A1c, and what it is not

What the test is measuring

Glucose in your blood attaches itself slowly and irreversibly to haemoglobin inside your red cells. HbA1c measures the proportion of your haemoglobin that has been sugared in this way. Because a red cell lives roughly three months before it is recycled, the test gives an average of your blood glucose over about that period.

Two consequences fall straight out of that sentence, and both of them matter.

First, it is weighted. The most recent weeks contribute considerably more to the result than the oldest ones, because the cells from three months ago are already being cleared. A very good fortnight moves it a little. A very good quarter moves it properly.

Second, it is a test of your red cells as much as your glucose. Anything that changes how long red cells live changes the result without a single thing having happened to your blood sugar. That is not a technicality; it is the reason a substantial number of people get a result that does not match how they feel, and it is the part almost nobody is told.

The units, because this trips everybody

There are two units in circulation and they are not interchangeable, which produces a reliable weekly confusion in our type-2 and prediabetes circles.

The older DCCT unit is a percentage — you will see results like 6.2 per cent. The IFCC unit, standard across the UK and much of Europe since around 2011, is millimoles per mole — you will see 44 mmol/mol. They describe the same measurement on different scales. Somebody quoting 5.6 and somebody quoting 38 may be describing an identical result.

The thresholds differ by country too, and that is a genuine source of anxiety rather than a trivial detail. In the UK, a result of 48 mmol/mol (6.5 per cent) or above on the appropriate testing is used to diagnose type 2 diabetes, and the band from 42 to 47 mmol/mol (6.0 to 6.4 per cent) is treated as a high-risk or non-diabetic hyperglycaemia range. American practice uses a wider high-risk band, starting at 5.7 per cent. So a result that reads as unremarkable in one country reads as pre-diabetic in another, with no biology having changed. Sam found this out by reading an American forum, and lost a weekend to it.

When you compare your number with someone else’s, check the unit and check the country. Half the alarm in this community evaporates at that step.

Things that make it read wrongly

Mira keeps this list because it comes up constantly. In every case the honest position is that the result should be interpreted cautiously, not that it is worthless.

• Iron-deficiency anaemia. Red cells live longer, accumulate more glucose, and the result reads higher than the true average. Treating the anaemia can make an A1c fall with no change in glucose at all.
• Haemolysis, blood loss, or recent transfusion. Shorter cell life or borrowed cells; the result reads lower or simply becomes uninterpretable.
• Haemoglobin variants. Sickle cell trait, thalassaemia trait and other variants interfere with some assays and not others. This disproportionately affects people of African, Mediterranean, Middle Eastern and South Asian heritage, and it is a good reason to ask which assay your lab uses if your result and your glucose readings disagree.
• Advanced kidney disease. Alters red-cell turnover, and the result becomes less reliable.
• Pregnancy. Red-cell turnover changes across the trimesters; A1c is not the test used for gestational diabetes.

If your A1c and your finger-prick or sensor readings tell different stories, that disagreement is information. Bring both, say plainly that they do not match, and ask whether a fructosamine or a different approach is worth doing. That is a reasonable question, not a challenge.

What the trials showed, in outline

GLP-1 and dual-incretin medicines lower A1c substantially in people with type 2 diabetes, and the effect is one of the better-evidenced things in this whole field.

In SUSTAIN-6 (Marso, New England Journal of Medicine, 2016), semaglutide lowered A1c well below placebo over two years in people with type 2 diabetes and high cardiovascular risk, alongside its cardiovascular findings. In SURPASS-2 (Frías, NEJM, 2021), a head-to-head trial over 40 weeks, all three tirzepatide doses lowered A1c by around two percentage points or a little more, and each was superior to semaglutide 1 mg, which itself produced a reduction close to two points.

Sunil, who runs our reading group, always adds the same caveat and he is right to. These are averages in trial populations who were selected, monitored and supported in ways ordinary care cannot match. The spread around the average is wide, the baseline A1c of the population strongly influences how much a drug appears to lower it, and a bigger number on a chart is not automatically a better outcome for a particular person.

Why it is a bad scoreboard

People arrive in the circles wanting to use A1c as feedback. It is a poor instrument for that, for reasons that have nothing to do with willpower.

It moves slowly by design. Testing it more often than roughly every three months mostly measures noise. It is a single average, so it cannot distinguish a steady middling glucose from a life of spikes and crashes that happen to average out. And in the specific case of these medicines, a falling A1c often reflects several things at once — the drug’s direct effect on insulin and glucagon, slower gastric emptying, changed eating, changed activity — which cannot be separated by the test.

There is also a harder point, and it belongs in this community more than most. Turning a lab value into a personal score is a well-worn route into a miserable relationship with your own body. Priya wrote in her journal about checking a number she could not influence in the timescale she was checking it, and how much better she felt when her clinician agreed to test it twice a year and talk about how she was actually living in between. If that pattern sounds familiar, this guide and support resources are here.

Use it as a slow instrument for a slow question, brought to a conversation with your clinician. That is what it is good at.

If you are not diabetic

A lot of members here do not have type 2 diabetes and get an A1c anyway, either as part of a private screen or because the prescribing service checks one. A few things worth knowing.

An A1c in the high-risk band is a risk marker, not a diagnosis and not a moral event. It moves in both directions over the years in large cohorts, and a meaningful proportion of people in that band never progress to diabetes. It is a prompt for a conversation about how you are being followed, and it is the kind of thing worth repeating rather than acting on once.

Also: if you have no diabetes, your A1c will usually change only modestly on these medicines, because there is much less room to move. That is not a sign the drug is not working. It is simply what a normal number does — very little.

And if you take insulin or a sulfonylurea alongside a GLP-1 medicine, a falling A1c is exactly the situation in which doses of those other drugs may need to come down, because that is where the real hypoglycaemia risk lives. That is a prescriber conversation, and it is one to have early rather than after an episode. There is more in the type 2 summary.