P122 Comparison of sleep estimation using Apple Watch accelerometry against polysomnography

Abstract

Abstract Introduction Consumer wearables offer new ways to improve our health and well-being, including sleep. Researchers are interested in consumer wearables because their widespread adoption creates the potential for larger studies than could be run with clinically validated measurement methods, as those are more expensive or less convenient. This study investigates sleep tracking using sensor data from Apple Watch in comparison to the gold standard polysomnography (PSG). Method We used Apple Watch accelerometer data to establish both activity and heart rate (using ballistocardiography). Thirty participants (13 female, 17 male) wore the Apple Watch on their non-dominant wrist during clinical PSG. We compared predicted sleep status at the epoch level and overall sleep parameters, taking PSG as the ground truth. Results Our method achieved sleep-wake classification accuracy of 84%, sensitivity of 95%, and specificity of 47%. Apple Watch overestimated total sleep time (mean+SD) by 39.4 + 57.7 mins, underestimated WASO by 45.5 + 54.6 mins and the number of awakenings by 5.0 + 6.9. We observed worse performance for participants who had PSGs exhibiting frequent respiratory events. Discussion Accelerometry cannot replace PSG for diagnostic purposes. However, the Apple Watch results compare favourably to previously published Actiwatch-PSG comparisons. The performance we measured suggests that Apple Watch based accelerometry could be used in longitudinal studies to gather information similar to clinically validated accelerometers, potentially on a larger scale for lower cost. Further study is needed to understand how sleep disorders affect this kind of measurement.