Abstracts

Rationale: Use of continuous EEG monitoring in ICU settings has been expanding rapidly. However, reliable interpretation of prolonged EEG recordings in critically ill patients requires significant expertise and is extremely time consuming, which severely limits its rapid and widespread use. As an initial step in building an accurate seizure detector for use in ICU settings, we investigated the sensitivity and specificity of quantitative EEG (qEEG) trending for rapidly identifying frequent non-convulsive seizures in ICU patients during continuous EEG monitoring. Methods: EEG recordings recorded from five ICU patients were studied. The total duration of the EEG analyzed was about 48 hours, which contained 142 non-convulsive seizures. The qEEG trending consisted of measures of signal regularity (pattern-match regularity statistic, PMRS), amplitude variation (AV), and maximal local frequency (MLF), all calculated for each 5.12 second epoch. Since the PMRS value decreases with the increase of signal regularity, it was hypothesized that PMRS values would drop significantly during ictal activity. AV and MLF were used to reject drops of PMRS values due to rhythmic artifacts or normal physiological movements. In addition, since seizures in ICU patients vary more spatially compared to those in EMU patients, trending in multiple recording regions was calculated simultaneously. The resulting qEEG trending was analyzed to investigate how PMRS drops (at least 2 standard deviations below the mean of the preceding 60 seconds) correlated with the occurrences of non-convulsive seizures. Results: Figure 1 demonstrates an example of PMRS trending derived from four recording regions in a 120-minute recording with 10 seizures recorded. Each drop in PMRS values coincided with a 15-20 second ictal discharge. Overall, the PMRS trending accurately detected 141 out of the 142 non-convulsive seizures (99.3% sensitivity). The single seizure missed by PMRS trending was due to the presence of an electrode artifact (high AV) during the ictal period. In addition, there was only one false positive detection during the entire 48 hours of recording. Conclusions: These findings suggest that it is feasible to develop a robust and reliable ICU seizure monitoring system based on multi-region qEEG analysis. Application of these trending measures to a larger population of ICU EEG recordings will help further define the strengths of this novel qEEG technique.