FHIR-GPT Enhances Health Interoperability with Large Language Models

Area of Science:

• Health Informatics
• Natural Language Processing
• Artificial Intelligence in Healthcare

Background:

• Health data interoperability is crucial for research, clinical trials, and public health surveillance.
• The Fast Healthcare Interoperability Resources (FHIR) standard is promoted by federal agencies to advance interoperability.
• Transforming heterogeneous electronic health record data, especially unstructured text, into FHIR resources presents significant challenges.

Purpose of the Study:

• To assess the capability of large language models (LLMs) in converting clinical narratives into FHIR resources.
• To develop and evaluate a specific LLM, FHIR-GPT, for transforming clinical texts into FHIR medication statements.

Main Methods:

• Development of FHIR-GPT, a generative pretrained transformer model tailored for clinical text to FHIR medication statement transformation.
• Experimental evaluation using 3671 snippets of clinical texts to compare FHIR-GPT performance against existing methods.

Main Results:

• FHIR-GPT achieved an exact match rate exceeding 90% in transforming clinical texts into FHIR medication statements.
• FHIR-GPT demonstrated substantial improvements over existing NLP pipelines, particularly in extracting timing schedules (over 50%), forms (42%), and reasons (35%).

Conclusions:

• Large language models show significant potential for enhancing health data interoperability by efficiently converting unstructured clinical narratives into structured FHIR resources.
• FHIR-GPT offers a promising solution to overcome the challenges associated with health data transformation, reducing development costs and improving accuracy.