1 - (SYM 37) Application of Temporal Network Analysis to Sleep Diary Data: Testing the Limits by Adhering to Best Practices?

The interplay between sleep parameters, as well as variables such as stress and depressed mood, is a well-documented complex relationship. Sleep disturbances one night, such as shorter sleep duration, detrimentally affects next day mood, which can result in impaired sleep the following night. Although these relationships can be delineated using single-timepoint questionnaires, sleep diaries may more validly capture these unfolding processes. Using sleep diaries, researchers can analyze repeated-measures assessments and document changes in sleep parameters, mood, and stress over time. One promising method for analyzing these data is temporal network analysis, which estimates relationship between variables by assessing whether one variable predicts another (and itself) at each time point. In doing so, one can pinpoint variables with high “outstrength,” variables that are highly predictive of other variables at the following timepoint(s).

To our knowledge, no studies have assessed the relationship between sleep parameters and self-reported depressed mood and stress via temporal network analysis. To this end, we analyzed diary data from 401 nurses who completed assessments of sleep functioning, depressed mood, and stress, each morning upon awakening. Nurses completed these diaries once daily for 14 days. The following variables from the diaries were used for the present analyses: sleep onset latency (SOL), wake after sleep onset (WASO), sleep efficiency (SE), total sleep time (TST), depressed mood, stress, and a “midpoint” variable representing the halfway point between bedtime and waketime, accounting for sleep timing.

Overall, TST showed high outstrength centrality, promoting greater stress and reduced SE. Thus, TST may be a candidate target for intervention for patients exhibiting higher levels of depressed mood and stress. However, this presentation will present these results against the backdrop of cutting-edge methodological advances to analyzing these data, focusing on the parameters of temporal network analysis. Specifically, rule-of-thumb recommendations suggest that temporal networks are best constructed with at least 20 time points, which do not match the parameters of our study. Although other methods are available to estimate networks with longitudinal data, these methods are best fit for panel data. As such, this presentation will focus on a balance between best practices for temporal network analysis, and how to consider estimation procedures for data that may not “fit” the parameters of the statistical model.