Predictive factors for postoperative deformity in thoracolumbar burst fractures: a statistical approach.

OBJECTIVE To assess, using statistical analysis, if and to what extent the final outcome of surgical treatment for burst fractures depends on operation type, fracture level and initial deformity severity. METHODS A database of 287 patients with single-vertebral-level thoracic and lumbar spine fractures analysed using simple and multiple linear regression analyses models. The dependent variable was last follow-up (LFU) kyphotic angle and the predictor variables were operation type [anterior approach (AA), posterior short-segment fixation (PSSF) and posterior monosegmental fixation (PMF)], fracture level (T11-L1, L2-L3 and L4-L5) and preoperative kyphotic angle. The models were applied on either the whole sample or on the operation type subgroups. RESULTS In simple linear regression analysis models, fracture level accounted for 32% and 18% of the variation in LFU kyphotic angle in the AA and PMF subgroups, respectively. In the multiple linear regression models for the same subgroups, up to 40% of the variation in LFU kyphotic angle was accounted for by fracture level. Surgical treatment, as predictor variable, indicated that patients treated by PSSF developed a post-surgical kyphotic angle 8.51° more severe than those treated by AA. However, the model accounted for only 2% of the variation in LFU kyphotic angle. Simple linear regressions performed on each subgroup with preoperative kyphotic angle as the independent variable revealed that the variable accounted for 15% (PSSF subgroup), 17% (AA subgroup) and 34% (PMF subgroup) of the variation in LFU kyphotic angle. CONCLUSION All valid regression models displayed modest explanatory power, suggesting that factors other than those taken into consideration are involved.