ABSTRACT
OBJECTIVES: To assess the interreader agreement and reader performance in the evaluation of patients with pancreatic cancer (PC) in two classification systems of local resectability status prior to initiation of therapy, namely the National Comprehensive Cancer Network (NCCN) and Karolinska classification system (KCS). METHODS: In this ethics review board-approved retrospective study, six radiologists independently evaluated pancreatic CT-examinations of 30 patients randomly selected from a tertiary referral centre's multidisciplinary tumour board database. Based on well-defined criteria of tumour-vessel relationship, each patient was assigned to one of three NCCN and six KCS categories. We assessed the intraclass correlation coefficient (ICC) and compared the percentages of correct tumour classification of the six readers in both systems (Chi-square test; a P-value <0.05 was considered significant). The standard of reference was a consensus evaluation of CT-examinations by three readers not involved in the image analysis. RESULTS: The ICC for NCCN and KCS was 0.82 and 0.84, respectively (very strong agreement). The percentages of correct tumour classification at NCCN and KCS were 53-83% and 30-57%, respectively, with no statistically significant differences in the overall reader comparison per classification system. In pair-wise comparison between readers for NCCN/KCS, there were statistically significant differences between reader 5 vs. readers 4 (Pâ¯=â¯0.012) and 3 (Pâ¯=â¯0.045)/ reader 5 vs. reader 4 (Pâ¯=â¯0.037). CONCLUSION: Interreader agreement in both PC classification systems is very strong. NCCN may be advantageous in terms of reader performance compared to KCS.
