Listening to Women's Comments: Text Mining Will Revolutionize How We Organize and Act on Women's Concerns [23G

Abstract

INTRODUCTION: Listening and acting upon women's comments is central to the relationship we have with patients. Doing this systematically and scientifically is difficult and sometimes involves thematic analyses which are time consuming and costly to perform. We present a new way of analyzing thousands of comments and seeing how each comment affects the HCAHPs rate hospital score (RHS). METHODS: We utilized text mining to analyze 27,335 free text comments. Comments were provided by women from Jan 2008 to June 2014 as supplemental comments to the HCAPHs survey. We used SAS text miner (v13.1). Similar words were combined. We then calculated the frequency of each word and the frequency of its relationships to other words using an expectation maximization algorithm. This produced a set of most frequently linked words which are statistically significantly related to each other. We then performed multiple regression analyses to determine magnitude of positive or negative effect on HCAHPs RHS. RESULTS: Negative comments have a larger effect on rate hospital score than positive comments. Comments regarding midwives had the highest positive effect on RHS. Topics generated were helpfulness of staff, cleanliness, feeding, pain control and postpartum care. All these areas had a statistically significant effect (p=0.05) on the RHS given by the person giving the comment. CONCLUSION: The systematic nature of this analysis is important. Our output is not subject to interpretation bias or comment reading fatigue. Whilst we present national level data the analysis could be performed on different subsets of data, providing results at regional or hospital level.