In Annals of neurology ; h5-index 85.0
OBJECTIVES : REM sleep behavior disorder (RBD) is a potentially harmful, often overlooked sleep disorder affecting up to 70% of Parkinson's disease patients. Current diagnosis relies on nocturnal video-polysomnography, which is an expensive and cumbersome exam requiring specific clinical expertise. Here, we explored the use of wrist actigraphy to enable automatic RBD diagnoses in home settings.
METHODS : Twenty-six Parkinson's patients underwent two-week home wrist actigraphy, followed by two in-lab evaluations. Patients were classified as RBD vs. non-RBD based on dream enactment history and video-polysomnography. We comprehensively characterized patients' movement patterns during sleep using actigraphic signals. We then trained machine learning classification algorithms to discriminate patients with or without RBD using the most relevant features. Classification performance was quantified with respect to clinical diagnosis, separately for in-lab and at-home recordings. Performance was further validated in a control group of non-PD patients with other sleep conditions.
RESULTS : To characterize RBD, actigraphic features extracted from both (i) individual movement episodes and (ii) global nocturnal activity were critical. RBD patients were more active overall, and exhibited movements that were shorter, of higher magnitude, and more scattered in time. Using these features, our classification algorithms reached an accuracy of 92.9±8.16% during in-clinic tests. When validated on home recordings in Parkinson's patients, accuracy reached 100% over a two-week window, and was 94.4% in non-PD control patients. Features showed robustness across tests and conditions.
INTERPRETATIONS : These results open new perspectives for faster, cheaper, and more regular screening of sleep disorders, both for routine clinical practice and clinical trials. This article is protected by copyright. All rights reserved.
Raschellà Flavio, Scafa Stefano, Puiatti Alessandro, Martin Moraud Eduardo, Ratti Pietro-Luca