Introduction: Post-traumatic epilepsy (PTE) affects as many as one-third of severe traumatic brain injury (TBI) survivors, impacting long-term outcomes. Identifying early EEG biomarkers that portend PTE may improve risk stratification and lead to effective therapy. Here, we sought to evaluate the predictive performance of paroxysmal slow-wave events (PSWEs) in determining the risk of PTE in patients with severe TBI.
Methods: This case-control study included 45 severe TBI patients (17 with PTE and 28 without PTE), matched for age and Glasgow Coma Scale (GCS) score at admission. EEG data from the first 20 minutes of recordings obtained during acute care admission were preprocessed and analyzed for PSWEs detection. Descriptive statistics were used for exploratory analysis. Logistic regression and leave-one-out cross-validation (LOOCV) techniques were used to assess predictive performance.
Results: Patients with PTE had significantly longer time in PSWEs on initial EEG recordings (p = 0.040), an increase in the time PSWEs were detected between initial and follow-up EEGs (p = 0.035), and a lower median power frequency (MPF) of PSWEs on the initial EEG (p = 0.020). Lower MPF of PSWEs was associated with increased PTE risk (OR 5.88, 95% CI, 1.08-32.01; p = 0.041). Multivariate logistic regression identified time in PSWEs, MPF of PSWEs in the initial EEG, and decompressive hemicraniectomy (DHC) as significant predictors of PTE (AUC of 0.866, 95% CI: 0.794-0.982, p < 0.0001). This model maintained strong predictive performance under LOOCV (AUC of 0.834, 95% CI: 0.682-0.964, p < 0.0001, accuracy 80%).
Conclusion : Paroxysmal slow-wave events on EEG predict the development of PTE following severe TBI. The model’s robust performance suggests good generalizability and resistance to overfitting.
.jpg)
