Enhancing Patient Safety Event Analysis Using Artificial Intelligence: A Pilot Study of an Artificial...

To address the challenge of analyzing large volumes of patient safety event (PSE) reports, we developed and evaluated an AI-powered software tool. The primary goal was to assess the tool's potential to support analysts and uncover novel trends in patient safety databases.

A pilot evaluation was conducted with seven organizations (4 health care facilities and 3 patient safety organizations) to assess the tool's impact on analysts' workflows and their ability to uncover insights. Feedback was gathered through interviews with patient safety analysts using the tool. Two human factors experts analyzed the findings using a human cognition framework for information visualization to identify strengths and areas for improvement. Novel insights from PSE data were systematically recorded, capturing trends and themes that emerged during the analysis process.

Participants from 6 of 7 institutions reported that the tool helped identify valuable insights, such as trends in procedural errors, inconsistencies in event categorization, and emerging issues with specific medications and devices. The emerging themes algorithm effectively highlighted previously undetected patterns by grouping related events and emphasizing novel keywords. However, participants noted some irrelevant keywords due to limitations in narrative data quality. The tool's design principles, including chunking information and highlighting key terms, improved efficiency in reviewing reports.

The AI-driven tool demonstrated potential to enhance patient safety by supporting analysts in detecting trends and patterns in PSE reports. Future iterations will address identified limitations and further refine its ability to organize data around user mental models for improved usability.