 32. Importantly, accuracy showed a “nice consistency,” with no degradation over the very impressive six-month duration (see below). 83% of the readings were within 15 mg/dl (in hypoglycemia) or 15% (in hyperglycemia) of the reference value, and 99.6% of readings were within the A and B sectors of the consensus error grid overall. Participants demonstrated a median transmitter wear time of 23 hours/day, showing very strong adherence to the unique form factor. As Dr. Aronson pointed out, given that the transmitter is easily removed, the high wear time provides strong evidence that the device is comfortable and perceived as useful by patients. To this end, 82% of participants “agreed” or “highly agreed” that the Eversense XL was easy to use, and 90% “agreed” or “highly agreed” that the mobile app was easy to use. The vast majority of participants (90%) indicated that they liked the ability to display glucose readings on a smartphone – there is no dedicated receiver for the Eversense receiver (smartphone-only display). Importantly, no serious adverse events or infections were reported. A few mild skin reactions occurred but were resolved within two to eight weeks of sensor removal; one patient experienced a limited skin reaction to the adhesive. Preliminary results presented at ATTD showed a solid 78% of the sensors made it to the full six months without having to be extracted. Of the eight sensors that were removed prior to 180 days: one participant withdrew consent one day after insertion; one participant’s sensor was removed at 93 days due to connectivity issues; and there were six early sensor replacement alarms after day 130. We’re excited to see the first signs of expanding Eversense into pediatrics, as we think this population could massively benefit from the unique features and form factor of the device. It’s currently unclear where the Health Canada regulatory submission process stands (for adult and pediatric), as well as whether Senseonics intends to pursue the 180-day indication directly in this geography. 

  • As a reminder, the 90-day Eversense was just approved by the FDA yesterday for adjunctive use in adults with two fingerstick calibrations/day. The 180-day Eversense XL received its CE mark in September and is now available in all OUS markets except South Africa. Senseonics plans to have the entire existing userbase converted to the XL by the end of the year and expects to begin recruitment for a US XL clinical trial this summer. The jump from 90-day implantation to 180-day implantation (plus “reduced calibration” in the US) could be very meaningful for patients, particularly young ones.

4. Dr. Irl Hirsch’s Group: 33.5% of Patients in Study (N=1,039) Have “Discordant” Avg Glucoses That Don’t Align with Expected A1c-Derived Avg Glucose (Using ADAG Formula)

University of Washington’s Dr. Jordan Perlman, a pupil of the great Dr. Irl Hirsch, presented a very interesting retrospective analysis on a group of 1,039 patients, finding that 33.5% had “discordant” average glucoses. “Discordant,” in this case, means that the patients’ CGM/SMBG-derived average glucoses differed from the average glucose derived from their lab A1cs (using the ADAG study as a reference) by at least 15%. A small portion of this discrepancy can be attributed to comorbidities previously proven or hypothesized to invalidate A1c, such as stage of CKD (GFR <45 increases the odds of ADAG discordance) or NAFLD (trend toward greater discordance in individuals with NAFLD). However, in the study population that had no comorbidities (n=682), 28.7% of individuals still had discordant mean glucoses. What could be causing this discrepancy? Dr. Perlman didn’t delve into all of the possible causes of discordance – that could take all day – but did detail two. The majority of patients studied were on SMBG (n=756), while the remainder were on CGM (n=283). 18% of CGM-derived mean glucoses were discordant, while 39% of SMBG-derived mean glucoses were discordant, suggesting that CGM decreases the odds of ADAG discordance. We’d a likely reason: CGM captures more postprandial and overnight values and better reflects true average glucose. The authors also broke the population down by type of diabetes. Of the 681 type 1s in the database, 29% had discordant mean glucoses, while 42% of the 319 type 2s had discordant mean glucoses – type 2 diabetes increases the odds of ADAG discordance. Dr. Perlman cautioned against relying on A1c alone to make treatment decisions because it’s unclear when discordance gains clinical relevance – in what patients and at what level? She proposed that professional consensus is needed to define acceptable discordance before ADAG-derived mean glucose is expanded. We happened to identify the most with one of her off-hand remarks: “Maybe there’s more of a problem with our test, and less of a problem with the individuals taking the test.” Comorbidities were excluded from the ADAG study because authors didn’t want to skew the results of the study, but these newer data suggest comorbidities are only part of the story, along with “LIMITLESS” potential for other confounders. Add this well-done analysis to the list of reasons the field is going beyond A1c to consider glycemic metrics like time-in-range, hypoglycemia, hyperglycemia, and glycemic variability.

Decision Support and Additional Technology Sessions

1. UVA CGM-Based Decision Support for Type 1s Significantly Reduces Glucose Variability (CV 36->33), Time <70 md/dl (3%->0.9%) in N=24 Pilot Study

UVA’s Dr. Marc Breton fleshed out promising pilot data from a study of a CGM-based decision support system based on the UVA DiAs platform in 24 type 1 adults (16 pumpers, 8 on MDI). [Dr. Bruce Buckingham flashed the summary figure from this pilot – shown below – at DTM]. In its current form, the system is composed of three main modules: automated treatment parameters adaptation (recommends changes in therapy based on retrospective risk zones), insulin sensitivity-informed bolus calculator (adjusts boluses based on ratio of real-time insulin sensitivity to historical insulin sensitivity), and exercise advice (determines whether patient should snack/wait or proceed with exercise depending on age, insulin on board and initial glucose). Through the ~four-week study, which included clinic-based meals and exercise, the decision-support system significantly improved a number of glucose parameters relative to standard of care: Primarily, coefficient of variation fell from 36% at baseline to 33% (34% -> 30% at mealtime), largely from less hypoglycemia, as time ≤70 mg/dl decreased by ~33 minutes per day (from 3.2% to 0.88%). Time in range (70-180 mg/dl) and time above 250 mg/dl did not significantly change from standard care, a surprising finding (potentially because of the small study size). The plot below of “quality” glucose – average vs. time below 70 mg/dl – does show a much tighter band in the treatment group towards less hypoglycemia and a lower average. Notably, this decision support system is in the midst of a three-month, n=120 study at Stanford, UVA, and Mt. Sinai to explore applications in MDI – we first heard about this study, which is using Novo Nordisk-supplied smart insulin pens, at DTM 2017. We look forward to two preliminary posters from this study group at ADA, including one on insulin pen priming trends and a second on late and missed meal boluses. The team will also conduct studies to parse out the impact of each decision support module during 2018, attempt to untangle the impact of information without support vs. prescription (“take X units of insulin”) and their acceptability to patients, and extrapolate the system to type 2s. 

  • Dr. Breton provided granular data from the in-clinic days assessing the effectiveness of the decision support in helping participants handle exercise and meals. When individuals exercised two hours post-meal, the system (red) appears to greatly reduce hypoglycemia; three hours post exercise, it appears to prevent early hypoglycemia, though may put the user at slightly higher risk of hypoglycemia starting ~40 minutes after beginning activity. For sessions when individuals began exercise with blood glucose <180 mg/dl, Dr. Breton remarked that the glucose traces resemble those from hybrid closed loop – significantly less hypoglycemia, and significantly less variability. Regarding postprandial excursions, the decision support system improved patient handling of high fat/protein meals, without significantly impacting handling of meals overall or large meals. In the postprandial excursions slide below, a distribution to the left (lower area under the curve) is favorable – higher red bars on the left indicate lower and/or shorter duration hyperglycemic excursions after a meal. We’ve never seen meal data plotted as histogram of AUC; a simple time-in-range metric might have summarized it better?

  • The other modules to be included in the decision support system are: Bedtime advice, average glycemia indicator, and hypoglycemia prediction (up to three hours in advance). We’re not sure if these are incorporated into the system currently in evaluation at UVA, Mount Sinai, and Stanford.

    • Dr. Boris Kovatchev, Dr. Breton’s mentor, showed impressive data from the hypoglycemia prevention module at DTM.

2. OHSU AID/Decision Support Work with Exercise/Meals; T1-DEXI Pilot Study to Inform n=300-500 Study; “DailyDose” Decision Support for Type 1s on MDI Funded by HCT, Co-Developed with Dr. Cafazzo

OHSU’s Dr. Jessica Castle gave an update on her team’s work with closed loop and automated decision support to improve adaptation, including the Helmsley Charitable Trust-funded, Jaeb-coordinated T1-DEXI pilot study. The pilot study will enroll 60 individuals between ages 15-70 with type 1 diabetes, and collect one month of insulin, CGM, food, and physical activity data. Data will be collected with Dexcom G5, DiabNext’s Clipsulin dose capture device (an uncommon choice), a Garmin activity tracker, and a custom app developed at OHSU for food photos and exercise logging. The app asks patients to take photos before and after eating, estimate carbohydrate content, and provide a qualitative estimate of how much protein and fat is consumed. Participants will be randomized to complete two in-clinic and four home sessions of either aerobic, anaerobic, or high intensity interval exercise. Every seven days, providers will review CGM and insulin data, and make insulin dose recommendations. This study will eventually inform a larger study of 300-500 subjects with a goal of building better exercise and food models for automated insulin delivery and decision support.

  • Dr. Castle detailed a couple of other current and future OHSU studies in the domains of dual-hormone closed loop and decision support:

    • (i) The team is in the midst of a three-day outpatient study with in-clinic aerobic exercise (n=20) assessing a dual-hormone closed loop system consisting of Insulet Omnipod and Dexcom G6. Notably, this study is using Xeris glucagon, which Dr. Peter Jacobs previously told us is stable for the duration of the study and does not need to be changed. The system will also be informed by OHSU-developed hypoglycemia prediction and postprandial hyperglycemia prevention algorithms.

    • (ii) “DailyDose” is a smartphone-based decision support system for type 1s on MDI therapy that provides bolus, basal, and behavioral recommendations. The application was co-developed with University of Toronto’s Dr. Joe Cafazzo and funded by the Helmsley Charitable Trust. No further details were shared on this platform, including study timing, blood glucose input (CGM, SMBG, both?), other data inputs, etc.

  • Dr. Castle quickly reviewed three OHSU-developed algorithms, two for hypoglycemia prevention, and one for minimizing postprandial hyperglycemia. The first hypoglycemia prediction algorithm is a heuristic algorithm that uses heart rate and glucose at the start of exercise to help patients predict hypoglycemia risk in real time. The second hypoglycemia prediction algorithm is a “random forest algorithm,” meaning that it runs through a group of decision trees and the outcome of each tree is averaged to get a higher predictive accuracy. Inputs for this second, more complex algorithm, include glucose and heart rate at the start of exercise, insulin on board, average TDD, sex, weight, height, and BMI. The algorithms were trained on 154 exercise observations (some closed loop, some open loop), and were validated in 90 individuals: The first heuristic algorithm more accurately predicted hypoglycemia than the decision tree algorithm (86.7% vs. 79.6% accuracy), driven by greater specificity. Regarding meals, the OHSU team is developing ALPHA – adaptive learning postprandial hypo-prevention algorithm.

3. Imperial College London Bolus Calculator (ABC4D) Auto Adjusts Insulin:Carb Ratio Based on Last Week’s Post-Prandial Outcomes; Integration into Hybrid Closed Loop System to be Tested in RCT (n=20)

Imperial College London’s Dr. Pau Herrero-Viñas detailed the Advanced Bolus Calculator for Diabetes Management (ABC4D), an insulin bolus calculator that uses Run-to-Run (R2R) control and Case-Based Reasoning (CBR) to automatically adjust patients’ insulin:carb ratios depending on the post-prandial outcomes of the previous week. The ABC4D patient-facing app and clinician platform was tested in a small pilot study (n=10) of adults with type 1 diabetes on MDI. Participants used the ABC4D smartphone app for six weeks in their home environment, returning to the clinical research facility weekly for data upload, revision, and adaptation of the CBR case base. Although Dr. Herrero-Viñas alluded to a trended reduction in postprandial hypoglycemia and improvements in other glycemic outcomes, no changes were found to be significant. Still, he was encouraged by the results, believing the non-significant outcomes to likely be the result of the small sample size. Since the study’s publication in 2016, Dr. Herrero-Viñas and his team have designed an updated ABC4D app intended to be more user-friendly: all features are accessible on the patient’s smartphone (i.e. no need to go to the clinic for data download or algorithm revision), and CGM integrates directly with the bolus calculator to account for metrics like rate of glucose change. A non-inferiority, crossover RCT investigating the new version is currently being planned in a larger study population – Dr. Herrero-Viñas displayed a picture of the user app with a Dexcom G6, but we’re not sure if that is the only CGM that can be integrated. Dr. Herrero-Viñas’s team is also planning to incorporate the bolus calculator into a hybrid closed loop system called “Biap.” The slide introducing Biap depicted Dexcom’s G6 and Tandem’s t:slim X2 pump, although these devices were not directly named during the presentation – this name is new to us! An open-label, randomized study (n=20) with three arms (sensor-augmented pump vs. Biap with standard bolus vs. Biap with adaptive bolus) is in the works.


4. A Diabetes Specialist’s Message to Primary Care Providers: “CGM is Here to Stay,” Will Replace BGM in Next Five Years (Particularly for Those on Intensive Insulin Therapy)

In a very practical session aimed towards educating primary care providers on current diabetes tech offerings, Dr. Jim Chamberlain (St. Mark’s Diabetes Center) emphasized that “CGM is here to stay” and has “taken over the type 1 diabetes world.” In fact, Dr. Chamberlain opined that CGM will likely replace BGM “completely” in the next five years, especially for those on intensive insulin therapy. While we firmly believe CGM is the new standard of care and are elated to see uptake increase in those with type 1 and type 2 diabetes, things will need to accelerate quite precipitously in the next five years to completely replace BGM – even in the US. (T1D Exchange CGM penetration stood at 24% as of data last fall at the best US type 1 centers, though with Libre and Guardian Sensor 3 CGM penetration is likely higher now. Dr. George Grunberger recently estimated at AACE that ~95% of people with diabetes are not on CGM, and expansion into developing nations will not be easy.) To this end, we’re hopeful to see more patients at least using connected BGM so as to facilitate data download. Dr. Chamberlain provided the example of the One Touch Verio Flex and One Touch Reveal App. When analyzing data, Dr. Chamberlain advised attendees to consider not only the average blood glucose and A1c, but also variability and time-in-range, denoting a “good” standard deviation as one-third to one-half of the average blood glucose (Dr. Irl Hirsch frequently cites one-third as the desired upper limit, in line with Monnier’s work on a CV <36%.) He briefly emphasized the limitations of A1c, made especially clear by CGM data. Dr. Chamberlain noted that he has a “handful of patients” using Companion Medical’s InPen, which he jokingly characterized as “a poor man’s pump” (this is actually a big positive, considering InPen’s ability to deliver a pump’s benefits in a lower-cost package). He was especially excited about InPen’s ability to track insulin on board, as well as the 0.5 unit dose increments. We can’t wait for more smart pens to hit the market and scale, and are pleased to hear that providers are similarly enthusiastic. Dr. Chamberlain briefly detailed the now two (!) automated insulin delivery systems approved in the US – Medtronic’s 670G (as of yesterday approved in those 7+ years-old) and Tandem’s t:slim X2 pump with Basal-IQ and Dexcom’s G6 iCGM (approved yesterday in those 6+ years-old). Dr. Chamberlain mentioned that the 670G requires “quite a bit of training” and acknowledged the practice of adding “fake carbs” as “fine, it’s where we’re at right now.” He also noted that he has two patients on DIY systems, admitting that they are “doing pretty well.” We were glad to see a session dedicated to informing primary care providers on diabetes technology – given that the majority of people with diabetes do not receive their care from endocrinologists or diabetologists, many thought leaders advocate for better incorporation of technology in the primary care setting. 

  • Dr. Chamberlain, who has type 1 diabetes himself, has worn Dexcom’s G6 and Abbott’s FreeStyle Libre. He had to calibrate the G6 a few times during the first couple of days and encouraged providers to warn their patients that some initial calibration may be necessary. He found the Libre “very comfortable” and emphasized that patients must be informed that the system lacks alarms. Read diaTribe’s test drive of Dexcom’s G6 here.


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