Citation

Greenan K, Taylor SL, Fulkerson D, Shahlaie K, Gerndt CH, Krueger EM, Zwienenberg M. J. Neurosurg. Pediatr. 2019; ePub(ePub): 1-10.

Affiliation

Department of Neurological Surgery, University of California, Davis.

Copyright

(Copyright © 2019, American Association of Neurological Surgeons)

DOI

10.3171/2019.1.PEDS18293

PMID

30952132

Abstract

OBJECTIVE: A recent retrospective study of severe traumatic brain injury (TBI) in pediatric patients showed similar outcomes in those with a Glasgow Coma Scale (GCS) score of 3 and those with a score of 4 and reported a favorable long-term outcome in 11.9% of patients. Using decision tree analysis, authors of that study provided criteria to identify patients with a potentially favorable outcome. The authors of the present study sought to validate the previously described decision tree and further inform understanding of the outcomes of children with a GCS score 3 or 4 by using data from multiple institutions and machine learning methods to identify important predictors of outcome.

METHODS: Clinical, radiographic, and outcome data on pediatric TBI patients (age < 18 years) were prospectively collected as part of an institutional TBI registry. Patients with a GCS score of 3 or 4 were selected, and the previously published prediction model was evaluated using this data set. Next, a combined data set that included data from two institutions was used to create a new, more statistically robust model using binomial recursive partitioning to create a decision tree.

RESULTS: Forty-five patients from the institutional TBI registry were included in the present study, as were 67 patients from the previously published data set, for a total of 112 patients in the combined analysis. The previously published prediction model for survival was externally validated and performed only modestly (AUC 0.68, 95% CI 0.47, 0.89). In the combined data set, pupillary response and age were the only predictors retained in the decision tree. Ninety-six percent of patients with bilaterally nonreactive pupils had a poor outcome. If the pupillary response was normal in at least one eye, the outcome subsequently depended on age: 72% of children between 5 months and 6 years old had a favorable outcome, whereas 100% of children younger than 5 months old and 77% of those older than 6 years had poor outcomes. The overall accuracy of the combined prediction model was 90.2% with a sensitivity of 68.4% and specificity of 93.6%.

CONCLUSIONS: A previously published survival model for severe TBI in children with a low GCS score was externally validated. With a larger data set, however, a simplified and more robust model was developed, and the variables most predictive of outcome were age and pupillary response. ABBREVIATIONS: AUC = area under the curve; ER = Emergency Room; GCS = Glasgow Coma Scale; GOS = Glasgow Outcome Scale; GOS-E = Extended GOS; ICP = intracranial pressure; MVA = motor vehicle accident; NAT = nonaccidental trauma; RH = Riley Hospital; TBI = traumatic brain injury; tSAH = traumatic subarachnoid hemorrhage; UCD = UC Davis.

Language: en

Keywords

AUC = area under the curve; ER = Emergency Room; GCS = Glasgow Coma Scale; GOS = Glasgow Outcome Scale; GOS-E = Extended GOS; ICP = intracranial pressure; MVA = motor vehicle accident; NAT = nonaccidental trauma; RH = Riley Hospital; TBI = traumatic brain injury; UCD = UC Davis; outcome prediction; pediatric; severe TBI; tSAH = traumatic subarachnoid hemorrhage; trauma