Automated clinical coding: what, why and where we are
A perspective on why coding clinical records is hard to automate, covering seven challenges and why knowledge-based methods should be combined with deep learning.
The question
Clinical coding turns the information in a patient’s record into structured codes, such as ICD-10, so it can be used for statistics and research. It is a cognitive, time-consuming task that follows a standard process to stay consistent. Can AI take it on?
What we did
This perspective piece, led by Hang Dong, introduces the idea of automated clinical coding and summarises its challenges from an AI and language-processing point of view. It draws on the literature, our project experience from late 2019 to early 2022, and discussions with clinical coding experts in Scotland and the UK.
What we found
The piece covers seven challenges, including the lack of high-quality benchmark datasets, heterogeneous, incomplete and noisy data sources, and learning from low resources. It argues there is a gap between current deep learning approaches and the explainability and consistency that real practice needs, so knowledge-based methods that capture the standard process may need to be combined with deep learning.
Why it matters
Coders need to be involved in development. There is much to achieve before an AI system can support coding in routine use, which is the direction our later human-AI coding work took.