Abstracts

Rationale: Due to the increased awareness of nonconvulsive seizures (NCS) in patients admitted to the Neuro-ICU, the utilization of continuous EEG monitoring (cEEG) is rising. Although quantitative EEG (qEEG) software is widely available, there are few studies evaluating the utility of qEEG in the adult population; furthermore, most of these studies evaluate individual qEEG tools, when, clinically, panels of tools are often used. The aim of this study is to evaluate the sensitivity and specificity of qEEG trends for seizure detection in adult patients in the Neuro-ICU.Methods: The dataset for this study consisted of 180 qEEG slides that were collected retrospectively from 45 patients who were admitted to the Duke Neuro-ICU (30 patients with NCS and 15 patients without NCS). Each qEEG slide contained one hour of data and consisted of the following qEEG tools: rhythmicity, fast Fourier transform displayed as a density spectral array, amplitude integrated EEG, and EEG asymmetry index. The corresponding raw EEG segments were reviewed independently by the study authors to identify seizures and placed in the following categories: no seizures, 1-2, 3-5, 6-10 or >10 seizures. The randomized qEEG panels (n=180) were distributed to 5 electroencephalographers and these reviewers were asked to determine the number of seizures by selecting one of the aforementioned categories. The reviewers did not have access to the corresponding raw EEG data. Results: There was a sensitivity of 0.76 (95% CI 0.72-0.79) and specificity of 0.54 (95% CI 0.49-0.59) for the reviewer s ability to detect the presence of seizures on qEEG panels when compared with the gold standard of independent raw EEG review. The positive predictive value for a seizure was 71% (95% CI 67-74%) and the negative predictive value was 61% (95% CI 56-66%). However, reviewers correctly identified the number of seizures only 36% of the time. The interrater reliability for the detection of presence or absence of seizures was 0.80. When comparing short seizures (<25th percentile, <41 s) with long seizures (>75th percentile, >121 s), there was no difference in the reviewer s ability to detect the presence of seizures (p=0.55). In addition, there was no difference in the reviewer s ability to detect the presence of seizures based on the spatial distribution of seizures.Conclusions: Although isolated review of commonly-used qEEG trends by 5 electroencephalographers demonstrates acceptable sensitivity and specificity for the detection of the presence of seizures, their ability to quantify the number of seizures was poor. Even when used as a panel, qEEG trends do not appear to be sufficient as the sole method for reviewing cEEG data. Additional training for users, use of other trends and simultaneous availability of raw EEG with qEEG trends might be helpful. In addition, using qEEG trends as a means for non-neurophysiologists may still be a reasonable means to identifiy periods of concern during live recording.