Explainable ICD Coding via Entity Linking

TL;DR

This work reframes ICD coding as an explainable entity linking task to provide explicit textual evidence for each code, enabling better human–AI collaboration. It compares three BioMistral-based approaches—ICL-BioMistral, SFT-BioMistral, and InsGenEL-BioMistral—employing constrained decoding and evaluating on public explainable ICD datasets. Fine-tuned models outperform in-context learning, with InsGenEL offering multi-entity generation suitable for production, and SFT showing robust performance in few-shot settings. Limitations include reliance on gold mentions for evaluation and data scarcity; future work targets end-to-end mention detection, broader data collection, efficiency, and ethical deployment in clinical environments.

Abstract

Clinical coding is a critical task in healthcare, although traditional methods for automating clinical coding may not provide sufficient explicit evidence for coders in production environments. This evidence is crucial, as medical coders have to make sure there exists at least one explicit passage in the input health record that justifies the attribution of a code. We therefore propose to reframe the task as an entity linking problem, in which each document is annotated with its set of codes and respective textual evidence, enabling better human-machine collaboration. By leveraging parameter-efficient fine-tuning of Large Language Models (LLMs), together with constrained decoding, we introduce three approaches to solve this problem that prove effective at disambiguating clinical mentions and that perform well in few-shot scenarios.