The measurement that produced one of the more talked-about prevention findings of the past year took seven days. Participants in the Harvard Aging Brain Study clipped an Omron pedometer to the waistband, wore it during waking hours, and handed it back. Days registering under 100 steps or over 30,000 were thrown out as bad data, and anyone with at least four usable days was counted in. That single ordinary week, for 296 cognitively unimpaired adults aged 50 to 90, was then set against the slowest possible outcome measure: annual cognitive testing and repeated PET scans running from April 2010 to February 2025, up to 14 years, with a median of nine years of cognitive follow-up. The results were published in Nature Medicine on 3 November 2025 by Wai-Ying Wendy Yau, Jasmeer Chhatwal and colleagues at Mass General Brigham.

What the team found is unusual because of where the signal landed. Among people whose baseline amyloid PET was elevated, that is, people in the preclinical stage of Alzheimer's with plaques but no symptoms, higher baseline step counts were associated with slower decline on the PACC-5 cognitive composite and slower progression on the Clinical Dementia Rating Sum of Boxes. Higher activity was not associated with less amyloid, at baseline or over time. Instead, in the subset of 172 participants who had repeat tau scans (a further 241 had repeat amyloid scans), higher step counts were associated with slower accumulation of tau in the inferior temporal cortex, one of the first neocortical places tangles appear. Formal mediation analysis put a number on it: at elevated amyloid, slower tau accumulation accounted for 84 per cent of the association between activity and slower PACC-5 decline, and 40 per cent of the association with slower functional decline. Activity was tracking a step inside the pathological cascade, not sitting beside it.

Then there is the dose. The authors split the cohort into four bands: inactive at 3,000 steps a day or fewer, low activity at 3,001 to 5,000, moderate at 5,001 to 7,500, and active at 7,501 or more. Among those with elevated amyloid, estimated PACC-5 scores over the median nine years fell by 2.5 points in the inactive group, 1.5 in the low activity group, 1.1 at moderate activity and 1.2 in the most active group: decline lower by 40, 54 and 51 per cent against the inactive band. The functional scores moved the same way, rising 2.2 points in the inactive group and 1.1 in the most active. The curve bends, then flattens. Alzforum, reading the study tables, noted the cohort's own average was 5,719 steps a day, which is to say the plateau sits roughly where ordinary daily life already sits for many people in their seventies, and well below the 10,000 that fitness trackers have made a moral standard.

This matters for families because prevention advice has for years stopped at a verb. Move more. Here is a dataset that attaches a range and a mechanism, and that is precisely why it is being read as the evidence base under the lifestyle arm of the current generation of combination prevention trials. But the boundary needs saying in the same breath: this is a longitudinal observational cohort, not a trial. Nobody was assigned to walk. Step counts came from one week at baseline, and people who walk more differ from people who walk less in a hundred ways that no model fully removes. Independent commentators making that point at publication included Masud Husain of the University of Oxford, who noted the cohort was relatively small and the study was not designed to measure the effects of exercise, in remarks gathered by the UK Science Media Centre.

The trial evidence, so far, is more sobering. In the POINTER Imaging ancillary study, published in JAMA Neurology in June 2026, 983 participants from the US POINTER trial underwent MRI or PET scanning, and after two years there were no differences between the structured and self-guided lifestyle arms in amyloid burden, entorhinal tau, hippocampal volume or white matter hyperintensity volume, even though the parent trial (JAMA, 28 July 2025) had found a cognitive advantage for the structured program. Two years may simply be too short to move a biomarker that the Harvard cohort watched drift for a decade. And a Rush Memory and Aging Project analysis in The Lancet Healthy Longevity in March 2026 sharpened a different worry: baseline accelerometer activity was not associated with incident dementia at all (hazard ratio 1.00), while activity measured at years two, four and six was (hazard ratio around 0.61), a pattern the authors read as evidence that the timing of any benefit, and the possibility that early disease itself reduces movement, still needs defining.

So the honest picture is a strong, biologically coherent association in a well-characterised cohort, sitting next to randomised imaging data that has not yet confirmed it and epidemiology that keeps raising reverse causation. For someone living with an elevated amyloid scan and no symptoms, the useful takeaway is not a step target issued by a newspaper. It is that the number the researchers found meaningful was modest, that the curve flattened rather than kept climbing, and that the whole question of what any of it means for one person's walking, mobility, joints and heart belongs in a conversation with their own clinician.