Predicting Neurological Recovery after Traumatic Brain Injury in Children: A Systematic Review of Prognostic Models

Abstract

Predictive modeling is foundational to treatment and long-term management of children with traumatic brain injury (TBI). Assessment of injury severity in the acute-care setting enables early stratification of patients based on their risk of death, lifelong disability, or unfavorable outcome. This review evaluates predictive models that have been developed or validated for pediatric TBI patients. The predictive accuracy of these models, the outcomes and time points predicted, and the variables and statistical methods utilized in model development were compared. Embase, Scopus, MEDLINE^®, and Web of Science were searched for studies that developed statistical models for predicting patient outcomes following pediatric TBI. Studies were excluded if they focused on adults or non-traumatic brain injury, or if they did not assess classification accuracy. A total of 4538 entries were identified and screened, with 7 studies included for analysis. This included five studies in which adult predictive models were validated for use in the pediatric setting, and two in which new models were derived from a pediatric cohort. Trials of adult prediction tools in pediatric cohorts, including the International Mission for Prognosis and Analysis of Clinical Trials in TBI (IMPACT) and Corticoid Randomisation After Significant Head Injury (CRASH)-TBI models, showed comparable accuracy between classification of adults and children. Models derived from pediatric cohorts showed improved accuracy. Most studies solely focused on clinical variables, with two studies incorporating biochemical and imaging variables. Predictive models for pediatric TBI are primarily based on methods and variables identified in adult studies. Although adult models have proven effective in select pediatric cohorts, they may be suboptimal when compared with models derived or adjusted for children.