1259 Should We Trust Native Apple Algorithms for Detecting Sleep Onset in Children?

Abstract Introduction Sleep trackers such as Apple watches (AW) have proliferated in the consumer market, spurring the public onto sleep tracking. AW is currently the dominant smartwatch both globally (22%) and in the US market (59%). The public health impact of this is significant for youth because sleep is vital during critical windows of neural development. However, current users can only rely on information derived from black-box algorithms, many of which have not been tested for accuracy. The objective of this study was to assess the output of the native Apple algorithms in determining sleep onset times in children compared against a research-grade device and algorithm. Methods A community sample of 10 children aged 8-17 years (Mage = 13.70 ± 2.79) completed daily sleep diaries and wore an AW and Actiwatch on their non-dominant wrist for 14 days. A difference score was calculated for sleep onset time between each device and sleep diaries. One sample t-tests were conducted on the difference score for each device, and a paired samples t-test compared the two devices. False positive (device reporting sleep during wake) and false negatives (device missing sleep events) rates were calculated for both devices. Results Both devices showed significantly different sleep onset times compared to sleep diaries (AW: mean = 5.9 hrs ± 2.3 SD, p < .001; Actiwatch: mean = 0.7 ± 0.4 SD, p < .001); however, the error for AW was significantly larger compared to the Actiwatch, t(9) = 6.47, p < .001. Whereas the Actiwatch had no false negative or positive events, the Apple native algorithms exhibited a false positive rate of 29.8% and a false negative rate of 6.0%. Conclusion Our findings suggest sleep onset times derived from native Apple algorithms are significantly less accurate compared to Actiwatch; however, given the scalability of AW, one solution may be to process raw accelerometry data from AW with validated open-access sleep algorithms, particularly in pediatric populations. This approach has shown promise in adults and needs to be tested in children. Support (if any) The authors thank the staff at the Henry Ford Sleep Disorders and Research Center and the Moreno Lab for their support.