September 10-15, 2017; Lisbon, Portugal; Days #5-6 Highlights – Draft

There was no slowing down in the final two days of EASD 2017, as you’ll be able to tell from this packed highlights report on days #5-6! Full results from AZ’s EXSCEL trial for GLP-1 agonist Bydureon (exenatide) stole the show on Thursday (we posted our breaking news coverage here moments after the data was reported), but Friday featured no shortage of new data, this time on DEVOTE (Novo Nordisk’s CVOT for basal insulin Tresiba) and CANVAS (J&J’s CVOT for SGLT-2 inhibitor Invokana). Read on for detailed highlights spanning diabetes therapy, diabetes technology (the CONCEPTT trial of CGM in pregnancy), and big picture sessions (lots on hypoglycemia!). In case you missed it, you’ll find our EASD reports from days #1-2, day #3, and day #4 online, plus a piece on DEPICT 1 results for AZ’s SGLT-2 inhibitor Farxiga in type 1.

Diabetes Therapy Highlights

1. Dr. Adrian Hernandez presented CV outcomes data from AZ’s EXSCEL trial of GLP-1 agonist Bydureon (exenatide) vs. placebo (n=14,752). He unveiled the hazard ratio for three-point MACE of 0.91 (95% CI: 0.83-1.00, p<0.001 for non-inferiority, p=0.06 for superiority) trending toward exenatide, but missing the statistical threshold for superiority by a razor-thin margin. Talk about the “edge” of superiority with an upper bound of 1.00. These results trended in a much more positive direction than many may have thought after learning that EXSCEL had not met superiority from AZ’s topline announcement in May. From our view, this certainly seems to put the punch back into the view that GLP-1 is a positive therapy – there was already so much to be said for it given glycemic impact, glycemic-dependent nature (no hypoglycemia), and weight loss.

2. We heard three new analyses from the DEVOTE CVOT of Novo Nordisk’s next-gen basal insulin Tresiba (insulin degludec) vs. Sanofi’s Lantus (insulin glargine): (i) Dr. Bernard Zinman correlated glycemic variability to severe hypoglycemia and all-cause death (paper published online in Diabetologia), (ii) Dr. Thomas Pieber showed a possible temporal relationship between severe hypoglycemia episodes and all-cause death (paper published online in Diabetologia), and (iii) Dr. John Buse discussed development of the DEVOTE hypoglycemia risk score and accompanying app, DEVOTE, which we saw at Novo Nordisk’s booth in the exhibit hall.

3. A Friday symposium provided an update on the CANVAS trial of J&J’s SGLT-2 inhibitor Invokana (canagliflozin), and Dr. Bruce Neal shared more granularity on the amputations seen with canagliflozin vs. placebo. While he reinforced that canagliflozin, independent of all other risk factors, still nearly doubled amputation risk in this uber-high-risk group, he presented univariate and multivariate analyses that revealed other factors to look out for in real-world risk mitigation.

4. Dr. Juan Frias presented full results from a phase 2b study of Elcelyx’s Metformin Delayed Release (DR), showing phase 3-readiness overall with modest efficacy though excellent tolerability.

5. During a Sanofi-sponsored symposium on real-world evidence behind basal insulins, we were treated to an overview of Sanofi’s suite of real-world studies for Toujeo (insulin glargine U300), consisting of the four DELIVER studies as well as a project called LIGHTNING, which will leverage machine learning to compare Toujeo vs. other basal insulin products.

6. New data from Sanofi/Regeneron’s ODYSSEY DM-INSULIN study demonstrated a 52% reduction in LDL cholesterol with PCSK9 inhibitor Praluent (alirocumab) vs. placebo in people with type 1 diabetes (p<0.0001), comparable to the 49% placebo-adjusted LDL reduction previously reported at ADA 2017 for the subgroup with type 2 diabetes.

Diabetes Technology Highlights

1. The CONCEPTT RCT testing CGM in pregnant women (n=215) showed positive neonatal outcomes with Medtronic’s older Guardian CGM. CGM drove a significant reduction in the incidence of large for gestational age (OR=0.51, p=0.02), fewer NICU admissions lasting 24+ hours (OR=0.48, p=0.02), fewer incidences of neonatal hypoglycemia (OR=0.45, p=0.03), and one-day shorter length of hospital stay (p=0.01). With CGM, time-in-range improved by 100 minutes/day, and A1c declined an additional 0.2%. These are spectacular results, which were published in the Lancet. With this study, we can’t imagine that responsible guidelines would dictate that any pregnant woman with diabetes should not have access to CGM.

2. Insulet’s corporate symposium shared real-world Omnipod user data from over 38,000+ Glooko users, German/Austria DPV registry members, and outcomes with different pumps at Manchester Diabetes Center. The takeaway was clear – Omnipod outcomes look similar to other pumps, and retention is excellent. We also saw an encouraging n=1 trace from the ongoing hotel study of the Horizon automated glucose control system.

3. 12-month data from Glytec’s outpatient Glucommander decision support system demonstrated a significant A1c drop of 2.5% (baseline: a very high 10.3%), sustained with adjustments every ~57 days after goal was reached. It was an uncontrolled study, but Glytec continues to show strong results.

4. Dr. William Tamborlane spelled out the “Inconvenient Truth” of today’s diabetes management (poor), illuminated the light at the end of the tunnel (new tools, including adjunct therapies), and shared a 670G wish list from two colleagues who were in the pivotal study. Wow!

5. A BD poster showed that local inflammation, mimicking that seen from extended infusion set wear and insulin excipients, alters insulin PK profile in a large animal model.

6. A four-month, multicenter, pragmatic clinical trial (n=415) demonstrated that use of the Valeritas V-Go insulin delivery device is associated with significant reductions in A1c amongst type 2 adults, alongside significant decreases in the insulin total daily dose (TDD). 

7. Integrity Applications found no accuracy differences in use of its non-invasive ear-clip glucose monitor in people with prediabetes and type 2 diabetes (MARDs of ~16%-18%). While there is still lots to prove here, we think it’s notable that there is this good work focused on prediabetes and type 2 diabetes prevention.

8. A small randomized controlled trial (n=32) found that use of an automated bolus calculator improves glycemic variability, but does not change A1c, instances of hypoglycemia, quality of life or glucose levels in type 1 adults with pump experience and “reasonable” glucose control (A1c <10%). The characterization of “reasonable” is lacking, from our view.

Big Picture Highlights

1. Dr. Bart Van der Schueren (University of Leuven) provided an EMA perspective on measuring hypoglycemia in trials. The EMA’s updated diabetes drug guideline will be open for comment soon, likely defining “clinically important hypoglycemia” as <54 mg/dl and a “glucose alert level” at <70 mg/dl. It was great to see the enthusiasm for using CGM in drug trials, and we expect comments will focus on moving to “<” instead of “<” to align with consensus.

2. Dr. Stephanie Amiel (King’s College London) provided a compelling snapshot of why <54 mg/dl (<3 mmol/l) is an evidence-based cutoff for serious/clinically important hypoglycemia.

3. The International Hypoglycemia Study Group hosted an information-packed symposium at EASD, imparting some of their endless knowledge on hypoglycemia prevalence, pathogenesis, therapeutic pipeline, and more. There was so much good learning here!

4. It was an absolute pleasure to attend the Novo Nordisk-sponsored Rising Star Symposium, featuring some of the best and brightest young minds in diabetes research presenting their work in immunology, islet plasticity, gene-behavior interactions, and favorable adiposity.

5. Dr. Angus Jones detailed the efforts of the UK MASTERMIND consortium, which hopes to develop a precision medicine model capable of improving the accuracy of diabetes diagnosis and treatment decision-making.

Diabetes Therapy Highlights

1. GLP-1 Agonist Bydureon (Exenatide) Narrowly Misses Superiority in EXSCEL CVOT (HR=0.91, 95% CI: 0.83-1.00)

Dr. Adrian Hernandez presented CV outcomes data from AZ’s EXSCEL trial of GLP-1 agonist Bydureon (exenatide) vs. placebo (n=14,752). Immediately, without wasting a breath, he unveiled the hazard ratio for three-point MACE of 0.91 (95% CI: 0.83-1.00, p<0.001 for non-inferiority, p=0.06 for superiority) trending toward exenatide, but missing the statistical threshold for superiority by a razor-thin margin. Talk about the “edge” of superiority with an upper bound of 1.00. These results trended in a much more positive direction than many may have thought after learning that EXSCEL had not met superiority from AZ’s topline announcement in May. The trial did not reach superiority for any individual component of the three-point MACE composite, but all of these secondary outcomes trended in favor of exenatide, with hazard ratios of 0.86 for non-fatal stroke (95% CI: 0.70-1.07), 0.95 for non-fatal MI (95% CI: 0.84-1.09), and 0.88 for CV death (95% CI: 0.73-1.05). Dr. Hernandez also presented several subgroup analyses that probed for possible heterogeneity of effect among different patient populations. Interestingly, there was a significant interaction for age (p=0.005 for interaction), whereby Bydureon was associated with statistically significant CV risk reduction in people older than 65 (HR=0.80, 95% CI: 0.71-0.91). Turning to secondary endpoints, Dr. Hernandez reported a hazard ratio for all-cause mortality of 0.86 (95% CI: 0.77-0.97, nominal p=0.016) favoring exenatide vs. placebo. By the trial’s pre-specified hierarchical testing structure, this can only be considered exploratory since exenatide did not meet superiority for its primary endpoint. Nevertheless, we find this suggestion of a 14% risk reduction for all-cause mortality to be more than noteworthy from a population perspective. The question now is whether the neutral overall result in EXSCEL had more to do with the exenatide molecule (compared to Novo Nordisk’s liraglutide) or with elements of pragmatic trial design, such as no run-in period, a larger primary prevention cohort (27%), a wide range of concomitant medications allowed, and a “less sick” population. See our detailed discussion and commentary section for insights from Drs. Angelyn Bethel, Rury Holman, independent commentator Dr. Francesco Giorgino, and others.

2. New DEVOTE Analyses Reveal Links Between Glycemic Variability/All-Cause Mortality, Severe Hypoglycemia/CV Death

We were treated to three new analyses from the DEVOTE CVOT of Novo Nordisk’s next-gen basal insulin Tresiba (insulin degludec) vs. Sanofi’s Lantus (insulin glargine): (i) Dr. Bernard Zinman correlated glycemic variability to severe hypoglycemia and all-cause death (paper published online in Diabetologia), (ii) Dr. Thomas Pieber showed a possible temporal relationship between severe hypoglycemia episodes and all-cause death (paper published online in Diabetologia), and (iii) Dr. John Buse discussed development of the DEVOTE hypoglycemia risk score and accompanying app (which we saw at Novo Nordisk’s booth in the exhibit hall).

• (i) In DEVOTE 2, investigators found day-to-day glycemic variability to be significantly associated with all-cause mortality (HR=1.58, 95% CI: 1.23-2.03, p=0.0004). After adjusting for A1c and baseline characteristics, this finding retains its statistical significance with a hazard ratio of 1.33 (95% CI: 1.01-1.75, p=0.0432). Dr. Zinman presented additional data on glycemic variability and its relationship to severe hypoglycemia, which after adjusting for A1c and baseline characteristics, outputs a hazard ratio of 3.37 (95% CI: 2.52-4.50, p<0.0001). In contrast, he announced that there was no significant association between glycemic variability and three-point MACE after adjusting for A1c and baseline characteristics (HR=1.21, 95% CI: 0.98-1.49, p=0.0811). Dr. Zinman concluded that Tresiba’s benefits may arise from lower glycemic variability, which could substantially decrease frequency of hypoglycemia and risk for all-cause death, per this analysis. As a reminder, the relative risk reduction for severe hypoglycemia was 40% with Tresiba vs. Lantus in the original DEVOTE presentation at ADA 2017, while relative risk reduction for severe nocturnal hypoglycemia was even more impressive at 53%. In our view, hypoglycemia is a critical outcome beyond A1c and any therapy that reduces this risk is enormously valuable in our diabetes treatment arsenal. The hard evidence linking glycemic variability to hypoglycemia and all-cause death will also be valuable as we push for more consideration (from clinical trialists, regulators, patients, providers, and payers alike) of outcomes beyond A1c.
• (ii) Next, Dr. Pieber presented the DEVOTE 3 analysis, showing a possible temporal relationship between severe hypoglycemia and all-cause death. In the 15 days following a severe hypoglycemia episode, hazard ratio for all-cause mortality in DEVOTE was 4.20 (95% CI: 1.35-13.09). In the first 30 days post-event, this hazard ratio was 3.66 (95% CI: 1.51-8.84). The increase in risk for all-cause death seems to decline as more time passes following a severe hypoglycemia episode (at one year – HR=2.78, 95% CI: 1.92-4.04). Dr. Pieber noted the wide confidence intervals, acknowledging that the temporal relationship of severe hypoglycemia and death is suggestive rather than conclusive at present. That said, severe hypoglycemia still showed a significant increase in risk for death from any cause at any time (HR=2.51, 95% CI: 1.79-3.50), and this is a very important finding. Severe hypoglycemia did not show statistically significant time course correlations with MACE events overall, but CV death was more likely following a severe low (HR=2.14, 95% CI: 1.37-3.35). These findings give yet another reason that hypoglycemia risk reduction is key in diabetes care – if the exceptionally high cost of hospitalizations and the quality of life impact wasn’t enough, this data now also shows a clear link to all-cause death.

• (iii) Dr. John Buse discussed development of the DEVOTE hypoglycemia risk score and accompanying app. He reviewed rigorous analyses to show that the risk score is indeed accurate, predicting risk for severe hypoglycemia or a MACE event based on input variables such as age, A1c, duration of diabetes, gender, and insulin status (naïve, bolus, or basal-bolus). Dr. Buse suggested one potential application of the app in facilitating patient/provider conversations when it comes to insulin initiation – people with type 2 diabetes are often resistant to starting insulin treatment, but he showed how adding bolus insulin on the app only modestly changes hypoglycemia risk for certain other pre-set variables. All in all, though the purpose of this app is not definitely decided, we thought it was a brilliant tool, and we’re excited to see where Novo Nordisk goes with this.

3. Dr. Neal Presents Granular Amputation Data from CANVAS

A Friday symposium provided an update on the CANVAS trial of J&J’s SGLT-2 inhibitor Invokana (canagliflozin), and Dr. Bruce Neal shared more granularity on the amputations seen with canagliflozin vs. placebo. He discussed a dozen or so risk factors that were associated with lower-extremity amputations according to a univariate analysis of the CANVAS safety data, explaining that “by far the strongest relative risk came from having a prior amputation” (HR=21.4), though he also listed peripheral vascular disease (HR=2.5), neuropathy (HR=3.4), history of CV disease (HR=2.9), nephropathy (HR=2.2), baseline insulin use (HR=2.4), baseline A1c >8% (HR=2.0), male gender (HR=2.6), retinopathy (HR=2.3), use of loop diuretics (HR=2.1), baseline eGFR <45 ml/min/1.73m² (HR=1.8), and diabetes duration ≥10 years (HR=1.6). His slides then advanced to display findings from a multivariate analysis: prior amputation (HR=20.9), peripheral vascular disease (HR=3.1), male gender (HR=2.4), neuropathy (HR=2.1), A1c >8% (HR=1.9), presence of CV disease at baseline (HR=1.5), and canagliflozin treatment (HR=1.8) remained as variables significantly increasing an individual’s amputation risk. Dr. Neal reinforced that canagliflozin treatment, independent of all other risk factors, still nearly-doubled amputation risk – in other words, these new results don’t explain away the amputation signal seen in CANVAS, attributing it to something other than the molecule, but they do pinpoint factors to consider as we push for more diligent monitoring and enhanced patient education around foot care in diabetes. Moreover, the multivariate analysis identifies possible reasons that CANVAS may have found an amputation signal not seen in EMPA-REG OUTCOME (in which, Dr. Neal reminded everyone, amputations were collected retrospectively) – if, for example, Lilly/BI’s CVOT featured much less peripheral vascular disease, neuropathy, or prior history of amputations in its baseline study population (we’re eager to do a deeper dive on this). Looking ahead, Dr. Neal suggested that AZ’s DECLARE CVOT for Farxiga (dapagliflozin) will shed more light on SGLT-2 inhibitors and amputations. He underscored that other phase 3 trials of canagliflozin (not counting CANVAS) have reported only 10 lower-extremity amputation events total, in 8,114 participants, and that Truven observational results reflect no imbalance (HR=0.98, 95% CI: 0.68-1.41). This symposium also featured new renal data from CANVAS (presented by Dr. Dick De Zeeuw), a suggestion of beta cell protection from Dr. David Matthews, and fascinating independent commentary by Dr. Ele Ferrannini (this talk was worth waiting till the very end of EASD 2017!). Read about all of this in our detailed discussion and commentary section below.

4. Long-Awaited Phase 2b Read Out on Elcelyx’s Metformin Delayed Release (DR) Demonstrates Modest Efficacy, Excellent Tolerability, Readiness for Phase 3

Dr. Juan Frias presented full results from a phase 2b study of Elcelyx’s Metformin Delayed Release (DR). These findings have been highly-anticipated since topline data from the study (n=542) was announced in November 2016 (also the subject of a late-breaking abstract at ADA 2017). After 16 weeks of treatment and from a baseline A1c of 8.6%, type 2 diabetes patients randomized to the highest tested dose of Metformin DR (1,500 mg) experienced a mean A1c reduction of 0.62% vs. a ~1.1% reduction with the comparator dose of metformin immediate release (IR) 2,000 mg (p<0.05). Fasting glucose reduction with Metformin DR was >75% of that observed with Metformin IR. PK analysis found that Metformin DR caused substantially lower plasma exposure (~1/3), as intended (this therapy candidate acts on the gut with an aim toward less systemic exposure than regular metformin immediate release or extended release). Taking into account the efficacy and exposure results, Dr. Frias reported that the glycemic improvement per amount of systemic drug exposure was actually higher with Metformin DR than with metformin IR. The safety/tolerability profile was superb, with clinically-meaningful and statistically significant ~50% reductions in GI adverse events vs. standard metformin IR 2,000 mg daily.

• Elcelyx may pursue a targeted phase 3 program for Metformin DR in patients with type 2 diabetes and chronic kidney disease (CKD) stages 3B and 4 – in which metformin is currently restricted or contraindicated. This would be a letdown for the broader population that could benefit from the apparent GI tolerability benefit (if covered by payers), but isn’t entirely surprising in the challenging current regulatory and commercial climate for new diabetes drugs. In light of the modest efficacy but excellent tolerability and low systemic exposure in this study, Elcelyx is likely to test higher doses of Metformin DR in phase 3 than were tested in phase 2. The company de-risked this approach by conducting a pharmacokinetic (PK) study with population PK modeling showing that Metformin DR up to 2,100 mg daily should be safe in CKD stages 3B/4.

5. Sanofi’s Ambitious Real-World Evidence Campaign for Next-Gen Basal Insulin Toujeo

During a Sanofi-sponsored symposium on real-world evidence behind basal insulins, we were treated to an overview of Sanofi’s suite of real-world studies for Toujeo (insulin glargine U300), consisting of the four DELIVER studies as well as a project called LIGHTNING. A preliminary analysis from the DELIVER-D study comparing Toujeo vs. Novo Nordisk’s next-gen Tresiba (insulin degludec) in patients who switched over from Sanofi’s Lantus (insulin glargine U100) showed no statistically significant difference in A1c reduction or hypoglycemia between Toujeo and Tresiba arms, though hypoglycemia appeared to trend ever-so-slightly in Toujeo’s favor. We also got our first introduction to the intriguing LIGHTNING project, which will use machine learning on real-world data from 156,000 US patients from Humedia/Optum to compare Toujeo vs. Tresiba, Levemir (Novo Nordisk’s insulin detemir), and Lantus. Specific goals include: (i) finding patient segments in which Toujeo provides the greatest hypoglycemia differentiation vs. comparators, (ii) evaluating medical cost-savings for those segments, and (iii) estimating differences in glycemic control between Toujeo and the basal insulin competitors for those segments. Data is expected in the next 6-12 months, and we’ll keep our eyes and ears peeled. Although DELIVER and LIGHTNING are not RCTs, we think many providers – and possibly payers – would still be eager to hear the results. We also appreciate Sanofi’s emphasis on real-world evidence, especially when it comes to critical outcomes beyond A1c like hypoglycemia, and we certainly hope payers are compelled to improve reimbursement prospects for Toujeo as a highly-effective, next-generation basal insulin. Although findings from SWITCH 1, SWITCH 2, and DEVOTE point to Tresiba’s hypoglycemia benefit over Toujeo, Toujeo’s benefits are substantial as well – far more so than Lantus and Levemir, to say nothing of NPH. All in all, Sanofi seems to be putting more commercial and clinical development resources behind its next-generation basal insulin product, which makes sense in light of waning sales of flagship product Lantus.

6. Dr. Colhoun Presents ODYSSEY DM-INSULIN Analysis in Type 1 Diabetes

New data from Sanofi/Regeneron’s ODYSSEY DM-INSULIN study demonstrated a 52% reduction in LDL cholesterol with PCSK9 inhibitor Praluent (alirocumab) vs. placebo in people with type 1 diabetes (p<0.0001), comparable to the 49% placebo-adjusted LDL reduction previously reported at ADA 2017 for the subgroup with type 2 diabetes. This double-blind, 24-week study randomized 517 insulin-treated patients (441 with type 2 diabetes, 76 with type 1 diabetes) with starting LDL levels >70 mg/dl despite maximally-tolerated statins to alirocumab 75 mg or to placebo in a 2:1 fashion. Alirocumab dose was titrated up to 150 mg at 12 weeks for patients who did not achieve an LDL cholesterol <70 mg/dl at week 8. University of Edinburgh’s Dr. Helen Colhoun presented the results specific to type 1. The impressive 52% drop from baseline LDL 121 mg/dl translated to a mean 62 mg/dl decline (as we understand it, this is very clinically-meaningful). Notably, 63% of participants achieved this LDL reduction with the lower, 75 mg dose of alirocumab, while 37% of participants were titrated up to 150 mg. In terms of the time course for LDL reduction, a massive drop occurred by week eight, after which point levels largely plateaued. Dr. Colhoun pointed out that 70% of participants with type 1 diabetes achieved LDL goal of <70 mg/dl with Praluent vs. 76% of people with type 2 diabetes (who came in with a lower mean baseline LDL of 110 mg/dl, and a smaller variance). Similarly, high levels of target achievement were seen for the non-HDL cholesterol goal of <100 mg/dl with Praluent – 71% for type 1 participants and 79% for type 2 participants. Overall, we are thrilled to see a dedicated analysis of type 1 diabetes in this clinical trial program, and we’re even more pleased that Praluent’s potent LDL-lowering effect also spans to this often under-studied population (in addition, of course, to type 2 diabetes). Given that CV disease is the main cause of morbidity and mortality in people with diabetes, we see great potential for Praluent as an effective tool for lipid management in this patient population. We are eagerly awaiting results from the ODYSSEY OUTCOMES CVOT of Praluent, expected to complete in December 2017 with results anticipated early 2018, to learn whether this LDL-lowering effect translates to a lower incidence of CV events. In the FOURIER CVOT, Amgen’s Repatha (evolucmab) recently became the first PCSK9 inhibitor to report a cardioprotective effect (demonstrating a 15% risk reduction the composite primary endpoint of CV death, non-fatal MI, non-fatal stroke, hospitalization for unstable angina, or coronary revascularization), and we think this bodes well for Praluent’s CVOT results.

• Beyond the primary endpoint of superior LDL-lowering with Praluent vs. placebo, the trial’s type 1 diabetes subgroup also met several key secondary endpoints: placebo-adjusted, alirocumab was associated with a 43% reduction in non-HDL cholesterol (p<0.0001), a 29% reduction in total cholesterol (p<0.0001), a 39% reduction in ApoB (p<0.0001), a 19% reduction in Lp(a) (p=0.0039), and a 40% reduction in LDL particle number (p<0.0001). Dr. Colhoun highlighted the Lp(a) reduction as particularly exciting, since this key risk factor for peripheral vascular disease is not affected by statin therapy. Consistent with the rest of the ODYSSEY program, adverse event rates were low, with no appreciable differences between the alirocumab and placebo arms. 

Diabetes Technology Highlights

1. CONCEPTT Trial – CGM in Pregnancy Drives Improved Neonatal Outcomes (LGA, NICU Admissions, Hypo); -0.2% A1c Advantage, +100 minutes/Day in Range

The JDRF-funded CONCEPTT RCT testing CGM in pregnant women (n=215) showed positive neonatal outcomes with Medtronic’s older Guardian CGM. Though not the primary endpoint, significantly improved neonatal outcomes were the headline – CGM drove a significant reduction in the incidence of large for gestational age (OR=0.51, p=0.02), fewer NICU admissions lasting 24+ hours (OR=0.48, p=0.02), fewer incidences of neonatal hypoglycemia (OR=0.45, p=0.03), and one-day shorter length of hospital stay (p=0.01). The numbers needed to treat (NNT) were compelling – NNTs of just 6-8 women with CGM to prevent one of those negative outcomes. The primary A1c endpoint showed a small -0.2% A1c advantage for CGM at 34 weeks (p=0.02). However, mothers on CGM spent a significant 100 more minutes/day in range (68% vs. 61%; p=0.003), 72 fewer minutes/day in hyperglycemia (27% vs. 32%; p=0.03), and a non-significant ~14 fewer minutes per day in hypoglycemia (3% vs. 4%; p=0.1). Results were published in The Lancet, a major visibility win! As expected with the older Guardian sensor, wear time was lower than in more recent CGM studies – 70% of pregnant participants used CGM for 75%+ of the time. In addition, ~80% of women reported frustrations with the CGM device. We brought this limitation up in Q&A (it was not mentioned), and would guess the trial probably underestimated current CGM’s potential benefit in pregnancy. What would outcomes have looked like with G4/G5, FreeStyle Libre, or Guardian Sensor 3? Since the trial took three years to run, a year to plan, and spanned six countries (Canada, UK, Spain, Italy, Ireland, and the US), getting the latest devices in was obviously a challenge. The study concludes that “CGM should be offered to all pregnant women with type 1 diabetes using intensive insulin therapy” – hear, hear! According to The Lancet publication, it’s also the first study to indicate potential for improvements in non-glycemic health outcomes from CGM use. Nice! There were a few big surprises from this trial: (i) pump+CGM outcomes looked worse on a few notable endpoints vs. MDI+CGM outcomes (see below); (ii) CGM had no significant benefit on severe hypoglycemia; and (iii) the difference in outcomes was quite large between some countries (a fascinating source of commentary from Dr. Elisabeth Mathiesen). See more details below. We look forward to cost-effectiveness data when it is published. The trial also included a second arm testing CGM in women planning pregnancy, but showed no A1c benefit at 24 weeks or conception.

• An editorial from Dr. Satish Garg in The Lancet was very positive, with a very favorable beyond-A1c mention: “We believe that the CONCEPTT results support CGM use during pregnancy for all women with type 1 diabetes and time in range might become an important measure in pregnancies associated with type 1 diabetes; thus endocrine and obstetric medical societies could consider advocating or recommending revising their guidelines accordingly.” We wonder if this study could help validate time-in-range as a meaningful surrogate endpoint, independent of A1c. The study follows a very positive hybrid closed loop study during labor/pregnancy from Dr. Helen Murphy and colleagues at Cambridge, which was published in NEJM in 2016 – automation is a definitely an exciting frontier for pregnancy, along with use of next-gen devices! Dr. Murphy did mention in Q&A that a FreeStyle Libre in pregnancy study is ongoing.
• Neonatal outcomes – paramount in a diabetes pregnancy study – were very positive in favor of the CGM group. Babies from mothers who wore CGM were less likely to be large for gestational age (LGA; >90^th percentile – 53% in CGM group vs. 69% in control group; p=0.02), lower incidence of neonatal hypoglycemia requiring IV glucose (15% vs. 28%; p=0.03), were less likely to require NICU admissions >24 hours (27% vs. 43%; p-0.02), and had significantly lower median customized centile (a measure of birthweight standardized for maternal ethnicity, height, weight, and neonatal sex and gestational age at delivery – 92% vs. 96%; p=0.05). Further, infants from mothers in the CGM group had hospital stays reduced by nearly a full day (3.1 days vs. 4.0 days; p=0.02). Co-PI Dr. Denice Feig pointed out that not only were the median customized centiles lower in the CGM group (across each of the four study sites), but the lower portion of the box plot was much wider, indicating that many more babies were closer to the normal weight range than in the control group. The numbers needed to treat (NNT) were quite compelling: Six women with CGM prevented one event of LGA; eight women with CGM to prevent one event of neonatal hypoglycemia; and six women with CGM to prevent one NICU admission over 24 hours. Economic analysis wasn’t presented, but we’d guess that 72-96 months of CGM use (nine months per pregnancy * NNT) would be cost-effective relative to those expensive negative events. There were no differences in serious adverse pregnancy outcomes (miscarriage, stillbirth, termination, or congenital anomaly), obstetric outcomes (hypertensive disorders in pregnancy, C-section, maternal weight gain, maternal length of hospital stay), or gestational age at delivery between groups.
• Get many more details below on the glycemic and pump vs. MDI outcomes, along with commentary from Dr. Elisabeth Mathiesen.

2. Insulet Shares Real-World Data from 38,000+ Omnipod users on Glooko + DPV + Manchester – Similar Outcomes to Other Pumps; No Pipeline Updates

Insulet’s corporate symposium shared real-world Omnipod user data from over 38,000+ Glooko users, German/Austria DPV registry members, and outcomes with different pumps at Manchester Diabetes Center. The takeaway was clear – Omnipod outcomes and retention look similar to other pumps, if not slightly better. Overall, there was nothing truly compelling or shocking in the cross-sectional, retrospective, observational data (see key outcomes below), but it does show what is possible with direct device data downloads – characterizing real-world outcomes, understanding populations, and benchmarking. The Glooko partnership is clearly going to become a bigger Insulet asset over time, helping the company keep up with Medtronic’s CareLink. Indeed, Insulet-provided Glooko is now in 2,800+ clinics, with over 50,000 users uploading. There were no big pipeline updates, though the Horizon Automated Glucose Control system has now been tested in 113 patients (n=7,104 hours), up notably from 92 patients (n=4,584 hours) as of AADE in August. Medical Director Dr. Trang Ly did not give a timing update on Horizon, though did confirm that a pre-pivotal study is the next step after the ongoing five-day hotel study is complete. She did show a CGM trace from one 10-year-old participant in the hotel study, who entered with a mean A1c of 9.8%; on day #4, mean glucose had dropped to ~144 mg/dl, predicting an estimated A1c drop to ~6.7% (an impressive 82% time-in-range). No other days or participants were shown, though this was an encouraging n=1 plot in one user, and we look forward to full outcomes from this first outpatient study with Horizon. Presumably the previous timing is on track for a pivotal in 2018 and launch in 2019, though this was not confirmed. Otherwise, President Shacey Petrovic confirmed that Lilly U500 Omnipod clinical/development work is complete, data analysis is in progress, and Insulet is “preparing for submission” and running human factors studies. (A launch is expected in 2019, per 2Q17.)  As of AADE, Dr. Trang Ly hoped data from this phase 3 study (VIVID; n=416) could be shared at ADA 2018. U200 work remains in “technical development” and an FDA submission is expected in the “next 1-1.5 years,” in line with the planned 2020 launch. There was no update on the Dash PDM beyond what we saw at AADE and ADA; as of the 2Q17 call, FDA submission was expected in 4Q17. On the commercial front, there was definite excitement from Ms. Petrovic concerning Insulet’s move to directly distribute Omnipod in Europe, starting in mid-2018 with the Ypsomed agreement expiration (see our previous coverage).

• Dr. Trang Ly shared data from 38,778 Omnipod patients in the US with at least three months of data downloaded using Insulet-provided Glooko. The average glucose level was 186 mg/dl, equivalent to an estimated A1c of 8.1%. Dr. Ly emphasized that this “compares favorably” to data from the T1D Exchange Registry, where A1c is 8.4%. The more interesting analysis showed data in a subset of Omnipod CGM users (n=3,394) vs. non-CGM users (n=35,384) – those using CGM had a slightly higher percentage of glucose tests in-range (52% vs. 48%), a slightly lower estimated A1c (8.1% vs. 7.8%), and a lower total daily insulin dose. Dr. Ly mentioned that this data likely underestimates CGM use in Insulet’s user base, since it does not include G5/Glooko integration. A separate slide showed a subset of Insulet users with self-reported demographic data, which included a notable n=496 type 2s– they were slightly older (mean age: 56 years), but had a lower estimated A1c (7.8%) and higher daily insulin dose (59 units per day) than the full cohort. This certainly bodes well for Insulet’s U500 and U200 work! We’ve include both slides below.

• Drs. Thomas Danne (DPV registry) and Lalantha Leelarathna (Manchester Diabetes Center) provided reassuring real-world Omnipod data from Europe – outcomes with Omnipod looked very similar to other pumps. The DPV registry showed increasing use of Omnipod since 2012, with the highest use in youth with type 1 diabetes (10-15 years). A1c of Omnipod users in DPV was an encouraging 7.2%-7.7% (depending on age group); it was not compared to other pumps. On average, <10% of patients stopped using Omnipod in the DPV, retention in line with what Insulet has shared historically. At the Manchester Diabetes Center, Omnipod retention at three years was an impressive 99%, better than Medtronic (96%), Animas (96%), and Roche (81%). Said Dr. Danne, “For people who say Omnipod has a delivery problem – we don’t see this with the data and with the retention rate. We can feel good about it.” Dr. Danne concluded that “switching to a tubeless pump appears to be an effective alternative to MDI.” 

3. 12-Month Data on Glytec’s Outpatient Glucommander Insulin Titration Software – 2.5% A1c Drop (Baseline: 10.3%)

12-month data from Glytec’s outpatient Glucommander decision support system demonstrated a significant A1c drop, sustained with adjustments every ~57 days after goal was reached. 74 patients with insulin-dependent diabetes (baseline A1c: 10.3%; 78% type 2; average age = 56 years) were included in the non-randomized, uncontrolled study. At three months, A1c had dropped 2.3% points to 8.0%; at six months, A1c stayed flat at 8.0%; at nine months, A1c crept up to 8.3%; and at 12 months, A1c fell to 7.8%, representing a 2.5% absolute drop from the high baseline of 10.3%. The magnitude of this drop closely resemble that shown in six-month data from a separate study (presented at ATTD 2o17). Over the course of the study, mean blood glucose dropped from 214 mg/dl to 162 mg/dl. Impressively, the median time to goal (three consecutive days with average blood glucose <180 mg/dl) was just seven days. These rapid and sustained improvements in glycemia came at a very low risk of SMBG-detected hypoglycemia. The percentage of readings <54 mg/dl was o.37%, and the percentage <40 mg/dl was 0.05% – no episodes required assistance. These values were captured in a mean 3.0 fingersticks per day in the first three months, and 2.5 per day in the final three months. It’s impressive that Glytec can titrate insulin with so little glucose data; how would it look with CGM? After the first three months of treatment, Glucommander recommended a dose update every ~57 days – the poster reasons that the best-case three-to-four months between in-person clinic visits doesn’t meet the titration needs of most patients, and we would add that software is likely better at adjusting insulin than a provider with limited time. Average total daily dose per kg rose by 38% (0.66 U/kg to 0.91 U/kg), reflecting a failure to intensify prior to the intervention. Interestingly, the percent of daily dose coming in the form of basal fell from 50% to 31% - this ratio of basal:bolus continues to vary widely in studies and we hope the field moves away from one-size-fits-all expectations for 50%/50%. The poster notes that larger prospective RCTs are needed to confirm these results – great to hear since a number of studies have shown solid data but with no control group – and also that Glucommander will be used to treat over 100,000 patients in 2017.

• We would have loved to see CGM in this study, at least in a subset of participants, to truly capture time spent in hypoglycemia (especially overnight). Richer CGM data could also be leveraged to guide the titration software more precisely. Glytec has existing BGM integrations with Roche, Livongo, Agamatrix, and Telcare.
• Participants were given a Telcare cellular-enabled meter and asked to check four times a day. Blood glucose data was automatically sent to the clinic, where patients due for titration were flagged in the Glytec software so that a nurse could make the recommended adjustment, which appeared either on the patient’s meter or via text message.

4. Dr. William Tamborlane: Enthusiasm for Adjunct Therapies in T1; Colleagues’ 670G Wish List

Dr. William Tamborlane reviewed much of the back half of his lecture from AACE back in May, spelling out the “Inconvenient Truth” of today’s diabetes management (poor), illuminated the light at the end of the tunnel (new tools, including adjunct therapies), and shared a 670G wish list from two colleagues who were in the pivotal study. Dr. Tamborlane, a diabetes technology pioneer, was enthusiastic about adjunctive therapies in type 1 diabetes – “they’re obviously not a cure, but there’s lots of excitement at this meeting, particularly about SGLT-2 inhibitors in type 1.” He was of course referencing DEPICT 1 (AZ’s trial of Farxiga (dapagliflozin) in type 1 diabetes) and inTandem 3 (Lexicon’s SGLT-1/2 inhibitor, sotagliflozin.) We hope to see Dr. Tamborlane’s Yale group study SGLT-2s on top of closed loop, something his colleague Dr. Eda Cengiz has mentioned in the past. Dr. Tamborlane later shared some MiniMed 670G feedback provided by two of his colleagues who were in the pivotal trial: They’d like to: (i) be exited from closed loop (Auto Mode) less often; (ii) adjust target glucose levels (it’s fixed at 120 mg/dl); (iii) manually set temporary basal rates for periods of stress; (iv) manually give correction doses that correct to glucose levels <150 mg/dl; and (v) track glucose and insulin delivery remotely on smartphone apps. We note that this list has grown by one since we last heard him speak at AACE, with the latest addition of “fewer exits.” 

5. BD Animal Study on Local Inflammation Altering PK Insulin Profile – Implications for Extended Infusion Set Wear

A BD poster showed that local inflammation, mimicking that seen from extended infusion set wear and insulin excipients, alters insulin PK profile in a large animal model. Swine (n=5) were given bolus subcutaneous injections of lipopolysaccharide (LPS; a documented activator of inflammatory cascades) or saline. 24 hours later, four units of U100 insulin lispro was injected into the same site. Plasma measurements recorded over the following six hours demonstrated significantly less total circulating insulin over the first 60 minutes, significantly lower peak plasma insulin concentration, and significantly slower time to peak in the LPS condition. This means that local inflammation may cause decreased and slower insulin absorption into the vasculature. The poster claims that this is the first time that local subcutaneous inflammation has been shown to significantly alter insulin PK, and BD is interested in looking more representatively at the effects of device- and insulin excipient-induced insulin. This has obvious implications in extended infusion set wear, often called the Achilles heel of pump therapy. A number of groups, including BD with funding from JDRF, are currently looking at ways to extend set wear time to seven days or beyond. JDRF is also funding TJU/Capillary Biomedical’s efforts ($1.5 million) to develop a novel seven-day wear set,  and last June, Medtronic shared that it is also working on an extended wear set that it is hoping to launch “in the next three years.” Addressing this issue with novel mechanical or molecular approaches will greatly reduce hassle associated with pump wear, and make a single-site CGM/pump closed loop system more realistic.

6. Valeritas V-Go Use Significantly Reduces A1c in “Pragmatic Controlled Trial,” With Higher Baseline A1c Reflecting Tougher Patients

In a four-month, multicenter, pragmatic clinical trial, use of the V-Go insulin delivery device induced significant reductions in A1c amongst adults with type 2 diabetes (n=415), alongside significant decreases in total daily dose of insulin (TDD). 52 participating clinical sites across the US were randomized to offer either the V-Go or continue offering standard care to their patients (we’re not sure how enrollment was determined within a clinic assigned to V-Go). Overall, the V-Go group saw a significantly greater A1c decrease than the control group (0.95% vs. 0.46%; P=0.002), though from a higher baseline A1c (9.9% vs. 9.3%). Presumably, this reflects some selection bias in prescribing – those given V-Go were more likely to be having a harder time with their diabetes. There was a 32% decrease in the number of V-Go patients with A1c greater than 9% vs. a 17% reduction in the control group, though the higher baseline A1cs with V-Go makes comparative conclusions challenging. Indeed, by the study’s end, the control group actually had lower average A1cs than did the V-Go group (8.8% vs. 9.0%). Average TDD decreased significantly in V-Go patients by 0.2 U/kg (baseline: 71.3 units), while TDD in the control group remained unchanged. Similar results were achieved in a sub-population of MDI users, with V-Go patients achieving significantly larger decreases in A1c (-1.0% vs -0.4%; p=0.006) alongside declines in TDD. Again, baseline A1c levels were significantly greater in the V-Go group. An interesting A1c vs. TDD density plot for the overall study population (below) shows leftward and downward shifts in the solid blue line (V-Go group at end of the study) compared to the dotted blue line (V-Go at the beginning of the study) – this indicates a lower A1c with less insulin. Meanwhile, the control group only saw the red dotted line (pre-study) shift down slightly (to the solid red line), reflecting improved A1c without decreased TDD. Promisingly, a questionnaire administered to patients at the end of the study indicated that 94% of patients used V-Go as directed, and 85% intend to continue using V-Go. Valeritas has published a number of positive posters and studies in the recent past showing declines in A1c and TDD, and we’ll be interested to see if the newly public company can scale adoption and reimbursement. 

7. Integrity’s Non-Invasive GlucoTrack Sensor Performs with MARD in Mid-to-High Teens in Prediabetes and T2D

A single-arm study (n=32) sponsored by Integrity Applications found that there were no significant differences in the accuracy of the GlucoTrack, the company’s non-invasive blood glucose monitoring device, between people with prediabetes (n=7; MARD=18.3%), people newly diagnosed with type 2 diabetes (duration 5 years; n=9; MARD=15.6%), and people with long-duration type 2 diabetes (duration >5 years; n=16; MARD=16.6%). Participants wore the ear-clip GlucoTrack for a “calibration day,” comprised of three measurements every 10 minutes, followed by a trial day consisting of a pre-prandial measurement, standardized breakfast, and six additional measurements every 30 minutes. All groups demonstrated similar proportions of measurements in the A and B zones of the Consensus error grid (at least 92% of the measurements were in Zone A across all groups). The sample sizes are small, so we’re not sure how results would look in a more robust pivotal study. If the main application is for people with prediabetes and those on orals, is this accuracy good enough? GlucoTrack is already CE marked and has launched in China, South Korea, Turkey, and some European countries, according to a rep in the exhibit hall. There is still a lot to prove here. The product’s non-invasive aspect is appealing (after a few calibration fingersticks twice a year), but the ear clip is bulky and attached to an industrial-looking handheld by a wire, and each reading takes one minute to register. The form factor and user friendliness aspects are lacking at the moment (save for the lack of fingersticks), while the likes of Dexcom and Abbott continue to move quickly on innovating subcutaneous sensors – slimmer on the body, less painful to insert, more integrated with consumer electronics, and lower cost. Until GlucoTrack becomes more discreet and less time-consuming to use, adoption will likely be difficult.

8. Automated Bolus Calculator Improves Glycemic Variability in T1 Adults with No Change in A1c, Hypo, QOL, or Glucose Levels

A randomized controlled trial (n=32) found that use of an automated bolus calculator improves glycemic variability, but does not change A1c, instances of hypoglycemia, quality of life, or glucose levels in type 1 adults with good A1c control and pump experience. While previous studies have shown that automated bolus calculators can reduce A1c and improve quality of life, the majority have consisted of patients new to pumps and with high A1c values. In this four-month study, 32 type 1 adults who had used pumps for at least six months and with A1c <10% (a bit high to constitute “reasonable control” in our view), were randomized to receive either an automated bolus calculator or continue with standard care. All participants received instructions on carb counting and took an exam prior to randomization. Although glucose variability did significantly improve in the experimental group while it remained unchanged in the control, there were no observed significant differences in glucose levels, LBGI and HBGI scores, hypoglycemia frequency, and quality of life both within and between groups. It’s interesting that improvements in glycemic variability were achieved, yet this did not impact anything else – we can’t recall seeing that in a study before. This was a small study and both groups consisted of experienced pumpers with verified carb counting ability, so it’s perhaps not surprising that the calculator didn’t augment A1c or other parameters drastically. We expect that the meal dosing paradigm is going to change quite drastically in the coming years with automated insulin delivery – what’s the best way optimize the patient-automated algorithm interface to achieve maximum time-in-range? Meanwhile, how will bolus calculators benefit MDIs when they are more widely integrated into apps paired with smart pens? Interestingly, the researchers had originally intended to enroll type 2 adults as well, but they were only able to find two eligible participants and therefore decided to exclude them.

Big Picture Highlights

1. Dr. Bart Van der Schueren on Upcoming EMA Guideline for Clinical Trials – Likely to Include Hypo Thresholds <54 and <70 mg/dl (open for COmment), Continued Rec to Use CGM

Dr. Bart Van der Schueren (University of Leuven) provided an EMA perspective on measuring hypoglycemia in trials. Notably, the Agency’s updated diabetes drug guideline will be open for public comment soon, confirming his remarks at the Outcomes Beyond A1c meeting in August. In the draft guideline (no timeline on release), EMA is leveraging recommendations from the International Hypoglycemia Study Group – “clinically important hypoglycemia” will be defined as <54 mg/dl, while a “glucose alert level” will be <70 mg/dl (but does not have to measured routinely in trials). Dr. Van der Schueren admitted that use of the “less than or equal to” will likely be a source of public comment: “I don’t know why we did that, but we might get comments on that!” We hope it is changed to simply “less than” so that it aligns with the consensus achieved at the Outcomes Beyond A1c meeting in August: <54 mg/dl and <70 mg/dl. (This may seem small, but it is very critical for the field to agree on.) Dr. Van der Schueren mentioned that if a drug shows less of these non-severe hypoglycemia events, it “will be taken into account” in the review – we assume such CGM data would be reported in the label, just like severe hypoglycemia for products like Tresiba. He echoed his August comments that “better analytic tools have changed what we can measure” – his slide showed a picture of time-in-range on FreeStyle Libre and the FDA press release approving Dexcom’s G5 for non-adjunctive use. We love seeing this CGM enthusiasm from EMA and we seriously hope FDA gets the message. Dr. Van der Schueren highlighted patient-reported outcomes as an area where more consensus is needed, since “a lot of questionnaires are not sufficiently validated.” The upcoming EMA guideline will not recommend a specific PRO, though EMA did include the hypoglycemia fear survey and diabetes therapy-related quality of life questionnaire in the Trulicity label (see page 16 here). Last, he reprised his comments from August on composite/co-primary endpoints, which are difficult for regulators – they typically mix efficacy and safety endpoints within the same statement (e.g., achieving A1c <7% without episodes of severe hypoglycemia or DKA). “It becomes difficult to disentangle that into a primary indication,” said Dr. Van der Schueren. The EMA is seeing more of these outcomes proposed for adjunctive therapies in type 1 diabetes (e.g., SGLT-2s), and we’ll be interested to see how companies and regulators cope with them for benefit-risk decisions and product labels. 

2. Dr. Stephanie Amiel on Why <54 mg/dl (<3 mmol/l) is An Evidence-Based Cutoff for hypoglycemia

Dr. Stephanie Amiel (King’s College London) provided a compelling snapshot of why <54 mg/dl (<3 mmol/l) is an evidence-based cutoff for serious/clinically important hypoglycemia. As she did at the August 29 outcomes beyond A1c meeting, Dr. Amiel shared study after study showing the link between negative health outcomes and glucose levels <54 mg/dl – impaired cognitive function (e.g., Gonder-Frederick Diabetes Care 2009), strong correlation with severe hypoglycemia (e.g., DAFNE data), and evidence for CV harm and arrhythmias (e.g., ACCORD, Pistrosch et al., Acta Diabetol 2015). Much of the evidence is summarized in the International Hypoglycemia Study Group paper, published in Diabetologia and Diabetes Care earlier this year (see the reference section for many of the studies she covered). The paper has already been cited 20 times! Dr. Amiel again noted that 70 mg/dl is an “alert level,” meaning it is the lower limit of the target range (Dr. Amiel said this level was “usually asymptomatic”) and alerts patients that they need to do something. Interestingly, in many of the studies Dr. Amiel covered, <70 mg/dl was NOT linked to the host of adverse outcomes like <54 mg/dl. We thought her summary of the data was compelling and wonder how FDA will interpret all the evidence linking <54 mg/dl to meaningfully negative outcomes. Will further validation be needed? Hopefully, the field will start to use CGM widely to capture the occurrence of hypoglycemia in trials, further building a case. Ultimately, we hope non-severe hypoglycemia data (e.g., <54 mg/dl) makes it into product labels, changes incentives for drug development, and even redefines the bar for reimbursement.

3. International hypo Study Group Symposium Highlights: Prevalence, Pathogenesis, Sleep, Hypo Unawareness Pipeline

The International Hypoglycemia Study Group hosted an information-packed symposium at EASD, imparting some of their endless knowledge on hypoglycemia prevalence, pathogenesis, therapeutic pipeline, and more. There was so much good learning here, and we’ve condensed some of the most impactful and interesting points below.

• Hypoglycemia is certainly recognized to be a huge problem, but statistics presented by Dr. Ulrik Pedersen-Bjergaard were still alarming. These figures demonstrate how, even though hyperglycemia hospitalizations have gone down and overall A1c has decreased, hypoglycemia remains a big limitation to better therapy. As CGM use rises further, we expect a lot more hypoglycemia that has traditionally gone undetected will surface, hopefully adding more urgency to the discussion. Could we ever see a day where drugs are indicated for the reduction of hypoglycemia (independent of A1c)?

• Denmark’s Dr. Pedersen-Bjergaard emphasized that the correlation between hypoglycemia and A1c is not strong, likely due to differential glycemic variability from patient-to-patient. In the DCCT, intensive control did increase the risk of severe hypoglycemia relative to the control arm, but people with recurrent hypoglycemia or impaired awareness of hypoglycemia were excluded from the study. Recently, the HAT study demonstrated that the association between A1c and severe hypoglycemia is actually much less significant in the broader population – there is a slightly significant relationship (with a small effect size) between A1c and any hypoglycemia in type 1, but not in type 2. Nocturnal and severe hypoglycemia were not associated with A1c at all.
• Dr. Pedersen-Bjergaard introduced the IHSG patient and provider hypoglycemia risk stratification infographic tool, which will be available here soon. The tool separates patients by low, moderate, and high risk, offers solutions for each risk range, describes acute and long-term outcomes and risk factors, and providers risk-reduction strategies. This could be a very helpful resource, especially for those who aren’t seen after by one of the luminaries of the IHSG.
• Read more details below!

4. Annual Rising Star Symposium Highlights Immunology, Islet Plasticity, Gene-Behavior Interactions on BMI, & Favorable Adiposity

It was an absolute pleasure to attend the Novo Nordisk-sponsored Rising Star Symposium, which featured some of the best and brightest young minds in diabetes research today. As one of the very last sessions of the week, we cannot think of a more fitting end to the conference than learning from the newest generation of diabetes scientists. This year’s brilliant recipients were Drs. Gustaf Cristoffersson (Uppsala University, Sweden), Teresa Mezza (Cattolica University, Rome, Italy), Andrew Wood (University of Exeter, UK), and Hanieh Yaghootkar (University of Exeter Medical School, UK) who discussed their work in immunology, islet plasticity, genetic and environmental interactions on BMI, and favorable adiposity respectively. See below for some of our major takeaways from this fascinating symposium.

• Dr. Cristofferson detailed his work leveraging video microscopy to examine the cell to cell immune response in type 1 diabetes. He found that only 1%-2% of the CD8+ T-cells shown to cluster around beta cells in a diabetes mouse model are actually specific for islet antigens. By tagging islet-specific T-cells red and non-islet-specific T-cells green, Dr. Cristofferson discovered that these non-islet-specific T-cells can actually suppress the immune response, potentially by limiting trafficking or proliferation. These findings hint at a new avenue for developing immunotherapies combatting diabetes closer to the source.
• Dr. Mezza presented her work investigating the relationship between beta cell function and islet morphology in hopes of identifying new therapeutic targets to preserve beta cell mass in diabetes patients and prevent beta cell degeneration in those at risk. She found that islet size is inversely correlated with glucose uptake – the more insulin resistant a patient is, the larger the islet. Not surprisingly, insulin sensitivity is inversely related with alpha cell area, with five-times greater alpha cell mass in insulin resistant patients indicating a decrease in beta cells. Additionally, she found an increase in small islets in insulin resistant patients, suggesting that insulin resistance prompts the production of new islets as part of a compensatory mechanism. This is further supported by the observation that insulin sensitivity is inversely correlated with the proportion of double cells (cells that include markers for alpha and beta function), indicating proliferation. Importantly, duct cells in particular were found to secrete both insulin and glucagon, suggesting that they may be progenitors for endocrine cells. Dr. Mezza hopes to investigate potential regulators of duct and alpha cell trans-differentiation to identify how to intercept the conversion from duct to alpha cell and potentially redirect the process to generate more beta cells.
• Next up was Dr. Wood, who sought to determine whether genetic effects on BMI are accentuated given specific behaviors. It’s important to investigate this area, because it’s possible that we’re underestimating the contribution of genetics to obesity risk, which is currently estimated at 40%-70%. Dr. Wood described a Gene x Behavior interaction study (GxB) recently published by Tyrrell et al. in the International Journal of Epidemiology demonstrating that watching TV and self-reported activity exhibit an interaction between genetics and the environment. We know that watching more TV is associated with a greater BMI, but Tyrrell et al. were able to find that this difference in BMI actually increases by 35% as the genetic risk increases, showing that the environment is able to enhance genetic risk. Likewise, self-reported low activity exacerbates genetic susceptibility to a high BMI. Dr. Wood is currently conducting a GxB study examining interactions due to sleep patterns and physical activity data derived from wearable accelerometers. The researchers haven’t been able to identify evidence of interactions with interrupted sleep patterns, although self-reported data suggest it is a significant factor. This area of research could be critical in curbing the global obesity epidemic, as it may serve to identify which environmental factors have the greatest impact on BMI.
• Wrapping up the symposium, Dr. Yaghootkar discussed her research examining the physiological and molecular basis for why some people with obesity never develop type 2 diabetes, a condition which is termed “favorable adiposity.” Using data from the UK Biobank, Dr. Yaghootkar identified genetic variants associated with body fat percentage and then tested the genetic variance against multivariable metabolic outcomes using summary statistics from published GWASs (genome-wide association studies). These analyses revealed 620 independent genetic loci associated with body fat percentage and 24 alleles associated with favorable adiposity. These favorable adiposity alleles were shown to be associated with reduced risk of type 2 diabetes, heart disease, and hypertension. Interestingly, favorable alleles are associated with more lower body fat in women (corresponding to a pear body-type) and fat all over in men (corresponding to an apple body-type). Still, when looking at individual alleles, Dr. Yaghootkar noted heterogeneous effects in body shape, clouding the pciture. Dr. Yaghootkar hopes to use the genetic variance to understand the role of favorable adiposity on non-metabolic disease outcomes. As a parting teaser, she mentioned that favorable adiposity is associated with a higher risk of depression in women – which we find surprising at first glance because people with diabetes are more susceptible to depression, but then again, that certainly has a lot to do with the environment (stigma, etc.).

5. UK MASTERMIND Consortium Using Precision Medicine Models to Improve Accuracy of Diagnosis and Treatment Decision-Making

Exeter’s Dr. Angus Jones detailed the efforts of the UK MASTERMIND consortium, which hopes to develop a precision medicine model capable of improving the accuracy of diagnosis and treatment decision-making. Dr. Jones considers differentiation between type 1 and 2 to be a major error of practice in diabetes management, a topic we first heard him speak about at last year’s EASD Rising Star symposium. Currently, this critical decision places individuals in one of two boxes using a combination of clinical judgement and islet antibodies results, and all future management decisions are based on that initial box. But as Dr. Jones aptly noted, “patients behave like their endogenous secretions, not like their BMI or diagnosis.” Dr. Jones and his team have created a model which integrates an individual’s clinical features, antibody results, and biomarkers for insulin need to generate a predicted diagnosis with a very sensitive and specific AUC (area under curve) score of 0.97 (perfect is 1). The UK MASTERMIND consortium also aims to develop prediction models for treatment options by leveraging EHR to identify clinical predictors for therapy response. Dr. Jones and team have found that obese females are more likely to respond positively to TZDs and negatively to SFUs than non-obese females. These trends indeed appeared in a retrospective analysis of the GSK ADOPT trial. Fascinatingly, no such differences exist for males by BMI. Still, Dr. Jones acknowledged that side-effects must be considered, as TZDs are associated with a 7% weight gain, which is not ideal for women with a high BMI. Dr. Jones performed a similar analysis comparing DPP-4 inhibitors and SGLT-2 inhibitors, finding that a lower BMI and triglycerides <2.3 mmol/l (42 mg/dl) are independently associated with a better response to DPP-4 inhibitors, while a higher BMI and good renal function can predict a positive response to SGLT-2 inhibitors. Dr. Jones  hopes for more tailored treatment options, with models generating a few options that physicians can review with their patients. He also emphasized the need to replicate these preliminary findings in real-world and clinical trials to minimize false positives.

• This work is so important for two reasons: (i) Clinical decision support will greatly augment medicine in the not-so-distant future, handling the pharmaceutical aspect of care (better than any practitioner could) and freeing up provider time to focus on other factors that impact diabetes quality of life; and (ii) Parsing out biological mechanisms in the lab is time-consuming and expensive, while mining existing clinical data for patterns and trends (who responds to what therapy?) is relatively cheap and quick. The UK MASTERMIND consortium’s strategy could guide future treatment regimens and also possibly generate hypotheses to be tested in clinical trials or at the lab bench.
• Dr. Jones has released a free diagnostic app available on iTunes and Google Play stores, Diabetes Diagnostics, which uses the individual’s clinical features to distinguish between type 1 and type 2. He hopes to incorporate early insulin requirement biomarkers into the model within the next year. His team is also working on creating diagnosis prediction calculators integrated directly into the EHR and are beginning to create provider-facing tools for therapy prediction. Nice!
• VU University Medical Center Vice Chair Dr. Giel Nijpels provided a fascinating perspective on precision medicine as a general practitioner who has been in the field for 40 years. He believes that new technologies will provide the basis for effective diabetes management and appreciates the current work exploring biomarkers. The next step, in his mind, is to include environmental factors so as to create an “exposome model” based on human and environmental exposures. In just one example of the need to consider the environmental impact, Dr. Nijpels discussed how  patients of low socioeconomic status tend to have the highest diabetes risk, experience difficulties with treatment, and struggle with glycemic control. Dr. Nijpels also called for more integrated care and better stratification of people with diabetes by their susceptibility and likely response to treatment.
• Dr. Jones is a disciple of the 2016 EASD/Novo Nordisk Foundation Diabetes Prize for Excellence winner, Dr. Andrew Hattersley. See our take on Dr. Hattersley’s 2016 lecture, which touched on many of the same themes.

Detailed Discussion and Commentary

Exenatide Study of Cardiovascular Event Lowering (EXSCEL): Primary Results

Study Rationale, Design, and Conduct

Robert Mentz, MD (Duke University, Durham, NC)

Dr. Robert Mentz discussed key aspects to the design and execution of EXSCEL for AZ’s GLP-1 agonist Bydureon (2 mg exenatide once-weekly). Like previous type 2 diabetes CVOTs, this study was double-blind and placebo-controlled (n=14,752, including 7,356 randomized to exenatide and 7,396 randomized to placebo). Unlike trials that have come before it, EXSCEL was meant to be a pragmatic outcomes study, more closely mimicking real-world conditions. As such, Bydureon treatment was integrated with usual care, and any concomitant anti-hyperglycemic therapy besides another GLP-1 agonist was allowed at the discretion of a patient’s usual diabetes care provider (rather than an HCP specially-appointed to execute EXSCEL). According to Dr. Mentz, annual calcitonin measurements were the only piece of data not collected from a usual care setting. We learned in a separate conversation with AZ’s VP of US Medical Affairs Dr. Jim McDermott that Bydureon was administered via single-dose reconstitution kits instead of the multi-use prefilled product pen. The more cumbersome dosing process could have contributed to lower medication adherence in this trial vs. LEADER for Novo Nordisk’s GLP-1 agonist Victoza (liraglutide) – otherwise, we might expect a once-weekly agent like Bydureon to show greater adherence numbers vs. a once-daily agent like Victoza. Moreover, this trial featured no run-in period, which typically identifies and excludes patients who are likely to show poor medication adherence. Median follow-up was 3.2 years, and study visits occurred at one week, two months, six months, and 12 months post-randomization, after which visits were every six months until end of treatment to minimize interference with usual care. Dr. Mentz further outlined how the trial included 70 days of safety follow-up immediately after the treatment period. He described inclusion criteria as broad: a wide range of CV risk was accepted, and there was no enrichment of the study population with elderly patients (participants were accepted down to age 18). Exclusion criteria included current or prior GLP-1 agonist use, eGFR <30 mL/min/1.73m², prior pancreatitis, ≥two episodes of severe hypoglycemia in the past 12 months, calcitonin >40 ng/L, and personal or familial history of MEN-2. The primary endpoint was time to first occurrence of three-point MACE (non-fatal MI, non-fatal stroke, or CV death). Secondary endpoints were analyzed by conditional hierarchy, and included all-cause mortality, individual components of three-point MACE, hospitalization for acute coronary syndrome, and hospitalization for heart failure. The trial aimed for a minimum 1,360 primary endpoint events, and in the end collected 1,744. Lastly, Dr. Mentz established EXSCEL as a truly global study, enrolling 6,788 participants from Europe at 329 clinical sites (46%), 3,708 patients from North America at 190 sites (25%), 2,727 people from Latin America at 64 sites (18%), and 1,529 people from Asia Pacific at 104 sites (10%).

• Simultaneous with the start of this EASD symposium, full EXSCEL results were published online in NEJM.

Participant Characteristics and Risk Factor Changes

Angelyn Bethel, MD (University of Oxford, UK)

Oxford’s Dr. Angelyn Bethel presented baseline characteristics of the study population, highlighting some key differences in this regard between EXSCEL and LEADER, Novo Nordisk’s positive CVOT for GLP-1 agonist Victoza (liraglutide). AZ’s trial enrolled 14,752 adults with type 2 diabetes, making it the largest diabetes CVOT completed to-date (notably, the company’s DECLARE CVOT for SGLT-2 inhibitor Farxiga will be even larger, with >17,000 participants, while LEADER enrolled 9,340 participants). At baseline, the average EXSCEL participant was 62 years-old with BMI 32 kg/m², eGFR 76 ml/min/1.73m², A1c 8%, and diabetes duration ~12 years. Metformin was the most common background diabetes drug (~75% of both treatment arms), followed by insulin and sulfonylureas, and ~12% of people were taking a DPP-4 inhibitor at study start – in contrast, people on a DPP-4 inhibitor were excluded from LEADER. As planned, EXSCEL enrolled a participant pool that was ~30% primary prevention, ~70% secondary prevention, making this a lower-risk group overall. More specifically, ~73% of both the exenatide and placebo arms had a prior CV event at baseline (53% had a history of coronary artery disease, 17% of cerebrovascular disease, 19% of peripheral arterial disease, and 16% of congestive heart failure). Both statins and anti-thrombotics/anti-coagulants were used in >70% of participants at baseline, while ~55% of individuals were on statin therapy and 40% were taking an ACE inhibitor/ARB. Median follow-up time was 3.2 years. Treatment discontinuation was quite high in both the exenatide and placebo groups, at 43% and 45%, respectively. Dr. Bethel didn’t comment extensively on this (underscoring instead the very small number of people lost to follow-up, at <0.5% total), but we imagine use of the Bydureon reconstitution kit rather than the easier-to-use pen had some contributing influence on dropout rates. To better understand what stimulated dropouts, we’d be curious to see any data on patient experience or quality of life in both LEADER and EXSCEL (and other diabetes outcomes studies, like EMPA-REG OUTCOME, CANVAS, and SUSTAIN 6, for that matter).

• Dr. Bethel also shared data on changes in A1c, heart rate, body weight, blood pressure, and initiation of new therapies in EXSCEL. By the end of the study, there was a 0.5% A1c treatment difference between the Bydureon and placebo arms (p<0.001). Trial design aimed for glycemic equipoise, and people in both arms started from the same baseline A1c of 8%, but Dr. Bethel explained the significant difference by pointing out that HCPs were allowed to use any other diabetes drugs (with the exception of GLP-1 agonists) in treating their patients to target. Once again, she reinforced pragmatic trial design, and alluded to the fact that LEADER study protocol was more restrictive in what additional medications could be prescribed – no GLP-1 agonists, DPP-4 inhibitors, or pramlintide (AZ’s Symlin). As expected, heart rate increased by ~two beats/minute with exenatide therapy vs. placebo (p<0.001). From a baseline body weight just over 203 lbs, placebo-treated patients tended to gain weight by the end of follow-up while exenatide-treated patients experienced modest weight loss, leading to a significant ~3 lb treatment difference (p<0.001). Systolic blood pressure declined from a baseline ~136 mmHg in both arms, but Bydureon showed superior reductions by a mean ~2 mmHg (p<0.001).
• Dr. Bethel reported a 33% risk reduction for the addition of any new diabetes therapy with Bydureon vs. placebo (HR=0.67, 95% CI: 0.63-0.71, p<0.00). Similarly, relative risk reduction for new initiation of insulin treatment was 39% with Bydureon (HR=0.61, 95% CI: 0.54-0.68, p<0.001). This finding is clinically-meaningful in terms of simplifying medication regimens for people with diabetes in the real world. Moreover, it suggests that participants randomized to placebo in this trial were treated more aggressively with other agents besides once-weekly exenatide, including SGLT-2 inhibitors, which have demonstrated cardioprotection in their own CVOTs (see our coverage of EMPA-REG OUTCOME for Lilly/BI’s empagliflozin and CANVAS for J&J’s canagliflozin). Indeed, SGLT-2 inhibitors were given open-label to 6.5% of patients in the Bydureon group vs. 9.4% of patients in the placebo group.

Safety Data

Bernard Zinman, MD (University of Toronto, Canada)

Dr. Bernard Zinman presented safety data from the EXSCEL trial. Serious adverse events were well-balanced across treatment arms, affecting 17% of participants in both the exenatide and placebo groups. There were 56 individuals in the exenatide arm (0.8%) and 38 in the placebo arm (0.5%) who reported serious adverse events related to treatment, and 108 (1.5%) and 104 (1.4%) participants respectively experienced serious adverse events leading to permanent treatment discontinuation. Severe hypoglycemia (requiring assistance) occurred numerically fewer times in the exenatide group (404 events) than the placebo group (450 events). Respectively, 247 (3.4%) and 219 (3.0%) participants experienced severe hypoglycemia, though this difference did not reach statistical significance. Acute pancreatitis – historically, a concern surrounding GLP-1 agonists – occurred in 26 participants in the exenatide group vs. 22 in the placebo group. Dr. Zinman noted that these very low rates of acute pancreatitis, and their near equivalence between the exenatide and placebo group, are extremely reassuring. Malignancy was also similar between groups, affecting 355 patients in the exenatide arm and 361 patients in the placebo arm (4.8% vs. 4.9%). Within this, there were 15 cases of pancreatic cancer in the exenatide arm (vs. 16 in the placebo arm) and 14 cases of thyroid cancer in the exenatide arm (vs. six in the placebo arm). Medullary thyroid cancer (MTC) occurred in two participants on exenatide and one on placebo; in all three cases, participants had an elevated calcitonin levels at baseline, a known risk factor for MTC. Dr. Zinman explained that most safety data in EXSCEL was collected via adverse event reporting – only acute pancreatitis and malignancy were confirmed by adjudication.

Cardiovascular Outcomes

Adrian Hernandez, MD (Duke University, Durham, NC)

Dr. Adrian Hernandez took the stage for the symposium’s main event: presentation of the CV outcomes data. Immediately, without wasting a breath, he unveiled the hazard ratio for three-point MACE of 0.91 (95% CI: 0.83-1.00, p<0.001 for non-inferiority, p=0.06 for superiority) trending toward exenatide, but missing the statistical threshold for superiority by a razor-thin margin. Talk about the “edge” of superiority with an upper bound of 1.00. These results trended in a much more positive direction than many may have thought after learning that EXSCEL had not met superiority from AZ’s topline announcement in May. The trial did not reach superiority for any individual component of the three-point MACE composite, but all of these secondary outcomes trended in favor of exenatide, with hazard ratios of 0.86 for non-fatal stroke (95% CI: 0.70-1.07), 0.95 for non-fatal MI (95% CI: 0.84-1.09), and 0.88 for CV death (95% CI: 0.73-1.05). Dr. Hernandez also presented several subgroup analyses that probed for possible heterogeneity of effect among different patient populations. Interestingly, there was a significant interaction for age (p=0.005 for interaction), whereby Bydureon was associated with statistically significant CV risk reduction in people older than 65 (HR=0.80, 95% CI: 0.71-0.91). The p-values for interaction were non-significant for sex, race, and geographical region (Europe, North America, Latin America, and Asia Pacific). Similarly, there was no significant interaction for various risk factors, including diabetes duration, baseline A1c, eGFR, BMI, background anti-hyperglycemic oral therapy, prior congestive heart failure, or a history of CV events.

• Turning to secondary endpoints, Dr. Hernandez reported a hazard ratio for all-cause mortality of 0.86 (95% CI: 0.77-0.97, nominal p=0.016) favoring exenatide vs. placebo. By the trial’s pre-specified hierarchical testing structure, this can only be considered exploratory since exenatide did not meet superiority for its primary endpoint. Nevertheless, we find this suggestion of a 14% risk reduction for all-cause mortality to be noteworthy. No other secondary endpoints met superiority, but most had point estimates that trended in the “right” direction (to the left of unity), with a hazard ratio of 0.97 for fatal or non-fatal MI (95% CI: 0.85-1.10, p=0.622), 0.85 for fatal or non-fatal stroke (95% CI: 0.70-1.03, p=0.095), and 0.94 for hospitalization for heart failure (95% CI: 0.78-1.13, p=0.485). The secondary endpoint of hospitalization for acute coronary syndrome (ACS) trended in favor of placebo (HR=1.05, 95% CI: 0.94-1.18, p=0.402), but importantly, did not reach statistical significance.

Source: NEJM

Source: EXSCEL EASD presentation

EXSCEL in Perspective

Rury Holman, MD (University of Oxford, UK)

In a notable address, Dr. Rury Holman, first author on the EXSCEL paper published in NEJM, argued that this trial largely supports the CV benefits to the GLP-1 agonist class as a whole. He presented a meta-analysis of three GLP-1 agonist CVOTs (including EXSCEL, LEADER, and SUSTAIN 6 but excluding ELIXA for Sanofi’s lixisenatide because it evaluated four-point MACE as its primary outcome) showing a significant 12% risk reduction for three-point MACE with a GLP-1 agent vs. placebo (HR=0.88, 95% CI: 0.81-0.95, p=0.002). Dr. Holman reported a non-significant p-value for interaction of 0.28, suggesting no evidence for heterogeneity of effect between exenatide, liraglutide, and semaglutide. He called out that CV death accounts for two-thirds of all mortality in these large diabetes trials, and then discussed a similar meta-analysis looking specifically at CV death, which found a 13% relative risk reduction with GLP-1 agonist therapy (HR=0.87, 95% CI: 0.79-0.96, p=0.007). Here, the p-value for interaction was again non-significant at 0.43. Turning to all-cause death, Dr. Holman presented a meta-analysis showing 12% relative risk reduction with a GLP-1 agonist product (HR=0.88, 95% CI: 0.81-0.95, p=0.002). The p-value for interaction was 0.63, once more pointing to a lack of heterogeneity between these different GLP-1 agents. While it would be hard to assert that all GLP-1 agonists are identical in their effects – the variations in molecular structure are well-documented, and head-to-head studies like SUSTAIN 7 (semaglutide vs. dulaglutide) have demonstrated clinically-meaningful differences in A1c-lowering and weight loss – Dr. Holman’s overarching message was that Bydureon’s missing the mark on superiority had more to do with trial design and study population than with the exenatide molecule itself.

• He proposed four specific differences between CVOTs that may have influenced outcomes: (i) baseline A1c (8.7% in LEADER and SUSTAIN 6 vs. 8.1% in EXSCEL), (ii) primary prevention cohort (19% in LEADER, 17% in SUSTAIN 6, and 27% in EXSCEL), (iii) median follow-up time (3.8 years, 2.1 years, and 3.2 years, respectively), and (iv) median drug exposure (3.5 years, 1.8 years, and 2.4 years, respectively). The inclusion of more participants facing lower CV risk at baseline may have muted the cardioprotective benefit to Bydureon vs. placebo. In fact, a major theme emerging at the FDA Advisory Committee for Victoza’s CV indication was liraglutide’s differing effects in lower- vs. higher-risk cohorts, and the FDA-approved label update applies only to type 2 diabetes patients with established CV disease. This isn’t to say that liraglutide, exenatide, or any GLP-1 agent doesn’t confer any cardioprotective value for those without a prior history of CV disease. Rather, it implies that CV benefit may take much longer to appear in a clinical trial enrolling low-risk patients, and that EXSCEL may have been “less powered” vs. LEADER in this regard. The hazard ratio for three-point MACE was 0.90 (95% CI: 0.82-1.00) in favor of Bydureon among EXSCEL participants with a prior CV event, 0.99 (95% CI: 0.77-1.28) among those without any history of CV disease. Dr. Holman’s discussion of less overall drug exposure in the present study vs. LEADER alluded to the less-than-ideal adherence to once-weekly exenatide injections. We might expect longer exenatide exposure to correlate with greater CV benefit, and to this end, we’re eager for more in-depth analysis of the adherence data from all GLP-1 agonist CVOTs side-by-side. While some have noted that Intarcia’s FREEDOM-CVO trial of ITCA 650 (implantable mini pump offering three-six months of continuous exenatide release, circumventing the adherence issue) was also neutral, showing CV safety but not efficacy vs. placebo, we also think that’s more about trial design and interest in speed to show safety.
• Dr. Holman positioned EXSCEL as a reassuring safety dataset for Bydureon, as it should “dispel concerns” related to pancreatitis, pancreatic cancer, medullary thyroid cancer (MTC), and severe hypoglycemia. Moreover, he highlighted the 14% relative risk reduction for all-cause mortality in the trial (HR=0.86, 95% CI: 0.77-0.97). The pre-defined statistical testing plan precludes us from drawing conclusions about this as statistically significant, but all-cause death is clearly an important outcome (if not the most important outcome), and the lower frequency in exenatide vs. placebo arms is a valuable insight for clinical practice, according to Dr. Holman. We agree – this trial overall supports Bydureon as a safe and effective diabetes therapy, one that could benefit many patients in the real world (especially considering the very low proportion of the type 2 population that is on any GLP-1 agonist). While many questions remain about how Bydureon compares to Victoza and other in-class competitors, we pause to note incredibly important contributions from EXSCEL to the diabetes field: The study lends additional evidence in support of this advanced therapy class, and in our view, doesn’t conclusively refute (at least not yet) that CV benefits could be a GLP-1 class effect (in fact, we think it reinforces this).

Commentator

Francesco Giorgino, MD (University of Bari Aldo Moro, Italy)

Tasked with providing independent commentary on EXSCEL, Dr. Francesco Giorgino weighed positive and neutral results from the Bydureon CVOT, sharing a fairly balanced view overall. The suggestion of risk reduction for all-cause mortality (HR=0.86, 95% CI: 0.77-0.97) is certainly a positive, according to Dr. Giorgino, even though this finding cannot be deemed statistically significant because exenatide didn’t show superiority on the primary endpoint. He pointed to the neutral effect of Bydureon on heart failure hospitalization as another positive (HR=0.94, 95% CI: 0.78-1.13), given some lingering concerns in the diabetes community over incretin therapies and heart failure risk (this stems primarily from the SAVOR-TIMI CVOT of AZ’s DPP-4 inhibitor Onglyza, as neither LEADER nor SUSTAIN 6 found a significant signal for heart failure hospitalization). That the hazard ratio for individual components of three-point MACE (non-fatal MI, non-fatal stroke, and CV death) all trended in the right direction is also good news, as this offers a more compelling and comprehensive case for CV safety and falls in line with the CV efficacy data seen in LEADER for Novo Nordisk’s GLP-1 agonist Victoza (liraglutide). Indeed, Dr. Giorgino likened EXSCEL most closely to LEADER among all GLP-1 agonist CVOTs. He reported a number needed to treat (NNT) of 98 for liraglutide treatment over three years to prevent one death from any cause vs. an NNT of 106 for exenatide treatment over three years. As to why the primary endpoint result in EXSCEL missed superiority, Dr. Giorgino circled back to pragmatic trial design (no run-in period) and a larger primary prevention cohort (27% vs. 19% in LEADER, 0% in ELIXA for Sanofi’s lixisenatide, and 17% in SUSTAIN 6). He mentioned that while A1c-lowering was similar with Bydureon in EXSCEL and with Victoza in LEADER, the former demonstrated no significant risk reduction for severe hypoglycemia, which may have influenced observed outcomes (liraglutide was associated with a 20% risk reduction for confirmed hypoglycemia <54 mg/dl, p<0.001). Further, he reminded the audience that Bydureon reconstitution kits probably decreased adherence from where it would have been with a prefilled pen. That said, Dr. Giorgino found fault with some key arguments in defense of EXSCEL’s pragmatic design as well. For one, SGLT-2 inhibitors were also used at a similar rate in the placebo arm of SUSTAIN 6, which was positive for cardioprotection regardless. Moreover, investigators emphasized the infrequent clinical visits to minimize interference with usual care, but Dr. Giorgino pointed out that provider visits in LEADER were equally infrequent, once every six months after the first six months. These were important criticisms to be articulated, and we look forward to much more discussion on RCTs vs. the real world going forward. We’re eager for a more in-depth look at concomitant medications, patient/provider interactions, and adherence in all GLP-1 agonist CVOTs side-by-side.

• Ultimately, Dr. Giorgino expressed his opinion that GLP-1 agonists are beneficial in terms of their CV effects, but it’s still unclear which target type 2 diabetes population is best. He described how the pleotropic effects of drugs in the class may lead to CV benefit, and outlined exenatide’s direct effects on the CV system including anti-inflammatory responses and anti-atherosclerotic responses. Glycemic efficacy has little to do with cardioprotection in EXSCEL or other diabetes CVOTs, in Dr. Giorgino’s view, since other large studies like ACCORD, ADVANCE, VADT, and UKPDS did not show a reduction in all-cause mortality despite substantial A1c drops.
• Very notably, Dr. Giorgino did not give much credence to the argument dividing GLP-1 agonists into exendin-4-based molecules (exenatide, lixisenatide) and human GLP-1-based molecules (liraglutide, semaglutide). While the variations on molecular structure may result in products with slightly different signaling and biological effects, he explained that these variations are unlikely to cause meaningful differences in CV effects. We’ll be following this debate closely in the weeks and months ahead.

Close Concerns Questions

Q: How will the diabetes community respond to EXSCEL, a convincing demonstration of Bydureon’s CV safety and a narrow miss for CV benefit? Do these results add or subtract from the notion of a cardioprotective class effect for GLP-1 agonists?

Q: How big a role did poor adherence play in EXSCEL’s neutral result? Would the same CVOT using a prefilled exenatide pen (or better yet, the upcoming Bydureon autoinjector) yield a positive result?

Q: How did elements of pragmatic trial design influence outcomes in EXSCEL? What was the impact of SGLT-2 inhibitors prescribed to placebo-treated patients, and how does this compare to SGLT-2 use in SUSTAIN 6?

Q: Will the field move toward standardization of CVOT design? Will this start with FDA?

Q: In what form (if any) will EXSCEL data be incorporated onto the Bydureon product label? How will this affect AZ’s marketing around its GLP-1 agonist franchise?

Q: Is there a meaningful divergence in CV effects between exendin-4-based vs. human GLP-1-based molecules? What kind of study (mechanistic or outcomes) might investigate this further?

Q: If EXSCEL’s neutral result does in fact have more to do with pragmatic trial design than with the molecule itself, how can we explain the neutral findings from FREEDOM-CVO for Intarcia’s ITCA 650 (implantable exenatide mini pump)? 

Canagliflozin Cardiovascular Assessment Study (CANVAS)

Design, Methods, and Patient Characteristics

Greg Fulcher, MD (University of Sydney, Australia)

Dr. Greg Fulcher reprised his role from ADA 2017, speaking first during this CANVAS symposium to review trial design and baseline characteristics of the study population. Notably, J&J’s CVOT for SGLT-2 inhibitor Invokana (canagliflozin) integrated data from two separate studies: (i) CANVAS (n=4,330), initiated in 2010, randomized type 2 patients to canagliflozin 100 mg, canagliflozin 300 mg, or placebo, while (ii) CANVAS-R (n=5,812), initiated in 2014, randomized people to canagliflozin 100 mg that could be up-titrated to 300 mg, or to placebo. In total, the CANVAS program enrolled 10,142 participants across 30 countries and 667 clinical sites in North America, Latin America, Europe, and Asia Pacific. The primary endpoint was a three-point MACE composite (non-fatal MI, non-fatal stroke, and CV death), and mean follow-up time when taking the two trials together was 188 weeks (~3.6 years). In discussing patient characteristics/disposition in CANVAS, Dr. Fulcher focused on the mean age of 63 years-old in both canagliflozin and placebo groups, with a mean diabetes duration of 14 years. He shared that ~36% of participants were female, attributing the majority male subject pool to higher CV risk in the latter. According to Dr. Fulcher, ~90% of patients had hypertension at baseline, while ~14% had a history of heart failure and ~66% had some prior history of CV disease. Nearly 80% of individuals enrolled were white, and almost as many were on background metformin treatment (~50% on insulin, ~43% on sulfonylureas, ~12% on DPP-4 inhibitors). Baseline A1c was 8.2% in both arms, with a mean BMI of 32 kg/m².

Effects on Intermediate Biomarkers and Beta Cell Function and CV Outcomes

David Matthews, MD (University of Oxford, UK)

Dr. David Matthews discussed the effect of canagliflozin on intermediate biomarkers and beta cell function in CANVAS. He revealed that canagliflozin dose (100 mg or 300 mg) had no effect on the previously-described mean A1c change of -0.6% with the drug vs. placebo. On the beta cell front, he reported that both doses of canagliflozin produced improvements in beta cell function as measured by HOMA2-%B after one year. From a baseline of ~56%, HOMA2-%B grew to 68% for canagliflozin 100 mg and 72% for canagliflozin 300 mg (vs. 60% for placebo, p<0.0001 for both comparisons). We are certainly intrigued by the potential for a beta cell protective effect with canagliflozin therapy, and we eagerly await longer-term data to determine whether this effect is real and endures over time. To our knowledge, this restoration of beta cell function has not been described in the SGLT-2 inhibitor class to-date, so we’re particularly keen to hear more commentary from thought leaders on how meaningful this is, and what the mechanism could be.

• Without missing a beat, Dr. Matthews transitioned straight into the CV outcomes from the CANVAS trial. After reviewing the original results first presented at ADA 2017, he pointed to some baseline differences between the primary prevention (n=3,486) and secondary prevention (n=6,656) cohorts: Namely, the secondary prevention cohort had a higher baseline rate of heart failure (18% vs. 8%), as well as significantly higher use of cardioprotective agents like statins (81% vs. 63%), antithrombotic agents (87% vs. 49%) and beta blockers (64% vs. 33%). Background use of other diabetes therapies (metformin, insulin, SUs, DPP-4 inhibitors, and GLP-1 agonists), as well as RAAS inhibitors and diuretics did not differ significantly between the primary prevention and secondary prevention cohorts, nor did mean age, sex, diabetes duration, hypertension, or a variety of risk factors such as starting A1c, BMI, blood pressure, and cholesterol. We learned earlier that there was no heterogeneity of benefit for higher-risk vs. lower-risk participants in the CANVAS program (that is, the p-value for interaction was non-significant), but these baseline differences are still important to note as they could have skewed the results in some way. In all likelihood, this is not the last we’ll hear on subgroups in CANVAS, and we’ll keep our eyes and ears peeled for further detailed analyses.

Renal Outcomes

Dick De Zeeuw, MD (University Medical Center Groningen, The Netherlands)

Though CANVAS was primarily a cardiovascular outcomes trial, the microvascular findings on Invokana’s renal benefits packed quite a punch as well. Dr. Dick De Zeeuw summarized renal data from the integrated program (highlighting very similar values to EMPA-REG OUTCOME), shared new insights specific to CANVAS-R, and emphasized the importance of the ongoing CREDENCE trial (expected to complete in June 2019) to fully grasp canagliflozin’s impact in diabetic nephropathy. In CANVAS-R, canagliflozin was associated with an increase in eGFR in year two of treatment. In a subsequent off-treatment period (median time 30 days), Dr. De Zeeuw showed how this line continues on its upward slope, indicating further increase in eGFR even after people have stopped taking the oral SGLT-2 inhibitor. Right now, this data is merely “suggestive of a renal protective profile,” as Dr. De Zeeuw put it, but CREDENCE could draw more definitive conclusions about Invokana’s beneficial effects on kidney function. Interest in SGLT-2 inhibitors for kidney disease has essentially tripled of-late, with Lilly/BI and AZ launching their own dedicated studies of empagliflozin (Jardiance) and dapagliflozin (Farxiga) respectively in chronic kidney disease (CKD). At ESC 2017, Dr. David Fitchett hinted that patients with eGFR <60 ml/min/1.73m² may actually stand to benefit the most from SGLT-2 inhibitor therapy in terms of renal and cardioprotection, and we’re excited about this new direction for an advanced diabetes therapy class.

Adverse Events and Safety Considerations

Bruce Neal, MD (The Georgia Institute for Global Health, Sydney, Australia)

Dr. Bruce Neal, first author on the CANVAS paper published in NEJM, shared new, more granular data on amputations occurring during the trial. This topic is of great interest to many in the diabetes field, including us, since it was revealed during the original CANVAS full results presentation at ADA that canagliflozin increased risk for lower limb amputations nearly two-fold vs. placebo (HR=1.97, 95% CI: 1.41-2.75, p<0.001). Dr. Neal discussed a dozen or so risk factors that were associated with lower-extremity amputations according to a univariate analysis of the CANVAS safety data, explaining that “by far the strongest relative risk came from having a prior amputation” (HR=21.4), though he also listed peripheral vascular disease (HR=2.5), neuropathy (HR=3.4), history of CV disease (HR=2.9), nephropathy (HR=2.2), baseline insulin use (HR=2.4), baseline A1c >8% (HR=2.0), male gender (HR=2.6), retinopathy (HR=2.3), use of loop diuretics (HR=2.1), baseline eGFR <45 ml/min/1.73m² (HR=1.8), and diabetes duration ≥10 years (HR=1.6). His slides then advanced to display findings from a multivariate analysis: prior amputation (HR=20.9), peripheral vascular disease (HR=3.1), male gender (HR=2.4), neuropathy (HR=2.1), A1c >8% (HR=1.9), presence of CV disease at baseline (HR=1.5), and canagliflozin treatment (HR=1.8) remained as variables significantly increasing an individual’s amputation risk. Dr. Neal reinforced that canagliflozin treatment, independent of all other risk factors, still nearly-doubled amputation risk – in other words, these new results don’t explain away the amputation signal seen in CANVAS, attributing it to something other than the molecule, but they do pinpoint factors to consider as we push for more diligent monitoring and enhanced patient education around foot care in diabetes. Moreover, the multivariate analysis identifies possible reasons that CANVAS may have found an amputation signal not seen in EMPA-REG OUTCOME (in which, Dr. Neal reminded everyone, amputations were collected using a different process) – if, for example, Lilly/BI’s CVOT featured much less peripheral vascular disease, neuropathy, or prior history of amputations in its baseline study population (we’re eager to do a deeper dive on this). Looking ahead, Dr. Neal suggested that AZ’s DECLARE CVOT for Farxiga (dapagliflozin) will shed more light on SGLT-2 inhibitors and amputations. He underscored that other phase 3 trials of canagliflozin (not counting CANVAS) have reported only 10 lower-extremity amputation events total, in 8,114 participants, and that Truven observational results reflect no imbalance (HR=0.98, 95% CI: 0.68-1.41).

• Moreover, Dr. Neal described common precipitating events that preceded lower limb amputations in CANVAS. Infection was the most common (67/140 amputations in the canagliflozin arm, 27/47 amputations in the placebo arm), followed by gangrene (53 and 13 events in the canagliflozin and placebo groups, respectively), peripheral arterial disease (46 and 13 events), ulcers (43 and 17 events), acute limb ischemia (18 and 3 events), and neuropathy (18 and 10 events). This fits with a recurring theme we’ve been hearing from several thought leaders – amputations are a soft endpoint (meaning patients/providers make a decision about whether or not to have one), and can often be avoided if care teams are diligently monitoring for these major precipitating factors. Once again, we note the tremendous opportunity here to improve foot care in real-world diabetes management (we’d love for J&J to show leadership in this initiative, alongside other SGLT-2 manufacturers), starting with education on what key features to monitor for.
• On fractures, Dr. Neal discussed conflicting trends: More fractures were seen with canagliflozin vs. placebo in CANVAS, but fewer were seen in the SGLT-2 arm in CANVAS-R. Among adjudicated low-trauma fractures, there was no imbalance, but among all adjudicated fractures, canagliflozin was associated with significantly more vs. placebo across the integrated program (HR=1.26, 95% CI: 1.04-1.52). According to Dr. Neal, there is no clear reason for this discrepancy as of yet, and so uncertainty persists around canagliflozin and bone fractures. As he said at ADA, “I’m not easily persuaded by things happening by chance, but this is a weird result. We don’t see a fracture signal in EMPA-REG OUTCOME, nor in CANVAS-R, nor in any of the other CANVAS trials, so it’s possibly by chance.” He concluded this portion of his talk with a mention of CREDENCE – this upcoming outcomes trial of canagliflozin in diabetic kidney disease will hopefully help clarify the fracture trend.

Implications for Clinical Practice

Bruce Neal, MD (The Georgia Institute for Global Health, Sydney, Australia)

Dr. Bruce Neal remained on stage to contextualize CANVAS trial results, and this talk focused on a forthcoming publication from the George Institute that analyzes 82 SGLT-2 inhibitor studies in conjunction. The meta-analysis covers canagliflozin, empagliflozin (Lilly/BI’s Jardiance), and dapagliflozin (AZ’s Farxiga), as well as three agents developed and marketed in Japan (luseogliflozin [Taisho Pharmaceuticals’ Lusefi], ipragliflozin [Astellas’ Suglat], and phase 3 tofogliflozin [Chugai Pharma]). The results suggest a favorable risk/benefit profile for SGLT-2 inhibitors overall, with trends toward cardioprotection and renal benefit, plus minimal safety signals (except elevation in the risk for amputation and fracture). Dr. Neal described how the meta-analysis also shows some key differences between the various SGLT-2 inhibitors, mostly on the safety front, but he was careful to preface his remarks by cautioning that it is difficult to compare between trials that involved different populations and different methodologies.

• The meta-analysis found an overall trend toward risk reduction for major CV events (HR=0.85, 95% CI: 0.77-0.93, p=0.504). Data from canagliflozin and empagliflozin indicates a significant risk reduction for CV death (HR=0.75, 95% CI: 0.65-0.87, p=0.024), as well as a trend toward risk reduction for non-fatal MI (HR=0.84, 95% CI: 0.73-0.98, p=0.648), for unstable angina (HR=0.95, 95% CI: 0.73-1.24, p=0.380), for heart failure hospitalization (HR=0.67, 95% CI: 0.55-0.80, p=0.936), and for all-cause death (HR=0.79, 95% CI: 0.70-0.88, p=0.110). Across all of these outcomes, Dr. Neal noted the “striking comparability” in individual point estimates for canagliflozin and empagliflozin, with two notable exceptions: CV death (which favored empagliflozin with a hazard ratio of 0.62 vs. 0.87 for canagliflozin) and non-fatal stroke (which favored canagliflozin with a hazard ratio of 0.91 vs. 1.24 for empagliflozin). He attributed these differences to chance, given the small number of events.
• On renal outcomes, the meta-analysis showed a trend toward risk reduction for albuminuria progression (HR=0.72, 95% CI: 0.67-0.77, p=0.488) and serious decline in kidney function (HR=0.59, 95% CI: 0.49-0.71, p=0.793). This is based on data from canagliflozin and empagliflozin alone.
• On safety, the meta-analysis revealed that the data thus far on the SGLT-2 inhibitor class shows no significant increase in the overall risk for cancer (HR=1.05, 95% CI=0.93-1.18, p=0.83), metabolic acidosis (HR=1.30, 95% CI=0.59-2.86, p=0.45), or thromboembolism (HR=0.91, 95% CI=0.62-1.34, p=0.68). On the other hand, the class is associated with an increased risk of fracture (HR=1.05, 95% CI=0.92-1.20, p=0.01) and amputation (HR=1.44, 95% CI=0.1.13-1.83, p=0.01). Furthermore, some interesting within-class differences emerged on the safety front, including:
  
  • A trend toward higher risk of metabolic acidosis with canagliflozin vs. empagliflozin and luseogliflozin;
  • A trend toward lower risk of thromboembolism with empagliflozin vs. canagliflozin, dapagliflozin, and luseogliflozin;
  • A trend toward higher risk of fracture with canagliflozin than dapagliflozin and empagliflozin;
  • And, famously, a trend toward higher risk of amputation with canagliflozin vs. empagliflozin.

Commentator

Ele Ferrannini, MD (University of Pisa, Italy)

In an engaging presentation to conclude EASD 2017 (this symposium was in the last Friday slot), University of Pisa’s Dr. Ele Ferrannini offered strong opinions on CVOT design and SGLT-2 inhibitor class effects. He suggested the field move away from three-point MACE as a primary outcome, referring to this endpoint as “a bit of a salad.” The field’s knowledge of SGLT-2 agents, for example, is complicated by the fact that canagliflozin and empagliflozin both showed 14% risk reduction for three-point MACE, while strokes trended in the wrong direction in EMPA-REG OUTCOME even though the relative risk reduction for CV death was greater (38% vs. 13% in CANVAS). Dr. Ferrannini also alluded to some irony in the fact that a non-fatal MI counts as “plus one” in terms of a CV event but “minus one” in terms of avoiding a CV death. Also, he pointed out that MACE lumps together MI and stroke, which have partially different sets of risk factors. Studies dedicated to MI alone, stroke alone, or heart failure alone might reveal more valuable and clinically-applicable information for diabetes practice, he argued. These are certainly clinical trials we’d like to see, and we’re glad Lilly/BI and AZ have launched dedicated investigations of their SGLT-2 inhibitor products in heart failure, but we also acknowledge the massive investment of time and resources that goes into a CVOT and the need for some practicality in selecting endpoints. Moreover, diabetes CVOTs published to-date are already exceptionally difficult to compare due to differences in study design and participant pool, and losing three-point MACE as a primary outcome might move us further away from standardization on this front (of course, this raises another debate over the advantages and drawbacks of CVOT standardization). Nonetheless, we thought this was a very interesting argument laid out by Dr. Ferrannini, and we look forward to collecting more insights and perspectives from other thought leaders. He continued his presentation by establishing the likelihood of cardioprotective and renal protective class effects among all SGLT-2 inhibitors, pointing to the congruence of findings between CANVAS and EMPA-REG OUTCOME on these key macrovascular and microvascular endpoints. Safety may be where SGLT-2 agents differ. That said, Dr. Ferrannini explained that amputations are a procedure-driven endpoint (or a soft endpoint, with decisions left up to patient/provider). He urged HCPs to look for precipitating factors (infection, gangrene, etc.) and to monitor carefully instead of shying away from Invokana (or worse yet, from all SGLT-2 inhibitor treatment options). We echo this view completely, and we also appreciated Dr. Ferrannini’s call for more real-world evidence on amputations, which could be most valuable in elucidating the signal.

Continuous Glucose Monitoring Before and During Pregnancy in Women with Type 1 Diabetes: Results from CONCEPTT; a Multicentre Multinational Randomized Controlled Trial

Results

The JDRF-funded CONCEPTT RCT testing CGM in pregnant women (n=215) showed positive neonatal outcomes with Medtronic’s older Guardian CGM. Though not the primary endpoint, significantly improved neonatal outcomes were the headline – CGM drove a significant  reduction in the incidence of large for gestational age (OR=0.51, p=0.02), fewer NICU admissions lasting 24+ hours (OR=0.48, p=0.02), fewer incidences of neonatal hypoglycemia (OR=0.45, p=0.03), and one-day shorter length of hospital stay (p=0.01). The numbers needed to treat (NNT) were compelling – NNTs of just 6-8 women with CGM to prevent one of those negative outcomes. The primary A1c endpoint showed a small -0.2% A1c advantage for CGM at 34 weeks (p=0.02). However, mothers on CGM spent a significant 100 more minutes/day in range (68% vs. 61%; p=0.003), 72 fewer minutes/day in hyperglycemia (27% vs. 32%; p=0.03), and a non-significant ~14 fewer minutes per day in hypoglycemia (3% vs. 4%; p=0.1). Results were published in The Lancet, a major visibility win!

As expected with the older Guardian sensor, wear time was lower than in more recent CGM studies – 70% of pregnant participants used CGM for 75%+ of the time. In addition, ~80% of women reported frustrations with the CGM device. We brought this limitation up in Q&A (it was not mentioned), and would guess the trial probably underestimated current CGM’s potential benefit in pregnancy. What would outcomes have looked like with G4/G5, FreeStyle Libre, or Guardian Sensor 3? Since the trial took three years to run, a year to plan, and spanned six countries (Canada, UK, Spain, Italy, Ireland, and the US), getting the latest devices in was obviously a challenge. The study concludes that “CGM should be offered to all pregnant women with type 1 diabetes using intensive insulin therapy” – hear, hear! According to The Lancet publication, it’s also the first study to indicate potential for improvements in non-glycemic health outcomes from CGM use. Nice!

There were a few big surprises from this trial: (i) pump+CGM outcomes looked worse on a few notable endpoints vs. MDI+CGM outcomes (see below); (ii) CGM had no significant benefit on severe hypoglycemia; and (iii) the difference in outcomes was quite large between some countries (a fascinating source of commentary from Dr. Elisabeth Mathiesen). See more details below. We look forward to cost-effectiveness data when it is published. The trial also included a second arm testing CGM in women planning pregnancy, but showed no A1c benefit at 24 weeks or conception.

• An editorial from Dr. Satish Garg in The Lancet was very positive, with a nice beyond-A1c mention: “We believe that the CONCEPTT results support CGM use during pregnancy for all women with type 1 diabetes and time in range might become an important measure in pregnancies associated with type 1 diabetes; thus endocrine and obstetric medical societies could consider advocating or recommending revising their guidelines accordingly.” We wonder if this study could help validate time-in-range as a meaningful surrogate endpoint, independent of A1c. The study follows a very positive hybrid closed loop study during labor/pregnancy from Dr. Helen Murphy and colleagues at Cambridge, which was published in NEJM in 2016 – automation is a definitely an exciting frontier for pregnancy, along with use of next-gen devices! Dr. Murphy did mention in Q&A that a FreeStyle Libre in pregnancy study is ongoing.
• Neonatal outcomes – paramount in a diabetes pregnancy study – were very positive in favor of the CGM group. Babies from mothers who wore CGM were less likely to be large for gestational age (LGA; >90^th percentile – 53% in CGM group vs. 69% in control group; p=0.02), lower incidence of neonatal hypoglycemia requiring IV glucose (15% vs. 28%; p=0.03), were less likely to require NICU admissions >24 hours (27% vs. 43%; p-0.02), and had significantly lower median customized centile (a measure of birthweight standardized for maternal ethnicity, height, weight, and neonatal sex and gestational age at delivery – 92% vs. 96%; p=0.05). Further, infants from mothers in the CGM group had hospital stays reduced by nearly a full day (3.1 days vs. 4.0 days; p=0.02). Co-PI Dr. Denice Feig pointed out that not only were the median customized centiles lower in the CGM group (across each of the four study sites), but the lower portion of the box plot was much wider, indicating that many more babies were closer to the normal weight range than in the control group. The numbers needed to treat (NNT) were quite compelling: Six women with CGM prevented one event of LGA; eight women with CGM to prevent one event of neonatal hypoglycemia; and six women with CGM to prevent one NICU admission over 24 hours. Economic analysis wasn’t presented, but we’d guess that 72-96 months of CGM use (nine months per pregnancy * NNT) would be cost-effective relative to those expensive negative events. There were no differences in serious adverse pregnancy outcomes (miscarriage, stillbirth, termination, or congenital anomaly), obstetric outcomes (hypertensive disorders in pregnancy, C-section, maternal weight gain, maternal length of hospital stay), or gestational age at delivery between groups.
  
  • We would’ve loved to see primary outcomes beyond A1c, but a JDRF representative told us that a primary neonatal outcome would’ve required much more statistical power – he estimated 10,000 participants, compared to the 215 pregnant women in this study! We’re not all that familiar with pregnancy outcomes and their frequency, but in this case, it seems to have made sense to not power for neonatal events. The same rep also mentioned that there will be some follow-up of both baby cohorts to look for long-term impacts of maternal CGM use on offspring development.
• A subgroup analysis of pump vs. MDI CGM users revealed that pump users achieved lower reductions in A1c (-0.32% vs. -0.55%; p=0.001) and slightly less time spent in hypoglycemia (according to the speaker, though we didn’t notice a large difference in the publication’s appendix) However, pumpers had concerning higher rates of gestational hypertension (14.4% vs 5.2%; p=0.02), preterm delivery (43.2% vs 36.2%; p=0.04), and NICU admissions for greater than 24 hours (44.5% vs. 29.6%; p=0.01) than MDI users. This result was surprising, especially since pumpers at baseline seemed to be more engaged with their health (lower rates of smoking, take preconception vitamins, booked appointments earlier, …). Infants of pump users also experienced higher rates of hypoglycemia requiring IV dextrose (31.8% vs. 19.1%; p=0.03), although once adjusted (not specified, but presumably for baseline maternal characteristics), the difference was no longer significant. There were no other observed significant differences in diabetes status or neonatal outcomes. These results are disappointing for pump users, as it is more expensive therapy that did not deliver better outcomes in this study. Why? We wonder if pumps psychologically loosen eating restrictions, while the hassle of injections might make expectant mothers more likely to hesitate before eating or choose different foods (pure speculation on our part). We look forward to seeing the cost-effectiveness data, and certainly, this is very positive news for MDI+CGM.
  
  • Could higher proportions of MDI users in clinics that tend to have better outcomes (see Dr. Mathiesen’s call to “Learn from Spain” below) underlie MDI vs. pump discrepancies? In other words, Spain had a significantly lower rate of LGA than a number of other countries (even in the control group vs. other countries’ CGM groups) – were Spanish patients simply more likely to be on MDI?  
• The 0.2% A1c advantage for CGM at 34 weeks was characterized as “small,” and in our view, outweighed by the time-in-range and time-in-hyperglycemia data. We must go Beyond A1c in studies like this! CGM users saw a 0.54% decline in A1c by 34 weeks (baseline: 6.8%) vs. a 0.35% decline in the control group (baseline: 6.95%) – the 0.2% difference was statistically significant (p=0.034), though it missed the goal for a 0.5% difference and comes in smaller than recent studies like DiaMonD and GOLD. Strangely, ~20% of A1c samples were missing, though the investigators do not believe it impacted results. The time in range (63-140 mg/dl) data were very strong: pregnant women on CGM spent 100 more minutes per day in the tight target range and 72 fewer minutes per day in hyperglycemia relative to the control group at 34 weeks. Time in hypoglycemia did not statistically favor CGM (~14 minutes better with CGM), given low rates in both groups (3% vs. 4%; p=0.1). Coefficient of variation favored the CGM group at 34 weeks (32% vs. 34%), but just barely missed statistical significance (p=0.058).
• Despite raising a number of concerns about the study, and especially, diabetes pregnancy care overall, Dr. Elisabeth Mathiesen (Head of Diabetes Treatment at the Copenhagen Centre for Pregnant Women with Diabetes) concluded that CGM “is the future” and that the CONCEPTT study “will change the future for pregnant women with diabetes.” She began by acknowledging that CGM proved effective in the study, reducing A1c greater than in the control group and preventing many adverse neonatal events. However, she found the lack of an effect on maternal severe hypoglycemia “a little disturbing” and pointed to the downsides of using CGM (80% of users reported problems and 30% used the sensors less than 75% of the time) – again, we believe the use of a better sensor would’ve made a difference on both of these points. Other points of weakness for Dr. Mathiesen were generalizability (the study took three years to enroll at 31 centers) and moderate endpoint data collection (86% of A1c values; 77% of sensor data). In light of this data, will Dr. Mathiesen implement CGM in all pregnant women as suggested by the study authors? In favor of “yes,” she said, many will ask for it, the study recommends it, and her “stomach feeling” is that CGM is the future. On the other hand, she questioned the quality of treatment in the study overall (more in first bullet below), and worried that the cost of CGM will prompt reduction in other aspects of care –  “the cost of CGM use in 20 pregnant women is the cost of the salary of one working nurse.” Conversely, if CGM prevents expensive neonatal outcomes, it should drop healthcare resource – it is an upfront investment with upside, especially in pregnancy, when there are two bodies who can benefit. In our opinion, pregnant women with diabetes, who clearly have a lot at stake and are asked to keep their blood glucose levels to very tight ranges (63-140 mg/dl in this study), need access to their glucose values at all times to stay in range most of the day . The glycemic findings from this trial were not as strong as some would have hoped, but they still suggest definite benefit (+100 minutes/day), in tandem with improved neonatal outcomes. We hope that economic analysis and/or follow-up studies with more accurate and user-friendly sensors will tip consensus more toward an unequivocal “yes.”
  
  • Dr. Mathiesen pointed out the care that pregnant women were getting in the trial, even in the CGM group, presenting a slide with the simple question: “Treatment of Excellence?” She questioned whether a lack of A1c guidelines for pregnancy in the two main countries of this study (Canada and UK) impacted quality of care. In the CGM group of CONCEPTT, A1c late in pregnancy was 6.4%, there was a 38% rate of preterm delivery, and 53% rate of LGA. While these figures compared favorably to the control group, “routine results” from her center in Copenhagen are significantly better: In pre-publication data from this clinic (n>400), A1c late in pregnancy is 6.1%, preterm delivery rate is 17%, and LGA rate is 38%. Dr. Mathiesen then showed a figure from earlier in the presentation depicting LGA rates from the control and CGM groups in Canada, the UK, Italy, and Spain (below), and the clear message “Learn from Spain.” Indeed, the rate of LGA in the Spanish control group was less than half that in the other countries’ control groups, and even lower than that of the CGM groups in two of the three other countries! This discrepancy, to her, is “probably more important” than the influence of CGM on pregnancies. She implored attendees to “please listen to how [Dr.] Rosa [Corcoy] (an endocrinologist in Barcelona) is treating her patients. It probably has to do with diet – carb counting, weight gain, and attitude.”
    • Dr. Murphy gave the perfect response to Dr. Mathiesen’s criticism: “No doubt, if we could clone Dr. Mathiesen and put her in every center, we could get much better glucose control. This took place in 31 centers. What we have is generalizable. A1c, with all standardizations, is still different. It’s not correct to compare a single center to labs across 31 sites. I do agree that CGM is coming and one of the important findings is the large difference across groups in CGM measures with small impact on A1c. In previous studies, we did not have CGM. An extra 100 minutes in target; most women would take that if they don’t have access to Dr. Mathiesen.”

• During Q&A, Dr. Feig noted that we don’t know the relative contributions of glucose control throughout the whole pregnancy vs. in the 24-48 hours pre-delivery on neonatal outcomes. Existing literature, she said, is “quite variable” about the contribution during labor, but she believes it may contribute a little. It’s an interesting scientific question, especially as we think about automated insulin delivery and the previous finding from Dr. Murphy’s NEJM study. Theoretically, better glucose control with less hypoglycemia would be good for the mother, so even if 100% of the contribution on neonatal glycemia were from the labor period and just prior, we still hope that the mother could be given access to CGM.
• Dr. Mathiesen began her commentary by applauding the PIs – Drs. Feig and Murphy – for their bravery in forging ahead and conducting an RCT in pregnancy. “Until these two women dared to take it on, no one would dare. These women are truly brave.” Hear, hear!

Symposium: Diabetic Hypoglycemia: New Thinking, New Tools

Summary

The International Hypoglycemia Study Group hosted an information-packed symposium at EASD, imparting some of their endless knowledge on hypoglycemia prevalence, pathogenesis, therapeutic pipeline, and more.

• Hypoglycemia is certainly recognized to be a huge problem, but statistics presented by Dr. Ulrik Pedersen-Bjergaard were still alarming. These figures demonstrate how, even though hyperglycemia hospitalizations have gone down and overall A1c has decreased, hypoglycemia remains a big limitation to better therapy. As CGM use rises further, we expect a lot more hypoglycemia that has traditionally gone undetected will surface, hopefully adding more urgency to the discussion. Could we ever see a day where drugs are indicated for the reduction of hypoglycemia (independent of A1c)?

• Denmark’s Dr. Pedersen-Bjergaard emphasized that the correlation between hypoglycemia and A1c is not strong, likely due to differential glycemic variability from patient-to-patient. In the DCCT, intensive control did increase the risk of severe hypoglycemia relative to the control arm, but people with recurrent hypoglycemia or impaired awareness of hypoglycemia were excluded from the study. Recently, the HAT study demonstrated that the association between A1c and severe hypoglycemia is actually much less significant in the broader population – there is a slightly significant relationship (with a small effect size) between A1c and any hypoglycemia in type 1, but not in type 2. Nocturnal and severe hypoglycemia were not associated with A1c at all. 2017 data from EDIC, the DCCT follow-up, now shows that the relationship between lower A1c and increased rate of severe hypoglycemia is blunted, even though in the same cohort of patients. Why? Dr. Pedersen-Bjergaard suggested that treatment improvements may underlie the differences, though patients outside of the intensive arm may have developed impaired hypoglycemia awareness or other risk factors.
  
  • Audience members seemed confused by this counterintuitive tenet, as one asked for clarification during Q&A. The panel dismissed explanations including glycemic variability, a “disengagement with one’s own self-management,” and a clustering of risk factors in a few individuals. We believe the evidence against a solid relationship between A1c and hypoglycemia should be preached from the mountain tops, as it is the best argument against relaxing glycemic control in hypoglycemia-prone individuals.
• Nocturnal hypoglycemia actually reduces sleep efficiency, increasing delta power and slow wave sleep at the cost of decreased REM sleep, increased wakes, and increased sleep latency (time to sleep) once awoken. This is why people with diabetes often feel exhausted when they wake in the morning, and why morning exhaustion is a possible indicator of prior hypoglycemia. Belgium’s Dr. Bastiaan de Galan also contested the “myth” of the Somogyi effect (elevated blood glucose in the morning as an indication of nocturnal hypoglycemia, or “posthypoglycemic hyperglycemia”). Low fasting glucose in the morning is still an accepted possible sign of nocturnal hypoglycemia, but the morning glucose rebound is no longer widely accepted.
• Scotland’s Dr. Rory McCrimmon showed data demonstrating that impaired awareness of hypoglycemia prevalence hasn’t changed in the last 30 years (~25%), and skimmed the surface of ongoing mechanistic and therapeutic research. The field hasn’t yet determined the main biological culprit(s) behind IAH, but it may involve glucose transport, alternate fuels (glycogen/lactate), glucose sensing (GK/AMPK/K_ATP), peripheral signals (opiates/steroids/cytokines), and/or neurotransmitter release. Whatever the mechanism, the net result is an increase in GABAergic (inhibitory) tone and a decrease in glutamatergic (excitatory) tone from glucose-sensing neurons. He showed preliminary evidence that IAH is a product of habituation to some repeated stimulus (a logical hypothesis): in a rodent model, his group was able to dis-habituate and restore physiological response to recurrent hypoglycemia with high-intensity exercise. There are countless other novel approaches to restore hypoglycemia awareness currently under investigation: (i) oral dehydroepiandrosterone; (ii) inhaled formoterol; (iii) opioid receptor blockade; (iv) naltrexone; (v) somatostatin type 2 antagonism; and (vi) oral diazoxide (K⁺ channel activator). Having a toolkit of agents that improve hypoglycemia awareness, and a pharmacogenomics approach that dictates which work best in which patients, would be a true game-changer. Of course, we also hope to see CGM and automated insulin delivery better studied in those with impaired hypoglycemia and high risk of severe hypoglycemia.
• Dr. Pedersen-Bjergaard introduced the IHSG patient and provider hypoglycemia risk stratification infographic tool, which will be available here soon. The tool separates patients by low, moderate, and high risk, offers solutions for each risk range, describes acute and long-term outcomes and risk factors, and providers risk-reduction strategies. This could be a very helpful resource, especially for those who aren’t seen after by one of the luminaries of the IHSG.
• Two days prior, IHSG member Prof. Brian Frier was given the honor of delivering the Camilo Golgi lecture, in which he indicated that consensus has been reached on thresholds for clinically relevant hypoglycemia bins – at 3.9 mmol/L (70 mg/dl; “alert level hypoglycemia”) and at 3.0 mmol/L (54 mg/dl; “significant hypoglycemia”). When we last heard discussion on hypoglycemia “bins” at the diaTribe Foundation’s Glycemic Outcomes Beyond A1c meeting in July, a breakout session moderated by Profs. Frier and Stephanie Amiel at that meeting expressed agreement on the cut-offs of <70 mg/dl and <54 mg/dl to divide the different classes of hypoglycemia, though had less unanimity on terminology. Prof. Frier told us that the International Hypoglycemia Study Group has agreed on how to call these bins since then but awaits consensus with other groups.
• Dr. McCrimmon explained why hypoglycemia may only be deleterious in the context of hyperglycemia, not on its own. Chronic hyperglycemia is an inflammatory stimulus that depletes the anti-oxidative response. When hypoglycemia comes along, it causes tremendous oxidative stress, but there are fewer reserves of anti-oxidants to restore homeostasis. This makes sense in theory, and was supported in a rodent study: Control mice performed fine on novel object recognition task (testing memory) after intermittent hypoglycemia, while STZ mice (who have chronic hyperglycemia) struggled immensely following the same intervention (while STZ mice not exposed to hypoglycemia did fine). At a molecular level, the STZ mice who had undergone the hypoglycemia treatment had significantly higher protein carbonylation, a biomarker of oxidative activity. Might eating a diet rich in anti-oxidants help to dampen the deleterious effects of hypoglycemia? The idea that the recovery from hypoglycemia and the control leading into it could be just as important as the event itself is fascinating.

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