What you need to gather

For each measurement window:

• Your laboratory HbA1c result (in mmol/mol; or in % and convert)
• Your CGM mean glucose for the 90 days that ended on the date of that HbA1c (in mmol/L)

Three pairs is the minimum for a reasonable estimate. Four or five is better. The reason: HGI is reproducible within an individual, but a single pair can be skewed by an unusual glucose period (illness, holiday, infection) or a lab assay running slightly off. Averaging across multiple windows smooths that out.

The calculation, step by step

Step 1, calculate predicted HbA1c (GMI) for each pair

The Bergenstal 2018 formula converts CGM mean glucose into the HbA1c that would be predicted if you were an average glycator:

Worked numbers: a mean glucose of 8.5 mmol/L gives a GMI of 12.71 + (4.70587 x 8.5) = 52.7 mmol/mol.

Step 2, calculate HGI for each pair

HGI is the difference between your measured HbA1c and the GMI:

Continuing the example: if measured HbA1c at the same date was 58 mmol/mol, then HGI = 58 minus 52.7 = +5.3 mmol/mol. Positive = you glycate faster than average.

Step 3, average the HGI values to get mHGI

Add up the HGI values from all your pairs and divide by the number of pairs. That mean HGI (mHGI) is your stable summary statistic.

Step 4, classify

• mHGI < -3 mmol/mol: low glycator
• mHGI between -3 and +3 mmol/mol: average glycator
• mHGI > +3 mmol/mol: high glycator

These thresholds come from the working draft of the Pemberton-Chalew mHGI framework (under peer review; ADA 2026 poster accepted). They produce clinically meaningful separation in the Birmingham cohort. They may shift slightly as the manuscript is finalised. The categories matter more than the exact decimal.

The science behind glycator status, and a methodological correction to a foundational paper

Before working through the example below, the science worth reading: the first paper to systematically test whether glycator status predicts complications was Lachin et al. (2007), a secondary analysis of the DCCT cohort. The paper’s title states that HGI is not an independent predictor of microvascular complications. That title has been cited for nearly two decades as evidence that glycator status does not matter beyond HbA1c. It is a methodological error, and the field has been propagating it.

What the paper actually shows (the finding that holds up)

Adjusting for mean blood glucose (the right comparison), Lachin 2007 showed that HGI predicted both retinopathy and nephropathy. At the same average glucose, fast glycators developed more complications than slow glycators. This is the foundational mean-glucose-adjusted finding of the HGI field, and it cleanly supports the matrix in Part 4.

Where the title’s claim came from (and why it does not hold up)

The “not independent” claim came from a separate model that adjusted for HbA1c when testing HGI’s effect. That adjustment is mathematically incoherent given how HGI is built. HGI is defined as the residual of HbA1c minus GMI:

Adjusting for HbA1c when testing HGI’s effect removes the very variance HGI is mathematically defined to carry. HbA1c absorbs the glycation signal because the glycation signal lives inside HbA1c by construction. This is the textbook Table 2 fallacy (or mediator-as-confounder error): in plain English, the model controlled for the very thing it was trying to measure. HbA1c sits on the causal pathway from (mean glucose plus glycation phenotype) to complications. Adjusting for the mediator nullifies the upstream cause’s coefficient mechanically. It is not a finding about biology; it is collinearity by construction.

The weight of evidence since: HGI is an independent risk factor

Eight subsequent studies, using cohort-, mechanism-, and ethnicity-based methods, confirm that glycation phenotype is biologically real and clinically consequential at the same mean glucose:

• Bovee 2024: HGI independently associated with retinopathy and nephropathy after adjusting for mean glucose (longitudinal T1D)
• Maran 2022: fast glycator phenotype predicts higher microvascular complication rates; phenotype stable over time
• Shah 2024 / 2025: high glycators have higher incident retinopathy at the same HbA1c
• Hempe 2024: provides the biological mechanism (GOP, vitamin C recycling, GLUT-1, G6PDH) for why glycation differs at the same glucose
• Hempe and Hsia 2022: synthesises decades of HGI evidence
• McCarter 2004: biological variation in HbA1c at given mean glucose exceeds analytical noise
• McCarter and Chalew 2024: HGI discordance is structured by ethnicity, not random
• Pemberton 2025: ethnicity HbA1c effect persists at the same mean glucose, in a UK paediatric T1D cohort

Eight of the nine studies cited here confirm HGI matters at the same mean glucose. None refute it.

How science gets to the truth

The Lachin 2007 title was not bad faith; it was a textbook Table 2 fallacy (mediator-as-confounder), a known statistical error that has subsequently been recognised and corrected for across the field. This is exactly how science is supposed to work: a theory is proposed, statistical models can mislead, replication and stress-testing across cohorts and methods correct the error, and the field moves closer to the truth. Calling out the original error plainly, while acknowledging the same paper’s correctly-specified analysis, is what intellectual honesty requires. The matrix in Part 4 rests on the corrected reading and the eight follow-on studies, not on the misleading title.

If you encounter the Lachin 2007 “not independent” claim cited as evidence that glycator status does not matter, this is the correction to share. The original paper’s mean-glucose-adjusted analysis is the finding that holds up.

Common pitfalls and caveats

• Pair the right windows. The 90 days of CGM data must be the 90 days before the HbA1c blood draw, not 90 days starting on that date. The HbA1c reflects the past, not the future.
• CGM coverage matters. The Bergenstal formula assumes you wore your CGM at least 70% of the 90 days. Sparse coverage can bias the mean glucose figure.
• Recent acute illness. An infection, a course of steroids, a holiday with very different glucose patterns can all skew a single pair. This is why averaging across 3 to 4 windows matters.
• Anaemia, haemoglobinopathies, recent transfusion, pregnancy. Any condition that changes red blood cell turnover changes how HbA1c reflects glucose. mHGI calculated during such a window may not reflect your stable glycator status. Discuss with your care team if any of these apply.
• HbA1c near assay limits. Very high or very low HbA1c values approach the edges of where the GMI regression was derived. The estimate is still informative but slightly less reliable.

None of these break the framework. More data points, more comparable conditions, more confidence in your category.