0099 Comparison of two actigraphy-based algorithms for detecting daytime and nighttime sleep. (25th May 2022)

• Record Type: Journal Article
• Authors: Gao, Chenlu; Li, Peng; Morris, Christopher; Zheng, Xi; Ulsa, Ma Cherrysse; Gao, Lei; Scheer, Frank; Hu, Kun
• Is Part Of: Sleep. Volume 45(2022)Supplement 1
• Page Start: A44
• Page End: A45
• Publication Date: 2022-05-25
• Subjects: Sleep -- Physiological aspects -- Periodicals; Sleep disorders -- Periodicals; Sommeil -- Aspect physiologique -- Périodiques; Sommeil, Troubles du -- Périodiques; Sleep disorders; Sleep -- Physiological aspects; Sleep -- physiological aspects; Sleep Wake Disorders; Psychophysiology; Electronic journals; Periodicals; 616.8498
• DOI: 10.1093/sleep/zsac079.097
• Languages: English
• ISSNs: 0161-8105

Abstract

Abstract: Introduction: The Actiware software that comes with Philips Respironics' actiwatches tends to overestimate sleep, due to its poor accuracy in distinguishing immobility from sleep. Re-scoring rules were introduced in the Cole-Webster algorithm to overcome this issue. Previous validation of the two algorithms was based on nighttime sleep, and their performance in daytime sleep detection is unknown. This study aims to test/compare the performance of the two algorithms in detecting daytime sleep and nighttime sleep. Methods: We analyzed actigraphy and polysomnography data that were simultaneously collected from 25 participants (14 non-shift-workers and 11 shift-workers; age: 30.93±8.96 [mean±SD]; female: 14 [56%]) each in two in-lab visits with scheduled nighttime or daytime sleep. The sleep/wake epochs scored by the Cole-Webster algorithm and Actiware (using medium wake threshold) were compared to those obtained from polysomnography. We conducted linear mixed-effects regression models to compare the sensitivity, specificity, and F1-score (a measure of performance less affected by imbalanced datasets) in detecting daytime and nighttime sleep and between the two algorithms. Results: The Cole-Webster algorithm (mean±SE: daytime=0.66±0.02, nighttime=0.60±0.02) yielded lower sensitivity than Actiware (daytime=0.96±0.02, nighttime=0.96±0.02; p<0.0001), which was consistent for both daytime and nighttime sleep (daytime/nighttime×algorithm interaction: p=0.2). The Cole-Webster algorithm (daytime=0.91±0.04, nighttime=0.94±0.05) yielded higher specificity than Actiware (daytime=0.45±0.04, nighttime=0.56±0.05; p<0.0001), which was consistent for both daytime and nighttime sleep (daytime/nighttime×algorithm interaction: p=0.2). Both sensitivity and specificity did not differ between daytime and nighttime sleep (p>0.05). F1 scores of the Cole-Webster algorithm were lower (daytime=0.77±0.02, nighttime=0.74±0.02) than those of Actiware (daytime=0.92±0.02, nighttime=0.97±0.02; p<0.0001) for both daytime and nighttime sleep. There was a significant daytime/nighttime×algorithm interaction on F1 score (p=0.02). Specifically, the Cole-Webster algorithm performed better in scoring daytime than nighttime sleep, whereas Actiware performed better in scoring nighttime than daytime sleep. Conclusion: For both algorithms, the performance was similar in detecting daytime and nighttime sleep. Compared to Actiware, the Cole-Webster algorithm was generally better at detecting wake (i.e., high specificity) but worse at detecting sleep epochs (i.e., low sensitivity) and yielded worse overall performance (i.e., low F1). Future studies should test/validate other Actigraphy-based algorithms' performance in scoring daytime sleep. Support (If Any): NIH R01HL094806, RF1AG064312, RF1AG059867, BrightFocus Foundation A2020886S