Enhancing MAUDE Database Utility by GPT-4 and Cause-Effect Visualization.

Area of Science:

• Medical device safety
• Health informatics
• Artificial intelligence in healthcare

Background:

• The Manufacturer and User Facility Device Experience (MAUDE) database contains vital information on medical device malfunctions and adverse events.
• The complexity and volume of MAUDE data present significant challenges for efficient analysis and insight extraction.

Purpose of the Study:

• To develop an automated analytical pipeline utilizing GPT-4 to streamline the extraction, categorization, and visualization of safety events from the MAUDE database.
• To reduce the need for extensive human annotation in analyzing medical device safety data.

Main Methods:

• An automated analytical pipeline was developed using GPT-4, a large language model.
• 4,459 colonoscopy reports from MAUDE (2011-2021) were analyzed.
• Safety events were categorized into operational, human factor, and device-related types. Ishikawa diagrams visualized data, stored in a vector database for similarity search.

Main Results:

• The pipeline successfully extracted and categorized safety events from colonoscopy reports with minimal human annotation.
• A subset of events was visualized using Ishikawa diagrams and stored for efficient retrieval and comparison via similarity search.

Conclusions:

• The automated GPT-4 pipeline offers an efficient method for analyzing the MAUDE database, overcoming data complexity.
• This approach significantly reduces the workload for human annotators and enhances the accessibility of critical safety insights from medical device reports.