Evaluating the performance of open-source and proprietary processing of actigraphy sleep estimation in children with suspected sleep disorders: a comparison with polysomnography

Study Objectives Evaluate the performance of actigraphy-based open-source and proprietary sleep algorithms compared to polysomnography in children with suspected sleep disorders.Methods In a sleep clinic, 110 children (5-12 years, 54% female, 50% black, 82% with sleep disorders) wore wrist-placed ActiGraph GT9X during overnight polysomnography. Actigraphy data were scored as sleep or wake using open-source GGIR and proprietary ActiLife software. Discrepancy and epoch-by-epoch analyses were conducted to assess agreement between algorithms and polysomnography, along with equivalence testing.Results The open-source vanHees2015 algorithm showed good accuracy (79.5% +/- 12.0%), sensitivity (81.1% +/- 13.5%), and specificity (66.0% +/- 23.8%) for sleep detection but was outperformed by the proprietary ActiLife algorithms. The magnitude and trend of bias for total sleep time (TST), sleep efficiency (SE), sleep onset latency, and wake after sleep onset were similar between algorithms. TST and SE were statistically equivalent for the Cole-Kripke (Actilife) and vanHees2015 algorithms compared to the Sadeh (Actilife) algorithm. The Cole-Kripke (ActiLife) demonstrated higher sensitivity (90.5%) to detect sleep but lower specificity (61.2%) than Cole-Kripke (GGIR) (sensitivity: 62.7%, specificity: 79.9%). Sadeh and Cole-Kripke estimated sleep outcomes were not statistically equivalent between implementations in ActiLife and GGIR.Conclusions The open-source vanHees2015 algorithm performed well but slightly worse than the proprietary ActiLife algorithms in children. The open-source nature vanHees2015 makes it ideal for clinical pediatric use. Implementation of the Sadeh and Cole-Kripke algorithms in the proprietary ActiLife and open-source GGIR software yield different sleep estimates, so comparisons between studies using these different implementations should be avoided.