Using large language models to extract information from pediatric clinical reports.

Area of Science:

• Medical Informatics
• Natural Language Processing
• Pediatric Healthcare

Background:

• Unstructured medical documentation, like clinical reports, impedes efficient data analysis and integration.
• Current data extraction methods are labor-intensive and lack flexibility.
• Automated analysis of medical documents can significantly benefit patient care and research.

Purpose of the Study:

• To assess the performance of Large Language Models (LLMs) in extracting structured data from unstructured pediatric clinical reports.
• To evaluate the accuracy and flexibility of LLMs for medical data extraction.
• To compare the efficacy of different LLMs in processing clinical documentation.

Main Methods:

• Nine distinct Large Language Models (LLMs) were evaluated.
• LLMs were tasked with extracting structured, patient-specific information from pediatric clinical reports.
• Performance was measured by accuracy in identifying key data points.

Main Results:

• Both commercial and open-source LLMs demonstrated high accuracy in identifying patient-specific information.
• Top-performing LLMs achieved over 90% accuracy on critical data extraction tasks.
• LLMs provide a flexible and effective method for structuring unstructured clinical data.

Conclusions:

• Large Language Models are a viable and accurate tool for transforming unstructured pediatric clinical reports into structured data.
• LLM-based data extraction enhances efficiency and flexibility in medical data analysis.
• This technology holds significant potential for advancing clinical research and patient care systems.