JMIR Publications

ABSTRACT

Background:

Contactless sleep technologies (CSTs) hold promise for longitudinal, unobtrusive sleep monitoring in health and disease at scale, particularly in older people where the increased incidence of sleep abnormalities with aging is considered a risk factor for several neurodegenerative disorders. However, few CST have been evaluated in older people.

Objective:

To evaluate the performance of three contactless sleep technologies (a bedside radar [Somnofy] and two under-mattress devices [Withings Sleep Analyser and Emfit-QS]) compared to polysomnography (PSG) and actigraphy [Actiwatch Spectrum] recorded during a first night in a sleep laboratory, 10-hour time in bed protocol, which induced mild sleep disturbance.

Methods:

Thirty-five older men and women (70.8±4.9 years; 14 women) several of whom had comorbidities and/or sleep apnoea, participated in the study. Devices were evaluated by estimating a range of performance metrics for classification of sleep vs wake, and NREM and REM sleep stages (sleep summary and epoch by epoch concordance) and comparing to PSG metrics.

Results:

All three CSTs overestimated total sleep time (bias [mean]: > 90 min) and sleep efficiency (bias: > 13 %) with an associated underestimation of wake after sleep onset (bias: > 50 min). Sleep onset latency was accurately detected by the bedside radar (bias: <6 mins) while the under-mattress devices overestimated this sleep parameter (bias: >16 mins). CSTs did not perform as well as actigraphy in estimating the all-night sleep summary measures. The bedside radar performed better in discriminating sleep vs wake (MCC [mean and 95% CI]: 0.63 [0.57 0.69]) than the under-mattress devices (MCC: =0.41 [0.36 0.46]; Emfit-QS =0.35 [0.26 0.43]). Accuracy of identifying REM and Light sleep was poor across all CSTs while deep sleep was predicted with moderate accuracy (MCC: >0.45) by both Somnofy and Withings Sleep Analyser. The deep sleep duration estimates of Somnofy was found to be significantly correlated (r2=0.6, p<0.01) with average slow wave activity (0.75-4.5 Hz) detected with PSG while for the under-mattress devices this correlation was not significant (Withings Sleep Analyser [r2=0.0096, p=0.21] and Emfit [r2=0.11, p=0.58]).

Conclusions:

Overall, the CSTs overestimated total sleep time and the sleep stage prediction was unsatisfactory in this group of older people in whom sleep efficiency was relatively low (71%). Our findings indicate that even though these devices provide information on sleep which may be useful, for particular use-cases, the performance of CSTs requires improvement before they can serve as an alternative to PSG in estimating most sleep variables in older individuals.