Sleep staging based on autonomic signals: a multi-center validation study.

STUDY OBJECTIVES

One of the most important caveats of ambulatory devices is the inability to record and stage sleep. We assessed an algorithm determining 4 different stages: wake, light sleep, deep sleep, and REM sleep using signals derived from the portable monitor Watch-PAT100 (PAT recorder).

METHODS

Participants (38 normal subjects and 189 patients with obstructive sleep apnea [OSA]) underwent simultaneous overnight recordings with polysomnography (PSG) and the PAT recorder in a study originally designed to assess the accuracy of the PAT recorder in diagnosing OSA. Light/deep sleep and REM sleep from the PAT recorder recording were automatically scored based on features extracted from time series of peripheral arterial tone amplitudes and inter pulse periods. The PSG scored sleep stages 1 and 2 were classified as light sleep for epoch-by-epoch comparisons.

RESULTS

The overall agreement in detecting light/deep and REM sleep were 88.6% ± 5.9% and 88.7% ± 5.5%, respectively. There was a good agreement between PSG and the PAT recorder in quantifying sleep efficiency (78.4% ± 9.9% vs. 78.8% ± 13.4%), REM latency (237 ± 148 vs. 225 ± 159 epochs), and REM percentage (14.4% ± 6.5% vs. 19.3% ± 8.7%). OSA severity did not affect the sensitivity and specificity of the algorithm. The Cohen κ coefficient for detecting all sleep stages: sleep from wake, REM from NREM sleep, and deep from light sleep were 0.48, 0.55, 0.59, and 0.46, respectively.

CONCLUSIONS

Analysis of autonomic signals from the PAT recorder can detect sleep stages with moderate agreement to more standard techniques in normal subjects and OSA patients. This novel algorithm may provide insights on sleep and sleep architecture when applying the PAT recorder for OSA diagnosis.