MedDCR: Learning to Design Agentic Workflows for Medical Coding

TL;DR

MedDCR addresses the challenge of automated medical coding by learning agentic workflows through a closed-loop design–execute–reflect cycle. It introduces a meta-agent architecture—Designer, Coder, and Reflector—augmented by a memory archive that enables reuse and progressive refinement of coding pipelines under guideline constraints. Experiments on MDACE and ACI-BENCH show MedDCR achieves state-of-the-art performance, e.g., Micro-F1 of 0.51 on MDACE and F1 of 0.52 on ACI-BENCH, with cost-efficient search relative to execution. The work demonstrates that automated workflow optimization yields higher accuracy and greater interpretability than fixed, hand-crafted pipelines, and offers plug-and-play integration with expert workflows for practical deployment in clinical coding.

Abstract

Medical coding converts free-text clinical notes into standardized diagnostic and procedural codes, which are essential for billing, hospital operations, and medical research. Unlike ordinary text classification, it requires multi-step reasoning: extracting diagnostic concepts, applying guideline constraints, mapping to hierarchical codebooks, and ensuring cross-document consistency. Recent advances leverage agentic LLMs, but most rely on rigid, manually crafted workflows that fail to capture the nuance and variability of real-world documentation, leaving open the question of how to systematically learn effective workflows. We present MedDCR, a closed-loop framework that treats workflow design as a learning problem. A Designer proposes workflows, a Coder executes them, and a Reflector evaluates predictions and provides constructive feedback, while a memory archive preserves prior designs for reuse and iterative refinement. On benchmark datasets, MedDCR outperforms state-of-the-art baselines and produces interpretable, adaptable workflows that better reflect real coding practice, improving both the reliability and trustworthiness of automated systems.

Figures (2)

• Figure 1: Overview of the MedDCR framework. (1) The memory archive is initialised with general reasoning strategies (e.g., self-refinement, multi-agent ensembles, chain-of-thought prompting) and coding-specific strategies (e.g., medical term extraction, weak code filtering, ICD tool use), together with other optional seed workflows. (2) In each optimisation loop, the Designer proposes new workflows, the Coder compiles and executes them (with self-fixing if needed), and the Reflector provides both evaluation scores and textual feedback. The memory archive stores all past workflows, enabling reuse, progressive refinement, and workflow selection from top-performing and recent designs. This closed-loop process discovers effective coding workflows under guideline constraints.
• Figure 2: Case study of the search process on ACI-Bench. The blue line tracks the best workflow discovered at each iteration, measured by F1. The figure illustrates how performance improves as the system explores diverse candidates, learns from high-performing workflows, and balances precision and recall to refine the final design.