AO Spine upper cervical injury classification system: a description and reliability study

Background ContextPrior upper cervical spine injury classification systems have focused on injuries to the craniocervical junction (CCJ), atlas, and dens independently. However, no previous system has classified upper cervical spine injuries using a comprehensive system incorporating all injuries from the occiput to the C2–3 joint.PurposeTo (1) determine the accuracy of experts at correctly classifying upper cervical spine injuries based on the recently proposed AO Spine Upper Cervical Injury Classification System (2) to determine their interobserver reliability and (3) identify the intraobserver reproducibility of the experts.Study Design/SettingInternational Multi-Center Survey.Patient SampleA survey of international spine surgeons on 29 unique upper cervical spine injuries.Outcome MeasuresClassification accuracy, interobserver reliability, intraobserver reproducibility.MethodsThirteen international AO Spine Knowledge Forum Trauma members participated in two live webinar-based classifications of 29 upper cervical spine injuries presented in random order, four weeks apart. Percent agreement with the gold-standard and kappa coefficients (ƙ) were calculated to determine the interobserver reliability and intraobserver reproducibility.ResultsRaters demonstrated 80.8% and 82.7% accuracy with identification of the injury classification (combined location and type) on the first and second assessment, respectively. Injury classification intraobserver reproducibility was excellent (mean, [range] ƙ=0.82 [0.58-1.00]). Excellent interobserver reliability was found for injury location (ƙ = 0.922 and ƙ=0.912) on both assessments, while injury type was substantial (ƙ=0.689 and 0.699) on both assessments. This correlated to a substantial overall interobserver reliability (ƙ=0.729 and 0.732).ConclusioNSEarly phase validation demonstrated classification of upper cervical spine injuries using the AO Spine Upper Cervical Injury Classification System to be accurate, reliable, and reproducible. Greater than 80% accuracy was detected for injury classification. The intraobserver reproducibility was excellent, while the interobserver reliability was substantial.