Alzheimer's blood tests are moving from "whether" to "when"
Three February papers push plasma p-tau217 past detection: clock models estimating years to symptom onset, four assays compared head to head, and a real-world audit of what the result changes.
The question a blood biomarker for Alzheimer's disease has been able to answer is whether amyloid pathology is present. The question people actually ask is when, if ever, it will start to affect them. Three papers published in February move plasma p-tau217 — the ratio-based tau marker that has become the field's front-runner — a step toward the second question, and give a partial answer to a third: does knowing the result change what a clinician does.
Clocks, not classifiers
The Nature Medicine paper builds what the authors call clock models [s1]. Rather than classifying a person as positive or negative, a clock uses longitudinal biomarker measurements to estimate the age at which that person crossed into positivity — and from there, how long they have until symptoms.
The models used longitudinal plasma %p-tau217, the ratio of phosphorylated to non-phosphorylated tau at position 217, from two independent cohorts of 258 and 345 participants [s1]. The estimated age at plasma %p-tau217 positivity was associated with the age at onset of Alzheimer's symptoms, with adjusted R² of 0.337 to 0.612 and a median absolute error of 3.0 to 3.7 years [s1].
A median error of roughly three years is the number to hold onto. The authors frame it as within a margin acceptable for use in clinical trials [s1] — which is a deliberately bounded claim, and a different claim from being useful for telling an individual patient what to expect.
One substantive finding sits inside the modelling: the time from p-tau217 positivity to symptom onset was markedly shorter in older individuals [s1]. If that holds, biomarker positivity does not carry a fixed lead time, and the same test result means something different at 60 than at 80.
The paper also constructed similar models using data from one p-tau217/Aβ42 immunoassay and four plasma p-tau217 immunoassays [s1], which matters for whether the approach survives contact with the several commercial platforms now in use.
Do the assays agree?
That question is the subject of the second paper, in Alzheimer's Research & Therapy [s2]. Prior head-to-head comparisons of p-tau217 assays in community-based cohorts — as opposed to specialist memory-clinic samples — have been limited [s2].
The authors evaluated two southwestern Pennsylvania cohorts: MYHAT-NI, with two-year longitudinal neuroimaging follow-up of amyloid, tau and cortical thickness, and the Human Connectome Project/CoBRA cohort, designed for a roughly 50:50 split of self-identified Black and non-Hispanic White participants [s2]. Plasma p-tau217 was measured on four platforms — Lumipulse, Johnson & Johnson, ALZpath and NULISA — with amyloid and tau pathology assessed by [11C]PiB and [18F]Flortaucipir PET [s2].
Across 344 participants (MYHAT-NI n = 111, median age 76; HCP/CoBRA n = 234, median age 62), all four assays showed moderate to strong cross-platform correlations, with Spearman correlations of 0.40 to 0.86, and statistically equivalent areas under the curve of 0.84 to 0.90 for determining amyloid positivity [s2].
Equivalent discrimination is the good news. A correlation range whose lower bound is 0.40 is the caveat: two assays can rank people similarly overall while disagreeing substantially about a given individual, which is exactly the situation in which a threshold applied to one platform does not transfer to another.
Does the result change anything?
The third paper asks the question that decides whether any of this reaches patients. Published in the Journal of Neurology, it followed 200 consecutive new patients evaluated for cognitive symptoms across two settings — general neurology consultations referred from primary care, and a specialised memory unit [s3]. The cohort was 38.5% subjective cognitive complaints, 47.5% mild cognitive impairment and 14% dementia [s3].
Attending neurologists recorded a pre-biomarker clinical diagnosis and rated their diagnostic confidence from 0 to 10, then repeated both after receiving the p-tau217 result [s3].
Fifty-one of 200 patients (25.5%) had a change in diagnostic category after the result [s3]. Mean diagnostic confidence rose from 6.90 ± 1.74 to 8.49 ± 1.68 [s3]. Changes were seen in both settings and across all clinical stages [s3].
The paper also reports agreement with the eventual final diagnosis, and here its own numbers do not reconcile. It states that the pre-biomarker diagnosis was maintained in "71/200 cases (75.5%)" with a kappa of 0.576, against a post-biomarker figure of 189/200 (94.5%) with kappa 0.906 [s3]. Seventy-one of 200 is 35.5%, not 75.5%; 151 of 200 would be 75.5%. The discrepancy is in the published abstract, and we report both figures rather than choosing one. The kappa statistics, which do not depend on that count, still show a substantial improvement in agreement.
What is and is not established
None of these papers is a demonstration that outcomes improve. Diagnostic confidence is a measure of how sure the clinician feels; a change in diagnostic category is a change in a label. Whether either translates into better care, or into harm from earlier labelling in people who would never have developed symptoms, is not addressed by any of the three.
The clock models were built and validated in research cohorts, and the authors' stated use case is clinical trials [s1]. The head-to-head comparison establishes that four assays perform equivalently for amyloid positivity in two US community cohorts [s2], not that any given laboratory's cut-off is transportable. The real-world study is a single before-and-after design in 200 consecutive patients across two clinical settings, without a control group in which the test was withheld [s3].
What to watch
The specific claim worth tracking is the age dependence of lead time [s1]. If the interval between biomarker positivity and symptoms genuinely compresses with age, then any framing of these tests as giving people "years of warning" needs to be age-specific, and prevention trials that enrol on biomarker status alone are enrolling a population with very unequal time horizons.
Sources
- [s1] Predicting onset of symptomatic Alzheimer's disease with plasma p-tau217 clocks. Nature Medicine, 19 February 2026. https://doi.org/10.1038/s41591-026-04206-y
- [s2] Head-to-head comparison of plasma p-tau217 immunoassays for incipient Alzheimer's disease in community cohorts. Alzheimer's Research & Therapy, 6 February 2026. https://doi.org/10.1186/s13195-026-01969-x
- [s3] Impact of blood p-tau217 testing on diagnosis and diagnostic confidence in cognitive disorders: a real-world clinical study. Journal of Neurology, 10 February 2026. https://doi.org/10.1007/s00415-026-13676-6
Sources
- Predicting onset of symptomatic Alzheimer's disease with plasma p-tau217 clocks — Nature Medicine , February 19, 2026
- Head-to-head comparison of plasma p-tau217 immunoassays for incipient Alzheimer's disease in community cohorts — Alzheimer's Research & Therapy , February 6, 2026
- Impact of blood p-tau217 testing on diagnosis and diagnostic confidence in cognitive disorders: a real-world clinical study — Journal of Neurology , February 10, 2026
More on
Alzheimer's blood tests exist. In people with no symptoms, they answer the wrong question.
The approved tests are for patients already showing cognitive decline. Used on the worried well, a positive result mostly signals a risk that may never become disease — and no treatment follows from it.
An FDA-cleared Alzheimer's blood test missed badly in one memory clinic cohort
At a Mayo Clinic cohort of 252 patients, 40% of amyloid-negative people tested positive. The manufacturer says it has found a manufacturing problem and put the product on hold.
Brief bursts of brain activity flagged who would go on to develop Alzheimer's
In 85 people with mild cognitive impairment, transient beta-band events recorded at rest differed years before diagnosis. The sample is small and the finding has not been replicated.
A tau antibody drove a tangle marker down 90% — in seven people
The biomarker effect reported at CTAD is large and sustained over two years. The sample is seven, the design is open dose-escalation, and imaging was available for three.