Tumor resectability and response on CT following neoadjuvant therapy for pancreatic cancer: inter-observer agreement study.

Objectives: First, to measure inter-observer agreement regarding tumor resectability and response, and second, to measure diagnostic performance in predicting negative resection margin, on re-staging CTs of patients who received neoadjuvant therapy for pancreatic cancer.

Methods: This retrospective study included patients who received neoadjuvant therapy for borderline resectable pancreatic cancer from 2017 to 2020. Six readers independently evaluated initial staging and re-staging CT images. They categorized the resectability on re-staging CT based on the NCCN guideline, and evaluated tumor response to neoadjuvant therapy according to our proposed criteria on a 5-grade scale. For inter-observer agreement, Gwet's agreement coefficients were used. A crossed random effect model was used to pool the sensitivity and specificity of six readers in predicting negative resection margin.

Results: Seventy-seven patients with the median age of 66 (59-70) were included. The pooled agreement for tumor resectability was 0.64 (95% CI, 0.56-0.71) for differentiating the three categories, and 0.84 (0.77-0.91) for differentiating resectable or borderline resectable cancer vs. unresectable cancer. Agreement for tumor response grade was 0.89 (0.85-0.92). The pooled sensitivity and specificity for predicting negative resection margin were 48% (43-52%) and 61% (57-64%), respectively, when only "resectable" on re-staging CT was considered as index test positive. When either "resectable"' or "borderline resectable" was considered as positive, the pooled sensitivity and specificity were 91% (89-94%) and 5% (4-6%), respectively.

Conclusion: CT can be used reliably with a high inter-observer agreement in selecting candidates for surgery after neoadjuvant therapy of pancreatic cancer.

Key Points: • On CT following neoadjuvant therapy of pancreatic cancer, six readers showed high agreement in differentiating resectable or borderline resectable vs. unresectable cancer (Gwet's coefficient, 0.84). • Inter-observer agreement was also high for our proposed tumor response grade (Gwet's coefficient, 0.89). • Specificity was very low (5%) while sensitivity was high (91%) when either resectable or borderline resectable cancer on re-staging CT was considered as predictive of negative resection margin status.