When does HbA1c lie? My A1c dropped 0.7 with no change in my CGM average

Two data streams disagreeing, and I cannot reconcile them. Over eleven weeks my HbA1c fell from 7.1% to 6.4%. Over the same eleven weeks my sensor mean glucose was 8.6 mmol/L in the first half and 8.5 mmol/L in the second — statistically indistinguishable. Time in range barely moved. My weight is down 6 kg, so something is happening, but not to my glucose.

In between those two draws I was treated for iron deficiency and my haemoglobin went from 108 to 131 g/L. I am now suspicious that the A1c fall is a haematological artefact rather than a glycaemic improvement, but I do not know the mechanism or the direction well enough to be sure I have it the right way round.

Broader question, since I would like to know when to distrust this test in general: what are the conditions under which HbA1c systematically misreads, in which direction does each one push it, and what do you use instead? I have seen fructosamine mentioned but never explained, and I do not know whether a sensor-derived figure is a substitute or just a different measurement.

On the alternatives, since the question asked what to use instead and the answer differs by how long a window you need.

Fructosamine

Fructosamine measures glycated serum proteins, overwhelmingly albumin. Albumin's half-life is around 14 to 20 days, so fructosamine integrates glucose over roughly the preceding 2 to 3 weeks rather than 3 months. Its virtues and vices both follow from that:

• Virtue: completely independent of red cell lifespan. Haemoglobinopathy, transfusion, haemolysis and iron status do not touch it. This is the reason it exists.
• Virtue: a shorter window, so it responds to a change in weeks rather than months.
• Vice: it depends on albumin turnover instead. Nephrotic syndrome, protein-losing enteropathy, cirrhosis, severe hypoalbuminaemia and thyroid dysfunction all shift it — nephrotic syndrome markedly downwards, because albumin is being lost before it finishes glycating.
• Vice: assay standardisation is poor compared with A1c, reference intervals vary between labs, and there is no outcome literature relating fructosamine values to complications. There is no equivalent of "48 mmol/mol means diabetes".

Glycated albumin is a refinement of the same idea that expresses the result as a percentage of albumin and so partially corrects for albumin concentration. It is more useful than fructosamine and much less widely available.

Practical positioning: fructosamine is a cross-check, not a replacement. It answers "is this person's glucose roughly where their A1c suggests" when you have reason to distrust the A1c. It does not answer "what should this person's target be".

Continuous glucose monitoring

A sensor measures a different quantity and it is the right tool when A1c is confounded. The standardised metrics worth knowing:

• Mean glucose over at least 14 days with at least 70% sensor wear. Shorter or patchier records are not representative.
• Time in range, conventionally 3.9 to 10.0 mmol/L (70 to 180 mg/dL), reported as a percentage.
• Coefficient of variation, with 36% as the conventional boundary between stable and labile glycaemia. Two people with the same mean and CVs of 22% and 48% are in very different situations and have the same A1c.
• Glucose management indicator, a sensor-derived A1c-like number computed as GMI(%) = 3.31 + 0.02392 × mean glucose in mg/dL [1].

Worked, for a mean of 8.5 mmol/L: convert to mg/dL first, 8.5 × 18 = 153 mg/dL. Then GMI = 3.31 + (0.02392 × 153) = 3.31 + 3.66 = 6.97%. Note that this lands within 0.03 points of the eAG-inverted figure in the accepted answer, which is expected — the two equations were fitted to overlapping data and both are linear in mean glucose.

The critical point about GMI: it is deliberately not called an estimated A1c, because in a substantial minority of people it differs from the measured A1c by more than 0.5 points in either direction, and that discrepancy is reproducible within a person. It reflects the same glycation-gap phenomenon. When GMI and A1c disagree persistently, neither is wrong; you have learned something about that person's glycation.

The obvious limitation of a sensor is that it only knows about the period you wore it. A1c integrates the whole quarter whether you were paying attention or not, which is precisely why it is harder to game and why it remains the trial endpoint.

A specific warning for this population, since the question is nominally about a general topic but the context is weight loss with a GLP-1 receptor agonist.

Several of the confounders in the accepted answer's table are more likely, not less, in someone eating substantially less for months:

• Iron deficiency. Intake of haem iron falls with total intake, absorption is unaffected by the drug but the substrate is reduced, and menstruating women start from a lower reserve. Iron deficiency without anaemia is common and inflates A1c before the haemoglobin ever drops. This is the confounder most likely to be silently present.
• B12 and folate. Reduced intake, plus metformin co-prescription in the type 2 population, plus in some cases prior bariatric surgery. All three depress erythropoiesis.
• Blood donation. Unremarkable and easy to forget, but a donation eight weeks before a draw depresses A1c measurably.

The practical consequence is that a full blood count and iron studies belong in the same tube as the A1c, at least at baseline and annually. Interpreting a series of A1c values without knowing whether haemoglobin and ferritin were stable across the series is guesswork, and this is not a rare edge case — it is the ordinary situation in a person whose intake has halved.

One further note on direction of bias that cuts against the drug rather than for it. If someone becomes iron deficient during treatment, their A1c will drift upwards for haematological reasons while their glucose improves. That is the opposite of the case in the question, and it produces the frustrating pattern of an A1c that refuses to fall despite obvious weight loss and better sensor data. Before concluding that a drug has stopped working, check whether the denominator moved.

None of this is clinical advice. Anaemia in an adult always has a cause, some of the causes are serious and unrelated to diet, and "it is probably my intake" is not a diagnosis anyone should make about themselves.