By age 61, 6% of these adults already had Alzheimer's pathology in their blood. None of them had dementia. Most of them had no idea anything was wrong.
A new study from Kristine Yaffe and her team at UCSF, published in The Lancet, used blood biomarkers to detect Alzheimer's-related amyloid and tau changes in 1,350 dementia-free middle-age adults from the CARDIA cohort. The findings reshape how we should think about the disease's timeline.
Five years after the baseline blood draw, the adults who tested positive had more than twice the risk of rapid decline in verbal memory and nearly four times the risk of rapid decline in processing speed.
That isn't "early." That is decades early.
What the CARDIA Cohort Showed
CARDIA, the Coronary Artery Risk Development in Young Adults study, has tracked U.S. adults since 1983. The Yaffe team measured plasma phosphorylated tau 217 (p-tau217), amyloid-beta 42, and amyloid-beta 40 using the FDA-approved Fujirebio Lumipulse assay. The most clinically informative result comes from the ratio of p-tau217 to amyloid-beta 42, which captures both the production of tau pathology and the brain's failure to clear amyloid.
Mean age was 61. Forty-five percent of participants were Black, 55% were white, 58% were women. Six percent tested positive for Alzheimer's pathology by the p-tau217/Aβ42 ratio. Fifteen percent were positive by the Aβ42/Aβ40 ratio alone. Four percent by p-tau217 alone.
Now look at the baseline cognitive testing. The pathology-positive group already scored lower on processing speed and executive function. Their global cognition was normal. Their memory was normal. But the timed, frontal-system tests, the ones that depend on a brain that can move information cleanly and quickly, already showed the strain.
Five years later, the gap widened. Rapid decline in verbal memory: OR 2.44 (95% CI 1.16–5.13). Rapid decline in processing speed: OR 3.98 (95% CI 1.71–9.3). The associations were stronger in women, in Black participants, and in carriers of the APOE4 allele.
I wrote earlier this year about a Washington University paper using the same p-tau217 biomarker to estimate symptom onset 15 to 20 years in advance. The CARDIA study extends that work into a younger window. Together they redraw the disease curve: amyloid and tau changes are not the late event we used to imagine, they are a midlife event with a long preclinical tail.
Why "Subtle" Doesn't Mean "Wait and See"
The pathology-positive group looked fine on a screening visit. A 30-minute MMSE or MoCA would not have caught them. Their referring physician likely saw a healthy adult, took a normal history, and ordered no further workup.
But the blood already knew.
In my practice I see executives in their late 50s and early 60s who say things like: "I'm fine, I just can't multitask the way I used to." Or: "I have to read emails twice." Or the spouse who walks in and says, "He's slower." The conventional neurology response is reassurance and the offer of a future visit. That window, the years between subtle slowing and a clinical diagnosis, is precisely when intervention has the most effect on the disease's trajectory. That is also exactly the window CARDIA just lit up.
This is why precision diagnostics matter. A normal MRI and a normal cognitive screen do not rule out Alzheimer's pathology in midlife. The Intensive Brain Health Program exists to run the workup conventional care doesn't: plasma p-tau217, amyloid ratios, neuroinflammation markers, vascular risk panel, APOE status, sleep architecture, hormonal and metabolic substrates. The cost of catching a 6% case at 61 is trivial compared to the cost of catching it at 73, by which point the structural damage is irreversible.
The False-Positive Problem the Editorial Names
The accompanying Lancet editorial by Tiia Ngandu and Anna Rosenberg of the Finnish Institute for Health and Welfare made an important point. In a population with low pre-test probability (healthy 50-year-olds with no symptoms), the positive predictive value of these blood tests drops. False positives go up. They argued, correctly, that the assays are not yet appropriate for untargeted, community-wide screening.
That nuance is being lost in the direct-to-consumer ad market. Yaffe herself raised the concern: "The tests are supposed to be for those with symptoms but many people are getting these tests without symptoms off-label."
The answer is not to abandon the biomarkers. The answer is to use them inside a real clinical context, alongside cognitive testing, vascular risk profiling, APOE status, and structural imaging. A p-tau217 result is a data point, not a diagnosis. Used inside a precision workup it tells you which patients need aggressive risk modification now. Used as a stand-alone direct-to-consumer test, it generates anxiety and confusion in roughly proportionate measure.
What This Means for the People Reading This
If you are in your 50s or early 60s and you have noticed subtle slowing (finding the right word takes a beat longer, you reread paragraphs, your reaction time on tennis or driving feels different), that is worth taking seriously. Most of the time it isn't Alzheimer's pathology. Sometimes it is. The CARDIA data shows that the subtle stuff has a signal underneath it that the right tests can find.
If you have a family history of dementia, APOE4 status, hypertension, insulin resistance, or untreated sleep apnea, the urgency is higher. Those are the modifiable upstream contributors that translate biomarker positivity into clinical decline. Catch the biology in the window where the brain can still recover, and the trajectory bends. Some of that recovery work is metabolic and pharmacologic, some of it is nutritional. Action Potential Supplements exists for the nutritional layer of that work.
The Neuroeconomy thesis sits underneath this. Your brain is the asset that compounds, or depreciates, through the second half of your career. The CARDIA finding is that the depreciation is measurable before it is visible. That is the period that matters. Brain protection in midlife is the highest-yield investment a high-performer can make, and the data now supports moving on it sooner than the field has been willing to admit.