Twenty years between the question and the blood test

A handful of yes-or-no questions about sleep, answered by 1,325 Americans at an average age of 40, tracked with Alzheimer's blood markers measured two decades later. The study pushes the sleep question earlier than it has ever been asked, and shows how thin our measure of sleep still is.

In 2000 and 2001, thousands of American adults in their late thirties and early forties sat down at study visits in Birmingham, Chicago, Minneapolis and Oakland and answered some very ordinary questions. How would you rate your sleep? How many hours do you usually get? Do you have trouble falling asleep, wake in the night, wake too early? Are you sleepy during the day? They were part of CARDIA, the Coronary Artery Risk Development in Young Adults study, which enrolled 5,115 people aged 18 to 30 in 1985 and 1986 and has examined them every two to five years since. Nobody in that room was thinking about dementia. The questions took a minute. Then twenty years passed.

What happened next is the point of a paper published on April 14, 2026, in Alzheimer's & Dementia by Clemence Cavailles, Mercedes Carnethon, Kristen Knutson, Stephen Justin Thomas and Kristine Yaffe. At the Year 35 visit, blood from 1,325 of those participants (mean age at the sleep questions: 40.3) was assayed for three markers now used to stage Alzheimer's biology: the plasma amyloid beta 42/40 ratio, tau phosphorylated at threonine 217, and neurofilament light chain, a general marker of nerve cell damage. A subgroup of 1,044 also had brain MRI scored for SPARE-AD, a machine-learning index of how closely a brain's atrophy pattern resembles early Alzheimer's disease. The associations held up after adjustment for age, sex, race, education and study centre: longer sleep duration went with a lower amyloid 42/40 ratio, poor sleep quality with higher NfL, daytime sleepiness with both, and insomnia symptoms with higher SPARE-AD scores (mean -0.776, 95% CI -0.807 to -0.746, compared with -0.841, 95% CI -0.874 to -0.809, in those without them). In the MRI subgroup, insomnia symptoms were not a rarity: 544 people, 52.4%, reported at least one.

The honest surprises sit in the gaps. p-tau217, the most Alzheimer's-specific of the three blood markers, showed no association with any sleep measure. The duration signal ran the unexpected way, with longer rather than shorter sleep tied to the worse amyloid ratio, and the longer-sleeping group was small (33 people, 3.2% of the MRI subgroup), which is a thin foundation to build anything on. And when the authors added body mass index, smoking, alcohol, exercise, diabetes, hypertension and depressive symptoms to the model, most findings survived but the sleepiness-amyloid link faded to a confidence interval that crosses zero (beta -0.08, 95% CI -0.21 to 0.04). Adjusting for APOE e4 status or kidney function did not change the picture, and there was no sign the results differed by sex, race or APOE e4 carriage.

This is the missing metric, and it cuts both ways. Sleep here was captured by self-report at a single visit: no overnight polysomnography, no wrist actigraphy, no measure of how much slow-wave sleep anyone got, no sleep apnoea diagnosis, no record of whether the person who ticked "poor" in 2001 slept badly for the next twenty years or slept beautifully from 2003 onward. The same group's earlier work, in Neurology in October 2024, found that CARDIA participants reporting two to three poor sleep characteristics had brains that looked 1.6 years older on MRI fifteen years later, and those with more than three looked 2.6 years older, again from questionnaire answers. That crude instrument keeps producing signals, which is either evidence that self-reported sleep carries real information, or evidence of how much sharper the answer might be if anyone had been measuring properly in 2001.

For a family reading this, the practical content is narrower than the headline suggests, and that is worth saying plainly. Nobody in this study was reported as having dementia. These are group-average shifts in blood chemistry and scan patterns among people in late midlife, not predictions about individuals, and a blood marker is not a diagnosis. An observational association also cannot tell you which way the arrow points: early Alzheimer's biology can itself disturb sleep, and the disease's silent phase is thought to run ten to twenty years before symptoms, which is exactly the window this study straddles. What the paper does support is the framing the authors give it, sleep as a potential modifiable risk factor worth studying at forty rather than seventy. Whether a particular person's insomnia or daytime sleepiness needs attention is a conversation for their own clinician, who has reasons to take both seriously regardless of what amyloid is doing.

The questions were cheap, the wait was long, and the answer arrived twenty years after anyone could have acted on it. That is the argument for asking better questions now.