How does Libre calculate A1c ?

I found the libre very inaccurate and quit using it. It’s good to see trends, but for me that was about it.

Nick25 said:

Not sure if Libre is providing me and medical staff with misleading record of my A1c. My continuous Libre figures fluctuate ( of course) but I am unclear whether the quite good average A1c which Libre produces, is calculated using the continuous monitoring figures , or only those when I carry out scans.
I have periods when my graph shows it has gone higher , but if I haven’t scanned then does that higher period get included in the overall A1c?

Click to expand...

Hi @Nick25 - A1c on the Libre is estimated using the continuous monitoring figures - not just your scans.

The Hb in HbA1c is hemoglobin and the blood test measures how many glucose units have stuck to your Hb during the last 3 months to make glycated-Hb. The blood lab can quantify glycated-Hb and gives you an HbA1c value.

Libre sensor does not measure anything in blood and has no idea what your glycated-Hb looks like, so they developed an algorithm to estimate A1c (the Glucose Management Indicator, GMI). GMI estimates A1C based on the average glucose level from the continuous monitoring sensor readings during the report period. This is what you see on the Libre under "Estimated A1c"

The Estimated (GMI) value and the true blood HbA1c are rarely the same and a 20-35% difference is not unusual (my Libre A1c is 4.8% but my blood A1c is 5.7%).

Nick25 said:

I have a second problem that my blood glucose readings are often 2 or more higher than Libre readings. Not sure whether this is just because one is anticipating a rise before the other, or a semi permanent under reading - in which case the Libre A1c figure that satisfies my nurse is that accurate.

Click to expand...

.Could be either - I would check your Libre after the BG value and see if it appears later. This lag is well known and can be a few minutes to >20 minutes. I've tracked BG and Sensor readings in parallel after an outragous carb intake (a honey croissant - but all in the name of science!). BG rose to 13 over 60 mins and the Libre followed but was 5-10 mins behind.

If yous suspect an underreading then you should check the position and seal of the sensor and/or do the BG test again with a different finger. The best time to check is when you have a stable BG over 30 min as this should be reflected on the Libre.

The figures from the libre are just an estimate which can vary quite a bit from the actually figure, your team will only use it as a rough guide to your general control but will use an actual blood test when they want the hba1c

Nick25 said:

Not sure if Libre is providing me and medical staff with misleading record of my A1c. My continuous Libre figures fluctuate ( of course) but I am unclear whether the quite good average A1c which Libre produces, is calculated using the continuous monitoring figures , or only those when I carry out scans.
I have periods when my graph shows it has gone higher , but if I haven’t scanned then does that higher period get included in the overall A1c?
I have a second problem that my blood glucose readings are often 2 or more higher than Libre readings. Not sure whether this is just because one is anticipating a rise before the other, or a semi permanent under reading - in which case the Libre A1c figure that satisfies my nurse is that accurate.

Click to expand...

I don’t know how the Libre estimates A1C but it was pretty close.
Libre estimated 37
A1C provided 38

Close enough for me.

For me, the Libre predicted hba1c has been lower than the lab ones consistently for 7 years. Every time, my hba1c has been 8 mmol/mol higher than Libres prediction (so I'm still quite sure what my hba1c would be by looking at the Libre and adding 8.)
This difference perfectly adds up with Libres for me usually reading between 0.8 and 2.2 mmol/l lower than blood.

Jasmin2000 said:

If yous suspect an underreading then you should check the position and seal of the sensor and/or do the BG test again with a different finger. The best time to check is when you have a stable BG over 30 min as this should be reflected on the Libre.

Click to expand...

For some people Libre simply underreads.

Antje77 said:

For me, the Libre predicted hba1c has been lower than the lab ones consistently for 7 years. Every time, my hba1c has been 8 mmol/mol higher than Libres prediction (so I'm still quite sure what my hba1c would be by looking at the Libre and adding 8.)
This difference perfectly adds up with Libres for me usually reading between 0.8 and 2.2 mmol/l lower than blood.

For some people Libre simply underreads.

Click to expand...

I get that and as we're all different it's not surprising that for some it underreads (including me).
The real shame is that the medical profession continue to cite TIR and estimated A1c on the CGM instead of true BG for which these recommended levels were given by NICE/NHS.

Jasmin2000 said:

The real shame is that the medical profession continue to cite TIR and estimated A1c on the CGM instead of true BG

Click to expand...

I've never heard of medical professionals not doing bloodwork anymore for hba1c's.
The TIR is very useful, as long as you know the pitfalls, hba1c doesn't tell enough.

I use DiaBox alongside the official Librelink app.
Unlike Librelink, DiaBox can be calibrated.
My endo is perfectly happy to accept DiaBox's TIR instead of the lying Librelink.
Currently the difference is that DiaBox thinks I'm low 4% of the time over the last 3 months (most not lower than 3.7, which my endo finds acceptable), Librelink thinks I've been low for 22% of the time. Which would of course make all alarms go off with my endo should she insist to use Libre data instead of DiaBox.

It's only a mild calibration, usually less than 1 mmol/l difference, and I like to err on the side of caution to not miss hypo alarms on DiaBox, and there's always that pesky first day before calibration, so the real % is likely lower than that 4%.

That said, I didn't even show her my phone at my last appointment, she simply believes what I tell her. I am very, very lucky with my endo.

Nick25 said:

quite good average A1c which Libre produces, is calculated using the continuous monitoring figures , or only those when I carry out scans.

Click to expand...

Abbott don't say exactly how they calculate it, but we know that they record your average blood sugar every 15 minutes. I exported my data from the LibreView website and calculated the A1c based on those 15 minutes blocks and it matches what the libre suggests.

Also, it wouldn't make sense to base it only on scans, as there are some people that only scan at meal-times (and this number has increased since they changed the app to display your figures every minute which means you don't need to scan).

Accuracy seems to differ from person to person. My libre very rarely differs from a finger prick reading, but my HbA1c is consistently higher than the libre's estimate. Finger prick meters are not all that accurate either, but it is impossible to estimate A1C based on finger prick readings alone so nobody really know what their HbA1c is going to be.

I have say that the diabetic nurse at hospital clinic demonstrates rather greater faith in Libre figures than I!

Click to expand...

A doctor/diabetic nurse being a fan of the Libre does not necessarily mean they are blind to the possible problems. It is a tool which can give both you and your doctor/nurse a lot of information and that information may be helpful.

My own view is that nothing is accurate, including the HbA1c test.

There is also the little thing of LibreLink giving a pretty different estimated hba1c than their own platform LibreView.
So both services by the same provider must be using different calculations on the same raw data...

My Librelinks 'estimated hba1c' is 8 mmol/mol lower than LibreViews GMI, I just checked.

Antje77 said:

There is also the little thing of LibreLink giving a pretty different estimated hba1c than their own platform LibreView.
So both services by the same provider must be using different calculations on the same raw data...

My Librelinks 'estimated hba1c' is 8 mmol/mol lower than LibreViews GMI, I just checked.

Click to expand...

Oh that's depressing if true. I'll have to have a look. Would it be so difficult for them to sync the Libreview with LibreLinK?

Jasmin2000 said:

Oh that's depressing if true. I'll have to have a look. Would it be so difficult for them to sync the Libreview with LibreLinK?

Click to expand...

I'm curious if it's the same for you.
Of course it wouldn't be difficult for Abbott to sync them, but which one should they choose?
For me, the Librelink one matches with the difference between fingerprick and sensor, they both read too low.
The LibreView one is close to my lab hba1c, but doesn't match the average BG according to Libre.

Antje77 said:

My Librelinks 'estimated hba1c' is 8 mmol/mol lower than LibreViews GMI, I just checked.

Click to expand...

Most people seem to thing that GMI and A1c/HbA1c are different names for the same thing, but they aren't and they have different values (unless your values to be 7% or around 53 mmol/mol which is the same on both scales).

No idea what your figures are, but if your Average was 4.85 (is that even possible ?) that would mean a GMI of 35.53 and a HbA1c of 27.53 which is lower by 8

If your average BS was about 12.3 it would be the other way around and your HbA1c would be 8 higher than GMI.

When I checked them, they matched for me, i.e. taking my estimated A1C for the libre and converting to GMI gave the figure on the libreview after I changed the report to be based on 90 days.

sgm14 said:

No idea what your figures are, but if your Average was 4.85 (is that even possible ?) that would mean a GMI of 35.53 and a HbA1c of 27.53 which is lower by 8

Click to expand...

Wow, I'm now completely puzzled in a very interested way!
I asked about the difference in Libreview/Librelink a couple of years ago (from memory), and no-one seemed to have the answer.

As for my own numbers, which I don't usually share but they seem to be relevant here, you're pretty close with your guess. Measured hba1c is usually 35 or 36, last time - januari - it was lower due to having had to dial my basal down during the month before the blood draw. Dialling Tresiba down takes a bit of time, and I spent more time in the lowish 4's than I like.
If I were to test now I'd expect to be back at a more comfortable 34-36.

According to the chart I use, an hba1c of 31 equals an average BG of 5.4, 36 should be an average BG of around 6.0. Which seems to add up with my fingerpricks and graphs.

sgm14 said:

Most people seem to thing that GMI and A1c/HbA1c are different names for the same thing, but they aren't and they have different values (unless your values to be 7% or around 53 mmol/mol which is the same on both scales).

Click to expand...

Before I dive down the google rabbit hole to never emerge again, do you happen to have a link to something explaining the difference between (estimated) hba1c anf GMI?
Or if you'd like, explain it yourself of course, not doubting you, just wanting to spare you some work!

sgm14 said:

Most people seem to thing that GMI and A1c/HbA1c are different names for the same thing, but they aren't

Click to expand...

OK so I get that they aren't the same (and shame on Abbott for calling it "estimated A1c") - it is GMI that the LIbre determines - right?

Antje77 said:

I'm curious if it's the same for you.
Of course it wouldn't be difficult for Abbott to sync them, but which one should they choose?
For me, the Librelink one matches with the difference between fingerprick and sensor, they both read too low.
The LibreView one is close to my lab hba1c, but doesn't match the average BG according to Libre.

Click to expand...

I would choose the one that made most sense when I see how it's calculated. Ignoring the "estimated A1c" term and calling it GMI from now on, I still have to understand why Librelink and LIbreView give different GMIs. Libreview can give you a GMI for different time periods, so maybe that needs checking.

Jasmin2000 said:

OK so I get that they aren't the same (and shame on Abbott for calling it "estimated A1c") - it is GMI that the LIbre determines - right?

Click to expand...

Estimated hba1c is what you get on LibreLink, GMI is what you get on LibreView.
So apparently different things with different names, I had no idea either.

I'm sure @sgm14 will get back with us with something to help us understand the difference.

Antje77 said:

Before I dive down the google rabbit hole to never emerge again, do you happen to have a link to something explaining the difference between (estimated) hba1c anf GMI?
Or if you'd like, explain it yourself of course, not doubting you, just wanting to spare you some work!

Click to expand...

Some references.



I don't know whether I am misinterpreting the latter, but it looks like GMI was chosen because the estimated A1C that CGM's give can't really ever match the HbA1C and they thought it was better to give it a completely different to avoid the confusion. Unfortunately I think it has just added to the confusion.


Most of the references I found where starting from average blood sugar measured in mg/dl, so if you are starting in mmol/L you need to divide by 0.0555 to get mg/dl

My C# code (which I am hoping is correct)

var GMI_Perc = Math.Round(3.31m + (0.02392m * (average / 0.0555m)), 2);
var GMI = Math.Round(12.71m + 4.70587m * average , 2);

var HbA1c_Perc = Math.Round(((average / 0.05555m) + 46.7m) / 28.7m , 3);
var HbA1c = Math.Round(((HbA1c_Perc - 2.15m) * 10.929m) , 2);

Jasmin2000 said:

OK so I get that they aren't the same (and shame on Abbott for calling it "estimated A1c") - it is GMI that the LIbre determines - right?

Click to expand...

Actually there are calling it by the correct name.
The libre app is trying to estimate your A1C
The LibreView website is calculating your GMI.

The shame (as you put it) is that they should have either used the same term and the same values in both places or else used both terms and both values

I guess that when cgms were introduced they thought they could estimate what you A1C would be, but over time realised that they are not and never could be accurate way to tell what your HbA1C would be and so now they are trying to move away from using the term A1C. At that same time they decide to change the algorithm which means that not only are the names different, but the values are different.

From healthline.com (2022), "Leading diabetes experts have determined that previous term “estimated A1C” wasn’t helpful because people with diabetes might assume that it’s comparable to their 3-month A1C results." Looks like Abbot didn't get the memo!