Reproduce “PRISM: A Transformer-based Language Model of Structured Clinical Event Data” 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
Lionel Levine, John Santerre, Alex S. Young, T. Barry Levine, Francis Campion, Majid Sarrafzadeh
PRISM, a transformer-based language model, was designed to model the sequential progression of clinical decision-making using structured event data. It achieved a perplexity of 2.01 on a 10,000-token clinical vocabulary when trained on MIMIC-IV, demonstrating its capability to reduce uncertainty in predicting subsequent clinical events and generate coherent diagnostic pathways.
Key insights
- Clinical decision-making processes can be effectively framed and modeled as sequential generative tasks, analogous to natural language modeling.
- Transformer-based architectures are well-suited to capture complex, long-range dependencies and contextual relationships within longitudinal patient clinical histories.
- By predicting 'what's next' rather than merely classifying static outcomes, AI systems can offer more dynamic and nuanced support for clinical reasoning.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.