Prediction of outcome in traumatic brain injury with computed tomographic characteristics: a comparison between the computed tomographic classification and combinations of computed tomographic predictors

Abstract

Background and objective: The Marshall computed tomographic (CT) classification identifies six groups of patients with traumatic brain injury (TBI), based on morphological abnormalities on the CT scan. This classification is increasingly used as a predictor of outcome. We aimed to examine the predictive value of the Marshall CT classification in comparison with alternative CT models.

Methods: The predictive value was investigated in the Tirilazad trials (n = 2269). Alternative models were developed with logistic regression analysis and recursive partitioning. Six month mortality was used as outcome measure. Internal validity was assessed with bootstrapping techniques and expressed as the area under the receiver operating curve (AUC).

Results: The Marshall CT classification indicated reasonable discrimination (AUC = 0.67), which could be improved by rearranging the underlying individual CT characteristics (AUC = 0.71). Performance could be further increased by adding intraventricular and traumatic subarachnoid hemorrhage and by a more detailed differentiation of mass lesions and basal cisterns (AUC = 0.77). Models developed with logistic regression analysis and recursive partitioning showed similar performance. For clinical application we propose a simple CT score, which permits a more clear differentiation of prognostic risk, particularly in patients with mass lesions.

Conclusion: It is preferable to use combinations of individual CT predictors rather than the Marshall CT classification for prognostic purposes in TBI. Such models should include at least the following parameters: status of basal cisterns, shift, traumatic subarachnoid or intraventricular hemorrhage, and presence of different types of mass lesions.