• Carter McInnis, GarciaMaría José SolanaMJSNeuroscience and Mental Health Research Program, Hospital for Sick Children, Toronto, Ontario, Canada., Elysa Widjaja, Helena Frndova, HuyseJudith VanJVNeuroscience and Mental Health Research Program, Hospital for Sick Children, Toronto, Ontario, Canada., Anne-Marie Guerguerian, Adeoye Oyefiade, Suzanne Laughlin, Charles Raybaud, Elka Miller, Keng Tay, Erin D Bigler, Maureen Dennis, Douglas D Fraser, Craig Campbell, Karen Choong, Sonny Dhanani, Jacques Lacroix, Catherine Farrell, Miriam H Beauchamp, Russell Schachar, James S Hutchison, and Anne L Wheeler.
• Faculty of Health Sciences, Queen's University, Kingston, Ontario, Canada.
• J. Neurotrauma. 2021 Sep 1; 38 (17): 240724182407-2418.

AbstractThe identification of children with traumatic brain injury (TBI) who are at risk of death or poor global neurological functional outcome remains a challenge. Magnetic resonance imaging (MRI) can detect several brain pathologies that are a result of TBI; however, the types and locations of pathology that are the most predictive remain to be determined. Forty-two critically ill children with TBI were recruited prospectively from pediatric intensive care units at five Canadian children's hospitals. Pathologies detected on subacute phase MRIs included cerebral hematoma, herniation, cerebral laceration, cerebral edema, midline shift, and the presence and location of cerebral contusion or diffuse axonal injury (DAI) in 28 regions of interest were assessed. Global functional outcome or death more than 12 months post-injury was assessed using the Pediatric Cerebral Performance Category score. Linear modeling was employed to evaluate the utility of an MRI composite score for predicting long-term global neurological function or death after injury, and nonlinear Random Forest modeling was used to identify which MRI features have the most predictive utility. A linear predictive model of favorable versus unfavorable long-term outcomes was significantly improved when an MRI composite score was added to clinical variables. Nonlinear Random Forest modeling identified five MRI variables as stable predictors of poor outcomes: presence of herniation, DAI in the parietal lobe, DAI in the subcortical white matter, DAI in the posterior corpus callosum, and cerebral contusion in the anterior temporal lobe. Clinical MRI has prognostic value to identify children with TBI at risk of long-term unfavorable outcomes.

27 keywords...

hide…