Background: Recognition and diagnosis of delirium in ICU patients is difficult due to the fluctuating character and the subjective and discontinuous nature of clinical assessment such as CAM-ICU screening. Hence, there is clear need for an objective, continuous measurement method. It is known that delirium is associated with changes in the electroencephalogram (EEG), reflected as an increase in theta and delta and reduction in alpha power. This makes EEG a candidate tool for delirium detection and monitoring in the ICU.
Objective: To determine if changes in EEG activity are indicative for delirium and whether these signals can be used for continuous delirium detection at the ICU.
Methods: In a case-control observational study, continuous EEG recording was applied (maximum 5 days). Exclusion criteria were RASS ≤-3 and neurological disorders. Patients were screened using CAM-ICU 3xday. Around each CAM-ICU score a window of 1 hour was defined within 5 minutes of artifact-free EEG was selected and divided in 30 epochs of 10 seconds. Relative power was computed for the four frequency bands delta, theta, alpha and beta. Furthermore, a subset of five EEG features, including alpha to delta (AD) ratio, theta to alpha (TA) ratio, center of gravity (COG), approximate entropy (AE) and spectral variability (SV) were calculated for each EEG epoch and combined into a learning K-Nearest Neighbor Classifier (KNNC).
Results: 21 patients, mean age of 67±7 years, were included of which 10 delirious patients and 11 non-delirious controls. Spectral analysis of the regular frequency did not show differences between the delirium and control group, however, four of the five EEG features (COG; p=0.01, SV; p=0.01, TA-ratio; p=0.02 and AD-ratio; p=0.01) used for classification revealed significant differences. KNNC achieved an accuracy of 67% and a sensitivity and specificity of 47% and 89% respectively.
Conclusions: As opposed to other studies, spectral analysis of the frequency bands revealed no significant differences between delirious and control patients. This could be due to sedatives, the heterogeneous ICU population or severity of illness for which no correction was made. Although the classifier only reached a moderate sensitivity and specificity, significant differences are seen in four EEG features to distinguish delirious and control patients. Future research is necessary to increase the performance of EEG classification; however this method could have potential for continuous delirium detection.
Wednesday 10 September 2014, 16:30 - 17:30 h
Building Carré - room CR 3.718