New blood test predicts when Alzheimer’s symptoms will begin

A single protein in your blood can rise for years before Alzheimer’s symptoms appear. In a new study, researchers argue that rise is steady enough to function like a clock.

Not a perfect clock, and not one meant for personal forecasting yet. Still, the promise is hard to ignore: one blood draw, taken once, could help estimate when symptoms are likely to begin.

The work comes from Washington University School of Medicine in St. Louis and was published in Nature Medicine. The team reports that their models predicted the onset of Alzheimer’s symptoms with a median absolute error of about three to four years.

More than 7 million Americans live with Alzheimer’s disease, and the Alzheimer’s Association projects that health and long-term care costs for Alzheimer’s and other dementias will reach nearly $400 billion in 2025. With no cure, the best near-term leverage may come from catching the disease process earlier and testing preventive treatments faster.

“Our work shows the feasibility of using blood tests, which are substantially cheaper and more accessible than brain imaging scans or spinal fluid tests, for predicting the onset of Alzheimer’s symptoms,” said senior author Suzanne E. Schindler, MD, PhD, an associate professor in the WashU Medicine Department of Neurology.

The model centers on a tau-related blood biomarker: p-tau217, measured in plasma. Levels of plasma p-tau217 can already help doctors diagnose Alzheimer’s in people who have cognitive impairment. The study notes that these tests are not currently recommended for cognitively unimpaired people outside research studies or clinical trials.

Schindler’s group aimed at a harder question. They did not ask only who has Alzheimer’s-related pathology. They tried to estimate when symptoms are likely to begin.

To do it, they analyzed data from two long-running initiatives: the WashU Medicine Knight Alzheimer Disease Research Center (Knight ADRC) and the Alzheimer’s Disease Neuroimaging Initiative (ADNI), which spans multiple U.S. sites. The participants described in the source included 603 older adults who lived independently in the community.

The study grew out of a project developed and launched by the Foundation for the National Institutes of Health Biomarkers Consortium, a public-private partnership that includes WashU Medicine.

The team focused on a related measure, plasma %p-tau217, because large longitudinal datasets were available in both cohorts.

In the Knight ADRC cohort, 506 people had longitudinal plasma %p-tau217 measures. Their median age at baseline was 67.7 years. About 54.2% were female, 35.8% were APOE ε4 carriers, and 8.5% were cognitively impaired, defined here as a Clinical Dementia Rating (CDR) greater than 0. The median span from first to last plasma collection was 7.1 years, and 61.3% provided three or more samples.

In ADNI, 406 people contributed longitudinal plasma %p-tau217 measures. Their median age at baseline was 72.7 years. About 49.3% were female, 34.2% were APOE ε4 carriers, and 48.3% were cognitively impaired. The median span from first to last plasma collection was 5.0 years, and nearly all participants provided three or more samples.

The study measured p-tau217 using PrecivityAD2, a clinically available diagnostic blood test from C2N Diagnostics, described as a WashU startup co-founded by WashU Medicine researchers David M. Holtzman, MD, and Randall J. Bateman, MD, both coauthors on the study. In the ADNI cohort, plasma p-tau217 was also measured using tests from other companies, including a Fujirebio Lumipulse p-tau217/Aβ42 measure described as cleared by the U.S. Food and Drug Administration, plus four commercially available p-tau217 assays listed as C2N Diagnostics, Janssen LucentAD Quanterix, ALZpath Quanterix, and Fujirebio Lumipulse.

The researchers leaned on a prior idea from brain imaging. Amyloid and tau PET scans show that Alzheimer’s pathology tends to accumulate in fairly consistent trajectories over time. Blood markers could offer a cheaper route to a similar timing signal.

“Amyloid and tau levels are similar to tree rings, if we know how many rings a tree has, we know how many years old it is,” said lead author Kellen K. Petersen, PhD, an instructor in neurology at WashU Medicine. “It turns out that amyloid and tau also accumulate in a consistent pattern and the age they become positive strongly predicts when someone is going to develop Alzheimer’s symptoms. We found this is also true of plasma p-tau217, which reflects both amyloid and tau levels.”

Before estimating time, the researchers first looked for a range where %p-tau217 changed with relatively consistent rates across people. They identified an overlap interval from 1.06% to 10.45% where the rate of change had relatively low variance.

That constraint matters. The study reports unstable time estimates at very low values in ADNI, sparse data at high values, and rapid increases at the high end that could make time estimates unstable. So the clock, as built here, does not apply cleanly outside that range.

They defined plasma %p-tau217 “positivity” as greater than 4.06%, chosen to align with an amyloid PET Centiloid value of 20.

Then they used two approaches to create time-based models: Temporal Integration of Rate Accumulation (TIRA) and Sampled Iterative Local Approximation (SILA). Both methods estimated years from plasma %p-tau217 positivity and produced broadly aligned results across cohorts.

One of the clearest storylines came from age. The older someone was when they became %p-tau217 positive, the shorter the time to symptoms tended to be.

For example, using TIRA-based estimates in the Knight ADRC cohort, participants who became plasma %p-tau217 positive at age 60 had a median time until symptom onset of 20.5 years. Participants who became positive at age 80 had a median time until symptom onset of 11.4 years.

That pattern showed up again when the team modeled age at symptom onset directly. They report that models predicting age of symptom onset based on estimated age at %p-tau217 positivity had a median absolute error ranging from 3.0 to 3.5 years within the same cohort. Cross-cohort applications produced median absolute errors around 3.0 to 3.7 years, depending on the direction of the test.

The researchers also measured how well risk models ranked people by likelihood of developing symptoms, using a concordance index. In Knight ADRC, the bootstrapped C-index was 0.784 for TIRA-based models and 0.790 for SILA-based models. In ADNI, the C-index was 0.730 for TIRA-based models and 0.750 for SILA-based models. Additional analyses that incorporated people already cognitively impaired at enrollment produced different C-index values, and the authors describe those analyses as addressing potential survivor bias.

The team shared code for model development and notes that Petersen built a web-based application for researchers to explore the clock models in more detail.

“These clock models could make clinical trials more efficient by identifying individuals who are likely to develop symptoms within a certain period of time,” Petersen said. “With further refinement, these methodologies have the potential to predict symptom onset accurately enough that we could use it in individual clinical care.”

The paper draws a bright line between research utility and personal medical forecasting.

First, the clock only works, as defined here, inside a specific biomarker range where change stays consistent, 1.06% to 10.45% for %p-tau217. Very high values could signal high risk, but the model does not provide stable timing estimates there. Very low values suggest low likelihood of symptoms for many years, but the study says it cannot provide precise timing at that end either.

Second, the error margin, three to four years, limits individual use. The authors say that level of precision could still help group-level studies.

Third, the study warns against applying the approach to personal decision-making today. It notes that Alzheimer’s biomarker testing for cognitively unimpaired people is not recommended outside research studies or clinical trials because of uncertain benefits and potential risks, and the authors discourage people from using these models to determine a personal estimated age of symptom onset.

The paper also lists broader limitations. Participants largely identified as non-Hispanic White, which may limit generalizability to other groups. The analysis did not explicitly model dropout or death, which could introduce survival bias. The study also points to the problem of co-pathologies in older adults, including cerebrovascular disease and other neurodegenerative diseases, which could affect how %p-tau217 relates to symptoms as people age.

The authors suggest a path forward that stays within what their results can support: combining %p-tau217 with other blood biomarkers associated with cognitive symptoms in Alzheimer’s to refine estimates.

Research findings are available online in the journal Nature Medicine.

The original story "New blood test predicts when Alzheimer’s symptoms will begin" is published in The Brighter Side of News.

Like these kind of feel good stories? Get The Brighter Side of News' newsletter.