Interrater Reliability of NIH TBI Imaging Common Data Elements for Brain MRI in Mild TBI

The National Institutes of Health-National Institute of Neurological Disorders and Stroke (NIH-NINDS) Traumatic Brain Injury (TBI) Imaging Common Data Elements (CDE) are standardized definitions for pathological intracranial lesions based on their appearance on neuroimaging studies. The NIH-NINDS TBI Imaging CDEs were designed to be as consistent as possible with the FDA definition of biomarkers as "an indicator of normal biological processes, pathogenic processes, or biological responses to an exposure or intervention." However, the FDA qualification process for biomarkers requires proof of reliable biomarker test measurements. We determined the interrater reliability of TBI Imaging CDEs on subacute brain MRI performed on 517 mild TBI patients presenting to 11 U.S. Level 1 trauma centers. Three U.S. board-certified neuroradiologists independently evaluated brain MRI performed 2 weeks postinjury for the following CDEs: traumatic axonal injury (TAI), diffuse axonal injury (DAI), and brain contusion. We found very high interrater agreement for brain contusion, with prevalence- and bias-adjusted kappa (PABAK) values for pairs of readers from 0.92 [95% CI 0.88-0.95] to 0.94 [0.90-0.96]. We found intermediate agreement for traumatic and diffuse axonal injury (TAI and DAI), with PABAK values of 0.74-0.78 [0.70-0.82]. The near-perfect agreement for brain contusion is likely due to the high conspicuity and distinctive appearance of these lesions on T1-weighted images. Interrater agreement for TAI and DAI was lower, because signal void in small vascular structures, and artifactual foci of signal void, can be difficult to distinguish from the punctate round or linear areas of slight hemorrhage that are a common hallmark of TAI/DAI on MRI.