# Decomposing Error and Style in Automated Clinical Coding

arXiv: https://arxiv.org/abs/2609.24877 · HTML: https://mlsanity.com/paper/2609.24877/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 10 of its 18 references that are in the map.
- First posted on arXiv 21 Sep 2026.

## Builds on (most cited first, 10 of 10)

- [UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction](https://mlsanity.com/paper/1802.03426/) (2018)
- [Explainable Prediction of Medical Codes from Clinical Text](https://mlsanity.com/paper/1802.05695/) (2018)
- [The 'Problem' of Human Label Variation: On Ground Truth in Data, Modeling and Evaluation](https://mlsanity.com/paper/2211.02570/) (2022)
- [ACI-BENCH: a Novel Ambient Clinical Intelligence Dataset for Benchmarking Automatic Visit Note Generation](https://mlsanity.com/paper/2306.02022/) (2023)
- [PLM-ICD: Automatic ICD Coding with Pretrained Language Models](https://mlsanity.com/paper/2207.05289/) (2022)
- [Automated Medical Coding on MIMIC-III and MIMIC-IV: A Critical Review and Replicability Study](https://mlsanity.com/paper/2304.10909/) (2023)
- [VariErr NLI: Separating Annotation Error from Human Label Variation](https://mlsanity.com/paper/2403.01931/) (2024)
- [Automated clinical coding using off-the-shelf large language models](https://mlsanity.com/paper/2310.06552/) (2023)
- [MedCodER: A Generative AI Assistant for Medical Coding](https://mlsanity.com/paper/2409.15368/) (2024)
- [Toward Reliable Clinical Coding with Language Models: Verification and Lightweight Adaptation](https://mlsanity.com/paper/2510.07629/) (2025)

## Abstract (arXiv)

In automated clinical coding, where the label space spans tens of thousands of diagnosis and procedure codes, models are currently evaluated against a single gold annotation, treating any deviation as error. But we find when two teams code the same 110 ACI-Bench encounters, they agree on only 73% of codes (Jaccard similarity) for the same note; even after an independent clinical audit removes erroneous codes, agreement rises only to 77%. Is that gap error or something systematic? We model the systematic component as coding style psi, a coder- or site-specific policy over what to code and how much to document, and recast coding as p(codemidnote,psi), estimating psi with a 10-dimension rubric. If style were noise, conditioning on it would do nothing. Instead, across five datasets a model conditioned with a data-matching style raises ICD F1 by up to 26 points and an extreme mismatched one lowers it by up to 21. Four prompt based coding methods spanning 39-49 F1 converge to 52-56 once style is supplied (All p<0.05). Much of what single-gold evaluation charges to model error is recoverable, unmodeled style.