Blood tests for Alzheimer's proteins predicted dementia only modestly over 12 years

Diagnosing Alzheimer's disease used to need a brain scan or a sample of the fluid around the spinal cord. In the last few years, blood tests have arrived that measure proteins linked to the disease. In memory clinics and research studies they match brain scans remarkably well. That raises a hope: could a simple blood test pick out, years in advance, which healthy older people will develop dementia?

A study of more than 5,000 British civil servants, followed for about 12 years, gives a sobering answer. The proteins were clearly linked to later dementia, but as a forecast for individuals they were only modestly accurate, and not much better than what doctors already know about a person's age and health. It is an observational study: it tracks people over time rather than testing a treatment.

What they did

The Whitehall II study has followed London office workers since the late 1980s. In 2012–2013, 5,225 participants without dementia, on average 69 years old, gave a blood sample that was frozen. In 2024, a laboratory that did not know who the samples came from measured four proteins:

• Amyloid-beta 42/40 ratio. Amyloid is the protein that clumps into plaques in Alzheimer's brains; a lower ratio in blood is the warning sign.
• p-tau217, a form of the protein tau, which forms tangles inside nerve cells in Alzheimer's.
• NfL (neurofilament light), a piece of the skeleton of nerve fibres that leaks into the blood when nerve cells are damaged.
• GFAP, a protein released by astrocytes, support cells of the brain, when they react to damage.

The team then used national health records to see who was diagnosed with dementia of any kind up to March 2025. They asked two questions: how strongly each protein was linked to dementia, and how well the proteins could actually predict it.

What they found

• 516 people (about 10%) developed dementia during a median follow-up of 12.2 years.
• Clear links. People in the worst quarter for each protein had a higher risk, after taking age, sex, education, smoking, exercise, weight, blood pressure, diabetes, heart disease and kidney function into account. The risk was 3.7 times higher for p-tau217, 2.3 times for GFAP and 1.9 times for both NfL and the amyloid ratio.
• Modest prediction. Forecasting accuracy is measured with a score called the C-index: 0.5 is a coin toss, 1.0 is perfect. p-tau217 alone scored 0.69, the amyloid ratio 0.59, and all four proteins together 0.76. In memory-clinic studies, p-tau217 often scores above 0.90.
• Many cases missed. Using the worst quarter as a "positive" result, the tests flagged at most about 60% of the people who later developed dementia. Of those flagged, fewer than a quarter went on to get it. A negative result was more reassuring: 92–95% of people who tested negative stayed free of dementia.
• No better than known risk factors. A model using only age, background, lifestyle and health scored 0.77, about the same as all four proteins. Adding the proteins to it raised the score to 0.81, a small improvement.
• Better in the short term. The proteins predicted dementia in the first seven years better than in the years after.

What to keep in mind

• Dementia of any kind. Health records rarely said which type of dementia people had, so the team could not test Alzheimer's separately. The blood tests target Alzheimer's, while many older people have a mix of causes.
• Mild cases may be missing. Diagnoses came from health records, which can miss milder dementia.
• Who was studied. Participants were mostly White former office workers, and about three in four were men. Whether the results apply to more diverse groups is unclear.
• Other illnesses blur the picture. The authors note that common conditions such as kidney disease affect blood levels of these proteins, which may be one reason the tests work less well in the general population than in carefully selected research groups.
• Prediction, not diagnosis. The study looked at healthy people years before symptoms. The tests may perform better in people who already have memory problems, which is where they are used in clinics.

Why it matters

Blood tests for Alzheimer's proteins are moving into medical practice, and there is interest in using them to screen healthy older people. This study suggests caution: in an ordinary population, the tests missed many future cases and raised many false alarms, and they added little to what age and health already reveal. The authors conclude that the tests have limited use for early detection in the general population, but could serve as a first step to decide who should be sent for further, more specialised examination, especially where little else is known about a person.