Implementing a natural language processing pipeline to expedite clinical chart review to assess germline genetic testing in veterans with breast cancer.

• DOI: 10.1200/jco.2024.42.16_suppl.e13640
• Publication Date: 2024-05-29T17:43:22Z
• Authors: Vikas Patil; Richard Mcshinsky; Ahmad Sami Halwani; Deborah Morreall; Chunyang Li; Kelli M Rasmussen; Christina Yong; Zachary R. Burnin...; Carolyn S. Menendez; Akiko Chiba; Haley Moss; Sarah Violet Colonna; Kerry G. Rowe; Michael J. Kelley

e13640 Background: The Making Advances in Mammography and Medical Options for Veterans Act of 2022 mandated that the Department Affairs (VA) perform a study germline genetic testing with breast cancer (BC). Assessing whether Veteran BC was offered based on electronic health records requires time-consuming review unstructured clinical notes by experts who are also extremely busy providing care. To expedite this process, we implemented productionized natural language processing (NLP) pipeline to identify relevant concepts highlight them within custom-made chart interface. Methods: Clinical subject-matter (SMEs) were convened keywords suggestive testing. These categorized into 3 sets: gene-related concepts, VA-specific services, exams tests. We then used entity recognition linking link these knowledge base. enriched entities similarly extracted from an open-source reference text BC. resulting visualized as word cloud reviewed SMEs. Power BI Apps interface display 200 randomly selected Veterans, highlighted. Each Veteran’s independently 2 5 SMEs, assessed testing, referred counseling service, or otherwise had evidence Disagreements adjudicated whole team. Results: 11622 (mean 58/Veteran, 3028 characters/note) reviewed. Table shows most prevalent among genes, mutations, exam names. Of reviewers, annotated 80 patients each single session, lasting less than 50 minutes average. Reviewers inter-annotator agreement 85.5%. Mention target associated accuracy 80.4%, precision 76.4%, recall 99.1%, F1 86.3%. Germline rates detailed separate submission. Conclusions: Modern informatics tools such NLP, Microsoft Platform, high-performance computing environments allow rapid implementation information technology can enable clinicians gain critical insights quickly efficiently record data. [Table: see text]