Design: pooled participant-level analysis of seven prospective cohorts, with survival and random survival forest models
Population: 8,582 adults, mean age 70, 66% female; 16% Black, 47% non-Hispanic White, 37% Hispanic or other
Primary outcome: cognitive impairment, defined as mild cognitive impairment or dementia
Effect: incident impairment HR 1.41 (95% CI 1.22–1.64) per SD; 1.76 (1.36–2.26) in APOE-e4 carriers vs 1.26 (1.12–1.42) in non-carriers

Plasma p-tau217 is moving from research tool towards clinical use, and the question this analysis asks is whether the number means the same thing in everyone. Participant-level data were pooled from seven multi-ethnic prospective cohorts in Canada, the Dominican Republic and the United States, collected between 1992 and 2025: 8,582 adults at baseline, mean age 70, of whom 4,569 had no cognitive impairment and entered the longitudinal analysis. Sixteen per cent were Black and 37% Hispanic or of other ethnic origin, which is unusual for this literature and part of the point.

Higher p-tau217 was associated with cognitive impairment at baseline (odds ratio 1.77, 95% CI 1.42 to 2.20) and with incident impairment (hazard ratio 1.41, 1.22 to 1.64 per standard deviation). Stratified by genotype, the association was stronger in APOE-e4 carriers — odds ratio 2.25 (1.52 to 3.34) against 1.52 (1.35 to 1.72), and hazard ratio 1.76 (1.36 to 2.26) against 1.26 (1.12 to 1.42). Each standard deviation rise shortened time to impairment by 24% in carriers and 13% in non-carriers. Survival curves separated three to four years after the blood was drawn, before any symptoms.

The practical consequence is that a p-tau217 value cannot be reported as a single risk figure. The same concentration carries a different prognosis and a different expected interval depending on genotype, which means a service offering the test needs to decide in advance whether it will also genotype, and what it will tell someone who has not yet developed symptoms. That is a counselling problem before it is a laboratory one.

Decide, before offering p-tau217, whether APOE genotyping will accompany it
Do not quote a single risk figure from a p-tau217 concentration alone
Explain that the horizon is years — curves separated three to four years after testing
Establish what the person wants to know before the result exists, not after
Remember that this cohort was recruited in North America and the Caribbean; Indian data are absent

Why it matters

The same biomarker value predicts a different risk and a different timeline depending on genotype, which a single reference range cannot convey.

The statistics, in plain English

These are hazard and odds ratios per standard deviation, not thresholds, so they describe a gradient rather than a positive-or-negative test. Note that the carrier and non-carrier intervals overlap in places — 1.52 to 3.34 against 1.35 to 1.72 — so the difference in strength is real but not as separated as the point estimates suggest. The timing figures, 24% against 13% shorter time to impairment, are the more useful ones in clinic, because what people ask about is when, not whether.