GMI – is it useful?

As I posted recently, my CGM data has been estimating my HbA1c will be 5.2% (which it was: 33 mmol/mol).¬†However while Sugarmate also calculated an estimated HbA1c of 5.2%, its “GMI” value was 5.8%. I’ve always struggled to comprehend what GMI is meant to indicate.

What is GMI?

It turns out that “GMI” (Glucose Management Indicator) is a measure that was created because some CGM users were distressed that their estimated A1c values did not match their lab A1c values (see the paper abstract describing it).

As referenced in earlier articles, I use BG meters that are reasonably accurate, and use those to calibrate my CGM. I run a closed-loop insulin pump system that depends on my CGM data being as accurate as possible. An inaccurate CGM is not safe, and I can not accept that. It is occasionally slightly”off” (nothing stays stable forever) but that’s soon rectified. “Fasting glucose” lab tests have consistently come in quite close to my meters and CGM data too.

This might explain why my lab HbA1c results have always been close to my CGM estimations, as they’ve been working off a hopefully-accurate base. The actual and estimated (HbA1c and “eHbA1c”) values have always been within 2-3 mmol/mol (and exactly the same on this last example). That’s within 0.2% or so in DCCT units. This has been the case for values between 33-55 mmol/mol (5.2-7.2%).

There are multiple reasons why the eHbA1c and lab HbA1c can differ, including:

  • Biological differences, including haemoglobin that doesn’t have a 120-day life. See this discussion at NGSP.
  • Significant changes in average BG during the sample period. The eHbA1c is a simple calculation from the average BG over the last 90 days, while the HbA1c test measures haemoglobin which is made up of a mixture of ages back as far as¬†120-days. The approximation of a 90-day average for a 120-day linear decay breaks down when there’s been a significant difference in average BG across those 90 days (in the lab test the more-recent average has a larger effect).

What is supposed to mean?

The GMI paper mentions using “a new formula for converting CGM-derived mean glucose to GMI based on recent clinical trials using the most accurate CGM systems available”. But somehow it doesn’t seem to match the data from my own CGM. It doesn’t try to take into account viariability like Dexcom’s GVI and PGS metrics: it’s just a different formula converting from the same 90-day average BG reading as eHbA1c.

Is it just a “fudge” to placate people who don’t like the eHbA1c numbers? It certainly smells a bit that way: at least in the range of values I’m used to seeing.

By using a name that doesn’t include the letters “A1c”, GMI manages to avoid too much association/comparison with the HbA1c measurement. But then how are we supposed to interpret the GMI percentage value?

All I can really say is that GMI has not yet shown any meaning or use for me.

2 thoughts on “GMI – is it useful?”

  1. Both eA1c and GMI are based off of your CGM data Average BG for the time period and are expressed as a % value. With GMI you can use as few as 12 days. It is not meant to do anything other than give a guideline or indicator of how control is doing shorter term than an actual lab A1c test. It will be unusual for either of these to match a lab test and that is not the goal. GMI was never meant to be used to evaluate 90 days of data and compare this value to a lab A1c. For those who have a rather unstable BG profile, GMI may be a better tool. If you have sable BG and a low SD for a full 90 days, the GMI number calculates a higher result than eA1c does. Of course, lab A1c is not a simple average of your BG levels. It is affected profoundly by the BG levels during the most recent 6 weeks.

    eA1c Calculation: (46.7 + average_blood_glucose in mg/dL) / 28.7

    GMI Calculation: 3.31 + (0.02392 x [mean glucose in mg/dL])

    1. In my experience it IS usual for the “estimated A1c” to be very close to the lab HbA1c.

      Yes I am aware of the HbA1c test being affected by the 120-day (average) life of the cells, and I’ve done enough modelling of the data to know that a 90-day average is a reasonable approximation of the results. As long as the average hasn’t changed dramatically recently.

      And if the average hasn’t changed much, it doesn’t really matter that much how many days you calculate the average of. But the GMI figure doesn’t seem to have any useful relationship to any other metric that I can see, whether the overall control has changed recently or not.

Leave a Comment

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.