Early seizures are common among infants and toddlers hospitalized with mild-moderate traumatic brain injury (TBI) and are associated with unfavorable outcomes. We aimed to develop a risk prediction model for early seizures in this population to support clinical care. Patients <3 years old admitted to our pediatric intensive care unit after mild to moderate TBI (initial Glasgow Coma Scale score 9–15) between 2011 and 2024 were included. The outcome was early electrographic or clinical seizures within the first 3 days of admission. Variable selection was performed using Lasso regression. Variables selected by Lasso were subsequently included in a multivariable logistic regression model, trained using repeated fivefold cross-validation. Model discrimination was assessed using area under the receiver operating characteristic curve (AUROC) and area under the precision-recall curve (AUPRC). Calibration was evaluated with Hosmer–Lemeshow tests and reliability plots. Thresholds for continuous variables were chosen using locally estimated scatterplot smoothing. Coefficients were rounded to the nearest integer to create a point system to facilitate bedside model risk calculation. All values are reported with 95% confidence intervals in parentheses. Of 796 patients who met inclusion criteria, 52 (6%) had early seizures, electroencephalography (EEG) was obtained in 180 (23%), and antiseizure medication prophylaxis was administered to 299 (38%). A 9 variable, point-based prediction model was developed that included (points): age <12 months (+2), presentation concerning for seizure (+2), bruising on physical exam (+0.5), subdural hemorrhage (+1.5), intraparenchymal hemorrhage (+1.5), initial hemoglobin ≤ 7.5 g/dL (+1.5), initial prothrombin time ≥ 15 sec (+1.5), sodium ≥ 145 mmol/L (+2), and glucose ≥ 200 mg/dL (+0.5). AUROC was 0.89 (0.84, 0.95), and AUPRC was 0.46 (0.32, 0.59). The goodness-of-fit test showed acceptable calibration ( p = 0.22), as did visualization of a reliability plot. Point thresholds of 1 and 3 had sensitivities of 0.96 (0.87, 0.99) and 0.65 (0.51, 0.78) with specificities of 0.44 (0.40, 0.47) and 0.91 (0.89, 0.93), respectively. This predictive model using nine variables in pediatric patients less than 3 years old with mild to moderate TBI could help identify patients at high risk for early seizures and who should receive seizure prophylaxis or EEG monitoring. This model merits additional validation and refinement.
McNamara et al. (Sun,) studied this question.