Blood Sugar Markers

HbA1c

Haemoglobin A1c, glycated haemoglobin

How much sugar has attached to the haemoglobin inside red blood cells. Those cells live about 120 days, so the measurement carries a memory. Recent weeks are represented in every cell in the sample, and older weeks only in the ones that have lasted. So the result reflects average blood sugar over roughly the previous 2 to 3 months, not any single moment. The Body Guide sets out why the weighting falls out of the age spread of your red cells, under Blood cells, in the Body Guide.

This is the marker the NHS uses to diagnose and monitor type 2 diabetes.

No fasting, any time of day. That is its practical advantage over a glucose, and it is the reason it can be run on a sample taken whenever suits you. It is read differently in pregnancy, after a recent transfusion, and in conditions that shorten red cell lifespan, where guidance says to use a glucose instead.

Where the numbers come from

The World Health Organization sets 48 mmol/mol (6.5%) as the level at or above which diabetes is diagnosed. In someone without symptoms a second sample is required to confirm it, so a single reading is never diagnostic on its own.

What the World Health Organization does not do is interpret anything below that. Its position, set out when it accepted the marker for diagnosis and unchanged since, is that there was insufficient evidence to make any formal recommendation about HbA1c levels below 6.5%. It noted that values just below the line may indicate intermediate hyperglycaemia. It reported what others had proposed, without adopting a band of its own. So the familiar band underneath the diagnostic line is not the World Health Organization's, and describing it as theirs is a common misattribution.

The band came from a UK expert group working out how to implement the World Health Organization's decision, and NICE adopted it. NICE describes 42 to 47 mmol/mol (6.0 to 6.4%) as non-diabetic hyperglycaemia, meaning a raised risk of developing type 2 diabetes and not a diagnosis of it. It is the figure the NHS Diabetes Prevention Programme is built around, and Diabetes UK use the same one. Diabetes Canada use it too.

The American band starts 3 units lower, and that is the whole disagreement. The American Diabetes Association sets 5.7 to 6.4%, which is 39 to 47 mmol/mol. The top of the 2 bands is the same and the diagnostic line is the same. The bottom is not. So 39, 40 and 41 mmol/mol are prediabetes in America and normal in the UK, outside the high-risk band, outside the prevention programme and outside the register a GP keeps. A single result can carry 2 different labels depending on which country's article you read it against.

One detail in the small print of both. Neither body's paired figures are exact conversions of each other. Run the conversion properly and 42 mmol/mol is 5.99%, 47 is 6.45% and 48 is 6.54%, while 5.7% is 38.8 and 6.4% is 46.4. Each body has rounded outward to give itself round numbers in its own preferred unit. So the pairs are approximations, and the boundaries sit slightly differently depending on which scale your report uses.

Below 42 there is no committed figure anywhere. No body has set a target, an optimal value or a reference range for someone who does not have diabetes, and the American label of normal below 5.7% is the floor of a diagnostic category, not a recommendation. It is one of the markers where an optimal number gets quoted most confidently and sourced least often.

What a change between 2 results has to be before it means anything

This is the part that decides whether a second test has told you something. Combine the biological swing in a healthy person with the imprecision of a good analyser, and 2 HbA1c results from the same person have to differ by roughly 4 to 5 mmol/mol (based on the % scale) before the difference is more likely real than noise and depends on the method. That figure is a floor. It assumes the same lab, the same method and a person whose glycaemia is actually steady. Figures lower than that are in circulation, and they come from taking a variation calculated on the percentage scale and applying it to a result in mmol/mol, which is the trap set out under Units.

Set that next to the band. Non-diabetic hyperglycaemia runs from 42 to 47, which is 5 units wide. So the amount by which a result has to move to mean anything is the whole width of the band it is being read against. A shift from 43 to 46 is not a deterioration. It is the same result twice.

There is a second figure, and it is much larger. Follow people through routine care instead of a controlled study and the total variation in repeat HbA1c results is several times the lab figure, and it rises the higher someone's average sits. Most of that extra is real change in blood sugar and not measurement error, along with samples going to different labs on different methods. The 2 numbers answer different questions. Roughly 4 to 5 mmol/mol is the answer to whether a difference is measurement noise. The larger figure is the answer to whether your next result will look like this one, and it will not.

One sample, 2 labs, 2 different answers. Certification allows each lab to sit within 5% of the reference value, and that 5% is set on the percentage scale. Converted across, it is about 3.6 mmol/mol either side of the diagnostic line. Two labs that both pass can therefore report the same sample as 44.4 and 51.6. Measured instead of allowed, the European quality-assessment scheme that covers thousands of labs found 1 lab in 20 failing its quality criterion when it first ran. On 2024 data that is down to about 1 lab in 50. The criterion is not a simple distance from the true value. It allows a total error of 5 mmol/mol and asks how often a lab would fall outside that, so a lab can fail it while most of its results sit well inside 5 mmol/mol. Whole countries and whole manufacturers still fail it, up to 1 country in 10 and up to 3 manufacturers in 10 on the 2024 data, and variation between labs remains the largest single component of the total error. That is the error that shows up when a private result is compared with a GP result.

The same HbA1c does not mean the same average blood sugar

Two people can have identical HbA1c and quite different average glucose, and the gap is not small. Work out each person's true average blood sugar from long runs of their own glucose measurements, and 2 people at the same HbA1c can differ by more than 3.3 millimoles per litre. For scale, 3.3 millimoles per litre is the difference between a fasting glucose of 5.0 and one of 8.3.

The main reason is red cell age. Label red cells and track them and the average age of the cells in circulation runs from about 38 to about 60 days in people with nothing wrong with their blood at all. Older cells have had longer to accumulate sugar. So a person whose cells last longer reports a higher HbA1c at the same glucose, and a person whose cells turn over faster reports a lower one. Modelling each person's own red cell turnover instead of using the population average cuts the error in estimating their average blood sugar by more than half. There is a second, separate mechanism as well, which is that the concentration of glucose inside the red cell is not the same fraction of the concentration outside it in everybody.

The gap is consistent within a person. Measure it, wait the better part of a year, measure it again, and the direction and size hold. That consistency is the strongest argument that it is a real property of the person and not a run of bad analyses.

Whether the gap predicts anything on its own is unsettled, and the argument is worth following. On one side, a high gap travels with more retinopathy, more kidney disease and more large-vessel disease at the same HbA1c. The death rate is raised at both ends and not one. In the largest trial where this was examined after the fact, the cardiovascular benefit of intensive glucose lowering appeared in people with a low or middling gap and not in those with a high one. The excess deaths in the intensive arm sat almost entirely in the high-gap group. On the other side, the objection is structural and it is a fair one. The gap is calculated as the difference between a measured HbA1c and an HbA1c predicted from a glucose measurement. So a noisy or unrepresentative glucose automatically manufactures a gap, and the group labelled high-gap is partly just the group whose glucose happened to read low that day. Different cohorts have also published different prediction equations, so the size of a person's gap depends on which one is applied to them. A 2026 review argues for sidestepping the arithmetic entirely. It would measure the reversible, loosely attached fraction of glycated haemoglobin directly and use its ratio to the stable fraction, which appears to flag roughly 1 sample in 7 as one where the HbA1c does not reflect the expected blood sugar. It is not standardised and it has no outcome data behind it yet.

Our own reading of that, stated as ours. The gap is real, it is person-specific and it is stable, and none of that has been turned into anything you can act on. What it does establish is that the population range fits this marker poorly. Formally, the spread between people is several times the spread within one person. When that is true a published range is a weak comparator and your own previous result on the same method is a strong one. That is an argument for tracking and not for a target. It is also why we would rather you had 3 HbA1c results a year apart on the same method than one result read against a table.

Ethnicity moves this number, and the effect runs both ways in the same population

At the same measured blood sugar, HbA1c runs higher in Black participants than in White ones, and the size of the gap depends entirely on where in the range you look. In people with normal glucose tolerance it is about 0.13 to 0.21 percentage points, which is roughly 1.4 to 2.3 mmol/mol. In the raised band it is about 0.26 to 0.30, and in people with diabetes about 0.47, which is roughly 5 mmol/mol. So the gap widens as glucose tolerance worsens, and the figure most often quoted, 0.4 percentage points or roughly 4 mmol/mol, comes from work in people with type 1 diabetes whose authors said plainly that it could not be generalised to anyone below the diagnostic line. For anyone reading an HbA1c below the diagnostic line, the smaller figures are the right ones.

The mechanism looks like red cell and haemoglobin biology and not worse blood sugar control. Glycated albumin and fructosamine measure glycation of things other than haemoglobin, and they show no such difference at the same blood sugar.

In the UK, the clearest South Asian figure comes from a Leicester population given full glucose tolerance testing alongside HbA1c. After allowing for fasting and post-load glucose, HbA1c ran about 0.2 percentage points higher, which is roughly 2 mmol/mol. That is a smaller gap and it rests on weaker ground. Fasting and post-load glucose are not a 3-month average, so the part that is a glycation difference and the part that is a real difference in blood sugar are less clearly separated. Nobody has yet done the continuous-monitor version of that study in South Asian people, and until they do the figure should be read as provisional.

Sickle cell trait runs the other way and more than cancels the difference at the bottom of the range. Carrying it lowers HbA1c by about 3 mmol/mol at the same blood sugar, which is larger than the 1.4 to 2.3 gap in people with normal glucose tolerance, and about 8% of the African American participants carried it in the work where this was measured. Prevalence varies widely across African populations and runs considerably higher in parts of West and Central Africa. So both statements are true at once, they point in opposite directions, and they are present in the same population. Several of the genetic variants known to shift HbA1c independently of glucose are red cell genes and not sugar-handling genes, and their frequencies differ markedly between ethnic groups.

No body has changed the threshold for anyone. The stated reason is that the risk of complications at a given HbA1c appears to be similar in Black and White populations, which is the argument against moving the line. The American guidance adds a point we would repeat and not improve on, which is that ethnicity is a poor stand-in for a genetic variant that could be measured directly.

The threshold is the same for everyone, and what changes is who gets tested

The UK applies the same HbA1c diagnostic threshold to every ethnic group. There is no ethnicity-adjusted cut-off in NICE guidance, in World Health Organization guidance or in NHS practice, and 48 mmol/mol means the same thing whoever the sample came from. What does change is who is offered a test and how early. NICE starts risk assessment at 40 in the general population and at 25 in people of South Asian, Chinese, African-Caribbean and Black African background, and it drops the body mass index that prompts a blood test from 30 to 23 for South Asian and Chinese people. So ethnicity moves the door into testing and not the line drawn once you are through it.

Type 2 diabetes also arrives earlier. In UK primary care the average age at diagnosis is 58 in White European people, 48 in African-Caribbean people and 46 in South Asian people. Average body mass index at diagnosis is lower as well, 32.5, 31.1 and 29.2 in that order, and more than a third of South Asian people are diagnosed below a body mass index of 30 against about a quarter of White European people. The physical picture that prompts a test in a White European person can be absent in a South Asian person who already has the condition.

Prevalence is the figure most often overstated. Diagnosed type 2 diabetes ran at about 5.0% in White, 7.7% in Asian and 5.6% in Black adults registered with London general practices. Once age, sex and deprivation are accounted for, the odds come out about 2.4 times higher in Asian people and about 1.65 times higher in Black people than in White people. The raw gap and the adjusted gap are different quantities, and the 3 to 5 times figure in general circulation is the raw kind. Adjusted for those things the gap is smaller, and it is still large.

The lower body mass index thresholds are blunter than they are usually described. NICE does not publish a figure for each group. It publishes a single pair, overweight from 23 and obesity from 27.5, and applies that same pair to South Asian, Chinese, other Asian, Middle Eastern, Black African and African-Caribbean people together. The evidence underneath it says those groups are not alike. Work in English primary care put the body mass index carrying the same type 2 diabetes risk as 30 in White people at about 23.9 for South Asian, 26.6 for Arab, 26.9 for Chinese and 28.1 for Black participants. NICE has not adopted that detail, so the same number is doing very different work depending on who it lands on.

There are no ethnicity-specific waist thresholds in UK guidance at all. NICE took waist measurements in centimetres out of its obesity guideline and replaced them with a waist-to-height ratio of 0.5, stated to apply to both sexes and all ethnicities. The 90 cm figure quoted for South Asian men comes from the International Diabetes Federation's metabolic syndrome definition, an international consensus and not UK guidance, and that definition sets no African figure at all and tells you to use the European figure instead. So no validated waist number exists anywhere for Black African or African-Caribbean people.

None of that is on a blood panel. Where fat sits in the body is measured by imaging and a tape measure and not by any blood marker, and Blood sugar and metabolism, in the Body Guide covers what differs there.

Haemoglobin variants, and the 2 different problems they cause

Carrying an unusual haemoglobin can wreck an HbA1c result, and it does so by 2 separate routes that need separating because they have different fixes.

The first is the assay. Several methods separate haemoglobin by electrical charge or by chromatography. A variant carrying a different charge can be counted as glycated when it is not, or be missed entirely. Whether this happens depends on the exact analyser and, on some of them, the software version. The body that certifies these assays publishes a table of which methods are affected by which variant, updated as methods change. The honest summary is that most current platforms handle the common traits and a minority do not. Sickle cell trait, haemoglobin C, E and D traits and a raised proportion of fetal haemoglobin are the ones that appear on it. The fix here is a method that does not suffer the interference, and the result is a number that was simply wrong, and not a number that was right about the wrong thing.

The second is red cell survival, and no change of method fixes it. In conditions where red cells are destroyed early, including sickle cell disease and thalassaemia, not the traits, the cells never live long enough to accumulate their share of sugar, so the HbA1c reads low and it reads low correctly. The measurement is accurate and the inference from it is not. Guidance in both the UK and the US says to use a glucose in that situation instead.

Glucose-6-phosphate dehydrogenase deficiency belongs in the second group. It shortens red cell life and lowers HbA1c at the same blood sugar. It runs in the same populations where HbA1c is otherwise raised, so it works against the ethnicity effect and does not explain it, and people carrying it can have real problems with blood sugar that their HbA1c does not show.

This is not rare in the UK. Across UK Biobank and a British Pakistani and Bangladeshi cohort, about 1 in 7 Black UK males and about 1 in 63 Asian UK males carry a deficiency allele, against fewer than 1 in 10,000 White males. Fewer than 1 in 50 of those men have it recorded anywhere in their health records. Those counts cover the 2 most common alleles only, so they are a floor, and the Asian figure rests on cohorts that are mostly Indian, Pakistani and Bangladeshi.

The variant is not the same in every group. People of African background mostly carry the milder version, in which only the oldest red cells are vulnerable. People of South Asian, Middle Eastern and Mediterranean background mostly carry the Mediterranean variant, which leaves less enzyme behind and can take red cells of any age.

The effect on HbA1c is large. Male carriers average about 10 mmol/mol below non-carriers while their actual blood sugar is barely different, and 10 mmol/mol is the distance between 48 and 38. Men who carry an allele are diagnosed with type 2 diabetes about 4 years later than men who do not. Women carrying a single copy sit between the 2, because of which X chromosome each cell switches off, so this is not a male-only effect.

UK guidance already says not to use HbA1c for diagnosis where red cells are being destroyed. Between episodes there is nothing to notice, so that exclusion almost never reaches the people it was written for. An HbA1c that looks comfortable alongside symptoms, strong risk factors or a raised glucose is a reason to read the glucose rather than the HbA1c.

Low iron pushes this number up, and that matters more than it sounds

Iron deficiency raises HbA1c without blood sugar having moved, and correcting the iron brings it back down on its own. The direction is the one that causes trouble, because it pushes a result up into the raised band and not down out of it.

The size of the effect is the part usually overstated. The best-controlled comparison, run on 2 different assay principles side by side so that assay interference could be ruled out, puts it at 3 to 5 mmol/mol in iron-deficiency anaemia. That is enough to move 39 to 42, or 45 to 48, but it sits at about the same size as the amount a result has to move before the move is real, so it is not clearly above the noise. It tracks the severity of the anaemia, and mild anaemia has not been shown to have a significant effect. Pool the studies of what happens when the iron is replaced and the estimate loses statistical significance, with an enormous spread between them. Whether iron deficiency without anaemia moves HbA1c outside pregnancy has not been settled either way.

The practical consequence is that an HbA1c in the 42 to 47 band alongside a low ferritin is a different finding from the same HbA1c with normal iron, and repeating it after the iron is corrected is the way to separate the 2. B12 and folate deficiency are described as working the same way, by lengthening the average life of the red cell, though the evidence there is mechanistic and not measured. Iron, in the Body Guide sets out why ferritin falls long before a blood count shows anything, which is the reason the 2 belong on the same sample.

The rest of what moves it, and what does not

Anything that shortens the average life of a red cell lowers the number. Haemolysis of any cause, recent blood loss, a recent transfusion, and treatment with erythropoietin or with intravenous iron all do this. The effect of a transfusion is to drag the result towards the donor's blood sugar and away from yours. Removal of the spleen lengthens red cell life and raises it, and an enlarged spleen shortens it and lowers it. Ribavirin has been reported to lower it. In advanced kidney disease and on dialysis the number understates blood sugar, and the reason is a combination of shortened red cell life and treatments given for the anaemia. HIV infection and some of the drugs used to treat it also cause it to understate. Chemically modified haemoglobin in kidney failure can raise it on charge-based methods specifically, which is a method problem and not a biological one.

In pregnancy it runs below the non-pregnant range in both early and late pregnancy, falling in the second trimester and rising again in the third. NICE says not to use HbA1c routinely to assess blood sugar control in the second and third trimesters, and to use glucose criteria instead.

What does not move it is worth stating because it is asked about constantly. High-dose vitamin C and vitamin E at the doses people actually take have no clinically relevant effect. Aspirin at an ordinary daily dose has been shown to raise HbA1c by a small amount that is not clinically relevant, and on other methods it makes no measurable difference at all. And biotin, which does distort a long list of hormone tests, does nothing here at all, because no HbA1c method uses the capture chemistry that biotin jams.

Low HbA1c is not automatically better, and this is the honest version

The assumption that lower is better within the normal range does not survive contact with the mortality data. Across general populations followed for a decade or more, the relationship between HbA1c and death from any cause is a U and not a line. The lowest risk sits at roughly 36 to 38 mmol/mol. Risk is raised above about 46, which is unsurprising, and it is also raised below about 31, which is not what anybody expects.

What it means is a different question and the answer is probably not what it looks like. People with a very low HbA1c have, on average, larger red cells, more anaemia, higher liver enzymes, lower cholesterol and lower body weight. A low HbA1c also predicts being admitted with liver disease. That pattern says the low number is a marker of something else going on and not a cause of anything. Adjusting for the things that travel with a low result does not make the association go away, so they do not fully explain it either. No single cause of death accounts for it. There is no trial evidence and no genetic evidence, and no guideline body treats a low HbA1c as something to act on.

Two limits on all of that. The excess sits at values well below anything a healthy person is likely to produce, and the effect between roughly 31 and 46 is flat and not sloping. And none of the cohorts behind it are British, so there is no UK figure.

What the raised band predicts, and what it does not

Being told you are prediabetic sounds like being told you are on a path, and the evidence is more equivocal than the label. Of people with an impaired fasting glucose, about 2 in 3 do not have diabetes 10 years later, and between a third and 6 in 10 revert to normal blood sugar on their own within 5 years. The same population data that produced the U-shaped mortality curve show neither half of the raised band carrying any excess death rate at all over more than a decade, while both diagnosed and undiagnosed diabetes clearly did.

Against that, the marker does predict progression, and it predicts it better than a fasting glucose does. The reasonable reading is that the band identifies a group at raised risk of developing diabetes. The risk within the band is modest and reversible, and the word prediabetes oversells it. The European epidemiology group that objected to the lower American boundary made the same point about the glucose version of the band, and the glucose entry sets out where UK guidance ended up on that, which is not where you would expect.

In the post

HbA1c is one of the more dependable markers on a posted sample, and the reason is structural. It is not a substance floating in plasma that the cells can consume. It is a chemical modification of the haemoglobin locked inside the red cell, and it stays attached for the life of the cell.

The one real hazard is heat. No assay method holds a whole blood sample steady beyond 3 days once the temperature is at or above 30 degrees. Refrigerated, the windows run to 2 or 3 weeks. At ordinary room temperature the manufacturers' own windows differ substantially between platforms, so the figure that matters is the one validated on the analyser your sample actually reaches. Which way a warm sample drifts depends on the method, and on several of them it drifts downwards, which is the falsely reassuring direction. A hot car, a hot postbox and a sample posted on a Friday are the risk, not the transit itself, and collecting Monday to Thursday and posting the same day is what the instruction exists for.

What this test cannot tell you

It cannot tell you what your blood sugar does during a day, because it is an average and averages hide shape. Two people with the same HbA1c can have completely different patterns of rise and fall, and nothing on this measurement separates them. The Body Guide covers what continuous monitoring adds and what it does not.

It also cannot be fixed by one careful morning, which is its main advantage over a glucose and the reason it is the marker the NHS built its diabetes pathway around.

Finger-prick devices that give an HbA1c on the spot are a separate question from a finger-prick sample posted to a lab. Independently evaluated against reference methods, most of the point-of-care analysers on the market failed to meet the certification standards they carried certificates for. That testing was done under ideal conditions. Diabetes UK's position is that a finger-prick HbA1c should not be used for diagnosis and must be confirmed against a lab result.

Units

The UK reports HbA1c in mmol/mol and the US reports it as a percentage. They are the same measurement expressed 2 ways, which is why every threshold is given both ways instead of one being converted into the other.

There is a trap in converting between them that catches professionals. The 2 scales do not share a zero. The percentage scale has an offset built into it, so 0 mmol/mol converts to 2.152% and not to nothing. The consequence is that a percentage change on one scale is not the same percentage change on the other. A variation of 1% expressed in the percentage scale is about 1.5% expressed in mmol/mol. Anyone quoting a figure for how much HbA1c varies, or how precise an assay is, has to say which scale they mean. A figure lifted from an American source and applied to a UK result in mmol/mol will understate the variation by about a third.

Glucose

Fasting glucose, random glucose

The concentration of sugar in blood at the moment of sampling. It is one number from one moment, and it moves with sleep, illness, stress, exercise and the previous day's food.

Fasting glucose and random glucose are the same measurement and different tests. The lab does not run 2 assays. What changes is whether you had eaten, and that changes which thresholds the number can be read against and how much it is worth. A glucose taken after an overnight fast reflects how well the body holds sugar steady at rest. That is a question with published answers. A glucose taken at any other time reflects what you ate and when, and nobody has published an answer to that. What the test is called matters less than what you actually did. A number produced from a properly fasted sample can be read against the fasting thresholds whatever the request was called, and a number produced from a fed sample cannot be, whatever it is called.

Fast for 10 to 12 hours, water only, if you want the fasting thresholds to apply. This is the only marker where fasting changes which published thresholds the number can be read against. Everywhere else a fast just tidies the number up. That is why the panels carrying it ask for one.

What a random glucose can establish, and what it cannot

It can establish one thing. A random glucose at or above 11.1 millimoles per litre meets the diagnostic criteria for diabetes on its own, in someone who also has the classic symptoms. Those are excessive thirst, passing large amounts of urine and unexplained weight loss. Without those symptoms the same number needs a second abnormal test to confirm it.

Below that it establishes nothing formal, and this is the part that surprises people. There is no recognised impaired random glucose band. The World Health Organization does not define one. Neither does the American Diabetes Association, and neither does the lab medicine guidance both bodies endorse. A random glucose of 6.5 is not a diagnosis, is not prediabetes, and is not a normal result either, because no criteria exist to call it any of those things. Equally, a normal random glucose rules nothing out, because there are no criteria under which it could.

What it is good for is triage, and it is better at that than its reputation. A single random glucose above 5.6 is more strongly associated with undiagnosed diabetes than any single traditional risk factor, and the association climbs steeply with the number. But measured against a full glucose tolerance test, a random glucose finds a little under 3 in 4 of the people it should and wrongly flags almost a third of the people it should not. That is a useful signal and a poor test. The honest summary is that a high random glucose is a reason to get a fasting glucose or an HbA1c, and a normal one is not a reason to relax.

Where the numbers come from

On a fasting sample the World Health Organization sets 7.0 millimoles per litre as the diabetes threshold, again with a second sample required where there are no symptoms, and calls 6.1 to 6.9 impaired fasting glucose.

Three bodies, 3 different lower bounds, and the UK has the lowest of them. American guidance works in milligrams per decilitre. The American Diabetes Association draws its band at 100 to 125 mg/dL, which is 5.6 to 6.9 millimoles per litre, against the World Health Organization's 110 to 125 mg/dL, or 6.1 to 6.9. NICE is lower than both. Its own definition of being at high risk of type 2 diabetes is a fasting plasma glucose of 5.5 to 6.9. All 3 agree exactly on the diabetes threshold, at 126 mg/dL or 7.0 millimoles per litre. The whole disagreement is at the bottom.

So a fasting glucose of 5.8 is normal by the World Health Organization's criteria, prediabetes by American criteria, and high risk in the eyes of the guideline a UK GP actually works to. That is not an error in any of them. It is 3 organisations drawing a line at different points on the same continuous relationship.

Two things about that are worth having straight. The American and NICE lines are not really different, because 100 mg/dL converts to 5.55, so 5.6 and 5.5 are the same boundary printed 2 ways. And it means the UK did not hold the conservative line here. When the American threshold was lowered from 6.1 to 5.6 the European diabetes epidemiology group formally objected, arguing that it created a pandemic of impaired fasting glucose without improving prediction of either diabetes or cardiovascular disease, and recommended keeping 6.1. In Denmark the change moved the proportion of adults meeting the criteria from about 1 in 8 to nearly 2 in 5. NICE then went slightly lower still.

Set that next to the HbA1c side and UK guidance points in opposite directions on the same question. On HbA1c, NICE has the highest lower bound of the bodies that publish one, at 42 where America sets 39. On fasting glucose, NICE has the lowest. The same guidance is the most cautious about calling a raised HbA1c a risk and the least cautious about calling a raised glucose one. Nobody has reconciled that, and it is a real reason the 2 markers disagree about the same person.

There is a further wrinkle inside American guidance itself. The American Diabetes Association's clinical standards print 5.6 as the lower bound, while the lab medicine guideline the same body co-authors prints the World Health Organization's 6.1. Both are current.

What counts as fasting, and the 2 bodies differ here too

The American definition is no calories for at least 8 hours, with water permitted. It is a floor with no ceiling. The World Health Organization does not set a general figure for a standalone fasting glucose at all. Where it does specify a fast, which is the preparation for a glucose tolerance test, it gives a window of 8 to 14 hours with water permitted. The reason a window has a ceiling is that a long enough fast eventually pushes glucose back up as the body switches to making its own, though how much and after exactly how long has not been quantified in a way anyone should put a number on. The practical point is that 10 to 12 hours satisfies everybody and a 20-hour fast is not a better test.

What a difference between 2 results has to be before it means anything

Fasting glucose swings much more in a healthy person than HbA1c does. Putting the biological variation together with a good analyser, 2 fasting glucose results from the same person have to differ by roughly 0.8 to 1.0 millimoles per litre before the difference is more likely real than noise.

The impaired band runs from 6.1 to 6.9, which is 0.8 wide. So the amount a fasting glucose has to move to mean anything is the entire width of the band. A result of 6.2 and a later result of 5.6 are, statistically, the same morning twice.

This shows up directly when people are retested. Of those whose fasting glucose was at or above the diabetes threshold on one morning, only about 7 in 10 were still above it a fortnight later. Requiring 2 abnormal fasting results instead of one cuts the apparent rate of undiagnosed diabetes by about a quarter. That is why the guidelines all insist on a second sample, and it is not a formality.

This is the one number on the menu where the tube decides the answer

Glucose is not a fixed feature of blood the way haemoglobin is. It is fuel, and the cells in the sample carry on burning it after the needle comes out. In a tube where the cells are still sitting in the plasma, glucose falls by roughly 5 to 7% an hour at room temperature. Starting from a normal fasting level that is around 0.3 millimoles per litre every hour, and more than that from a higher starting point. Nothing about the result looks wrong. It is simply lower than the blood was.

Everything in the guidance for this marker is about that one problem. The current lab recommendation is a tube containing granulated citrate buffer, which acidifies the blood and stops glycolysis almost immediately, and if that is not available then the tube goes into iced water and the cells are spun out within 15 to 30 minutes. The same guidance says specifically that fluoride tubes, which are what most UK labs have used for decades, should not be relied on for this, because fluoride blocks the pathway too far down. Glucose falls at or near the same rate in a fluoride tube as in a plain one for the first hour, and full inhibition is delayed for up to 4. Citrate tubes are far better and they are not free. They carry a positive bias of their own, reading measurably higher than a chilled heparin sample, enough on its own to raise the apparent rate of diabetes in a population. The different citrate brands are not interchangeable with each other either, so serial results from different tube types are not comparable. A serum gel tube is perfectly good for glucose if it is spun within the hour and worthless if it is not, which is why studies of it look contradictory until you check whether it was centrifuged before storage.

The scale of this is not theoretical. One regional health service shortened the time samples spent in transit and started separating them sooner. The average fasting glucose across its whole adult population rose by about 0.6 millimoles per litre, and the proportion of results above 5.55 millimoles per litre went from about 1 in 10 to about 1 in 4. Their HbA1c barely moved over the same period, which is what proves the shift was in the handling and not in the people. Separately, correcting for room-temperature glycolysis in one maternity service would have roughly doubled the rate at which gestational diabetes was diagnosed.

So the practical position for anyone reading a glucose result, here or anywhere, comes to 3 things. The number depends on what tube it was drawn into and how long the cells sat in the plasma. Neither of those facts is printed on a report. And delay runs the number one way only, which is downwards, while the tube chemistry that prevents the delay pushes it slightly the other way. A glucose is at its most trustworthy analysed the same day it is taken. Of the 2 markers on a blood sugar panel it is the fragile one and HbA1c is the dependable one.

A finger-prick glucose and a venous glucose are the same fasted and not the same after a meal

Fasted, the difference between capillary and venous plasma glucose averages 0.1 millimoles per litre, which is nothing. During a glucose tolerance test the capillary number runs about 1.7 millimoles per litre higher, which is 20 to 25%. The tissues have taken their share out of the blood by the time it reaches a vein. The World Health Organization publishes separate post-load thresholds for capillary samples for that reason and treats venous plasma as the standard.

Separately from the sample type, handheld meters are not accepted for diagnosis by anybody, including for diabetes in pregnancy, and the stated reason is the variability between devices. A meter is for managing known diabetes, not for finding it.

HbA1c and glucose disagree often, and neither is the tiebreak

Applied to the same population, the fasting glucose criteria flag nearly twice as many people as being in the intermediate band as the HbA1c criteria do, and the 2 groups only partly overlap. Among people with impaired glucose tolerance on a formal tolerance test, only about a third have an HbA1c in the raised band. Which test you happened to have done is therefore a large part of what determines whether you are told anything at all.

On reproducibility the ranking is clear and useful. HbA1c repeats most closely, fasting glucose next, and the 2-hour tolerance test worst by a wide margin. On predicting what happens next, the 2 are similar for developing diabetes, and HbA1c is the stronger of the 2 for cardiovascular disease and for death from any cause, to the point that once HbA1c is accounted for, a fasting glucose adds nothing for those outcomes. Why raised blood sugar damages arteries in the first place is set out under Heart and blood vessels, in the Body Guide. For someone testing privately and not on the day, that combination of arguments points at HbA1c. The reason to have a glucose alongside it is that the 2 do not measure the same thing and the discordant cases are real, not that the glucose refines the HbA1c.

What sits beyond a fasting glucose, including things we do not sell

The formal test is a glucose tolerance test, where a measured 75g glucose drink is given and blood taken at 2 hours. At or above 11.1 millimoles per litre meets the diabetes threshold and 7.8 to 11.0 is impaired glucose tolerance. It is the only test that identifies impaired tolerance at all, and a fasting glucose on its own misses roughly 30% of previously undiagnosed diabetes. It is also the least reproducible of the 3 tests. It cannot be done by post because it needs timed sampling. A GP can arrange one.

The direction of travel is towards reading that test at 1 hour rather than 2. In 2024 the International Diabetes Federation proposed 8.6 millimoles per litre at 1 hour as intermediate hyperglycaemia and 11.6 as diabetes, on the argument that the 1-hour value outperforms fasting glucose, HbA1c and the 2-hour value for predicting both diabetes and its complications. No other body has adopted it. The thresholds are already being contested. Work in African populations suggests a lower diabetes cut-off is more appropriate there. In Brazil the 1-hour test doubled detection of unknown diabetes, while the proposed staging did not behave consistently. It is a real and promising change that is not settled, and that is how we would read it.

Fasting insulin and HOMA-IR

This is the measurement that would catch the problem earliest, and it is the one almost nobody runs. Insulin climbs for years to hold glucose steady. So a normal glucose and a normal HbA1c can sit on top of a pancreas working far harder than it should. A fasting insulin paired with a fasting glucose gives a score called HOMA-IR, and it picks that up while both of those numbers are still comfortable. The physiology, and the shape of the curve that makes it possible, are set out under Blood sugar and metabolism, in the Body Guide. What follows is how to read the number if you get it elsewhere.

We do not sell it, and the reason is the same one that runs through the glucose entry. Insulin has to be spun out of contact with the red cells within about an hour. Red cells carry an enzyme whose job is to break insulin down, and once they are damaged it does exactly that inside the tube. At a degree of haemolysis too slight to see, insulin is already down by about 14%. Heavily haemolysed samples lose over 90% of it. The loss only runs one way, which is downwards, so a mishandled sample returns a reassuring number instead of an alarming one. A posted self-collected sample cannot control the tube chemistry, the time to separation, the temperature and the handling all at once. Insulin needs all 4.

The formula, and the trap inside it

HOMA-IR is fasting insulin multiplied by fasting glucose and divided by 22.5, with insulin in milliunits per litre and glucose in millimoles per litre. That divisor only works for those units. With glucose in the American milligrams per decilitre the divisor is 405, and with insulin in picomoles per litre it is 135. All 3 are in circulation, so a HOMA-IR is meaningless without knowing which units produced it. Using the millimole divisor on a milligram glucose inflates the answer 18-fold.

There is also a second, better version. The original 1985 model came with a hand formula as an approximation. Its authors later replaced it with a computer model that accounts for how the liver and the rest of the body differ, for what happens at higher glucose levels, and for the proinsulin circulating alongside the insulin. Oxford, where the model comes from, states that the hand formula is not appropriate for use with the insulin assays now in use, and gives the calculator away free for non-commercial use. The number a private lab or clinic hands you is almost always the hand version. The 2 are not comparable with each other. The computer model is scaled so that a healthy young adult comes out at about 1.0, while the hand version puts the same person nearer 2.

Where the numbers come from, and this is the short part

Nowhere. No professional body publishes a HOMA-IR threshold or an optimal fasting insulin. The strongest statement on it comes from the American Diabetes Association's own working group on insulin measurement. It concluded that these measures are done for research only, that they are comparable within a single study and not between studies, and that there are no criteria by which an individual could be classified as insulin sensitive or insulin resistant. The International Diabetes Federation lists HOMA-IR among research measures and says insulin resistance is not an essential requirement for the diagnosis it was defining. NICE and the World Health Organization do not address it at all, which for a UK reader means it is not in any NHS pathway.

What exists instead is a scatter of thresholds from individual populations. Published figures from primary sources run from about 1.7 to about 3.8, and they move with ethnicity, age and sex. A HOMA-IR of 2.5 is insulin resistant against some of those figures and normal against others, from the same blood. The figure most often quoted in consumer material is an optimal fasting insulin below 5. It is not from a professional body, a cohort or a published paper. It appears on testing company and wellness sites and nowhere traceable. It is repeated everywhere and sourced nowhere.

The assay problem, which is the reason for all of the above

Insulin immunoassays are not standardised against each other. Run one sample across the commercial methods and the spread between them is about 24% at the midpoint and reaches 66% at the extreme. A shared calibrator was tried twice and failed to close it both times. There is still no listed higher-order reference material for insulin, which there is for the closely related C-peptide. Two of the most widely used analysers differ by about 11% on the same population. The American national health survey publishes conversion equations so its own figures can be compared across the years it changed platform.

There is a subtler problem on top. Most of these assays cross-react by more than 40% with one of the partly processed forms of proinsulin. That form rises disproportionately in insulin-resistant people, so the bias runs upward in exactly the group being measured most often. The computer model accounts for circulating proinsulin. The hand formula does not.

How much a single result can tell you

Less than most people assume, and the original authors said so first. Their own paper put the precision of the estimate at about 31% and stated plainly that this limits its use.

Measured properly in healthy people, fasting insulin varies by about 25% within the same person from week to week, against a spread between people of about 52%. The consequence is arithmetic and stark. A single fasting insulin locates your own true average to within roughly half. It takes 7 samples to pin it down to within 20%, and 25 samples to get to within 10%.

HOMA-IR inherits that and adds to it, at about 27% within the same person. The people who measured it concluded that it is highly variable within an individual in a way that limits the value of single measurements. Put through the arithmetic, a HOMA-IR has to roughly double, or nearly halve, before the change is real. That has been measured directly as well as calculated, and the 2 agree.

The instructive comparison is a score called QUICKI, which is built from the same 2 numbers and varies by about 4% within a person instead of 27%. The whole difference is that it takes logarithms instead of multiplying. So a good part of HOMA-IR's instability is the shape of the formula and not the biology. That is an argument for reading it on a log scale, or for not reading single values at all.

What to do instead, and you may already have it

Two substitutes need no insulin at all. The ratio of triglycerides to HDL cholesterol, and the triglyceride-glucose index, which multiplies a fasting triglyceride by a fasting glucose and takes a logarithm. Both are computed from numbers on an ordinary lipid profile plus a fasting glucose. Anyone who fasted for a panel carrying lipids and a glucose already has what is needed.

They are not a poor man's version. Measured against the reference method that actually infuses insulin and glucose and watches what happens, the triglyceride-glucose index performs at least as well as HOMA-IR. In the largest direct comparison it came out slightly ahead. The group whose work defined insulin resistance testing recommended triglycerides or the triglyceride to HDL ratio over insulin themselves, for the specific reason that insulin assays are not standardised. And because triglycerides and glucose are both standardised, traceable measurements, neither substitute carries the assay problem, the haemolysis problem or most of the variability.

The same caveats apply to them as to HOMA-IR, and 2 of them matter. The triglyceride to HDL ratio fails badly in people of Black ethnicity. At the threshold usually quoted it misses most insulin-resistant Black adults, because triglycerides tend to run lower in that group at the same degree of insulin resistance. And the published cut-offs for both are as population-specific as HOMA-IR's. There is an added trap in the triglyceride to HDL thresholds, which differ depending on whether the triglyceride is in millimoles per litre or milligrams per decilitre, a factor of nearly 2.

Our own reading, stated as ours. The idea behind measuring insulin is right and the measurement is not ready. If you want to track insulin resistance, the useful move is to pick one of the substitutes, compute it from a panel you were having anyway, and watch it in yourself over years. If you want the insulin itself, a GP can request it and most private labs sell it. The number is worth having as a trend in you and not as a score against somebody else's threshold.

Fructosamine and glycated albumin measure glycation of proteins in plasma and not of haemoglobin, over roughly 2 to 3 weeks instead of 2 to 3 months. Because they do not involve red cells at all, they sidestep every red cell problem the HbA1c entry describes. That is precisely why they were the control measurements in the ethnicity comparisons. They are available privately in the UK and we do not currently offer either. The trade-off is a shorter memory and far less outcome evidence behind them.

In the post

Of everything on a blood sugar panel, glucose is the marker most changed by the journey, and the tube paragraph sets out why. Delay moves it downwards, and how much depends on the tube chemistry and on how long the cells stayed in contact with the plasma and not on anything about you.

Units

The UK reports glucose in millimoles per litre and the US in milligrams per decilitre. Unlike HbA1c there is no offset, so the conversion is a straight division by 18.016. What there is instead is rounding. 126 mg/dL is 6.99 millimoles per litre and 100 mg/dL is 5.55, and both bodies print the rounded figures, which is why the same boundary can appear as 5.55, 5.5 or 5.6 depending on where you read it.