Multi-Night Validation of a Sleep Tracking Ring in Adolescents Compared with a Research Actigraph and Polysomnography

Abstract

Background: Wearable devices have tremendous potential for large-scale longitudinal measurement of sleep, but their accuracy needs to be validated. We compared the performance of the multisensor Oura ring (Oura Health Oy, Oulu, Finland) to polysomnography (PSG) and a research actigraph in healthy adolescents.

Methods: Fifty-three adolescents (28 females; aged 15-19 years) underwent overnight PSG monitoring while wearing both an Oura ring and Actiwatch 2 (Philips Respironics, USA). Measurements were made over multiple nights and across three levels of sleep opportunity (5 nights with either 6.5 or 8h, and 3 nights with 9h). Actiwatch data at two sensitivity settings were analyzed. Discrepancies in estimated sleep measures as well as sleep-wake, and sleep stage agreements were evaluated using Bland-Altman plots and epoch-by-epoch (EBE) analyses.

Results: Compared with PSG, Oura consistently underestimated TST by an average of 32.8 to 47.3 minutes (Ps < 0.001) across the different TIB conditions; Actiwatch 2 at its default setting underestimated TST by 25.8 to 33.9 minutes. Oura significantly overestimated WASO by an average of 30.7 to 46.3 minutes. It was comparable to Actiwatch 2 at default sensitivity in the 6.5, and 8h TIB conditions. Relative to PSG, Oura significantly underestimated REM sleep (12.8 to 19.5 minutes) and light sleep (51.1 to 81.2 minutes) but overestimated N3 by 31.5 to 46.8 minutes (Ps < 0.01). EBE analyses demonstrated excellent sleep-wake accuracies, specificities, and sensitivities - between 0.88 and 0.89 across all TIBs.

Conclusion: The Oura ring yielded comparable sleep measurement to research grade actigraphy at the latter's default settings. Sleep staging needs improvement. However, the device appears adequate for characterizing the effect of sleep duration manipulation on adolescent sleep macro-architecture.

Keywords: actigraphy; adolescents; polysomnography; validation; wearable.

Figure 1

Bland−Altman plots of mean bias with upper and lower bands of agreement between polysomnography (PSG), Oura and Actiwatch H5 and M10 settings for each TIB condition (Red: 6.5-hour, green: 8-hour, and blue: 9-hour). A mean bias line above and below zero demonstrates overestimation and underestimation of the device against PSG, respectively. Bland-Altman plots of (A) total sleep time (TST), and (B) wake after sleep onset (WASO) for Actiwatch M10, H5, and Oura, respectively. The solid line indicates the mean value of bias and the dashed line represent 1.96 SD limits of agreement.

Figure 2

Bland–Altman plots of (A) light, (B) deep, and (C) REM sleep, respectively between polysomnography (PSG) and Oura. Each TIB condition is color-coded with red: 6.5-hour, green: 8-hour, and blue: 9-hour sleep. The solid line indicates the mean value of bias and the dashed line represent 1.96 SD limits of agreement.

Figure 3

TST, WASO, and sleep stages measured by PSG (blue lines) and Oura ring (red lines) for the Continuous (dotted lines) and Split (solid lines) sleep groups across the manipulation nights. Error bars denote 95% confidence intervals. Blue asterisks denote significant differences between groups with PSG measures, while red asterisks denote significant differences between groups with Oura measures. *P < 0.05; **P < 0.01; ***P < 0.001.