Profiling continuous sleep representations for better understanding of the dynamic character of normal sleep

Abstract

The amount and quality of sleep substantially influences health, daily behaviour and overall quality of life. The main goal of this study was to investigate to what extent sleep structure, as derived from the polysomnographic (PSG) recordings of nocturnal human sleep, can provide information about sleep quality in terms of correlating with a set of variables representing the daytime subjective, neurophysiological and cognitive states of a healthy population without serious sleep problems. We focused on a continuous sleep representation derived from the probabilistic sleep model (PSM), which describes the microstructure of sleep by a set of sleep probabilistic curves representing a finite number of sleep microstates. This contrasts with approaches where sleep is characterised by a set of one-dimensional sleep measures derived from the standard discrete sleep staging. Considering this continuous sleep representation, we aimed to identify typical sleep profiles that represent the dynamic aspect of sleep during the night and that are associated with a set of studied daily life quality measures. Cluster analysis of sleep probabilistic curves has proven to be a helpful tool when identifying specific sleep temporal profiles, but it faces problems when curves are complex and time misalignment is present. To overcome these problems, we proposed and validated a novel 2-step iterative clustering and time alignment method. We compared the quality of alignment and cluster homogeneity produced by the method with existing approaches in which (i) the time alignment of curves precedes the clustering step, and (ii) time alignment and clustering are performed simultaneously. The obtained homogeneous clusters of REM, Wake and Slow Wave Sleep resembled the clustering structure of subjects with significantly different subjective scores of sleep quality and mood, as well as more objective cognitive test scores. Moreover, the sleep profiles associated with individual clusters help to better understand the existing associations between the overnight dynamics of specific sleep states and daily measures.