Central Curation of Glasgow Outcome Scale-Extended Data: Lessons Learned from TRACK-TBI
Kim Boase, Joan Machamer, Nancy Temkin, Sureyya Dikmen, Lindsay Wilson, Lindsay D. Nelson, Jason Barber, Yelena G. Bodien, Joseph T. Giacino, Amy J. Markowitz, Michael McCrea, Gabriela Satris, Murray B. Stein, Sabrina R. Taylor, Geoffrey T. Manley, and the TRACK-TBI Investigators, Opeolu Adeoye, M. Ross Bullock, John D. Corrigan, Ramon Diaz‐Arrastia, Richard Ellenbogen, V. Ramana Feeser, Adam R. Ferguson, Raquel C. Gardner, Dana P. Goldman, Shankar Gopinath, J. Claude Hemphill, C. Dirk Keene, Frederick K. Korley, Joel H. Kramer, Natalie Kreitzer, Harvey S. Levin, Christopher J. Lindsell, Christopher J. Madden, Alastair J. Martin, Thomas W. McAllister, Randall E. Merchant, Pratik Mukherjee, Laura B. Ngwenya, Florence Noël, Amber Nolan, David O. Okonkwo, Eva Palacios, Daniel P. Perl, Ava M. Puccio, Miri Rabinowitz, Claudia Robertson, Jonathan Rosand, Angelle M. Sander, David M. Schnyer, Seth A. Seabury, Mark Sherer, Arthur W. Toga, Alex B. Valadka, Mary J. Vassar, RN MS, Paul Vespa, Kevin Wang, John K. Yue, Esther L. Yuh, Ross Zafonte
Abstract
The Glasgow Outcome Scale (GOS) in its original or extended (GOSE) form is the most widely used assessment of global disability in traumatic brain injury (TBI) research. Several publications have reported concerns about assessor scoring inconsistencies, but without documentation of contributing factors. We reviewed 6801 GOSE assessments collected longitudinally, across 18 sites in the 5-year, observational T ransforming R esearch and C linical K nowledge in TBI (TRACK-TBI) study. We recorded error rates (i.e., corrections to a section or an overall rating) based on site assessor documentation and categorized scoring issues, which then informed further training. In cohort 1 ( n = 1261; February 2014 to May 2016), 24% of GOSEs had errors identified by central review. In cohort 2 ( n = 1130; June 2016 to July 2018), acquired after curation of cohort 1 data, feedback, and further training of site assessors, the error rate was reduced to 10%. GOSE sections associated with the most frequent interpretation and scoring difficulties included whether current functioning represented a change from pre-injury (466 corrected ratings in cohort 1; 62 in cohort 2), defining dependency in the home and community (163 corrections in cohort 1; three in cohort 2) and return to work/school (72 corrections in cohort 1; 35 in cohort 2). These results highlight the importance of central review in improving consistency across sites and over time. Establishing clear scoring criteria, coupled with ongoing guidance and feedback to data collectors, is essential to avoid scoring errors and resultant misclassification, which carry potential to result in “failure” of clinical trials that rely on the GOSE as their primary outcome measure.