We recently reported the news that Anthem would soon be covering FDA approved pTau217 blood tests, with conditions, as part of dementia work-ups in the clinic.
And today, I saw headline news at CNN that there's a new study, pTau217 is a strong marker of 5-year risk of developing MCI or dementia in healthy elders. The CNN article said that high pTau217 matched a 38% increased risk of dementia. I thought this sounded misleading; do I care if my ten year dementia risk is 10 chances ini 100 or 13.8 chances in 100? It turns out the CNN piece was misquoting, and pTau217 is actually a much stronger predictor (from 12% to 38% absolute risk or 3X relative risk).
Here's a dialog with Chat GPT to puzzle it out.
###
Can you find this paper? There is recent evidence that p-tau171 can be a marker for elevated dementia risk in “healthy elders,” somewhat as high blood pressure is a risk factor for stroke. I had seen a five-year figure of 38% and a ten-year figure of 78%. However, relative-risk numbers can be misleading: if my underlying chance were one in ten, a 38% increase would raise it only to about 1.38 in ten. The information may have come from a recent news report or perhaps an abstract at the upcoming international Alzheimer’s meeting.
Chat GPT
The paper can indeed be found, although the biomarker is p-tau217, not p-tau171. It is:
Buckley RF, Townsend DL, Birkenbihl CJ, et al. “Prognostic Value of Blood-Based P-Tau217 Levels for Progression to Cognitive Impairment.” JAMA. Published online July 14, 2026. doi:10.1001/jama.2026.12556.
The paper was published in conjunction with its presentation on July 15 at the 2026 Alzheimer’s Association International Conference in London. It pooled 2,684 people who were cognitively unimpaired at baseline across six longitudinal research cohorts. Their median age was 69.6 years, and median follow-up was 5.4 years. During follow-up, 478 participants reached the study’s composite endpoint of cognitive impairment. (JAMA Network)
The CNN sentence gets the central statistic wrong
My initial impression came from a CNN report by Sandee LaMotte, which said that cognitively healthy people with high p-tau217 had an “estimated 38% greater chance” of developing early signs of dementia over five years. As written, that is not what the study found. (Facebook)
The study found that people in the very-high p-tau217 group had an estimated 38% absolute probability of progressing to cognitive impairment within five years. It did not find merely that their risk was 38% higher than that of some comparison group.
That distinction is substantial. The estimated five-year risk in the low p-tau217 group was 12%, while the estimate in the very-high group was 38%. Therefore:
The absolute increase was 38% minus 12%, or 26 percentage points.
The risk ratio was 38 divided by 12, or approximately 3.2 times the risk.
Expressed as a relative increase, the very-high group’s risk was about 217% higher than the low group’s risk.
By contrast, a “38% greater chance” applied to a baseline risk of 12% would produce a risk of only about 16.6%:
12% × 1.38 = 16.6%.
Thus, the CNN wording does not merely simplify the result. It transforms a 38% absolute risk into an apparent 38% relative increase, which is a very different and much smaller effect.
There may be an identifiable source for the mix-up. The paper separately reports that each one-standard-deviation increase in p-tau217 was associated with a hazard ratio of 1.38 for progression. That means a 38% increase in the instantaneous event rate per standard-deviation increase, within the time-to-event model. It is not the same thing as saying that people with very high p-tau217 had a 38% increase in their five-year probability. The paper’s HR of 1.38 and its five-year absolute risk of 38% happen to contain the same digits, but they answer different statistical questions. (JAMA Network)
The complete risk table—and how many people were in each group
The study divided participants into four p-tau217 categories. These were not quartiles. The investigators standardized p-tau217 measurements within each cohort and chose thresholds corresponding approximately to biologically meaningful amyloid-PET Centiloid levels: 10, 25, and 60 Centiloids. Consequently, the four groups were unequal in size. (JAMA Network)
| Baseline p-tau217 category | Approximate amyloid-PET correspondence | Participants, N (%) | 5-year absolute risk | 10-year absolute risk |
|---|
| Low: <−0.5 SD | <10 Centiloids | 516 (19.2%) | 12% (95% CI, 9%–15%) | 40% (31%–49%) |
| Intermediate: −0.5 to <1.1 SD | 10–24 Centiloids | 1,087 (40.5%) | 15% (12%–17%) | 45% (37%–52%) |
| High: 1.1 to <2.5 SD | 25–60 Centiloids | 598 (22.3%) | 24% (20%–28%) | 62% (52%–69%) |
| Very high: ≥2.5 SD | >60 Centiloids | 483 (18.0%) | 38% (33%–43%) | 78% (69%–84%) |
The group sizes are important. The 38% estimate did not arise from a handful of extreme outliers: the very-high category initially included 483 participants. At the same time, the fact that 18% of this particular study population fell into the very-high category should not be interpreted as showing that 18% of ordinary cognitively healthy older Americans have such levels. These were selected Alzheimer’s research cohorts, not a representative population sample. (JAMA Network)
The table also clarifies another ambiguity in the CNN account. The formal high group had a five-year risk of 24%, twice the 12% risk in the low group. The 38% estimate belonged specifically to the very-high group. Referring simply to participants with “high levels” collapses two distinct categories.
At two years, the modeled absolute risks remained low in all four groups, ranging from approximately 1% to 4%. The sharper separation appeared over five years, and still more dramatically in the ten-year projections. (JAMA Network)
The endpoint was not dementia alone
The study’s outcome was broader than “developing dementia.” Participants were counted as having progressed if they developed any one of the following:
Mild cognitive impairment;
Dementia; or
Two consecutive global Clinical Dementia Rating scores of 0.5 or greater.
Therefore, “progression to cognitive impairment” is the accurate description. CNN’s phrase “early signs of dementia” is understandable as popular shorthand, but it blurs the distinction between MCI, a low CDR score, and established dementia. The reported 38% should not be presented as a five-year probability of receiving a dementia diagnosis. (JAMA Network)
The secondary cognitive analysis told a consistent story. Over five years, modeled cognitive performance was relatively stable in the low and intermediate groups, while the high and especially the very-high groups showed greater decline on the harmonized Preclinical Alzheimer Cognitive Composite. (JAMA Network)
Why the ten-year number is less secure
The 78% ten-year estimate is striking, but it is much less firmly supported than the five-year result. Only 139 participants—about 5% of the entire sample—had at least ten years of clinical follow-up, and most of the long-term information came from one cohort. (JAMA Network)
The survival figure shows how sparse the tail became. Of the 483 participants who began in the very-high group, 425 were still under observation and event-free at two years, 285 at four years, 143 at six years, 27 at eight years, and only 11 at ten years. The corresponding numbers at ten years were also small in the other groups: 19 low, 65 intermediate, and 16 high.
The 78% figure is therefore a model-based extrapolation informed by a dwindling number of long-observed participants. Its reported confidence interval—69% to 84%—does not fully communicate every source of uncertainty, including reliance on selected cohorts and the scarcity of ten-year observations.
The models also did not explicitly treat death as a competing risk because mortality information was not consistently available. Someone who died or left the study without developing cognitive impairment was censored. In an older population, failing to model death as a competing event may make long-range disease probabilities appear somewhat higher than the probability a living person would actually experience before death. (JAMA Network)
“Healthy elders” were cognitively healthy, but not necessarily population-typical
The participants were cognitively unimpaired at enrollment, but “healthy elders” can give an overly broad impression. The pooled population was assembled from six observational studies and clinical-trial cohorts. One component was the 952-person A4 prevention trial, which required participants to be amyloid-positive. Overall, 43% of the pooled study population was amyloid-positive at baseline. (JAMA Network)
That enrichment was useful scientifically because it provided enough people with substantial Alzheimer-related pathology to study subsequent progression. It also means the observed distribution of low, intermediate, high, and very-high p-tau217 should not be treated as a prevalence survey of the general older population.
The accompanying JAMA editorial noted additional limitations: the cohorts were generally highly educated and had limited racial, ethnic, and socioeconomic diversity. Medical conditions that can affect measured p-tau217 concentrations, including chronic kidney disease, were not incorporated fully into the prognostic estimates. Cognitive impairment could also arise from vascular disease, other neurodegenerative conditions, or mixed pathology rather than Alzheimer disease alone. (JAMA Network)
Is p-tau217 analogous to blood pressure?
The blood-pressure analogy is useful in one limited respect: both measurements may help stratify the future probability of a clinically important outcome in a person who currently has no symptoms.
But biologically they are not quite equivalent. High blood pressure is itself a causal and modifiable contributor to stroke risk. Elevated p-tau217 is better viewed primarily as an indicator that an Alzheimer-related pathological process—particularly amyloid-associated tau biology—may already be advancing. It is closer to a measure of underlying disease burden or disease stage than to a stand-alone risk factor imposed upon an otherwise pathology-free brain. The study found that p-tau217 retained prognostic information even after accounting for amyloid PET, suggesting that it captures an integrated aspect of advancing Alzheimer-related pathology. (JAMA Network)
For now, the most defensible application is research: identifying cognitively normal individuals at sufficiently high near-term risk to make Alzheimer prevention trials feasible. The authors and editorialists caution that the estimates are not yet precise or generalizable enough to tell an individual cognitively healthy person, based on a commercial blood result, that his or her personal five- or ten-year risk is a particular percentage. Current guidance continues to recommend against routine p-tau217 testing of cognitively unimpaired people outside research studies or clinical trials. (JAMA Network)
The bottom line is therefore considerably more dramatic—and more nuanced—than the CNN sentence. The study did not find a mere 38% relative increase. It estimated a 38% five-year absolute risk of broadly defined cognitive impairment in the very-high p-tau217 category, compared with 12% in the low category: a 26-percentage-point difference and roughly a threefold risk ratio. That is a substantial prognostic signal. But it was derived from selected, pathology-enriched research cohorts, the endpoint was broader than dementia, and the ten-year estimate rests on comparatively sparse follow-up.
Primary paper: Buckley et al., JAMA, July 14, 2026
Accompanying editorial: Schindler SE, Wolk DA. “Predicting Risk of Cognitive Impairment With Alzheimer Disease Blood Biomarkers.” JAMA. Published online July 14, 2026. doi:10.1001/jama.2026.13322.
