Accuracy of computed tomography in assessing thoracic pedicle screw malposition

Transpedicular instrumentation of the thoracic spine is potentially dangerous because of the close proximity of vital structures and the morphologic variability seen. Computed tomography has become the gold standard in evaluating the position of thoracic pedicle screws in vivo. Unfortunately, despite its common use, the accuracy of computed tomography has not been adequately investigated. The objective of this work was to evaluate the accuracy of computed tomography in evaluating the position of thoracic pedicle screws. One hundred ninety-four thoracic pedicles in nine cadaveric specimens were instrumented and evaluated postoperatively with computed tomography and open dissection. Computed tomography films were assessed by three blinded observers who noted the position of each pedicle screw on two separate occasions. These data were subsequently compared with the open dissection data. Computed tomography was found to be 76 +/- 16% sensitive and 75 +/- 13% specific when compared with open dissection. Overall accuracy was 76 +/- 8%. Intraobserver accuracy was 79 +/- 5% in assessing thoracic pedicle screws by computed tomography. Fair to moderate degree of agreement was demonstrated for both interobserver and intraobserver data using kappa values. Computed tomography was found to be relatively insensitive in assessing thoracic pedicle screw position. In the face of postoperative complications, surgical exploration and hardware removal may still be necessary despite negative computed tomography.