Prognostic scores in pediatric gunshot-induced traumatic brain injury: comparative analysis

The aim of this study was to compare the prognostic performance of the Baylor, St. Louis, and Surviving Penetrating Injury to the Brain (SPIN) scoring systems for mortality and neurosurgical intervention in children with gunshot-induced traumatic brain injury (GTBI), clarifying how their differing variable compositions influence accuracy and clinical applicability in pediatric GTBI.

This was a retrospective cohort study of pediatric patients with firearm-related intracranial injuries presenting to a tertiary pediatric trauma center from January 2014 to April 2023. Clinical, laboratory, and neuroimaging variables were abstracted to calculate Baylor, St. Louis, and SPIN scores. Primary outcomes were mortality and neurosurgical intervention (external ventricular drain [EVD] placement, intracranial pressure [ICP] monitor placement, craniotomy, decompressive hemicraniectomy [DHC]). Associations were assessed with univariate and multivariate logistic regressions (α = 0.05). Discrimination was evaluated with receiver operating characteristic (ROC) curves and the area under the ROC curve (AUC). Optimal score cut points were derived for clinical utility, and confusion matrix metrics summarized performance. Clinical utility was assessed quantitatively by statistical performance and qualitatively by considering input availability (e.g., CT dependence), motivating a pragmatic, two-step workflow.

Eighty-two children (mean age 8.59 ± 4.73 years) met inclusion criteria; mortality occurred in 25.6%, and 54.9% underwent a neurosurgical procedure. All 3 scores were associated with mortality in univariate analyses. In multivariate modeling with age and the scores, the St. Louis score remained the strongest mortality predictor (adjusted OR 1.33, p = 0.013) and showed the highest discrimination (AUC 0.85) with a clinical threshold of 6.5. The SPIN score showed good discrimination (AUC 0.81) with a threshold of 31 and outperformed the Baylor score despite lacking imaging inputs; the Baylor score had lower overall discrimination. For neurosurgical intervention, only EVD placement showed significant univariate associations (St. Louis, SPIN, and age). In multivariate models, St. Louis and SPIN did not retain significance for EVD, but both demonstrated modest discrimination (SPIN AUC 0.671; St. Louis AUC 0.652) and high sensitivity/negative predictive value, supporting use as rule-out tools. No significant associations were observed for ICP monitor placement, craniotomy, or DHC.

The authors' findings support a pragmatic, two-step workflow: use the SPIN scale at first contact, before neuroimaging, to inform triage, then apply the St. Louis scale after imaging to refine prognosis. Predefined cut points for mortality (SPIN score ≤ 31; St. Louis score ≥ 7) and EVD placement (SPIN score ≤ 41; St. Louis score ≥ 2) can standardize bedside decision-making. Their complementary nature suggests their use at different points of patient management to optimize their clinical utility.