HbA1c, Fasting Glucose or OGTT: Choosing the Test

Three tests can diagnose type 2 diabetes, and most people are only ever offered one of them. Which one depends largely on local practice, on whether fasting is convenient, and on what the clinician happened to tick. It is easy to assume the choice is administrative, because all three appear on guidelines as acceptable routes to the same diagnosis.

They are not interchangeable. A fasting glucose measures the concentration of sugar in blood at a single moment after an overnight fast. An oral glucose tolerance test measures how the body handles a defined load of sugar over two hours. HbA1c measures how much glucose has attached itself to haemoglobin inside red blood cells over the preceding weeks. These are three different physiological quantities, related but not equivalent, and a person can sit above the diagnostic threshold on one while sitting comfortably below it on another.

That discordance is not rare or exotic. When populations are tested by more than one method, a substantial fraction of those identified by any test are identified by only one of them. The practical result is that the test you are given partly determines whether you receive a diagnosis at all, and if so, when.

This piece sets out what each test actually measures, where they disagree and why, and which situations make one clearly preferable to the others.

Three Tests Measuring Three Different Things

Fasting plasma glucose is the oldest of the three and the most direct. After eight or more hours without food, the glucose in blood reflects the balance between what the liver is releasing and what tissues are taking up under basal conditions. Insulin is at its lowest working level, and the main thing being probed is whether the liver’s glucose output is being adequately suppressed. A raised fasting value therefore points particularly towards hepatic insulin resistance.

The oral glucose tolerance test probes a different capacity. A standard dose of glucose is drunk, and blood is sampled at baseline and two hours later. What is being tested is the ability of the pancreas to release insulin quickly enough, and of muscle and fat to take up glucose fast enough, to clear the load in a reasonable time. This is a stress test, and like all stress tests it exposes limitations invisible at rest. Someone whose basal regulation is intact but whose insulin response is slow or blunted can have a normal fasting glucose and a clearly abnormal two-hour value.

HbA1c measures neither a moment nor a response. Glucose attaches non-enzymatically to proteins, and haemoglobin inside red cells is a convenient one to measure because red cells circulate for a predictable period and are easy to sample. The proportion of haemoglobin carrying this attached glucose reflects average exposure over the cell population’s lifespan, weighted towards the more recent weeks because older cells are steadily removed. It is a retrospective average, insensitive to when the glucose was high and insensitive to yesterday’s meal.

Each answers a distinct question. Is basal regulation intact? Is the response to a load adequate? What has average exposure been recently? Those questions have related but non-identical answers.

How Often the Three Disagree

Studies comparing tests within the same individuals consistently find substantial disagreement, and the pattern is systematic rather than random.

The two-hour value from a tolerance test typically identifies the largest number of people with abnormal glucose regulation, including many with entirely normal fasting values. Fasting glucose identifies a partly overlapping group. HbA1c identifies a third group, overlapping both but not contained in either. The overlap between any two is substantial and far from complete, and the group identified by all three is smaller than the group identified by any single one.

Some of this is measurement noise, but most of it is biology. A person whose problem is a delayed insulin response has an abnormal tolerance test and a normal fasting glucose because their fasting regulation is fine. A person with mildly raised glucose across the whole day may have an abnormal HbA1c while never crossing the fasting threshold, because HbA1c integrates exposure that no single measurement captures. Conversely, someone with high post-meal peaks but low values the rest of the time can have a near-normal average and an abnormal tolerance test.

Thresholds contribute too. The diagnostic cut-offs were derived from the level at which the risk of diabetes-specific eye damage begins to rise, and they were derived separately for each test. Because the underlying distributions differ between populations, cut-offs that align well in one population align less well in another.

Fasting Requirements and Practical Compliance

The requirements look small on paper and shape real-world results considerably.

A fasting glucose requires no caloric intake for at least eight hours, which in practice means an early morning appointment after an overnight fast. Water is permitted. Coffee, even black, is not, and this is a common source of invalid samples because people do not consider it food. Where the fast is incomplete, the value is uninterpretable rather than merely slightly raised, since a recent meal can raise glucose well above the diagnostic threshold in someone entirely healthy.

Compliance falls further when appointments are hard to schedule early. People who work shifts, who care for others in the morning, or who must travel some distance are systematically disadvantaged by a test that requires arriving unfed. Screening that is inconvenient is screening that some groups do not receive.

The tolerance test asks considerably more. It requires the same overnight fast, then a baseline blood sample, then drinking a sweet solution, then remaining at the facility without eating, smoking or walking about for two hours, then a second sample. The drink is unpleasant for many, and a proportion of people feel nauseated or occasionally vomit, which invalidates the test. Preparation also requires adequate carbohydrate intake in the preceding days, because a low-carbohydrate diet reduces glucose tolerance temporarily and can produce a falsely abnormal result.

HbA1c requires nothing. A sample can be taken at any time of day in any state of feeding, which is why it can be added to a blood draw already happening for another reason. That convenience explains most of its adoption in routine screening, and it also explains why HbA1c catches people who would never have returned for a fasting appointment.

What an OGTT Reveals That Others Miss

The tolerance test is the most burdensome of the three and remains valuable precisely because it stresses a system that looks normal at rest.

The typical sequence of deterioration in type 2 diabetes begins with rising insulin resistance compensated by increased insulin output, so both fasting glucose and average exposure remain normal. What fails first, in many people, is the early insulin response: the rapid burst of insulin release that normally occurs in the first minutes after glucose enters the circulation. Without that early burst, the glucose load clears slowly, producing a high two-hour value while fasting regulation remains adequate for years.

This means a normal fasting glucose and a normal HbA1c do not exclude impaired glucose tolerance, and the group with isolated post-load abnormality is not small. It carries elevated cardiovascular risk and a high rate of progression, so identifying it is not merely academic.

It also has a particular role in ambiguous cases. Where a person has a strong family history, features of insulin resistance, or a screening result sitting just below a threshold, a tolerance test can resolve the question the simpler tests left open.

Its main weakness is reproducibility. Repeat tolerance tests in the same person, days apart, can differ enough to move someone across a diagnostic threshold. Physical activity in preceding days, illness, sleep, stress, and carbohydrate intake all influence the result. This poor reproducibility is a genuine limitation and one reason the test is not used routinely for population screening despite its sensitivity.

Populations Where HbA1c Underperforms

HbA1c assumes that red cells live a predictable length of time and that haemoglobin is normal. Where either assumption fails, the number stops meaning what it appears to mean, and this affects more people than is generally appreciated.

Anything that shortens red cell survival lowers HbA1c independently of glucose, because cells are removed before they have accumulated their expected glucose burden. Haemolytic anaemias, significant blood loss, and the period following a transfusion all do this. So does treatment of iron or vitamin deficiency anaemia, because the marrow floods the circulation with young cells that dilute the older, more glycated population. Advanced kidney and liver disease alter red cell turnover in ways that can move the value in either direction.

Anything that lengthens red cell survival raises HbA1c. Untreated iron deficiency is the most common example and works in the opposite direction to its treatment, which is a source of genuine confusion when a patient is tested before and after iron replacement.

Haemoglobin variants form a separate category. Some variants interfere with particular measurement methods, giving results that are wrong for analytical rather than biological reasons, and the direction of the error depends on which method the laboratory uses. Laboratories are generally aware of the variants common in their population and select methods accordingly, but a result that conflicts sharply with the clinical picture should prompt the question.

There is also a persistent finding that average HbA1c differs slightly between ancestry groups at the same measured glucose level, for reasons that are not fully explained and may involve red cell characteristics. The difference is small relative to the diagnostic range, but at a threshold it can matter.

Gestational Screening as a Special Case

Pregnancy is the one setting where the tolerance test remains clearly dominant, and the reasons illustrate the wider principles well.

Pregnancy alters glucose handling by design. Placental hormones produce progressive insulin resistance across the second and third trimesters, which is a normal adaptation that increases nutrient availability to the fetus. Gestational diabetes is what happens when the pancreas cannot compensate for that imposed resistance. Because the problem is specifically one of compensating for a load, it is a post-load abnormality by nature, and fasting values often remain unremarkable.

HbA1c performs poorly here for a different reason. Red cell turnover increases in pregnancy and blood volume expands, both of which lower HbA1c relative to actual glucose exposure. The condition also develops over weeks, whereas HbA1c reflects the preceding weeks, so the measurement lags the physiology at exactly the point where timely detection matters. Testing typically occurs in a window when intervention can still change outcomes, and a lagging indicator is poorly suited to that.

The tolerance test in pregnancy also uses different thresholds from those applied outside it, and in some protocols a different glucose load and additional sampling times. Those thresholds were set against risks to the pregnancy rather than against long-term eye damage, which is why a value that would be unremarkable in a non-pregnant person can warrant intervention in a pregnant one.

The practical consequence is that a normal HbA1c should not be treated as reassurance about gestational diabetes, and that people offered a tolerance test in pregnancy are receiving the test best matched to the question.

Confirming a Diagnosis Before Treatment

A single abnormal result is a screening finding rather than a diagnosis, and the distinction has practical consequences.

The standard approach is confirmation on a second sample, ideally taken on a different day. Where the initial test was HbA1c, confirmation may use HbA1c again or a glucose measurement. Where the initial abnormality was a fasting glucose, a repeat fasting sample or an HbA1c serves. Two different tests both above threshold is generally accepted as sufficient without a repeat, since the probability of two independent tests being spuriously abnormal is low.

The exception is the person with unmistakable symptoms, meaning marked thirst, high urine output, weight loss and a clearly raised random glucose. In that situation the diagnosis is not in doubt and waiting for confirmation delays treatment that is needed immediately.

Confirmation matters because of what a diagnosis brings with it. It carries medication, monitoring, dietary changes, implications for insurance and driving in some jurisdictions, and a durable label. Analytical imprecision and biological variation together are large enough that a value sitting marginally above a threshold has an appreciable chance of falling below it on a repeat, particularly for glucose-based tests with their higher day-to-day variability.

Where results sit in the intermediate range that indicates elevated risk without meeting diabetes criteria, the response is different in kind. That finding identifies a state that frequently progresses but frequently does not, and the interventions that alter its course are largely behavioural. Repeat testing at intervals, rather than immediate treatment, is the usual approach, and the interval is chosen against how close the value sat to the threshold.

The useful framing is that these three tests answer three questions, and the question you need answered should drive the choice. For unselected screening where getting tested at all is the main hurdle, HbA1c wins on practicality. Where basal regulation is the concern, or where a haemoglobin or red cell condition makes HbA1c untrustworthy, a fasting glucose is the more honest measurement. Where the suspicion is early failure of the insulin response, or where the setting is pregnancy, the tolerance test remains the one that sees what the others cannot. Knowing which of the three you were given, and what it therefore did not look at, is the part that most often goes unmentioned.