Sleep duration tracks with an Alzheimer's-type pattern of brain atrophy and with cognitive test performance in a U shape, meaning that people who reported long sleep looked no better than those who reported short sleep, according to a peer-reviewed analysis of 38,816 UK Biobank participants published in Alzheimer's & Dementia on 8 July 2026. Anagnostakis and colleagues, working across Columbia University and the University of Pennsylvania, used self-reported sleep collected at the baseline visit between 2007 and 2010 in adults aged 40 to 70 who had no dementia diagnosis at that point, and compared it with SPARE-AD, a machine-learning imaging measure derived from brain MRI that scores how closely a person's pattern of atrophy resembles that of people with Alzheimer's disease. The European Academy of Neurology named the paper its research paper of the month on 7 August 2026.

After adjustment for age, sex, body mass index, APOE-e4 status, systolic blood pressure, education, smoking, LDL cholesterol and HbA1c, the association between sleep duration and SPARE-AD was non-linear and statistically significant (p = 0.001). The same U shape appeared for all three cognitive measures, Trail Making Test A, Trail Making Test B and the Digit Symbol Substitution Test (all p < 0.001), with the best scores falling at 6.7, 6.7 and 7.2 hours respectively. Short sleep of 5 to 6 hours and long sleep of 9 to 10 hours were both associated with poorer cognitive performance. Mean reported sleep in the sample was 7.2 hours. In a separate analysis using TriNetX electronic health records in the United States, 125,444 propensity-matched people with insomnia had a higher recorded risk of an Alzheimer's diagnosis over a median 9.2 years (hazard ratio 4.3, 95 percent confidence interval 4.07 to 4.54), and 27,660 matched people with hypersomnia over a median 9.3 years (hazard ratio 5.05, 95 percent confidence interval 4.48 to 5.69). The authors imposed a four-year lag before counting diagnoses to reduce reverse causation.

The brain-imaging signal was not uniform. Splitting by sex, the link between sleep duration and SPARE-AD held in men (p < 0.001) but was not significant in women (p = 0.239). Splitting by age band, at 40 to 60 and at 60 to 80 years at the imaging visit, no association between sleep duration and Alzheimer's-type atrophy reached significance; the authors attribute this to reduced sample size and note the U-shaped pattern persisted mainly in men. They list the study's own limits plainly: the cross-sectional design precludes causal inference, and sleep duration was self-reported and therefore open to recall bias and misclassification. The health-records arm is observational too, and hazard ratios drawn from diagnostic codes can reflect how often people with a sleep disorder see a doctor as much as what is happening in their brains.

The reportable point for families is the shape of the curve rather than its direction. Public messaging about sleep and dementia has tended to warn about too little sleep; this analysis reports that in the same cohort, on the same measures, long sleepers sat on the unfavourable side as well. Whether long sleep contributes to brain change, or is one of its earliest visible consequences, is a question this study was not designed to answer.