Objective:

We aim to develop a new compact wearable EEG device and paired, automated analysis pipeline to monitor continuous EEG and predict delirium. 

Background:

Delirium is a highly prevalent yet underdiagnosed neuropsychiatric condition which primarily affects older, hospitalized patients. Better objective diagnostic measures are needed to detect delirium reliably. 

Design/Methods:

We applied a wireless, wearable single-channel EEG device to hospitalized patients admitted for at least one night and evaluated for delirium using the 3D-CAM. We evaluated device comfort and recording yield. Recordings were preprocessed, segmented into 4-second epochs, and used to extract 3,444 time-based, frequency-based, and non-linear EEG features. We trained machine-learning classifiers using XGBoost to predict delirium, using stratified cross-validation.  

Results:

We enrolled 155 adult inpatients, 40 of whom met 3DCAM criteria for delirium (25.8%). Over 80% of participants found the device to be “very positive” or “positive” for comfort and willingness to wear again. 97% of patients found the device to have no impact on sleep. Our device reliably detected delirium with an AUROC of 0.80 using our comprehensive quantitative EEG feature set.  Spectral features derived from the Catch22 and FOOOF packages as well as relative theta consistently ranked highly on Shapley analysis.

Conclusions:

Continuous wearable EEG is well-tolerated in hospitalized patients and yields high quality signals from which delirium can be reliably predicted with competitive accuracy. This wearable EEG system may serve as a crucial tool for neural monitoring, thereby accelerating the development of preventative strategies and targeted treatments for patients with delirium or other acute neurological illnesses.