Building a Natural Language Interface for FHIR Clinical Terminology Server

Area of Science:

• Health Informatics
• Artificial Intelligence
• Natural Language Processing

Background:

• Fast Healthcare Interoperability Resources (FHIR) clinical terminology servers facilitate coded medical data access.
• Current FHIR systems face challenges with search accuracy due to errors and strict query syntax requirements.
• Users need to understand FHIR API formats and attribute codes for data retrieval.

Purpose of the Study:

• To develop an improved natural language interface for the FHIR clinical terminology system.
• To address limitations in search accuracy and ease of use associated with the standard FHIR API.
• To enhance the retrieval of coded medical data through more intuitive interaction methods.

Main Methods:

• Integration of two advanced large language models (LLMs).
• Implementation of speech-to-text conversion for voice input.
• Utilization of vector semantic searching for enhanced data matching.
• Development of a conversational artificial intelligence component for user interaction.

Main Results:

• Preliminary evaluations indicate promising performance for the natural language interface.
• The interface demonstrates potential to overcome FHIR search inaccuracies caused by typos or word variations.
• Improved user experience is anticipated by simplifying data retrieval processes.

Conclusions:

• A novel natural language interface shows potential to significantly enhance FHIR clinical terminology system functionalities.
• The integration of LLMs and other AI technologies offers a robust solution for more accessible medical data retrieval.
• Further development and evaluation are warranted to fully realize the benefits for healthcare professionals and researchers.