Reliability and reproducibility of dens fracture classification with use of plain radiography and reformatted...

The classification system of dens fractures by Anderson and D'Alonzo has been widely used in clinical studies. Of the three types of fractures, Type II and Type III are of particular importance because the distinction between them may affect treatment decisions. The purposes of this study were to assess whether this classification is reliable and reproducible and to determine whether computed tomography can improve its reliability and reproducibility.

Plain radiographs and spiral computed tomography images of dens fractures in eleven patients were assessed, and the fractures were assigned a classification of Type II or Type III at two readings, separated by six months, by two spine surgeons and three neuroradiologists. Kappa coefficients of agreement between the raters as well as the reproducibility of the classifications made by the individual raters were calculated independently for the fracture classifications based on the plain radiographs and those based on the reformatted computed tomography scans.

The kappa coefficient for classifications based on plain radiographs was 0.30 and 0.25 (fair agreement) at the first and second readings, respectively. For classifications based on computed tomography scans, the corresponding kappa coefficients were 0.46 (moderate agreement) and 0.67 (substantial agreement). The kappa coefficients for intrarater reliability among the five raters averaged 0.56 (moderate agreement) when computed tomography scans were used and 0.28 (fair agreement) when plain radiographs were used.

Substantial variation with regard to the classification of dens fractures was found within our group of raters. Greater agreement occurred when reformatted computed tomography scans rather than plain radiographs were used as the basis for classification. When classifying dens fractures according to the system of Anderson and D'Alonzo, one should consider using reformatted computed tomography scans and reaching a consensus with multiple raters.