Kristen M Saad, MS1, Paige L Martinez, MS2, Larissa A Mcgarrity, PhD3, Ellen H Morrow, MD2, Eric T Volckmann, MD2, Juliana S Simonetti, MD4, Anna R Ibele, MD2. 1University of Utah School of Medicine, 2University of Utah Department of Surgery, 3University Hospital Rehabilitation Psychology, 4University of Utah Department of Internal Medicine
INTRODUCTION: Bariatric surgery is one of the best-evidenced treatments for obesity, a complex medical condition with both physiologic and psychosocial contributors. As such, it’s become standard of care at some institutions to elicit pre-surgical written personal statements regarding patient motivation for seeking bariatric surgery. While these statements are often used as a “jumping-off point” for clinical discourse, they also represent a rich data source for text analytics.
This pilot study (n=50) presents novel interdisciplinary methodological approaches and best practices for bariatric personal statement text mining to uncover and elucidate lexical and psychosocial patterns in presurgical patients.
METHODS AND PROCEDURES: A proof-of-concept dataset of pre-surgical personal statements and associated demographic factors was constructed using an IRB-approved database, and subsequently analyzed with R. The dataset was cleansed, normalized, and descriptive statistical techniques applied to better understand the population of statements. The statements were then analyzed for word frequency, topic affiliations, and sentiment to better understand underlying patterns in patient motivation and psychological state.
RESULTS: Word clouds, comparative graphs, and sentiment charts are used to effectively communicate the results of text mining studies using the pilot dataset. Methods, lessons learned, and best practices for database design, data cleansing, descriptive analytics, advanced computational techniques, and result integration and operationalization will be presented.
CONCLUSIONS: By combining a wide range of computational methods including word counts, lexical diversity, collaborative filtering, and sentiment analysis, underlying text patterns can be detected and analyzed for clinical use as well as potential association with post-surgical and long-term outcomes. While bariatric survey statements have been evaluated and subjectively scored by teams of researchers, this is the first known investigation utilizing computational text analytics to evaluate statements elicited as standard-of-care. This project represents a step forward in the field as it leverages a novel application of interdisciplinary techniques to enable clinicians to understand patient motivations and population patterns using patients’ own words.
Presented at the SAGES 2017 Annual Meeting in Houston, TX.
Abstract ID: 95195
Program Number: P131
Presentation Session: Poster Session (Non CME)
Presentation Type: Poster