Draw the usual picture of Alzheimer's biology and you get a smooth ramp: amyloid creeping up over twenty years, tau following, memory giving way at the end. It is a useful picture and it is also a little too tidy. In the April 7, 2026 issue of Alzheimer's & Dementia, a team led by first author Mengyao Hu with senior author Jonathan Graff-Radford, working with the Mayo Clinic Study of Aging, asked a blunter question of the same data: if you let the statistics look for the exact age where a line changes slope, where does it find one? The answer was not a single ramp. It was a set of hinges, and they do not all open at once.
The analysis covered 2082 participants in the Mayo Clinic Study of Aging, a population-based study of residents of Olmsted County, Minnesota. Median age was 71, with half the group between 62 and 79; 87 percent were cognitively unimpaired at the time of testing, 11 percent had mild cognitive impairment, and 22 people, about 1 percent, had dementia. Twenty-eight percent carried an APOE e4 allele. The researchers measured plasma amyloid beta 42/40, p-tau181, GFAP and neurofilament light chain, plus amyloid PET, tau PET, hippocampal volume adjusted for head size, and a global cognition score. Then they used generalized additive models to describe the age curves and Davies' test with piecewise regression to locate the breakpoints. One crucial detail: this was cross-sectional. Each person contributed a snapshot, and the curve is built across people of different ages, not by following one person from 50 to 85.
The hinges landed in a pattern. Plasma amyloid beta 42/40 broke earliest, at 48.5 years, and then declined steadily from there. Global cognition changed slope at 59.8 years (95 percent confidence interval 55.4 to 66.0). Amyloid PET followed at 62.3 years (56.2 to 69.3). Then came a cluster: p-tau181 at 67.2 years (60.3 to 70.3), GFAP at 68.1 (63.5 to 71.8), hippocampal volume at 68.1 (63.1 to 71.9), and neurofilament light chain at 70.7 (65.9 to 75.6). Four markers of tau, glial response and neurodegeneration all bending inside roughly a five-year window around age 70. Tau PET, notably, did not bend at all; it rose with age at a steady rate, with no breakpoint the models could support.
A second check used a subsample of 462 participants whose blood had been run on C2N's mass-spectrometry platform rather than Quanterix's immunoassay. The broad story held, and the authors note the findings replicate across the two platforms, though the ages shifted later: both p-tau217 and p-tau181 broke at 72.6 years, while their ratio measures and the mass-spectrometry amyloid beta 42/40 showed no clear breakpoints. When the team repeated the models in cognitively unimpaired participants only, most breakpoints stayed within the original confidence intervals. The biggest move was neurofilament light chain, which shifted from about 70.7 years to 61.6 (57.2 to 65.9) once impaired participants, who skew older, were removed. That sensitivity analysis is a reminder of how much these ages depend on who is standing in the room.
For a family reading this, the temptation is to hear a countdown, and that is the wrong reading. These are population averages, the age at which the group's curve changes direction, and they say nothing about when any individual's blood marker will move or whether it ever will. Half the people in this cohort were over 71 and cognitively fine. What the numbers do offer is a sense of sequence: the blood amyloid ratio shifting decades before anyone would think to look, cognition and amyloid imaging bending around 60, and the markers of injury and glial activation clustering near 70. The authors' own framing is cautious, concluding that breakpoints cluster near late midlife and suggest windows for screening and monitoring. That is a hypothesis about when to measure, not a schedule for anything else, and the paper does not test whether screening or intervening at any of these ages changes what happens next.
The cohort's shape also matters. The Mayo Clinic Study of Aging draws from one Minnesota county, overwhelmingly cognitively healthy at enrollment, with a median of 14 years of education and very few people with advanced dementia. Blood biomarker research has leaned for years on exactly this kind of sample, which the authors themselves flag as a limit on generalizability. Whether the hinge near 70 sits at 70 in a different county, a different country or a different population is an open question, and one worth asking before anyone turns these ages into a protocol.