Performance of wearable finger ring trackers for diagnostic sleep measurement in the clinical context

Ring-trackers are a growing consumer wearable category that provide a number of sleep metrics, yet their measurement accuracy remains poorly understood. Previous validation studies have mainly focused on healthy individuals, while a significant part of the potential present and future value lies in applications on non-healthy subjects. To enable applications in research and medical applications, rigorous evaluation of performance in clinical settings against the gold-standard polysomnography is needed. To address this knowledge gap, we investigated how the measurements of three commercially available ring trackers (Oura, SleepOn, Circul) perform against polysomnography in a university sleep lab population with a diverse set sleep-related disorders as well as sleep-unrelated medical conditions. We evaluated individual-level and group-level discrepancies of standard sleep measures and conducted an epoch-by-epoch analysis of sleep staging classification performance using a standardized analysis framework. While average group-level sleep measures are similar (e.g., TST differences between rings and gold standard were below 12 min for the Oura ring), individual-level differences often remained large. Ring-derived sleep metrics were characterized by complex bias, indicating that their correction is non-trivial. Sleep/Wake distinction of the Oura and SleepOn rings reached similar performance as previously reported for healthy individuals (~ 85% accuracy), but was worse for the Circul ring (~ 65% accuracy). Sleep stage classification (Wake, Light, Deep, REM sleep) sensitivity ranged from 0.14 (REM sleep classification of the Circul ring) to 0.58 (Light sleep classification of the SleepOn ring). Across all sleep stages, the Oura and SleepOn ring performed similarly (53.18% and 50.48% accuracy), whereas the Circul ring performed worse (35.06% accuracy). Our findings confirm recent descriptions of device-related bias and additionally uncover practical limitations in the application in a real-world sleep laboratory patient cohort. Critically, while some devices may demonstrate reasonable agreement with PSG on average, this agreement masks substantial individual-level inaccuracies, prohibiting their use in clinical sleep medicine, as accurate assessment of individual nights, including both nights with exceptionally low or high sleep quality and quantity, is essential for patient care.