Abstracts

Rationale: Interictal epileptiform discharges (IEDs) are reliable markers of seizure recurrence risk on routine electroencephalogram (EEG). Unfortunately, IEDs can be elusive on routine 30-minute recordings. Sample Entropy (SampEn) can estimate the entropy of biological signals and has shown promise in the identification and real-time prediction of seizures (Biomed Signal Process Control 2012: 7(4), 401). In this work, we evaluated if SampEn on routine EEG could predict long-term seizure recurrence risk.

Methods: We selected consecutive patients undergoing 30-minute ambulatory EEG at the Centre Hospitalier de l’Université de Montréal between December 2017 and July 2018. Exclusion criteria were absence of follow-up after the EEG, unclear epilepsy diagnosis at the end of the follow-up period, and electrical seizure(s) on EEG. Medical charts were reviewed for seizure recurrence after the EEG and factors associated with seizure recurrence: age, abnormal slowing on EEG, presence of IEDs on EEG, focal brain lesion on neuroimaging, and a number of epilepsy risk factors (developmental delay, traumatic brain injury, stroke, brain tumor, history of febrile seizure, history of central nervous system infection, and first-degree relative with epilepsy). EEG signals were band-pass filtered in the delta (1 – 4 Hz), theta (4 – 8 Hz), alpha (8 – 13 Hz) and beta (13 – 40 Hz) frequency ranges. SampEn was calculated for all channels using 9-s non-overlapping windows and values aggregated at the 25th percentile to obtain a single SampEn value for each EEG. We performed a multivariate logistic regression to test the association between SampEn and seizure recurrence after the EEG while controlling for the selected co-predictors. We also tested the predictive performance of SampEn alone for seizure recurrence in patients of different age groups compared to the presence of IED using a receiver-operating characteristic (ROC) curve.

Results: 335 EEGs from 271 patients were included. The median follow-up period was 84 weeks (IQR 32.5 – 116.5). Out of these EEGs, 208 (62%) were from patients diagnosed (167 EEGs) or later diagnosed (41 EEGs) with epilepsy. For 106 EEGs (31.6%), the patient had an epileptic seizure following the EEG. Patients with seizure recurrence had lower SampEn (0.813 vs. 0.836, p = 0.009). This difference was primarily attributable to patients ≤ 40-year-old (0.81 vs. 0.86, p = 0.0005, Figure 1A). The multivariate model demonstrated that each 1SD reduction in SampEn was associated with a 1.64-fold increased risk of seizure recurrence (95% CI 1.25 – 2.12, p < 0.0001), independently of other risk factors. 129 (38.5%) EEGs were from patients ≤ 40-year-old, and prevalence of seizure recurrence within this subgroup was 46.5%. In this subgroup, the area under the ROC curve (ROC AUC) for the prediction of seizure recurrence was 0.66 (95% CI: 0.57 – 0.76, Figure 1B). By contrast, the presence of IEDs (27 EEGs, 20.9%) had a ROC AUC of 0.61 (0.53 – 0.67). Linear combination of SampEn and IED yielded a ROC AUC of 0.71 (0.62 – 0.80).