O033 Predicting subjective sleep quality using multi-day actigraphy data: A machine learning approach. (7th October 2021)

• Record Type: Journal Article
• Authors: Kao, C; D'Rozario, A; Lovato, N; Bartlett, D; Postnova, S; Grunstein, R; Gordon, C
• Is Part Of: Sleep advances. Volume 2:Supplement 1(2021)
• Page Start: A14
• Page End: A15
• Publication Date: 2021-10-07
• Subjects: Sleep disorders -- Periodicals; Circadian rhythms -- Periodicals; 616.8498
• DOI: 10.1093/sleepadvances/zpab014.032
• Languages: English
• ISSNs: 2632-5012

Abstract

Abstract: Objectives: Insomnia is diagnosed using clinical interview but actigraphy is often used as a consecutive multi-day measurement of activity-rest cycles to quantify sleep-wake periods. However, discrepancies between subjective complaints of insomnia and objective actigraphy measurement exist. The aims of the current study were to (i) predict subjective sleep quality using actigraphic data and, (ii) identify features of actigraphy that are associated with poor subjective sleep quality. Methods: Actigraphy data were collected for 14-consecutive days with corresponding subjective sleep quality ratings from participants with Insomnia Disorder and healthy controls. We fitted multiple machine learning algorithms to determine the best performing method with the highest accuracy of predicting subjective quality rating using actigraphic data. Results: We analysed a total of 1278 days of actigraphy and corresponding subjective sleep quality ratings from 86 insomnia disorder patients and 20 healthy controls. The k-neighbors classifier provided the best performance in predicting subjective sleep quality with an overall accuracy, sensitivity and specificity of 83%, 74% and 87% respectively, and an average AUC-ROC of 0.88. We also found that activity recorded in the early morning (04:00-08:00) and overnight periods (00:00-04:00) had the greatest influence on sleep quality scores, with poor sleep quality related to these periods.. Conclusions: A machine learning model based on actigraphy time-series data successfully predicted self-reported sleep quality. This approach could facilitate clinician's diagnostic capabilities and provide an objective marker of subjective sleep disturbance.