- Design
- Retrospective analysis of a prospective population cohort (Whitehall II)
- Population
- 5,225 UK adults, mean age 69, without dementia at baseline
- Primary outcome
- Incident all-cause dementia over a median 12.2 years
- Effect
- p-tau217 worst quartile HR 3.74 (95% CI 3.12–4.50); C-index 0.76 for all biomarkers vs 0.77 for risk factors
Whitehall II, a long-running cohort of UK civil servants, assayed blood stored in 2012–13 from 5,225 people with a mean age of 69 for the amyloid-beta 42/40 ratio, p-tau217, neurofilament light and GFAP. Over a median 12 years, 516 developed dementia.
Each marker was associated with higher risk, most strongly p-tau217 (worst quartile HR 3.74). But as predictors they were modest: sensitivity was 42–60% and positive predictive value 16–24%, so most people with abnormal results did not develop dementia in that time. Negative predictive value was good, above 92%. All four markers together predicted about as well as a model built from age, education, behaviour and health conditions; adding them to that model improved prediction somewhat.
Blood tests for Alzheimer pathology are moving into clinics and direct-to-consumer markets. They perform well in memory clinics, where the question is whether symptoms are due to Alzheimer disease. This study asks a different question — predicting dementia years ahead in people without symptoms — and finds them unready for it.
The cohort was mostly male, white and well educated, and samples were frozen for a decade, so results may differ elsewhere.
- Do not order blood dementia biomarkers to screen people without cognitive symptoms.
- Explain to worried patients that an abnormal result in an asymptomatic person often does not lead to dementia.
- Keep blood biomarkers for symptomatic patients being assessed in a specialist setting.
- Address modifiable risk factors — blood pressure, hearing, inactivity, smoking — which predicted about as well.
Why it matters
It sets a limit on a test that is being marketed for exactly this use.
Don't overread it
This tested prediction in asymptomatic adults; it says nothing against biomarker use in memory clinics.
The statistics, in plain English
A C-index of 0.5 is a coin toss and 1.0 perfect ranking. p-tau217 alone reached 0.69, all biomarkers 0.76 — the same as ordinary risk factors (0.77). A positive predictive value of about 20% means four in five people flagged did not develop dementia over 12 years.
Read the rest in the app
You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.
Scan to keep reading on your phone. No account needed to start.
Tomorrow morning, before your first patient
One edition a day for neurology, written by the desk, every claim tied to its paper. Six minutes.
Get the app — free