Reproduce “Code Like Humans: A Multi-Agent Solution for Medical Coding” 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
Andreas Motzfeldt, Joakim Edin, Casper L. Christensen, Christian Hardmeier, Lars Maaløe, Anna Rogers
Corti.ai researchers introduce "Code Like Humans" (CLH), a multi-agent, LLM-based framework that mimics human medical coding workflows to handle the full US ICD-10-CM system with over 70,000 labels. This approach, leveraging external ICD resources, demonstrates improved performance on rare codes and establishes a new baseline for full-scale medical coding.
Key insights
- Emulating human multi-step reasoning and explicitly integrating external knowledge resources through an agentic LLM framework is crucial for addressing the full complexity of medical coding beyond constrained label spaces.
- Enriching LLM input context with structured guideline information and employing 'thinking-enabled' decoding significantly enhances agent performance, especially for refining and validating candidate codes.
- Performance analysis reveals specific 'blind spots' of the system, such as abbreviations, social qualifiers, and psychological terms, highlighting areas for future targeted improvements in LLM reasoning abilities.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.