It takes about ten minutes to fill in the Pittsburgh Sleep Quality Index. Nineteen questions, a pen, a kitchen table: what time did you go to bed in the past month, how long until you fell asleep, what time did you get up. Nobody asks what happened in the dark. Nobody asks whether you dozed off in the recliner at four in the afternoon, whether you took something to help, whether you stopped breathing forty times an hour. Those ten minutes are what much of the world's sleep-and-dementia research is built on, and a new paper from the 1Florida Alzheimer's Disease Research Center takes them about as far as they can honestly go.
Writing in Alzheimer's & Dementia on June 3, 2026, Layaly Shihadeh, Monica Rosselli and colleagues used PSQI answers from 326 older adults enrolled at the center's Florida sites: 113 cognitively normal, 192 with mild cognitive impairment, 21 with dementia, mean age 66.4 years. Clinical and functional status was tracked with the Clinical Dementia Rating Sum of Boxes, the measure that scores memory, judgment, home life and personal care and gets used as an endpoint in drug trials. The question was not whether sleep and cognition travel together, which is old news. It was whether Alzheimer's biology changes the exchange rate: whether the same reported sleep buys you a steeper slope if amyloid PET, plasma p-tau217 or a smaller hippocampus says the disease is already underway.
It did, in the cross-sectional snapshot. Longer self-reported sleep duration and a later wake time were associated with worse CDR-SB scores, and those associations were stronger in people with higher amyloid, higher p-tau217 and smaller hippocampal volume. Over time, the pattern held up rather than dissolving: longer sleep duration predicted faster worsening on CDR-SB (time by sleep duration interaction, B = -0.15, p = 0.031), and the three-way interaction of time, wake time and amyloid was significant (B = 0.003, 95% confidence interval 0.002 to 0.004, p < 0.001), with the steepest trajectories in people who both woke later and carried more amyloid. This is an observational cohort, not an experiment, and nothing here shows that sleeping longer causes decline. The arrow could easily run the other way, with lengthening sleep and drifting wake times being an early symptom of the same disease the biomarkers are detecting.
Here is the part that belongs in this section. The finding that survived was not the global sleep score everyone quotes; it was two narrow components, duration and timing, out of seven the questionnaire collects, screened first by correlation with a Bonferroni threshold of p < 0.007 and then carried into regression models adjusted for age, education and depressive symptoms, with false discovery rate correction. Meanwhile the authors list plainly what the instrument does not capture: daytime napping is not in the PSQI, and sleep medication use was not examined, though both could move cognitive outcomes. There was no actigraphy here and no overnight sleep study. The biomarkers, for all their precision, were measured once, so the paper cannot follow amyloid and tau and sleep as they change together. Retention thinned as such studies do: 182 participants (55.8 percent) reached a second visit, 74 (22.7 percent) a third. Education stood in for socioeconomic status. The dementia subgroup, 21 people, is too small to carry firm estimates on its own.
What does a family do with that? Not very much on a Tuesday night, and the paper does not pretend otherwise. But there is something useful in the shape of it. If biology really does set the exchange rate, then two people reporting identical sleep on identical forms are not in identical situations, and the thing that distinguishes them is sitting in a PET scanner and a blood tube. That is an argument for sleep questions being asked properly in biomarker-characterized cohorts, with the naps and the pills and the apnea written down, rather than an argument for changing anything about your own night. The same center has separately reported, in a conference abstract published in the journal in January 2026, that sleep apnea status altered how strongly amyloid PET tracked with plasma p-tau217 in 288 of its participants, and only in non-Hispanic/Latino participants at that: another hint that the untested variables are doing real work.
The authors say external validation in independent cohorts is needed before any of this generalizes beyond three Florida sites. That is the correct place to stop. What they have delivered is a testable model with longitudinal follow-up attached, which is more than the field usually gets, and a very clear inventory of the metrics that were missing while the test was being run.