Large discrepancy between estimated A1c on Libreview vs Libre Link App

[Ronancastled]

I have been reviewing my sensor data from my 14 day Libre trial.
My Average Glucose on both the Libreview online portal & the Libre Link App was 4.6mmol/L
Yet they've used some weird algorithm called GMI to estimate my A1c as 5.3% or 34.

If I do an online conversion of my average glucose to A1c I come out with the much lower estimate of 4.5% or 26.

So I consult the Libre Link App which agrees with the conversion above.

I suppose what I must ask is what is this GMI & how accurate have others found it in relation to your lab A1c vs CGM A1c ?

 

[Antje77]

Same here.
LibreView said 35.
LibreLink said 27.
Lab said 35.

So LibreView was right. The weird thing is that my Libres tend to read significantly lower than blood for me, so I would expect it predicting LibreLink's 27 when the lab said 35.

LibreView must use some very strange algorithm to reach its conclusions.

 

[Rokaab]

This is an article where it talks about HbA1c and GMI, they are not the same - which is probably why they don't always match https://care.diabetesjournals.org/content/41/11/2275

 

[Ronancastled]

LibreView must use some very strange algorithm to reach its conclusions.

I expect the same yet I'm drawn back to the CGM data of the 153 non-diabetics I posted last week.
Their average glucose, according to the Dexcom, was 99mg/dl (5.5mmol/L)

So that converts to an A1c of 5.1% which is exactly the average of the non-diabetic group.

Are the Dexcoms more accurate than the Libre or do they use a GMI as well ?

 

[In Response]

I pay no attention to the HbA1C estimate from Libre (or xDrip which I prefer to use).
For me the Time in Range is much more important and relevant: I could have a really low HbA1C if I spend more time hypo.
As my Libre is funded by NHS, they are becoming more interested in Time in Range as they see the value.

 

[Antje77]

Are the Dexcoms more accurate than the Libre or do they use a GMI as well ?

Dexcom is said to be more accurate than Libre.
I've heard people report both dexcom and libre are inaccurate, and I've heard people report their sensor is on spot. So apparently it differs from person to person. And of course there are many people who hardly ever do a fingerprick, so they won't know the accuracy of their sensors except for the difference in hba1c.

At the moment I'm using the 3rd party app DiaBox together with the LibreLink app. In DiaBox you can calibrate, so now my sensors seem to be much closer to fingerpricks.

 

[Antje77]

I pay no attention to the HbA1C estimate from Libre (or xDrip which I prefer to use).
For me the Time in Range is much more important and relevant: I could have a really low HbA1C if I spend more time hypo.
As my Libre is funded by NHS, they are becoming more interested in Time in Range as they see the value.

But TIR is affected by inaccuracies as well. According to LibreLink and LibreView I'm low 6% of the time. I know I'm not because I usually do a fingerprick to double check.
If I'd share my data with my HCP they would see a diabetic who has lots of hypo's and they'd likely urge me to get my numbers higher.
The calibrated DiaBox app says I'm low 0% of the time, which isn't correct either, it misses some hypo's.
The truth lies in the middle, I'm pretty sure I'm not below 4 much more than 1% of the time.

So I agree TIR is very important, but it's prone to the same weaknesses in the system as estimated hba1c's and GMI's.

None of this answers the question on why the Libre itself shows a difference in estimated hba1c and GMI of 8mmol/mol for both Ronancastled and myself. You'd think they would decide on an algorithm which would come up with the same results on both their platforms.

 

[Ronancastled]

None of this answers the question on why the Libre itself shows a difference in estimated hba1c and GMI of 8mmol/mol for both Ronancastled and myself. You'd think they would decide on an algorithm which would come up with the same results on both their platforms.

I would infer from the existence of an algorithm & the invention of a new term, GMI, that the manufacturer admits that the sensor is inaccurate but don't want to admit it publicly or alter the production/software to suit. rather they fix it in the mixing if anyone remembers live albums.

Remember the old venous vs capillary blood debate from finger sticking, then the meter's software was altered to read 12% lower (from memory) so they were more accurate to a venous lab draw.

I'm actually overdue lab work so it will be interesting to see the real HbA1c.

Then there's a bigger debate, what's more important, TIR, AUC, Glucose Variability, SD . . . twas simpler back in the olden days

 

[Goonergal]

would infer from the existence of an algorithm & the invention of a new term, GMI, that the manufacturer admits that the sensor is inaccurate but don't want to admit it publicly or alter the production/software to suit. rather they fix it in the mixing if anyone remembers live albums.

I quite like the GMI figure - it was far closer to my last HbA1c than the straight line average previously used. Since the Libre uses interstitial fluid, not blood, it makes sense that the numbers aren’t entirely aligned and this seems to be a formula that gives a closer match.

All said and done, meters, Libre, HbA1c all working in different ways. So long as this is understood, I don’t see any issues.

 

[Antje77]

All said and done, meters, Libre, HbA1c all working in different ways. So long as this is understood, I don’t see any issues.

The issue is that the same Libre data are used by the same company (Abbott) to give different results for the same person. That's just weird!
And an 8mmol/mol difference is not a small difference either.

 

[Goonergal]

The issue is that the same Libre data are used by the same company (Abbott) to give different results for the same person. That's just weird!
And an 8mmol/mol difference is not a small difference either.

But they’re labelled as two different things. I don’t find it an issue.

 

[Antje77]

But they’re labelled as two different things. I don’t find it an issue.

They're both labeled as an estimation of your hba1c. Both estimations are based on the same measurements of the same sensor in the same interstitial fluid at the same time by the same company. I think it's as strange as a single lab giving you two different hba1c's from the same blood draw. Especially as the difference is large enough to either decide perfectly healthy, no prediabetes at all or full blown diabetes.

Such a difference between Libre prediction and lab hba1c, yes I can understand. But such a difference based on the exact same data suggests something's off with the system interpreting those data.

 

[Goonergal]

They're both labeled as an estimation of your hba1c. Both estimations are based on the same measurements of the same sensor in the same interstitial fluid at the same time by the same company. I think it's as strange as a single lab giving you two different hba1c's from the same blood draw. Especially as the difference is large enough to either decide perfectly healthy, no prediabetes at all or full blown diabetes.

Such a difference between Libre prediction and lab hba1c, yes I can understand. But such a difference based on the exact same data suggests something's off with the system interpreting those data.

But one is an average of the readings and the other is an algorithm based on the actual HbA1c scores of those with similar averages/data to you. So yes, using the same data but in different ways. That’s not unusual with any data analysis.

Think we may have to agree to differ!

 

[LittleGreyCat]

Same here.
LibreView said 35.
LibreLink said 27.
Lab said 35.

So LibreView was right. The weird thing is that my Libres tend to read significantly lower than blood for me, so I would expect it predicting LibreLink's 27 when the lab said 35.

LibreView must use some very strange algorithm to reach its conclusions.

So LibreView appears to be accurate?

 

[Jaylee]

This is an article where it talks about HbA1c and GMI, they are not the same - which is probably why they don't always match https://care.diabetesjournals.org/content/41/11/2275

Hi @Rokaab ,

Thanks for posting this link.
From what I can gather. The "GMI" from a T1 insulin managment perspective means an HCP could possibly acertain if a low HbA1c, or even lower than their "ideal" figures in the hand book was accomplished by "too many hypos?"
"Glucos Managnemt Indicator."

Many years ago, (way back in the dark times before sensors?) I had to argue with a DSN after a glowing A1c result that I was actually getting no more than usual.
Lol, she was wrongly adamant I must be..

 

[Ronancastled]

But they’re labelled as two different things. I don’t find it an issue.

@Goonergal I think I have you, you're saying that the GMI is just a new unit of measurement which may or may not reflect your true A1c ?
What I don't understand is how the A1c in the non-diabetic group exactly matched their average glucose measured by the Dexcom. Perhaps they also had GMI's higher than their A1c, now I think I'm just confusing myself.

 

[Antje77]

So LibreView appears to be accurate?

For me it does. But I've seen many people reporting that the estimated hba1c from LibreLink was quite accurate for them. If they had the same discrepancy between LibreLink and Libreview, LibreView must have been way out for them.

now I think I'm just confusing myself.

Not only yourself, this whole thing is confusing me as well!

the other is an algorithm based on the actual HbA1c scores of those with similar averages/data to you.

Would this mean the established link between average BG and hba1c is wrong?

 

[Jaylee]

Would this mean the established link between average BG and hba1c is wrong?

The HbA1c, is like an "average speed check" on the motorway. It can flag up someone booting it down the road, but not where they were doing it?
Of course a "model driver" could have been flooring it over a short distance & done the rest pootling along at 20mph to compensate in order not to get a fine after seeing the signs..?

Hence why the "GMI" was devised. But I'm only guessing. It's a little like having your HCPs following you about.
Or having your vehicle fitted with tha tracker..

 

[Goonergal]

Would this mean the established link between average BG and hba1c is wrong?

I think even that is an estimate as the blood analysis is done differently with the HbA1c. What @Jaylee says above is a much clearer explanation than I could give!