Unveiling Endoscopic Procedures in MAUDE: An ETL-LLM Method

Area of Science:

• Medical Informatics
• Patient Safety Research
• Health Data Science

Background:

• The MAUDE database is vital for patient safety studies, but inconsistent data extraction methods hinder analysis.
• Lack of standardized methods for event report analysis leads to unreliable insights for clinicians.

Purpose of the Study:

• To develop and validate a standardized method for extracting and analyzing medical device event reports from the MAUDE database.
• To improve the consistency and reproducibility of patient safety data analysis.

Main Methods:

• An ETL (Extract, Transform, Load) pipeline was combined with Large Language Models (LLMs).
• The method was applied to endoscopic clip reports for extracting and analyzing categorical and narrative data.
• Focus on standardizing identification and interpretation of adverse event reports.

Main Results:

• The ETL-LLM method effectively extracted and analyzed both categorical and narrative reports.
• Identified key insights related to patient complications, surgical procedures, and device usage.
• Demonstrated improved data standardization and interpretability.

Conclusions:

• The proposed ETL-LLM method offers a transparent and effective approach to analyzing MAUDE data.
• This standardization has the potential to promptly inform clinicians and advance patient safety research.
• Encourages further research utilizing open-access databases for enhanced patient safety.