EMA clinical guidelines for diabetes drugs encourage use of CGM in clinical trials; key benefits and barriers for drug companies – July 3, 2013

Executive Highlights

  • The EMA clinical investigation guidelines for developing diabetes drugs mention CGM in five separate instances, providing a mild endorsement of the technology and signaling a growing industry trend.
  • We believe CGM has immense potential to improve data collection in clinical trials of drugs. This report includes our thoughts on the key advantages and key barriers, a brief list of recent drug trials using CGM, and our key questions.

Last year, the EMA clinical investigation guidelines for developing diabetes drugs had a noteworthy five mentions of CGM – we took this as an encouraging sign that awareness is growing of the benefits of CGM data as companies develop new drugs. The guidelines note that CGM “is encouraged and regarded as useful in adults and children to describe overnight glucose profiles and postprandial hyperglycemia.” Not exactly a mandate, though certainly a positive regulatory sign from where things were a few years ago. We are big proponents of this trend and hope to see it explode in drug trials in the coming years.

Following a review of the EMA guidelines, this report outlines what we think are the key benefits to using CGM in clinical trials: 1) much richer glycemic data than A1c; 2) additional glycemic data that SMBG alone cannot capture; 3) helps companies – especially those with limited resources – maximize data collection from early clinical studies; 4) potentially increased protocol compliance since CGM is less of a hassle for patients than frequent fingersticks; 5) captures more real-world data than SMBG (i.e., patients can advantageously time fingersticks; CGM is 24/7); 6) provides highly understandable and compelling marketing messages for patients and HCPs; and 7) could get more patients and clinicians comfortable with CGM and increase penetration of the technology.

This report also addresses what we think are some of the key barriers delaying wider use of CGM in clinical studies of drugs: 1) the FDA, the EMA, and other regulators must start to embrace CGM as a gold standard on par with A1c; 2) misperceptions that CGM is time consuming and not worth the effort; incorrect views that CGM is costly and not worth the upfront expense; 4) misperceptions that CGM isinaccurate, in its early stages, and not that reliable; and 5) a general lack of awareness that CGM could be a valuable option for use in clinical trials.

Just as when GLP-1 agonists came into the fold and suddenly everyone was asking, with novel new drugs, “What was the change in weight?”, we feel like we are getting to a similar point with novel drugs and CGM – audiences and KOLs are increasingly asking companies, “Did you use CGM?” or, “What happened with CGM?” As the table below makes clear, CGM can and is being used for many drug classes, and we certainly expect to see much more use in the next five years. We believe that the additional data has the potential to speed drug development, improve regulatory review times, and support unique labels and compelling marketing messages. Most of all, we hope it improves therapies and helps patient outcomes.


  • The EMA clinical investigation guidelines for developing diabetes drugs mention CGM in five separate instances. The first mention ( is the most ringing endorsement of CGM (“encouraged and regarded as useful”), though all mentions were positive. The language “should be considered” is of course non-binding and could be stronger, though we believe that with better CGM and easier data upload, the guidelines will more stridently recommend use of CGM.
    • Measures of glycemic control – Plasma Glucose. “The use of devices allowing continuous blood glucose monitoring is encouraged and regarded as useful in adults and children to describe overnight glucose profiles and postprandial hyperglycemia. Currently these methods still require traditional blood glucose measurements for calibration and it needs to be taken into consideration that glucose measurements from the interstitial fluid lag temporally behind blood glucose values. However, depending on the mode of action of the test agent and risk for hypoglycemia (particularly nocturnal hypoglycemia) of the study population, continuous blood glucose monitoring should be considered to provide additional relevant information.”
    • 4.4.2: Hypoglycemia. “Use of continuous glucose monitoring, providing more complete information on night profiles, should be considered, especially in patient groups at increased risk for hypoglycemia.”
    • 5.2: Assessment of Efficacy. “In addition to the evaluation of the overall blood glucose control by A1c, at least seven-point capillary-blood glucose profiles (before and after each meal and at bedtime) at regular intervals are necessary, particularly in type 1 diabetic patients. Alternatively, continuous glucose monitoring could be considered.”
    • 5.4.2: Children. “A1c is the recommended primary efficacy endpoint. Glycemic variability and hypoglycemic episodes are important secondary endpoints (see section 5.2). Both should be documented, preferably by continuous glucose measurements.”
    • 5.5.1: Safety Aspects – Hypoglycemia. “Hypoglycemia is the biggest obstacle to tight glucose control and is considerably more frequently observed in patients with type 1 diabetes than those with type 2 diabetes. Incidence and rate of both overall and severe hypoglycemia should be determined in all clinical trials. In order to assess nocturnal hypoglycemia, the use of continuous glucose monitoring devices should be considered. A relevant reduction of documented episodes of hypoglycemia, particularly severe events (see 7.2), if studied in appropriately controlled trials, could support a claim of superiorityover the insulin used as a comparator provided that this is not achieved with simply allowing A1c to rise.”
    • The 28-page EMA Guidelines document is posted at etail.jsp?webContentId=WC500129256&mid=WC0b01ac058009a3dc.


  1. CGM captures much richer glycemic data than A1c, the most typical efficacy endpoint in diabetes trials. Importantly, CGM can help investigators differentiate between fasting and postprandial reductions in glucose – these are key for understanding a drug’s mechanism of action, dosing, and benefits for patients. Additionally, A1c over a three-month period is too crude a metric in our view to adequately characterize glycemic control that changes on a minute-to-minute and day-to-day basis.
    • We hope to see time-in-range more extensively used, since it can truly give context to an A1c value (what we would call “quality of A1c”). For instance, if a drug reduces A1c by 1%, a company might consider that very robust efficacy and worth pursuing. But what if that same drug increases hypoglycemia by 50%? In our view, a drug’s effect on time spent in range is much more important than an A1c value – CGM is the only way to understand that effect.
    • Drs. Aaron Kowalski and Sanjoy Dutta recently wrote an excellent commentary entitled “It’s Time to Move from the A1c to Better Metrics for Diabetes Control” (Diabetes Technology & Therapeutics 2013). The article comes down strongly on using A1c alone, since it gives little information about hypoglycemia or day-to-day variability (e.g., “A mean speed of 55 mph over the past three months of commuting will never reflect times when a vehicle is racing at 100 mph and far in excess of the speed limit or is slowed to 10 mph in traffic congestion. Similarly, the clinician only receives the most basic of information from the HbA1c measurement and masks the occurrence and frequency of dangerous highs and lows”). We couldn’t have said it better. Drs. Kowalski and Dutta also remind readers of the JDRF CGM trial, where patients with an A1c level of 6.4% spent nearly 100 minutes a day – every day – at levels<70 mg/dl – if that’s not a clear and convincing reason to go beyond A1c, we don’t knowwhat is.
  2. CGM allows companies to obtain additional glycemic data that SMBG alone cannot explain. This is an especially key factor for detecting and understanding reductions in hypoglycemia, which can go undetected at night, in patients who are hypoglycemia unaware, or in those who treat a low instead of checking their blood glucose. This topic of discussion came up at the advisory committee for Novo Nordisk’s for insulin degludec – at the time, the FDA raised a number of objections to the hypoglycemia data collected by Novo Nordisk (e.g., were episodes captured accurately, was nocturnal data of high quality). We wonder if more robust and widespread CGM data would have cleared this issue up. For more information on the objections, see our full report on the insulin degludec advisory committee at
  3. CGM allows companies (especially those with limited resources) to maximize data collection from early clinical studies. In small phase 1 and phase 2 proof of concept studies, little is often known about a drug. Adding CGM can give companies a much richer sense of acompound’s product profile and glycemic efficacy. It can inform things like dose-timing and drug concentration, allowing companies to iterate and optimize compounds for subsequent trials. CGM data is also a huge asset in early trials, as it can give companies a sense of the A1c reductions that can be expected in larger studies. It may also aid small companies in obtaining partnerships or securing funding – time-in-range is pretty easy to understand and quantify, while A1c is more abstract and subject to limitations.
  4. Wearing CGM is less hassle for patients than frequent daily SMBGs. Many trials continue to require seven-point daily SMBG profiles – this is a lot to ask of many patients, especially many type 2s that are not used to checking blood glucose so often (and in probably the vast majority of type 1s that are testing around four or five times per day). In our view, some CGM systems are accurate enough that seven-point profiles aren’t needed to get a true sense of glycemic efficacy – it will be just as good to get enough SMBGs to calibrate the system and combine that data with the most accurate CGM on the market.
  5. CGM captures more real-world data than SMBGs alone (i.e., patients can advantageously time fingersticks, while CGM is 24/7). There are anecdotes of patients cheating seven point SMBG profiles by using the same blood sample again, using sugar water, changing the time on their meter, etc. CGM helps avoid these real-world limitations, since there is no way to cheat the data collection.
    • We would note that there is ongoing debate in the field concerning how ethical blinded CGM is. The debate stems from reports in the literature of hypoglycemic seizures while patients were wearing blinded CGM. Some argue that blinded CGM allows true and unbiased collection of glycemic data, preventing any confounding factors arising from changes in patients’ behavior. Others believe that blinded CGM is completely unethical, since the device can alert patients to potentially life threatening hypoglycemia. We would note that if a trial is properly controlled and randomized, any such behavior confounds should wash out between the control and intervention groups. Additionally, the same thing about changes in behavior could be said about using “real-time” SMBG, and no one would argue that patients should perform “blinded” SMBG.
  6. CGM can provide highly understandable and compelling marketing messages. We expect many patients will appreciate hearing that “Drug X” keeps you in range 40% more of the time or reduces hypoglycemia by 60%, a statistic that is much easier to understand (and much more convincing) than a 0.5% drop in A1c. We suspect HCPs that are not diabetologists would also appreciate hearing these more intuitive CGM-based metrics.
  7. More use of CGM in clinical trials could get more patients and clinicians comfortable with CGM and increase broad penetration of the technology. The vast, vast majority of patients are not on CGM, though they could certainly benefit from greater glycemic information – in our view, that’s independent of usage patterns (i.e., 24/7 real-time, intermittent real-time use a few times per year, or blinded professional use). We believe greater CGM use in clinical trials will, at minimum, help providers and patients become more familiar and comfortable with the technology. It could also help clear up any bad perceptions of inaccuracy stemming from early CGM generations.


  1. For CGM to be widely adopted and used in clinical trials, we believe the FDA, the EMA, and other regulators must start to embrace it as a gold standard on par with A1c. For now, it is still seen as nice-to-have, meaning companies will not prioritize it as much as A1c. Certainly, the EMA guidelines discussed above are progress in the right direction, but they are cautious and non-binding. The FDA has also come around a bit on the device side, as it will accept time-in-range as a valid endpoint for artificial pancreas studies (per the final AP guidance; see our report at; yet, we think the drug side of FDA is slower on this front, as the focus for so long has been on A1c. We would note that the FDA does accept CGM metrics as secondary endpoints. We hope it does not take a DCCT-style time-in-range trial to make it clear that this a very valid metric. In our view, more accurate technology and more studies to support the benefits of CGM will change the paradigm, as will more companies coming to FDA with both A1c and CGM data for their compounds.
    • Lack of a long-term trial linking CGM and glycemic variability to hard clinical outcomes may be a key limiting factor. In the past, landmark trials like the JDRF CGM trial demonstrated modest improvements in A1c ~0.5% or less – we attribute the efficacy to early-stage technology that was not that accurate, harder to use, and more frustrating for patients. As a result, rates of sensor wear were quite low in many of these studies, particularly in younger populations. We believe the latest generation of CGM has improved considerably – we hope newer studies will demonstrate larger reductions in A1c.
    • Dr. Irl Hirsch’s FLAT-SUGAR study should grant insight into the feasibility and clinical relevance of a larger glycemic variability study. FLAT-SUGAR randomizes 120 patients with type 2 diabetes to receive either exenatide, glargine, and metformin or basal-bolus therapy and metformin for six months to maintain A1c at 6.7- 7.3% during the trial. Primary outcome is change of coefficient of variation at zero, three, and six months as measured by CGM ( Identifier: NCT01524705). The study’s main aim is feasibility (showing that two groups can have the same A1c but different glycemic variability), but we hope it will pave the way for a larger outcomes study that could fundamentally change the way diabetes is treated and reimbursed. The study is currently recruiting and slated to complete in July 2014.
  2. Perceptions that CGM takes extra time. From what we’ve heard, there is still a perception that CGM is time-consuming for providers, especially for training. We think this is changing with systems like the Dexcom G4 Platinum and Medtronic iPro2 that require very little start up time, have less errors and reliability problems, require little or no upfront training, and in the case of Medtronic (likely soon for Dexcom with the SweetSpot platform and Qualcomm 2netHub), offer web-based systems that reduce the hassle of downloading and data compilation. There is also a snowball effect here – as more and more companies begin using CGM in clinical trials, Medtronic and Dexcom will continue to optimize and simplify the process (e.g., fast online training of clinical sites and investigators).
  3. Perceptions that CGM is costly and not worth the upfront expense. In our view, there are several potential benefits of CGM data that could outweigh the upfront cost: 1) more data out of trials that could reduce the need for large/long studies; 2) faster regulatory approval (or more positive advisory committees) because the drug is better characterized; 3) reduced burden on patients who will not need to perform as many SMBGs; and 4) additional data that can support marketing claims and label inclusions (e.g., 50% reduction in hypoglycemia).
  4. Views that CGM is inaccurate, in its early stages, and not that reliable. We think this is already changing with newer generation systems, though there’s still an uphill battle to overcome from very early-generation systems. Some newer systems are approaching fingerstick accuracy – this is not widely recognized or appreciated in our view. Next-gen systems will be even more accurate, and one day, CGM might even be calibration free or require a single calibration at startup.
    • Comparing the accuracy of different CGM systems is often challenging due todifferences in study design; 2) differences in calibration techniques/frequency; and 3) retrospective vs. prospective algorithms. As of this writing, the two systems best positioned for use in clinical trials are Medtronic’s iPro2 and Dexcom’s G4 Platinum.
      • As reported in the Journal of Diabetes Science & Technology, Letter to the Editor (Welsh et al., 2012), the iPro 2’s MARD is 9.9% in adults and 10.1% in children (retrospective calibration). Technically, the iPro2 only needs two fingersticks per day, but Medtronic recommends four for clinical trials to ensure two fingersticks are actually taken. The iPro2 separately captures and stores the raw sensor signal data, and upon returning to the clinic, it is downloaded along with a patient’s glucose meter. The iPro2’s raw sensor signal is then retrospectively fit to the blood glucose readings and translated into standard CGM traces.
      • Dexcom’s G4 Platinum pivotal study demonstrated a MARD of 13.2% (prospective calibration). This is expected to drop to 11% once Dexcom’s updated algorithm is out in 2014. In thinking about this topic of retrospective vs. prospective calibration, we were reminded of Dr. David Price’s presentation at ADA 2012 on CGM study design and analytic techniques. Using the aforementioned pivotal study data, he showed that the G4 Platinum’s MARD drops to 11% with retrospective calibration (twice per day). In short, we think it’s difficult to directly compare Medtronic’s iPro2 and Dexcom’s G4 Platinum on accuracy, though they are likely in the same ballpark around 10-11%.
  5. General lack of awareness that CGM could be a valuable option for use in clinical trials. CGM is still a relatively new technology that has emerged in the last decade, so there is still awareness building that needs to happen. We are confident this will change drastically in the next decade as more patients wear CGM, as the technology improves, and as more trials use it. As penetration of the technology expands into more and more centers across the globe, it should also be easier to integrate CGM into clinical trials at increasingly experienced centers.


  • CGM is drug class-agnostic in our view. See the table for a non-exhaustive list of several companies that have or are using CGM to gather additional data about their compounds.
Drug Class Company Compound Study/ Identifer

Novo Nordisk

Eli Lilly


Insulin degludec

LY2605541 (basal)

Insulin glargine


346-OR at ADA 2012

ORIGIN substudy at EASD 2012






DPP-4 inhibitors




GLP-1 agonists




Glucokinase activators

Eli Lilly




Q: What is the biggest benefit for companies that use CGM in clinical trials?

Q: What is the most challenging part of using CGM in a clinical trial? How can this be overcome?

Q: What is the biggest barrier preventing more use of CGM in clinical trials – Cost? Hassle factor? Data upload? Still considered a nice-to-have? Lack of enthusiasm from regulators?

Q: What drug classes stand to benefit the most from CGM data?

Q: The EMA seems to be further ahead than the FDA regarding CGM use in clinical trials – why is that? What would it take to get FDA more on board with CGM use in clinical trials?

Q: Will a large-scale time-in-zone/glycemic variability outcomes study ever be conducted? Is this needed to truly put CGM data on par with A1c?

Q: Will a day come when regulators require CGM data for all drugs in development? Will that take factory calibrated CGM?

Q: Will greater use of CGM in clinical trials increase broader penetration of the technology?

-- by Adam Brown, Kira Maker, and Kelly Close