2010
American Epilepsy Society Poster (Abst. 1.096)
Analysis of Continuous Electroencephalography in the Intensive Care Unit for Nonconvulsive Seizure Activity- A Comparison of Four Methods
Authors: Y. Taher, Evan Fertig, A. Paige, J. Politsky, C. Lambrakis, O. Laban, J. Lee and M. Lancman
Rationale:
Nonconvulsive seizures (NCS) are more prevalent in the ICU than previously thought, and Continuous Electroencephalography (CEEG) is increasingly employed to detect and respond to them. Collecting CEEG digitally generates copious data, and the most efficient way to review it is unknown. Complete review by an encephalographer is the gold standard but is time consuming. Other methods include automatic seizure detection algorithms which are included most CEEG programs, scheduled review (i.e., review 5 out of every 30 minutes), and Compressed Spectral Array (CSA). The purpose of this investigation was to compare the sensitivity, specificity, and positive predictive value (PPV) of these methods using complete review as the gold standard.
Methods:
Nine patients were identified with NCS (patients who had clinically obvious events were excluded) who were monitored in 3 New York State ICU’s since 2009. Synchronized digital video and digital EEG (Ceegraph Vision, version: 7.03.06) with International 10-20 system was used. The indication for all was altered mental status. A 24 hour period with at least 1 NCS was selected. Comprehensive review was performed by an electroencephalographer, and seizure onset and offset times were recorded. Automated seizure detection was performed with the following parameters: Amplitude Threshold: 2.7, Detection Threshold: 1, Min. Frequency: 3, Max Frequency of Variation: 40. The CEEG was reanalyzed using only the first 5 minutes of each 30 minutes of the selected 24 hour period. FFT frequency analysis will be used to generate a spectrogram, and then another encephalographer who is blinded to the location of seizures will review the spectrograms and identify peaks of interest. Peaks which include seizure activity 2.5 minutes before or after them will be considered positive. True positive, true negative, false positive, and false negative rates are calculated.
Results:
The mean age was 55 years (3 days to 81 years). Etiology was unknown in 4 cases, post-anoxic in 3 cases, and ICH in 2 cases. The mean seizures per patient/day was 9.4 (1-16). Most seizures lasted approximately 60 seconds with the longest lasting over 120 minutes. For some, there were multiple scheduled reviews or automatic detections. The sensitivity, specificity, and PPV of the seizure detection algorithm were 86%, 99%, and 84%. The sensitivity, specificity, and PPV of scheduled review were 27%, 97%, and 5%. CSA analysis is ongoing.
Conclusions:
This investigation indicates that relying on automatic seizure detection or scheduled review alone will result in missing a significant number of NCS in the ICU, and as a consequence, may lead to incorrect therapeutic decision-making. The yield of scheduled review in particular is very low. These early results suggest that complete review of CEEG data should be performed. Future analysis will examine a larger patient number, different automatic seizure detection parameters and programs, the utility of CSA, and yield of combining these methods.