Medical AIEstimated average glucose (eAG) from HbA1c
Estimated average glucose (eAG) converts an HbA1c percentage into the average glucose it corresponds to, expressed in the same units patients see on a meter (mg/dL or mmol/L). It comes from the ADAG study regression. eAG is a population-level estimate of average glucose over the preceding 2 to 3 months, not a substitute for actual glucose monitoring, and it should not be confused with the CGM-derived glucose management indicator (GMI).
How to measure each input
- HbA1c value
- Use a laboratory HbA1c reported in NGSP / DCCT-aligned percentage units. The ADAG regression was derived across this range; enter the percentage, not mmol/mol.
- Units of the result
- eAG is shown in mg/dL (primary) and mmol/L (secondary), the same units used on glucose meters, so patients can compare it directly with their readings.
- When A1C is reliable
- The estimate is only as good as the A1C. It assumes normal red-cell turnover and no interfering haemoglobin variant.
Interpretation
| Band | Meaning |
|---|---|
| eAG below ~114 mg/dL (HbA1c < ~5.7%) | Corresponds to an HbA1c in the normal (non-diabetes) range. An A1C below 5.7% is generally considered normal by ADA criteria; interpret in clinical context. |
| eAG ~114 to 153 mg/dL (HbA1c ~5.7 to 6.9%) | Spans the prediabetes range (A1C 5.7 to 6.4%) and the lower diabetes range. An A1C of 6.5% or above on confirmed testing meets the diabetes threshold; for many treated adults a target below 7% is reasonable. |
| eAG ~154 mg/dL or above (HbA1c >= ~7%) | Average glucose at or above the level corresponding to an A1C of about 7%, which exceeds the common general target for many non-pregnant adults. Reassess the management plan, individualising the goal. |
Pitfalls, exclusions and caveats
- Conditions that invalidate HbA1c also invalidate eAG: haemoglobinopathies and variant haemoglobins, haemolytic or iron-deficiency anaemia, recent blood transfusion or significant blood loss, pregnancy, and advanced chronic kidney disease or recent erythropoietin use.
- eAG is a population-level estimate from a single A1C. Within the ADAG cohort there was a range of average glucose values for any given A1C, so an individual's true mean glucose can differ from the estimate.
- eAG is not the same as the glucose management indicator (GMI). GMI is calculated from continuous glucose monitoring (CGM) data over a recent period, whereas eAG is derived from a laboratory A1C; the two can diverge.
- It is not a substitute for glucose monitoring. It cannot show variability, hypoglycaemia, or post-meal excursions, which matter for safety and dose adjustment.
- The ADAG cohort was predominantly non-Hispanic White adults; the A1C to glucose relationship can vary by ethnicity and other factors, so apply judgement in diverse populations.
eAG (mg/dL) = 28.7 x HbA1c(%) - 46.7. eAG (mmol/L) = eAG(mg/dL) / 18.018 (equivalently 1.59 x HbA1c - 2.59). From the A1c-Derived Average Glucose (ADAG) study, Nathan et al. 2008.The eAG regression is from the A1c-Derived Average Glucose (ADAG) study (Nathan et al., 2008) and is endorsed in ADA guidance. This implementation is educational and not affiliated with the ADA or the ADAG investigators.
Frequently asked
What is the difference between eAG and HbA1c?
They describe the same thing in different units. HbA1c is the percentage of glycated haemoglobin; eAG re-expresses it as an average glucose in mg/dL or mmol/L using the ADAG study formula, so it lines up with the numbers seen on a glucose meter.
Is eAG the same as GMI from my CGM?
No. The glucose management indicator (GMI) is calculated from continuous glucose monitoring data over a recent window, while eAG is derived from a laboratory HbA1c reflecting the past 2 to 3 months. They often differ, and a gap between them can itself be informative.
When is eAG unreliable?
Whenever HbA1c is unreliable: haemoglobin variants, haemolytic or iron-deficiency anaemia, recent transfusion or blood loss, pregnancy, and advanced kidney disease. In these settings, direct glucose monitoring is more trustworthy than an A1C-derived estimate.
A score is a start. The product is asked.
Calculators give a number. When the patient in front of you needs the reasoning behind it, with the sources, the product is what does the looking-up.
Medical AI returns evidence-grounded answers backed by real citations. It is a reference tool, and these terms describe how it should and should not be used.
01 Educational purpose
Medical AI is an information and reference tool intended for educational use only. The answers it returns are not medical advice, diagnosis, or treatment. Always consult a qualified doctor or healthcare professional with any question concerning a medical condition.
02 Professional use
Medical AI is designed for use by practicing clinicians. It is not intended for direct patient use and is not a substitute for professional clinical judgment. Apply your own training and current guidelines to every decision an answer informs.
03 Privacy and data
We do not collect, store, or process personally identifiable patient information (PHI or PII). Do not enter names, dates of birth, medical record numbers, or any other patient identifiers into the composer.
04 Information currency
Medical knowledge evolves rapidly. Citations carry their publication date, so consult the primary source and the most recent clinical guideline before acting on anything material.