AI-assisted medical coding: suggestions your coders can defend

The article discusses the potential benefits of AI-assisted medical coding, specifically how such a tool could help coders defend their work in the face of increasing complexity and staff shortages in the healthcare industry. The National Efficient Price for activity-based funding in Australia has risen by 5.9% since 2025-26, while the number of separations from hospitals has grown by about 7% between 2018 and 2022.

Many coding programs have been discontinued due to low enrollment, leading to a need for more efficient coding processes. However, fully automatic coding is not defensible at this time, as AI models like GPT-4 still struggle with exact matches for ICD-10-CM codes, often producing non-billable or fabricated codes. Additionally, Australian coding editions change every three years, requiring code indexes to be versioned accordingly.

The article proposes an architecture for an AI-assisted coding system that incorporates documents from electronic health records (EMR) and pathology systems (PAS). The system retrieves these documents, queries a code index based on the separation date, and generates suggested codes from a retrieved set. The code index is built from licensed IHACPA code lists for each edition, and a rules layer applies Australian Coding Standards.

A coder-friendly interface displays suggested codes with the exact text span they came from, facilitating accurate review and auditing. The system also maintains an append-only audit log that demonstrates human assignment of every code, ensuring compliance and providing a quality feedback loop for the coding team.