Abstract
To the Editor:
We have read with interest the article by Chen and colleagues. 1 In their retrospective study, they attempted to develop and validate a predictive model of neurologic deterioration (ND) in individuals with moderate traumatic brain injury (mTBI). Overall, 28% of the population suffered ND; median time from presentation to ND was 39 h and 80% of patients who deteriorated did so within 72 h of admission. Causes of ND are not presented. The proposed prediction model is composed of eight prognostic factors derived from their multi-variate analysis, including: 1) clinical parameters (history of arterial hypertension and scores on the injury severity scales (ISS) and Glasgow Coma Scale (GCS); 2) neuroimaging: location (subdural, fronto-temporal contusions) and lesion type (Marshall scale); and 3) laboratory: D-dimer and platelet count. A figure included in Chen and colleagues' article shows the individual discrimination power of the components of the model; while the graphic is difficult to interpret, it is clear that the variables with the greatest predictive power were the Marshall scale and the GCS.
We wholeheartedly agree with the Chen and colleagues article that patients presenting with mTBI offer us with a unique opportunity to prevent or rapidly react to ND to minimize functional repercussions. 2 Key to achieve this goal is the correct risk stratification of these patients. An optimal predictive model should be accurate and simple. We have recently developed and proposed a simple model based on traffic lights (Figure 1), for which only the determination of GCS and the Marshall tomographic scale are necessary—both parameters that are universally used and amply validated. 3 The crucial aspect of our proposed risk stratification is to consider patients with GCS 9-11 with abnormal CT scan as a high-risk category. 3
Traffic light method for risk stratification and categorization of potentially severe traumatic brain injury based on Glasgow Coma Scale and Tomographic Marshall's classification. Reproduced with permission of publisher. 3 Color image is available online.
Beyond methodological concerns regarding its development (such as how multi-collinearity was defined or the lack of details about the characteristics of the validation cohort), the normogram proposed by Chen and colleagues may have incorporated more variables than those strictly necessary for adequate triaging, including one that may be unknown at the time of presentation (history of hypertension). It would be pertinent to compare our proposed, much simpler categorization scheme with the predictive normogram developed by Chen and colleagues.
If the differences in prediction performance are not substantial, it would be reasonable to argue that the simpler scheme, even if at the expense of a slightly lower predictive accuracy, would be more useful than a more complex model.