Reproduce “Structured Extraction of Real World Medical Knowledge using LLMs for Summarization and Search” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Edward Kim, Manil Shrestha, Richard Foty, Tom DeLay, Vicki Seyfert-Margolis

Researchers at RespondHealth and Drexel University developed a method to transform unstructured Electronic Health Records into structured patient knowledge graphs using Large Language Models. This approach facilitates the discovery and characterization of rare diseases, demonstrated by identifying patients with ultra-rare BPAN and highlighting discrepancies in phenotypic frequencies for Dravet Syndrome when compared to standard ontologies.

Key insights

  • Large Language Models (LLMs) significantly improve entity extraction and multi-label phenotype classification from clinical text, especially when employing few-shot prompting and dynamic few-shot techniques.
  • Knowledge graphs constructed from real-world EHR data can reveal discrepancies between observed patient phenotype frequencies and standardized ontology descriptions, as shown in the Dravet Syndrome case study.
  • The knowledge graph approach enables the identification of patients with rare diseases lacking specific diagnostic codes, by searching for consistent phenotypic presentations.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.