A single blood draw now detects Alzheimer’s pathology as accurately as a lumbar puncture. That is a genuine breakthrough and it is going to change dementia diagnosis. It has also been validated in a much narrower group of people than the coverage implies — and if you are well, you are not in it.
I had planned to write this as “the blood test that detects Alzheimer’s ten years early.” That was the working title, and it is roughly how the story has been told.
Reading the papers properly, the title was wrong — and wrong in a way that matters, because people are already asking to buy this test privately when they feel perfectly well.
Tau is a protein that stabilises the internal scaffolding of neurons. In Alzheimer’s disease it becomes abnormally phosphorylated, detaches, and forms the tangles that give the disease its pathology.
Phosphorylation at position 217 turns out to be unusually specific to Alzheimer’s, and a small amount leaks into the bloodstream. Which means the pathology inside the brain can be measured from an ordinary blood draw — no lumbar puncture, no PET scanner, no specialist centre.
That is a genuinely large change. Until very recently, confirming Alzheimer’s pathology meant a spinal tap or a scan costing thousands.
1,767 people with cognitive symptoms across five cohorts — four in secondary care (Malmö, Gothenburg, Barcelona, Brescia) and one in Swedish primary care. Measured on the fully automated Lumipulse platform, against cerebrospinal fluid as the reference standard.
Areas under the curve of 0.93 to 0.96. In secondary care, accuracy 89–91%. In primary care, 85%.
Using two cut-offs rather than one, accuracy rose to 92–94% — at the cost of leaving 12–17% of results in an intermediate zone that the test cannot classify.
And a set of negatives that matter: performance was unaffected by chronic kidney disease, diabetes, sex, APOE genotype or cognitive stage. It was lower in people aged 80 and over (83%).
The comparison that makes the case, though, is not against spinal fluid. It is against clinicians.
In an earlier JAMA study by the same group, after clinical examination, cognitive testing and a CT scan:
Primary care physicians identified Alzheimer’s correctly 61% of the time. Dementia specialists managed 73%.
Using the blood test, both reached 91%. That is not a marginal improvement over expert judgement. It is the difference between a coin-weighted guess and a diagnosis.
Every study above was conducted in people who already had cognitive symptoms and had been referred for assessment. That is the population the test was designed for, validated in, and recommended for.
The Alzheimer’s Association clinical practice guideline published in 2025 is explicit about the scope: blood biomarkers are for individuals with objective cognitive impairment, being evaluated by clinicians experienced in memory disorders, where Alzheimer’s is the suspected cause. Tests meeting a high enough standard can triage, or substitute for PET and spinal fluid testing.
The guideline also says something that ought to be quoted more often: there is significant variability between tests, and many commercially available blood biomarker tests do not meet the required thresholds — particularly when a single cut-off is used.
From a genuinely remarkable finding, in a genuinely unusual population. In a Colombian family carrying a PSEN1 mutation that causes autosomal dominant Alzheimer’s — where onset age is largely predictable — p-tau217 was raised in carriers from around age 25, roughly twenty years before expected cognitive symptoms.
That tells us the pathology builds long before symptoms. It does not tell us that a blood test performs the same way in a 55-year-old with no symptoms and no family history. Those are different questions, and only the first has been answered.
Four reasons, and none of them is that the test is poor.
Predictive value collapses when prevalence falls. This is the arithmetic that undoes most screening enthusiasm. A test with 90% accuracy performs well when half the people tested have the disease — which is roughly the case in a memory clinic. Apply the same test to a well population where a small fraction have detectable pathology, and a large share of positives become false ones. The test has not changed. The population has.
The intermediate zone is not small. Between 12 and 17% of results land in a band the test cannot classify. In a memory clinic that triggers a lumbar puncture and an answer. In a well person it produces months of anxiety and no resolution.
Pathology is not disease. A meaningful number of people have Alzheimer’s pathology in their brains and never develop dementia. A positive result in someone with no symptoms tells you something is present. It does not tell you when, or whether, it will matter.
And there is a specific false-positive trap. Real-world work has found p-tau217 elevated in people with severe renal impairment. In a symptomatic patient a clinician weighs that. In a direct-to-consumer test result it is invisible.
If you have cognitive symptoms, this test is one of the most useful things to arrive in medicine this decade, and you should ask about it. If you are well and worried, it is not the reassurance you are hoping to buy.
Here is what I think is the more interesting problem, and the reason this article sits on a nutrition site at all.
A diagnostic test tells you what is there. It says nothing whatever about what is modifiable. And for dementia risk, the modifiable layer is substantial and almost entirely absent from the diagnostic conversation.
Homocysteine. The most directly actionable marker in this space, and I have written about it at length elsewhere. The VITACOG trial showed B vitamin supplementation slowed brain atrophy in people with raised homocysteine and mild cognitive impairment. It is a cheap blood test and the intervention is inexpensive.
Insulin resistance. Hyperinsulinaemia reduces the availability of insulin-degrading enzyme, which also clears amyloid. Fasting insulin and HbA1c together tell you more here than either alone — and fasting insulin is rarely run.
The kynurenine pathway. Inflammation diverts tryptophan toward quinolinate, an NMDA agonist implicated in excitotoxicity. Organic acid testing maps this, and the kynurenate-to-quinolinate balance is more informative than either value alone.
APOE genotype — and here the framing matters more than the result. APOE4 is a risk modifier, not a sentence. Notably, the Nature Medicine data found p-tau217 accuracy was unaffected by APOE genotype, which is worth knowing for anyone who has had a genetic panel and been frightened by it.
None of this is an alternative to a diagnosis. It is the part of the picture a diagnosis does not contain — and, unlike the pathology, some of it responds to being changed.
If you have cognitive symptoms: see your GP and ask about referral. This test is coming into UK practice and it is a very good one. Do not order it privately instead of getting assessed.
If you are worried but well: the honest answer is that the modifiable markers are the better use of your money and your attention. Homocysteine, HbA1c, fasting insulin, vitamin D, thyroid function and inflammatory markers — most of it available free from a GP with a reason.
And if you have already had a positive result privately: take it to a doctor rather than to the internet. A result outside the population it was validated in needs interpreting by someone who can weigh it, and that is not a website.
Educational content, not medical advice. p-tau217 testing is validated for use in people with cognitive symptoms as part of a specialist assessment, and is not currently recommended as a screening test in asymptomatic people. If you are concerned about your memory or thinking, see your GP — that is the appropriate route and it is free.
Which includes knowing when a test cannot answer your question. The modifiable markers are cheaper, and they are the ones you can act on.
Ask AIdan → See testing options