A meal raises blood glucose. The pancreas releases insulin, and insulin moves that glucose into cells. Where insulin sensitivity is good, the rise is moderate, the insulin response is proportionate, and glucose settles back toward baseline without much drama.

Where insulin sensitivity is reduced, the same meal requires more insulin to achieve the same effect. A larger insulin response does not always stop neatly at baseline. Glucose can overshoot downward, producing a trough somewhere between two and three hours after eating.

The trough is the part you feel. Attention narrows, work that felt easy at 11am takes visible effort, and the body starts asking for fast fuel. At 3pm that request is persuasive.

This has been measured directly. In a study of 1,070 participants who ate 8,624 standardised meals while wearing continuous glucose monitors, the size of the glucose dip at two to three hours predicted self-reported hunger, how soon people ate again, and how much they ate over the following 24 hours.¹ The dip was a better predictor of all three than the peak was. The individual correlations were modest, and they held in the same direction across both a UK and a US cohort.

That finding reframes the usual advice. Most conversations about post-meal energy focus on the spike. The evidence points at what happens on the way back down.

A standard panel usually measures fasting glucose, and sometimes HbA1c. Both are useful. Neither is designed to see an afternoon crash.

Fasting glucose describes one moment, taken first thing in the morning after an overnight fast. The 3pm crash happens seven or eight hours later, after food. The test and the symptom are not observing the same event. HbA1c has the opposite limitation: it averages roughly three months of glucose exposure, so a daily pattern of sharp rises followed by sharp falls can average out to a reassuring number.

Continuous monitoring shows how much this misses. When researchers tracked people who were classified as normoglycaemic by standard measures, those participants still spent around 15% of monitored time in the prediabetic glucose range, and about 2% in the diabetic range.² The same work found that people fall into distinguishable patterns of glucose response, and that two people with identical fasting results can behave quite differently after the same meal.

So a normal fasting glucose is genuine information. It rules out a great deal. It does not describe what your metabolism does for the rest of the day.

This is the part that matters most for anyone whose bloods keep coming back clear. It is also the most common version of the conversation I have with people who have been told, accurately, that nothing is wrong.

Glucose is held stable by insulin. Early in the development of insulin resistance, the pancreas compensates by producing more insulin, and that extra effort succeeds. Glucose stays inside the reference range precisely because something is working harder to keep it there. Measure only glucose and you measure the outcome while the effort stays invisible.

The Whitehall II cohort followed 6,538 British civil servants without diabetes, 505 of whom were diagnosed over a median of about ten years. Working backwards from diagnosis, HOMA insulin sensitivity declined steeply across the five years beforehand, while fasting glucose showed its sharp quadratic rise only in the final three.³ Insulin sensitivity was the earlier signal. A parallel analysis in the same cohort found these trajectories differ between ethnic groups, with South Asian participants showing a faster decline in insulin sensitivity than white participants.⁴

The observation is not new. Longitudinal work in Pima Indians described insulin resistance preceding fasting hyperglycaemia decades ago, and noted it was not explained by obesity alone.⁵

The practical consequence: fasting insulin is an ordinary blood test that is rarely ordered. Paired with fasting glucose it produces HOMA-IR, a calculation validated against euglycaemic clamp measurements in 1985 and still in routine research use.⁶ The original authors were explicit that the model has limited precision in any single individual, so it is best read alongside symptoms and other markers rather than treated as a verdict on its own.

Surrogate markers are convenient and they disagree with each other more than is usually acknowledged. One study comparing people with prediabetes against matched controls found fasting insulin and HOMA-IR clearly separated the two groups, with median HOMA-IR of 3.12 against 1.21. The same study found no correlation between the triglyceride to HDL ratio and HOMA-IR within its prediabetic group.⁷

That is a useful caution. A lipid ratio can suggest the pattern, and it is not a substitute for measuring insulin directly. It also explains why a single reassuring marker rarely settles the question when symptoms are persistent.

The order you eat food in matters more than it sounds like it should. In a randomised crossover trial, 16 healthy adults ate identical meals in different sequences. Putting the vegetables and protein before the carbohydrate blunted the glucose response, and did so with lower early insulin release and greater GLP-1 stimulation.⁸ Sixteen people is a small trial. Treat the size of that effect as provisional and the direction as consistent with a wider literature.

Then there is what you do afterwards. Breaking up prolonged sitting with two minutes of walking every 20 minutes suppressed post-meal glucose compared with sitting throughout, in a randomised crossover study of sedentary adults.⁹ A short walk after lunch costs nothing and needs no equipment, which makes it the easiest thing on this page to test on yourself.

Sleep is the one that surprises people who are working long weeks, and it has the most striking data behind it. When healthy young adults spent four nights at 4.5 hours in bed rather than 8.5, fat cells sampled afterwards needed nearly three times the insulin concentration to reach half-maximal signalling, and whole-body insulin sensitivity fell in parallel.¹⁰ Calorie intake and activity were controlled throughout, so the change came from the sleep restriction itself.

None of this needs a diagnosis to be worth doing. Measurement earns its place by answering the two questions guessing cannot: whether the pattern is actually there, and whether what you changed moved it.

Insulin is one explanation and it is not the only one. A mild dip in alertness after lunch is a normal circadian feature, and plenty of people experience it without any metabolic abnormality at all. What makes it worth investigating is severity, persistence, and whether it is new.

Other patterns produce a similar afternoon picture. Low free T3 tends to bring cold intolerance and slowed thinking alongside the fatigue. Low ferritin produces exercise intolerance and shortness of breath on stairs. A flattened cortisol curve tends to show a hard time waking rather than a specifically post-lunch collapse. Chronic under-eating at lunch will do it mechanically. So will genuine sleep debt, with or without the insulin effect described above.

These overlap, and they are distinguishable on testing. The reason to measure rather than guess is that the correct next step differs entirely depending on which one is operating.