LLM-assisted clinical coding audit through an interpretable coding pipeline.

AI / informaticsInternational journal of medical informatics· Jul 2026

LLM-assisted clinical coding audit through an interpretable coding pipeline.

Supriya Khadka, Xiaorui Jiang, Vasile Palade

Read paper

Large Language Models in CareAI / informaticsClinical practiceQ1IF 6.2

Large Language Models-assisted audit found undercoding in 76.3% of a benchmark coding dataset

LLM-assisted clinical coding audit through an interpretable coding pipeline.

Supriya Khadka … Vasile Palade

International journal of medical informatics · Jul 2026 · Q1 in Health Informatics

Purpose

Clinical coding is vital yet complex, often hindered by imperfect training data.

Methods

We developed a novel and effective interpretable coding pipeline that integrates Large Language Models (Large Language Models) for evidence extraction and code verification with a multiclass classifier trained on a large-scale silver-standard evidence dataset for code prediction.

Results

76.3%

more than three in four coded records were missing codes they should have had

More results

Re-evaluating existing models on the corrected datasets improved performance, consistently across all metrics on MDACE, and in recall on CodiEsp.

More results

Models and results that are publishable in compliance with the data privacy requirements of the datasets have been made publicly accessible in the acocmpanied Github repository: https://github.com/Supriya090/LLM-Clinical-Coding-Audit.

“

Conclusion · 1 of 2

Our findings demonstrate that substantial errors in benchmark datasets significantly impact model evaluation.

Conclusion · 2 of 2

The results underscore a critical need to shift the research focus from purely model-centric approaches to data-centric solutions in clinical AI, highlighting the effectiveness of AI-assisted audits in improving data integrity.

LLM-assisted clinical coding audit through an interpretable coding pipeline.

LLM-assisted clinical coding audit through an interpretable coding pipeline.

Related papers