Different predictors of neurosurgical intervention and mortality in moderate traumatic brain injury: a nationwide cohort study.

Introduction: Moderate traumatic brain injury (TBI) is heterogeneous. Admission and injury-related variables may predict outcomes differently, but their relative importance is unclear. Research question: Do admission and injury variables differ in predicting neurosurgical intervention and 30-day mortality in moderate TBI? Material and methods: In a nationwide retrospective cohort study using the Swedish Trauma Registry (SweTrau; 2018-2023), we included adults with moderate TBI (Glasgow Coma Scale 9-13). The primary outcome was neurosurgical intervention; the secondary outcome was 30-day mortality. For each, two multivariable logistic regression models were fitted: an admission model (demographics, comorbidity, physiology) and an injury model (injury severity and intracranial lesions). Discrimination was assessed using the area under the ROC curve (AUC), compared with DeLong's test, and reassessed within age strata (<65 and ≥ 65 years). Results: Among 1761 patients, 198 (11%) underwent neurosurgical intervention and 299 (17%) died within 30 days. The injury model discriminated intervention better than the admission model (AUC 0.84 vs 0.67), in both younger and older patients. For mortality the admission model discriminated better overall (0.84 vs 0.78), but only in older patients: among those <65, injury severity discriminated better (0.88 vs 0.72), whereas among those ≥65 the models were equivalent (0.73 vs 0.71). Increasing age was associated with higher mortality but lower odds of intervention. Discussion and conclusion: In moderate TBI, the information that identifies who needs surgery is not the information that predicts who will survive, and for survival that information itself changes with age. A single severity measure is therefore insufficient.