LAUSR

To the Editor: I ntraductal papillary mucinous neoplasms (IPMNs) account for approximately 25% of all cystic pancreatic neoplasms. Although they are often benign, IPMNs have garnered attention for their malignant potential given the highly fatal nature of pancreatic cancer. Studies of the natural history of IPMNs are disproportionately focused on higher-risk cases that have undergone surgical resection, leaving out the small IPMNs that comprise approximately 80% of cases. Longitudinal, generalizable data on a representative sampling of IPMN patients are required to improve our efforts to personalize surveillance recommendations, standardize society guidelines, and more appropriately recognize which patients should or should not undergo surgical resections. The presence of large-scale electronic health records (EHRs) provides an opportunity to assemble needed IPMN cohorts; however, the lack of specific diagnostic International Classification of Diseases, Ninth and Tenth Revisions (ICD-9/10) codes for IPMNs is a major roadblock to studying this condition. Natural Language Processing (NLP) is an artificial intelligence tool designed to intelligently process text for the purpose of building automated algorithms to help classify patientswhen standard codes are unavailable or inaccurate. Researchers have successfully used NLP technology in the study of other gastrointestinal conditions.We tested the diagnostic accuracy of multiple, automated key text word searches to already collected radiology imaging reports as a proof of concept thatNLPalgorithms could be used to identify important IPMN study cohorts.