Altered metabolism and machine learning for pancreatic cancer early detection

Pancreatic cancer is the 3rd leading cause of cancer death in the United States. The high mortality of pancreatic ductal adenocarcinoma (PDAC) is largely a consequence of diagnosis at an advanced stage when the tumor is no longer treatable for cure. Currently, asymptomatic screening for PDAC is not recommended for the general population, with screening pursued only for a small subset of patients with pancreatic cystic lesions or strong genetic risk for PDAC. Even when cancer is identified early, patients can have rapid recurrence after surgical resection, most often with liver metastases. To improve outcomes for patients with PDAC, a number of important advancements are urgently needed, including improved risk assessment to identify those at elevated risk for PDAC, new non-invasive biomarkers to select individuals for intensive imaging surveillance, and better strategies to identify those with localized tumors who are at risk for rapid recurrence after surgical resection. In the current proposal, we directly address these critical areas of need, focusing on: (a) machine learning models for risk assessment from electronic medical record data (Aim 1), (b) development of non-invasive biomarkers from stool and computed tomography (CT) imaging that measure metabolic alterations caused by early PDAC (Aim 2), and (c) characterization of CT imaging and circulating cell-free DNA methylation patterns to predict presence of occult metastases at the time of surgical resection (Aim 3). Furthermore, we will collect and make available clinical data, blood samples, stool samples, imaging studies, and tumor tissue from multiple patient populations critical to PDAC early detection research, including those with early-stage PDAC, chronic pancreatitis, genetic PDAC risk, pancreatic cystic lesions, and non-cancer controls (Aim 4). To accomplish the proposed work, we have assembled a highly experienced and collaborative team that is fully committed to working together and with other Pancreatic Cancer Detection Consortium units. Thus, we will leverage cuttingedge machine learning approaches, develop multiple innovative, non-invasive biomarker technologies, and collect a large array of data and clinical samples for collaborative activities within and outside the Pancreatic Cancer Detection Consortium. With a highly dedicated expert team and clear scientific plan, we expect to achieve our near-term goal of reducing pancreatic cancer mortality by finding PDAC earlier and treating it more effectively for cure.