RAG-Coding: Enhancing LLM Medical Coding with Structured External Knowledge

Researchers introduce RAG-Coding, an AI system using multiple LLM agents enhanced with retrieval-augmented generation to automate ICD-10-CM medical coding. The method outperforms baseline LLM approaches by 8-13% in accuracy and maintains clinical compliance by grounding decisions in official coding guidelines, while a newly released updated dataset enables evaluation against 2025 standards.

RAG-Coding represents a meaningful advancement in applying generative AI to healthcare administration, specifically medical coding—a critical but labor-intensive task affecting billing accuracy and clinical documentation. The system's architecture orchestrates multiple LLM agents that query external knowledge sources including official ICD-10-CM coding guidelines, creating a feedback loop that improves decision quality beyond what individual language models achieve independently. This approach addresses a fundamental limitation of large language models: their tendency to hallucinate or deviate from specialized domain requirements without grounding in authoritative references.

The healthcare coding landscape has faced longstanding inefficiencies, with manual coding remaining error-prone and resource-intensive. Traditional machine learning methods struggle with the complexity of medical terminology and the evolving coding standard updates. The introduction of MDACE-2025, an updated dataset with expert re-annotations aligned to current 2025 guidelines, signals commitment to maintaining benchmark relevance—a critical consideration in rapidly evolving regulatory environments.

For healthcare providers and technology vendors, RAG-Coding's performance metrics (8-13% F1 improvement) suggest potential cost reduction through automation while maintaining or exceeding accuracy thresholds necessary for billing compliance. The comparable performance to state-of-the-art pretrained models while using less specialized training indicates broader applicability across different LLM architectures. Healthcare IT companies and revenue cycle management platforms may adopt this methodology to enhance their coding automation tools, potentially disrupting traditional coding software vendors.

The technical validation across multiple LLM backbones demonstrates robustness, though real-world deployment requires integration with existing electronic health record systems and extensive clinical validation before full-scale adoption.

• →RAG-Coding achieves 8-13% accuracy improvement over baseline LLM methods by grounding decisions in official medical coding guidelines
• →The system's multi-agent architecture with retrieval-augmented generation addresses hallucination problems in specialized healthcare applications
• →MDACE-2025 dataset update ensures benchmarking reflects current 2025 ICD-10-CM standards, critical for regulatory compliance evaluation
• →Comparable performance to specialized pretrained models suggests the approach generalizes across different LLM backends
• →Methodology may reduce coding errors and operational costs for healthcare providers while maintaining clinical compliance standards