Predictive ability and interobserver reliability of computed tomography findings for angioembolization in patients with pelvic fracture.

Background: The purpose of this study was to identify computed tomography (CT) findings that predict the need for angioembolization in patients with pelvic fracture.

Methods: This retrospective cross-sectional study was performed between April 2006 and October 2015 at two urban emergency medical centers in Japan. The study included patients who underwent CT within 3 hours of arrival and were diagnosed with a pelvic fracture. The study outcome was undergoing angioembolization within 24 hours of arrival. Four independent readers blinded to all clinical information interpreted the CT scans for blush, thickness of retroperitoneal hematoma, and diameter and laterality of muscle swelling around the pelvis. Receiver operating characteristics analyses provided area under the receiver-operating characteristic curves (AUROC) and diagnostic indicators at the estimated thresholds including sensitivity, specificity, positive and negative predictive values and diagnostic odds ratio (DOR). The interobserver reliability of all radiographic findings was also evaluated.

Results: Fifty-two of 244 eligible patients underwent angioembolization. The predictive ability in terms of DOR was relatively better with blush on CT scan (sensitivity, 0.57; specificity, 0.86; DOR, 8.05) than with laterality of muscle swelling of 12.9 mm or greater (sensitivity, 0.79; specificity, 0.55; DOR, 4.60; AUROC, 0.75) and thickness of retroperitoneal hematoma of 22.7 mm or greater (sensitivity, 0.65; specificity, 0.74; DOR, 5.39; AUROC, 0.73). The interobserver reliability of blush, laterality of muscle swelling, and thickness of retroperitoneal hematoma was 0.43, 0.54, and 0.70, respectively.

Conclusion: All of the tested CT findings failed to show both sufficient predictive ability and sufficient interobserver agreement. Further diagnostic accuracy studies to validate these findings or establish a prediction model incorporating these findings are expected.

Level Of Evidence: Diagnostic, level V.