Preoperative CT Classification of the Resectability of Pancreatic Cancer: Interobserver Agreement

Background Accurate assessment of local resectability of pancreatic cancer at initial workup is critical to determine the most appropriate management strategy among up-front operation, neoadjuvant treatment, or palliative treatment. Purpose To investigate the interobserver agreement of the preoperative CT classification of the local resectability of pancreatic cancer and to determine if radiologist experience level impacts evaluation, and to evaluate the reader performance in assessing resectability at CT in a subset of patients with a reference standard for local resectability. Materials and Methods This retrospective study was composed of patients with pathologic-analysis-confirmed pancreatic cancers between January 2013 and December 2014 who underwent baseline multiphasic contrast agent-enhanced CT. Eight board-certified radiologists with different levels of experience (more experienced, ≥6 years, n = 4; less experienced, 1st- or 2nd-year fellows, n = 4) reviewed the CT images and classified cancers as resectable, borderline resectable, or unresectable. Interobserver agreements were determined for all reviewers and subgroups of reviewers stratified according to experience (more vs less) by using Fleiss κ statistics. In patients with reference standards for local resectability, diagnostic performances of each reviewer were assessed by using receiver operating characteristic curve analysis. Results There were 110 patients (mean age, 61 years ± 11; 60 men) who were evaluated. Overall interobserver agreements were moderate for resectability classification (κ = 0.48; 95% confidence interval: 0.45, 0.50). Only 30.0% of patients (33 of 110) were given the same resectability classification from all reviewers. More experienced reviewers demonstrated higher agreement in category assignments than less experienced reviewers (κ = 0.55 [95% confidence interval: 0.50, 0.60] vs 0.43 [95% confidence interval: 0.38, 0.49], respectively). For prediction at CT of margin-negative (ie, R0) resections (n = 82), areas under the receiver operating characteristic curve of all reviewers were greater than 0.80 (range, 0.83-0.96). However, borderline resectable cancers showed diverse R0 rates ranging from 0% to 74% depending on the reviewers. Conclusion Considerable interobserver variability exists in the assignment at CT of the local resectability of pancreatic cancer, even among experienced radiologists. © RSNA, 2019 Online supplemental material is available for this article.