Reliability of the NINDS common data elements cranial tomography (CT) rating variables for traumatic brain injury (TBI).

Brain injury : [BI]

PubMedID: 27936952

Harburg L, McCormack E, Kenney K, Moore C, Yang K, Vos P, Jacobs B, Madden CJ, Diaz-Arrastia R, Bogoslovsky T. Reliability of the NINDS common data elements cranial tomography (CT) rating variables for traumatic brain injury (TBI). Brain Inj. 2016;1-11.
BACKGROUND
Non-contrast head computer tomography (CT) is widely used to evaluate eligibility of patients after acute traumatic brain injury (TBI) for clinical trials. The NINDS Common Data Elements (CDEs) TBI were developed to standardize collection of CT variables. The objectives of this study were to train research assistants (RAs) to rate CDEs and then to evaluate their performance. The aim was to assess inter-rater reliability (IRR) of CDEs between trained RAs and a neurologist and to evaluate applicability of CDEs in acute and sub-acute TBI to test the feasibility of using CDE CT ratings in future trials and ultimately in clinical practice. The second aim was to confirm that the ratings of CDEs reflect pathophysiological events after TBI.

METHODS AND RESULTS
First, a manual was developed for application of the CDEs, which was used to rate brain CTs (n = 100). An excellent agreement was found in combined kappas between RAs on admission and on 24-hour follow-up CTs (Iota = 0.803 and 0.787, respectively). Good IRR (kappa > 0.61) was shown for six CDEs on admissions and for seven CDEs on follow-up CTs. Low IRR (kappa < 0.4) was determined for five CDEs on admission and for four CDEs on follow-up CT. Combined IRR of each assistant with the neurologist were good on admission (Iota = 0.613 and 0.787) and excellent on follow-up CT (Iota = 0.906 and 0.977). Second, Principal Component Analysis (PCA) was applied to cluster the rated CDEs (n = 255) and five major components were found that explain 53% of the variance.

CONCLUSIONS
CT CDEs are useful in clinical studies of TBI. Trained RAs can reliably collect variables. PCA identifies CDE clusters with clinical and biologic plausibility.