Analysis of Tissue Architecture to Identify Lethal Prostate Cancer in the Veteran Population
Abstract
The VA hospital system is ideally suited for population-based studies in prostate cancer (PCa) because it includes the full range of racial/ethnic groups represented in the U.S. population. The VA system was the first in the US to deploy an electronic health record, and the vast majority of medical notes, imaging reports, laboratory results, and demographic information are stored in large VA data repositories. These data are accessible for clinical operations and research personnel through the VA Informatics and Computing Infrastructure (VINCI). Currently, the number of veterans diagnosed and/or treated for PCa within the VA Healthcare System is approaching 1 million. Since many of these Veterans receive all of their cancer care within this system, from PSA screening through diagnosis and treatment, priceless long-term follow-up and outcomes are available. These aspects of VA Healthcare make it the ideal institution to assemble cohorts of men with PCa for biomarker discovery. By conducting research in VA cohorts, the results are likely to be more representative of various populations in the U.S., and the racial disparities that afflict most biomarker discoveries are greatly reduced. Therefore, we believe that taking advantage of the unprecedented large size of the VA PCa cohort of more than 1 million men will allow to make novel biomarker discoveries that will improve the medical treatments of Veterans in the future. All approaches using data analysis based on artificial intelligence (AI) require very large cohorts. The VA hospital system is a unique place where it is possible to put together a population of the size necessary for cutting-edge AI-driven biomarker development. To identify metastatic PCa cases in the VA, members of our research team previously constructed, tested, and validated a natural language processing tool and classified the entirety of PCa cases within VINCI, which included approximately 876,000 PCa cases, of which 275,409 were diagnosed at the VA and presumed to have diagnostic prostate needle biopsy (PNBX) specimens stored within the VA. Of these men, there are 35,802 metastatic cases and 21,481 (60%) will have been diagnosed with de novo metastases, while the others developed metastatic disease later in the disease course. Approximately 50% of the metastatic PCa cases are deceased. The goal of our project is to identify men at risk for metastatic PCa at the earliest possible time. Previous studies published by our team, which used a genomic analysis to characterize metastatic PCa, demonstrated that features of lethal PCa can be detected in diagnostic PNBX. In this project, we hypothesize that digital images of PNBX also contain metastatic features of PCa that can be identified by computers. We propose to collect regular pathology slides that were prepared at the time of the diagnostic PNBX from Veterans diagnosed with high-grade PCa who are included in the VINCI database. The large number of cases to choose from will allow our team to train and test AI algorithms that may be more accurate indicators of metastatic risk. For every case, we will generate a metastasis risk (MR) score. By studying the medical record, we will know that patient s outcome and whether or not metastatic progression occurred. This way, we can match the MR-score to the patient s disease course and determine its accuracy. Even though pathologists train for many years, they can encounter difficulties and a lack of consensus in PCa grading. Published studies demonstrated that computers can assist with grading, which is critical for treatment decisions. We will refine and test a previously developed computer-assisted algorithm for the Veteran population that can determine the grade and volume of cancer in every PCa slide. This algorithm will be subjected to extensive testing and, if successful, can be implemented in pathology practices at VA sites throughout the country to assist pathologists in diagnosis
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Dec 05, 2021
- Source ID
- W81XWH2110725
Entities
People
- Beatrice S Knudsen
Organizations
- United States Army
- University of Utah