Aging is associated with decline in cognitive performance, often leading to multimorbidity and frailty. Because changes in sleep architecture are both strongly associated with aging and a wide range of neurocognitive diseases, sleep may provide a general-purpose window into brain health. This association makes sleep-physiology based metrics, like BAI, attractive as indicators of brain health and suggests that sleep-based biomarkers associated with cognitive performance may promote early diagnosis of cognitive impairment.
Although changes in sleep have been linked with both aging and cognition, the relationship between brain age and cognition has not yet been evaluated. In this study, our first aim was to examine the degree to which a previously proposed brain age biomarker, “brain age index” (BAI), is associated with cognitive performance. We hypothesized that participants with elevated BAI would perform worse on cognitive assessments. Nevertheless, because BAI was developed to predict age, we reasoned that its correlation with cognition is likely nonspecific. Our second aim was to develop a series of novel markers of brain health termed Brain Cognitive Indices (BCIs), designed to serve as indicators of cognitive health. We hypothesized that specific combinations of sleep-EEG features would be correlated with performance on specific cognitive tasks, and that it may thus be possible to develop EEG-based indicators specifically correlated with each of these tasks.
To evaluate our aims, we recruited 150 adults with an average age of 48.8 years (+ 17.7 years) who underwent overnight electroencephalography (EEG) as part of diagnostic polysomnography at the Massachusetts General Hospital sleep lab. Patients completed the NIH Toolbox Cognition Battery, which measures total, crystallized, and fluid intelligence. We then evaluated the correlation of BAI with cognitive test scores and used a machine learning approach to develop a series of new sleep EEG-based indices that are optimized to correlate with specific cognitive scores. The new models, termed “brain cognitive indices” (BCIs), were then externally validated using data from the Sleep Heart Health Study.
Our findings showed that BAI was weakly correlated with crystallized intelligence (r=-0.25, p=0.002). In contrast, BCIs correlated with total (r= 0.37, p<0.0001) and fluid (r= 0.56, p<0.0001) intelligence and did not correlate with crystallized intelligence (r= -0.07, p=0.38). Both total and fluid BCI correlated with cognitive scores in external validation with a similar strength of correlation (Total: r=0.31, p<0.0001; Fluid: r=0.32, p<0.0001). All significant BCI models were able to differentiate low from high test scorers with good discriminatory abilities at the group and individual levels. Key features contributing to brain health information include delta-to-theta and delta-to-alpha band power ratios, kurtosis, spindle density, coupling between slow oscillations and spindles, and percentage of REM sleep.
Overall, our results indicate that information about both total and fluid intelligence is decodable from an overnight sleep EEG, and that overnight sleep EEG is a promising source of easily accessible and repeatable indicators of brain cognitive health.