Prognostic models for predicting posttraumatic seizures during acute hospitalization, and at 1 and 2 years following traumatic brain injury

• Published in: Epilepsia (Copenhagen) 2016-09, Vol.57 (9), p.1503-1514
• Main Authors: Ritter, Anne C., Wagner, Amy K., Szaflarski, Jerzy P., Brooks, Maria M., Zafonte, Ross D., Pugh, Mary Jo V., Fabio, Anthony, Hammond, Flora M., Dreer, Laura E., Bushnik, Tamara, Walker, William C., Brown, Allen W., Johnson‐Greene, Doug, Shea, Timothy, Krellman, Jason W., Rosenthal, Joseph A.
• Format: Article
• Language: English
• Subjects: Adolescent; Adult; Aged; Brain Injuries, Traumatic - complications; Brain Injuries, Traumatic - therapy; Craniectomy; Craniotomy - methods; Epilepsy; Female; Hospitalization; Humans; Imprisonment; International Classification of Diseases; Logistic Models; Longitudinal Studies; Male; Middle Aged; Prognosis; Prognostic modeling; Risk Factors; Seizures - diagnosis; Seizures - etiology; TBI Model System; Time Factors; Tomography Scanners, X-Ray Computed; Traumatic brain injury; Young Adult
• ISSN: 0013-9580; 1528-1167
• DOI: 10.1111/epi.13470

Summary Objective Posttraumatic seizures (PTS) are well‐recognized acute and chronic complications of traumatic brain injury (TBI). Risk factors have been identified, but considerable variability in who develops PTS remains. Existing PTS prognostic models are not widely adopted for clinical use and do not reflect current trends in injury, diagnosis, or care. We aimed to develop and internally validate preliminary prognostic regression models to predict PTS during acute care hospitalization, and at year 1 and year 2 postinjury. Methods Prognostic models predicting PTS during acute care hospitalization and year 1 and year 2 post‐injury were developed using a recent (2011–2014) cohort from the TBI Model Systems National Database. Potential PTS predictors were selected based on previous literature and biologic plausibility. Bivariable logistic regression identified variables with a p‐value < 0.20 that were used to fit initial prognostic models. Multivariable logistic regression modeling with backward‐stepwise elimination was used to determine reduced prognostic models and to internally validate using 1,000 bootstrap samples. Fit statistics were calculated, correcting for overfitting (optimism). Results The prognostic models identified sex, craniotomy, contusion load, and pre‐injury limitation in learning/remembering/concentrating as significant PTS predictors during acute hospitalization. Significant predictors of PTS at year 1 were subdural hematoma (SDH), contusion load, craniotomy, craniectomy, seizure during acute hospitalization, duration of posttraumatic amnesia, preinjury mental health treatment/psychiatric hospitalization, and preinjury incarceration. Year 2 significant predictors were similar to those of year 1: SDH, intraparenchymal fragment, craniotomy, craniectomy, seizure during acute hospitalization, and preinjury incarceration. Corrected concordance (C) statistics were 0.599, 0.747, and 0.716 for acute hospitalization, year 1, and year 2 models, respectively. Significance The prognostic model for PTS during acute hospitalization did not discriminate well. Year 1 and year 2 models showed fair to good predictive validity for PTS. Cranial surgery, although medically necessary, requires ongoing research regarding potential benefits of increased monitoring for signs of epileptogenesis, PTS prophylaxis, and/or rehabilitation/social support. Future studies should externally validate models and determine clinical utility.