Guide series, Part 1 of 5
HbA1c, the Three-Month Average That Is Not Average
HbA1c is treated as a simple summary of glucose. It is not. Two people with identical mean glucose can produce HbA1c values 10 mmol/mol apart. That gap is biology, not measurement error, and it changes what your TIR target should be.
Ask Grace
Want to understand why your HbA1c and your CGM average disagree? Ask Grace.
What HbA1c actually measures
HbA1c is the proportion of haemoglobin (the oxygen-carrying protein in red blood cells) that has glucose chemically attached to it. Glucose binds to haemoglobin slowly and irreversibly across the lifespan of a red blood cell (around 120 days). The result is a weighted average of glucose exposure over roughly three months, with the most recent few weeks counting most.
Think of a school report that counts the whole year but weights the summer term most heavily. Autumn is in there; it just carries less. That is the shape of the HbA1c window.
This is why HbA1c is reported in millimoles per mole (mmol/mol) or as a percentage. It is a measure of how much haemoglobin has been glycated, not a direct measure of glucose. The translation from one to the other depends on biology, and that biology is not the same in everyone.
The unit conversion: 48 mmol/mol = 6.5%. 53 mmol/mol = 7.0%. 58 mmol/mol = 7.5%. The international consensus 70% TIR target corresponds approximately to an HbA1c of 53 mmol/mol (around 7.0%) for an average glycator on a Zone P device. That target comes from Battelino et al., Diabetes Care 2019;42(8):1593-1603 (doi 10.2337/dci19-0028), Grade D.
HbA1c is not always reliable. Certain conditions can distort HbA1c independently of glucose.
- Iron deficiency anaemia: can falsely raise HbA1c, making glucose control appear worse than it is.
- Vitamin B12 or folate deficiency: can falsely raise HbA1c through altered red blood cell turnover.
- Haemoglobin variants (HbS, HbC, HbE; including sickle cell trait): can raise or lower HbA1c depending on the laboratory assay method, producing results that do not reflect glucose at all.
- Chronic kidney disease: shortened red blood cell lifespan can falsely lower HbA1c.
- Haemolytic anaemias: any condition that accelerates red blood cell destruction shortens the glycation window and can falsely lower HbA1c.
- Recent blood transfusion: introduces donor haemoglobin with a different glycation history, disrupting the 90-day window.
These are analytical interferences, not the biological glycation variation (HGI) discussed later in this guide. If your HbA1c and GMI disagree, the first clinical step is to rule out these analytical causes before attributing the gap to glycation biology. If any of these conditions apply to you, discuss with your diabetes care team whether your HbA1c can be interpreted at face value before using your personalised target in Part 4 (Lenters-Westra et al, 2025, Diabetic Medicine).
The DCCT backbone, why HbA1c is still indispensable
The Diabetes Control and Complications Trial (DCCT, 1993) randomised 1,441 people with Type 1 diabetes to intensive or conventional insulin therapy and followed them for 6.5 years. Intensive therapy reduced sustained three-step retinopathy progression by 76% in the primary prevention cohort, who entered with no retinopathy, and by 54% in the secondary intervention cohort, who entered with mild retinopathy. Across the trial, microalbuminuria fell by 39%, macroalbuminuria by 54%, and clinical neuropathy by 60%.
The relationship between HbA1c and complication risk was continuous, with no threshold below which risk stopped falling. Those are relative reductions, and relative reductions sit on top of absolute event rates that are worth seeing: sustained retinopathy progression in the primary prevention cohort ran at 1.2 events per 100 patient-years on intensive therapy against 4.7 per 100 patient-years on conventional therapy, while severe hypoglycaemia ran the other way, at 61 per 100 patient-years against 19 (Nathan 2021, the DCCT/EDIC review by the trial’s principal investigator).
A 76% reduction is real, and it is a reduction from a small number to a smaller one, bought at a threefold rise in severe hypoglycaemia. The observational follow-up (EDIC) later reported a 50% reduction in advanced kidney disease, meaning an eGFR below 60, in the original intensive group. That is a finding from decades of follow-up rather than from the 6.5-year randomised period, and it is a different endpoint from the albuminuria measures above. The microvascular benefit stayed measurable for around a decade after the two groups’ HbA1c had converged (Nathan 2021), and the cardiovascular benefit is still measurable at 30 years (Braffett 2025).
This is Grade A evidence. It established HbA1c as the gold standard outcome measure in diabetes and has not been overturned. Any TIR framework still has to account for what HbA1c uniquely captures.
Lachin 2022 (DCCT/EDIC analysis): when HbA1c and estimated TIR were directly compared as predictors of retinopathy in the DCCT dataset, HbA1c was the stronger predictor. TIR added very little above HbA1c. This is the result that stops “TIR will replace HbA1c” being a clean argument.
Same glucose, different HbA1c
If HbA1c were a pure function of mean glucose, two people with the same average glucose would have the same HbA1c. They do not. The classic observation, established by Hempe, McCarter and colleagues across the early 2000s, is that two people with identical mean glucose can record HbA1c values that differ substantially, in some reported observations by up to around 10 mmol/mol.
McCarter (2004) used DCCT data to show that this between-person variation is real biology. The within-person biological variation (CV around 3.6%) actually exceeds analytical imprecision, confirming HGI is real biology, not measurement noise. The between-person variation is much larger again and reproducible across years. People are not noisy; they are different.
Two kinds of discordance. When HbA1c and CGM mean glucose disagree, the cause falls into one of three categories (Lenters-Westra et al, 2025): (1) analytical interference (haemoglobin variants, iron deficiency, renal disease; see the warning in the section above), (2) biological glycation variation (HGI, the subject of this guide), or (3) CGM wear and calibration issues. The clinical workflow is: rule out analytical causes first, then assess glycation phenotype, then check CGM coverage. This guide majors on category 2, but categories 1 and 3 must be excluded with your care team before the glycation interpretation applies.
This stable per-person tendency to glycate faster or slower than average has a name: the haemoglobin glycation index (HGI).
HGI = Measured HbA1c minus predicted HbA1c (where predicted HbA1c, also called GMI, is calculated from mean CGM glucose using the Bergenstal 2018 formula: GMI mmol/mol = 12.71 + 4.70587 x mean glucose mmol/L). A positive HGI means you glycate faster than the average; a negative HGI means slower.
Three glycator phenotypes
Across populations of people with Type 1 diabetes, HGI distributes across a wide range. The Pemberton-Chalew framework groups people into three working categories based on their mean HGI across multiple visits (mHGI, the average of HGI from 4 to 6 paired HbA1c and 90-day CGM windows, collected over at least 9 months)[1]. The boundary between those categories, plus or minus 3 mmol/mol, is GNL modelling and is Grade E: no published paper states it, and that includes the ADA 2026 abstract from the same group. What the abstract does establish is that a person’s glycator phenotype is stable enough across repeated measurements to be worth categorising at all, which is what makes any boundary useful. Part 4 sets out the reasoning behind where this one sits, and the hub notes carry the full provenance for each figure.
Low glycator (mHGI < -3 mmol/mol)
Your HbA1c runs lower than your CGM mean glucose would predict. Same glucose, lower HbA1c. You may achieve good complication protection with a lower TIR target than the population consensus.
Average glycator (mHGI -3 to +3 mmol/mol)
Your HbA1c and glucose match the standard population relationship. The international 70% TIR target was modelled around your biology.
High glycator (mHGI > +3 mmol/mol)
Your HbA1c runs higher than your CGM mean glucose would predict. Same glucose, higher HbA1c. The standard 70% TIR target may not be enough; you may need a higher TIR to achieve the same complication protection as an average glycator.
The biological mechanism, why some people glycate faster
The mechanism is well characterised, largely through Hempe and colleagues at LSU Health Sciences Center.
The three biological features that set your glycation rate
Three biological features influence individual glycation rate:
- GLUT-1 transporter density on red blood cells. Red blood cells take up glucose passively via GLUT-1 transporters. The density of these transporters varies between individuals and is partly genetic. More transporters means more glucose entering the cell at a given blood glucose concentration, and more substrate available for glycation.
- The glucose oxidative pathway (GOP) and vitamin C recycling. Glucose inside the red blood cell can either enter the glycation reaction or be diverted into the pentose phosphate / GOP pathway. The GOP pathway also helps recycle vitamin C, which protects haemoglobin from glycation. Faster GOP flux means less glucose available for glycation; slower GOP flux means more.
- G6PDH enzyme activity. Glucose-6-phosphate dehydrogenase is the rate-limiting enzyme of the pentose phosphate pathway. Its activity influences how much glucose is diverted away from glycation. G6PDH activity has known genetic variants and varies between ethnic groups.
These are largely outside behavioural control. They are stable, trait-like properties of your red blood cell biology. This is why HGI is reproducible within an individual across years, and why it has a substantial heritable component.
The ethnicity effect, what Pemberton 2025 found
Pemberton, Uday, Krone, Fang and Chalew (2025, BMJ Open Diabetes Research and Care) studied 168 children and young people with Type 1 diabetes in Birmingham, UK, with paired laboratory HbA1c and 90-day CGM mean glucose data. After adjusting for mean blood glucose, CGM use, insulin delivery method, and socioeconomic deprivation, the Black ethnic group had HbA1c approximately 4 mmol/mol higher than White or South Asian peers.
This is not a measurement artefact and it is not explained by differences in technology access or care. On average across this Birmingham cohort, participants in the Black ethnic group showed faster glycation for the same glucose exposure than White or South Asian peers. This is a population-level mean difference with substantial within-group variation; it does not predict any individual’s glycation rate. McCarter and Chalew’s 2024 analysis using the DCCT cohort plus a separate non-Hispanic Black dataset reaches the same conclusion.
An honesty note on sickle cell trait and how laboratory assays read it
An intellectual honesty note on HbS trait. Sickle cell trait (HbS) is more prevalent in Black populations and can affect HbA1c assay results depending on the laboratory method used (some assays read falsely low in the presence of HbS). The Pemberton 2025 study adjusted for mean blood glucose, so the +4 mmol/mol finding reflects a glucose-independent biological difference in glycation. However, the possibility that assay-level interference from haemoglobin variants contributes to ethnic HbA1c differences in other cohorts using different laboratory methods should not be dismissed. This is an area where individual verification with your care team matters: if you carry a haemoglobin variant, your HbA1c may need interpreting through that lens as well as the glycation biology lens.
The treat-to-target trap: the same Pemberton 2025 cohort showed the Black group experienced significantly more time below 3.9 mmol/L and below 3.0 mmol/L. The likely explanation: clinical teams treat all patients to the same HbA1c target, which forces more aggressive insulin dosing in people whose biology produces higher HbA1c for the same glucose. The result is more hypoglycaemia, not better outcomes.
Treating everyone to the same HbA1c number ignores this. Recognising glycator status changes what good care looks like. The Via Negativa lens: remove the universal target before it creates harm. What you subtract (the one-size-fits-all assumption) matters more than what you add.
Why both HbA1c and TIR are needed, and neither alone is enough
HbA1c captures the biological consequence of glucose exposure for that individual: how much haemoglobin (and, by reasonable extension, other proteins relevant to complications) has actually been glycated. TIR captures the glucose exposure itself: how much time was spent in different ranges, and (with derivative metrics) how variable that exposure was.
If HGI were zero for everyone, TIR alone would be enough. It is not. Lachin’s 2022 finding that HbA1c outperforms estimated TIR as a predictor of retinopathy is exactly what you would expect when individual glycation biology contributes meaningfully to who develops complications. The implication is not “stop measuring TIR”; it is “interpret TIR through the lens of your glycator status”.
That interpretation is what Part 2 (calculate your status) and Part 4 (your personalised target) are about.
Notes
Note [1], collected from the section above
- Lachin (2007), using DCCT data and the correctly-specified mean-glucose-adjusted analysis, showed that high glycators had higher rates of retinopathy and nephropathy at the same mean blood glucose. (The same paper’s title-line claim that HGI was “not independent” came from a separate model that adjusted for HbA1c when testing HGI’s effect, a Table 2 fallacy because HGI is constructed as the residual of HbA1c minus GMI; Part 2 walks through this in detail.) The HGI signal has since been replicated in modern cohorts: Maran (2022), Bovee (2024), Shah (2024/2025), Hempe (2024), McCarter and Chalew (2024), Pemberton (2025). Eight of the nine studies cited here confirm; none refute. In Taleb’s framing, if you are a high glycator, the population average is not your probability; your individual risk accumulates along your own trajectory, not the ensemble’s. ← back to text
Part 1 of 5
HbA1c, the Three-Month Average That Is Not Average