Early Prognostication in Pediatric Severe Traumatic Brain Injury in South America: Development of a Local Pediatric-Specific Model and Validation of Established Models.

Prognostication in severe traumatic brain injury (sTBI) is important, but few models are pediatric-specific and from low- and middle-income countries where head computed tomography (CT) scans may not be routinely available. We assessed intensive care unit admission risk factors for early mortality and unfavorable outcome in a secondary analysis of 115 children (mean 7.0 years, standard deviation [sd] 5.3) receiving sTBI (Glasgow Coma Scale [GCS] total score ≤8 or GCS motor ≤5) care in South America who participated in the 16 hospital Pediatric Guideline Adherence and Outcomes (PEGASUS) Argentina trial between September 1, 2019, and July 13, 2020. Outcomes were 14-day mortality and 3-month Glasgow Outcome Scale-Extended for Pediatrics (GOS-E Peds). First, we examined univariate associations of predictors with the two outcomes. Then, two PEGASUS logistic regression models (core model with only clinical variables and full model with both clinical and CT variables) for each of the outcomes were derived. Models were examined for fit and compared for prediction. The locally derived PEGASUS model shows a good core prediction of 14-day (area under the receiver operating characteristic curve [AUROC]: 0.92; confidence interval [CI]: 0.85-0.99) and 3-month (AUROC 0.82 CI 0.73-0.91) outcomes; findings are similar to the International Mission on Prognosis and Analysis of Randomized Controlled Trials in TBI (IMPACT), Corticosteroid Randomization after Significant Head Injury (CRASH), and Petroni models. There was no difference between core and full models in prognosticating 14-day mortality, but IMPACT ( = 0.01) and PEGASUS ( = 0.01) full models outperformed their respective core models for 3-month GOS-E Peds. Core models, including PEGASUS, can be used but full models are preferred to prognosticate outcomes after pediatric sTBI in South America. PEGASUS model validation against external datasets is needed.