Conscious status predicts mortality among patients with isolated traumatic brain injury (TBI) in administrative data

Abstract Background Outcome studies in trauma using administrative data traditionally employ anatomy-based definitions of injury severity; however, physiologic factors, including consciousness, may correlate with outcomes. We examined whether accounting for conscious status in administrative data improved mortality prediction among patients with moderate/ severe TBI. Methods Patients meeting CDC guidelines for TBI in the 2006-2011 Nationwide Emergency Department Sample (NEDS) were identified. Patients were dichotomized as having no/brief loss of consciousness (LOC) vs extended LOC > 1 hour using ICD-9 fifth-digit modifiers. Receiver Operating Curves (ROC) compared the ability of logistic regression to predict mortality in models that included LOC vs models that did not. Results Overall, 98,397 individuals met criteria, of whom 25.8% had extended LOC. In univariate analysis, AIS alone predicted mortality in 69.6% of patients (AUROC 0.696, 95% CI 0.689-0.702), extended LOC alone predicted mortality in 76.8% (AUROC 0.768, 95% CI 0.764-0.773) and a combination of AIS and extended LOC predicted mortality in 82.6% of cases (AUROC 0.826 95% CI 0.821-0.830). Similar differences were observed in best-fit models. Conclusions Accounting for LOC along with anatomical measures of injury severity improves mortality prediction among patients with moderate/severe TBI in administrative datasets. Further work is warranted to determine whether other physiological measures may also improve prediction across a variety of injury types.