Interobserver Variability Limits Exact Preoperative Staging by Computed Tomography in Bladder Cancer

Abstract

To evaluate the agreement between radiologic staging of bladder cancer using multidetector row computed tomography (CT) and histopathologic staging and estimate the influence of interobserver variability of the CT findings as a potential limitation of this imaging modality. The available data on the value of multidetector row CT in clinical staging before cystectomy are controversial. The multidetector row CT reports of all patients undergoing radical cystectomy at our institution from 2004 to 2008 were retrospectively reviewed and compared in a blinded expert review by an experienced abdominal/genitourinary radiologist. The results of both radiologic reviews were subsequently correlated with the pathologic findings of the surgical specimens. The interobserver variability of radiology reports was estimated using κ statistics. Preoperative CT scans were available for 276 patients who underwent radical cystectomy. The accuracy of the primary and reference radiologists in predicting the correct local tumor stage was 49% (κ 0.23, P < .001) and 51% (κ 0.24, P < .001), respectively. The accuracy in predicting the presence of lymph node metastases was 54% (κ 0.04, P = .297) and 58% (κ 0.15, P = .011). The agreement between both radiologists was fair with regard to the local tumor stage (κ 0.23, P < .001) and the presence of lymph node metastases (κ 0.35, P < .001). The overall agreement between the local bladder cancer stage between CT and pathologic staging was poor to fair. Significant interobserver variability was found in the CT findings that might contribute to the limited accuracy of CT in the detection of extravesical tumor spread, infiltration of extravesical organs, and lymph node involvement.