Premium
Validation of the Munich Actimetry Sleep Detection Algorithm for estimating sleep–wake patterns from activity recordings
Author(s) -
Loock AnnSophie,
Khan Sullivan Ameena,
Reis Catia,
Paiva Teresa,
Ghotbi Neda,
Pilz Luisa K.,
Biller Anna M.,
Molenda Carmen,
VuoriBrodowski Maria T.,
Roenneberg Till,
Winnebeck Eva C.
Publication year - 2021
Publication title -
journal of sleep research
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.297
H-Index - 117
eISSN - 1365-2869
pISSN - 0962-1105
DOI - 10.1111/jsr.13371
Subject(s) - polysomnography , sleep (system call) , actigraphy , wakefulness , sleep onset , interquartile range , slow wave sleep , algorithm , medicine , audiology , circadian rhythm , insomnia , computer science , electroencephalography , psychiatry , operating system
Periods of sleep and wakefulness can be estimated from wrist‐locomotor activity recordings via algorithms that identify periods of relative activity and inactivity. Here, we evaluated the performance of our Munich Actimetry Sleep Detection Algorithm. The Munich Actimetry Sleep Detection Algorithm uses a moving 24–h threshold and correlation procedure estimating relatively consolidated periods of sleep and wake. The Munich Actimetry Sleep Detection Algorithm was validated against sleep logs and polysomnography. Sleep‐log validation was performed on two field samples collected over 54 and 34 days (median) in 34 adolescents and 28 young adults. Polysomnographic validation was performed on a clinical sample of 23 individuals undergoing one night of polysomnography. Epoch‐by‐epoch analyses were conducted and comparisons of sleep measures carried out via Bland‐Altman plots and correlations. Compared with sleep logs, the Munich Actimetry Sleep Detection Algorithm classified sleep with a median sensitivity of 80% (interquartile range [IQR] = 75%–86%) and specificity of 91% (87%–92%). Mean onset and offset times were highly correlated ( r = .86–.91). Compared with polysomnography, the Munich Actimetry Sleep Detection Algorithm reached a median sensitivity of 92% (85%–100%) but low specificity of 33% (10%–98%), owing to the low frequency of wake episodes in the night‐time polysomnographic recordings. The Munich Actimetry Sleep Detection Algorithm overestimated sleep onset (~21 min) and underestimated wake after sleep onset (~26 min), while not performing systematically differently from polysomnography in other sleep parameters. These results demonstrate the validity of the Munich Actimetry Sleep Detection Algorithm in faithfully estimating sleep–wake patterns in field studies. With its good performance across daytime and night‐time, it enables analyses of sleep–wake patterns in long recordings performed to assess circadian and sleep regularity and is therefore an excellent objective alternative to sleep logs in field settings.