# Automated Clinical Coding: What, Why, and Where We Are?

**Authors:** Hang Dong, Matúš Falis, William Whiteley, Beatrice Alex, Joshua Matterson, Shaoxiong Ji, Jiaoyan Chen, Honghan Wu  
**arXiv:** [2203.11092](https://arxiv.org/abs/2203.11092)  
**Date:** 2022-03-21  
**Summary page:** https://www.summarizepaper.com/en/paper/2203.11092v3/

## Key points

- Automated clinical coding is a promising task for AI and NLP to improve the efficiency and accuracy of transforming medical information into structured codes.
- Challenges need to be addressed to develop an AI-based automated system that is human-centered, explainable, intelligent, and robust to complex real-world scenarios.
- One of the main challenges is handling long documents with over 10 thousand tokens.
- Text redundancy or "Note Bloat" problem may impede the performance of deep learning models for code prediction.
- Achieving better performance than traditional methods such as CNN-based approaches for multi-label classification applied to clinical coding is another challenge due to inefficiency in modelling concept-level information and long documents.
- Future deep learning-based systems need to integrate knowledge reasoning with rules and ontologies for improved and more explainable results.
- Organizational challenges also need addressing before deploying an AI-based coding tool into the clinical coding environment.
- Coders need to be involved in model development and deployment stages despite being occupied with their coding work.
- Data availability should also be made available within papers related to this topic.

## Summary

Automated clinical coding is a promising task for Artificial Intelligence (AI) and Natural Language Processing (NLP), as it has the potential to improve the efficiency and accuracy of transforming medical information in a patient's health records into structured codes for statistical analysis. However, there are several challenges that need to be addressed to develop an AI-based automated system that is human-centered, explainable, intelligent, and robust to complex real-world scenarios. One of the main challenges is handling long documents. The recent Transformer-based pre-trained language models have limited input length due to their memory-demanding self-attention mechanism, while clinical notes can have up to over 10 thousand tokens. Additionally, text redundancy or "Note Bloat" problem may impede the performance of deep learning models for code prediction. Another challenge is achieving better performance than traditional methods such as CNN-based approaches for multi-label classification applied to clinical coding. This limitation may be due to inefficiency in modelling concept-level information and long documents. Moreover, manual coding is largely based on a standard process with rules applied to the healthcare system. Future deep learning-based systems need to integrate knowledge reasoning with rules and ontologies to achieve improved and more explainable results. Organizational challenges also need addressing before deploying an AI-based coding tool into the clinical coding environment. Coders need to be involved in model development and deployment stages despite being occupied with their coding work. Data availability should also be made available within papers related to this topic. In conclusion, while there are technical and organizational challenges that need addressing before deploying an AI-based automated system for clinical coding, there is a clearer path forward thanks to growing numbers of studies and projects in academia and industry. With further research support on projects in medical informatics and computer science, we look forward to seeing more advances in AI-assisted clinical coding in the next five years and beyond.