Page 38 - WNS 2022 E-Program Booklet

Prediction of seizure risk following repeated mild traumatic brain injury in
children
Laura M. Prolo, MD, PhD, Department of
Neurosurgery, Stanford University School of Medicine, Stanford CA
Co-authors: Michael C. Jin BS 1 , Adela Wu MD 1 , Cesar A. Garcia BA 1 ,
Adrian J. Rodrigues BA 1 , Sol Savchuk BS 1 , Gabriela D. Ruiz Colón BA 1 ,
Bina W. Kakusa MD 1 , Jonathon J. Parker MD, PhD 1 , Gerald A. Grant MD 2
1 Department of Neurosurgery, Stanford University School of Medicine,
Stanford, CA
2 Department of Neurosurgery, Duke University School of Medicine, Durham, NC
Introduction: While moderate to severe traumatic brain injury is a known risk fac-
tor for seizures, the impact of repeated mild head trauma on long-term seizure
risk is less clear. We developed a modeling tool to determine what clinical char-
acteristics predict long-term seizure risk following mild TBI (mTBI).
Methods: Utilizing a national longitudinal healthcare claims database, a total of
156,118 children (age < 18) diagnosed with mTBI without prior seizures were
identified from 2003 to 2021. Comorbidities linked to epilepsy were collected
during the six months preceding the initial mTBI event. Each mTBI event was
defined as a diagnosis code indicating head injury with ≤1 hour loss-of-
consciousness. Repeated mTBI was defined as a subsequent mTBI diagnosis
code > one month following the prior event. Follow-up was censored at antiepi-
leptic drug initiation, moderate or severe TBI, or loss-to-follow-up. Time-varying
multivariable Cox regression was used to assess impact of repeat mTBI. Three
distinct machine learning approaches were evaluated.
Results: Median duration of follow-up was 22.6 months and median time-to-
seizure was 306 days. Seizures among those with radiographic findings (median
time-to-seizure 112.5 days, IQR 5 to 526.25), loss-of-consciousness (80 days,
IQR 7 to 652 or both (22 days, IQR 5 to 192) occurred earlier. Both mTBI without
and with short loss-of-consciousness resulted in increasing seizure risk with re-
peated events (HR=1.196, 95%CI 1.082-1.322; HR=2.025, 95%CI 1.828- 2.244;
respectively). The survival random forest approach performed best, achieving
fixed-time AUROCs of 0.780 and 0.777 at 30- and 90-days post-mTBI.
Conclusion: Utilizing a machine learning approach, children with elevated seizure
risk following mTBI can be identified. Repeated mTBI accompanied by loss-of-
consciousness doubles the risk for future seizures and a small subset of children
were identified with a nearly 9- fold elevated seizure risk. These results suggest a
subset of children who have multiple mTBI events may benefit from prophylactic
antiepileptic drugs.

38