The Pupil Blinks, the Wrist Keeps Time

At a Copenhagen memory clinic, a handheld light and a week of wearable sensors were asked to forecast the next year of Alzheimer's disease. Neither asked the patient to remember a thing.

The test takes about as long as a yawn. A nurse holds a device the size of a small torch to the eye, a light pulses, and the pupil does what pupils do: it snaps shut and then creeps open again. Nobody is asked to recall three words, name the month, or copy a pentagon. At the Danish Dementia Research Centre at Rigshospitalet in Copenhagen, Mathias Holsey Gramkow and colleagues have spent several years asking a blunt question about that reflex and about the week of movement data gathered from body-worn sensors afterwards: can two signals a patient does not have to perform tell you where the disease is heading?

The pupil half of the answer is published and peer reviewed. In Alzheimer's Research & Therapy on 11 July 2025, the group reported an exploratory prospective longitudinal cohort study of early Alzheimer's patients examined with quantitative light reflex pupillometry (a handheld PLR-3000) at baseline, three months, twelve months and 18 or 24 months. A decrease in resting pupillary diameter was associated with a higher risk of clinically evaluated progression at one year, with an odds ratio of 4.3 and a 95 percent confidence interval running from 1.2 to 16.9. That interval is the story as much as the number is: the cross-validated area under the curve was 0.66, sensitivity 98 percent with a confidence interval of 15 to 100 percent, specificity 31 percent with an interval of 13 to 91 percent. A test with that specificity flags a great many people who will not progress. Separately, less relative pupillary change after the light stimulus at baseline tracked with decline on the Mini Mental State Examination (beta -5.1, 95 percent CI -1.6 to -8.6, p = 0.004).

The wrist half has a published diagnostic chapter and a newer, still preliminary prognostic one. In the same journal on 21 May 2025, the team reported a cross-sectional memory clinic study in which 70 patients with Alzheimer's, 29 with dementia with Lewy bodies, 23 with cerebrovascular disease and 48 healthy controls wore two SENS Motion sensors at home for a full week, yielding 510 activity features fed to a logistic regression classifier. Separating Alzheimer's from healthy controls was modest, at 68.8 percent accuracy. Telling one dementia from another did better, at 80 to 89 percent accuracy and 69 to 84 percent precision. The study was registered as IPATCH-AD (NCT05175664).

Then comes the piece that gives this angle its name, and it must be read for what it is: a conference abstract, published in the Alzheimer's & Dementia supplement on 24 December 2025, not yet a full peer-reviewed paper. In it the same single-centre cohort is reported with 86 patients in the pupillometry arm (mean age 75.7, mean MMSE 25.8) and 58 in the actigraphy arm (mean age 74.8, mean MMSE 25.5), followed for 18 to 24 months, with clinical progression, visually read [18F]FDG-PET and MMSE assessed at one year. A fall in resting pupil diameter between baseline and three months carried an odds ratio of 4.28 (95 percent CI 1.24 to 16.9) for clinical progression and correlated with cognitive decline (r = 0.17, p = 0.008); baseline relative pupil change predicted neuroimaging progression with an AUC of 0.64 (0.50 to 0.77). On the wrist side, intra-daily variability in step count predicted clinical progression (p = 0.016), lower evening and night activity predicted cognitive decline (p = 0.01), and falling intensity during the least active five hours of the day tracked MMSE decline over time (r = 0.20, p = 0.029). One caveat sits in the methods and deserves to be on the front page of any summary: the pupil features and the activity features were fitted in separate models. The two channels were measured in the same study, not fused into a single score.

Why a reader living with early Alzheimer's might still care is a matter of what these measures are made of. The light reflex is an autonomic circuit routed through the midbrain, which is why the authors treat it as a window on something other than cortical memory networks; actigraphy reads circadian robustness and fragmentation, the territory this section has followed all year through sleep and rest-activity work. Both are passive. There is no score to fail, no practice effect, no bad morning that becomes a bad result. For families who dread the clinic visit precisely because of the testing, that is not a small thing. It is also, for now, a research tool: the group's own scoping review in the Journal of the Neurological Sciences (15 February 2026) mapped the literature on wearable and portable digital biomarkers in early Alzheimer's across 3,019 people with MCI due to Alzheimer's and 1,942 with mild Alzheimer's, and found that rest and activity measures dominate the field at 39 percent of studies while only 3 percent reported anything about prognosis at all.

So the honest summary is narrow and useful at once: in one Danish memory clinic, a pupil that stops opening as wide and a day that loses its evening shape both leaned in the same direction as decline over the following year, measured without a single cognitive question. Whether that holds in another clinic, in another country, in a combined model, is work that has not been done yet.