European Association for the Study of Diabetes

48th Annual Meeting September 30, 2012 – October 5 2012; Berlin, Germany – Diabetes Technology

Table of Contents 

Executive Highlights

On the device front, Dexcom’s star-studded G4 Platinum corporate symposium was the highlight in CGM at EASD 2012, though late news on the European launch of Abbott’s Navigator II was also buzzworthy. The excitement for Dexcom’s new CGM (see our report on the post-EASD FDA approval call at was tangible during the company’s very well-received corporate symposium on the second day of the conference. We saw new cuts of G4 Platinum accuracy data from Dr. Thomas Peyser (VP Science and Technology, Dexcom, San Diego, CA), preliminary results from Dr. Bruce Buckingham (Stanford University, Stanford, CA) on a fantastic nocturnal remote monitoring study at a diabetes camp using the G4 Platinum CGM in pediatrics, and a sneak peak at Dexcom’s future pipeline (e.g., predictive algorithms, the accuracy of the G4 AP version) from the eloquent Dr. Jay Skyler (University of Miami Miller School of Medicine, Miami, FL). For the first time that we can recall, Dexcom’s future technology – combining remote monitoring, highly accurate CGM, and potentially predictive alerts – was positioned as an alternative to low glucose suspend (LGS). This would reinforce competition between Dexcom and Medtronic based not only on CGM device quality, but an approach to insulin therapy and hypoglycemia mitigation. Next-generation CGM news also came in an outstanding presentation by the esteemed Dr. Roman Hovorka (University of Cambridge, UK), where we learned that Abbott “silently” launched its FreeStyle Navigator II two to three weeks before EASD. We’re excited to hear this given the impressive accuracy and AP researcher enthusiasm for the original FreeStyle Navigator and hope to hear more about the device soon. Roche also emphasized its CGM in development (which it hardly ever mentions!), and sales representatives at the exhibit hall said to expect data at ATTD 2013. News on Medtronic’s Enlite was relatively absent this year, though this was likely to be expected since the sensor has been on the European market since April 2011. While an intermittent CGM study in pregnancy had disappointing results, the study without question speaks more to the need for continuous use and next-generation technology rather than an inherent lack of efficacy.

While the sessions brought little new news on the artificial pancreas, data on Medtronic’s predictive low glucose management (PLGM) algorithm and a per-protocol (PP) analysis from the CAT trial showcased the benefits of closed-loop algorithms on hypoglycemia. Using computer simulations, Dr. Barry Keenan (Medtronic Diabetes, Northridge, CA) showed how the new PLGM algorithm is expected to reduce the number of hypoglycemia events (<70 mg/dl) by 18% and the average duration of hypoglycemia by 50%, a significant improvement over the Veo’s corresponding reductions of 1% and 28%. We cannot wait to see the clinical data on the system, hopefully at ATTD 2013 in Paris. We also saw the PP analysis of the CAT Trial, which tested closed-loop control by two algorithms (Padova/Pavia/UVa’s MPC algorithm vs. Cambridge’s MPC algorithm) against open-loop control. The presentation confirmed the algorithms’ benefits in reducing hypoglycemia that was seen in the intent-to-treat analysis presented at ADA 2012 (see page 86 of our coverage at Last but certainly not least was Dr. Hovorka’s impassioned and highly comprehensive argument in favor of pursuing a “mechanical” solution to type 1 diabetes. In addition to the FreeStyle Navigator II launch news (see above), he provided a valuable update on the three-week overnight closed-loop home studies that the Cambridge team is currently conducting. The glucose traces from the first patients look quite good, and the study has recruited six out of a planned 16 participants. Titles highlighted in yellow are new additions to the report that were not initially included in our daily updates from Berlin. Additionally, titles highlighted in blue belong to our list of top 12 talks.


Diabetes Technology

Oral Presentations: Devices, Algorithms and Their Application


Barry Keenan, PhD (Medtronic Diabetes, Northridge, CA)

Dr. Barry Keenan presented three computer simulations of Medtronic’s predictive low glucose management (PLGM) algorithm. The algorithm, which we believe will be incorporated into the MiniMed 640G insulin pump, suspends insulin delivery based on whether hypoglycemia is predicted (30-minute time horizon) – Dr. Keenan emphasized that the advantage of PLGM is that it can reduce both the time spent in hypoglycemia (i.e., like the Veo, but more effectively) and the incidence of hypoglycemia (i.e., something the Veo cannot do since it only suspends once hypoglycemia is reached). Computer simulations compared measures of hypoglycemia in a control condition (no pump suspension), using the MiniMed 530G (suspending at 69 mg/dl), and using the PLGM algorithm. In one simulation using the FDA-approved UVA/Padova simulator (n=300), the PLGM reduced the number of hypoglycemia events by 18% and reduced the average duration of hypoglycemia by 50% (compared to reductions of 1% and 28% for the Veo algorithm). This is a significant difference and should really be appealing to patients and providers. The PLGM suspended insulin delivery at an average blood glucose of 83 mg/dl (vs. 69 mg/dl for the Veo) and the average max glucose after insulin resumption was 150 mg/dl (vs. 160 mg/dl for the Veo). As measured by YSI, the algorithm falsely suspended about 20% of the time, but Dr. Keenan explained that the impact of these false positives “is minimal” due to the system’s logic and automatic pump resumption. Medtronic has also simulated poor sensor performance and the algorithm performed similarly. Overall, the results look encouraging and we look forward to seeing the clinical data currently being collected in Germany by Dr. Thomas Danne (we expect to see this at ATTD 2013 in Paris) and by Dr. Tim Jones (Australia).

  • Medtronic’s predictive low glucose management (PLGM) algorithm suspends insulin delivery based on a 30-minute prediction of blood glucose. The algorithm will suspend insulin delivery if glucose is predicted and will automatically resume insulin infusion based on unspecified heuristics and logic. As a result of this predictive suspend, the PLGM can reduce both the incidence of hypoglycemia as well as the time spent in hypoglycemia. This builds upon the Veo, which only suspends insulin delivery once the hypoglycemic threshold is crossed (e.g., 69 mg/dl).
  • Medtronic tested the PLGM algorithm using the FDA-approved UVA/Padova computer simulator in 300 virtual patients. Hypoglycemia (targeting a glucose level of 60 mg/dl) was induced using a manual bolus. The PLGM algorithm was compared to the MiniMed 530G (i.e., the US version of the Veo that suspends at 69 mg/dl) and a control group (no pump suspension).
  • The PLGM reduced the number of hypoglycemic events by 18% and reduced the average duration of hypoglycemia by 50%. This builds on top of reductions of 1% and 28% for the MiniMed 530G (the reason for the 1% reduction in the total number of hypoglycemia events was due to simulated sensor error).
  • The sensitivity and specificity of the PLGM algorithm was modeled using data from the six-day Enlite accuracy study. The data set included 6,404 paired CGM-YSI points. The prediction horizon was set at 30 minutes and hypoglycemia was defined as <70 mg/dl. The algorithm had a sensitivity of 99.5%, meaning it detected nearly every hypoglycemia event. The tradeoff was a high false positive rate – 11% of suspensions were false alerts as measured by CGM and 20% were false alerts as measured by YSI. Dr. Keenan emphasized that the “impact of false positives is minimal” due to the algorithm’s intelligence in resuming insulin delivery.
  • To further test the PLGM algorithm, a range of good and poor performing sensors were modeled. MARDs ranged from 4% to 20% and true sensor noise and drift were acquired from the six-day Enlite accuracy study. Ten subjects were virtually evaluated and each repeated the study three times (i.e., three different sensors). Encouragingly, the data was in line with the numbers shown in the table above, suggesting that the PLGM algorithm is robust to poor sensor performance.

Questions and Answers

Q: What is the comparator in the clinical study?

A: There is a control arm in the study with Tim Jones. They are using exercise to create hypoglycemia.

Q: So it’s experimental and not in clinical events?

A: Yes, it’s in the CRC.



Dr. Emanuele Bosi presented accuracy results on the Glycolaser, a non invasive glucose monitoring device which uses two diode LED/lasers (the exact mechanism, he said, was an “industrial secret”). In our opinion, the picture of the Glycolaser revealed a form factor closer to hospital point-of-care meters than personal blood glucose meters. The device was tested in either fasting or postprandial conditions in 171 adults (31 healthy controls, 136 patients with type 1 or type 2 diabetes, and four patients with hypoglycemia syndrome). Only 49% of measurements were within ISO limits – 7.7% when glucose < 75 mg/dl and 52.5% when glucose >75 mg/dl. And if the proposed tighter ISO standards are implemented in early 2013 as Dr. Lutz Heinemann suggested on Day #1 (see page 7 at, Glycolaser will have to make even more substantial accuracy gains before pursuing regulatory approval. Certainly, as documented in the history of non-invasive methods that Dr. Bosi presented (“to date no one has been able to bring non invasive glucose measurement to reality”), achieving these improvements will be very challenging. Even Dr. Bosi (who was not involved in the development of Glycolaser) seemed relatively unconvinced about the meter’s viability during the Q&A that followed his presentation. He did note that if commercialized, the price could be low due to no “consumables” required.

Questions and Answers

Q: This seems to be interesting device. Some measurements were not acceptable, was this in specific patients? And does testing on different areas of the body play a role?

A: We didn’t identify what factor was making measurements be wrong. We need to work on the core technology. This kind of technology works only on the finger because it requires a lot of blood according to the inventor, who is a single and independent inventor, so there is not a great amount research behind it. I’m trying to convince the inventor to expand and get knowledge from more advanced researchers.

Q: You were smart enough to outline the critical history of non-invasive glucose monitoring. Unfortunately, all of these developments have failed. In a given patient, have you measured multiple blood samples in a row? Who has analyzed the data measured, and were patients aware of their blood glucose at the time?

A: The analysis was done separately. We are doing repeated measurements in the same individuals, and we have a few data, but more or less, the results are always the same. It’s relatively easy to reach the level of accuracy we achieved; the last gain in accuracy is always the most difficult to achieve.

Q: Can you elaborate on calibration and on the title? The title said in men – were results only in men?

A: Results were in humans in general – men and women. There was a slightly greater proportion of males. As to calibration, the inventor gave us the prototype, and according to him it was already calibrated, so there was not a calibrating phase. Apparently it doesn’t need it, but you know the results are still unsatisfactory. The matter of calibration has to be better investigated.

Q: What would the cost be, approximately?

A: That is impossible to answer. The cost is almost nothing, because once you have the instrument working, you have very little consumables. If it would work, the cost would be low. You’d still have to buy the machine, but I don’t know about the cost on that.

Q: So it’s not a Ferrari.

A: No, no.



Eric Renard, MD, PhD (Montpellier University Hospital, Montpellier, France)

Building on Dr. J. Hans DeVries’ intent-to-treat analysis of the CAT Trial at ADA 2012 (see page 86 of our coverage at, Dr. Eric Renard presented per-protocol results. As a reminder, the CAT Trial was a crossover-design comparison of 24 hours under open-loop control, closed-loop control with the Padova/Pavia/UVa (iAP Consortium)’s MPC algorithm, and closed-loop control with the Cambridge University team’s MPC algorithm (CAM). Dr. Renard explained that the per-protocol analysis excluded time during technical difficulties due to pump, sensor, or operational issues. Overall, the per-protocol analysis showed similar results to the intent-to- treat analysis: 1) time in range was comparable among all three conditions; 2) time in hypoglycemia was significantly lower during closed-loop control; 3) mean average glucose was higher with closed- loop control; 4) time in hyperglycemia was significantly higher with closed-loop control. However, Dr. Renard seemed to believe that the higher average glucose and time in hyperglycemia were preferable to the hypoglycemia observed during open-loop control. Additionally, Dr. Renard stressed that patients were achieving comparable time in range during closed-loop without the management burden of an open loop. Automated control works, he concluded. We believe that as the “closed loop” becomes closer, some patients may prefer “their” control to “closed loop” and this will likely create some conflicts in how the control is assessed.

  • As a reminder, the CAT Trial assessed the primary endpoint of time in target range in 48 patients (47 completed) with type 1 diabetes; secondary endpoints included 1) time spent in hypoglycemia; 2) time spent in hyperglycemia; and 3) mean plasma glucose. The study was conducted in six centers (three of which were naïve to closed-loop systems), with eight patients at each center. Patients used the OmniPod, Dexcom Seven Plus, and either the UCSB artificial pancreas system or manual validation from a nurse. Patients were studied in three non- consecutive 24-hour periods under three different conditions (open-loop, closed-loop with CAM, closed-loop with iAP; order of the conditions was randomized). The 24-hour period included three meals and an exercise excursion meant to challenge the algorithm. For additional protocol detail, please see our ADA 2012 report.
  • The per protocol analysis excluded 0.4% of time in the open-loop arm, 13% of time in the iAP arm, and 17% of time in the CAM arm. Time was excluded for technical difficulties including pump, sensor, or operational failures.
  • There was no significant difference between time in range between open-loop or closed-loop control, which was also observed in the intent-to-treat analysis. Dr. Renard explained that this finding should be considered in the context of the reduced burden on patients during closed-loop control. Automated control works, he said, with the benefit of reduced hypoglycemia (see below). Thus, he argued constant validation of the algorithms’ suggestions is not needed.
  • In line with results from the intent-to-treat analysis, patients achieved significantly less time in hypoglycemia with iAP and CAM compared to open-loop control (p=0.001); however, this was at the expense of significantly higher mean blood glucose (p=0.01)and time in hyperglycemia (p=0.01). iAP and CAM were comparable across all measures. Dr. Renard stressed that insulin levels were higher during open-loop control and that open-loop patients required a significantly higher number of hypoglycemia rescues (p=0.001). During Q&A he observed that though the average blood glucose was better during open-loop control, one has to ask whether mean blood glucose or time in hypoglycemia is more important. It seemed to us that Dr. Renard would take the latter view.
  • Dr. Renard argued that the trial demonstrated the feasibility and safety of using the closed-loop in a large number of patients. Notably, he said that the system’s performance would carry over to other centers irrespective of those centers’ previous closed-loop experience.

Questions and Answers

Q: Did you look at any changes in variability? We saw reduction in hypoglycemia but the same amount of time in target. Could the same glycemic results have been achieved in open-loop control by simply reducing the basal rate? And what was the algorithms’ target?

A: For patients in open-loop, insulin delivery was a patient decision, so no interventions in basal rate were made. It is common to see patients pushing too-high basal levels. For the algorithms, we targeted normal glucose control close to 100 mg/dl.

Q: Did you measure variability by standard deviation?

A: It has not been computed, but it seems that variability was reduced during closed-loop control. Another important remark – we wanted first to show the safety of the closed-loop, so we think that by choosing a better algorithm we could have less time in hypoglycemia and improve blood glucose control. The goal of the study was to show that this was feasible in a large number of patients. The next trial will do better at targeting the good control.

Q: Patients on open-loop achieved blood glucose targets better than the algorithms. Were you surprised?

A: The average glucose was better, but it all depends on whether you think what is important is mean blood glucose or time in hypoglycemia. To reach a good blood glucose with open-loop control means you are often in hypoglycemia. This is the difference with algorithms.

Q: There was no difference between SMBG and YSI calibration – was that retrospective?

A: It was randomized from the beginning. It was prospective. There were patients calibrating with YSI and others with SMBG.


Roman Hovorka, PhD (University of Cambridge, Cambridge, United Kingdom)

Dr. Hovorka presented results of a 36-hour crossover-design study that tested whether smaller meal- time boluses might reduce total insulin dose and risk of late postprandial hypoglycemia, without compromising glucose control. Adolescents with type 1 diabetes (n=8) spent alternating clinic visits under semi-closed-loop insulin delivery, with different amounts of prandial insulin (either 75% or 100% of the amount recommended by the pump’s bolus calculator). The smaller mealtime boluses led to similar glucose control as measured by mean, time within the target of 70-180 mg/dl, time below 70mg/dl, and time above 180 mg/dl – all with a total 36-hour insulin dose that was 10.6 units smaller overall, including lower basal rate. Dr. Hovorka hypothesized that acute postprandial insulin resistance may have been the reason that the larger size of the standard boluses did not translate to better control; he said that further studies were warranted to clarify why different insulin levels led to similar glycemia in adolescents (the finding was not replicated in adults in an AP@home study).

  • In this semi-closed-loop study, patients received mealtime insulin boluses of either 100% or 75% the dose recommended by the pump’s bolus calculator. Each patient in the crossover-design study underwent both test conditions, in randomized order. These eight adolescents with type 1 diabetes had mean age 15.9 years, mean A1c 8.9%, and median total daily insulin dose 0.9 U. The 36-hour study included two nights and one day, with moderate exercise on a stationary bike in the morning and afternoon. Manual boluses were given for meals (50-80 g carb) but not snacks (15-30 g carb).
  • The closed-loop system consisted of an Animas 2020 insulin pump, a model predictive control algorithm, and a single FreeStyle Navigator sensor calibrated per manufacturer’s instructions; Dr. Hovorka noted that the Navigator’s accuracy during the study was about as high as he’d ever seen (99.8% of reference-paired measurements in the Clarke A zone, 0.19% in the D zone). Control actions were taken by the system every 15 minutes.
  • Smaller mealtime boluses led to similar glycemic control with significantly lower 36-hr insulin dose (61.9 vs. 72.5 U, p=0.1) – reflecting both a ~25% decrease in mealtime bolus and a slight decrease in closed-loop basal rate – and significantly lower mean plasma insulin concentration (186 vs. 252 pmol/l). The smaller-bolus approach led to a slight increase in percentage of time spent above 10.0 mmol/l (180 mg/dl), mainly due to a larger glycemic excursion after dinner on the study’s second night. However, the change in hyperglycemia was not statistically significant, mean plasma glucose was identical at8.4 mmol/l (155 mg/dl), and the only instance of hypoglycemia below 2.5 mmol/l (<45 mg/dl) occurred under the standard-bolus condition. Total glucose appearance and glucose disposal, estimated using intravenous glucose with a stable radio-labeled tracer ([6,6-2H]glucose), were similar for each bolus strategy.

Smaller bolus

Standard bolus

p- value

Time within 3.9-10.o mmol/l (70-180 mg/dl)




Time below 3.9 mmol/l (<70 mg/dl)




Time above 10.o mmol/l (>180 mg/dl)




Total glucose appearance (umol/kg/min)




Glucose disposal (umol/kg/min)




Note: Percentages do not sum to 100% because they represent group medians rather than means

Questions and Answers

Comment: I think you are describing a clinically well-known phenomenon – that if your patients are slightly over-insulinized, then a slight reduction of dose does not lead to worse glucose control. As you showed, the incoming doses of 0.9 U/kg were high for adolescents. But I can offer no mechanistic understanding, of course.

Q: Do you think that this finding applies to other age groups, as well?

A: We tested this approach in the AP@home study and found that it does not apply to adults.

Q: Do you think that exercise might have been what reduced the need for basal in the afternoon or morning?

A: I don’t think there was interaction with the exercise – the observed drop in insulin concentration was already apparent overnight, after the first dinner. The exercise may have had an additional effect.

Q: Many reports show reduced accuracy of CGM in the hypoglycemic range. Was this something you found?

A: We were fortunate enough not to have much hypoglycemia, so I can’t really speak to that from this study. Our collective data suggest that coefficient of variation is greater at lower glucose – though this is only because you have a smaller denominator. We have found that Navigator accuracy is similar across the range.


Tao Yuan, MD (Peking Union Medical Hospital, Beijing, China)

Dr. Tao Yuan presented a small comparison of two ways for insulin-allergic diabetes patients to be desensitized to insulin. The patients were treated with diluted preparations of insulin using either an insulin pump (n=5; starting basal rate of 0.01 u/hr gradually increased to 1.4/hr over the course of three days) or subcutaneous injection (n=8; dose increased from 0.00001 to 20 u over the course of roughly seven days); treatment assignments were based on patient preference rather than randomization. Doses that provoked immune response were held steady until the immune response disappeared. Pump use was subjectively more convenient for the clinical staff, and it was associated with a slightly better success rate (100% vs. 87.5%). Rare though insulin allergy is, we are glad that attention is going toward optimizing its clinical treatment.

Questions and Answers

Q: Is insulin allergy more common in China than elsewhere?

A: It is rare, but my hospital is the best in China, so patients come from all over the world.


Oral Presentations: Profiling Glucose and Clinical Trials


Martin Prázný, PhD (Charles University in Prague, Czech Republic)

Dr. Martin Prázný presented data demonstrating that standard deviation, as a measure of glycemic variability, strongly correlates with the presence of microvascular complications (specifically, retinopathy, nephropathy, and peripheral neuropathy). In Dr. Prázný’s observational, case-control trial of 32 patients with type 1 diabetes, he compared 12 or 14 consecutive day CGM records of patients with microvascular complications (n=16) to those without complications (n=16). The CGMs used were the iPro2 and Dexcom Seven Plus and were used in masked mode. Standard deviation (SD), coefficient of variation (CV), and mean amplitude of glucose excursions (MAGE) of total CGM records were all significantly higher in patients with complications compared to those without complications (both when all complications were pooled and when each complication was analyzed separately). In a multivariate analysis, A1c, duration of diabetes, and age all failed to explain the presence of complications, but standard deviation did with an odds ratio of 4.53. Additionally, SD exhibited a significant positive correlation with vibration perception threshold (a measure of neuropathy), and exhibited a borderline significant correlation (p=0.056) with microalbuminuria. CV and MAGE exhibited a significant correlation with microalbuminuria. Neither mean absolute glucose (MAG) or continuous overall net glycemic action (CONGA) as measures of glycemic variability distinguished well between patients with and without complications. Dr. Prázný concluded that, since SD is very easy to calculate and is strongly associated with microvascular complications, it was the most efficient method of measuring glycemic variability. We believe that since this study correlates glucose variability with the presence of concomitant complications and not glucose variability as a predictor of complications, the clinical relevance is limited; the study assumes that, if glucose variability is to explain the presence of complications, that patients that had higher variability during the trial also had high variability before developing complications.

  • Patients in the trial without microvascular complications (n=16) had similar characteristics to those with microvascular complications (n=16). Respectively, average age was 43 and 39; diabetes duration was 21 and 18 years; mean blood glucose was 9.2 mmol/l (~166 mg/dl); A1c was 8.6%; BMI was ~26 kg/m2.

Questions and Answers

Q: I agree that SD is fast and easy, but I’m afraid I have a few problems with the trial, especially with its size. When you did your correlation you found no relation between complications and A1c, which we know isn’t true. I think in the analysis of the DCCT cohort you’ll find that there was a relationship between SD and A1c, meaning higher A1c was associated with greater SD. Rather disappointingly to me, SD is not independently related to glycemic complications directly. But I think what really matters at the end of the day is A1c.

A: The patients in our study did have similar A1cs, but that doesn’t say anything about where they will be in five to 10 years from now.

Q: Speaking of the relationship between glycemic variation and complications, it seems like a chicken/egg question. Which comes first?

A: I must say it’s a question of believing or not believing. I think there are no clues that glycemic variability induces complications directly. Variation can induce oxidative stress, etc., which may explain its association with complications.


Oral Presentations: Profiling Glucose and Clinical Trials


Carsta Koehler, PhD (Dresden University, Dresden, Germany)

Dr. Carsta Koehler presented a study which investigated whether an individual’s CGM profile was reproducible after a four-year period during which glycemic control remained stable. Though well presented, the findings were not as meaningful as we had hoped, especially because the less-accurate and higher-hassle Gold Medtronic CGM was used. The study included CGM data from 44 type 2 diabetes patients from Germany who, four years earlier, had participated in the 2007 ORIGIN CGM substudy (the results of which are published in Hanefeld et al., Diabetic Med 2010). Participants’ characteristics remained stable during the four-year follow-up period, with the exception of a slight increase in fasting plasma glucose (mean BMI of ~30 kg/m2, A1c of 5.7-6%; FPG changed from 104 mg/dl to 111 mg/dl) The study collected 72-hour data on interstitial glucose (i.G.) and compared several CGM parameters to those obtained during the 2007 ORIGIN CGM substudy (parameters included: average, fasting, minimal, and maximal i.G.; standard deviation of i.G. [SD]; mean average glucose excursions [MAGE]; AUC for 24-hour i.G. and 2-hour post-meal i.G.; time in mild or severe hypoglycemia). Participants’ individual CGM profiles mirrored those obtained four years prior, and correlation analysis found a correlation across several parameters (including SD, MAGE, AUC two hours post-meal, and time in severe hypoglycemia) between baseline and follow-up.

Questions and Answers

Q: What were you expecting to find? You reported stability in terms of CGM and glycemic variability. What was your hypothesis? What did you anticipate to find over four years?

A: Our hypothesis was that if you have stable glycemic control in the patient (no change in treatment), the pattern of your glycemic profile does not change – if you have a rapid increase in blood sugar after the meal and a long decrease in the postprandial phase, you have the same thing two or three years later.

Q: CGM is quite expensive. Do you see subgroups which would have a particularly high benefit from using CGMS and where it is clearly realistic to use CGM?

A: I don’t think we can perform CGM in all type 2 diabetes patients, but I think that we have a chance to use this diagnostic method in patients with a risk for hypoglycemia on in patients with cardiovascular risk. Other studies have investigated these patients – patients with diabetes and cardiovascular risks – and the presentation of this data is tomorrow.


Bruce Wolffenbuttel, PhD (University Medical Center Groningen, Groningen, The Netherlands)

Dr. Bruce Wolffenbuttel and colleagues conducted a post-hoc analysis of the DURABLE study for two purposes: 1) to assess the relationship between A1c and self-monitored blood glucose (SMBG) across different racial and ethnic groups; and 2) to evaluate the use of estimated average glucose (eAG) values derived from A1c as an alternate way to report blood glucose levels. The study included data from 1941 participants from four racial/ethnic groups (64% Caucasian; 16% Asian 12% Hispanic; and 6% African descent; data from “other” not included). The investigators found that at each mean blood glucose (MBG) level (108, 162, 216, and 288 mg/dl), Hispanics and those of African descent exhibited higher A1c levels relative to Caucasians (difference of ~0.3-0.4%); Asians also exhibited higher A1c levels vs. Caucasians, though to a lesser degree. Calculating eAG values and plotting them against SMBG showed that at MBG levels ≤11.4 mmol/l (205.2 mg/dl), eAG overestimated actual blood glucose levels while at MBG >11.4 mmol/l (205.2 mg/dl), eAG underestimated true blood glucose levels. After citing possible reasons why the A1c vs. MBG relationship differs by racial/ethnic group (details below), Dr. Wolffenbuttel concluded that doctors should consider a patient’s racial/ethnic background when assessing A1c and that eAG has limited clinical value.

  • The study first assessed the relationship between A1c and self-monitored blood glucose (SMBG). Plotting baseline A1c vs. the mean of all SMBG measurements for each participant (one data point per person) yielded a rather diffuse scatterplot with several corresponding A1c measurements for each mean SMBG value; nevertheless a positive linear relationship was observed for each racial/ethnic group. Mean regression analysis allowed investigators to estimate how A1c levels differ by mean blood glucose levels (MBG) in different racial/ethnic groups.
  • The investigators then calculated an eAG value for every participant based on his/her baseline A1c, using an equation obtained from the A1c-Derived Average Glucose (ADAG) study (Nathan et al., Diabetes Care 2008): eAG (mg/dl) = [28.7 * A1c (%)] – 46.7. Plotting each person’s eAG vs. his/her mean of all SMBG measurements yielded a regression equation that deviated substantially from the expected regression equation. The investigators then calculated a “mean blood glucose index (MBGI)” for each participant using the equation [MBGI = eAG – {mean of all SMBG values}], with the expectation that MBGI would equal zero (for every mean SMBG value) if eAG is a correct representation of mean SMBG. However, plotting MBGI vs. mean SMBG for each person yielded a linear regression equation with a strong negative slope instead of the expected horizontal line positioned at MBGI=0. The plot showed that at MBG levels ≤11.4 mmol/l (205.2 mg/dl), eAG overestimated actual blood glucose levels while at MBG >11.4 mmol/l (205.2 mg/dl), eAG underestimated true blood glucose levels.
  • Dr. Wolffenbuttel cited several possible reasons why the relationship between A1c and mean SMBG differs by racial/ethnic group. He explained that several factors beyond blood glucose levels independently influence A1c, including age, gender, BMI, MCH, MCHC, smoking, and alcohol consumption. Furthermore, both nutrition and genetic differences between racial/ethnic groups may play a role in the observed discrepancy. Dr. Wolffenbuttel also noted that studies have reported glucose-independent differences in A1c between blacks and whites.

Oral Presentations: Can We Improve Outcomes in Diabetic Pregnancy?


Anna Secher, MD, PhD (University of Copenhagen, Copenhagen, Denmark)

Dr. Anna Secher presented a very disappointing study testing intermittent use of CGM in pregnant women with type 1 and type 2 diabetes. Patients in the CGM group (n=79) wore the Medtronic Guardian real-time CGM (Sof-Sensor) for six days at a time during weeks eight, 12, 21, 27, and 33 of pregnancy and were compared to a non-CGM control group (n=75). Patients had a well-controlled A1c at baseline (~6.7%), though a high rate of severe hypoglycemia (~18%) that one of the doctors in Q&A pointed out probably reflects significant glycemic variability. Overall, there were no significant differences between the two groups in any glucose parameters at any point in pregnancy. Further, rates of macrosomia, pre-term delivery, and neonatal severe hypoglycemia were not significantly different between the two groups. We were quite surprised to see such neutral results though we think the study suffered from a number of major limitations: 1) the trial was done before wide trial use of the Enlite or Dexcom Seven Plus or G4 was widely available; 2) intermittent use of CGM for only six day periods following five study visits (rather than continuous use); 3) low compliance with the prescribed intermittent wear (64%); and 4) high rates of baseline severe hypoglycemia that did not improve by the end of the study. Given these limitations, we are highly skeptical of these results and feel very strongly that CGM can be a very valuable tool for pregnant women with diabetes. We look forward to further studies of continuous, real-time CGM using next generation technologies.

  • This study assessed the use intermittent real-time CGM as part of routine pregnancy care in 154 women with pre-gestational diabetes. Women with type 1 diabetes (80% of the study population) or type 2 diabetes were included and were randomized to an intermittent real- time CGM group or a non-CGM control group. In the CGM group, the Medtronic Guardian CGM (Sof-Sensor) was prescribed to be worn for six days at a time after study visits at weeks eight, 12, 21, 27, and 33. Both groups saw a diabetologist every second week and seven-point SMBG profiles were analyzed. Women willing to wear CGM between study visits were allowed to do so – the speaker did not mention how many women choose to do this, though we suspect a low number did. Patients’ real-time CGM readings were used to adjust diet and insulin using locally developed guidelines,. No specifics were provided on the guidelines, though in Q&A, Dr. Secher suggested that severe hypoglycemia is a real problem at their clinic and this directs a lot of their focus. Patients were advised to conduct SMBG before and after meals (90 minutes) and at bedtime. Pre- prandial targets were 72-108 mg/dl and post-prandial targets were 72-144 mg/dl.
  • There were no significant differences at baseline between the CGM (n=79) and non- CGM (n=75) groups. Women were in fairly good control and had a baseline A1c of 6.6% in the CGM group and 6.8% in the control group. However, we note that the rate of severe hypoglycemia was quite high in both groups (18% and 17% respectively) at baseline. A questioner in Q&A pointed out that this suggests a high rate of glycemic variability. Duration of diabetes was 10 years in the CGM groups and 12 years in the control group. Only 22% of type 1 patients in the study were on insulin pumps.
  • There was no significant difference in A1c at any point in pregnancy between the two study arms. Patients in the two groups tested SMBG equally frequently throughout the study. In both groups, 16% of women had at least one event of severe hypoglycemia throughout the study. We were surprised and disappointed not to see a significant improvement in severe hypoglycemia in the CGM group, though given the intermittent use and low compliance, this was probably not surprising – technology has to be easier to use in order to expect positive results..
  • Rates of macrosomia, pre-term delivery, and severe neonatal hypoglycemia were not significantly improved with CGM use – surprisingly, these trended (non- significantly) in the opposite direction. In the CGM group, 45% of pregnancies had macrosomia, compared to 34% in the control group. Pre-term delivery occurred in 29% of CGM group pregnancies vs. in 22% of control pregnancies. Severe neonatal hypoglycemia occurred in 13% of CGM group pregnancies and 14% of control group pregnancies. All results were not statistically significant.
  • A subgroup analysis of type 1s and those who followed the protocol did not change the overall findings. In women with type 1 diabetes, macrosomia occurred in 50% of CGM pregnancies vs. 36% in the control group. Pre-term delivery happened in 32% of CGM pregnancies and 27% of control pregnancies. In a per-protocol analysis, the results were similar – macrosomia in 49% of patients for CGM vs. 34% in the control group and pre-term delivery/severe neonatal hypoglycemia in 24% of CGM patients vs. 22% of those in the control group. We would have been interested to see data for those patients that wore the CGM 24/7.

Questions and Answers

Q: I’m wondering about the type of population selected. I was surprised by the low A1c – the controls and study group were very close to 6.5%. The frequency of severe hypoglycemia was 16-17%. This was probably a population with lots of glycemic variability. That probably explains the low A1c and the frequency of hypoglycemia. Based on the results of CGM, did you intervene?

A: We did intervene based on the results of CGM readings. We analyzed these results together with each woman and we used them to adjust insulin therapy. With the present form of CGM, you must use it together with plasma glucose measurement.

Q: What was the time duration you used CGM – three days or six days?

A: It was used for six days.

Q: And it was the iPro2?

A: No, the Sof-sensor.

Q: Was it approved for six-day use?

A: Yes.

Q: Can you explain the large difference in macrosomia in spite of similar control? The CGM group numbers were higher than in the control group.

A: We were disappointed with the results and very surprised. We were very surprised that the factual numbers were higher, though they were not statistically significant. We have had many considerations about why. One reason is that in our center, we have a lot of data on severe hypoglycemia and pregnancy. We really want to avoid severe hypoglycemia. Perhaps if we’re focusing too much on severe hypoglycemia, we might pay a price on maternal hyperglycemic complications.

Q: This was a courageous trial. But I guess it was designed to fail from the beginning. You used intermittent real time CGM. Dr. John Pickup showed that you need to wear the sensor at least 75% of the time. Why didn’t you use it continuously or as continuously as possible?

A: We did indeed use it as much as possible. Women were encouraged to use it extensively and free of charge. The trial design was with intermittent use. That was for several reasons. We have much experience with CGM and we knew that numerous alarms and other limitations in the system limit compliance. By asking the women to use it continuously, it would have been hard to find an unselected population and sufficient numbers of women.

Q: With real-time CGM, the results are better in patients on pump therapy. They can take boluses easier and reduce basal rates. What proportion of your patients were on pumps?

A: In the women with type 1 diabetes, 22% were on insulin pumps, and the majority of them had insulin pumps that could be connected to the CGM system. We hoped for even better results in that subgroup of women. We did not find any improvement in that group either.


Posters: Pumps and New Devices


Whitehurst, A.E. Colvin, A.D. DeHennis, J.C. Makous, M. Mortellaro, S. Rajaraman, J. Schaefer, D. Smith, S. Tankiewicz, O. Tymchyshyn, S. Walters, X. Wang (Senseonics, Inc., Germantown, MD)

This poster detailed the design and performance of Senseonics’ (formerly Sensors for Medicine and Science) fluorescence-based implantable glucose sensor in 12 people with diabetes that participated in three 28-day pilot studies (the sensor is designed for implantation for at least six months). Sensors for Medicine and Science/Senseonics was founded in 1999 and they have worked for almost 15 years on the development of their implanted sensor system. This is the best human clinical data that they have presented to date. Sensors were subcutaneously placed in the wrist (n=4) or upper arm (n=16) and were powered by an externally worn armband reader or wristwatch. The studies included six clinic visits of eight or more hours each and sensors were calibrated with one blood glucose value at the beginning of each clinic day. A second blood glucose value was used for calibration after 12 hours if the clinic visit was longer than 12 hours. CGM readings were prospectively determined and compared to YSI sampled every 15 minutes in clinic. Overall MARDs for the 20 sensors were 14.1% (upper arm), 16.8% (“improved” sensors in wrist), and 12.3% (“improved” sensors in upper arm), ranging from a low in one patient of 8.4% to a high in one patient of 29.6%. The “improved” sensors in the upper arm had 82% of points in the Clarke Error Grid A-Zone, 17% in the B-Zone, and 1% in the C- and D-Zones. This early feasibility data is encouraging, though we look forward to an upcoming six-month long trial in Germany and the UK that will help clarify some of the unanswered questions about the technology (see below).

  • Senseonics has developed a fluorescence-based implantable subcutaneous glucose sensor “designed to remain inserted for at least six months” (we note that this series of feasibility studies only tested the sensor for 28 days). The cylindrical sensor is 3 mm in diameter and 14 mm long – based on the poster’s picture, three of the sensors would fit on a 2-cent euro coin (quite small!). The sensors are inserted in the doctor’s office under lidocaine in an average of four minutes.
    • The implantable sensor uses an external reader (worn on an armband or wristwatch), which provides power to the implanted sensor, receives CGM data, alarms the user based on CGM values, and stores data for USB upload. The reader wirelessly provides power to the sensor using a wireless inductive link (13.56 Mhz). The reader also processes the implantable sensor data to determine sensor glucose values and rates of change. It can store up to six months of data, includes a beeper and a vibration motor to alert the patient (e.g., when glucose passes a threshold value), has a Bluetooth Low Energy link for wireless communication with a smartphone app, and features a USB port for charging and data exchange. The reader is pictured on the poster and appears as a small plastic box (about the size of a small matchbox) with a power button, a small screen (we presume sensor values and trends appear on it), and an elastic armband that secures the device to the user’s arm.
    • Senseonics has developed apps for the iPhone and iPod Touch that wirelessly operate as a user interface device for the reader, a display for CGM data, and a mechanism for providing user input. The poster noted that the apps are not used to process sensor data or store data long-term – this seems key from a regulatory perspective. The poster also showed color screenshots of the apps, which look quite sharp: a large sensor glucose reading in big font, a trend arrow, colored graphs, and warnings (e.g., Caution: Glucose above target!). The apps also allow the user to enter data on daily events (e.g., meals, insulin, and exercise), which is immediately transmitted to the reader.
  • A total of 20 implantable sensors were tested in 12 patients across three 28-day feasibility studies. Sensors were inserted in the wrist and upper arm and some patients woretwo sensors at a time. Eight sensors were tested in an initial cohort of patients, followed by testing of twelve “improved” sensors in second and third cohorts (the poster notes a slight change was made to improve long-term stability). The 12 patients were between 22 and 65 years old, had type 1 or type 2 diabetes, an A1c <10%, and a BMI < 35 kg/m2 (mean values and further population characteristics were not provided).
    • The studies included six clinic visits of eight or more hours each (days three, six, 12, 18, 24, and 29 [post-implant]). During the in-clinic visits, blood samples were taken every 15 minutes and processed using a YSI blood glucose analyzer. A single blood glucose value from the beginning of each clinic day was used to calibrate each sensor for each session from a Roche Accu-Chek Aviva SMBG, and the sensor glucose values were calculated prospectively for the session. A second glucose value was used for calibration after 12 hours if the clinic visit was longer than 12 hours.
    • Two reader prototypes were developed for the pilot clinical studies: a wristwatch reader (designed for use with a sensor inserted subcutaneously in the dorsal wrist) and an armband reader (designed for use with a sensor inserted subcutaneously in the upper arm). The readers were each programmed to read the implanted glucose sensor every two minutes.
  • Combined MARDs for the 20 sensors were 14.1% (upper arm), 16.8% (improved sensors in wrist), and 12.3% (improved sensors in upper arm). MADs ranged from 12 mg/dl to 19 mg/dl. The highest MARD experienced by any subject was 29.6% and the lowest MARD was 8.4%. The highest MAD was 27 mg/dl and the lowest was 7 mg/dl. The worst overall accuracy came in the four sensors implanted in the wrist, though the highest single instances of inaccuracy came in the initial upper arm cohort. The distribution of glucose values in the study was not reported on the poster. There were some points as low as 50 mg/dl and some as high as 350 mg/dl, but the Clarke Error Grid Analysis presented contained most points in the euglycemic range.
  • The four improved sensors placed in the upper arm had 82% of points in the Clarke Error Grid A-Zone (n=892), 17% in the B-Zone (n=184), 0.18% in the C-Zone (n=2), and 0.5% (n=5) in the D-Zone. Clarke Grid analyses were not presented for the other cohorts. We note that this cohort had the best performance of the three studied.
  • Other companies have shown good implanted sensor data for up to 30 days, only to have problems with encapsulation over subsequent days and weeks. Performance, reliability, and sensor chemistry are certainly challenged in the body’s environment as wear time increases, and it will be valuable to see how Senseonics performs over time. Additionally, we look forward to learning more about explantation after the six-month implant. As we understand it, explant takes about four minutes and there have not been concerns with exercising a large mass thus far.
  • Senseonics recently submitted a six-month clinical trial protocol to MHRA in the UK and BfArm in Germany. The trial protocol was pre-approved by TUV SuD (the selected notified body) and it contains both accuracy (in-clinic) and home use components. As we understand it, if the end points are met, regulators have agreed that this is an appropriate amount of clinical data to support approval.
  • From a regulatory perspective, we wonder how long of a study the FDA would require for a six-month implantable CGM; we imagine approximately nine to twelve months. Senseonics plans to discuss the European clinical trial protocol with the FDA through thenew “Pre-Submission” filing process that replaces the IDE. The hope is that the company can use the European data in the US submission, though this may be optimistic given the recent history with other devices (e.g., Medtronic’s Veo). Given the challenges in the US, we expect Senseonics will focus on Europe initially and then turn to the US at some point down the road.
  • It will be interesting to assess demand for implantable sensors from patients.cGenerally speaking, we imagine if reliability is high, the interest in implantable would expand the market to at least some degree. To what degree this cannibalizes the current market, which is only about 5% of type 1 patients and a very small minority (well less than 1%) of type 2 patients, will be interesting to see.
  • On the reimbursement front, the plan is to do insertions and explants in the doctor’s office – as a result, it will be key for the company to get those processes covered by insurance. Senseonics will begin those effectiveness trials after the pivotal trials are complete.

Posters: Blood Glucose Self-Monitoring


F. Kulozik, I. Platten, and C. Hasslacher (Diabetesinstitut Heidelberg, Heidelberg, Germany)

This poster, which was supported by Roche, compared the accuracy of 25 commercially available blood glucose meters in a wide range of glucose values (60-300 mg/dl) in 37 insulin-dependent patients with diabetes. All 25 meters were accurate within the requirements of the current ISO 15197 standards, though 14 of the 25 meters failed to meet the proposed ISO standards (95% within 15 mg/dl for <100 mg/dl and 15% for >100 mg/dl). The five most accurate meters according to this study were: 1) the Roche Accu-Chek Compact and Bayer Contour Next USB (tie); 3) Roche Accu-Chek Mobile; 4) Abbott FreeStyle Lite; and 5) Sanofi iBGStar. Meters that failed to meet the proposed standards were mostly from smaller companies, though the LifeScan OneTouch Verio, Bayer Breeze and Contour Plasma, and Sanofi BGStar came up short of the new standards. Obviously, results vary based on strip lots, though we were encouraged to see fairly high sample numbers (250-300) in this study. It was positive to see a study with the newest meters included together, as some recent studies we’ve covered did not compare all in one study (e.g., Freckmann et al., Journal of Diabetes Science and Technology 2012; Bayer’s North American Comparator Trial on pages 30-31 of our AADE 2012 report at The poster did not mention any study sponsor, surprisingly – we found out after our original viewing that Roche was the sponsor. Oddly, the Roche meter results were adjusted by 5% due to the strips’ hexokinase technology – we are in the midst of trying to get background on the rationale for this. In any case, we’ll be interested to see what happens with the new ISO standards – if they are indeed tightened, it should help push companies to further improve accuracy and could even bode well for CGM (e.g., calibration).

  • This study compared the accuracy of 25 commercially available glucose meters in five different blood glucose ranges in 37 insulin-dependent patients with diabetes (n=24 type 1, n=13 type 2). Patients had a mean age of 50 years, a mean A1c of 7.5%, a mean duration of diabetes of 17 years, and no concomitant use of substances that could affect blood glucose readings. Blood glucose levels ranged from 60-300 mg/dl in the study. For each SMBG system, 230-300 paired values were obtained. Results were compared to an internally and externally validated laboratory reference method (Hitado Super GL). Results from the referencestandard were converted from whole blood glucose values to plasma equivalent blood glucose values using the formula: plasma equivalent blood glucose (mg/dl) = 1.11 x whole blood glucose (mg/dl). Results were compared to current ISO 15197 standards (95% within 15 mg/dl and20% for <75 mg/dl and >75 mg/dl) and proposed standards (95% within 15 mg/dl for <100mg/dl and 15% for >100 mg/dl).
    • Since the Accu-Chek devices are calibrated by the hexokinase method, results from these meters were specifically adjusted by an increase of 5% (according to the poster, several studies have shown that blood glucose levels measured this way are 3.5-6.7% higher than glucose oxidase values). It struck us as somewhat odd that only the Roche meter results were adjusted. This also seemed somewhat unfair given that patients using the meters in a home-use environment would not see “adjusted” results on the meters.
  • The five most accurate meters were: 1) the Roche Accu-Chek Compact and Bayer Contour Next USB (tie); 3) Roche Accu-Chek Mobile; 4) Abbott FreeStyle Lite; and 5) Sanofi iBGStar. There was a wide range of overall accuracy between the glucose meters (80.4%-99.6%) and only three meters reached >98%. Accuracy was better at higher blood glucose values than at lower values. Data was collected and evaluated in the ranges of 50-99 mg/dl, 100- 149 mg/dl, 150-199 mg/dl, 200-249 mg/dl, and 250-300 mg/dl. For brevity in the table below, we have omitted the breakdowns and only included the overall data. Results below are presented as the percentage of values that meet the proposed ISO standards: within 15 mg/dl for <100 mg/dl and 15% for >100 mg/dl. The bold divider line denotes meters that failed to meet the 95% threshold.


Overall Accuracy (n)

Accu-Chek Compact

99.6%, n=282

Bayer Contour Next USB

99.6%, n=275

Accu-Chek Mobile

99.3%, n=294

Abbott FreeStyle Lite

97.7%, n=301

Sanofi iBGStar

97.6%, n=249

MyLife Pura

97.4%, n=271

Accu-Chek Aviva Nano

97.0%, n=303

LifeScan OneTouch Verio IQ

96.8%, n=278

LifeScan OneTouch Ultra Easy

96.8%, n=284

Bayer Contour USB

96.4%, n=248

LifeScan OneTouch Vita

96.2%, n=290

Beurer GL 40

94.9%, n=254

Bayer Contour Plasma

94.7%, n=284


92.6%, n=258

LifeScan OneTouch Verio

92.5%, n=265

A. Menarini Glucomen LX

92.0%, n=251

GlucoSmart Swing

91.7%, n=253

A. Menarini Glucomen LX Plus

91.5%, n=246

Wellion Calla

91.3%, n=252

Sanofi BGStar

90.1%, n=302

SmartLab Mini

88.5%, n=253

Smart Lab Sprint

87.1%, n=280

Bayer Breeze

86.7%, n=249

Beurer GL50

85.5%, n=228

Omnitest 3

80.4%, n=245

  • All 25 meters tested met the current ISO 15197 requirements, though 14 of the 25 meters failed to meet the proposed ISO requirements. The investigators also looked at accuracy per the proposed ISO thresholds for the five different glucose range buckets evaluated in the study. Four meters met the proposed ISO standards for every single glucose range: Accu-Chek Compact, Accu-Chek Mobile, Bayer Contour Next USB, and the OneTouch Ultra Easy. Another six meters met the proposed ISO standards for four out of the five glucose ranges: Accu-Chek Aviva Nano, Sanofi iBGStar, Bayer Contour USB, FreeStyle Lite, MyLife Pura, OneTouch Verio IQ, and OneTouch Vita.
  • Roche gave financial support for this study.


Stephanie Roze (Health EVAluation SAS, Lyon, France), Mark Cook (Medtronic UK, Watford, UK), and Peter Lynch (Medtronic International Trading Sarl, Tolochenaz, Switzerland)

This study evaluated the lifetime health and economic benefits of using CGM vs. SMBG alone. The results were quite powerful in favor of CGM: an additional two years of quality adjusted life expectancy and an additional three years alive and free of diabetes complications. Moreover, CGM was cost effective, with an incremental cost-effectiveness ratio (ICER) of just 17,932 per quality adjusted life year (QALY) gained, well under the NICE threshold of 20,000 per QALY. We note that these cost-effectiveness results were substantially better than those presented by Dr. Michael O’Grady at ADA 2012 (see pages 48-50 of our full report) – based on the JDRF CGM trial, he found ICERs ranging from a low of $57,170 (A1c <7% and wearing the sensor for seven days) and a high of $130,060 (A1c >7, age >25, and five days of wear). We believe much of this poster’s analysis was driven by the high baseline A1c used (10%) and the fact that wear time did not seem to be taken into account. Still, we think the results are powerful from a payer perspective and hope cost-effectiveness studies for CGM continue to show benefits, especially with next-gen technology.

  • This study used a computer simulation model to estimate the lifetime impact of CGM vs. SMBG alone on health complications and economic outcomes in type 1 diabetes in the UK. The researchers used the Core Diabetes Model, an internet-based, “highly validated” computer simulation model. Inputs came from John Pickup’s meta-analysis on CGM (BMJ 2011), a real-life observational study of CGM (Rose et al., ISPOR USA Conference 2012), hypoglycemia and quality of life data from the JDRF CGM trial (Diabetes Care 2010), and quality of life data from Currie et al., Curr Med Res Opin 2006.
  • The following model assumptions and inputs were used: baseline A1c of 10% (we believe this is what was used though it was not explicitly clear from the poster; we note this is quite a high A1c); a 0.9% reduction in A1c with CGM use (again, not totally clear from the poster’s methods); mean age: 27 years; mean diabetes duration: 13 years; mean SMBGs per day: 7.1 (SMBG alone) and 4.4 (CGM); annual rate of major hypoglycemic events: 27.7 events per 100 patient years (SMBG alone) and 15 events per 100 patient years (CGM); an adjusted quality of life to account for reduced fear of hypoglycemic events in the CGM arm (U=0.0152); diabetes related treatment and complication costs (UK specific) taken from various published sources; yearly discount rates of 1.5% (clinical outcomes) and 3.5% (economic outcomes). The analysis did not include indirect costs, which would have led to even better health economic outcomes.
  • CGM had an incremental cost-effectiveness ratio (ICER) of 17,932 per quality adjusted life year (QALY) gained, well under the NICE threshold of 20,000 per QALY. Quality adjusted life expectancy was 1.9 years higher with CGM (11.9 years with SMBG alone vs. 13.8 years with CGM). Total costs of SMBG alone were 59,193 vs. 93,523 for CGM. In a one-way sensitivity analysis considering no effect on the rate of major hypoglycemic events for CGM, the ICER was 23,067 per QALY. The one-way sensitivity analysis based on no reduction of fear of hypoglycemic events due to CGM led to an ICER of 21,336 per QALY. By using an equal discount rate for both health and economic parameters of 3.5%, the ICER was 25,975 per QALY. By varying the treatment costs by 10%, the ICER varied by 17% (14,883-20,980). It’s good to see these sensitivity analyses did not drastically affect the main finding.
  • Use of CGM was linked to nearly three more years alive and free of diabetes complications vs. SMBG alone.

Time Alive and Free of Complications (Years)


SMBG Alone


Change with CGM

End-Stage Renal Disease




Myocardial Infarction




Severe Vision Loss












Average for all Complications




Posters: Continuous Glucose Monitoring


Price and K. Nakamura (Dexcom, San Diego, CA); T. Bailey (AMCR Institute, Escondido, CA), M. Christiansen (Diablo Clinical Research, Walnut Creek, CA), E. Watkins (Profil, Chula Vista, CA), D. Liljenquist (Rocky Mountain Diabetes and Osteoporosis Center, Idaho Falls, ID)

This poster gave a concise, useful summary of the accuracy and reliability improvements of the Dexcom G4 Platinum Sensor over the Dexcom Seven Plus. We’ve seen snippets of the new data at ADA 2012, EASD 2012, and ISPAD 2012, though this poster put it all together and did a nice job comparing a broad array of metrics from the two sensors’ pivotal studies. The short story is that the G4 Platinum has improvements in every measure of accuracy and reliability over the Seven Plus (with the exception of Day 1 accuracy, which is marginally worse with the G4 Platinum). Especially striking are the G4 Platinum’s improvements in the hypoglycemia range and the strong durability of accuracy at the seven- day mark. Moreover, this poster made it clear that the G4 Platinum pivotal study was more robust than the Seven Plus pivotal study in a number of ways – it included more patients (72 vs. 53), more and longer in-clinic days for each subject (three 12 hours vs. one eight hour), nearly double the sensors (108 vs. 67), and ten times more points in the YSI range <55 mg/dl (361 vs. 33). For more details, please see tables below. As a reminder, the G4 Platinum was approved by FDA on the Friday of EASD (October 5, 2012) and will begin shipping in late October (see our coverage of the FDA approval call at

  • The G4 Platinum has reduced the number of outlier sensors vs. the Seven Plus. The poster statistically summarized this as follows: G4 Platinum – mean ARD: 13.6%, median ARD 12.5%, standard deviation: 6.7% vs. Seven Plus – mean ARD: 16.7%, median ARD 14.4%, standard deviation: 8.6%. A histogram also summarizes individual sensor MARD data, illustrating that the G4 Platinum has far fewer sensors with a MARD >20%. This is encouraging from a patient perspective and we believe it should encourage more patients to stay on CGM and wear it a greater proportion of the time. The landmark JDRF CGM study (NEJM 2008) found that patients in all age groups (adults, adolescents and children) who wore CGM at least six days a week had a clinically significant reduction in A1c. We look forward to patients using the G4 Platinum more and, as in the JDRF CGM study, getting greater clinical benefit.

Dexcom Seven Plus vs. G4 Platinum – Accuracy vs. YSI


G4 Platinum

Seven Plus

Sensors in Pivotal Study



Mean Absolute Difference (MAD)

20.8 mg/dl

24.7 mg/dl

Mean Absolute Relative Difference (MARD)



MAD within Biochemical Hypoglycemia (YSI <70 mg/dl)

10.6 mg/dl

16.0 mg/dl

MARD within Biochemical Hypoglycemia (YSI <70 mg/dl)



Number of Samples in Severe Hypoglycemia (YSI <55 mg/dl)



Percentage of CGM values that were

<70 mg/dl when YSI <55 mg/dl



Clarke Error Grid (CEG) A-Zone Overall



CEG A-Zone in Hypoglycemia (YSI

<70 mg/dl)



Mean Paired Sensor Coefficient of Variation



Sensor Display Rate



Sensors Lasting Seven Days




Dexcom G4 Platinum – Durability of Accuracy


Matched CGM- YSI Pairs

MARD 40-400


Percent within 20% of YSI*

Percent >40% of YSI*

Day 1





Day 4





Day 7






Dexcom Seven Plus – Durability of Accuracy

Day 1





Day 4





Day 7





*For YSI <80 mg/dl, the absolute difference is presented as the difference between CGM value and YSI, rather than the percent.

Corporate Symposium: Dexcom G4 Platinum Continuous Glucose Monitoring: Revolutionary Technology Bringing Patient Care to the Next Level (Sponsored by Dexcom)


Thomas A. Peyser, PhD (VP Science and Technology, Dexcom, San Diego, CA)

Dr. Peyser gave an excellent data-filled presentation on Dexcom’s new G4 Platinum sensor (the first time we’ve seen the branding “Platinum” to refer to the Gen 4 sensor) with one major takeaway: this is a whole new level of CGM accuracy for Dexcom – and patients will surely benefit. He gave a deeper dive into the pivotal trial data presented at ADA by Dr. David Price (see page 28 of our report at and presented some very unique cuts of the accuracy data that we hadn’t previously seen. We also appreciated his commentary on how current accuracy metrics fail to account for patients’ varied experiences with individual sensors – in short, there are limitations to presenting overall averages. Dr. Peyser emphasized that the G4 Platinum’s improved accuracy and ease of use will result in a more positive experience for patients wearing the sensor, which will hopefully result in more sustained use and a greater clinical benefit. Considering the low rates of CGM penetration, we certainly hope the Dexcom G4 Platinum and Medtronic Enlite sensors can usher in a new era of accelerated CGM adoption. Of course, it’s tough to disentangle how much of CGM adoption is related to accuracy vs. reimbursement/cost vs. hassle factor, though we have no doubt that the improved technology will help convince more patients that the technology is worth wearing 24/7.

  • Regarding the regulatory status of the G4 sensor in the US, Dr. Peyser stated, “I’m confident that it will be approved soon. It could be next week, it could be next month, or it could be three months from now.” As a reminder, the system received CE Mark on June 15, 2012 and was submitted to the FDA on March 31, 2012. As of Dexcom 2Q12 (see our report at, approval was expected before the end of 2012.
  • Dr. Peyser reviewed the G4 accuracy data from the recent pivotal study, which he characterized as a “huge advancement” versus the Seven Plus and for the whole category. With the Seven Plus, 74% of values fell in the Clarke Grid A-Zone, compared to 80% with the G4. The overall MARD has also improved from 15.7% to 13.2%. Historically speaking, this compares to 22% for the GlucoWatch, 20% for the Guardian Real-Time, and 26% for the Dexcom STS – very good for recent comparisons, though the Abbott Navigator did achieve 82% A and 12.6% MARD in a similar 2007 study. As a reminder, the pivotal study occurred at four sites in 72 patients over seven days. It featured three 12-hour in-clinic sessions with YSI on days one, four, and seven (>9,000 paired CGM-YSI points). For more background on the pivotal study, see page 28 of our ADA 2012 full report
  • The G4 Platinum features significant improvements over the Seven Plus in the hypoglycemia range. For values <70 mg/dl, the G4 Platinum has a mean absolute deviation (MAD) of 11 mg/dl vs. 16 mg/dl for the Seven Plus. MARD in hypoglycemia has improved to 19.1% with the G4 Platinum compared to 27.3% for the Seven Plus. Clarke A Zone readings (<70 mg/dl) were 80% for the G4 Platinum vs. 62% for the Seven Plus. The percentage of CGM values<70 mg/dl when YSI read under 55 mg/dl is also improved with the G4 Platinum: 88% vs. 73%for the Seven Plus.
  • Dr. Peyser discussed how current accuracy metrics “often ignore patients’ experience while using CGM.” He noted that patients’ confidence or lack of confidence in CGM is often based on their own experience with individual sensors; e.g., “I had a good sensor” or “I had a bad sensor.” As a result, accuracy metrics should capture this variation in individual sensor performance, something that averages fail to capture. It was great to hear this patient perspective. (to illustrate the point, Dr. Peyser jokingly referred to the old statistics joke that has a man’s feet in the refrigerator and his head in the oven – he feels “pretty good on average”).
    • The G4 sensor has reduced the variability in sensor-to-sensor performance. Overall, the Seven Plus had a mean ARD of 15.9% and a standard deviation of 8.6%, which has been narrowed to 13.2% and a standard deviation of 6.7% with the G4 Platinum. In Dr. Peyser’s words, “This has important consequences in terms of how patients experience the use of CGM” – essentially, fewer “bad” sensors. He also showed an interquartile analysis that breaks down individual sensor MARDs. This was a unique way to display the data that we had not seen before – it basically gives a broader picture of the full spectrum of sensor accuracy, which we appreciated instead of just a single mean. Dr. Peyser noted that Dexcom was “surprised and delighted by the results.” Indeed, the table below shows that 50% of G4 sensors have a single digit MARD – quite impressive indeed.

All Sensors

Top 75% of All Sensors

Top 50% of All Sensors

Top 25% of All Sensors

Seven Plus Average MARD





G4 Platinum Average MARD





  • Dr. Peyser discussed the performance of the G4 Platinum as measured by the Clarke Error Grid, including an interquartile analysis. While some have criticized the Clarke Grid for being too loose overall, too tight in hypoglycemia, or just unsuitable for continuous data, Dr. Peyser believes it’s a useful metric to put accuracy in the context of clinical decision-making. However, he cautioned that CGM companies have abused this measure in the past by providing A+B Zone data together, which masks A Zone performance. The table below presents A Zone data for quartiles of individual sensors – Dr. Peyser emphasized that it’s “pretty incredible given where this technology was a decade ago” and this higher accuracy means users will have a “positive experience from most sensor wears.


All Sensors

Top 75% of All Sensors

Top 50% of All Sensors

Top 25% of All Sensors

G4 Platinum Percentage in A Zone





  • CGMs are very susceptible to calibration error from SMBG, especially on day one. Dr. Peyser showed an example of a G4 sensor with a MARD of 21.4% and 69% in the Clarke Grid A Zone. The low overall accuracy was due to a highly inaccurate first day, which improved to a MARD of 7.2% on day seven. It turns out that the blood glucose meter reading used to initially calibrate the Dexcom was off YSI by 54 mg/dl (!), which set the sensor session up for inaccuracy from the get-go. After simulating the correct YSI value retrospectively, Dr. Peyser noted that the accuracy would have been spot on for the entire first day. This underscores the importance of teaching patients that getting a good fingerstick is really important for calibrating CGM. Of course, Dexcom has a goal of eliminating or significantly reducing fingerstick calibrations in the future, so we hope this becomes less and less of a problem over time.
  • The precision of the G4 Platinum is on par with SMBG (Dexcom “is nipping at the heels of the BGM companies”). In the G4 pivotal trial, 36 patients wore two sensors (left and right abdomen). The coefficient of variation (CV) between the two sensors was 7%, on par with SMBG and a favorable comparison to previous CGMs (CV of 12-20%). Dr. Peyser explained that this is a “whole different level of performance and precision that has never been seen before in this field.”
  • The G4 Platinum’s improved performance reflects changes in the sensor, the transmitter, and the receiver. The sensor features a 60% reduction in volume and an improved biocompatible membrane (i.e., reduced wound response, more consistent performance across a wider range of patients). The transmitter is “a little larger” though “still very small” and has upgraded to a 2.4 GHz radio frequency. The transmission range has increased to 20 feet from just five feet on the Seven Plus. Dr. Peyser also highlighted some of the reliability data presented at ADA: 97% data capture over seven days (i.e., 279/288 possible readings per day) and 94% of sensors lasting seven days. Lastly, the new receiver has been redesigned to look like the original iPod nano. The screen is now color and the receiver incorporates new algorithms that adaptively adjust over time to account for changes in the environment of the sensor.


Martin Prázný, MD, PhD (Charles University, Prague, Czech Republic)

Dr. Martin Prázný gave the audience a first look at Dexcom’s new Studio software, which will accompany the new G4 sensor. The most significant change from the current DM3 software is the addition of a pattern recognition tool that automatically detects low and high patterns throughout the day – this was excellent to see and should really improve the value of CGM data. The approach reminded us of LifeScan’s OneTouch Verio IQ blood glucose meter, and in some ways it also resembled Medtronic’s CareLink Pro 3.0. Besides the new pattern tool, the software looks largely similar to the older version with sections for hourly stats, daily trends, distribution graphs, modal day reports, daily statistics, and a success report (based on the screenshots, we assume it is PC-only, which is disappointing). An example use of the pattern recognition feature is summarized in the table below – target ranges and nighttime/daytime periods can be customized. Like CareLink Pro 3.0, the new Studio software also graphically shades areas of hyperglycemia and hypoglycemia on the modal day report. Further, the pattern tool provides an associated column entitled “Some Possible Considerations” – these are fairly basic things like “If before meals, adjustment to basal insulin,” “If after meals, adjustment to meal time insulin,” “Review carb counting, effect of exercise, alcohol, and or food choices” – however, they are useful reminders for patients and clinicians. We look forward to hearing more about Dexcom’s plans for software, especially the ongoing integration with SweetSpot.

Pattern Insights Summary
Nighttime Lows (0 Found) No significant patterns detected.
Daytime Lows (2 Found) Most significant pattern of lows found between 6:45 pm and 7:40 pm
Nighttime Highs (0 Found) No significant patterns detected
Daytime Highs (1 Found) Most significant pattern of highs found between 8:05 AM and 11:35 AM.


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

Dr. Buckingham shared brand new data on a nocturnal remote monitoring study using the Dexcom G4 CGM in forty-one patients at two diabetes camps over one week. Using a USB cable, the G4 receiver was connected to an Android cell phone running the UVA Diabetes Assistant software. The cell phone sent the CGM data to a central server and to doctors’ computers and iPads at the camp. Patients were randomized to either control nights (no remote monitoring and fingersticks to determine hypoglycemia treatment) or remote monitoring nights. Notably, remote monitoring led to a 79% reduction in events <70 mg/dl (seven events vs. 33 events), a 100% reduction in events < 50 mg/dl (zero events vs. nine events), and an improvement in attendants’ response time to nocturnal alarms (no p-values included). Positive data aside, Dr. Buckingham was also highly positive on the G4 sensor itself: “These kids really liked this sensor. They really found it to be accurate…Dexcom has a real winner here.” Part of the randomization involved treatment with either mini-doses of glucagon or carbs. Interestingly, glucagon was associated with more recurrent hypoglycemia within three hours (i.e., it failed more often relative to carbs). It will be interesting to watch this over time and see if future studies are consistent with this finding.

  • There are two strategies to prevent severe hypoglycemia: 1) suspend insulin delivery using low glucose suspend (LGS) or 2) actively intervene with fast-acting carbs using CGM and remote monitoring. We thought this was an interesting way to position CGM on a similar playing field to LGS.
  • This pilot study tested the impact of remote monitoring with the Dexcom G4 sensor on nocturnal hypoglycemia at two diabetes camps. Forty-one patients (n=29 at Chinnock and n=12 at De Los Ninos; mean A1c ~8.4%) were randomized to a night with remote monitoring or a control night. On remote monitoring nights, patients used the G4 sensor and receiver, the latter of which was connected via a USB cable to a Samsung Android cell phone running the University of Virginia’s Diabetes Assistant application (for more on the Assistant, see page 14 of our ATTD 2012 report at CGM data while patients were sleeping was sent from the cell phone to a server at the University of Virginia through a cellular network or local WiFi (one of the camps did not have cell service). The UVA server then transmitted the data to the doctor’s computers in camp cabins and to portable iPads. The camp’s doctors could view all patients’ CGM readouts at once and intervened once blood glucose dropped below 70 mg/dl. The master display uses the red-yellow-green traffic lights for hypoglycemia and hyperglycemia and displays the CGM reading and trend arrow for each patient. On control nights, patients also wore a CGM, though it was not remotely monitored and their treatment was based on standard nighttime SMBG testing.
  • Remote monitoring decreased the number of events <70 mg/dl by 79% and events<50 mg/dl by 100%. The number of nights with remote monitoring (161) was comparable to the number of control nights (179). No p-values were reported on the slide. For more details please see table below.


Remote Monitoring

Control Nights

<70 mg/dl


>1 hour



>2 hours



<50 mg/dl


>30 minutes



>1 hour



  • Encouragingly, remote monitoring also reduced the response time and increased the response rate to nocturnal alarms. Seventy-seven nocturnal alarm events occurred in the remote monitoring group, and 100% of alarms were responded to. By contrast, the control group had 119 events and only a 54% response rate. Dr. Buckingham also displayed a box plot graph (unfortunately not numerically labeled) to demonstrate that response time to nocturnal alarms was lower and less variable in the remote monitoring group, and outliers were less extreme – a max response time of 80 minutes in the remote monitoring group compared to 118 minutes in the control group.
  • “These sensors worked really well. We had a lot of faith in the data and the results we were getting. It was impressive….it was very rare we got called and it wasn’t low.” Notably, the Dexcom G4 had a true positive alarm rate of 79% in this study, substantially better than historical data from other CGMs: 60% for the Navigator, 54-67% for the Guardian RT, and 54% for the Dexcom Seven Plus. It’s great to see the Dexcom G4 has indeed made serious improvements in the hypoglycemic region and this data seems to hold in a more real world setting. We believe detection/avoidance of hypoglycemia represents an important reason why many patients choose to go on CGM in the first place, and it’s certainly frustrating for patients when false low alarms occur. We would guess that improving the true positive alarm rate could have a noticeable benefit both on patients’ clinical experience with the device (i.e., reducing hypoglycemia) and potentially on CGM attrition.
  • Despite all the rigors of the camp (e.g., swimming, sweating, sports), 81% of the sensors remained on until the completion of camp (five to seven days). Dr. Buckingham noted that 5% of patients had mild erythema from the adhesive and there was no significant edema or inflammation at the sensor insertion sites.
  • Part of the randomization process included treatment with either mini-doses of glucagon or carbs – interestingly, glucagon failed much more often within three hours of the initial treatment. Mean glucose rose to ~180 mg/dl with mini glucagon vs. ~160 mg/dl with carbs (no significant difference). However, Dr. Buckingham noted that glucagon treatment was associated with many more recurrent lows within three hours.


Jay Skyler, MD (University of Miami Miller School of Medicine, Miami, FL)

Dr. Jay Skyler provided an exciting glimpse into Dexcom’s pipeline, which includes integration with SweetSpot Diabetes Care, insulin pump partnerships, the Gen 5 smartphone compatible sensor, involvement in several artificial pancreas projects, predictive algorithms, and remote monitoring. Notably, Dr. Skyler shared new accuracy data on the special edition of the G4 sensor designed for the artificial pancreas – an overall MARD of 11.3%, 96% of sensors with a MARD <20%, and in two example sensors, a sub-5% MARD that even beats out fingerstick monitoring accuracy. We also appreciated Dr. Skyler’s discussion of remote monitoring and predictive alerts, which he believes is a valuable alternative to LGS for two main reasons: 1) less risk of a roller coaster pattern because insulin is not being suspended; and 2) MDIs can use predictive CGM, unlike LGS systems that only pumpers can use. We thought these were both valuable points and it will be interesting to see how Medtronic and Dexcom position their respective next-gen products against each other in the years to come.

  • Dr. Skyler reviewed Dexcom’s partnership with SweetSpot Diabetes Care to build an Internet-based data platform. He first reminded attendees of the many advantages of the SweetSpot platform: it is cloud-based, compatible with several glucose meters and all insulin pumps (except Medtronic), integration with electronic medical records, and advanced glucose data analytics. Dr. Skyler believes the SweetSpot system will help transform diabetes care by enabling new models of care and moving diabetes care from the clinic into patients’ homes.
  • Turning to Dexcom’s insulin pump partnerships with Animas, Insulet, Roche, and Tandem, Dr. Skyler highlighted the benefits of integrated pump and CGM data – a recent study published in Diabetes Technology and Therapeutics (Frontino et al., 2012) demonstrated a 1% improvement in A1c in pediatric patients (<7 years old) with an A1c >7.5% using an Animas pump and Dexcom Seven Plus CGM. These young patients wore the sensors 83% of the time, demonstrating their wide utility even in such a young patient population. Dr. Skyler noted that the Gen 4 sensor will be integrated into the OmniPod and Animas pumps and will hopefully “be available sometime soon” (as of Dexcom 2Q12, a PMA supplement for the Animas Vibe is expected to be filed before the end of 2012, while the timeline for the Insulet product has not been announced – see our report at Meanwhile, the Gen 5 sensor will be part of future Roche and Tandem pumps. Dr. Skyler also said that it will only be a matter of time before the “Paradigm people” (i.e., Medtronic) want to have a “real glucose sensor with their pump.”
  • The Gen 5 sensor’s direct smartphone connection “looks like a really good way to go.” Dr. Skyler showed a picture of both a smart phone and a watch displaying a Dexcom CGM reading in a sleek interface. He was most positive on the potential of the Gen 5 CGM data to go directly to the cloud from the smart phone, a particularly nice advance for parents. He also expressed optimism in referring to how far the technology has come since the first gen came out in 2006.
  • Dr. Skyler shared new data on the special version of the G4 sensor designed for use in the artificial pancreas (“a remarkable device”). As we heard at ATTD 2012, the G4 AP uses the same sensor, transmitter, and receiver as the G4, but it includes new algorithms for improved accuracy and reliability. It will be made available to closed-loop researchers under an IDE or equivalent. Dr. Skyler showed a few examples of the accuracy improvement on days one to seven – in both instances, the day seven MARD was better than fingerstick monitoring. For more details, please see table below.
  G4 AP G4
Overall MARD 11.3% 13.2%

Percentage of Sensors with a MARD <20%



Example 1: Accuracy on Day One and Day Seven

11.1% to 4.7%

32.8% to 7.1%

Example 2: Accuracy on Day One and Day Seven

7.2% to 4.1%

12.7% to 5.3%

  • Dexcom is working with the University of Padova to develop a predictive algorithm and increase the warning time prior to hypoglycemia (<55 mg/dl). Dr. Skyler compared using a CGM threshold of 70 mg/dl alone to the threshold plus a prediction algorithm. The threshold alone resulted in a median alert time for hypoglycemia (<55 mg/dl) of 15 minutes, which increased to 20 minutes with the predictive algorithm. Additionally, the number of alerts that gave less than 15 minutes of notice for hypoglycemia dropped from 38% to 16% with the new algorithm. Encouragingly, the prediction algorithm only added an average of one nuisance alarm per week and 46% of sensors had no additional nuisance alerts. We're very glad to see Dexcom investing in this area given its potential to truly enhance patient quality of life and safety – as a reminder, Medtronic has had predictive alerts for some time in its Revel and Veo insulin pumps and is developing a predictive LGS system, the MiniMed 640G (see our Medtronic analyst day report at
    • The future of Dexcom technology will combine remote monitoring with advanced CGM – an “alternative to the need for LGS.” According to Dr. Skyler, using the most accurate and advanced CGM technology alleviates the need to suspend insulin delivery since the patient will be alerted in time to treat or prevent hypoglycemia. Indeed, he noted that shutting off the pump may lead to a rebound high, followed by overcompensation that drives glucose too low, followed by a rebound high, etc. The “best way to do it,” he believes, is to anticipate the hypoglycemia and intervene by taking appropriate carbs. Moreover, Dr. Skyler made the valuable point that LGS only works if you have a pump, while pre-warning with CGM would work in patients on MDI. Since​ ~90% of insulin requiring patients in Europe don’t use pump (~70% in the US), this is a huge fraction of the market.


Dorothee Deiss, MD (Endokrinologikum am Gendarmenmarkt, Berlin, Germany); Thomas Peyser, PhD (Dexcom, San Diego, CA); Martin Prázný, PhD (Charles University, Prague, Czech Republic); Bruce Buckingham, MD (Stanford University, Stanford, CA); and Jay Skyler, MD (University of Miami Miller School of Medicine, Miami, FL)

Dr. Boris Kovatchev (University of Virginia, Charlottesville, VA): Dr. Skyler, the outpatient study that you mentioned is now complete.

Dr. Skyler: Do you have the data? Are you presenting it here? Dr. Kovatchev: No.

Dr. Skyler: ATTD 2013?

Dr. Kovatchev: Yes.

Dr. Gary Steil (Children’s Hospital Boston, Boston, MA): Dr. Skyler, you showed some Howard Zisser data on not using meal boluses. I agree that meal boluses can be problematic. You said you think it’s better to give carbohydrate than shut the pump off. I’m a bit concerned about weight gain from excessive use of corrective carbohydrate.

Dr. Skyler: Yes, the best thing is to avoid hypoglycemia altogether, and I think that can be done with some of the CGM algorithms in development.

Dr. Steil: I have more confidence than you in turning the pump off in a timely manner, as long as it’s turned back on in a timely manner. This is why I have more faith in a closed- loop system than simple LGS. It would seem that if you prevent hypoglycemia with pump suspension, and turn it back on, you could prevent hyperglycemic rebounds.

Dr. Skyler: I think you’ve got to do all those things. Bruce, you have looked at suspension.

Dr. Buckingham: I think even with the Veo, the blood glucose goes up to about 150 mg/dl after several hours. With our system the pump turns back on immediately past the hypoglycemic nadir. Even with that, we get a bit of rebound to the 120-140 mg/dl range, though I wouldn’t call this hyperglycemia. It depends on the duration of suspension.

Dr. Skyler: How long did you typically suspend for?

Dr. Buckingham: Usually about 30-40 minutes. There are limits: the system can’t turn off for more than three hours at night, for example. If someone takes a bedtime snack and over-boluses, there is so much insulin on board that suspension can’t prevent hypoglycemia. Once you are past bedtime insulin, then you can be very effective because you don’t have much insulin on board.

Dr. Skyler: My bias is I worry about shutting off the pump. It could take three hours before you re-initiate basal to get the blood glucose back up. That can be ages.

Dr. Buckingham: We find that upon restarting, you get back to pretty effective insulin after 60-90 minutes.

Dr. Skyler: But that’s for shutting off for a 30-40 minutes at a time. But a two hour suspend is a pretty long time.

Q: I’m from BD. I wanted to ask you about calibration on day one being the hardest. You said when it comes to the first calibration, you have to get it right. Does it become easier over the sensor wear because of some rating scheme or other processes?

Dr. Peyser: Calibration on day one is done anew. Each sensor is calibrated based on blood glucose meter readings that are input into the receiver. On day one, you begin with two initial calibrations, followed 12 hours later with another calibration. On days three or four, you have a cumulative history of readings. The sensor system is more susceptible to error from SMBG on day one than on subsequent days. We think this is a large source of error in clinical studies.

Q: I think that with current economic conditions, it’s very important to have studies that show the same benefit for patients with MDI or pump use. Maybe we can spend the money on CGM instead of pumps, because we don’t have them for both.

Dr. Skyler: I think you raise an important point: that CGM is inexpensive compared to pumps, and that having the glucose information is probably more important than having continuous insulin delivery. I agree with you that we need to do more studies to make sure that everyone gets the message that you can improve glycemic control by adding CGM in the absence of pumps.

Q: Some of my patients who have lost reimbursement continue to pay out of pocket for CGM: they say, “I can live without my pump but not without my CGM.”

Dr. Skyler: I think it’s a pity to have to choose, but I agree that if you have to choose, you are going to choose the CGM.

Dr. Nicholas Argento (Maryland Endocrine and Diabetes Center, Columbia, MD): Could the Gen 5 transmitter send signals to an alarm clock? For adults that are hard of hearing, those with sleep apnea, or just hard sleepers, hearing alarms is a major challenge.

Dr. Peyser: We are actively looking into that. If you look at sleep arousal thresholds, adults are around 65- 70 decibels. Children are 70-80 decibels. Many adolescents are 105 decibels. That’s basically a jet engine. We’re looking at louder alarms – certainly not a jet engine alarm – but something like that. It’s an active area of research and development at Dexcom.

Comment: Maybe for adolescents you need electric shock therapy. [Laughter]

Dr. Deiss: Patients are enthusiastic about mealtime data. Do your patients also use retrospective analysis? Do you think this new Studio software will help patients?

Dr. Prázný: If we educate patients well, they will be able to use this and profit from it. They can do retrospective analysis on their own, but they may need our help in that from the beginning.

Dr. Deiss: In your practical experience, how many patients are really downloading their data and looking at it at least once per week? In my experience, a very small number of patients are doing this.

Dr. Skyler: It’s more important for patients to be looking at the receiver all the time and letting it guide their life instead of looking at it at the end of the week. So I don’t get on them when they don’t download. I’m more concerned that they look at it on a frequent basis and take ongoing actions rather than after the fact.

Dr. Steil: Following Dr. Peyser’s answer about calibration – how far is Dexcom from a sensor that does not need calibration, or at least not daily calibration?

Dr. Peyser: I am hesitant to say an exact date. I will tell you that we have an active program and have done clinical studies with the Gen 6 sensor, which appears to be good enough to go with an initial calibration and last seven, 10, even 14 days. The issue is to develop a manufacturing process capable of producing sensors of that quality at the 5-10 million per year number. It is at least a few years out, but I am convinced that it is doable.


Corporate Symposium: The Challenge to Optimize Insulin Therapy: How New Diagnostic Concepts and Technology Can Support People with Diabetes and Their Healthcare Professionals (Sponsored by Roche Diagnostics GmbH)


Matthias Schweitzer, MD, PhD (Head of Medical and Scientific Affairs, Roche Diabetes Care, Mannheim, Germany)

Dr. Schweitzer emphasized that “glucose information is the key driver for any kind of diabetes management.” He pointed to the often-cited STeP and PRISMA studies to show the relationship between structured SMBG adherence and diabetes control, and he noted that CGM delivers even more and different quality information on glycemia. Calling for wider access to these technologies, he argued that “there is no justification to withhold patients with diabetes from access to glucose information.” Interestingly, in his conclusion he stressed that Roche Diabetes Care differentiates itself from other companies – especially from “low-cost (product-only) suppliers” – through constant investment in the development of medical concepts, clinical research, and new technologies. This seemed like a veiled reference to Walmart’s recent introduction of low-cost ReliOn Prime blood glucose strips (~$9 for a 50- count box of strips), which some industry players have pointed out will threaten R&D investment in the entire BGM field (R&D investment is generally determined as a percentage of profit, which stands to decrease with the additional pricing pressure from Walmart’s low-cost strips.)


Ralph Ziegler, MD (Dr. med. Ralph Ziegler und Kollegen, Muenster, Germany) and David Cavan, MD (Royal Bournemouth Hospital, Dorset, UK)

Dr. Ralph Ziegler and Dr. David Cavan provided additional detail and color on the first results of the automated bolus advisor control and usability study (ABACUS). As seen at Roche’s Media event in the morning, t he key finding of the study was that compared to the active control group (standard MDI), a greater percentage of patients in the intervention group (MDI with trained use of blood glucose meter containing a built-in bolus calculator) reached the target of more than a 0.5% A1c decrease from baseline. Both interventions showed a reduction in the number of subjects who reduced insulin to avoid hypoglycemia, although those who did so in the bolus advisor group had a greater A1c reduction than controls (0.8% vs. 0.4%). For our full discussion, please coverage of the Roche Diabetes Care Scientific Symposium below.

Questions and Answers:

Q: Do you have information on differences between those with type 1 diabetes and type 2 diabetes and are there differences between age groups related to different types?

Dr. Cavan: The short answer is no. There were relatively few patients with type 2 diabetes in study, but we haven’t analyzed differences between the two groups.

Q: Does the expert calculate insulin-on-board?

Dr. Cavan: Yes

Dr. Richard Bergenstal (International Diabetes Center, Minneapolis, MN): Is there any data on the percent who got to A1c less than 7.0%?

Dr. Cavan: Not yet, but there is more analysis to come.


Eric Renard, MD, PhD (Montpellier University Hospital, Montpellier, France)

Dr. Eric Renard gave a high-level, thoughtful overview of insulin pump therapy and a look toward the artificial pancreas – which he suggested should use intraperitoneal insulin delivery (he showed promising closed-loop data with the transcutaneous Accu-Chek DiaPort) and algorithms running on a smart phone. Drawing encouragement from the outpatient, ambulatory closed-loop studies that began in 2011 (including one run by his team in Montpellier), he said that the “closed loop is no more a dream…it can happen”).

  • Dr. Renard endorsed several design approaches for commercial insulin pumps. First, he spoke to the relative benefit of bolus calculators that target glucose levels at the midpoint of a patients’ target range as opposed to the margin of the target range. Second, he highlighted the value of pumps that allow users to change the profile of insulin boluses (i.e., single bolus, double bolus, square-wave bolus, or dual-wave bolus) according to the type of meal (i.e., high in fat vs. mixed carbohydrate and fat). Third, he suggested that infusion sets could be improved. While pump therapy has been developed and well researched for the past 30 years, he said that there has been little research on the infusion set.
  • To demonstrate the benefits of intra peritoneal insulin delivery, Dr. Renard showed a subject’s glucose profile from the JDRF DiaPort Closed Loop Trial comparing intra peritoneal vs. subcutaneous insulin delivery using a model predictive control algorithm. Intraperitoneal delivery showed less glucose variability and greater time in zone (Eric Renard, Howard Zisser, et al., personal data).
  • To demonstrate the feasibility of translating CGM information into patient advice, Dr. Renard pointed to the DIAdvisor system, which is designed to predict forthcoming glucose profiles and provide therapy advice (EASD poster 1029: “Clinical assessment of DIAdvisor Device Shows High Accuracy in Glucose Prediction at 20-min Horizon and a Coherence of most advices on therapy in patients with type 1 diabetes”).

Questions and Answers

Dr. Richard Bergenstal (International Diabetes Center, Minneapolis, MN): Does data on intraperitoneal insulin tell us we need more rapid acting insulin? Does it work faster?

A: The difference is the efficiency when you infuse it towards the liver. With the intra peritoneal route, you modulate better glucose prediction from the liver.

Dr. Bergenstal: How many basal rates does a typical patient need to have effective control? Of course there is no such thing as a typical patient…but should we be worried if a patient has only one or two levels or should we be worried if they have 16?

A: I would be worried if there are more than three or four basal rates in a day. It would be a dream to think when we change the basal we will have immediate actions – it takes a couple hours.


Richard Bergenstal, MD (International Diabetes Center, Minneapolis, MN)

The esteemed Dr. Richard Bergenstal concluded the symposium with his perspective on CGM. He opened with T1D Exchange data showing that despite high frequencies of severe hypoglycemia, CGM use has remained remarkably low. Encouragingly, Dr. Bergenstal believes that the recent 2012 ADA Clinical Practice Recommendations’ mention of CGM would advance the discussion around CGM use and CGM reimbursement. He further emphasized that facilitating discussion between clinicians and patients about CGM would improve CGM use and patient adherence. Notably, Dr. Bergenstal called for a redefinition of “good” glucose control. He said that good control should be based on more than just A1c and consider time in range, hypoglycemia, and glucose variability – we agree! Dr. Bergenstal concluded his presentation by presenting (very) preliminary findings from the REACT 3 study of CGM vs. structured SMBG in type 2 diabetes. Both tools could improve glucose control, but CGM may be more effective in minimizing hypoglycemia at the same level of A1c reduction. Given the lack of consensus on the role of CGM in patients with type 2 diabetes, we are eager to follow up when the full REACT 3 analysis is completed and hope that it inspires more prospective studies on glucose monitoring interventions for behavior change.

  • Echoing previous presenters, Dr. Bergenstal emphasized the importance of using glucose information to inform therapeutic decisions. While he said that this point should be obvious, for many clinicians the importance of obtaining and using glucose information is not as engrained in practice as it should be. We appreciate that Dr. Bergenstal stressed the need to spread this message – it seems familiar to us, which reminded us how lucky we are to consistently have access to learning opportunities like EASD.
  • Dr. Bergenstal highlighted the mention of CGM in the 2012 ADA Clinical Practice Recommendations, which says that CGM use should be considered with intensive insulin therapy in adults with type 1 diabetes (A level evidence), some children (C level evidence), and in patients with hypoglycemia unawareness (E level evidence; Dr. Bergenstal noted that the low evidence grade reflects a lack of randomized control studies, but he said that in practice CGM use is well-established for this subpopulation). He also expressed his hopes that the evaluation in the ADA Clinical Practice guidelines would help get payers and providers thinking more seriously about reimbursement and prescription, respectively.
  • To improve clinical practice and patient adherence with regards to CGM use, Dr. Bergenstal emphasized the importance of: 1) explaining the connection between CGM and A1c; 2) setting clear goals; 3) giving a consistent message about CGM use (i.e., real-time vs. retrospective use); 4) teaching the difference between individual readings and patterns; and 5) teaching how to respond to the data. Interestingly, Dr. Bergenstal said that the question on when to use CGM (real-time vs. retrospective) was being hotly debated and that more studies were needed for a definitive answer.
  • Dr. Bergenstal asked the audience for help in expanding the definition of “good” glucose control. To demonstrate, he displayed the glucose profiles of two patients who, despite having identical A1cs had vastly different glucose variability. Glucose control needs to be defined by more than just A1c, he said, and proposed that the definition should also consider: 1) time in target range; 2) hypoglycemia; and 3) glucose variability.
  • Preliminary finding from the REACT 3 study suggested that in patients with type 2 diabetes both SMBG (collected and analyzed in a structured way) and CGM can provide similar improvements in glucose control, but CGM may be more effective in minimizing hypoglycemia whilst improving control. The study randomized patients to receive either real-time CGM or SMBG for 16 weeks. Every two to four weeks, patients met with physicians to discuss blood glucose data and make necessary adjustments in medication. Blinded CGM data was collected from both groups at baseline, eight weeks, and 16 weeks to assess the primary endpoint of time in range.
    • Both groups showed substantial reductions in A1c. The SMBG group’s mean A1c decreased from 7.8% to 7.0%, and the CGM group’s mean A1c decreased from 8.1% to 7.1%.
    • Both groups showed similar patterns in improvement for area under the curve, time in range, and percent of time >180 mg/dl; however, CGM showed greater reductions in time spent hypoglycemic. The SMBG group in fact trended towards increased percentage of time spent <70 mg/dl, <60 mg/dl, and <50 mg/dl over the 16 weeks, though the percentages themselves were very low (i.e., for <70 mg/dl, SMBG at 16 weeks was less than 2.5% vs. less than 1.0% for CGM; for <50 mg/dl, SMBG at 16 weeks was ~0.6% vs. ~0.2 % for CGM).

Questions and Answers

Q: In REACT 3, was the CGM used on a real-time basis for the patients in that arm or was it just used by the health care professional?

A: Patients had it real-time, but on monthly to two-week intervals, information was printed out and discussed with providers.

Q: Was there a difference in nocturnal hypoglycemia for SMBG patients, since they wouldn’t be getting 7-point profiles then?

A: I’m not sure yet. This is brand new data. I know a few instances of that, but can’t tell you statistically yet.


Scientific Media Symposium: Personalized Diabetes Management – Cutting-Edge Therapy Approach & Technological Innovation for Enhanced Patient Benefit (Sponsored by Roche Diabetes Care)


Luc Vierstraete (President Roche Diabetes Care, Roche Diagnostics GmbH, Mannheim, Germany)

Previously a bank in 1901, the Humboldt Carre served as the intimate setting for the Roche Diabetes Care Scientific Media Symposium. The room was set with eleven tables donning sparkling water, juices, and delicious German chocolate. Mr. Luc Vierstraete took the podium briefly to welcome the ~30 media representatives from across the globe to the early morning event that would feature a series of presentations on personalized diabetes management.



Antonio Ceriello, MD (University of Udine, Udine, Italy)

Dr. Ceriello reviewed (and encouraged) the ongoing trend toward personalization of type 2 diabetes treatment guidelines; however, he argued that clinicians could use more advice on establishing treatment goals and encouraging patient cooperation, relative to the open-ended ADA/EASD 2012 position statement. As a template for personalized diabetes management, he presented a six-step cycle based heavily on structured self-monitoring of blood glucose (SMBG) and collaborative review of the results: structured education, structured SMBG, documentation, analysis, personalized treatment, and treatment efficacy assessment – which feeds a new round of education to start the cycle anew (Ceriello et al., Diabetes Res Clin Pract 2012). We appreciated Dr. Ceriello’s frank acknowledgement of the difficulty of preventive medicine (“I am a diabetologist, and I am clearly overweight”), and we enthusiastically support the practice of ongoing, empirically driven diabetes management.

  • Dr. Ceriello presented a Personalized Diabetes Management Cycle based on what he called a “very naïve but important idea” – that a successful treatment regimen requires the patient’s cooperation and that measurement is needed in order to gain useful feedback for guiding that regimen. The cycle was heavily based around collection and analysis of SMBG data – the accompanying graphic included testing and/or a computer (to display results) in nearly every panel.


Ralph Ziegler, MD (Dr. med. Ralph Ziegler und Kollegen, Muenster, Germany)

Dr. Ralph Ziegler presented first results from the Automated Bolus Advisor Control and Usability Study (ABACUS). The study randomized patients with poorly controlled type 1 or type 2 diabetes on MDI to receive either standard MDI therapy or bolus advisor (BA) supported MDI therapy, which came in the form of the Accu-Check Aviva Expert, a BGM with an integrated bolus advisor. At six months, a greater percentage of patients in the BA achieved the A1c reduction target (>0.5% change from baseline) compared to the standard MDI group (p <0.01). However, the benefits were statistically significant only among patients with perfect baseline competency scores in MDI and carbohydrate counting, underscoring the need for interventions that work in less-competent patients. Another education-related finding was that 38.1% of patients (n=83) at baseline reported reducing insulin doses independently of insulin level due to fear of hypoglycemia: fortunately this fear declined (in both groups) without a rise in hypoglycemia. We think that built-in bolus calculators certainly stand to add value by encouraging more appropriate bolus dosing. However, they seem likely to face serious FDA scrutiny – particularly on the meters’ insulin-on-board calculations, considering that these depend on the user to accurately report his or her insulin doses.

  • The multi-national, prospective Automated Bolus Advisor Control and Usability Study (ABACUS) randomized 218 patients with poorly controlled type 1 or type 2 diabetes (A1c >7.5%) on MDI to receive MDI standard therapy (control) or bolus advisor (BA) supported MDI therapy (intervention). The latter arm used the Accu-Chek Aviva Expert, a BGM with an integrated bolus advisor (BA) BA that determines prandial and correction insulin doses. The primary endpoint assessed was six-month change in A1c from baseline. Secondary endpoints included additional measures of glycemic control, patient use of BA, and psychosocial measures including fear of hypoglycemia.
  • Baseline assessments showed that A1c was generally lower among patients with higher competency scores in MDI and carbohydrate counting. Patients with perfect scores in baseline assessments of both categories (n=66) had a significantly lower mean A1c than patients with no perfect score in either MDI or carbohydrate counting (n=79; 8.6% vs. 9.1%; p<0.01). Dr. Ziegler explained that this finding underscores the importance of structured education in diabetes management. Following competency assessments, participants received remedial education in areas where deficiencies were identified.
  • Overall, 38.1% of patients (n=83) at baseline reported reducing insulin doses independently of insulin level due to fear of hypoglycemia. Survey results showed: 1) 45.9% of participants (n=100) were worried about hypoglycemia; 2) 65.1% (n=142) engaged in hypoglycemia avoidance behavior (i.e., reducing insulin dose or eating additional carbohydrates independent of glucose level); 3) 37.6% (n=82) were worried and engaged in avoidance behavior; and 4) only 26.6% (n=58) showed no worry or avoidance behavior.
  • A greater percentage of the intervention group achieved over a 0.5% A1c reduction from baseline compared to the control group (56% [n=56] vs. 34.4% [n=32]; p<0.01); the average A1c reduction in the intervention group was 1.2%. However, when considered by baseline level of MDI/carbohydrate counting competency, the between-group difference (control vs. intervention) was significant only in the group with perfect competency scores. (First results were assessed only for completers.)
  • Both groups showed a significant decrease in the number of patients who reduced their insulin dose out of fear of hypoglycemia (p <0.05). Within the control group, the percentage of patients employing this avoidance behavior decreased from 37.5% (n=42) to 25.6% (n=23). Within the intervention group, the percentage of patients decreased from 39.8% (n=41) to 27.4% (n=26). While the between group difference was not significant, Dr. Ziegler emphasized that the intervention group avoidance behavior change was accompanied by a greater reduction in A1c. And importantly, the use of the bolus advisor was not associated with increased frequency of severe hypoglycemia over the six months compared to controls.
  • We sense the FDA to be quite wary of meters with bolus calculators, not least because bolus calculators depend on patients to have comprehensively logged their insulin injections. According to, a UK ease-of-use study comparing Abbott’s FreeStyle InsuLinx with built-in bolus calculator to other glucose meters was recently completed (updated September 10, 2012; Identifier: NCT01432275) and we look forward to hearing results, which could potentially substantiate the value of the built-in-bolus calculator and help with FDA approval (Editors note: the study was originally slated to complete in January 2012). As a reminder, the FreeStyle InsuLinx is approved ex-US with a built-in bolus calculator, but the approval timeline in the US for this feature remains unclear (the FreeStyle InsuLinx meter without the calculator was FDA cleared in March 2011). For additional discussion on the FreeStyle InsuLinx, please see our Abbott 2Q12 report at Similarly, we hope to learn more about any wider launch plans for the Accu-Chek Aviva Expert at the exhibit hall.


Sanjoy Dutta, PhD (JDRF, New York, NY)

Dr. Dutta summarized JDRF’s groundbreaking efforts to encourage development of an artificial pancreas and outlined some of the main ongoing challenges. He said that the first generation of semi- automated-insulin-delivery products can be developed with today’s technologies but that the second generation will depend on better sensors and faster insulin action. He briefly mentioned longer-term efforts toward the third-generation artificial pancreas features (e.g., multi-hormone delivery, CGM sensors that are either implantable or on the same port as insulin delivery), and he noted that JDRF is also working on “out-of-the-box” approaches such as a self-regulating insulin that could recognize ambient glucose concentration and accordingly dose itself at a molecular level in real time. The overarching goal of JDRF’s Treat Therapies division, Dr. Dutta said, is to take a “holistic approach at achieving glucose control and overall metabolic balance” while keeping in mind the “changing landscape of type 1 diabetes and its etiology” (e.g., greater prevalence of overweight/obesity and concomitant insulin resistance).

  • Dr. Dutta said that the second generation of artificial pancreas products will require better glucose sensors and faster insulin action. To the former, JDRF and the Helmsley Charitable Trust to fund sensors with tighter accuracy and better form factor; so far they have partnered with Medtronic and BD. As for insulin delivery, Dr. Dutta highlighted Roche’s Accu- Chek DiaPort (a port that allows insulin pumps to reach the intraperitoneal cavity and thus act more physiologically than subcutaneous insulin), BD’s intradermal microneedles (which give “as low a pain level as you can imagine with invasive insulin delivery” but which are in the “very early” stages of clinical testing [Pettis et al., Diabetes Technol Ther 2011]), and Insuline’s InsuPatch (a heating device to improve vascular flow of subcutaneous insulin that has been studied at Yale University).


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

Dr. O’Neal argued that in order to improve diabetes outcomes, primary care physicians must be empowered to take a larger role in insulin initiation. To this end he is helping to conduct the Stepping Up study, a cluster-randomized trial of 58 primary care practices (290 patients with type 2 diabetes). The researchers will study whether A1c can be lowered more effectively than with standard care using insulin initiation and titration based on the Stepping Up protocol (which, in turn, is based on three days of seven-point SMBG profiles using the Accu-Chek 360-degree View paper-based tool). Dr. O’Neal noted that structured SMBG provides much more detail than “traditional” fasting morning glucose tests at a much lower cost than continuous glucose monitoring (CGM), making it viable for near-term reimbursement in Australia. He closed with an analogy of CGM vs. structured SMBG: “Instead of a Ferrari, we’ve got a Volkswagen; I think that’s what we need right now, and that’s why we’ve chosen it for the Stepping Up study.”



Michael Schoemaker, PhD (Roche Diagnostics GmbH, Mannheim, Germany)

Dr. Michael Schoemaker posited that the major limiting factor preventing widespread CGM use is lack of CGM accuracy and reliability, and he highlighted Roche Diabetes Care for their investigation into the sensor-to-tissue interface in order to improve both of these limiting factors. Details were scarce, but Dr. Schoemaker said that Roche’s sensor, in early-stage development, shows “outstanding” accuracy, precision, and reliability, especially in the hypoglycemia range. This update follows on the heels of the company’s 2Q12 announcement that it was restructuring Roche Diabetes Care division – a project that includes increasing R&D investment in insulin pump and CGM technologies. Undoubtedly, the shift in R&D focus is timely for Roche considering the persistent pricing pressures in the BGM market and growing recognition of the benefits of glycemic trend data. (For more discussion of Roche’s restructuring initiatives, please see our Roche 2Q12 report at and Roche Investor Day report at Dr. Schoemaker concluded that CGM can become the standard of care if certain improvements are made, in particular: 1) better accuracy and precision; 2) translation of complex CGM information into medically relevant and actionable information; and 3) improved user- friendliness.


-- by Adam Brown, Hannah Deming, Jessica Dong, Kira Maker, Nina Ran, Joseph Shivers, Tanayott