Abstract Introduction Accurate sleep staging in the ambulatory setting with EEG is challenging and increases patient burden. The WatchPAT (WP, ZOLL Itamar), is a widely used PAT-based home sleep apnea test. The company recently applied modern supervised machine learning techniques to develop a new algorithm for comprehensive sleep staging based on clinical features and accumulated knowledge gleaned over 20 years of field experience. Over 900 simultaneous nocturnal WP and polysomnography (PSG) recordings were used in algorithm development to generate all stages (N1, N2, N3, REM and wakefulness). A Long Short-Term Memory (LSTM) network and a Hidden Markov Model based decoder were trained on a wide range of clinically significant features (hemodynamic, HRV, respiratory, oximetry, atonia, etc.), derived from WP channels. This study aimed to evaluate the performance of the new algorithm compared to gold standard PSG for sleep stage classification. Methods Prospective multi-center study with simultaneous WP and PSG recordings were conducted to validate the algorithm. PSG recordings were staged according to AASM guidelines by two independent scorers. We examined the agreement between WP to the consensus of PSG scorers (approximately 80% of all epochs) and to each scorer, individually, with Cohen’s kappa (k). Results The validation set of 231 subjects (141 Males, Age 52.4±17.8 years, BMI 32.0±8.6) was included in the final analysis. The average percentage of each sleep stage was similar between WP and consensus. Percentages SD for WP and PSG respectively were: Wake: 26.57% 16.63, 26.38% 16.96; N1: 2.5% 1.79, 3.72% 3.81; N2: 47.53% 13.34, 48.83% 15.76; N3: 8.69% 8.42, 9.38% 8.65 and REM: 14.72% 8.76, 11.69% 8.36. Epochs with consensus between scorers had substantial agreement with the algorithm for AASM stages (k=0.65), wake/sleep separation (k=0.77), and REM/NREM (k=0.73). When including epochs without scorer consensus, K between the WP and individual scorers for AASM stages were 0.51 and 0.59 (moderate agreement), suggesting that mixed or ambiguous features in some epochs accounted for reductions in agreement between scorers and the algorithm. Conclusion The new WP staging algorithm demonstrated substantial agreement with PSG for sleep stage classification. WP holds promise for detailed sleep characterization in the ambulatory setting, without EEG. Support (if any) ZOLL-Itamar, Caesarea, Israel
Pham et al. (Fri,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: