External Validation of an Imaging-Based Biomarker of Pancreatic Ductal Adenocarcinoma

Abstract

Listen

Translate article

We have previously identified a prognostic imaging-based biomarker of pancreatic ductal adenocarcinoma (PDAC) on computed tomography (CT) scans. At baseline, patients with conspicuous (high delta) PDAC are more likely to have mesenchymal biology, less stroma, and poorer clinical outcomes compared to those with inconspicuous (low delta) tumors. The objective of this study is to validate this classification system on CT scans in a second geographic population at an outside institution. Pre-treatment intravenous (IV) contrast-enhanced computed tomography (CT) images were reviewed for 21 patients receiving treatment for PDAC between 2010 and 2017 in Atlanta, Georgia. The patient population sample was 52% male, 48% female, 42% White, 33% Black or African American, and 25% Other/Unknown with a mean age of 65 years. Two experienced raters independently scored all images. The tumor/parenchyma interface was visually classified as either high delta (the interface was well defined) or low delta (the interface was poorly defined). These groups were then analyzed for survival differences using the log-rank test and Kaplan-Meier survival estimates. Interobserver concordance was measured using Cohen’s kappa test. Interobserver concordance was substantial (κ=0.78) with agreement on 19 of 21 total images. Both observers witnessed a trend toward significantly lower overall survival in those with high versus low delta tumors. One observer reported a 2-year survival rate of 53% for patients with low delta tumors versus 12% for those with high delta tumors (p=0.046). This study provides a step towards external validation of a prognostic CT biomarker of PDAC, corroborating previously reported results with new data from a second geographic population at an outside institution. Important considerations affecting broad applicability of this new prognostic tool include the availability of high-resolution contrast CT imaging as well as inter-observer variability in image scoring. Objective quantitative methods are in development to eliminate this variability. Additional images have been obtained from other institutions and are being analyzed for further external validation.