This article is for those of you curious about the GMI value shown on the Dexcom Clarity report.

[littlevoice359]

I have noticed that the Dexcom Clarity report now has a GMI value that looks like an estimated HbA1c value. I was curious as to what exactly this is, so I went digging. The link below is the best information source I have found to date, so I thought I would pass it along in case others among you are equally curious.

Glucose Management Indicator (GMI): A New Term for Estimating A1C From Continuous Glucose Monitoring

While A1C is well established as an important risk marker for diabetes complications, with the increasing use of continuous glucose monitoring (CGM) to hel

diabetesjournals.org

 

[Bruce Stephens]

That's a bit disappointing. It looks like it's basically mean BG, specifically GMI (mmol/mol) = 12.71 + 4.70587 × [mean glucose in mmol/L].

I'm not sure what value that gives beyond just mean BG (which also seems rather easy to understand).

 

[littlevoice359]

That’s pretty much it, insofar as the maths is concerned. One assumes those constants are the result of regression analysis of some description. I just wanted to satisfy my own curiosity so the question of disappointment didn’t come into it for me. I was able to tweak my own software to give me a GMI value outside of Dexcom Clarity so I’m satisfied with that.

 

[Bruce Stephens]

One assumes those constants are the result of regression analysis of some description.

Yes, I get the impression they were trying to get a value matching the HbA1c estimates, and maybe that makes sense. (I'm not convinced, but it's quite possible I'm wrong.)

 

[littlevoice359]

For what it’s worth, using this new equation with my own CGM data gives me a GMI value of 48, while the esimated HbA1c based on average blood glucose gives a value of 46. My lab result has never been under 48, so my hope is that GMI is a better predictor of my future lab results. Only time will tell…

 

[everydayupsanddowns]

Interesting @littlevoice359 thanks for the link

That's a bit disappointing. It looks like it's basically mean BG, specifically GMI (mmol/mol) = 12.71 + 4.70587 × [mean glucose in mmol/L].

I'm not sure what value that gives beyond just mean BG (which also seems rather easy to understand).

For me the GMI seems to track pretty well against my most recent actual A1c result (a couple of decimal points out).

The old ‘fingerstick’ empirically derived conversion formula I came across years ago
Est HbA1c % = (AvgBG+2.52) / 1.583

(a ‘best fit‘ calculation which came from hundreds of paired A1c/average BG datapoints) was always quite a long way out for me. At least 0.5-1% (5.5 - 11mmol/mol)

And Libre’s estimate, however that is derived, always seems to underestimate too.

It looks like this is an updated empirical conversion to me - that they have extracted pairs of CGM averages from modern systems for which they have matching lab A1c values from various trials - and then made a new ‘best fit’ formula. The old BG averages would have been fingersticks so CGM averages have the benefit of catching all the highs and lows I guess… Plus all overnight levels.

 

[littlevoice359]

Interesting @littlevoice359 thanks for the link



For me the GMI seems to track pretty well against my most recent actual A1c result (a couple of decimal points out).

The old ‘fingerstick’ empirically derived conversion formula I came across years ago
Est HbA1c % = (AvgBG+2.52) / 1.583

(a ‘best fit‘ calculation which came from hundreds of paired A1c/average BG datapoints) was always quite a long way out for me. At least 0.5-1% (5.5 - 11mmol/mol)

And Libre’s estimate, however that is derived, always seems to underestimate too.

It looks like this is an updated empirical conversion to me - that they have extracted pairs of CGM averages from modern systems for which they have matching lab A1c values from various trials - and then made a new ‘best fit’ formula. The old BG averages would have been fingersticks so CGM averages have the benefit of catching all the highs and lows I guess… Plus all overnight levels.

This matches my own thoughts on the subject. The attached excerpt from the paper describes the regression that led to the revised estimation algorithm.

 

[everydayupsanddowns]

My knowledge of maths rules is terrible, but does

GMI (mmol/mol) = 12.71 + 4.70587 × [mean glucose in mmol/L].

actually mean

GMI (mmol/mol) = (4.70587 × [mean glucose in mmol/L] ) + 12.71

because of the order of operations thing??

 

[littlevoice359]

My knowledge of maths rules is terrible, but does

GMI (mmol/mol) = 12.71 + 4.70587 × [mean glucose in mmol/L].

actually mean

GMI (mmol/mol) = (4.70587 × [mean glucose in mmol/L] ) + 12.71

because of the order of operations thing??

Yes. Multiplication happens before addition or subtraction. As a software engineer I personally use parentheses liberally to make my code less confusing.

Example:

Average CGM blood sugar over past 90 days: 7.6
GMI: (4.70587*7.6) + 12.71 = 48.474. ( rounded down = 48 )

 

[everydayupsanddowns]

As a software engineer I personally use parentheses liberally to make my code less confusing.

Don’t write so much code any more, but me too!

Writing sums in sentences with no ‘punctuation’ confuses the heck out of me. Especially as in that formula, as otherwise you basically have to read it right to left 🙄

 

[Sally71]

Not being a mathematician myself, I will never understand why multiplication and division have to take priority anyway, why would you not just read the sum from left to right? (Unless anything is in parentheses, I do understand what that’s for.). You read words from left to right without jumping about, so why not numbers? But if you ask that question to anyone who does a lot of maths they look at you as if you have just asked them the stupidest question in the world!

 

[Bruce Stephens]

Not being a mathematician myself, I will never understand why multiplication and division have to take priority anyway, why would you not just read the sum from left to right?

Almost always, the reason for such conventions is that things turn out more conveniently that way. As far as I'm aware that's the case for the various rules around precedence (including the more exotic rules like right associativity of exponentiation). It's fine to add parentheses if that makes things clearer; wouldn't hurt in this case but of course in more complex formulas the extra parentheses can get in the way (which is why we have the conventions).

 

[Inka]

I’m not a mathematician either @Sally71 so I’m probably the best-placed to explain to you not having much mathematical knowledge myself It’s simply an agreed way to do things to ensure you don’t get wrong answers. BODMAS is what I was taught, and although it might seem weird initially, it actually makes things a lot simpler (even for a non-mathematician like me!). It helps to break calculations down and do them in the right order:

https://www.mathsisfun.com/operation-order-bodmas.html

.