Abstract presented at “SLEEP 2006” – 20th Anniversary Meeting of the AOSS

Preliminary Validation of a New Device for Studying Sleep

Germain A, Buysse DJ, Kupfer DJ

Abstract

Introduction. This study examined the validity of the BodyMedia's SenseWear® armband as new tool to assess sleep and wakefulness, using polysomnography (PSG) as the gold standard. SenseWear measures body movement, like actigraphy, but also acquires other physiological signals, including surface body temperature, galvanic skin response, and heat flux, that may permit distinctions between NREM and REM sleep.

Methods. Twenty-seven participants (Mean age = 28.7 years, 16 women) underwent PSG recordings while wearing the armband. PSG and armband data were scored in 20-second epochs. BodyMedia's sleep algorithm was developed using techniques from statistical machine learning. Relevant features that distinguish characteristics of NREM and REM were first identified, including length of inactivity and rate of change in heat-flux. These features were then combined using a method that incorporates the probability of the feature for each sleep state, of each sleep state at each time point, and of different sequences of sleep states. Performance was evaluated using the method of "leave one out" cross-validation, which repeatedly trains the model on all but one of the subjects and tests on the remaining subject. Monte Carlo simulations were used to evaluate the rate of correct sleep stage detection attributable to chance.

Results. The algorithm correctly identified 93% of all sleep epochs, and 83% of all wakefulness epochs, for an overall epoch-by-epoch accuracy of 89%. The algorithm correctly identified 65% of all NREM sleep epochs (vs. 43.9% by chance), and 45.6% of all REM sleep epochs (vs.13.9% by chance), for an epoch-by-epoch accuracy of 70% (vs. 39% by chance).

Conclusion. BodyMedia's algorithm identified sleep and wakefulness with moderate to high sensitivity, specificity, and accuracy. Detection of NREM and REM substantially exceeds chance levels. Although further improvement is needed, SenseWear appears to be a promising, unobtrusive means for measuring NREM and REM as well as overall sleep and wakefulness.

Support (optional). The present work was support by the Mental Health Research and Intervention Center of the University of Pittsburgh through an NIMH funded grant (MH30195), the Fonds de la recherché en Santé du Québec, and the Canadian Institutes of Health Research.

Publication: Abstract presented at “SLEEP 2006” – 20th Anniversary Meeting of the Associated Professional Sleep Societies (AOSS), 2006 Salt Lake Ciry, Utah. USA