Classification and Regression Tree for Prediction of Outcome after Severe Head Injury Using Simple Clinical and Laboratory Variables

Publisher Website
Google Scholar
Abstract
Many previous studies have constructed several predictive models for outcome after severe head injury, but these have often used expensive, time consuming, or highly specialized measurements. The goal of this study was to develop a simple, easy to use a model involving only variables that are rapidly and easily achievable in daily routine practice. To this end, a classification and regression tree (CART) technique was employed in the analysis of data from 345 patients with isolated severe brain injury who were admitted to Asclepeion General Hospital of Athens from January, 1993, to December, 2000. A total of 16 prognostic indicators were examined to predict neurological outcome at 6 months after head injury. Our results indicated that Glasgow Coma Scale was the best predictor of outcome. With regard to the other data, not only the most widely examined variables such as age, pupillary reactivity, or computed tomographic findings proved again to be strong predictors, but less commonly applied parameters, indirectly associated with brain damage, such as hyperglycemia and leukocytosis, were found to correlate significantly with prognosis too. The overall cross-validated predictive accuracy of CART model for these data was 86.84%, with a cross-validated relative error of 0.308. All variables included in this tree have been shown previously to be related to outcome. Methodologically, however, CART is quite different from the more commonly used statistical methods, with the primary benefit of illustrating the important prognostic variables as related to outcome. This technique may prove useful in developing new therapeutic strategies and approaches for patients with severe brain injury.