Speaker 1 (00:09.944)
Welcome to the Glucose Never Lies podcast where science meets real life experience to empower diabetes management. I'm John Pemberton. I've lived with type 1 diabetes since 2008 and have spent nearly 20 years mastering both the science and art of managing it. Through personal experimentation, published research and my work as a diabetes specialist dietician, I've gained deep insights into what truly makes a difference. When my son Jude tested positive for type 1 diabetes antibodies,

I realised that all the knowledge in my head was wasted if I couldn't communicate it in a way that was clear, actionable and easy to come back to. So I built the Glucose Nevelise Education Programme, a free online resource designed to teach people diabetes management exactly the way I'd want people to understand it if they were looking after my son. After battling a functional motor disorder for many years and recently experiencing a major depressive episode, I was eventually pulled out of that hole by my friends, family and professionals who helped me get back to being me.

That experience taught me the power of giving and this podcast is my way of giving back. My co-host Louise is a highly experienced diabetes nurse with over 20 years in the field. She brings a wealth of knowledge and her superpower is making complex diabetes science accessible and practical for everyday life. She is the best diabetes nurse I have ever worked with and there have been some good ones. Most importantly, she keeps me in check and keeps the podcast on point. So if you're living with diabetes or supporting someone who is,

We want to make things easier, clearer, and importantly, more enjoyable. We hope you enjoy the content. If you do, please share it with those who may like it too. As a disclaimer, the information shared on the Glucose Nebulize podcast is for informational and educational purposes only. While we discuss strategies and insights for diabetes management, this podcast is not a substitute for professional medical advice. Always consult your healthcare team before making any changes to your diabetes plan.

That done with, let's get into the content.

Speaker 1 (02:14.062)
cast number 10. And this is the first part of a CGM series, a three part series where we're going to take a deep dive looking into the accuracy of continuous glucose monitoring, how you actually measure that. And then once we get past all that and our understanding of which CGMs are robust for people with type one diabetes, then we're going to move on to how look at the different features and options that they offer for different users. But first of all, let's say hello back to Lou, who's now after her dissertation for AMP. How's that?

relieved that it's been handed in.

Did you go alright?

I'll let you know in two weeks.

Okay, good. So she's been working hard. And in the meantime, I've been looking into continuous glucose monitoring a little bit further with the DSN forum. And we're putting something together to, I guess, really summarize everything that's going to be done within these three sessions, and to make it more publicly available. But for today, to frame this at the beginning, this conversation really is going to be about continuous glucose monitors for insulin dosing. So there are

Speaker 1 (03:15.736)
Continuous glucose monitor that are approved for insulin dosing and those which are not. Now, people who without diabetes who want to use those CGM devices is a completely separate topic that won't be covered in here. This is all about CGMs for insulin dosing, what is called non-adjunctive therapy decision. So that's the first thing. And the second thing is you'll notice on the contact page that I put my disclosures. So I've received money from ABAT, Dexcom, Rush, Insulet and others for various talks and bits and pieces.

So you can decide whether you feel like I've got a bias towards them or not. But it's important to be upfront about that because obviously some of the sensors and CGM systems we're to be discussing are those. And those, in my opinion, are the ones that are the most accurate and have the best features. So you can either take it that I've got a bias towards those, or you can take it that I speak for those because I believe in those. It's up to you. I'm just being open and honest at the beginning.

No worries. That's loud and clear.

Okay, so just just all get on the same page. So what we're going to talk about in this particular part, first part is going to be continuous glucose monitoring. What is it? And then before we get into trying to understand the accuracy of a device, we have to understand the study design conditions for the accuracy test, because if you don't understand that, the performance that you will get told about means nothing because you need to understand was it done on the right population, i.e. people with type 1 diabetes, people of the right age group?

Was it tested robustly across the full glucose range from as low as 2.2 to as high as 22.2 and some of the specifics within there. If you don't understand that, the performance is meaningless. It means nothing to you because you have to understand that first. So we're going to spend the first part going through that. Then the second part is going to be going through how do you actually measure accuracy? What numbers should we be looking at? What do they mean? And what should I be thinking about is useful for a CGM system that's good for insulin dose in that part too.

Speaker 1 (05:11.372)
And then part three, we'll be going through, OK, we're now happy with the systems that are deemed accurate enough and seem to have good data. What do we want to consider about the size of it, the alarms that it's got, the wear that it's got? All those things should be a last consideration once you feel that it's good enough for you to use for insulin dosing decisions. So I've got Lou with me today. So basically, I've been lost in this world for the last six or seven years and will probably be trying to explain things way too complicated. So I've got Lou here is going to be

bring me back in and bring things down to a level that hopefully everyone listening can take away. I'll try and do that myself, but the reality is I probably won't achieve that. So are up for that challenge?

Yeah, absolutely.

Okay, so pretty much on the, just a basic understanding. So when you do a finger prick, you get the glucose from the blood, because the blood comes up to the surface, and you measure the glucose that's in there. And that's what you see on your finger prick glucose meter. Now with a continuous glucose monitor, it's slightly different. It usually has a sensor that is sat in the fat layer of the tissue or the interstitial space, basically the fat tissue, and it's measuring the glucose that's in there. So they're measuring two different places.

But what they can do with that is we know the glucose level that's in the blood drifts out into interstitial space with about a five minute delay. So there are algorithms built within these continuous glucose monitors that kind of do the maths to determine what the blood glucose level is, do some kind of algorithm factor and adjustments, and then displays to you what your continuous glucose monitor is. And the systems have really improved from having a lag time of almost 20 minutes. It was 20 minutes behind what was going on in the blood.

Speaker 1 (06:46.86)
down to maybe two to five minutes, which is fantastic. So it's now moving to, as we're going to describe, insulin dosing.

So just to clarify, when you're doing a finger stick, which is sometimes called a capillary blood glucose level, so that's something to be aware of. And what we're saying now is that the sensors do something in between to almost bring that more in line with what is then coming out on your CGM, which will enable it to be closer matched to whatever the capillary is. So that lag time isn't as pronounced.

Yeah, that's right. And that's really where the companies tend to come into their own. So all companies measure the glucose that's in the fat tissue, but how the quality of their measuring instruments, whether it be the sense of filament, the enzyme that they use within there, and also the quality of the algorithm that can detect changes in noise, which is basically things that might be disrupting the filament.

That's where the real rubber hits the road with determining what you see on your continuous glucose monitor, whether that is a accurate representation of what the capillary glucose actually is. And that's when we get to talk about study design and accuracy. It's so important because not all sensors are the same because they're manufactured differently. They have different algorithms. Therefore, we will find very different results even with sort of like what appears to be on the surface, very similar products.

Okay, that all makes sense.

Speaker 1 (08:14.636)
Good. OK, so just a clear distinction that some of them are adjunctive, which means you can't use for insulin dosing decisions. We're not going to discuss those today. The ones that we are going to talk about is the ones that have been approved to determine insulin doses off specifically correction doses on the higher side is the most important thing. So what I want you to do now as a sort of listener is pretend that you are the football manager of a Premier League football team and you've got money to buy a striker, the best striker you possibly can get.

And what you're told is these three options that are about to be presented to you have all scored 800 goals and therefore are potentially equivalent. And this is what happens when you work in a diabetes team or maybe you look online for some of these CGM sensors and they will present a MAD value, which is mean absolute relative difference, which is on average, how far away is the sensor reading to the capillary glucose reading? And it's kind of...

Folklore that if her center has a mod of 10 % then it's pretty good It's all good for insulin dosing decisions that would mean on average is 10 % different So to put that into context if the finger prick was saying 10 then with a mod of 10 % on average The sensor reading would be anywhere between 9 and 11 because that's 10 % different. Absolutely. That makes sense. Yeah So if you just took that at face value, it wouldn't matter which one of these strikers these front players who scored 800 goals that you selected

But now if I told you that one was Ronaldo, one was Messi, and one was me, all of a sudden you would go, hang on a minute, that doesn't sound that those 800 goals are equivalent. And that's the exact same mental approach you need to take into thinking about sense of accuracy. If someone comes and tells you, we've got a mark of 10 % before they described what the study design was, it's like me just doing that with football. The first thing you would ask is, where did you score your goals? What was the level of competition?

Was it just in the park with your kids or was it in the Premier League or international football? Those 800 goals mean something very, very different. And that's how you have to think about MARD.

Speaker 2 (10:13.742)
So it's very much thinking about what the context of where that mar does come from, isn't it? So absolutely. So what the environment was, what the conditions were.

Exactly and simple thing is you need to think about how these sensors been tested in the way that the glucose behaves in real life. So for example, we know that you sometimes eat without insulin, your glucose level shoots up very quickly and then sometimes you give a correction dose and it hasn't worked in 20 minutes. You give another and then another and then another known as a rage ball. And next thing your glucose level is coming flying down and it's going down really fast.

So we need to make sure that these sensors have been tested all the way up to 22.2 or higher, going up fast. And we also need to make sure they're dropping, being tested when they're dropping very fast, because that's what's happened in real life from time to time. Send that a few times in practice.

Yeah absolutely and I guess it's thinking about let's say you're 10 it could be 11 you could be 9 but 10 % in the really high range is quite a big difference and also has it been tested and then obviously in the low range as well which will make a massive difference to what you do.

Yeah, and the simple way of thinking about this is if you were going to buy a car and you were told that this car had only been tested driving in the middle lane at 50 miles an hour, that wouldn't give you much confidence how it's going to perform when you've got a lot of traffic and it's got a zip in between lanes. That's how we need to think about how these sensors been tested. And therefore we need to ask about study design before we ask about accuracy. the basically four really simple questions to ask that will let you know if something has been

Speaker 1 (11:49.642)
and assess properly. And the first one is in the study that was done, were there at least 70 % or 70 % or more people with type 1 diabetes in the study? Because if you have less than that, it's unlikely that the sensor will have been tested in the very high and low ranges because people with type 2 diabetes generally don't go as high as people with type 1 and certainly don't go up there at the same speeds as people with type 1 because they have their own insulin production. And similarly, because they retain their own glucagon production, they don't drop as fast and they don't go as low.

So the more people with type 2 diabetes you have in a study, it's basically going to mean that the glucose levels haven't been tested in the full range and not at the right speeds. And it will make the accuracy look a lot more accurate than you will actually get within real life. So that's one simple question that first of all will help identify has the study, has the thing been tested robustly or not.

And guess the black and white of that is if it's done with people with type 1 diabetes where there's an absolute insulin deficiency, then you know the conditions are similar to what you want to apply in your life.

Absolutely. So the second thing is, did they do meal and insulin challenges within this study? So the meal challenge is just described, have a meal of about 60 grams of carbs, no insulin, make the glucose level rise very, very rapidly. And then once it gets high, give a big whacking dose of insulin and make it come down fast. Now, obviously this is done in laboratory and safe conditions, but you need to test a sensor like this, not because that's how the glucose behaves all the time.

but it can do occasionally. And those are the times when it's very high, when you do what? You give a correction dose. So you want it to be accurate when it's up there. That's when you're making the decisions when it's moving fast, so it needs to be tested. So the second question is, did you do meal and insulin challenges? And if the answer is yes, then we know it's a good chance it's been tested. And then the third one is when you have...

Speaker 1 (13:41.096)
and the sensor readings, they compare them to what we call a comparator or a reference measure, which is a measure of either capillary or venous glucose, which is assessed by a lab grade analyzer. And what you want is when you stick all the readings together, what we call paired reading, each sensor reading to each sort of reference measure that they've taken, and it's usually every 15 minutes in a study. And what they do is then they say, of all the readings, what percentage of those readings were low? Less than 4.4.

Now that number should be 8 % or more by international guidelines. That means it's been enough readings in the low range to determine whether it's accurate. Very similarly on a very high range, how many, what percentage of readings were above 16.7? There should be 5 % or more. Now, if you've got 70 % of people who type one and you've done meal and insulin challenges, that is easy to do. If you've got loads of people type two diabetes, you didn't do meal and insulin challenges, you're not going to have those numbers and you're going to make the sensor, we're not going to know how it performs like in those arenas.

And also you're going to get an accuracy which looks more accurate than it will actually perform in those situations. So those four key questions, percentage of people type one, 70 % or more. Meal and insulin challenges, yes or no. 8 % or more readings in the less than 4.4 range or 80 milligrams per deciliter or 5 % or more readings higher than 16.7, which is 300 milligrams per deciliter. If you get four ticks, we know then that the accuracy that is presented

you will know that it's kind of done in a way that will represent what you'll potentially get in real life. And if the answer is no to those, it doesn't mean that the sensor is inaccurate. It just means we don't know how it's going to perform. So it depends on how comfortable you are at taking risk. Personally, for me, when I'm doing instant decisions, I don't want to be taking too many risks. I'd rather use a sensor that I understand has been tested robustly and I know what the risk is going to be. I'm not that keen.

on me and the people who I support on using a system that hasn't been tested robustly. I don't know whether it will be accurate or not and I'm not sure that that's the right approach to take.

Speaker 2 (15:45.804)
Yeah, that makes perfect sense to me. So it's about ensuring that the sensor that you're potentially going to use has been tested in all of those conditions. So that obviously you're not taking those risks in your day to day life.

And in the show notes, I've put a little snippet of what I've done with the diabetes specialist nurse forum. And what we've done is put all the sensors that are available currently in the UK with brackets of them into non-adjunctive use, those for insulin making decisions and those non-adjunctive and adjunctive use those for not insulin decisions and basically scored them against these four criteria to get, sorry, the final one, which I didn't forgot to mention, which is really important is the data that you're showed, has it been peer reviewed? I, has it been published in a journal?

Or has it been reviewed by the FDA in America because they have a robust process for assessing? So if the answer is yes to those two, then you know the data has been thoroughly looked at. Sometimes when you have data on file or has only been presented as an abstract at a conference, you can't be as convinced that it's gone through the same rigorous process of data interrogation by people who are independent of that particular data. So what we've done...

It's put together a list of those their score out of five. And I'm not into naming and shaming different companies apart from it's there in the list. But for this perspective, all I want to tell you is the ones that have got a score of four and above that are there for insulin dose decisions. And that's the AcuCheck Smart Guide, all of the Dexcom products, Dexcom G6, G7, One, One Plus, the Freestyle Libre 2 Plus, the Freestyle Libre 3 Plus, the Simplera Medtronic and the Guardian 4.

All the other ones either have a score of less than four, either one or zero actually. And then obviously the adjunctive ones, which we're not really talking about again, have got pretty much a low score as well. So again, I'm not saying that those are not accurate devices. I'm saying with the current available data that we have, we don't know. The ones that I described formally, we can be confident they've been tested robustly. So they're the only ones that we're going to take forward.

Speaker 1 (17:53.016)
when we start to discuss accuracy because we know that the accuracy is meaningful in the context of insulin dose and decisions. And I guess what's also worth considering is now that continuous glucose monitors are linking with pumps to become hybrid closed loop systems, you would want a sensor that is very accurate to be driving those automated decisions. And to help us with that, there's something called the FDA integrated CGM CRITE classification.

And basically in America, they've created a new category, which is assessing sensors in a very robust way in terms of performance and study design. And if you have that marking, it then allows that sensor to be used for multiple sensors, sorry, for multiple hybrid closed loop systems. So the only ones available with that at the moment are the Dexcom G6, the Dexcom G7, the Freestyle Libre 2 Plus and the Freestyle Libre 3 Plus. Now, importantly, the Medtronic, Simplera and Guardian 4

are very accurate sensors, then they have American FDA Class 3 approval for the 780G. So that is clearly accurate and the data on that is very accurate overall. But there is one option that's available in the NHS technology appraisal and supply chain that doesn't have a sensor that either meets the FDA ICGIM criteria and actually on the scoring chart, you can see that that has a score of one out of five.

So I'm not going to suggest and again, name and shame. It could be that that is a very accurate device, but we don't have real data, certainly from the age of two on the sensor accuracy. And we also currently don't have any randomized control trial data or any effective real world data on the closed loop performance. And that has led to pediatric societies and adult specialist technology networks like the DCN to suggest that the moment let's put caution in place until we see further data. You can see the show notes for that. But again, from a

open and honest point of view, it's important to understand what the data is before we start thinking about what these things can do in real life. Don't you want to comment on that? I'll make any questions on that.

Speaker 2 (19:57.196)
think that's pretty clear to me. don't think you've, you haven't confused me. So, so, so that's good.

So ultimately now what we've got a situation is we understand that if it has been tested robustly with a study design score in our simple terms of four and above, we know the accuracy performance that we're going to look at is going to be robust and will likely represent real life performance. And now we're sort of going to move on and introduce the concept of accuracy now and then we're going to pause and we'll do that in part two. So just to put a primer for that.

When you're measuring accuracy, we need to do a solid study design just as we described. But then as I mentioned, you've got the CGM readings and then you've got a comparator or a reference. Now you can either do that for a capillary measure or you can do that from a venous measure. And there's a slight difference because the venous glucose is on average is about five to 10 % lower than the capillary glucose. And the reason for that is when we eat, the food goes into our belly, goes into our small intestine and into our blood.

And then it gets pumped out via the heart and it goes through the arteries to the arterioles to the capillaries. So therefore it's its highest concentration there. It gets used generally in the capillaries. So then the glucose concentration drops. So therefore when it then comes out into the veins and to go back to the heart, the concentration is lower than the capillary. And that is especially evident after eating. So the problem with using Venus as a comparator is it's always five to 10 % lower. And especially after meals, it's quite a bit lower.

which means then you're not actually measuring what your body is exposed to. And we know that it's the glucose that actually drives the microvascular problems with your eyes, your kidneys, et cetera. So if we kind of map the accuracy to venous glucose, we're potentially under-representing what's actually going on in the body and what the cells are exposed to. So in an ideal world, seeing how accurate these sensors are to capillary makes a lot more sense than venous glucose. Does that make sense?

Speaker 2 (21:58.7)
Yeah, so basically what you would say is that if we're looking at these studies and we're looking at comparators, it's preferable that the comparison has been done with the capillary blood glucose level rather than a venous blood glucose level.

That's right. But the downside is, as we're going to get into in the accuracy, is up till now, the vast majority have measured against Venus because that's what the regulatory bodies have required. But I sit on the IFCC working group for CGM, and there's a big push for an international standard and to move it to capillary just for the reasons that I've described. But at the moment, when we talk about accuracy, we're probably going to talk most of the comparators against

venous because that's what's been done although the latest data does show against capillary and we will discuss that but again it's just a little note of importance because if your sensor tracks very well to capillary glucose you're getting a good idea of what's actually going on in your body and you can make better decisions.

Yeah, and I guess so if we're teaching, I guess we teach families, the only way that they would be able to compare accuracy is with capillaries in it as well. So it's

Yeah, exactly. I mean, I don't know. haven't seen many people knocking about with an indwelling catheter in their arms. That's what I'm most... Exactly. again, if If your sensors aligned to your venous glucose, then when you do a capillary prick and it's a bit higher, you will think that it's wrong, but it's not that it's wrong. It's just actually supposed... It's aiming at the right place because that's what it's been measured against, really it should be measured against capillary. So I'm pretty confident that over the next five, five or six or seven or eight years,

Speaker 2 (23:12.942)
So it's sort of all...

Speaker 1 (23:37.612)
the CGM is going to be aligning more closely with CGM because that's where the international standard will go. And when we come to the accuracy shortly, you'll be really surprised at what a difference it makes with a CGM system that is aligned with Venus versus one that's aligned to capillary. Because then, for example, you will have two sensors where you've got 70 % timing range on both of them. So you as the user think fantastic and bang in the zone. then when you hit you and C comes back.

your HbA1c is half a percent lower for one of the sensors that is aligned to capillary glucose and half a percent higher when it's aligned to venus. So you as the user are then disappointed because you think you put the same and quite rightly have put the same effort in as the other person. Yet your HbA1c, the thing that's dictating your long term complication risk is going to be different. So having some standardization where the sensors align to capillary glucose that's measured effectively.

is essential as this space moves forward. And we're certainly not there yet. And once there is an international standard, a podcast like this will be completely defunct because if you have the ISO standard, it's like getting a stamp on your forehead, bang, we don't need to know about all this. But at the time being, whether you're a user, whether you're a healthcare professional, and even a manufacturer, you need to kind of understand this because we haven't got standardization yet, although it is coming on its way. That make sense? Good work. So...

We'll pause there before we get into how to measure accuracy, what different accuracy measures are, but just to summarize what we discussed. If someone tries to give you the accuracy of a performance before they tell you about a study design, you pause them and say, I'm afraid study design first, accuracy second. What percentage of people do you have with type 1 diabetes? 70 % or more or not? Tick yes or no. Did you use meal and insulin challenges? Tick yes or no. Did you have 8 % of readings less than 4.4? Yes or no. Do you have

5 % or more above 16.7 yes or no. And then finally, do you have a publication that you can point me towards so I know that that's been evaluated robustly? If you get five yeses, you can then absolutely look at the accuracy data with confidence that it's been tested across the full glucose range in real world conditions. If you're only getting a score of zero or one, again, it doesn't mean that the sensor isn't inaccurate and not helpful. It just means we don't really know what the performance is like. Therefore, the risk level goes up.

Speaker 1 (26:02.796)
And it depends on your appetite for risk. And certainly when we're talking about sensors that are driving automated insulin delivery systems on hybrid closed loop, I'm personally not in the business of taking risks with my own life. And I certainly wouldn't be doing that with the people that we support.

think that's really, really nice and clear. I guess the question for me would be that you would, if you were say a patient with type 1 diabetes, you would want your healthcare professional to sort of have this knowledge before advocating one system over another, wouldn't you? And I guess that would be a take home message, wouldn't

Yeah, I mean, obviously, if you've listened to this and go, oh, God, absolutely no way in the world that I'm going to answer all those questions. And I don't know. We've obviously done the work for you. When I say we, me and the DSM forum. So very soon over the next couple of weeks or so, there will be a chart downloadable. I'll put it on my site as well. But there'll be a chart downloadable from them that will be updated on a monthly basis, which will answer ask those five questions for you and give them a study design score out of five. And then it will actually tell you what the performance is in the metrics that we're going to discuss. And then

from that the systems that are available on prescription or on an HS supply chain what the features are. So everything will be covered in there. But if you've got this baseline understanding and understand it right now, you'll be able to go to that chart very easily and kind of depict what you think is important and how to approach this this CGM situation.

Good work.

Speaker 1 (27:26.816)
Okay, so we'll see you in part two for we'll get into the actual accuracy and we'll see whether someone's 800 goals is actually equivalent to someone else's 800 goals. So we'll see you in the next session.