From Text to Knowledge: An End-To-End Extraction Pipeline for Clinical Information

Area of Science:

• Medical Informatics
• Natural Language Processing

Background:

• Clinical free texts contain valuable patient data, but extracting structured information is challenging.
• Automating the extraction of allergy information is crucial for patient safety and clinical decision-making.

Purpose of the Study:

• To evaluate the efficacy of open-source Large Language Models (LLMs) in extracting and structuring allergic reaction data from German clinical free texts.
• To develop and test an end-to-end workflow for allergy data extraction, mapping, and formatting.

Main Methods:

• Utilized open-source LLMs (Llama 3.1, Qwen 2.5, Mistral NeMo) on 500 anonymized German discharge letters.
• Developed a workflow to extract allergy information, map to SNOMED CT codes, and format into HL7 FHIR resources.

Main Results:

• High accuracy achieved in detecting and encoding allergy substances.
• Challenges identified in reaction identification and encoding due to complex string-matching issues.
• Performance varied across different LLMs for specific tasks.

Conclusions:

• LLMs demonstrate potential for automating the extraction of domain-specific healthcare data.
• Further improvements are possible through specialized model training and advanced prompt engineering.
• This proof-of-concept highlights LLMs' utility in handling complex clinical data for improved healthcare automation.