Alzheimer’s is a disease in which timing is more brutal than diagnosis. Before symptoms appear, the brain has already been changing quietly for a long time. An attempt to estimate when symptoms will begin using blood biomarkers is opening an era of prediction.

A research team at Washington University School of Medicine in St. Louis (WashU Medicine), in St. Louis, Missouri, USA, directly addressed the question: “Can we predict the onset time of Alzheimer’s symptoms with a blood test?” The goal was not simply to measure biomarkers in blood and talk about risk, but to place those values on a time axis and calculate “whether symptoms are near or far.”

The study was not designed to end with a single test. In the Knight Alzheimer Disease Research Center cohort that Washington University has operated over the long term, participants provided blood repeatedly over multiple years and underwent regular cognitive and clinical assessments.

In other words, the researchers did not look only at “patients with clearly evident symptoms,” but tracked how time unfolds for people who appear, on the surface, to be doing fine. And to avoid relying only on data from a single institution, they additionally used data from ADNI (Alzheimer’s Disease Neuroimaging Initiative), a multi-center U.S. study, to check whether the same approach works in different groups as well.

The central material of the study was p-tau217 measured in plasma. However, the key was not to look at the p-tau217 number once and say “high/low.” The researchers sought to capture how the blood signal changes over time by including measures such as %p-tau217, an index that reflects the ratio of phosphorylated tau to non-phosphorylated tau. If that trajectory has a consistent form, a current blood value can be converted into “about where I stand on the timetable of disease progression.”

The next step is the “clock.” Based on repeatedly measured blood data, the researchers estimated when each individual transitioned out of the normal range and became classified as “positive” (more precisely, the age at which the transition likely occurred), and then tied that transition point to the time when symptoms actually began in clinical assessments as a model. As a result, the symptom onset time predicted by the “blood clock” was presented with an error range on the order of a few years.

This is because in Alzheimer’s, the question people truly want to ask is closer to “When will it come?” than “Do I have it?” When you hear “You are at risk” at a hospital, what changes your life is not the existence of risk itself but your sense of time. Depending on whether it is 20 years away or 3 years away, plans for work, retirement, insurance, savings, and family caregiving change. Medicine is the same. If there are treatments or interventions that slow progression, then “When should we start for it to matter?” becomes the central question of care.

There has already been research that tries to deal with “when.” Models have emerged that use expensive tests such as amyloid and tau PET scans or cerebrospinal fluid to precisely separate disease stages and estimate prognosis, but it has been difficult to apply them broadly because of cost and accessibility. The message of this approach is simple: let’s pose the same question with a tool that can be applied far more widely. When a blood test combines with the language of “prediction,” Alzheimer’s may move from a specialized test available only in certain hospitals to a strategic tool for clinical trials and everyday clinical care.

The key of this study is how it handles the “turning point.” For each person, p-tau217 rises gradually over time, and at some point it can cross the normal range and be classified as “positive.” The problem is that this transition is not stamped on a calendar as a single moment. Usually, the transition occurs somewhere between the last negative (normal) test and the first positive test. In other words, the transition time is observed not as a point but as an interval.

The researchers used this interval information to build a model that estimates each individual’s transition age. They then validated how well the model’s estimated transition age matched the observed interval (between the last negative and the first positive). They also performed cross-validation to check whether a clock built from one dataset also works in another, confirming that it was not an overfitted clock that works only in a specific group.

A clinically important observation followed as well. Even among people who are “positive,” the time to symptoms can differ depending on the “age at which they became positive.” A tendency was observed that when people become positive at an older age, the interval until symptoms is shorter. Accordingly, the model expands beyond simply identifying the transition point, incorporating age effects to estimate “symptom onset time.” In short, this clock is not a prophecy from a single indicator, but a tool that attempts to calculate a timetable created by a person’s age and blood signals.

It would be difficult for such a predictive model to become a tool that immediately confirms for an individual, “You will develop the disease in X years.” There is error, and individual lives are sensitive to error. But clinical trials are different in nature. A clinical trial is not about declaring an individual’s fate; it is about designing “which people to recruit at what time.”

The greatest cost of preventive clinical trials is not the drug but participant selection. You have to find, among people who are still functioning normally, the group that is likely to transition to symptoms in the near term. If the blood clock can find this group more efficiently, trials become shorter and cheaper. Then the development speed of preventive therapeutics can also accelerate. When trials change, clinical care follows. As evidence accumulates that “intervening at this timing has meaning,” the clinic’s core question shifts from “diagnosis” to “timing of intervention.” Ultimately, the change this research triggers is not a testing fad, but a timetable for how medicine deals with Alzheimer’s.

Even so, it is risky to imagine this technology immediately entering mass screening.

First, more validation is needed across diverse populations. A model that fits one cohort well is not guaranteed to work identically across the entire population.

Second, the way results are communicated matters. Prediction involves uncertainty. If uncertainty is not explained properly, prediction becomes fear rather than information.

Third, there must be a pathway after a positive result. If a blood test indicates high risk, what happens next, which additional tests confirm the diagnosis, what lifestyle or pharmacological interventions are recommended, and at what interval follow-up occurs must all be designed as a single service. Prediction does not end with a test. Prediction demands pathway design.

Finally, ethics and social rules follow. Alzheimer’s prediction is medical information and also future information. It can create ripple effects in insurance, employment, and family relationships. Therefore, this technology is not completed only inside hospitals. Society must evolve together, including the rules for handling prediction. Even so, one thing is clear. Alzheimer’s is increasingly moving from being “a disease of memory tests” to “a disease of managing time through biomarkers.”

Nature Medicine, 2026-02-19, “Predicting onset of symptomatic Alzheimerʼs disease with plasma p-tau217 clocks”, Kellen K. Petersen; Marta Milà-Alomà; Yan Li; Lianlian Du; Chengjie Xiong; Suzanne E. Schindler; et al.