Text-Mining Framework for Hospital Patient Experience Surveys: A Case Study

摘要

The healthcare industry generates a significant amount of data on an ongoing basis which includes historical medical, imaging, lab results, physician assessments, demographics, regulatory, insurance claims, and others. In many cases, this massive pool of data is used in day-to-day operational and clinical decision making process in a non-systematic approach. This research introduces a framework that aims to analyze a set of available free-text patient experience comments, using text mining techniques in order to 1) detect latent and hidden themes within collection of comments through topic modeling process, an unsupervised learning approach, 2) generate vector of features using topic distributions of topic modeling process, and 3) convert topic modeling outputs into a supervised learning problem to predict Top-Box and Non-Top-Box comments. In this framework, topic modeling acts as the feature reduction step for the supervised learning step, and the overall goal is to understand if topic distributions are robust enough to predict what comments generate negative or positive ratings. By analyzing more than four years of received comments from patients, important topics were extracted from textual data that was validated by subject matter experts in terms of consistency with practical understanding of patient experience. The proposed text mining framework provides the leadership with ideas about improvement opportunities in the hospital, while saving hundreds of hours of labor that would have been required for manual inspection and analysis of untapped textual feedbacks.