Leveraging LLMs to Understand Narratives in Maude Reports

Abstract

Interest in using the MAUDE database to investigate adverse events linked to medical devices has been growing. Yet, the narrative sections of these reports remain largely unexplored, leaving valuable insights unutilized and creating an incomplete understanding of these events. To bridge this gap, we employ large language models (LLMs) to analyze and interpret these narratives. Using OpenAI's GPT-4-turbo model, we focused on MAUDE reports involving endoscopic clips to identify uncoded surgical procedures and uncover additional insights. This approach showcases the potential of LLMs in processing narrative content, offering a more efficient and cost-effective alternative to previous methods and supporting the translation of MAUDE reports into actionable knowledge for clinical practice.