Reproduce “MedDec: A Dataset for Extracting Medical Decisions from Discharge Summaries” 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
Mohamed Elgaar, Jiali Cheng, Nidhi Vakil, Hadi Amiri, Leo Anthony Celi
Researchers from the University of Massachusetts Lowell and MIT developed MedDec, the first expert-annotated dataset for identifying and classifying medical decisions within clinical discharge summaries. This work established baseline performance using transformer models, achieving a span exact match F1 score of 34.8, and revealed significant challenges in extracting longer, more complex decision spans and rare decision categories.
Key insights
- The complexity of extracting medical decisions correlates with span length; longer spans are notably harder to identify and classify.
- Performance varies substantially across different patient phenotypes, influenced by training data volume and the inherent clinical clarity of the condition.
- Current general-purpose large language models (LLMs) show low performance for exact medical decision span extraction in zero-shot and one-shot settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.