Generating a Health Information Technology Event Database from FDA MAUDE Reports

Area of Science:

• Health Informatics
• Medical Device Safety
• Machine Learning in Healthcare

Background:

• Patient safety events (PSEs) are critical issues in healthcare systems.
• Health information technology (HIT) aims to improve care quality but can introduce unintended safety consequences.
• A comprehensive database of HIT events is crucial for understanding their impact.

Purpose of the Study:

• To develop and evaluate machine learning models for extracting HIT-related patient safety events from the FDA MAUDE database.
• To assess the accuracy and effectiveness of classic and convolutional neural network (CNN) models in identifying HIT events.

Main Methods:

• Utilized classic and CNN models to process and extract data from the FDA MAUDE database.
• Evaluated individual and combined model performance on a dedicated test set.
• Focused on identifying health information technology (HIT) events specifically.

Main Results:

• The best performing model achieved approximately 90% accuracy and a 0.87 f1 score in identifying HIT events.
• The model demonstrated capability in creating an HIT-exclusive database.
• The developed strategy can serve as a quality and error check for event reporting.

Conclusions:

• Machine learning models can effectively identify health information technology (HIT) events from the MAUDE database.
• This approach enhances the quality of HIT event reporting and analysis.
• The methodology shows potential for developing subtype-specific patient safety event databases.