2105 - Developing Natural Language Processing for Infection-related Safety Events

Session Description: Adverse events, near misses, and errors are underutilized reports for Infection Prevention departments that may be separate from Risk Management departments. These safety events have the potential to be harnessed to identify contributing factors and behaviors that led to adverse outcomes related to infection prevention activities. While Risk Management information systems may have a multitude of event reports, unstructured data — such as free text used for event descriptions—is difficult to rapidly mine, requiring many hours of manual analysis and potentially subject to personal bias.

This session will describe how Infection Prevention can collaborate with Patient Safety and Risk Management using qualitative analysis of safety events related to infection control breaches and infections. In order to categorize these events, most patient safety events are manually reviewed and classified. Using a dataset of over 4 million patient safety events, we built a Natural Language Processing (NLP) model to mimic the manual classification process to identify healthcare-associated infections and infection control breaches related to behaviors, process failures, and environmental factors. NLP promises to be a more timely, efficient, and robust method for data mining and analyzing reported safety events. This session seeks to serve as an introduction to potential semi-automated data extraction methods that can lead the way for infection control analytics.