Understanding How Blood-Based “Clock Model” May Predict Alzheimer Disease Symptoms

Emerging research is providing new insight into the ability of blood-based biomarkers to predict not only the risk of Alzheimer disease (AD) but also the timing of symptom onset. In a recent study from the Foundation for the National Institutes of Health (FNIH) Biomarkers Consortium published in Nature Medicine, investigators developed a statistical “clock model” using plasma levels of p-tau217, a protein associated with Alzheimer pathology, to estimate when individuals may begin experiencing cognitive symptoms.

By analyzing longitudinal blood samples from more than 600 adults who were initially cognitively unimpaired, researchers found that increasing levels of plasma p-tau217 followed a consistent trajectory that could be used to estimate the onset of AD-related symptoms with an average margin of error of approximately 3 to 4 years.¹The study also introduced a web-based application designed to help researchers visualize how p-tau217 levels change over time and how these changes relate to symptom development.

Susanne Schindler, MD, PhD, an associate professor of neurology at Washington University in St. Louis and Kellen Petersen, PhD, an instructor in the Department of Neurology at the same institution, were among the investigators involved in this research. In an interview with NeurologyLive^®, the 2 clinicians discussed the key findings of the study, the potential role of the p-tau217 clock model in estimating the onset of Alzheimer disease symptoms, and how these tools could help improve the design of future clinical trials.

NeurologyLive: What are the key takeaways from predicting the onset of symptomatic AD with the plasma p-tau217 clock study?

Susanne Schindler, MD, PhD: What we found is that after blood levels of a protein called p-tau217 increase beyond a certain point, they continue to increase in a consistent manner across individuals. That allows us to build a clock that relates blood p-tau levels to time and we can use that clock to estimate when blood p-tau217 levels reach positivity, an abnormal level that would be used in a clinic for diagnosis.

We also found that the age when people become p-tau217 positive is associated with the age they develop symptoms of AD, and that there is a relationship between this age and how long it takes to develop symptoms. If you're younger, it takes longer to develop symptoms, and if you're older, people develop symptoms after a shorter time. For those who progress from cognitively unimpaired to cognitively impaired, we could predict, with an error of about 3 to 4 years, when those symptoms would occur.

We did further models showing that when you looked across individuals, both those who developed symptoms and those who did not, the age of p-tau217 positivity was associated with development of AD symptoms. Overall, this suggests the feasibility of using a blood test to estimate when individuals will develop symptoms of the disease. There is still a wide error margin, so it's not good enough to use on an individual level. However, it is an error margin that could potentially make it useful for clinical trials.

We see that studies of individuals who have mutations that lead to AD show a similar error range, and those trials are much more efficient because they can roughly estimate symptom onset. We think at this point, the major application is clinical trials, but we’re interested in using additional biomarkers and clinical cognitive data to make these models more accurate.

What is the importance of conducting research of this nature?

Susanne Schindler, MD, PhD: The importance is that we have learned the further people are along with having symptoms of AD, the more progressed their symptoms are, and the harder it is to intervene and make a meaningful difference in disease progression. One way this is meaningful is that it allows us to look at the period before people develop symptoms and estimate where people are along the timeline. We know that amyloid plaques and tangles accumulate over many years, so there's a very long period when the disease biomarkers are positive, but people aren't yet symptomatic.

However, it has been quite uncertain where people are in that timeline. We knew they were at higher risk for developing symptoms, but this provides a time parameter for how far they are from symptoms. If we are able to further develop this, you can imagine that being ten years from developing symptoms is a very different thing to consider than being two years from developing symptoms. There is tremendous interest now in thinking about how we can prevent symptoms, both from the perspective of non-pharmacological interventions like exercise and lifestyle modifications, and pharmacological treatments.

This helps put into perspective where people are in terms of how far they are away from symptoms. Right now, the primary application is research and clinical trials, but the hope is to refine these models so that they could eventually become clinically meaningful.

Can you demonstrate the clock test and explain how it works?

Susanne Schindler, MD, PhD: We have the paper, but the paper isn't necessarily very user-friendly in terms of how all this works. Dr. (Kellen) Petersen created a web-based application meant for researchers to explore the data sets and analyses. We wanted clinicians to be able to access these models readily, in part to validate them and understand them, and with the hope that they could be applied more widely.

Kellen Petersen, PhD: The app lets you choose the model and input individual-level data, then the outputs are shown in the top left. It shows where an individual would be based on the input data, where they fall along this clock, and where they fall at a later follow-up period. There's also a Cox model plot that places where the person would fall in terms of likelihood of remaining cognitively unimpaired. At the bottom, it provides a summary output, including an estimate of age of symptom onset, the probability of being symptomatic at the time of plasma collection, and the probability at follow-up.

We modeled two different data sets, ADNI and Knight ADRC, and based on the cohort selected, users can choose from several assays, including the FDA-approved p-tau217/β-amyloid 42 ratio. You can input someone's plasma level and age, and the model estimates when they might have become biomarker positive and how far they are from symptom onset.

There were also two modeling approaches used to develop the clock. With those choices, you're able to enter values. For example, suppose someone had a plasma collection at age 80 and their value was 7% p-tau217. Then we might want to estimate where they would be on the clock and their probability of being symptomatic at another time.

Based on these values, the estimate is that this person would have been p-tau217 positive at age 68, which would have been about 12 years before and about 17 years before their follow-up time. That’s reflected on the clock. The red line in the application is the clock that relates the biomarker level to time since biomarker positivity. In the app, you're able to adjust the levels and see how that would change where the individual falls on the clock.

Susanna Schindler, MD, PhD: This is really meant for researchers to explore the data sets and analyses. We wanted clinicians to be able to access these models pretty readily, in part to validate them and understand them, and with the hope that they could be applied more widely. At this point, we wouldn’t recommend that people go get these tests and input their values to make personal decisions; that’s not what we want.

But many researchers around the world have data on these blood tests, and we would like them to be able to plug in their blood test values, understand where people are on the clock, and see how accurate our estimates are. It provides a way to look at our models. At this time, we know there is still a fair amount of variance in the estimates, but we think the error range would still be useful for research and clinical trials.

We also hope this encourages more people to work on this problem, because knowing when someone is likely to develop symptoms is a really important question that hasn’t been adequately explored. We also posted all the code used to develop the models, and the ADNI data set is quite easily available. That allows people to replicate what we’ve done and then take it further.

What makes this test unique?

Susanne Schindler, MD, PhD: People have been developing clock models for AD pathology for about 15 years, but this is the first time these types of models have been widely available for plasma blood tests, which is really unique. And this isn’t just a single blood test; it’s using data from several blood tests to estimate approximately when someone might develop symptoms.

Previously, blood tests have been used to determine whether someone has AD pathology in their brain. In research studies, we’ve looked at how these blood tests can predict risk for developing symptoms of AD. What’s unique here is estimating not just if someone might develop symptoms, but when.

What do you think this means for the future of predicting the onset of symptomatic AD?

Kellen Petersen, MD, PhD: We’re pretty confident that we can improve these models. Some ways to do that include better modeling approaches and incorporating other biomarkers, like imaging biomarkers, cognitive measures, and additional blood tests. We think that these can be improved and are pretty optimistic about it.

Susanne Schindler, MD, PhD: The goal is eventually to get these predictions to a point where they could be accurate enough for individual-level prediction. That’s one reason we made the application and shared the code. Different researchers may have different ideas, different approaches, and different biomarker modalities available to them. This is essentially version one using a single biomarker, but it seems likely these predictions could become much better in the future.

Transcript edited for clarity.

REFERENCES
1. Schindler SE, Peterson KJ, Li Y, et al. Plasma p-tau217 trajectories predict timing of symptomatic Alzheimer disease onset using a statistical clock model. Nat Med. 2026;32:1095-1094. doi:10.1038/s41591-026-04206-y