An automation framework for clinical codelist development validated with UK data from patients with multiple...

Area of Science:

• Health Informatics
• Artificial Intelligence in Healthcare
• Clinical Data Management

Background:

• Codelists are essential for standardized healthcare communication but are time-consuming to create.
• Existing literature emphasizes codelist transparency, often overlooking automation's potential.
• Developing high-quality codelists requires significant clinical expert input and time.

Purpose of the Study:

• To present an automated framework for generating clinical codelists with minimal expert input.
• To demonstrate the framework's utility through the DynAIRx project case study.
• To make the developed framework and codelists publicly accessible for future use.

Main Methods:

• Developed a Codelist Generation Framework to automate codelist creation.
• Applied the framework to the DynAIRx project, which aims to optimize medication prescribing for patients with multiple long-term conditions using AI.
• Generated and validated approximately 214 codelists for DynAIRx with clinical experts.

Main Results:

• The framework automated the shrinking of codelists using trusted sources and added new codes for review.
• The DynAIRx case study generated a codelist of ~14,000 codes requiring only 7-9 hours of clinician time, a reduction of over 80% compared to traditional methods.
• Validation by experts confirmed the appropriateness of the generated codelists, significantly reducing preparation time.

Conclusions:

• The automated framework substantially lowers workload, minimizes human error, and saves considerable time, especially for clinical experts.
• Emphasis on automation and trusted sources enhances transparency and reproducibility in codelist development.
• This approach offers a significant improvement over traditional methods for creating complex healthcare codelists.