A Blood Test Can Now Predict Alzheimer's Years Early — But It Reads a Probability, Not a Sentence
June 21, 2026 · Oxford VSI, Cognitive Neuroscience~5 min read
This spring the headlines arrived with the weight of a verdict: a simple blood draw can now spot Alzheimer's "years before symptoms." It is a real advance, and the science behind it is genuinely good. But the word "predict" is doing quiet, dangerous work in those sentences — because a number that predicts is not a number that has decided, and the gap between those two things is exactly where a healthy person can lose months of peace to a result they were never taught how to read.
What actually got measured
Two studies sit under the wave of coverage. In April 2026, a Mass General Brigham team followed 317 cognitively healthy older adults for about eight years and showed that a blood protein called p-tau217 rose, in many people, before an amyloid PET scan turned positive — meaning the chemistry shifts years ahead of the picture we used to wait for. In June, a Lancet study of roughly 1,300 midlife adults (average age 61) found that the small fraction with elevated amyloid and tau markers — about 6% of them — carried something like a 2.5-to-4-fold higher risk of faster memory and processing decline over the next five years. Both are careful, prospective, peer-reviewed work. Neither says what the headline says.
"Predicts" is not "diagnoses"
Here a small book pays off. The Oxford Very Short Introduction to Cognitive Neuroscience hammers one discipline into the reader: a brain image, or a brain marker, shows a state — never a cause, and never a fate. Its core rule for reading any brain headline is to ask "minus what?" — a scan only means something against a control, and a number only means something against a base rate. Translate that to the blood test and the whole frame changes. A raised p-tau217 says: the biology associated with Alzheimer's is, on average, further along in you than in someone with a low reading. It does not say you have the disease. It does not say you will. The Lancet authors put it plainly: a positive result "does not mean someone has dementia, and it does not predict the future with certainty."
New blood tests flag Alzheimer's biology years early, but a raised marker is a state and a probability, not a diagnosis or a fate: the much-quoted 2.5–4× is a relative risk over five years, and in the midlife study only ~6% tested positive. Framework: Cognitive Neuroscience: A Very Short Introduction (Oxford) — ask "minus what?" of any brain headline. Figures per the original April & June 2026 studies. Popular-science interpretation, not medical advice — for any health decision, consult a qualified doctor.
The number you were handed is relative
The scariest phrase in the coverage — "2.5 to 4 times the risk" — is a relative risk, and relative risk is a magician's number. It tells you how one group compares to another; it tells you nothing, by itself, about how likely you are. If a baseline five-year risk is small, several times a small number is still a smallish number. And the framing buries the cheerful half of the same study: in that midlife group, only about 6% had elevated markers at all. Ninety-four percent of people walked away negative. A test sold as a way to find the doomed is, for the overwhelming majority, a way to be told you are not. That is the same arithmetic, read honestly instead of for the fright.
The harder question is whether to look
So here is the part medicine is still working out, and it deserves more humility than a headline allows: there is, as of now, no approved drug that reverses Alzheimer's, and the researchers themselves call these tools "too early to recommend" for routine use in healthy people — closer to a screening instrument for clinical trials than a number to live by. Which means a healthy person can pay to learn a probability they cannot yet act on, and then carry it. For some, a frank risk number is a gift: it sharpens the things that genuinely move the odds — sleep, blood pressure, exercise, hearing, not smoking. For others it is a shadow that falls over every forgotten name. Knowing your odds is not automatically the same as being helped by knowing them.
What this means for you
If a test like this comes your way — and within a few years, one will — the move is not to refuse the science but to refuse the misreading. Ask the questions the marker can't answer for you: out of a hundred people with my result, how many actually develop the disease, and over how long? What would I do differently with the answer than without it? Is there a treatment to act on, or only a number to absorb? A biomarker is a weather forecast, not a calendar entry. It can tell you to carry an umbrella; it cannot tell you it will rain on you. The forecast is worth having — but only if you read it as a probability, and keep living as the person who hasn't been sentenced, because, statistically, you almost certainly haven't.
A marker that predicts is not a marker that has decided. It shifts the odds; it does not write the ending — and for most people who test positive on the scary number, the ending is still nothing at all.
Read it as a forecast, not a verdict. The umbrella is worth carrying; the certainty was never on offer.
Source: framework from Cognitive Neuroscience: A Very Short Introduction (Oxford) — a brain image or marker shows a state, never a cause or a fate; read any brain headline by asking "minus what?". Real-world basis: the April 2026 Mass General Brigham p-tau217 study (317 adults, ~8-year follow-up) and the June 2026 Lancet midlife biomarker study (~1,300 adults, ~6% marker-positive, 2.5–4× relative risk of faster decline over 5 years), where investigators stressed the result is not diagnostic. Specific figures are per the original studies. This is popular-science interpretation, not medical advice — for any health decision, including whether to test, consult a qualified doctor.
この報道の波の下には、二つの研究がある。2026年4月、マサチューセッツ総合病院ブリガムのチームは、認知的に健康な高齢者317名を約八年追い、p-tau217 という血中タンパク質が、多くの人でアミロイド PET 検査が陽性になる前に上昇することを示した——つまり化学的な変化は、私たちがこれまで待っていた画像より数年早く現れる。6月、『ランセット』の約1300名の中年者(平均61歳)の研究は、標識物が高い少数——約6%——が、今後五年で記憶と処理速度がより速く低下するリスクを、およそ2.5〜4倍背負うことを見いだした。どちらも丁寧で、前向きで、査読を経た仕事だ。だがどちらも、見出しが言うことは言っていない。