To determine whether combining epileptiform abnormality (EA) identification and quantitative electroencephalography (qEEG) spectral analyses predicts post-traumatic epilepsy (PTE).

EAs, including early seizures, sporadic epileptiform discharges (ED), lateralized or generalized periodic discharges (LPD, GPD), and lateralized rhythmic delta activity (LRDA), are more frequently present in PTE than non-PTE patients. In contrast, qEEG spectral analyses (rhythmicity, FFT Power, suppression) provide a comprehensive evaluation of electrophysiological brain signatures after traumatic brain injury (TBI). We hypothesize that combining early post-TBI EAs and spectral features predicts first-year PTE.

We conducted a multi-site, retrospective, case-control study of 56 PTE and 56 non-PTE patients matched by admission TBI severity, age and gender. We analyzed EEGs within 14 days post-TBI. EAs and qEEG features were computed using two independent algorithms – Persyst 14™ and a novel EA detector (Westover lab, unpublished). We performed bivariate analyses to identify EEG risk factors for PTE, controlling for recording duration. We also performed multivariate analyses leveraging forward model selection techniques and LASSO-regularization (8-fold cross-validated) for feature selection.

In bivariate analyses, FFT sub-band (delta, theta, beta) power variability (PDV, PTV, PBV), LRDA-GRDA ratio, and LPD significantly increase PTE risk (p<0.05). Rhythmicity delta asymmetry, suppression asymmetry, and GPD trend toward significant (p<0.1). Multivariate logistic regression including LRDA-GRDA ratio (p<0.01), GPD (p<0.05), and PBV (p<0.05) yields an AUC of 0.733 (95%CI 0.640-0.826). The resulting accuracy is 68% (sensitivity-70%, specificity-64%). LASSO-regularized logistic regression identifies a broader range of features (LRDA-GRDA ratio, GPD, PBV, suppression asymmetry, seizures). Adding PTE mechanism (penetrating) improves performance (Delong p=0.0575) with an AUC of 0.775 (95%CI 0.687-0.863), and an accuracy (also sensitivity, specificity) of 75%.

Our data suggest that combining machine-detected EAs and qEEG spectral features identifies patients at risk for PTE. Our results highlight the potential benefit of early EEG assessments in TBI patients.