Background: Sleep is increasingly recognized as a crucial component to rapid and successful rehabilitation, especially from traumatic brain injuries (TBIs). Assessment of longitudinal patterns of sleep in a hospital setting, however, are difficult and often the expertise or equipment to conduct such studies is not available. Actigraphy (wrist-worn accelerometry) has been used for many years as a simple proxy measurement of sleep patterns, but its use has not been thoroughly validated in individuals with TBI. Objective: To determine the validity of different sensitivity settings of actigraphy analysis to optimize its use as a proxy for recording sleep patterns in individuals with a TBI. Design: Comparison of actigraphy to criterion standard polysomnographic (PSG)-determination of sleep on a single overnight study. Setting: Six rehabilitation hospitals in the TBI Model System. Participants: Two hundred twenty-seven consecutive, medically stable individuals with a TBI. Interventions: Not applicable. Main Outcome Measure: Concordance between PSG-and actigraphy-determined sleep using different sensitivity threshold settings (low, medium, high, automated). Results: Bland-Altman plots revealed increasing error with increasing amounts of wake during the sleep episode. Precision-recall statistics indicate that with less sensitive actigraphy thresholds, episodes identified as "wake" are usually "wake," but many true episodes of "wake" are missed. With more sensitive actigraphy thresholds, more episodes of "wake" are identified, but only some of these are true episodes of "wake." Conclusions: In hospitalized patients with TBI and poor sleep, actigraphy underestimates the level of sleep disruption and has poor concordance with PSG-determined sleep. Alternate methods of scoring sleep from actigraphy data are necessary in this population.

No abstract

Introduction The measurable aspects of brain function derivable from polysomnography (PSG) that are correlated with sleep satisfaction are poorly understood. Previously, a weak association of PSG measures with subjectively rated sleep depth and restfulness was shown. Using recent developments in automated sleep scoring, which remove the within- and between-rater error associated with human scoring, we revisit whether whole night PSG measurements are associated with sleep satisfaction. Additionally, we investigate if PSG data collected closer to wake time explains the subjective sleep experience better than whole night PSG. Methods Random Forest machine learning was used to investigate the relationship between PSG data from the Sleep Heart Health Study (N=3,165, middle-aged and older adults) and self-reported sleep satisfaction (restfulness, depth). PSG were rescored using a novel automated algorithm that generates both a sleep stage for each 15-s epoch as well as a stage matching probability. Data were also parsed into 20 minute-fragments based on time relative to wake. Results Models explained 30% of subjective sleep depth and 27% of subjective sleep restfulness, with a similar top four predictors: minutes of N2 and wake after sleep onset (WASO), sleep efficiency, and age, capturing 28% (restfulness) and 26% (depth) of the relative model variance. With increasing subjective sleep quality, there was a progressive increase in N2 and decrease in WASO of similar magnitude, without systematic changes in N1, N3 or REM. In comparing those with the best and worst subjective experience of sleep, there is a range of approximately 30 minutes more N2, 30 minutes less WASO, an improvement of sleep efficiency of 7-8%, and an age span of 3-5 years. Random Forest models derived from PSG fragments closer to the offset of sleep did not provide better explanatory power compared to the whole-night data set. Conclusion Approximately one-third of the variance in two measures of self-reported sleep experience can be explained by whole-night PSG variables, notably an increase in N2 and decrease in wake that led to improved sleep efficiency. Interventions that specifically target these may be suitable for improving the self-reported sleep experience. Support (If Any)