Signatures for Acute and Chronic Insomnia from Locomotor Time Series Data

Sleep is an important part of human existence. During sleep the body functions restore and recharge. The sleep process is a complex multi-dimensional cycle that reflects the developmental changes in mental and physical health, along with the day-to-day state fluctuations. However, sleep disturbances, and insomnia in particular, affect a large part of the human population and their quality of life, work productivity and individual’s performance. Insomnia is characterised by the inability to fall asleep or stay asleep and/or waking too early and being unable to fall back asleep. It is a sleep disorder that remains under-diagnosed. Insomnia is strongly influenced by the brain activities, chronic conditions, physiological and cognitive health. Here, it is investigated as part of the brain activities, cardio and respiratory systems, blood flow and other physiological functions, namely in the context of network physiology [1] and as part of the neuronal, cardiorespiratory and blood networks.
We propose two new data driven and model free algorithms for investigation and classification of nocturnal awakenings in acute [2] and chronic [3] insomnia and normal sleep from nocturnal actigraphy collected from pre-medicated individuals with insomnia and normal sleep controls. The tri-axial accelerometer build in the Philips ActiWatch collects wrist movements (locomotor motion) data at epochs of 1 minute. Our algorithms are based on signals obtained from the ActiWatch capturing the locomotor activities. They do not require sleep diaries or any other subjective information from the individuals. They can be used as pre-screening tools for assessing insomnia at home. We derive the dynamical fingertips which form the signature of acute or chronic insomnia derived from actigraphy data and distinguish them from the normal sleep. We use spectral [4] and fractal [5] analysis, and obtain statistical, dynamical and sleep parameters (features) from the actigraphy time series data [1,2,3,4]. These features are then combined in machine learning models to classify individuals with acute or chronic insomnia from healthy sleepers [2,3]. The algorithms include classifiers and optimization that incorporate the predicted quality of each night of sleep for an individual to classify into acute or chronic insomnia or healthy group. The developed algorithms provide signatures for acute and chronic insomnia and healthy sleep and are very promising pre-screening tools for early detection of insomnia at home.
Acknowledgements
The authors acknowledge partial support from the Newton Advanced Fellowship, Academy of Medical Sciences, UK.
References
[1] A Bashan, RP Bartsch, JW Kantelhardt, S Havlin and P Ch Ivanov. Nature (2012) 3 702.
[2]M Angelova, C Karmakar, Y Zhu, SP Drummond, J Ellis, IEEE Access (2020) 8, 74413.
[3] S Kusmakar, C Karmakar, Y Zhu, S Shelyag, S Drummond, J Ellis and M Angelova, Sleep (2020) submitted.
[4] R Fossion, AL Rivera, JC Toledo-Roy, J Ellis, J and M Angelova (2017). PLoS ONE(2017) 12(7), 1-21.
[5] P Holloway, M Angelova, S Lombardo, A St Clair Gibson, D Lee and J Ellis, JRC: Interface (2014) 11(93), 1-8.