1075 Sleep-stage Independent Electroencephalography Features For Classification Of Veterans With Post-traumatic Stress Disorder. (27th May 2020)

• Record Type: Journal Article
• Authors: Laxminarayan, S; Wang, C; Oyama, T; Cashmere, D; Germain, A; Reifman, J
• Is Part Of: Sleep. Volume 43(2020)Supplement 1
• Page Start: A409
• Page End: A410
• Publication Date: 2020-05-27
• Subjects: Sleep -- Physiological aspects -- Periodicals; Sleep disorders -- Periodicals; Sommeil -- Aspect physiologique -- Périodiques; Sommeil, Troubles du -- Périodiques; Sleep disorders; Sleep -- Physiological aspects; Sleep -- physiological aspects; Sleep Wake Disorders; Psychophysiology; Electronic journals; Periodicals; 616.8498
• DOI: 10.1093/sleep/zsaa056.1071
• Languages: English
• ISSNs: 0161-8105

Abstract

Abstract: Introduction: Prior sleep studies have suggested that electroencephalography (EEG) spectral power and synchrony features in certain sleep stages differ significantly at the group-average level between subjects with and without post-traumatic stress disorder (PTSD). Here, we investigated whether a multivariate combination of sleep-stage independent EEG features could objectively identify individual subjects with PTSD. Methods: We analyzed EEG data recorded from 78 combat-exposed veteran men with (n = 31) and without (n = 47) PTSD during two consecutive nights of sleep. For each subject we computed 780 features from 10 EEG channels covering the whole brain, by averaging the values over the entire night regardless of sleep stage. Using a training set consisting of the first 47 consecutive subjects (18 with PTSD) of the study, we performed univariate feature selection and backward feature elimination using a logistic regression model. We then evaluated the model on the test set, which consisted of the remaining 31 subjects (13 with PTSD). We assessed model performance by computing the area under the receiver operating characteristic curve (AUC). Results: Feature elimination using the logistic regression model yielded three uncorrelated features that were consistently discriminative of PTSD across the two consecutive nights. When we trained the logistic model consisting of these three features using data from both nights of the training set, the model yielded test-set AUCs of 0.84 and 0.80 for Night 1 and Night 2, respectively. These values were considerably larger than the test-set AUCs of the three individual features, which ranged from 0.55 to 0.74 across both nights. Conclusion: We identified robust, stage-independent, whole-night features and combined them in a logistic regression model to discriminate subjects with and without PTSD. The model yielded AUCs above 0.80 on the test data, showing promise as an objective approach to diagnose PTSD at the individual level. Support: This work was sponsored by U.S. Defense Health Program (grant No. W81XWH-14-2-0145) and managed by the U.S. Army Military Operational Medicine Program Area Directorate, Ft. Detrick, MD. The study was also supported by the Clinical and Translational Science Institute at the University of Pittsburgh (UL1 TR001857).