Validation of Cerebral State Monitor Frequency Power Ratios for Detection of Delirium

Background

Delirium among hospitalized patients often goes undetected and is associated with increased length of stay and mortality. Cerebral state monitors (CSMs) collect limited lead electroencephalography (EEG) data and have shown promise in delirium detection.

Objective

This study compares three methods of detecting delirium: EEG using an Food and Drug Administration-approved CSM, a 3-minute diagnostic interview for the Confusion Assessment Method (3D-CAM), and the traditional reference standard clinical evaluation by a psychiatric consultation-liaison service.

Methods

Hospitalized patients >18 years of age evaluated by the consultation-liaison service were screened for inclusion. Consent was obtained from either the patient or authorized surrogate when appropriate. Participants were screened with 3D-CAM, followed by placement of 4 frontotemporal CSM leads and collection of 5 minutes of EEG data with eyes closed. The presence or absence of a delirium diagnosis on psychiatric evaluation using Diagnostic and Statistical Manual-5 criteria the day of data collection was recorded. A MATLAB-based program (Brainstorm) was used to calculate power in each EEG frequency band.

Results

There were 75 participants, 58 with complete 3D-CAM and EEG data. Twenty-three participants were found to be delirious by clinical diagnosis. The 3D-CAM differentiated between delirious and nondelirious participants with a sensitivity of 65%, specificity of 79%, and area under the curve of 0.720. EEG power differed between delirious and nondelirious participants in the HighDelta frequency band in the left temporal (L2) lead (false discovery rate-adjusted P = 0.020) and the LowAlphaHighDelta frequency band ratio in L2 (false discovery rate-adjusted P = 0.039). On receiver operator curve analysis, LowAlphaHighDelta and HighDelta outperformed 3D CAM, although not significant (all P > 0.18).

Conclusions

CSM frequency band power differed significantly between delirious and nondelirious patients but did not outperform 3D-CAM as a predictive test of delirium. Strengths of this study include the use of an Food and Drug Administration-approved CSM and freely available computational software and methods. Limitations include a small sample size, making the results vulnerable to skewing by outliers. Future studies should recruit larger sample sizes and explore the diagnostic utility of combined EEG variables obtained with CSMs at the bedside.