A Sentiment Analysis of Patient Safety Event Reports

Abstract

Patient Safety Event (PSE) reporting systems are widely utilized by healthcare providers across the country as a method for gathering information from front-line healthcare providers about patient safety. Automated processes that utilize natural language processing (NLP) techniques have been developed and applied to analyze these reports. The objective of this study was to explore and solidify methods on how to build a sentiment analysis (SA) library specific to patient safety event reports. Our hypothesis is that application of such a library to large data sets would illuminate trends in the relationship between workplace culture and patient safety that could lead to actionable interventions by decision-makers. We examined PSE reports, gathered from a single system over a period of approximately 15 months. The analysis uncovered 36 unique keywords that vary in degree of sentiment, and can serve as a basis for a PSE-specific sentiment library, including context specific keywords such as “hung up”, “volatile”, and “throw”. SA is a useful technique that, when used in combination with other indicators, may paint a picture of an institution's safety culture.