Pancreatic Cancer Risk Predicted from Electronic Health Records in U.S. Veterans Using Artificial Intelligence
Abstract
Fiscal Year 2022 (FY22) PCARP Focus Area(s) to be addressed in the proposed research: This proposal addresses the FY22 PCARP Focus Area Early detection research for pancreatic cancer, including the prevalence in individuals with pre-diabetes and diabetes and/or those in underserved ethnic and minority communities. Pancreatic cancer is an aggressive disease that typically presents late with poor patient outcomes. For example, 70% of Veterans with pancreatic cancer are diagnosed at a late stage of disease. However, patients who present with early-stage disease can be treated successfully. A better understanding of the risk factors for pancreatic cancer and detection at early stages coupled with the design of surveillance and intervention programs can improve patient survival and reduce overall mortality from this aggressive malignancy. The goal of the proposed research is to develop machine learning/artificial intelligence based predictive tools based on trajectories of data routinely available in the electronic health record (HER) to predict pancreatic cancer risk. We will develop a model in the Veteran population using nationwide data from the Veterans Affairs (VA) health care system and externally validate it in a civilian cohort from the Mass General Brigham (MGB, Boston) system. Innovative aspects of the proposed research project: This research is innovative in several respects. We propose to use the time sequence of clinical events, instead of only assessment at a single time point, in our artificial intelligence prediction model. In contrast, prior efforts to develop pancreatic risk prediction tools using patient records have used only the occurrence of disease codes, not the time sequence of disease states in a patient trajectory. We use a wide range of information from the EHR, including not only known risk factors but also less structured data that is routinely collected and could provide valuable clues under machine analysis. In addition, the proposed prediction methods predict not only whether cancer is likely to occur, but also provide risk assessment in incremental time intervals following the assessment, where time of assessment is defined as the day on which the risk prediction is performed based on the history of clinical records of the particular patient. This is important because the likely action resulting from a personalized positive prediction of cancer risk ideally should take into account the probability of the disease occurring within a shorter or longer time frame. Impact that the proposed research project s results might have on the field of pancreatic cancer research and/or patient care, including the goal of diminishing the burden of pancreatic cancer: Early identification of high-risk pancreatic cancer patients is an urgent need in current clinical practice, since detecting pancreatic cancer at an early stage is crucial toward avoiding poor outcomes and diminishing the burden of pancreatic cancer. Given the low predictive value of imaging and blood test screening, our high-performance AI model will provide an alternative and cost-efficient tool for pancreatic cancer risk assessment facilitating early detection. With completion of these aims, (a) we will be poised to implement a prediction-surveillance program that can be used to identify Veterans and civilians who are at elevated risk for pancreatic cancer and should be enrolled in surveillance and/or interception programs for disease detection, therapy, and prevention; (b) we will have characterized interactions among clinical risk factors on pancreatic cancer risk; and (c) we will have produced a prototype that applies AI technology to patient trajectory data for disease risk prediction for pancreatic cancer and beyond. Impact, in the short or long term, on individuals with pancreatic cancer, their families/caregivers, and/or the understanding of pancreatic cancer. The proposed research will have a major positive impact
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 04, 2024
- Source ID
- HT94252310463
Entities
People
- Chris Sander
Organizations
- Harvard University
- United States Army