Interobserver reliability of a CT-based fracture classification system

This study was designed to determine whether the interobserver reliability of a fracture classification scheme applied based on a single, carefully defined, computed tomography (CT) cut is greater than those previously reported for systems designed for use with plain radiographs.

Observer review of selected cases.

Four, level one, trauma centers.

Pretreatment CT scans of patients with calcaneus fractures were screened by the authors. Thirty cases were selected that had an appropriate semicoronal CT image. Ten orthopaedic traumatologists who were members of the Orthopaedic Trauma Association and had a minimum of 5 years postresidency experience were selected as reviewers.

The reviewers were provided with a digital CT image for each case as well as written and diagrammatic representations of the Sanders classification system. The observers then classified each fracture according to the Sanders classification.

: The mean kappa value for interobserver reliability for fracture types I-IV was 0.41 +/- 0.02 (mean +/- standard error of the mean; range, 0.07-0.64). Observers disagreed by more than 1 fracture type (ie, I vs. III or II vs. IV) in 10% of the cases. Observers agreed on the location of the fracture lines (A, B, C) in 90% of type II fractures and 52% of type III fractures.

The results indicate that in a carefully controlled paradigm, the interobserver reliability with a classification system based on interpretation of a single, carefully defined CT image was no better than the results reported for the same classification system used with full CT data or for other classification systems used for various fractures in the skeleton. Agreement in identifying the location of the fracture lines was very good for simple fractures but much worse for complex injuries. Additional study may determine whether the use of a full complement of CT images can improve reliability in classification of complex injuries.