A data driven approach to assess relationships between sleep, cognition and dementia: Findings from the Sleep and...

Sleep has been associated with cognition and risk of dementia. However, sleep is a highly complex and multi-dimensional state, and there is uncertainty about which aspects of sleep are most relevant to cognitive performance and dementia risk. We applied a data-driven approach to identify clusters of sleep variables that reflect meaningful sleep composites and examined their association with cognitive performance and dementia risk.

Data from the Sleep and Dementia Consortium, consisting of 5 US population-based cohorts were utilized. Participants had methodologically consistent, home-based polysomnography, self-report habitual sleep, neuropsychological assessments, and dementia risk surveillance. The pooled cognitive analysis included 5,958 participants aged ≥45 years, and the incident dementia analysis included 5,471 participants aged ≥60 years. A cluster around latent variables analysis was used to derive 9 latent sleep composites from 44 sleep metrics. Global cognitive composite z-scores were derived from principal component analysis. Linear regression models were used to assess associations between sleep composites and cognitive performance. Cox proportional hazard models assessed associations between sleep composites and incident dementia.

Mean (SD) age was 70 ± 11 and 74 ± 12 years for the cognitive and dementia analysis, respectively. There were 1,134 incident dementia cases (median follow-up time of 5-19 years). 9 sleep composites were identified, together explaining 49% of the total variance in the original 44 sleep metrics: Sleep quantity and efficiency, sleep fragmentation, light NREM predominance, N3 predominance, spindle number and duration, REM sleep bouts, respiratory disturbances, slow oscillation-spindle coupling and spindle amplitude. Of these, composites reflecting greater sleep quantity and efficiency (i.e., longer and more consolidated sleep; pooled β per one-unit change in composite, 0.03; 95% CI: 0.004 - 0.06; p=0.033) and stronger slow oscillation-spindle coupling (pooled β, 0.04; 95% CI: 0.003 - 0.07; p=0.039) were associated with better global cognition. However, no significant associations were identified between the 9 sleep composites and dementia risk.

Our data-driven approach identified longer, more consolidated sleep and stronger slow oscillation-spindle coupling as the composites of sleep most strongly related to cognitive performance. These composites may be useful in guiding further investigations of sleep-brain health relationships.