Antibiotic prophylaxis for penetrating brain injury.

Abstract

arrest), high-velocity CT negative TBI, and non-injured controls. Differences in GFAP and UCH-L1 concentrations were assessed using the ttest and Wilcoxon rank-sum test. Support vector machine learning was then utilized for the classification of the patient samples in our prediction tasks. Prediction accuracy was measured by the area under the curve (AUC), precision, recall, and F1 score. RESULTS: 111 matched GFAP and UCH-L1 samples were analyzed; 36 traumatic hemorrhage, 10 spontaneous hemorrhage, 16 oxygen deprivation, 10 high-velocity CT negative TBI, and 39 healthy controls. GFAP concentrations were statistically different (P < .05) in all but one comparison, high-velocity CT negative TBI and oxygen deprivation injury, while UCH-L1 concentrations were only statistically different for comparisons with non-injured control subjects. When GFAP and UCH-L1 concentrations were combined for prediction classification, the AUC for comparisons were as follows; 0.90 spontaneous vs traumatic hemorrhage, 0.93 oxygen deprivation vs spontaneous hemorrhage, 0.84 oxygen deprivation vs traumatic hemorrhage, 0.94 CT negative TBI vs traumatic hemorrhage, 1.00 CT negative TBI vs spontaneous hemorrhage, and 0.96 CT negative TBI vs oxygen deprivation. The classification prediction using both biomarkers for healthy controls and highvelocity CT negative TBI demonstrated an AUC of 0.93, precision 0.9, recall 0.84, and F1 score of 0.87. CONCLUSION: Serum concentrations of S100B and GFAP collected within 32 hours of injury have utility in classifying braininjured subjects based on the etiology of their injuries which has implications for early targeted management and prognostication of brain injury.