Classification and regression trees (CART) for prediction of function at 1 year following head trauma

Publisher Website
Google Scholar
Abstract
A cohort of 514 hospitalized head-injury survivors was identified based on their injury and 448 (87%) of them were followed for 1 year. Comprehensive neurobehavioral testing was performed 1 month and 1 year after injury. The authors developed predictions of six neuropsychological and two psychosocial outcomes 1 year after head injury. Prediction trees are presented for verbal IQ, Halstead's Impairment Index, and work status at 1 year. Early predictors of neurobehavioral outcome in survivors are similar to previously reported predictors of mortality. Extent (both depth and length) of coma and age are the medical and demographic variables most predictive of late outcome. Adding 1-month scores substantially improves prediction of neuropsychological variables. The classification and regression tree is a useful technique for predicting long-term outcome in patients with head injury. The trees are simple enough to be used in a clinical setting and, especially with 1-month scores, predictions are accurate enough for clinical utility.