Before I’d finished reading about the blood test that can flag cognitive decline years ahead, I’d already pictured holding out my arm for it. Then I pictured the drive home with a bad number and nothing to do about it. The scientists who ran the study seem to have pictured that same drive, because they say people without symptoms shouldn’t get tested at all. Eli Lilly, one of the study’s funders, had already used the same blood marker to sift through 63,124 people for a prevention trial of its own antibody.

The study, published in JAMA by a Mass General Brigham team led by Rachel Buckley and Reisa Sperling, pooled six cohorts from North America, Japan and Australia. That gave them 2,684 older adults who were cognitively unimpaired when they gave blood. Over a median 5.4 years of follow-up, 478 of them progressed to cognitive impairment. The paper defines that broadly: mild cognitive impairment, dementia, or two consecutive abnormal scores on a standard dementia rating scale. It isn’t a diagnosis of Alzheimer’s dementia.

The marker is p-tau217, and the higher it ran at baseline, the worse people tended to do. In the very high tier, more than 2.5 standard deviations above the reference, the estimated five-year risk of impairment was 38 percent. In the tier just below it was 24 percent. That 38 scares me, and it also means roughly six in ten people with the worst reading were estimated to stay clear of impairment for five years. Mass General Brigham’s release puts the ten-year estimate for the top tier at 78 percent, but much less data sits behind that number. These were research volunteers and trial participants, not a random slice of the public, and the authors say the numbers still need validating in unselected populations before anyone reads them as a personal forecast.

FIVE-YEAR RISK OF IMPAIRMENT (percent)
Figure data
GroupValue (percent)
Very high tier38
Tier just below very high24
Estimated five-year risk of progressing to cognitive impairment among cognitively unimpaired older adults, by baseline p-tau217 tier. Source: JAMA: Buckley, Sperling et al. (2026)
WHAT KISUNLA'S SCANS SHOWED
WHAT KISUNLA'S SCANS SHOWED
Scan findingKisunla (%)Placebo (%)
Brain swelling (ARIA-E)242
Microbleeds or iron deposits (ARIA-H)3113
Placebo-controlled trial results from the drug's US label, in patients who already had symptoms. Source: Kisunla (donanemab-azbt) US prescribing information

I expected the blood signal to shrink to almost nothing once the researchers accounted for the plaque, since amyloid PET is the gold-standard picture of what’s building up in the brain. It didn’t. In the JAMA analysis, each one-standard-deviation rise in p-tau217 carried a hazard ratio of 1.38 (95% CI, 1.30 to 1.46). In plain terms, that’s a bit over a third more risk of progressing at any point during follow-up. After adjusting for amyloid PET, it was 1.32 (95% CI, 1.24 to 1.41). Mass General Brigham says the prediction also held regardless of APOE4, the best-known genetic risk factor.

So why would a protein in your arm know something a direct picture of the plaque doesn’t? Tau lives inside neurons. It normally carries some phosphate tags, but in Alzheimer’s it picks up abnormal ones and clumps into the tangles that choke brain cells, and an amyloid scan can’t see tangles. I’d love to know whether p-tau217 is catching that second process early. This paper didn’t measure it.

The marker has a history, and Lilly is part of it. When p-tau217 results reached the Alzheimer’s Association conference in 2020, researchers called it the tau form most specific to Alzheimer’s and the earliest to change. It told Alzheimer’s apart from other neurodegenerative disorders with 89 to 98 percent accuracy and correlated closely with amyloid buildup. In people carrying a gene that causes Alzheimer’s, it started rising about 20 years before impairment. One of the researchers named on that work was Jeffrey Dage of Eli Lilly.

A cholesterol test with no statin

Sperling didn’t soften it for her own institution’s press office. “We do not yet have disease-modifying treatments for people who find out they are at high risk, which is why we don’t recommend currently available blood tests for asymptomatic individuals,” she told Mass General Brigham. Her advice stays the same whatever the result: exercise, eat well, sleep. She says the long-term goal is to get “to where cholesterol testing is in predicting your risk of a heart attack.”

A high cholesterol number is worth knowing because it leads somewhere: diet, exercise, a statin. In someone who feels fine, a high p-tau217 number leads to the same walk-eat-sleep advice everyone already gets. The only p-tau217 blood test regulators have cleared isn’t meant for that person either. Fujirebio’s Lumipulse G, which measures p-tau217 as a ratio against beta-amyloid 1-42, was cleared in May 2025 for older adults who already show signs and symptoms of Alzheimer’s. Mass General Brigham’s own release still notes that a p-tau217 test “recently received federal clearance” while describing a study of people with no symptoms. That pairing invites readers to make a leap the label doesn’t allow.

Sixty-three thousand arms

Sperling did name one place p-tau217 belongs today. It “can help identify people at high risk for future Alzheimer’s dementia,” she said, which makes it useful for choosing who enters prevention trials. According to the study’s funding and disclosures, backers included NIH, the Department of Defense, the Alzheimer’s Association, the GHR Foundation and Eli Lilly. Avid Radiopharmaceuticals provided in-kind support, and Avid is a PET-tracer maker that is a wholly owned Lilly subsidiary. Sperling reported consulting for several pharmaceutical and biotech companies and clinical trial funding from Eisai and Lilly, all outside the submitted work.

Lilly’s TRAILBLAZER-ALZ 3 shows what that use looks like at scale. The trial is testing donanemab, Lilly’s anti-amyloid antibody, in people without symptoms. Its Lilly-funded screening report describes 63,124 people screened over about two and a half years, many of them at mobile research units parked at health fairs and community events. Plasma p-tau217 was the only biomarker used to establish Alzheimer’s pathology, measured with a Lilly-run assay in Lilly’s own clinical lab. Screening failed 96.5 percent of them, mostly on their p-tau217 result, and 2,196 were randomized. Enrollment is closed now, and the registry lists primary completion in late 2027.

Kisunla, the marketed form of donanemab, is labeled for people who already have mild cognitive impairment or mild dementia. Its boxed warning says the brain swelling and bleeding known as ARIA can be serious, life-threatening or fatal, though the label notes ARIA usually causes no symptoms. In the placebo-controlled trial, scans showed swelling in 24 percent of treated patients versus 2 percent on placebo, and microbleeds or iron deposits in 31 percent versus 13 percent. If TRAILBLAZER-ALZ 3 succeeds, the drug’s market stops being people with symptoms and becomes people who feel fine and got flagged by a blood draw. The trial hasn’t yet shown whether those scan findings are a fair trade for a sharp 68-year-old.


Alzheimer’s research has run this experiment before. The A4 study gave Lilly’s solanezumab to more than 1,100 people with preclinical Alzheimer’s for about four and a half years. Higher baseline amyloid strongly predicted who went on to develop symptoms (p < 0.001), yet the drug didn’t slow their decline. It missed the primary endpoint (p = 0.26), every clinical outcome numerically favored placebo, and Lilly ended solanezumab’s development. The A4 study group is credited on the new JAMA paper. A sharper predictor gets you a better-selected trial, but it can’t make a drug work.

I love this biology. A phospho-tagged protein in an ordinary blood draw still predicts decline after you account for the plaque, and I want to know what it’s picking up. But if I were 68 and sharp, I would not order a p-tau217 test today. I’d walk, eat real food and protect my sleep, which is what the researchers tell everyone regardless of the number. I’d only hold out my arm once a prevention trial shows that people like me actually stayed sharp.

Sources

  1. JAMA: Buckley, Sperling et al. (2026)
  2. Mass General Brigham: Blood test predicts Alzheimer’s up to a decade early
  3. ScienceDaily: A simple blood test could reveal Alzheimer’s risk 10 years before symptoms
  4. Alzheimer’s & Dementia: TRAILBLAZER-ALZ 3 screening and baseline data
  5. ClinicalTrials.gov: TRAILBLAZER-ALZ 3 (NCT05026866)
  6. USC ATRI: A4 Study solanezumab results
  7. Kisunla (donanemab-azbt) US prescribing information
  8. BioSpace: FDA clears first Alzheimer’s blood test
  9. AAIC 2020: Blood biomarkers for tau
  10. News-Medical: Lilly, Avid receive FDA approval for Amyvid