A symptom is not the beginning of a problem. It is the point at which a problem became loud enough to notice — and for most chronic conditions, that point arrives years after the biochemistry started drifting. We can now demonstrate that rather than assert it.
Somebody gets a headache most afternoons. They take ibuprofen and the headache goes.
That is a reasonable thing to do, and I want to say clearly that I am not against it. But notice what has happened: the signal has been switched off and nothing has been asked about what was generating it. Dehydration, poor sleep, a magnesium-poor diet, jaw clenching under stress, screen posture, caffeine timing — or, very occasionally, something that needs a scan.
The ibuprofen works equally well in every one of those cases. That is precisely the problem with it as a strategy.
Symptoms are late. They are the point where a drifting system finally crosses a threshold you can feel, and the drift is usually measurable long before.
For years this was something people in my field asserted and could not really demonstrate. That has changed, and the demonstration did not come from us.
Delphi-2M, published in Nature in 2025, was trained on 400,000 UK Biobank participants and externally validated on 1.9 million Danish records. It forecasts more than a thousand diseases — often years ahead — because health trajectories contain signal about what is coming.
MILTON, from AstraZeneca, predicts 3,213 diseases from ordinary blood biomarkers, identifying incident cases that were undiagnosed at recruitment. And it largely outperformed available polygenic risk scores — your bloods predicted your future health better than your genome, in a cohort of half a million.
Neither project set out to prove anything about functional medicine. Both rest on the same premise: measurable change precedes diagnosis. If that were not true, neither model would work at all.
I wrote about what those studies do and do not establish in more detail here, including the part where a single cheap blood test outperformed the transformer. But the underlying point is this one: the drift is real, and it is detectable.
Iron. Ferritin falls first. Then transferrin saturation. Then red cell indices change. Haemoglobin drops last — which is why iron deficiency without anaemia exists as a category, and why someone can be exhausted for two years with a "normal" full blood count.
Glucose regulation. Insulin rises to maintain normal glucose long before glucose itself moves. By the time HbA1c crosses a diagnostic threshold, the compensation has been running for years. Measure insulin and you see it earlier; almost nobody does.
Alzheimer’s. In a Colombian family carrying a genetic form of the disease, the tau marker p-tau217 was raised roughly twenty years before expected cognitive symptoms. The pathology was building for two decades in silence. (With important caveats about what that does and does not mean for testing.)
Autoimmunity. Autoantibodies are frequently detectable years before clinical disease in rheumatoid arthritis and coeliac disease. The immune system had been signalling for a long time.
In every case the same shape: a system compensates, compensates, compensates — and then stops being able to. The symptom marks the end of the compensation, not the start of the problem.
The version of this argument I dislike says: your body was warning you and you ignored it. That is unfair and mostly untrue.
The early signals are usually non-specific and entirely reasonable to dismiss. Tiredness. Slightly worse sleep. Less tolerance for a heavy meal. Taking longer to recover from a cold. Every one of those has a hundred innocent explanations, and most of the time the innocent explanation is right. Nobody should be running to a clinic because they felt flat in February.
And plenty of illness has no meaningful prodrome at all. Trauma, acute infection, many cancers, genetic conditions with sudden presentation. Not everything was preceded by a signal you could have caught, and suggesting otherwise adds blame to people who are already unwell.
Here is the part that connects to nutrition rather than to diagnostics, and where I want to be careful about how strongly I put it.
Biological processes have running costs. Sustained catecholamine production consumes vitamin C, B6 and magnesium. Inflammatory signalling diverts tryptophan down the kynurenine pathway rather than toward serotonin. Detoxification pathways consume glycine, cysteine and glutathione. Thyroid hormone conversion needs selenium and iron.
So a system running hot for a long time is spending, and what it spends is measurable.
Where I would not overstate it: "every symptom has a nutrient cost" is a nice sentence and it is not literally true. Some symptoms are mechanical, some are structural, some are psychological, and attaching a nutrient story to all of them is exactly the overreach my field is known for. What is fair to say is that sustained physiological demand draws on nutrient reserves, and those reserves are often measurable before anything else is.
Not "test everything." Testing without a question generates anxiety and expense, and I have argued elsewhere that more data is not automatically better.
What it changes is the interpretation of a normal result. "Normal" means "not yet outside the range in which 95% of a reference population sits." It does not mean "unchanged." A ferritin of 18 and a ferritin of 90 are both normal, and they are not the same situation — particularly if yours was 90 three years ago.
The most useful measurement is rarely a single value. It is the same value, taken twice, a year or two apart.
Which is unglamorous advice and mildly inconvenient, because it means the best thing you can do with a set of blood results is keep them.
Noticing drift early is only useful if you can do something about it, and doing something is a more technical business than it looks. Dose, form, timing and titration decide whether an intervention does anything at all — and in one study the same iron tablet was absorbed four times better depending purely on what it was taken with.
That is the second half of this piece.
Educational content, not medical advice. New, persistent or changing symptoms should be assessed by your GP — the argument here is about what happens before symptoms, not a reason to delay investigating one you already have.
A normal result and an unchanged result are different things. The second one requires having measured before.
See testing options Ask AIdan →