Picture a woman of 36 with Down syndrome at a routine appointment. She is working, she is well, nobody in the room is thinking about dementia. Her bloods are unremarkable. Now run the same appointment two or three birthdays later and, according to a cohort study published in eBioMedicine on 11 July 2026, one number on that panel has quietly begun to lift away from the numbers of people her age without trisomy 21. Javier Arranz, Juan Lantero-Rodriguez and colleagues at Hospital de la Santa Creu i Sant Pau in Barcelona put the moment of departure for plasma p-tau217 at roughly 37.3 years, and for cerebrospinal fluid p-tau181 at roughly 38.1 years. Nothing has happened to her. Everything has started.

The study is cross-sectional, which matters and is stated here rather than buried: 461 participants drawn from the DABNI cohort of adults with Down syndrome and the SPIN cohort at Sant Pau, including cognitively normal euploid controls and people with sporadic Alzheimer's disease, each sampled once. The team fitted smoothed curves (LOESS) across age and asked when the Down syndrome curve stopped overlapping, on a 95 per cent confidence interval criterion, with the control curve. The answers arrive in a queue. Plasma p-tau217 first at about 37.3 years, then CSF p-tau181 at 38.1, CSF p-tau217 at 39.1, CSF NTA-tau at 39.6, plasma p-tau181 at 39.9, then CSF p-tau231 and p-tau205 around 40.8 and 40.9, CSF p-tau235 at 42.0 and plasma p-tau231 at 43.9. Frank abnormality, defined as two standard deviations above the control mean, came later and slowly: the earliest markers, CSF and plasma p-tau217 together with CSF p-tau181, crossed at around 42 years, and the whole set turned abnormal across roughly a twenty year span running from the fourth into the sixth decade.

Set that against the clinical clock and the shape of the problem is obvious. In the 2020 Lancet cross-sectional study of 388 adults with Down syndrome and 242 euploid controls, also from Barcelona and Cambridge, the median age of diagnosis was 50.2 years for prodromal Alzheimer's disease and 53.7 years for dementia. So the biology and the diagnosis are separated by something close to fifteen years of ordinary life: jobs, football, arguments about the remote control. That gap is not a tragedy in itself. It is the only room anyone has to work in.

Which is why the entry ages of trials are not administrative trivia. The ABATE study of the anti-amyloid vaccine ACI-24.060 enrols adults with Down syndrome aged 35 to 50, a window that, on these curves, starts just before the tau signal moves and closes long after it has. Alnylam announced on 15 July 2026 that it had begun APPlauDS, a global phase 2 study of intrathecal mivelsiran, an RNA interference therapy aimed at the APP gene, in adults who already have early-stage Down syndrome associated Alzheimer's disease. Enrolling people whose disease is already established is a legitimate and necessary thing to test. It is simply not the same experiment as intervening before the divergence, and this paper makes the distinction impossible to fudge.

There is a practical grace note for families and clinics in the rest of the findings. Blood held its own against spinal fluid. Plasma p-tau217 produced areas under the curve of 0.91 to 0.97 for the biological groupings, and in the symptomatic stages it did better than its CSF counterpart at telling prodromal disease from dementia (0.69 against 0.53, DeLong p = 0.019), because CSF markers flatten out once dementia arrives while plasma keeps moving. In a population for whom a lumbar puncture is a significant undertaking, a venepuncture that carries similar information is not a small thing. It is also an argument for a baseline drawn early, in the mid thirties, while the number still means nothing, so that a later number has something to be compared with.

None of this is a test a family can ask for and act on this afternoon, and none of it says that treating at 38 works. What it does is move the argument. For twenty years the debate in Down syndrome prevention has been whether to select participants by age or by biomarker. This cohort suggests the two questions collapse into one somewhere around the thirty-seventh birthday, and that a trial starting at 50 is not testing prevention at all.