Libre accuracy and predictive ability

[bofhvp]

I have been using it for the past 4 weeks and I have to say that I am NOT impressed. Its always saying my blood sugar level is way what my finger prick say. For exam, even though its scanning ever time and I can see the realtime graph. I would still manually scan and it would say its 9.7 However, my meter would give a totally different reading of 5.2
This has happened several time to the point am now asking how accurate IS Freestyle Libre ?
But I do love it for the graphs it give, but certainly NOT for the readings !

How have everyone else compared the Freestyle Libre reading to that of finger prick meter ?

(mod edit to say that this post and the following ones have been moved from https://www.diabetes.co.uk/forum/threads/has-anyone-quit-using-freestyle-libre.202578/page-2 so that posters can have a detailed discussion of libre accuracy without derailing that thread. )

 

[Jasmin2000]

I have been using it for the past 4 weeks and I have to say that I am NOT impressed. Its always saying my blood sugar level is way what my finger prick say. For exam, even though its scanning ever time and I can see the realtime graph. I would still manually scan and it would say its 9.7 However, my meter would give a totally different reading of 5.2
This has happened several time to the point am now asking how accurate IS Freestyle Libre ?
But I do love it for the graphs it give, but certainly NOT for the readings !

How have everyone else compared the Freestyle Libre reading to that of finger prick meter ?

I compare these all the time. Some things to be aware of:

- the realtime graph as you call it is fitted to the sensor readings, but is often corrected if there are outliers, e.g. high readings that don't fit on the curve. So when you see something oddly high, give the sensor 5-10 mins to correct for it and mostly the odd reading will disappear (unless you've scanned it at that time and then you'll see it hanging in mid-air and not on the curve - see my 9.5 scan below).

- the sensor is not measuring blood glucose but interstitial fluid glucose (IFG). Glucose moves from blood to ISF over 5-20 mins so your BG will be reflected in the sensor reading after some time.

Could be that you had a BG of 5.2 but at the same time the sensor recorded an outlier at 9.7.

 

[In Response]

the sensor is not measuring blood glucose but interstitial fluid glucose (IFG). Glucose moves from blood to ISF over 5-20 mins so your BG will be reflected in the sensor reading after some time.

The algorithm Libre uses to convert from Interstitial Fluid Readings (ISR) to BG, takes the lag into consideration by extrapolating the current trend. Therefore, the lag rarely impacts the readings on the Libre unless the current trend has changed in the last 15 minutes. This is why you may see “over shoots” as you illustrated - the algorithm predicted a higher reading as it was unaware your BG had started to fall in the last 15 minutes. As you see, it will correct this reading when it realises the trend change.
This is more likely than an “outlier” as the algorithm also smooths the readings to remove the impact of noise.

 

[Jasmin2000]

The algorithm Libre uses to convert from Interstitial Fluid Readings (ISR) to BG, takes the lag into consideration by extrapolating the current trend.

No, this is incorrect. The algorithm does not convert IFR to BG - the notation on the LibreLink calls it 'sensor glucose' and this is IFG.
Nor does it take the lag into consideration, as the curves are not shifted to an earlier time point - how would it even know how far ahead the BG is? My BG curves are always ahead of the final corrected sensor curves - sometimes a few minutes, sometimes 10-15 minutes.

Therefore, the lag rarely impacts the readings on the Libre unless the current trend has changed in the last 15 minutes.

The lag significantly impacts your readings if you're monitoring BG/IFG during intense exercise when BG is some time ahead of IFG.

This is why you may see “over shoots” as you illustrated - the algorithm predicted a higher reading as it was unaware your BG had started to fall in the last 15 minutes. As you see, it will correct this reading when it realises the trend change.
This is more likely than an “outlier” as the algorithm also smooths the readings to remove the impact of noise.

The overshoot is nothing to do with the lag - it is because IFG is not homogenous and can give odd local readings which disappear on curve fitting. The LIbre does not 'predict' readings but it corrects real local anomalies.

 

[Jaylee]

I compare these all the time. Some things to be aware of:

- the realtime graph as you call it is fitted to the sensor readings, but is often corrected if there are outliers, e.g. high readings that don't fit on the curve. So when you see something oddly high, give the sensor 5-10 mins to correct for it and mostly the odd reading will disappear (unless you've scanned it at that time and then you'll see it hanging in mid-air and not on the curve - see my 9.5 scan below).

- the sensor is not measuring blood glucose but interstitial fluid glucose (IFG). Glucose moves from blood to ISF over 5-20 mins so your BG will be reflected in the sensor reading after some time.

Could be that you had a BG of 5.2 but at the same time the sensor recorded an outlier at 9.7.

View attachment 66425

Hi, I’m loving your explanation. Does this anlso explain the odd scan error notification that can crop up sometimes regarding “try again in 10 minutes?”

 

[Ashintheuk]

Hi, I’m loving your explanation. Does this anlso explain the odd scan error notification that can crop up sometimes regarding “try again in 10 minutes?”

I attended a day last year near Coventry by the charity Diabetes Research & Wellness Foundation. Prof Parth Kar and Dr Iain Cranston gave a talk, and asserted that the testing strips aren't that reliable due to contamination of the sample, and that they age, whilst the sensor wires self clean contiuously

 

[SimonP78]

No, this is incorrect. The algorithm does not convert IFR to BG - the notation on the LibreLink calls it 'sensor glucose' and this is IFG.
Nor does it take the lag into consideration, as the curves are not shifted to an earlier time point - how would it even know how far ahead the BG is? My BG curves are always ahead of the final corrected sensor curves - sometimes a few minutes, sometimes 10-15 minutes.

I don't use the libre app (typo edited!) so I've no idea what it does, though I do hear people mention that it appears to try to predict what your current BG is based on the ISF trend and will sometimes change its mind when presented with new ISF data and changes the curve (i.e. historically). I'd be interested to see some data demonstrating this in practice (i.e. whether or not it is displaying a past ISF value or a predicted current "BG" value, and whether it goes back and changes the past curve based on current ISF value if it does a refit) and/or that it doesn't happen - do you have anything to hand?

I also assumed that ISF vs BG response rate changes (though I've never really dug into how predictable this is), however, in XDrip+, which is what I do use, there is always a constant time offset offset (and XDrip+ works in the past, in terms of current ISF value, so doesn't try to guess what your current BG might be and tell you want it thinks you are now, which is very odd when you first start using it, though becomes quite normal once you're used to it.) This constant lag seems to work pretty well, but again I do wonder what the typical ISF-BG time lags are and under what circumstances. Again, if you have data/pointers to papers I'd be interested to have a read.

 

[Jaylee]

I attended a day last year near Coventry by the charity Diabetes Research & Wellness Foundation. Prof Parth Kar and Dr Iain Cranston gave a talk, and asserted that the testing strips aren't that reliable due to contamination of the sample, and that they age, whilst the sensor wires self clean contiuously

Now oddly, I’ve had quite a few chances to check my Accu-chek mobile as well as the libre against NHS nurse issue meters when they use them on me. The last time was late last year whilst in A & E. (3 busted ribs & suspected internal injury.)
I’ve also heard this said before from Prof Kar.

I don’t subscribe to what he says on this. (Regarding the meters.) Regarding the probe on the sensor after a length of time could also get “crudded up”? A little like the body’s reaction to some piercings.. (possibly why there is a 14 day lifespan on the sensor.)

Just my take on it….

 

[Ashintheuk]

Now oddly, I’ve had quite a few chances to check my Accu-chek mobile as well as the libre against NHS nurse issue meters when they use them on me. The last time was late last year whilst in A & E. (3 busted ribs & suspected internal injury.)
I’ve also heard this said before from Prof Kar.

I don’t subscribe to what he says on this. (Regarding the meters.) Regarding the probe on the sensor after a length of time could also get “crudded up”? A little like the body’s reaction to some piercings.. (possibly why there is a 14 day lifespan on the sensor.)

Just my take on it….

He said that a chemical reaction between the sensor and the interstitial fluids keep it clean constantly

 

[Jaylee]

He said that a chemical reaction between the sensor and the interstitial fluids keep it clean constantly

Here's how they reckon it works. https://notjustapatch.com/how-does-freestyle-libre-work/

And a quote from it in layman's terms.

"This is the fun part. The filament contains an enzyme called glucose oxidase, this enzyme reacts with the interstitial fluid, converts glucose to hydrogen peroxide. A current is then generated which is electronically transmitted to your reader or smartphone app."

 

[CheeseSeaker]

I don't use the libre so I've no idea what it does, though I do hear people mention that it appears to try to predict what your current BG is based on the ISF trend and will sometimes change its mind when presented with new ISF data and changes the curve (i.e. historically).

Libre Itself is simply a CGM - it monitors interstitial fluid for current BG readings (usually 15-20 mins behind an actual BG reading), it doesn't know about ISF or insulin on board etc - just gives a BG reading.

I'd be interested to see some data demonstrating this in practice (i.e. whether or not it is displaying a past ISF value or a predicted current "BG" value, and whether it goes back and changes the past curve based on current ISF value if it does a refit) and/or that it doesn't happen - do you have anything to hand?

You'd need soemthing else to do this - xDrip doesn't do that (as far as I know) yes it works in the past - that's down to the readings from the CGM though - all CGMs work in the same way, so delay between fluid, and BG test

I also assumed that ISF vs BG response rate changes (though I've never really dug into how predictable this is),

Have a look at DynamicISF in AAPS - this alters ISF trends based on data from the last 5 day (5 days minimum) - collected from Nightscout

DynamicISF (DynISF) — AndroidAPS 3.2 documentation

androidaps.readthedocs.io

 

[Jasmin2000]

Hi, I’m loving your explanation. Does this anlso explain the odd scan error notification that can crop up sometimes regarding “try again in 10 minutes?”

That error notification is due to a technical issue with the sensor and I've heard that it can be due to the sensor being too cold, too hot or if ISF flow is reduced.

 

[In Response]

I find it interesting that some people report that their Libre is always higher than finger pricks and some people report the opposite. My experience is that the difference is random - sometimes higher and, just as often, the Libre is lower.
Over the years, we have come to assume accuracy with our finger prick meters but I wonder whether those who say that Libre always report high have finger prick meters that are closer to the lower threshold of accuracy and the Libre accuracy varies by sensor. Or whether it is something to do with the relationship between ISF and BG differs for different people.

 

[Jasmin2000]

I find it interesting that some people report that their Libre is always higher than finger pricks and some people report the opposite. My experience is that the difference is random - sometimes higher and, just as often, the Libre is lower.
Over the years, we have come to assume accuracy with our finger prick meters but I wonder whether those who say that Libre always report high have finger prick meters that are closer to the lower threshold of accuracy and the Libre accuracy varies by sensor.

Accuracy is not the question - I make sure my Glucometer is reading accurately using control solutions.

Or whether it is something to do with the relationship between ISF and BG differs for different people.

Bullseye! Some folk don't realise that there's a lag of 5-25 minutes between BG and it's appearance on the sensor IFG reading. So when they are reporting a higher level on BG it may be because it hasn't yet appeared on the sensor. This is different for different people and for different causes of BG changes (food, insulin, exercise, counter-regulatory hormones, etc).

 

[In Response]

Bullseye! Some folk don't realise that there's a lag of 5-25 minutes between BG and it's appearance on the sensor IFG reading. So when they are reporting a higher level on BG it may be because it hasn't yet appeared on the sensor. This is different for different people and for different causes of BG changes (food, insulin, exercise, counter-regulatory hormones, etc).

I am NOT talking about the lag. I am talking about when BG is stable but yet CGM is reporting high or low.
If it was anything to do with the lag, people would notice the same Libre (always) being higher AND lower because, just like investments, our BG "can go down as well as up"

 

[Jasmin2000]

I am NOT talking about the lag. I am talking about when BG is stable but yet CGM is reporting high or low.
If it was anything to do with the lag, people would notice the same Libre (always) being higher AND lower because, just like investments, our BG "can go down as well as up"

Yes, if it was anything to do with the lag BG and IFG would be the same but the latter would be delayed.
So what else is there that can give consistently higher readings on one but not the other?

IFG - badly calibrated or dysfunctional sensors, first day of sensor often not consistent, bad sensor placement, local IFG changes in sensor area, such as if you had your sensor on your arm and did a bunch of pushups.
BG - badly calibrated Glucometers, old or damp test strips, unwashed fingers with sugar on them.

There's also the CGM algorithm that assumes the ratio of blood volume/interstitial fluid volume is the same for everyone, but it isn't always and for some folk the algorithm consistently reads a bit higher than BG. However, the error is within the +/- 15% that we expect from such devices.

 

[Jasmin2000]

I don't use the libre so I've no idea what it does, though I do hear people mention that it appears to try to predict what your current BG is based on the ISF trend and will sometimes change its mind when presented with new ISF data and changes the curve (i.e. historically). I'd be interested to see some data demonstrating this in practice (i.e. whether or not it is displaying a past ISF value or a predicted current "BG" value, and whether it goes back and changes the past curve based on current ISF value if it does a refit) and/or that it doesn't happen - do you have anything to hand?

If you look at the pic on my last post there is a scanned reading (yellow, 9.5) that is not on the curve. This point was retrospectively ignored as the algorithm fitted the curve to a bunch of points before and after the 9.5 and decided that it was anomalous. So yes, it changed it's mind.

What a CGM and its software does not do is to move the time axis so that the curve matches true BG. There is a real life lag between BG and ISF but the sensor cannot know how long this is - 5-25 mins and cannot change its own time axis.

I also assumed that ISF vs BG response rate changes (though I've never really dug into how predictable this is), however, in XDrip+, which is what I do use, there is always a constant time offset offset (and XDrip+ works in the past, in terms of current ISF value, so doesn't try to guess what your current BG might be and tell you want it thinks you are now, which is very odd when you first start using it, though becomes quite normal once you're used to it.)

Sorry I've no idea what you mean by time offset and working in the past?
Importantly, no CGM will attempt to tell you what your BG is as it has no access to blood. Medical device regulators would not allow a device that produces actionable information that does not reflect the source of the data. All CGMs report only on ISF.

This constant lag seems to work pretty well, but again I do wonder what the typical ISF-BG time lags are and under what circumstances. Again, if you have data/pointers to papers I'd be interested to have a read.

That's a good question - how long the lag? I've seen 2-45 mins depending on where you look and it can depend on where the CGM sensor is placed but mostly on the individual. The lag is due to movement of glucose from blood to ISF so if you have circulatory or cardio-vascular issues, ion imbalances, an infection or other illnesse. the lag may change.

What is always the case is that lag time is minimal when glucose levels are relatively consistent. It's when they change over a short period of time that there is a noticable difference between BG and ISF levels.

 

[In Response]

BG - badly calibrated Glucometers, old or damp test strips, unwashed fingers with sugar on them.

Exactly the point I was making in comment #47. But obviously not clear enough as it took us a few comments to get there.
Sorry for my lack of clarity and brevity.

 

[Jasmin2000]

Exactly the point I was making in comment #47. But obviously not clear enough as it took us a few comments to get there.
Sorry for my lack of clarity and brevity.

No worries - also my bad as 90% of comments that ask 'why my sensor not giving me the same as BG' haven't heard of the lag or assume that somehow the device can eliminate it.

 

[Jasmin2000]

Libre Itself is simply a CGM - it monitors interstitial fluid for current BG readings (usually 15-20 mins behind an actual BG reading), it doesn't know about ISF or insulin on board etc - just gives a BG reading.

CGMs do not give BG readings ever. They read glucose levels in ISF and report ISF levels.