Sleep as part of the 24-hour day: Methods and Applications in Oncology – Joshua Wiley

Abstract

Sleep is commonly analysed as an independent factor. However, because of the 24-hour constraints on a day, changes in sleep will co-occur with changes in remaining time use. This talk introduces compositional data analysis (CoDA) for sleep research. CoDA is illustrated using 24-hour sleep and activity data from accelerometry, first cross-sectionally showing associations between sleep and activity with daily emotions. Next, CoDA is extended to multilevel models, which commonly occur in sleep research as sleep across multiple days. A Bayesian implementation in R using the new multilevelcoda package is presented and results of a simulation study discussed. Multilevel CoDA is used to analyse how nightly sleep architecture, collected via at home EEG, predicts next day affect. CoDA solves a common constraint in sleep research with sleep stages that time in all stages sums to total sleep.