Optimising Human-AI Collaboration in Clinical Coding
Author Information
Author(s): Gao Yue, Chen Yuepeng, Wang Minghao, Wu Jinge, Kim Yunsoo, Zhou Kaiyin, Li Miao, Liu Xien, Fu Xiangling, Wu Ji, Wu Honghan
Primary Institution: Beijing University of Posts and Telecommunications
Hypothesis
How can human and AI work effectively together in clinical coding tasks?
Conclusion
The CliniCoCo framework significantly reduces coding time and improves accuracy in clinical coding tasks.
Supporting Evidence
- CliniCoCo reduces coding time by 40% on average compared to manual coding.
- The model achieves F1 scores around 0.80–0.84.
- Senior coders' performance improved from 0.72 to over 0.93 F1 score.
- CliniCoCo can suggest halving the annotations with minimal impact on accuracy.
- Human evaluations showed significant improvements in correction rates for EMR mistakes.
Takeaway
This study shows that using AI to help human coders can make coding faster and more accurate.
Methodology
The study used a human-in-the-loop framework with deep learning to optimize clinical coding processes.
Potential Biases
Potential biases may arise from the reliance on human coders' feedback and the quality of EMR data.
Limitations
The study's findings may not generalize to all clinical settings due to variability in EMR quality and coder experience.
Participant Demographics
Participants included clinical coders with varying levels of experience: seniors, juniors, and interns.
Statistical Information
P-Value
0.01
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website