Have you ever been suspicious of marketing claims on CGM accuracy for people with type 1 diabetes?
Me too.
After reading Bad Science, by Ben Goldacre, I decided to investigate continuous glucose monitoring (CGM) accuracy claims.
Ben is a doctor who uses evidence-based principles and critical appraisal to assess marketing claims.
If the claims stand up to scrutiny, he applauds, if they fall short, he explains why, so the consumer can make an informed decision.
This is my attempt to help people in the type 1 diabetes community better understand CGM accuracy.
Also, I am on a mission to prevent British and European diabetes care from being flooded with untested CGM devices (this is happening right now!), which may cause serious safety issues, considering their glucose values are being used to determine insulin doses and hypo treatment decisions!
Can the marketing claims be trusted, and are they helpful?
After reading, the reviewer will:
- Understand MARD % and why it does not tell us much!
- Understand 20/20 and why it is misleading.
- Discover 15/15 and why this is a better measure of CGM trustworthiness.
- Compare all CGMs to identify the most trustworthy.
- Find out which CGM devices have accuracy data on what type of people, adults vs children, type 1 vs type 2 diabetes.
- Compare all CGMs on features such as hypo protection, blood glucose testing requirements, and data sharing capability.
- Finally, be ready to take action to protect the diabetes community from untested CGM devices.
My hope is that:
- A few CGM marketers might read this blog. It will give them a dose of type 1 diabetes reality that is missing in their glossy brochures!
- The blog finds its way to the CE Mark regulators for CGM devices. A group of people who desperately need to read this!
- It reaches some local commissioners and health insurance reimbursement teams so they understand price should only be considered once there is parity in accuracy and functionality.
- The MHRA catch wind of what is going on in time to stop CGM devices with little or no accuracy data flooding the market causing a price war, where people with diabetes are the casualties. Just as happened with blood glucose strips not too long ago.
One can only hope.
Before we can compare CGMs we must understand how accuracy is ‘supposed’ to be measured and reported.
- A CGM company asks multiple diabetes centres across the world to independently measure the accuracy of their CGM.
- Several centres agree and hopefully, a large age range (2-70yrs) of type 1’s are recruited.
- Each person has CGM fitted as per manufacturer guidance.
- They come into the lab on multiple occasions during the sensor life e.g. days 1,3,7,10.
- Each visit they have a venous cannula inserted to take blood glucose readings by a very accurate measuring device – Yellow Spring Instrument (YSI).
- During the visit they have meals, hypos are induced, and ideally, do some exercise (though rarely).
- The CGM readings are compared with blood glucose readings every 5 minutes during their visit.
- The lead researcher analyses the data and writes up the key findings such as MARD%, 20/20, 15/15, and others.
- The results are published in a peer-reviewed journal for transparency and public scrutiny.
- The results should NOT be squirrelled away as data on file by the company (why would they do this? I bet you have some thoughts)
There are other details I am skipping such as the results of children aged 6 and under are compared against a finger-prick blood glucose monitor, not YSI.
However, they are not of primary importance for our discussion, and I have produced a detailed table at the end that shows all the accuracy measures and methodologies.
From an ivory tower perspective, we should not be comparing the accuracy studies against each other because they are so different, and many are inadequate.
For example, a review of CGM accuracy during exercise reported the time lag between blood glucose and sensor glucose increases from 5 minutes to 12-24 minutes causing deterioration in MARD to above 13%, due to fast-changing glucose levels.
Therefore it’s clear how to get very accurate CGM results.
Test CGM accuracy when the glucose level is very stable.
Who would be the ideal population for this?
Adults with type 2 diabetes, especially those not on insulin.
Coincidentally, this is the exact population used in the accuracy studies for the Medtrum and GluoRX Aidex.
In both studies, over 80% of the participant had type 2 diabetes, and of those for GlucoRX Aidex, less than 50% were using insulin. The Medtrum study does not report how many of the type 2’s were using insulin.
However, when it came to CE Mark, both CGM devices were granted approval for adults and children with type 1 diabetes.
Surely you are joking Mr Feinmann?
Let me be clear.
I do not blame the CGM companies, they are playing within the rules. The issue is the loopholes that allow approval to be obtained so easily, without peer-reviewed published data.
This article is not trying to blame and point fingers. It is aimed to raise awareness of the loopholes with real-life examples, and boost motivation within the type 1 diabetes community to plug the holes.
In May 2022, CGM systems with questionable accuracy data have CE Mark, and their accuracy data is being marketed to people with type 1 diabetes, health care professionals, local commissioners and reimbursement teams for health insurers.
Therefore, it is imperative we compare while being mindful of the difference in study designs and make every attempt to penalise inadequacies during comparison.
To ensure we are not misled by poor study design, I will build on the scoring system for CGM devices I published in 2020.
For now, let’s just say study design and reporting of results is not consistent and often key measures are missing.
In fact, some companies don’t publish the results in a peer-reviewed journal and still get CE Mark – say hello to the Glucomen Day and Medtronic Enlite 4.
I digress, let’s get back to marketing.
When the company gets the data from the researcher, they set about marketing their product.
First, they must show the accuracy of their product is close to the best on the market.
They cannot lie, but they can be economical with the facts.
This is where MARD % comes in, the number that flatters to deceive.
Mean Average Relative Difference % (MARD %) reports, on average, how different the paired blood glucose and sensor glucose readings were when tested in the study.
Take a glance at this table, it suggests the accuracy is similar for all devices, about 10% MARD.
| CGM | MARD % |
| Dexcom G6 (1) | 9.6-9.8% |
| Dexcom G7 (2,3) | 8.1-9.1% |
| Enlite 3 (4) | 9.6% |
| Enlite 4 (data on file) | 10.6-12.0% |
| Eversense (5) | 9.6% |
| Glucomen (6) & data on file | 9.7-11.4% |
| Libre 1-3 (7) | 9.2-9.7% |
| Medtrum A6 (8) | 9.1% |
Here is the trick,
allow the reader to interpret the MARD % without any guidance.
Most people interpret MARD % as shown in the two tables below, first in mmol/L and the second in mg/dL.
| BLOOD GLUCOSE 3.5mmol/L | BLOOD GLUCOSE 10mmol/L | BLOOD GLUCOSE 15mmol/L | |
| Accuracy range 10% | SENSOR GLUCOSE range 3.2 – 3.8mmol/L | SENSOR GLUCOSE range 9.0 – 11.0mmol/L | SENSOR GLUCOSE range 13.5 – 16.5mmol/L |
| BLOOD GLUCOSE 65mg/dL | BLOOD GLUCOSE 180mg/dL | BLOOD GLUCOSE 270mg/dL | |
| Accuracy range 10% | SENSOR GLUCOSE range 59 – 71 mg/dL | SENSOR GLUCOSE range 172 – 198 mg/dL | SENSOR GLUCOSE range 240 – 298 mg/dL |
The tables show the accuracy range for the sensor glucose at three different blood glucose readings for 10% accuracy.
As all the CGM MARDs are about 10%, therefore most people think all sensor glucose readings fall within the ranges shown in the tables.
I wish this was true, however, 10% is the average with a lot of variabilities.
The reality is, that only 50-60% of the readings fall within a 10% accuracy range!
Quite clearly, marketers would be sacked for writing,
Our product has a MARD of 10%, but remember, only 50% of the readings fall within a 10% accuracy range – that’s being too truthful!
Do people with type 1 diabetes need all CGM readings within 10%?
Here lies the pivotal question.
What level of accuracy is needed for a CGM reading to be trustworthy enough?
Trustworthy to allow effective treatment hypos, safe corrections from high glucose levels, and peace of mind when in target?
Is the 20/20 percentage trustworthy?
Let’s investigate.
Here are the 20/20 percentages of trustworthy CGM readings for the different CGMs.
| Sensor & Device | 20/20: Marketed percentage of trustworthy CGM readings |
| Dexcom G6 (1) | 92% |
| Dexcom G7 (2,3) | 93% |
| Enlite 3 (4) | 89% |
| Enlite 4 (data on file) | 85% |
| Eversense (5) | 93% |
| Glucomen (6) & data on file | 89% |
| Libre 1-3 (7) | 93% |
| Medtrum A6 (8) | 91% |
All about 90%, great, we can trust 9 out of 10 readings, right?
Not quite,
In the words of Ben Goldacre, “I think you will find it’s a little more complicated than that.”
First, we must define 20/20.
The first 20 speaks to the level of accuracy when the glucose level is less than 3.9mmol/L or 70mg/dL, what we will now call the hypo range.
Specifically, in the lab, when the sensor glucose reading was in the hypo range, it would be deemed accurate if the blood glucose reading was within 1.1mmol/L or 20mg/dL.
For example, if the sensor glucose reading is 3.5mmol/L or 65mg/dL, it would be deemed accurate if the blood glucose reading was within 2.4-4.6mmol/L or 45-85mg/dL.
The second 20 speaks to the level of accuracy when the glucose level is more than 3.9mmol/L or 70mg/dL.
Specifically, in the lab, when the sensor glucose was above 3.9mmol/L or 70mg/dL it would be deemed accurate if the blood glucose reading was within 20%.
For example, if the sensor glucose reading is 10.0mmol/L or 180mg/dL, it would be deemed accurate if the blood glucose reading was within 8.0-12.0mmol/L or 144-216mg/dL.
I believe 20/20 gives people in the diabetes community a false sense of confidence in CGM accuracy.
Let me expand on my belief of why false confidence, with three points.
First, the accuracy required in the hypo range is too wide. If a person has a sensor glucose of 3.9mmol/L or 70mg/dL and their alarm goes off, they could be treating a blood glucose level of 5.0mmol/L or 90mg/dL, as they are within 1.1mmol/L or 20mg/dL.
On the flip side, a person could be tootling around at 5.0mmol/L or 90mg/dL on their CGM, whilst their actual blood glucose level is 3.9mmol/L or 70mg/dL.
Practically, it seems 1.1mmol/L or 20mg/dL is just too generous and risks mismanagement of hypos.
I believe that using a 0.8mmol/L or 15mg/dL allowance reduces mismanagement: fewer unnecessary hypo treatments and missed hypos.
Secondly, the accuracy of within 20% when the glucose level is above 3.9mmol/L or 70mg/dL is problematic.
If a person has sensor glucose of 10.0mmol/L or 180mg/dL and their alarm goes off, they could be correcting a high blood glucose level of 8.0-12.0mmol/L or 144-216mg/dL, as they are both within 20%.
If the blood glucose level is on the lower side, say 8.0mmol/L or 144mg/dL, the correction dose calculated to drop the glucose level to 5.5mmol/L of 100mg/dL will be too aggressive leading to a hypo of 3.5mmol/L or 65mg/dL.
On the flip side, if the blood glucose level is on the high side, say 12.0mmol/L or 216mg/dL, the correction dose calculated will be insufficient.
Cutting the accuracy allowance to 15% minimises a lot of over and under correcting issues.
Cutting to 10% may even be better, however, 15% is accurate enough to make trustworthy decisions, in my opinion.
Finally, and most importantly, the 20/20 reported as one percentage does not tell the accuracy in the:
- Hypo range: less than 3.9mmol/L of 70mg/dL
- Target range: 3.9-10.0mmol/L or 70-180mg/dL
- High range: above 10.0mmol/L or 180mg/dL
This allows companies to hide that their CGM is inaccurate in the hypo range because it has good accuracy in the target range and high range.
This problem remains when the 15/15 percentage results are combined rather than broken down into ranges, see this table.
| Sensor & Device | 15/15: Marketed percentage of trustworthy CGM readings |
| Dexcom G6 (1) | 81% |
| Dexcom G7 (2,3) | 87% |
| Enlite 3 (4) | 81% |
| Enlite 4 (data on file) | * |
| Eversense (5) | 85% |
| Glucomen (6) & data on file | * |
| Libre 1-3 (7) | 86% |
| Medtrum A6 (8) | 82% |
*Data not reported
If you follow my logic, what we need is:
- Accuracy in hypo range within 0.8mmol/L or 15mg/dL
- Accuracy in the target range within 15%
- Accuracy in the high range within 15%
Not only that but robust standards on the percentage accuracy are required in all the above target ranges to gain approval for use.
Someone must have to have thought of this?
Of course.
Let me introduce you to the FDA interoperable CGM (iCGM) standards (9).
The standards are comprehensive and robust, however, to prevent boring you, I have pulled out what I believe to be the meat and potatoes, the iCGM 15/15.
- 85% or more CGM readings within 0.8mmol/L or 15mg/dL in the hypo range (<3.9mmol/L or 70mg/dL)
- 70% or more CGM readings within 15% in the target range (3.9-10.0mmol/L or 70-180mg/dL)
- 80% or more CGM readings within 15% in the high range (>10.0mmol/L or 180mg/dL)
Readers who want the full criteria and standards required to achieve FDA iCGM approval should read this excellent paper by Timothy Bailey and Shridhara Alva (10).
How do all the CGMs stand up to iCGM 15/15 criteria?
You would have thought this would be easy to answer.
Incorrect.
I have had to delve deep, very deep into the research papers, and data on file and harass marketing departments for information.
All the raw data is presented in the table at the end of this blog, but it’s overwhelming.
To simplify, I devised a scoring system to compare the accuracy of all CGMs in the table below.
I have included the age ranges and populations studied, as this should raise some eyebrows.
| Points | Hypo range trustworthiness: Accuracy of CGM within 15mg/dL or 0.8mmol/L when the glucose level is <70mg/dL or <3.9mmol/L | Target range trustworthiness: Accuracy of CGM within 15% when the glucose level is 70-180mg/dL or 3.9-10.0mmol/L | High range trustworthiness: Accuracy of CGM within 15% when the glucose level is >180mg/dL or >10.0mmol/L | Accuracy data population: Subjects age range of accuracy data in years |
| 0 | <50% | <50% | <50% | <50% type 1D |
| 1 | 50-75% | 50-59.9% | 50-74.9% | 24-70 |
| 2 | 75-79.9% | 60-64.9% | 70-74.9% | 18-70 |
| 3 | 80-84.9% | 65-69.9% | 75-79.9% | 6-70 |
| 4 | 85-89.9%* | 70-74.9%* | 80-84.9%* | 4 – 70 |
| 5 | ≥90% | ≥75% | ≥85% | 2 – 70 |
* = FDA iCGM standard (9)
What do all the CGM devices score out of 20?
This table puts the CGMs head-to-head on the iCGM 15/15 criteria.
| Sensor & Device | Hypo Accuracy: Out of 5 | In target accuracy: Out of 5 | High accuracy: Out of 5 | The age range of accuracy data: Out of 5 | Accuracy Score: Out of 20 |
| Dexcom G6 (1) | 4 | 5 | 5 | 5 | 19 |
| Dexcom G7 (2,3) | 4 | 5 | 5 | 5 | 19 |
| Enlite 3 (4) | 1 | 5 | 5 | 3 | 14 |
| Enlite 4 (data on file) | * | * | * | 5 | 7 |
| Eversense (5) | 5 | 5 | 5 | 2 | 17 |
| Glucomen (6) & data on file | * | * | * | 3 | 3 |
| Libre 1-3 (7) | 5 | 5 | 5 | 4 | 19 |
| Medtrum A6 (8) | 0 | 5 | 5 | 0 | 10 |
*Data not reported
Eyeballing the table, you will not be surprised to learn:
- To date (May 2022), only the Dexcom G6 (age 2+) and Libre 2 (age 4+) have received iCGM approval. This means they can be used with automated insulin delivery systems without a need for further testing. They have CE Mark for type 1 diabetes ages 2 and over (G6) and 4 and over (Libre 2), exactly in line with the published data.
- I have no doubt the G7 and Libre 3 will achieve iCGM approval once the paperwork has been submitted and assessed. They also have CE Mark for type 1 diabetes ages 2 and over (G7) and 4 and over (Libre 3), exactly in line with the published data.
- Potentially the Eversense could achieve iCGM status for adults, but I am not sure if the overall accuracy performance would meet the full standards (10). Eversense has CE MArk for adults with type 1 diabetes, exactly in line with the published data.
- The other CGMs have not filed an application for iCGM status, I wonder why?
What you will be amazed to learn:
- Medtrum A6 has been given a CE Mark (European approval) for use as standalone CGM and to be integrated with their pump to make automatic adjustments to insulin for people with type 1 diabetes aged 2 and over. The accuracy data is from adults with type 2 diabetes (84%) mainly with a small number of type 1 adults (16%), nothing on children, and the accuracy in the hypo range at 15/15 is 36%. To me, this means the regulator who approved this must really not understand CGM accuracy. If you know them, please forward this blog. The Medtrum sensor may be accurate for people with type 1, the point is, we just don’t have the data to say so.
- The Enlite 3 used with the Guardian Connect and Minimed 670G has poor accuracy in the hypo range. This may raise an eyebrow or two considering for the 670G, the sensor readings in the hypo range are used to stop insulin.
- The Enlite 4 used in the Minimed 780G does not have a published peer-reviewed paper and this is of concern considering the 20/20 of 85% is worse than the Enlite 3’s 89%. Knowing the CGM readings from the Enilte 4 are driving the automated insulin changes for the 780G algorithm, surely the accuracy data should be in public view?
- The Glucomen CGM has very limited data on file and one published study of eight people! Despite this, it has been given CE mark for children aged 6 years and older. Based on what data I ask? Again, if you know the CE mark regulator, please forward this blog.
- There is a new product to market I have not been able to fully assess, yet, the GlucoRX Aidex. Here is a summary of the accuracy study from the journal article (13), MARD of 9.1% measured in 114 adults (18-75yrs) with diabetes, only 13 (11%) had type 1 diabetes and 101 (89%) had type 2 diabetes. We will stop here as this is a study on adults with type 2 diabetes. Yet, it has CE Mark and is being marketed at type 1 diabetes aged 14 and over – how is this possible? Read this great blog for more depth – GluocRX Aidex.
If you have one of the CGMs with little or poor accuracy data, please do not panic!
If the CGM is working well and improving your life that’s fantastic. For example, the 670G and 780G have robust trial data showing they are effective and safe. I am not saying they are not.
However, I am saying, as a type 1 diabetes community, do we not deserve transparency, especially regarding approval for CE Mark?
I feel strongly that to achieve CE Mark there needs to be a robust criterion in place similar to the FDA’s iCGM.
I am not saying the standards have to be as high as the iCGM.
But I am saying the standards and criteria need to be published, and all CGMs with approval should have a document showing how their peer-reviewed published accuracy study data meets those standards.
If I was the regulator, the minimum standards I would enforce would be:
Accuracy study requirements:
- All studies must be published in a peer-reviewed journal before applying for approval.
- The study must report type 1 and type 2 diabetes data separately.
- The study must report under 18’s and 18 and over separately.
- The study must include activities that make the glucose go up and down – eating, hypos, activity.
- Accuracy must be reported in the hypo (<3.9 or <70mg/dL), target (3.9-10.0mmol/L or 70-180mg/dL) and high (>10.0mmol/L or >180mg/dL) ranges for 15/15, 20/20, 30/30 and 40/40.
- The study must report the percentage of sensors that last the full life.
Standards for approval:
- At least 5,000 (minimum 10% <3.9 or <70mg/dL) paired readings with YSI in each population approval is requested for (Type 1<18yrs, Type 1 >18yrs, Type 2 <18yrs, type 2 over 18yrs).
- Hypo range (<3.9 or <70mg/dL): 75% withing 0.8mmol/L or 15mg/dL, , 80% within 1.1mmol/L or 20mg/dL, 90% withing 1.7mmol/L or 30mg/dL, 95% within 2.2mmol/L or 40mg/dL.
- Target range (3.9-10.0mmol/L or 70-180mg/dL): 75% within 15%, 85% within 20%, 95% within 30%, 99% within 40%.
- High range (>10.0mmol/L or >180mg/dL): 80% within 15%, 85% within 20%, 95% within 30%, 99% within 40%.
- At least 75% of sensors last the full life.
Reporting of approval:
- For each approval population (Type 1<18, Type 1 >18, Type 2 <18, type 2 over 18), a document is produced showing the percentages of all accuracy requirements and citing the research papers, to allow the public to scrutiny.
If this was in place right now:
- The G6 would have approval for adults and children with type 1 diabetes.
- The G7 would need to publish the percentage of sensors lasting the full 10 days before approval.
- The Libre 1-3 would have approval for children with type 1 diabetes.
- The Libre 1-3 would be touch and go for adults, as only 71% of sensors last the full 14 days in adults. Maybe the approval would only be for 12 days when more than 75% of sensors last the distance?
- The Eversense would have approval for adults with type 1 diabetes.
If this was required for CE Mark historically, I feel confident that:
- Medtrum would not have received a CE Mark for type 1 diabetes and may be struggling for type 2 adults (Only 1,678 paired YSI readings and only 36% accuracy in hypo range within 0.8mmol/L or 15mg/dL).
- CE mark for the 780G would be pending until published data on the accuracy of the Enlite 4 is publicly available.
- CE mark for Glucomen would be pending until published accuracy data is publicly available.
- GlucoRX Aidex would not have received CE Mark for type 1 diabetes, only probably type 2 adults. Consider, the study reported 92% accuracy in the hypo range within 0.8mmol/L or 15mg/dL. However, only 2% (320 of 14,586) of the paired YSI readings were in the hypo range as they did not induce hypos in the study – probably because most were type 2’s not on insulin???
- New companies to the CGM market will not see the UK and Europe as a soft targets to introduce their experimental products.
I am all for market forces and healthy competition to drive down the cost, BUT, only if all products meet a minimum standard of accuracy – we need this desperately!
Without minimum standards, the UK especially, and some member states of Europe will be the experimental playground of companies trying to profiteer in the CGM space. The upside is that we will have first go on all the new technology. The downside is that we will play Russian Roulette with the lives of people with diabetes. In this case, taking the precautionary principle is required, especially considering we have CGM devices that have proven accuracy in specific populations.
Is there more than accuracy to consider?
You bet there is.
Everything from hypo alarms, blood glucose testing required, warm-up times, data sharing and compatibility with automated insulin delivery systems.
Rather than going through everything in detail, I have comprised another scoring system on the features I think are the most important.
The table is not exhaustive, you may value other things, that’s a personal choice, but you are stuck with these as it’s my blog.
| Points | Hypo Prevention alarms & alerts | Blood glucose testing required | Real-time data sharing | Warm-up time from insertion | Automated Insulin Delivery Systems compatibility |
| 0 | 0 | Confirm <4.0mmol/L & >13.9mmol/L | Data after scan | 2 hours | 0 |
| 1 | 1 | Calibrations essential | Real-time alarms by message to followers | 1 hour | 1 |
| 2 | 2 | Only if symptoms do not match the sensor reading | Real-time data & alarms for followers | 30 minutes | 2 |
| 3 | 3+ | Only if symptoms do not match sensor reading and optional calibrations to stop prolonged false alarming | † | † | 3+ |
The scoring system is self-explanatory, but I have highlighted one thing – optional calibrations to stop prolonged false alarming.
I have used most CGMs and the Dexcom and Libre products are the best in my opinion.
However, I think what sets the Dexcom apart is the ability to calibrate with finger-prick blood glucose if required.
You may wonder why I think this is so important?
Consider what happens to me at least once a week.
I am lying in bed enjoying a luxurious sleep when BEEP BEEP, the low alarm goes off at 3.9mmol/L (70mg/dL) with a steady arrow across. My wife is annoyed and I feel no symptoms, so I don’t want to crunch four dextrose tablets.
I check my blood glucose twice, 4.4mmol/L (79mg/dL) and 4.5mmol/L (80mg/dL), both within the 15/15 accuracy. I calibrate the Dexcom twice to 4.5mmol/L (80mg/dL), the number on the Dexcom changes to 4.5mmol/L (80mg/dL) and I enjoy a lovely night’s sleep.
If the glucose drops later, I am protected, I have a low alarm still on.
When I used Libre 2 and 3, I cannot calibrate, therefore, I have a decision to make.
Option 1, unnecessarily crunch four dextroses to stop the false alarm.
Option 2, turn off the low alarm for the night and hope I do not drop further.
Both are poor options!
I have seen this happen so many times for school kids too.
The low alarm goes off at school, the child says they feel ok, and blood glucose is checked and not hypo.
On the Dexcom they check the blood glucose again just to make sure they calibrate, problem solved, and confidence regained.
On the Libre, the alarm keeps going off and confidence is lost.
Being able to calibrate gives the user a quick way to restore confidence in the CGM when the discrepancy between the sensor and blood glucose is large.
Also, being able to calibrate stops persistent false alarming which can really cause alarm fatigue.
I am sure Libre will add this to their device in time.
I will be amazed if you cannot calibrate the Libre when used with automated insulin delivery systems in the near future.
Remember the CGM readings drive the insulin changes, so if there is a big discrepancy between blood glucose and sensor glucose it will need to be calibrated!
Once this is in place, the Libre products will be not far behind.
When calibrating please make sure:
- The blood glucose meter is an accurate one
- The blood glucose meter has been checked for accuracy in the last six months
- Test the blood glucose twice using proper technique and hand washing
Here is how the CGM’s scored on my features scoring system
| Sensor & Device | Hypo protection | Blood glucose testing required | Real-time data sharing | Warm-up time from insertion | Automated Insulin Delivery Systems connection | Score out of 13 |
| Dexcom G6 (1) | 3 | 3 | 2 | 0 | 3 | 11 |
| Dexcom G7 (2,3) | 3 | 3 | 2 | 2 | 0 | 10 |
| Enlite 3 (4) | 3 | 1 | 1 | 0 | 1 | 6 |
| Enlite 4 Data on file | 3 | 3 | 2 | 0 | 1 | 9 |
| Eversense (5) | 3 | 1 | 2 | 2 | 0 | 8 |
| Glucomen (6) & data on file | 3 | 1 | 2 | 1 | 0 | 7 |
| Libre (7) | 0 | 2 | 0 | 1 | 0 | 3 |
| Libre 2 (7) | 1 | 2 | 1 | 1 | 0* | 4 |
| Libre 3 (7) | 1 | 2 | 2 | 1 | 0 | 5 |
| Medtrum A6 (8) | 3 | 1 | 2 | 0 | 1 | 8 |
*Coming very soon
What should be clear is:
- The Dexcom products offer the best all-around features. Please note if you want to use the Dexcom with an automated insulin delivery system it will need to be the Dexcom G6 for now.
- The Libre products focus very much on minimal interruption to the user with no finger pricks and limited optional alarms. Many people with type 1 diabetes love this as too many bells and whistles are overwhelming.
Recent innovations?
The Dexcom G7 has a couple of innovations that are worth mentioning.
First, is the delayed first alarm for highs. It can be programmed so the high alarm only goes off after a specified time, to stop annoying after-meal alarms.
For example, the high alarm can be set at 10.0mmol/L (180mg/dL) but only go off when it’s been above that level for say 90 minutes. This means a brief high after breakfast will not trigger an alarm that you can do little about.
Second, the rate of change alarm can be tethered to a glucose level.
Previously if set the fall alert to go off with a double arrow down, it could go off at 12.0mmol/L (215mg/dL) after a correction or doing exercise, a useless alarm!
But now you can set it so the fall alert only goes off with a double arrow down below say 6.0mmol/L (110mg/dL) when hypo prevention is needed.
In fact, this tethering works perfectly for MATCH from Dynamic Glucose Management.
Similarly, you can tether the rise alert to a high glucose level.
For example, the rise alert only goes off if the double arrow up above 10.0mmol/L (180mg/dL). This means the alarm only goes off when you need to take action, reducing the alarm burden.
In fact, this works perfectly with GAME of Dynamic Glucose Management.
In this case, the rise alert with the double arrow above 10.0mmol/L (180mg/dL) would mean 30 minutes of walking is needed to get it back to target.
Let’s wrap this up.
You are probably amazed it’s possible to get CE Mark for adults and children with type1 diabetes, despite the accuracy data submitted being on adults with type 2 diabetes.
More importantly, there seems to be no UK specific appraisal of the products before they are marketed to patients, health care professionals and reimbursement teams.
It’s hard to say where the loophole is. Is it at CE Mark approval, or is it at country-specific regulation?
For me, it would make sense for CE Mark to create more robust regulations.
However, this is unlikely to change quickly, so national safety organisations, such as the MHRA will need to take this on for now.
By now it should be clear MARD % and 20/20 are inadequate accuracy measures to let people with type 1 diabetes know how trustworthy the CGM will be.
You may agree that the 15/15 criteria and FDA iCGM standards give a much clearer view of trustworthiness.
I hope you agree that something needs to change with how the CE Mark is obtained.
There need to be clear and published criteria, standards, and published performance data for each product that has received approval.
It should be obvious I am massively biased towards Dexcom products, but hopefully for good reason.
The Dexcom products have robust accuracy data, they offer the best choice of features, the G6 can be used with a wide range of automated insulin delivery systems, and you can calibrate to improve confidence.
Final thoughts
If you are a person with type 1 diabetes, I hope this helps guide your future decisions – knowledge is power.
If you are a health care professional, I hope you have better questions to ask the CGM companies when they come calling.
If you are a CGM marketer, I hope you start putting the full 15/15 accuracy data on your glossy brochures – we are not stupid, I promise.
If you are Abbott, I hope you allow optional calibrations soon for the Libre products.
If you are a company with data on file (Glucomen Day and Enlite 4) or have a CE Mark way beyond your accuracy data (Medtrum and GluoRx Aidex)I hope you will rectify this.
If you are a regulator for CE Mark for CGM devices, please implement criteria, and standards and publish product performance against them.
If you are a local commissioner or in the re-reimbursement team for health insurance, I hope you understand accuracy and functionality must have parity before considering the cost.
If you work for the MHRA, please step in before profiteering companies flood the market with unstudied CGM devices, leading to a price war where people with diabetes are the victims.
Recap, If I was the regulator for the MHRA, the minimum standards I would enforce would be:
Accuracy study requirements:
- All studies must be published in a peer-reviewed journal before applying for approval.
- The study must report type 1 and type 2 diabetes data separately.
- The study must report under 18’s and 18 and over separately.
- The study must include activities that make the glucose go up and down – eating, hypos, activity.
- Accuracy must be reported in the hypo (<3.9 or <70mg/dL), target (3.9-10.0mmol/L or 70-180mg/dL) and high (>10.0mmol/L or >180mg/dL) ranges for 15/15, 20/20, 30/30 and 40/40.
- The study must report the percentage of sensors that last the full life.
Standards for approval:
- At least 5,000 (minimum 10% <3.9 or <70mg/dL) paired readings with YSI in each population approval is requested for (Type 1<18yrs, Type 1 >18yrs, Type 2 <18yrs, type 2 over 18yrs).
- Hypo range (<3.9 or <70mg/dL): 75% withing 0.8mmol/L or 15mg/dL, , 80% within 1.1mmol/L or 20mg/dL, 90% withing 1.7mmol/L or 30mg/dL, 95% within 2.2mmol/L or 40mg/dL.
- Target range (3.9-10.0mmol/L or 70-180mg/dL): 75% within 15%, 85% within 20%, 95% within 30%, 99% within 40%.
- High range (>10.0mmol/L or >180mg/dL): 80% within 15%, 85% within 20%, 95% within 30%, 99% within 40%.
- At least 75% of sensors last the full life.
Reporting of approval:
- For each approval population (Type 1<18, Type 1 >18, Type 2 <18, type 2 over 18), a document is produced showing the percentages of all accuracy requirements and citing the research papers, to allow the public to scrutiny.
Skin in the game?
I am going to contact the MHRA – the UK Medicines & Healthcare products Regulation Agency [email protected] – To inform them of this issue and point them to this blog – https://theglucoseneverlies.com/whats-the-most-accurate-cgm-device-for-type-1-diabetes/
Also, to network with influential people in the UK diabetes space to raise awareness.
Getting your skin in the game?
If you live in the UK email the MHRA with your concerns: [email protected] – point them to this blog – https://theglucoseneverlies.com/whats-the-most-accurate-cgm-device-for-type-1-diabetes/
If you are using a product that has no peer-reviewed accuracy data or sub-standard accuracy data, and its caused you diabetes-related issues, please report via the yellow card system if you live in the UK – this is a very powerful tool
If outside the UK, report to the equivalent of the MHRA in your country.
Finally, raise awareness with health care professionals, patient groups and on your social networks.
You could share this article, or even better, share your experience if you have suffered at the hands-on an untested device.
If you enjoyed this blog, please share it and feel free to mention me, @mm640G
For other great reads on this topic please see:
- Tim Street’s Diabettech site, specifically one article on accuracy and one on a new product called GluocRX Aidex.
- A research paper that discusses criteria for how to select a CGM device from 2020.
- An excellent review on CGM accuracy by Timothy Bailey and Shridhara Alva
John Pemberton RD
Type 1D since 2008
Paediatric Diabetes Dietitian
Father of two kids at high risk of type 1D
If you want to subscribe to more nerdy blogs or contact me, please do so here.
If you want to check out the CGMs:
The full Accuracy data table as promised
| Sensor & Device | Age | n= (% type 1) | Paired (YSI or BLOOD GLUCOSE) | MARD | CEG Zone A | 2.2-22.2 mmol/L or 40-400 mg/dL 15/15 | Hypo 15mg or 0.8 mmol | Target 15% | High 15% | 2.2-22.2 mmol/L or 40-400 mg/dL 20/20 | Hypo 20mg or 1.1 mmol | Target 20% | High 20% | Cal & Conf (y/n) | Senor life (% last full life) | FDA & CE | Hypo alerts |
| Dexcom G6 (1,11) | 2-70 | 324 (99%) | 25,101 (YSI) | 9.8% | 90% | 81% | 87% | 76% | 86% | 92% | 92% | 88% | 95% | 0 (n) | 10 days (94%) | 2yrs | 4 |
| Dexcom G6 (1) | 6-17 | 165 (100%) | 5,722 (YSI) | 9.6% | 91% | 82% | 76% | 80% | 87% | 92% | 82% | 91% | 95% | 0 (n) | 10 days (77%) | 2yrs | 4 |
| Dexcom G6 (1) | 2-5 | 8 (100%) | 82 (BLOOD GLUCOSE) | 9.9% | * | * | * | * | * | 93% | * | * | * | 0 (n) | 10 days (77%) | 2yrs | 4 |
| Dexcom G7 Arm (2) | 18-78 | 308 (81%) | 40,000 (YSI) | 8.2% | * | 90% | 90%† | 86% | 93% | 95% | 94%† | 93% | 97% | 0 (n) | 10 days (*) | 2yrs | 4 |
| Dexcom G7 Abdomen (2) | 18-78 | 311 (81%) | 39,000 (YSI) | 9.1% | * | 86% | 85%† | 83% | 88% | 93% | 91%† | 92% | 95% | 0 (n) | 10 days (*) | 2yrs | 4 |
| Dexcom G7 Arm (3) | 7-17 | 122 (100%) | 8,068 (YSI) | 8.1% | * | 89% | 88%† | 87% | 92% | 95% | 93%† | 94% | 98% | 0 (n) | 10 days (*) | 2yrs | 4 |
| Dexcom G7 Abdomen (3) | 7-17 | 118 (100%) | 7369 (YSI) | 9.0% | * | 86% | 76%† | 84% | 93% | 93% | 85%† | 91% | 98% | 0 (n) | 10 days (*) | 2yrs | 4 |
| Dexcom G7 (3) | 2-6 | 28 (100%) | (BLOOD GLUCOSE) | 9.3% | * | 83% | * | * | * | 92% | * | * | * | * | 10 days (*) | 2yrs | 4 |
| Enlite 3 with Guardian Connect (4) | 14-75 | 88 (70%) | 11,664 (YSI) | 9.6% | 90% | 81% | 72%† | 80% | 87% | 89% | 82%† | 86% | 94% | 3-4(y) | 7 days (72%) | 2yrs | 3 |
| Enlite 4 Data on file | 17-80 | 150 (?) | 39,025 (YSI) | 10.6% | * | * | * | * | * | 88.8% | * | * | * | * | 7 days (*) | 2yrs | 3 |
| Enlite 4 Data on file | 7-17 | 108 | 15,326 | 12.0% | * | * | * | * | * | 83.5% | * | * | * | * | 7 days (*) | 2yrs | 3 |
| Enlite 4 Data on file | 2-6 | * | * | * | * | * | * | * | * | * | * | * | * | * | * | * | 3 |
| Eversense (5) | 18+ | 35 (71%) | 15,170 (YSI) | 9.6% | 91% | 85% | 90% | 81% | 88% | 93% | 95% | 90% | 96% | 2 (y) | 180 days (*) | 18yrs | 3 |
| Glucomen (6) | Adult | 8 | 450 (YSI) | 9.7% | 89% | * | * | * | * | * | * | * | * | 1 (y) | 14 days | 6yrs | |
| GlucoRx Adiex (13) | 18-76 | 114 (11%) | 14,586 | 8.1% | 95% | 86% | 92% (only 2% of YSI readings) | 86% | 86% | 96% | 98% (only 2% of YSI readings) | 96% | 96% | 0 (n) | 14 days (95%) | CE 14yrs | 3 |
| Libre 1-3 (12) | 18-70 | 146 (100%) | 18,926 (YSI) | 9.2% | 93% | 86% | 94% | 77% | 90% | 93% | * | * | * | 0 (n) | 14 days (71%) | 4yrs | 1 |
| Libre 1-3 (12) | 4-17 | 129 (100%) | 6,548 (YSI) | 9.7% | 92% | 86% | 96% | 74% | 89% | 93% | * | * | * | 0 (n) | 14 days (78%) | 4yrs | 1 |
| Medtrum (8) | 24-70 | 63 (16%) | 1,678 (YSI) | 9.1% | 90% | 82% | 36% | 78% | 86% | 91% | 72% | 88% | 98% | 3-4(y) | 7 days (*) | CE 4yrs | 3 |
* = Data not published, †= <4.2 mmol/L not <3.9 mmol/L as reported in IFU, ‡ = ≤4.4 mmol/L not <3.9mmol/L as reported in IFU, MARD = Mean Absolute Relative Difference, CEG = Clarke Error Grid, Cal & Conf = Calibrations and confirmatory blood glucose check required, CE = Conformitè Europëenne Mark and indication for approval, FDA = Food & Drug Administration indication for Approval, YSI = Yellow Spring Instrument venous glucose, BLOOD GLUCOSE = Whole Capillary Blood glucose, TBR = Time below range <3.9 mmol/L, TIR = Time in range 3.9-10.0 mmol/L, TAR = Time above range >10.0 mmol/L
References
1. Shah VN, Laffel LM, Wadwa RP, Garg SK. Performance of a factory-calibrated real-time continuous glucose monitoring system utilizing an automated sensor applicator. Diabetes Technol Ther. 2018;20(6):428–33.
2. Garg SK, Kipnes M, Castorino K, Bailey TS, Akturk HK, Welsh JB, et al. Accuracy and Safety of Dexcom G7 Continuous Glucose Monitoring in Adults with Diabetes. Diabetes Technol Ther. 2022 Feb 21;
3. Laffel LM, Bailey TS, Christiansen MP, Reid JL, Beck SE. Accuracy of a Seventh-Generation Continuous Glucose Monitoring System in Children and Adolescents With Type 1 Diabetes: https://doi.org/101177/19322968221091816. 2022 Apr 25;193229682210918.
4. Christiansen MP, Garg SK, Brazg R, Bode BW, Bailey TS, Slover RH, et al. Accuracy of a Fourth-Generation Subcutaneous Continuous Glucose Sensor. Diabetes Technol Ther. 2017;19(8):446–56.
5. Christiansen MP, Klaff LJ, Bailey TS, Brazg R, Carlson G, Tweden KS. A Prospective Multicenter Evaluation of the Accuracy and Safety of an Implanted Continuous Glucose Sensor: The PRECISION Study. Diabetes Technol Ther. 2019;21(5):231–7.
6. Hochfellner DA, Mader JK, Simic A, Taucher MT, Sailer LS, Kopanz J, et al. Accuracy Assessment of the GlucoMen ® Day CGM System in Individuals with Type 1 Diabetes: A Pilot Study. Biosensors. 2022 Feb 1;12(2).
7. Alva S, Bailey T, Brazg R, Budiman ES, Castorino K, Christiansen MP, et al. Accuracy of a 14-Day Factory-Calibrated Continuous Glucose Monitoring System With Advanced Algorithm in Pediatric and Adult Population With Diabetes. J Diabetes Sci Technol. 2020;
8. Zhou J, Zhang S, Li L, Wang Y, Lu W, Sheng C, et al. Performance of a new real-time continuous glucose monitoring system: A multicenter pilot study. J Diabetes Investig. 2018;9(2):286–93.
9. Freckmann G, Pleus S, Grady M, Setford S, Levy B. Measures of Accuracy for Continuous Glucose Monitoring and Blood glucose Monitoring Devices. J Diabetes Sci Technol. 2019;13(3):575–83.
10. Bailey TS, Alva S. Landscape of continuous glucose monitoring (CGM) and integrated CGM: Accuracy considerations. Diabetes Technol Ther. 2021 Sep 1;23(S3):S5–11.
11. Welsh JB, Gao P, Derdzinski M, Puhr S, Johnson TK, Walker TC, et al. Accuracy, utilization, and effectiveness comparisons of different continuous glucose monitoring systems. Diabetes Technol Ther. 2019;21(3):128–32.
12. Seibold A, Alva S, Feuchter L, Lazarus M, Liu H, Nada M. Performance of Freestyle Libre® 2 System in Adult and Pediatric Populations. Vol. 14, Diabetes – Nicht nur eine Typ-Frage – www.diabeteskongress.de. 2019.
13. Ji L, Guo L, Zhang J, Li Y, Chen Z. Multicenter Evaluation Study Comparing a New Factory-Calibrated Real-Time Continuous Glucose Monitoring System to Existing Flash Glucose Monitoring System. J Diabetes Sci Technol. 2021 Aug 11;19322968211037990.