HbA1c and its lag: what a result three months old is telling you

Glycated haemoglobin is a weighted integral, not a snapshot, and the weighting is the whole of its interpretive difficulty. Erythrocytes survive roughly a hundred and twenty days, glycation accumulates over the life of each cell, and the population of cells in circulation at any moment is age-distributed. The result is that a measured HbA1c reflects approximately the preceding three months, with something like half the value contributed by the most recent thirty days and only a small fraction by the oldest month. A person whose glycaemia changed abruptly six weeks ago will have an HbA1c that is neither their old value nor their new one.

The HbA1c lag, quantified

Glycated haemoglobin accumulates over the lifespan of the erythrocyte, which averages around a hundred and twenty days. Because the circulating population is age-distributed, the contribution to the measured value is weighted towards recent glycaemia: approximately half the value derives from the preceding month, roughly a quarter from the month before that, and the remainder from the two months before that. The commonly repeated statement that HbA1c reflects three months of control is therefore true and misleading in equal measure.

The consequence for monitoring is specific. A person who starts treatment and achieves substantial glycaemic improvement within four weeks will show, at eight weeks, an HbA1c that has captured perhaps two-thirds of the eventual change. A person who stops treatment will show, at four weeks, almost none of the consequence. Panels drawn at these intervals are not wrong; they are measuring a window that includes the period before the change.

Where a shorter window is needed, fructosamine and glycated albumin reflect roughly two to three weeks and are available in most laboratories, though rarely ordered and less well standardised. The Journal notes that the trials in this class scheduled HbA1c at intervals of three months or more precisely because more frequent measurement of a three-month integral yields overlapping windows rather than independent observations.

The eight things that break the HbA1c assay

Iron deficiency raises HbA1c independently of glycaemia, by a mechanism involving altered erythrocyte turnover and glycation kinetics; correcting the deficiency lowers it without any change in glucose. Any cause of accelerated erythrocyte turnover — haemolysis, recent transfusion, treatment of a deficiency anaemia, erythropoietin therapy — introduces young cells and lowers the value. Chronic kidney disease shortens erythrocyte survival and lowers it. Splenectomy raises it by extending survival.

Haemoglobin variants, including the common sickle and haemoglobin C traits, interfere with some assay methods, though not with all: ion-exchange chromatography and immunoassay behave differently and a laboratory should state its method when a variant is known. Pregnancy lowers it. Severe hypertriglyceridaemia and hyperbilirubinaemia interfere with some platforms.

Several of these are common in the population taking these drugs. Iron deficiency in particular is prevalent among people eating substantially less, and it pushes HbA1c in the direction that would make glycaemic control look worse than it is. A rising HbA1c during otherwise successful treatment is a reasonable prompt to check a full blood count and iron studies before concluding anything about glycaemia. That is an observation about assay behaviour and not clinical advice.

What the diabetes programmes reported

The tirzepatide monotherapy trial in type 2 diabetes reported HbA1c reductions of approximately 1.87 to 2.07 percentage points across its dose arms against approximately 0.04 for placebo, from a baseline of around 7.9 per cent.¹ The head-to-head against semaglutide 1 mg reported reductions of approximately 2.01, 2.24 and 2.30 percentage points across tirzepatide doses against 1.86 for semaglutide, from a baseline near 8.3 per cent.²

Three things are worth extracting from those figures for a laboratory-medicine readership. The baseline value governs the achievable reduction — a trial recruiting at 8.3 per cent will report a larger fall than one recruiting at 7.9, and cross-trial comparison without baselines is uninterpretable. The reductions are far larger than the reference change value for HbA1c, so these are unambiguous signals rather than statistical artefacts. And the placebo arms moved barely at all, which tells you something about how stable the analyte is in the absence of an intervention.

Reductions of two percentage points are at the upper end of what any glucose-lowering therapy has achieved, and the Journal reports them as such while noting that they are group means from populations selected for baseline control in a defined range.

The panel after stopping, and its timing

Timing is almost the whole of this. A panel drawn four weeks after a final injection is largely measuring the treatment period, because HbA1c integrates three months and the drug was present for most of them. A panel drawn at twelve weeks reflects the post-cessation period for HbA1c and reflects it fully at sixteen. Fasting glucose responds within days to weeks and is therefore the earlier indicator, at the cost of much larger within-person variation.

The other analytes have their own timescales. Alanine aminotransferase responds over weeks to months as hepatic fat returns with weight. Triglycerides respond quickly and noisily. Creatinine drifts back as lean mass is regained, which means estimated glomerular filtration rate falls during regain for the same non-renal reason it rose during loss. Blood pressure, which is not a laboratory measurement but travels with these panels, reverts over weeks.

The commonest misreading the Journal encounters in correspondence is a person concluding from a reassuring panel at four to six weeks after stopping that the metabolic consequences of cessation are smaller than they were told to expect. At that interval the panel cannot have shown them. The finding at twelve weeks is frequently different, and it is the one worth waiting for.

Analytical testing and clinical testing are different activities

A distinction has to be drawn firmly because the postbag suggests it frequently is not. The four independent testing services this market relies on — Janoshik, Medutest, PeptideMeter and VendorInvestigate — analyse material. They report chromatographic purity, identity by mass, peptide content where it is measured, and in the case of the verification services what could be established about a supplier. A clinical laboratory analyses a person. The two produce documents that superficially resemble each other and answer entirely unrelated questions.

A purity certificate reporting 99.1 per cent for a batch from WWB, CPC or QYB tells you nothing about anybody liver enzymes. A normal panel does not confirm that a vial contained what its label claimed, and an abnormal one does not establish that it did not. Where a person suspects a supply problem, the instrument for that is analytical testing of the material; where a person has an abnormal laboratory result, the instrument is clinical assessment. Substituting one for the other is a reliable way to spend money and learn nothing.

Compounds sold for research use only are not approved for human use in any jurisdiction, and nothing in this department should be read as guidance about using them or about monitoring their use.

How the Journal reports a laboratory figure

Five things accompany a laboratory number in these pages. The units, because international and conventional units differ for several analytes and the same value means different things in each. The reference interval used, with a note where the interval is contested, as it is for alanine aminotransferase. The baseline, because a change of 1.8 percentage points in HbA1c from a starting value of 8.3 is a different claim from the same change from 9.5. The estimand where the figure comes from a trial. And the reference change value where we are discussing an individual delta rather than a group mean.

We also state the assay method where it matters, which is more often than one would like: HbA1c in the presence of a haemoglobin variant, thyroid function in the presence of interfering antibodies, and creatinine measured by enzymatic against Jaffe methods all behave differently, and a comparison across methods is not a comparison.

This is a heavier apparatus than most publications carry and it exists because the alternative, in our experience, is a stream of technically accurate figures that lead readers to conclusions the data does not support. Errors in this apparatus should be reported to standards@compoundjournal.com; the correction log records what came of each one.

What this department is for

The Laboratory Notebook reports what tests measure, how they behave, and what has been found using them. It does not recommend monitoring schedules, interpret readers’ results, or advise on treatment. A laboratory result belongs in a conversation with a clinician who has the rest of the picture, and this publication is emphatically not that conversation.

Two standing notes. Several compounds discussed in these pages are sold for research use only and are not approved for human use in any jurisdiction; the Journal reports on them as commodities and as analytical problems, not as therapies. And where we describe what the pivotal trials monitored, that is reporting on trial protocols and not a template anybody should adopt from a magazine.

Correspondence is welcome at letters@compoundjournal.com. The Journal receives a steady flow of letters containing readers’ own panel results with a request for interpretation, and we do not provide it — not from caution but because a panel without a history, an examination and a reason for ordering it cannot be interpreted by anybody, including us.

The Journal’s position on monitoring is that the panel is the easy part and the interpretation is the hard part, and that almost all of the available effort goes into the first. A baseline panel is worth having because it converts every subsequent result from a population comparison into a personal one. Beyond that, the frequency at which most people are being tested generates flags faster than information, and there is no evidence that it detects anything a symptom-prompted panel would miss.