0107 A Data-Driven Multidimensional Approach Examining the Relationship Between Cognition and Sleep Composites in the Multi-Ethnic Study of Atherosclerosis

Abstract Introduction Sleep is critical for maintaining overall brain health and may influence the onset and perpetuation of cognitive impairment. Since sleep is a complex and multi-dimensional construct, we hypothesized that methods that incorporate multi-modal measurement of sleep across various domains (i.e. sleep continuity, macro-architecture, power spectral density) may identify groupings of correlated variables that serve as distinct phenotypes that predict cognitive performance outcomes. Methods Multimodal measurement of sleep (actigraphy, polysomnography, patient-reported outcomes) and cognitive testing were completed around the time of the 5th follow-up exam (2010 – 2013) from the Multi-Ethnic Study of Atherosclerosis (MESA), a large multiethnic cohort in the United States. A principal components analysis (PCA) with varimax rotation was utilized on 39 sleep variables from multiple domains of sleep health for 1,414 participants who had complete sleep data (overnight polysomnography, 7-day actigraphy, and questionnaires) and completed cognitive function testing. Cognition was measured by performance on the Digit Symbol Coding Task (DSCT), a subtest of the WAIS-III measuring processing speed, working memory, and visuospatial processing. Scores on the DSCT were modeled as the primary outcome and were regressed onto scores derived from principal components (PC, p <.05) while controlling for important covariates. Results The study sample had an average age of 68.9 years, was 54.3% female, and 38.7% White, 26.3% Black/African American, 24.3% Hispanic, and 10.7% Chinese. Thirteen novel sleep composites were extracted from the PCA that explained about 71% of variance within the sample. Scores from 4 PCs were significantly (p <.05) related to DSCT scores when controlling for demographic and lifestyle factors. The description of each of the 4 PCs were as follows: 1) generalized EEG power, 2) strength of rest-activity pattern, 3) spindle and slow oscillation coupling, 4) spindle density. Conclusion In this study, interpretable cognition-related sleep composites were identified and characterized by EEG features and actigraphy metrics that have been previously linked to cognition in prior research. Use of data-driven and multi-dimensional approaches may efficiently identify composite biomarkers of multi-dimensional sleep associated with cognitive impairment. Examining correlated dimensions of sleep may shed light on how sleep health and cognition are related. Support (if any) NHLBI HL56984 and NIA R01AG070867.