Pancreatic cancer detection using 5-hydroxymethylation signatures in plasma-derived cell free DNA in high-risk patients with new onset diabetes.

Abstract

539 Background: Pancreatic cancer (PaCa) is the third leading cause of cancer death in the United States despite a low incidence rate. It is often diagnosed when cancer has already metastasized to distant organs. Late diagnosis deprives patients of potentially curative treatments such as surgery and impacts survival rates. People with new onset diabetes (NOD) are at 6-8 fold increased risk for PaCa compared to the general population. Indeed, 0.85% of patients with NOD will be diagnosed with PaCa within 3 years. This population of PaCa patients with NOD constitute 25% of all new pancreatic cancer diagnoses. Surveillance of the NOD population for PaCa presents an opportunity to shift PaCa diagnosis to earlier stage. Methods: Whole blood was obtained from a cohort of 167 PaCa patients and 490 patients with cancers other than PaCa as well as 836 non-cancer controls with and without NOD. Plasma was processed to isolate cfDNA and 5hmC libraries were generated and sequenced. 5hmC data is used to generate models for PaCa detection using Bluestar Genomics’s technology platform. Results: To investigate whether PaCa can be detected in plasma, we interrogated plasma-derived cfDNA hydroxymethylation in PaCa patients and non-cancer controls. Models trained using 5hmC-based biomarkers from cfDNA consistently performed with a mean test sensitivity of 61.1% [95% confidence interval (CI): 35.7% to 82.7%] and a test specificity of 97.6% (CI: 93% to 99.5%) measured across 50 cross validation iterations within the training data set, which was composed of 48.3% early stage (Stages I & II) disease. The final model was trained using all of the training data, yielding 58.4% (CI: 47.5% to 68.8%) sensitivity at 98% (CI: 96.5% to 99.0%) specificity. This model was then tested on an independent test set with 22 PaCa patients (51.7% early stage, 15 of which was NOD) and 123 non-cancer control patients (53 of which were NOD) and yielded a classification performance of 59.1% (CI: 36.4% to 79.3%) sensitivity at 95.9% (CI: 90.8% to 98.7%) specificity. The model performance in the subset of patient cohort with NOD was 53.3% (CI: 26.6% to 78.7%) sensitivity at 94.3% (CI: 84.3% to 98.8%) specificity. Lastly, sensitivity observed on an independent validation set, composed of 56 PaCa and 117 ITTP samples, was 46.4% (CI: 33.0% to 60.2%) with 100% (CI: 96.8 to 100%) specificity. Conclusions: Our results demonstrate PaCa detection in plasma-derived cfDNA using 5hmC profiles. Overall, the model performed consistently between the training and independent validation datasets. A larger clinical study is under development to clinically validate the model described in this study with the goal of identifying occult PaCa within the NOD population in order to enable earlier detection and thus improve patient outcomes.