Abstracts

Rationale: EEG sensitivity in children after a first unprovoked seizure is reported to range from 0.34-0.81, specificity from 0.43-0.93, and up to 38% of children with a normal EEG after a first seizure will go on to develop epilepsy (Eur J Neurol 2016;23(3):455-463). Computational analysis may improve the diagnostic yield of EEG by objectively quantifying patterns and trends not readily apparent by visual inspection. Entropy is a nonlinear measure of the relative predictability and complexity of a signal, and an association between lower EEG entropy and epilepsy has been reported. Frequency band power is a quantitative measure often used to characterize EEG signals. We sought to determine whether EEG entropy and frequency band power are useful measures for identifying pediatric patients at risk of developing epilepsy following a first unprovoked seizure and a normal EEG. Methods: This retrospective study included 32 pediatric patients (mean age 5.9 years, range 10 months - 16 years) with a single unprovoked seizure and a subsequent clinical EEG read as normal. By chart review, 13 patients had a second seizure within two years of the first seizure and 19 patients did not. Five minutes of awake EEG recorded after the initial seizure was analyzed, in broadband (1-70 Hz), delta (1-4 Hz), theta (4-8 Hz), alpha (8-13 Hz), beta (13-30 Hz), and gamma (30-80 Hz) frequency bands. For each patient, Shannon entropy and frequency band log power of EEG signal amplitude were calculated for each channel and then averaged across all channels. Results: Mean entropy correlated highly with mean log power across all frequency bands (rho=0.80 to 0.99, p < 4e-8). Multiple linear regression showed that age and future seizure status individually and combined were significant predictors of mean entropy and mean log power in all frequency bands (e.g. in a linear regression model including both age and future seizure status, gamma band mean entropy F(1,29)=22.1, p=1.5e-6, adjusted R^2=0.58; gamma band mean log power F(1,29)=23.4, p=8.9e-7, adjusted R^2=0.59). Linear discriminant analysis with leave-one-out cross-validation demonstrated that mean entropy and mean log power measures in all frequency bands had a sensitivity between 0.46-0.62 and a specificity between 0.89-1 to identify patients who would develop future seizures (see figure for an example). Conclusions: EEG entropy and log power change with age and may increase the diagnostic yield of routine EEG to identify children at higher risk of developing epilepsy after a first unprovoked seizure. Prospective validation is needed. Funding: NIH NINDS 3R25NS070682-04S1 Training Grant, Matthew Siravo Memorial Foundation Epilepsy Research Grant, and Alpert Medical School Summer Research Assistant Program