13 August 2021
By Dr. Sarah Schoch
Postdoctoral Research Fellow at Donders Institute for Brain, Cognition and Behaviour
Sleep can be defined on levels of behavior, brain activity or cellular dynamics. In humans, the gold standard is to measure sleep by using electroencephalography (EEG). But behavioral measures can complement in-lab investigations to track sleep across several days and in the natural sleep environment. In a recent video, Dr. Sarah Schoch and Prof. Dr. Salome Kurth explain an approach to get more comparable and stable sleep variables.
While several methods exist for assessing behavioral sleep (videosomnography, observation, diaries), the most common is actigraphy. Actigraphs measure movement and other signals such as temperature and light. Algorithms are then applied to compute and estimate sleep and wake states from these movement-based measures. Despite the benefits that actigraphy yields in complementing laboratory EEG-studies, some aspects of actigraphy remain to be optimized for sleep estimation from actigraphs. Three particular challenges have hindered a global generalizability from actigraphy data.
Limited Applicability of Algorithms
A first obstacle is that several different algorithms are available, some only applicable to specific age groups (Haghayegh et al., 2019; Schoch et al., 2021). The use of different algorithms can lead to different sleep estimates, especially in populations with changed sleep and movement patterns, such as infants (Schoch et al., 2019). Differences in estimating, e.g., sleep duration based on different algorithms, are significant when considering normative values across development. While certain adjustments to sleep estimates from actigraphy have been suggested that reduce the deviance between algorithms (e.g., rescoring rules Webster et al., 1982), these are only selectively applied. One solution we propose for infant research is an analysis pipeline that incorporates different adjustments to ensure the generalizability of the results (Oakley, 1997; Sadeh et al., 1995; Schoch et al., 2019).
The second obstacle is that variables estimated from actigraphy are not applied in a standardized manner. Instead, several different variables are used, sometimes with distinct names or overlapping concepts (e.g., Sleep Opportunity and Sleep Period). If too many variables are incorporated in analysis, we run into the multiple testing problem; if we select only certain variables, we might overlook important dimensions of sleep. In a new article published in Sensors (Schoch et al., 2020), we applied a principal component analysis on 48 variables frequently reported in actigraphy studies and identified five sleep composites in infants. Sleep composites are Sleep Timing, Sleep Night, Sleep Day, Sleep Activity, and Sleep Variability. In this video, we explain our approach: