Alzheimer’s blood tests are becoming credible diagnostic tools, but their usefulness depends on the question being asked. Investigating memory symptoms differs from screening a healthy person. Validated tests can support a clinical assessment; NIH’s March 2026 summary cautions that p-tau217 testing is not currently recommended for cognitively normal people or as a stand-alone diagnostic tool.
What the blood sample is measuring
A biomarker is a measurable sign of a biological process. In Alzheimer’s research, blood tests can detect proteins associated with changes in the brain. Mayo Clinic’s guide to dementia diagnosis describes biomarkers alongside clinical history, cognitive testing, imaging and other investigations. Its central point is that diagnosis requires a combination of assessments. These measurements do not replace the account of how someone is functioning. For blood testing specifically, NIH describes p-tau217 as a protein marker associated with Alzheimer’s-related brain pathology.
One prominent marker is phosphorylated tau at position 217, usually shortened to p-tau217. Some tests measure it alone, whilst others use ratios involving tau or amyloid proteins. In Sebastian Palmqvist and colleagues’ 2024 study in JAMA, the researchers examined percentage p-tau217 and a combined amyloid probability score. These are specific laboratory methods, rather than interchangeable versions of a single universal blood test.
The distinction matters when reading a report. A result needs the name of the assay, its interpretation and the question it was intended to answer. A numerical value detached from that context offers much less information than its apparent precision suggests.
The accuracy findings deserve their context
Palmqvist’s 2024 JAMA investigation included 1,213 people in Sweden undergoing evaluation because of cognitive symptoms. Using predefined thresholds, the blood measurements performed well in identifying Alzheimer’s-related pathology. This was a diagnostic study in symptomatic people, not a trial showing that screening healthy adults improves their later health.
A further 2025 study led by Palmqvist in Nature Medicine evaluated an automated p-tau217 assay in 1,767 participants with cognitive symptoms. With a single threshold, accuracy was 89–91% in secondary care and 85% in primary care. These differences caution against carrying one headline percentage into every setting.
Using two thresholds increased accuracy to 92–94%, but that calculation excluded the 12–17% of participants whose results fell into an intermediate range. Those people still needed an answer. An intermediate result is a recognised feature of the testing approach, rather than a laboratory inconvenience that can be ignored.
The useful reading of an accuracy claim therefore includes three questions: who was tested, what counted as the reference standard, and how were uncertain results handled? Without those details, a percentage can look more conclusive than the underlying study.
A positive result depends on the starting probability
Sensitivity describes how often a test detects the target condition when it is present. Specificity describes how often it returns a negative result when that condition is absent. Neither percentage, by itself, tells a particular person the chance that their positive result is correct. The 2025 specialist-care guideline explicitly asks clinicians to consider the probability of Alzheimer’s pathology before testing.
Consider a hypothetical test with 90% sensitivity and 90% specificity. These numbers are chosen to illustrate the arithmetic, not to describe a named product. In 1,000 people, suppose 100 have the target pathology. The test would identify 90 of them and miss 10. Among the other 900 people, 90 would receive false-positive results. Of 180 positive results, half would be true positives.
Now suppose 500 of the same-sized group have the pathology. The test would produce 450 true positives and 50 false positives. Nine in ten positive results would be correct. The test characteristics have stayed the same; the mix of people being tested has changed.
This illustration does not estimate anyone’s personal risk. It explains why a service investigating cognitive impairment and a service offering unsolicited reassurance cannot simply borrow the same interpretation. The starting question influences what a result can reasonably mean.
Symptoms still need a wider investigation
The NHS description of dementia assessment begins with history, the effect of symptoms on daily life, and a review of medicines and existing conditions. Assessment may include cognitive testing, blood investigations and scans. The aim is to understand the person’s difficulties and their possible causes.
Similarly, NICE guideline NG97 calls for investigation of potentially reversible contributors, including depression, sensory impairment and medication effects, before referral when dementia remains suspected. NICE also warns against excluding dementia solely because someone scores normally on a cognitive instrument. Neither a reassuring score nor a worrying laboratory measurement should end the reasoning prematurely.
This is where a biomarker’s promise needs a clinical framework. A positive Alzheimer’s marker does not remove the need to consider other explanations or conditions occurring alongside it. Michelle Mielke and Nicole Fowler’s 2024 review in Nature Reviews Neurology (open author manuscript) highlights that Alzheimer’s pathology can be an incidental finding and must be interpreted with the symptoms and possible coexisting pathology.
There is also a separate safety issue. The NHS advice on sudden confusion is to go to A&E or call 999 if someone suddenly becomes confused. That situation should not wait for an elective biomarker appointment or an online test result.
Testing pathways have different purposes
The 2025 guideline distinguishes triage from confirmation in people with objective cognitive impairment seen in specialised care. A suitable triage test helps decide who needs further investigation; a positive triage result requires confirmation with another method. A sufficiently accurate confirmatory blood test can, in that setting, substitute for amyloid PET or spinal-fluid biomarker testing. Both recommendations are conditional, and they apply to tests meeting specified performance requirements.
The research possibility is broader. NIH’s March 2026 report on p-tau217 research describes a model intended to estimate when symptoms might emerge. It could help researchers select participants for prevention trials, but the report cautions that the model is not yet accurate enough for individual use. Research usefulness asks a different question from whether an individual clinical test will help someone today.
Before agreeing to testing, a reasonable conversation would cover what happens after each possible result: positive, negative or intermediate. It would also cover who explains the findings and whether they would change the proposed care. These questions keep attention on the decision the test is supposed to support.
The people behind the study average
Mielke and Fowler’s population-use review (open author manuscript) cautions that research participants can differ from typical patients in primary care. Comorbidities and demographic differences may affect interpretation, and evidence from a selected clinic population cannot automatically establish performance across the whole population. The review also raises consent, incidental findings and the consequences of recording a biomarker result.
This does not mean that every additional condition invalidates every assay. In the 2025 automated-assay study, accuracy was not affected by chronic kidney disease in the studied cohort, although it was lower in participants aged 80 or over. Such findings need to remain attached to the assay and population examined. A claim about one method should not become a promise about all products measuring a related protein.
The important development is better diagnostic access with an explanation of uncertainty. A service that can provide the measurement but cannot explain its limitations has left a central part of the work unfinished.
What this means in practice
- If gradual memory changes are worrying you, consider recording when they began and how they affect daily activities for a GP discussion; the NHS assessment guide explains why that history matters.
- For a planned assessment, bring a medicines list and, if you wish, someone who knows you well, as suggested in the same NHS guide.
- If a blood biomarker is proposed, ask for the assay name, its intended use and the plan for an intermediate result.
- Ask who will explain the result alongside the clinical assessment, and what decision it could change.
- Treat sudden confusion as urgent: follow the NHS advice to attend A&E or call 999.
What we don’t know
Diagnostic accuracy does not establish that routine screening of people without symptoms produces better long-term outcomes. The 2024 population-use review (open author manuscript) identifies unresolved questions around broad implementation and personal consequences. The 2024 JAMA study also calls for evaluation of how testing influences clinical care. There is a gap between reliably finding a biological signal and demonstrating benefit from every proposed use of that signal.
For now, the most constructive question is what a result would add to a careful assessment. Better measurement is valuable when it helps explain uncertainty and supports a decision that matters to the person being tested.
Photo: Polina Tankilevitch on Pexels. Laboratory sample containers; illustrative image, not a specific Alzheimer’s assay.
