The device weighs almost nothing. Participants at the Baltimore Longitudinal Study of Aging were fitted with an Actiwatch-2 on the non-dominant wrist just before leaving the clinic, told to keep it on for seven days, and asked to fill in a sleep diary morning and evening: when they got up, when they got into bed meaning to sleep. Then they went home and lived normally. The watch counted movement. It did not know it was measuring anything about a brain.
What researchers led by Marc Kaizi-Lutu and Daniel Callow at the Johns Hopkins Bloomberg School of Public Health, with colleagues at the National Institute on Aging Intramural Research Program, did next was set those movement counts against magnetic resonance imaging of the regions Alzheimer's disease tends to reach early: the parahippocampal gyrus, the hippocampus, the amygdala. The sample was 344 cognitively unimpaired adults aged 50 and over, mean baseline age 72.9 years, 57.3 percent women, 22.4 percent identifying as Black or African American, each contributing an average of 5.56 nights of usable actigraphy. Roughly two thirds went on to have at least one follow-up scan; among those, the mean follow-up was 1.18 years across an average of 2.27 visits. The paper appeared in Alzheimer's & Dementia on 14 April 2026.
The rhythm was described with three plain numbers: how strong the day-night contrast was (relative amplitude), how much the rhythm broke up within a day (intradaily variability), and how closely one day resembled the next (interdaily stability). Cross-sectionally, weaker and more fragmented rhythms went with smaller medial temporal volumes. Longitudinally, more regular and less fragmented rhythms went with better maintenance of amygdala volume over the follow-up. None of this is causation, and the authors do not claim it is. It is an association measured at one moment in a group of volunteers who were, by design, cognitively healthy when the watch went on.
The age finding is the part that makes this different from the rest of the circadian literature. Rest-activity rhythms showed no significant cross-sectional association with ventricular volume, the space that widens as brain tissue is lost. Over time, though, greater fragmentation predicted faster ventricular enlargement (B = 1.47, p = 0.022), and age changed the size of that link: the interaction was significant (p = 0.010), and fragmentation predicted faster ventricular expansion at ages 75 and 85 but not at 65. The same wrist number, in other words, appeared to carry different structural weight depending on how old the wrist was. Kaizi-Lutu offered one reading in the university's announcement, suggesting there may be "a spiral of neurodegeneration and circadian disruption" that intensifies with age. That is a hypothesis about mechanism, not a demonstration of one.
The authors are unusually direct about what would undercut them. Several associations that looked significant before correction did not survive Benjamini-Hochberg adjustment for multiple comparisons. The study carried no amyloid or tau measures, so the pathology underneath the volume loss is unobserved. Crucially for an age-stratified claim, the BLSA scans people on different schedules (every four years under 60, every two years from 60 to 79, annually at 80 and over), so the youngest participants simply had fewer follow-up scans, which the authors say limited confident comparison between age groups. And actigraphy is a behavioural proxy: as the paper puts it, "RARs are not measures of the endogenous circadian rhythm", which would require an intensive in-laboratory protocol.
For a family reading this, the practical content is narrow and worth stating without decoration. Nothing here identifies an individual at risk, and a fragmented week on a wrist monitor is not a finding about anyone's brain. What the work does is argue for a metric that is cheap, passive, and currently missing from most clinical pictures: seven days of movement, collected without a clinic visit, that behaves differently against brain structure at 85 than at 65. Whether that difference is a signal worth acting on is a question for larger samples with longer follow-up, repeated actigraphy, and amyloid and tau data. It is not a question a watch can answer yet.