Font Size: a A A

Classification And Regression Tree (CART) For Prediction Of Outcome After Severe Head Injury

Posted on:2008-07-10Degree:MasterType:Thesis
Country:ChinaCandidate:J L WangFull Text:PDF
GTID:2144360218455963Subject:Neurosurgery
Abstract/Summary:PDF Full Text Request
Objective: (1)To develop a predictive model for outcome after severe headinjury involving variables can be easily and rapidly achieved in daily routinepractice, requiring a high predictive accuracy. (2)To confirm the simplelaboratory variances which can predict the prognosis of severe brain injury. (3)To disting.uish the primary predictor from the secondary predictor in thepredictive model.Methods: A classification and regression tree (CART) technique isemployed in the analysis of data from 331 patients with severe brain injurybetween the predictors and prognosis. A total of 8 prognostic indicators (theinitial GCS at admission, age, glucose level and white blood cell count on the2~nd morning, the pupil reflexes, need craniotomy or not, mostly caused byepidural hematoma or not, accompany by the traumatic subarachnoidhemorrhage or not) are examined to predict neurological outcome. TheGlasgow Outcome Scale (GOS) at 6 months after severe head injury is adoptedas prognostic criterion.Results: Glasgow Coma Scale (GCS) is lying in the first step of theclassification and regression tree, and is the best predictor of outcome. Glucoselevel on the 2~nd morning, the head computed tomography manifests that theinjury mostly caused by epidural hematoma or not, and age prove to be strongpredictors, are standing at the second step. White blood cell count on 2~ndmorning is also correlated with the prognosis, while it is at the third step. Theoverall predictive accuracy of CART model for these data was 87.9%. All variables included in this tree have been shown to be related to outcome.Conclusions: (1)This study selects simple and easy to get variables, with ahigh predictive accuracy. The CART has been proved to be useful in developinga simple and effective predictive model for the outcome after severe head injury,with the primary benefit of illustrating the important prognostic variablesrelated to outcome. (2)The initial GCS at admission, glucose level on the 2~ndmorning, age, mostly caused by epidural hematoma or not, white blood cellcount on the 2~nd morning are correlated with the prognosis. (3)GCS is the mostimportant predictor.
Keywords/Search Tags:Severe head injury, Classification and regression tree, Prognosis
PDF Full Text Request
Related items