Classification of tibial plateau fractures using 3DCT with and without subtraction of unfractured bones

Background: The addition of 3D CT scans for observers evaluating tibial plateau fractures decreases agreement and diagnostic accuracy, possibly due to obstruction of the view by unfractured bones (i.e., patella, fibula, and femur) in 3D reconstructions. Questions/purposes: This study compared (1) the interobserver agreement between (subgroups of) observers classifying 3DCT reconstructions of tibial plateau fractures according to Schatzker, Luo, and AO/OTA, (2) recognition of different fracture characteristics, and (3) the level of confidence on a scale from 0 to 10 with and without subtraction of the unfractured bones. Methods: We selected twenty consecutive 3DCT reconstructions of tibial plateau fractures treated in a level I trauma center between 2013 and 2014. Members of the Science of Variation Group were randomized (1:1) to evaluate reconstructions with or without subtraction of the unfractured bone. Forty-nine surgeons completed the survey. The interobserver agreement was determined using Fleiss’ kappa, and bootstrapping (resamples = 1000) was used to calculate the standard error, z statistic, and confidence intervals. Results: Overall, subtraction of unfractured bones did not affect interobserver agreement in Schatzker, Luo, and AO/OTA classification systems, agreement on the presence of fracture characteristics, and level of confidence. In subgroup analysis, non-trauma surgeons had significantly higher agreement in the three classification systems with subtraction. For fracture characteristics, some subgroups had higher agreement while others had lower agreement with subtraction. Discussion: The finding that agreement on classification of tibial plateau fractures is limited on 3DCT scan and is not improved by subtraction of unfractured bones suggests that disagreement may be based on variations in understanding and bias. Reliability of these classification systems might benefit from a focus on accounting for these pre-existing conceptualizations rather than further emphasis on imaging detail. Level of Evidence: Diagnostic, level 1.