Optimising the paradigms of human AI collaborative clinical coding

Area of Science:

• Health Informatics
• Artificial Intelligence in Medicine
• Clinical Documentation Improvement

Background:

• Manual clinical coding is time-consuming and prone to errors.
• Automated clinical coding (ACC) offers a potential solution but requires human oversight for real-world application.
• Integrating human expertise with ACC systems is crucial for maximizing efficiency and accuracy.

Purpose of the Study:

• To propose and evaluate CliniCoCo, a novel human-in-the-loop (HITL) framework for effective collaboration between ACC systems and human coders.
• To optimize annotation workloads and improve the efficiency of the coding process using deep learning.
• To assess the impact of CliniCoCo on coding time, accuracy, and the performance of professional coders.

Main Methods:

• Developed a human-in-the-loop (HITL) framework named CliniCoCo, leveraging deep learning for automated clinical coding (ACC).
• Implemented collaborative strategies across annotation, training, and user interaction stages within the framework.
• Conducted experiments using real-world electronic medical record (EMR) datasets from Chinese hospitals.

Main Results:

• CliniCoCo achieved F1 scores of 0.80-0.84 with optimized annotation workloads.
• The system demonstrated the ability to halve annotation requirements for EMRs with 30% mistaken codes, with a minimal 0.01 F1 decrease.
• Human evaluations showed CliniCoCo reduced coding time by 40% and improved correction rates for EMR mistakes, including a threefold improvement in identifying missing codes.
• Professional coders' performance improved significantly, with F1 scores boosted to over 0.93 from 0.72.

Conclusions:

• The CliniCoCo framework effectively integrates automated clinical coding with human expertise, enhancing efficiency and accuracy in EMR processing.
• HITL approaches, like CliniCoCo, are vital for optimizing ACC systems in practical healthcare settings.
• CliniCoCo offers a significant advancement in clinical coding, reducing workload and improving data quality for better healthcare outcomes.