Midlife Sleep and Brain Health

How do sleep habits in midlife relate to dementia risk more than a decade later? A study of over 330,000 people examined several sleep characteristics together, including daily sleep duration, frequent insomnia, and daytime sleepiness, to assess their associations with subsequent dementia.

The study, conducted by Tao Wei, Jie Chang, and colleagues, was published in Alzheimer's Research & Therapy on September 30, 2025. It used UK Biobank data from participants aged 40 to 64 at enrollment, with a median follow-up of 13.8 years. Reading the full paper alongside its main results table makes it clear that the associations between sleep scores and different types of dementia need to be considered separately.

People with higher healthy sleep scores had a lower risk of all-cause dementia and vascular dementia, whereas no significant association was found for Alzheimer's disease. These findings offer clues about sleep and long-term brain health. However, this observational study cannot establish whether actively improving sleep would produce the same changes in risk.

Research Background

Assessing Multiple Dimensions of Sleep

The study looked beyond sleep duration. Researchers also asked whether participants had trouble falling asleep or woke during the night, whether they snored, and whether they fell asleep unintentionally during the day. These questions captured aspects of sleep that duration alone could not, allowing researchers to examine how a combination of sleep characteristics related to dementia risk.

The scoring system was straightforward. Participants received 1 point for each of five characteristics: sleeping 7 to 8 h a day, having an early chronotype, having no frequent insomnia, not snoring, and having no frequent daytime sleepiness. Total scores ranged from 0 to 5. An early chronotype meant identifying as a morning person or more of a morning than an evening person; it did not require a fixed bedtime or wake-up time. The absence of frequent symptoms allowed occasional symptoms, encompassing the responses never, rarely, or sometimes.

This score was used to classify participants for research purposes and is not a tool for individual diagnosis or treatment planning. Assigning a point for an early chronotype does not establish that forcing evening types to get up early will reduce their risk. The temporal relationship between sleep and dementia also needs to be clarified. Sleep disturbances may contribute to disease development, but they may also emerge after neurodegenerative changes have begun and before a diagnosis is made.

Methods

Cohort Follow-up and Brain Imaging

Of 406,364 participants in the eligible age range, the researchers excluded those who had withdrawn, had incomplete sleep data, or already had dementia at enrollment. The final sample comprised 333,014 participants with a mean age of 54.1 years. They enrolled between 2006 and 2010, after which the researchers tracked new dementia cases using hospital inpatient, death, and primary care records, among other sources. Follow-up continued until December 2022 or ended earlier at dementia diagnosis, death, or loss to follow-up.

Sleep data came from questionnaires completed at enrollment. What did sleeping 7 to 8 h mean in this study? The questionnaire asked about total sleep in every 24 h, including naps. Although the paper also used the term "per night" in its scoring definition, the questionnaire determines what was actually measured. Sleep was not recorded continuously, so the researchers could not determine whether participants maintained the same habits over the following decade or more.

The primary outcome was all-cause dementia, with Alzheimer's disease (AD) and vascular dementia (VD) analyzed separately. The researchers used Cox proportional hazards models to compare risk across sleep scores, taking participants who scored 0 to 1 as the reference group. The models included age, education, and lifestyle factors, as well as body mass index (BMI), relevant medical history, and apolipoprotein E (APOE) ε4 allele status, to reduce the influence of these differences on the comparisons.

Among the participants, 33,401 also underwent brain magnetic resonance imaging (MRI), a median of 9.4 years after their initial sleep assessment. The researchers examined measures such as grey matter volume and white matter hyperintensity (WMH) volume to determine whether sleep scores were also associated with brain structure.

Results

Risk Associations and Differences

During follow-up, 3,035 new dementia cases were recorded, including 1,304 AD cases and 597 VD cases. In the fully adjusted model, each 1-point higher healthy sleep score was associated with a hazard ratio (HR) of 0.93 for all-cause dementia, with a 95% confidence interval (CI) of 0.89 to 0.96 (P < 0.001). Compared with participants who scored 0 to 1, those who scored 5 had an HR of 0.75 (95% CI 0.61 to 0.92; P = 0.006).

Based on the HR, the group scoring 5 had approximately 25% lower relative risk than the group scoring 0 to 1. This figure comes from a comparison between two groups of people. The study did not ask the same individuals to improve their sleep and then observe whether their risk fell, so this 25% cannot be treated as the expected effect of improving sleep. Nor does it mean that the probability of developing dementia fell by 25 percentage points.

The differences became clearer when dementia was separated into subtypes. Comparing scores of 5 with scores of 0 to 1, the HR for VD was 0.55 (95% CI 0.37 to 0.82; P = 0.004), whereas the HR for AD was 1.08 (95% CI 0.75 to 1.55; P = 0.675). The study found no significant association between the sleep score and AD risk. These results do not support a conclusion that the sleep pattern prevents AD, but neither do they establish that sleep and AD are entirely unrelated.

Three of the five sleep characteristics were associated with lower all-cause dementia risk. The HRs for sleeping 7 to 8 h a day, having an early chronotype, and having no frequent daytime sleepiness were 0.83, 0.91, and 0.70, respectively. The corresponding 95% CIs were 0.77 to 0.89, 0.85 to 0.98, and 0.59 to 0.82, with P values of < 0.001, 0.017, and < 0.001. Neither the absence of frequent insomnia nor the absence of snoring showed a significant association. The findings for the composite score do not imply that every component has an independent protective effect. 

The age-stratified analysis also shows why relative and absolute differences need to be considered together. The relative association was stronger among participants aged 40 to 55. However, the reported 15-year cumulative incidence in that age group was 0.17% for those scoring 5 and 0.72% for those scoring 0 to 1, compared with 1.78% and 2.63%, respectively, among participants aged 56 to 64. The absolute difference was larger in the older group. Relative associations alone are therefore insufficient to determine which age group might benefit more from an intervention.

Mechanistic Discussion

Brain Structure and the Limits of the Evidence

The imaging findings offered another clue. Participants with higher sleep scores had larger grey matter volumes and smaller overall WMH volumes. WMHs are areas of high signal intensity in the white matter on specific MRI sequences. For each 1-point higher sleep score, the standardized regression coefficient for grey matter volume was 0.01 (95% CI 0 to 0.02), while the coefficient for overall WMH volume was −0.02 (95% CI −0.03 to −0.01). These comparisons used standardized measures. The values 0.01 and −0.02 cannot be interpreted as a 1% increase in brain volume or a 2% reduction in lesion volume.

The authors therefore proposed that healthier sleep patterns may be associated with less white matter damage. They also used mediation models to estimate how much of the association between sleep scores and dementia risk could be statistically explained by brain structure. The estimated proportions were approximately 0.96% for grey matter, 3.15% for overall WMH, and 4.23% for periventricular WMH. These estimates came from separate models and must not be added together.

Only 49 new dementia cases occurred after imaging in this analysis, leaving the estimates uncertain. Each participant also had only a single imaging assessment, so the researchers could compare brain structure between people but could not determine whose brain was atrophying more slowly. These results can help generate mechanistic hypotheses, but they do not establish that sleep reduces dementia risk by changing brain structure.

The researchers performed several checks, including excluding people who developed dementia within the first 2 or 4 years, changing the approach to missing data, and accounting for the effect of death on dementia ascertainment. The main results remained broadly consistent. Nevertheless, disease may have been developing for years before diagnosis, so its effects on sleep remain difficult to exclude. Participants were predominantly of European ancestry, sleep was self-reported, and medical records may have missed cases or misclassified dementia subtypes. Whether the findings apply to other populations still requires validation.

Future Directions

Longitudinal Validation and Intervention

Future studies could measure sleep repeatedly across different populations, combining wearable devices with sleep monitoring to clarify how habits change with age and health. Repeated brain imaging and more dementia cases occurring after imaging are also needed to test whether changes in sleep precede changes in brain structure. Prespecifying the primary analyses would help reduce uncertainty arising from multiple comparisons. Whether improving sleep can lower dementia risk will require randomized intervention studies. By targeting specific problems such as insomnia or daytime sleepiness, tracking cognition and dementia over the long term, and reporting both relative and absolute risks, these studies could establish the magnitude and duration of any benefit.

References

Wei, T., Chang, J., Zhao, Y., Li, A., Sun, W., Liu, X., . . . Tang, Y. (2025). Associations of adherence to a healthy sleep pattern with dementia risk in the UK biobank. Alzheimer's Research & Therapy, 17(1), Article 213. doi:10.1186/s13195-025-01864-x