Educational

Two people get an A1C result of 7.2% on the same afternoon.

The first has been running between 6 and 9 mmol/L most days, with occasional excursions. The second has been swinging from 3 to 16 and back, several times a week, with a handful of overnight lows nobody caught.

Same number. Same clinical impression, if the number is all anyone looks at. Very different three months.

What A1C measures

Glucose in the bloodstream attaches to haemoglobin in red blood cells, and it stays attached for the life of the cell. Measure what fraction of your haemoglobin is glycated and you get a picture of average glucose over the cells' lifespan — roughly 8 to 12 weeks.

It's a genuinely good test. It doesn't need fasting, it isn't thrown off by what you ate that morning, and it predicts long-term complication risk better than any single glucose reading. That's why it became the standard.

One nuance that gets missed: it isn't an even average across three months. Because red cells are constantly being replaced, recent glucose counts for more — roughly half the result comes from the previous month. A good four weeks before a test genuinely does move the number.

What it can't see

An average discards the shape of the data, and the shape is where a lot of the day-to-day experience lives.

Variability. A steady 8 and a swing between 3 and 13 can produce the same mean. There's a growing body of evidence that glucose variability contributes to complication risk independently of average glucose, and the mechanisms proposed — oxidative stress, endothelial damage — are plausible even where the evidence is still developing.

Hypoglycemia. Lows pull the average down. An A1C that improved partly because of overnight lows looks like progress and isn't.

Timing. Post-meal spikes and fasting levels can both be elevated, or one and not the other, and A1C reports the blend. The intervention for each is different.

When the number itself is unreliable

A1C depends on red blood cells behaving normally, and several common conditions mean they don't.

Anything that shortens red cell lifespan — haemolytic anaemia, sickle cell trait and other haemoglobin variants, recent significant blood loss, advanced kidney disease — tends to produce a falsely low A1C, because the cells haven't been around long enough to accumulate glucose. Iron deficiency anaemia can push it the other way. Pregnancy changes red cell turnover substantially. A recent transfusion makes the result close to meaningless.

This matters because someone with a haemoglobin variant can be told their control is excellent while their actual glucose says otherwise. If your A1C and your meter consistently disagree, that discrepancy is worth investigating rather than explaining away.

Where glycemic load comes in

A1C is an outcome. It tells you what happened and gives you no lever to pull.

Glycemic load is an input. It's a property of the meal in front of you, available before you eat it, and it acts on precisely the thing A1C can't show — the size of the rise after each meal. A day of low-GL meals is a day with smaller peaks, and enough of those days is what eventually moves the average.

They aren't competing measurements. One is the steering, the other is the odometer.

Time in range fills the gap

Continuous monitoring introduced a metric that captures what A1C misses: the percentage of time spent between 3.9 and 10.0 mmol/L, generally with a target above 70%, alongside the percentage spent below range.

Two people at 7.2% might be at 75% and 55% time in range. That's a real difference in daily experience and, most likely, in risk — and it's invisible to the blood test.

A note on the CGM readout: the glucose management indicator a CGM reports is an estimate derived from sensor data, and it frequently differs from a lab A1C by a few tenths. Neither is wrong. They're measuring different things.

How much change counts as change

The assay itself has measurement error. Repeat the same sample and results can differ by a couple of tenths of a percentage point, and different labs vary further.

So a move from 7.2% to 7.0% is not clearly an improvement — it's inside the noise. A move from 7.2% to 6.6% is a real signal. This matters because people adjust their whole approach on the strength of a change that never happened, and then adjust it back three months later.

Testing more often than every three months rarely helps either, since the measurement window itself is three months long. Two consecutive readings six weeks apart are largely made of the same data.

What we'd actually suggest doing with this

Keep the A1C. It remains the best available predictor of long-term risk and the number your care team will reasonably want.

Add something that shows shape — a CGM if you have access, or structured finger sticks before and 90 minutes after your most repeated meals if you don't. Then use glycemic load on the meals those readings flag.

That sequence is the useful one: the average tells you whether things are broadly working, the shape tells you where the problem is, and the composition of a specific meal is the only part of it you can actually change tonight.

And if your A1C is where you want it but your days feel like a rollercoaster, believe the rollercoaster. The number was never designed to describe how a Tuesday went.