Purpose: Classifications of fractures of the thoracolumbar spine are theoretically designed to help make therapeutic decisions. Three classifications (J. Laulan, F. Denis, F. Magerl) were compared to assess reproducibility for use by a surgery team.

Material and methods: The classifications were described during a SOFCOT symposium in 1995. Four observers examined 60 files reading them twice at a 1 month interval. The files included plain radiographs (AP and lateral view) and a scanner series and were read in random order. Intra- and interobserver concordance were measured with the kappa method.

Results: Intra- and interobserver reproducibility was good for the classification proposed by F. Denis (kappa = 0.6229 and 0.0795) for classification groups but was weak for subgroups (kappa = 0.028 and 0.571). Reproducibility was moderate for the classification proposed by J. Laulin (interob-server kappa = 0.460, intraobserver kappa = 0.541). The Magerl classification produced low to negligible reproducibility for classification groups and subgroups (intra- and interobserver kappa = 0.138 to 0.0343).

Discussion: Because of its low to negligible reproducibility, the Magerl classification would be difficult to use in clinical practice to make coherent therapeutic decisions or for scientific research to analyze series of fractures treated using this classification. The reproducibility of the F. Denis classification was good for groups but low for subgroups that include fractures resulting from different mechanisms requiring radically different treatment strategies. This is a good classification system for descriptive work but can lead to treatments poorly adapted to the causal mechanism of the fracture. The reproducibility of the J. Laulan classification is moderate but each group in this classification corresponds to fractures caused by the same mechanism. Therapeutic indications determined with this system would be more coherent.