Quantifying Patient Sentiment of Treatments for Lateral Elbow Tendinopathy
Beau McGinley, MBA, Daniel Whittingslow, PhD, Paul Ghareeb, MD, Joseph Lamplot, MD, Bowers L Robert, DO PhD, Charles Daly, MD, Eric R Wagner, MD MScR and Michael Gottschalk, MD, Emory University School Medicine, Atlanta, GA
Introduction: Sentiment analysis allows researchers to quantify public opinion on any topic. In the management of Lateral Elbow Tendinopathy (LET), formerly known as lateral epicondylitis, no clear best practices exist following failure of conservative management. In order to better understand patient perceptions and preferences in the setting of clinical ambiguity, we performed a sentiment analysis of tweets regarding the various treatments of LET. We hypothesized that newer, more expensive treatments would be portrayed more positively compared to more evidence-based, conservative therapies.
Methods: A computer program was written to search Twitter and extract data on all posts with specific key words The data collected from each tweet included the username, body of the tweet, and number of likes, comments, and retweets. A proof-of-concept test was performed before analyzing tweets involving LET with conservative management, platelet rich plasma (PRP), steroids, or surgery. Advertisements were filtered and removed.
Each tweet was input into multiple natural language processing (NLP) programs. The performance of each NLP program was assessed using the aforementioned proof-of-concept dataset to determine which NLP would work best for our LET dataset. We then analyzed the sentiment of the tweet dataset related to LET. Based on their key words, tweets were divided into four groups: conservative management, PRP, surgery, or steroids. Mean sentiment scores were calculated and statistically compared for each group.
Kolmogorov-Smirnov-Lilliefors test of normality was performed. Mean sentiment scores for tweets containing each key word were compared using the nonparametric Kruskal-Wallis one way test of variance with Tukey's Post Hoc Analysis.
Results: In the proof-of-concept analysis, one NLP program (NLTK VADER) successfully differentiated groups of negative, neutral, and positive tweets. Using the same NLP program, A total of 10,602 tweets were analyzed related to LTE. Mean compound scores were significantly different (p<0.05) between each group, descending from PRP (0.22) to conservative (0.06) to surgery (-0.09) to steroids (-0.17). The conservative management group had the highest number of likes, retweets, and comments per tweet on average.
Conclusions: For the management of LET, social media users portray PRP most positively, followed by conservative management, surgery, and steroids in descending order. Further, PRP and conservative therapy have positive sentiment on average, while surgery and steroids were negatively portrayed by twitter users. Sentiment analyses are an evolving tool with the potential to inform healthcare providers about patient's preconceived perceptions of treatment modalities as well as track their satisfaction post-intervention. Level of Evidence IV.
Back to 2022 ePosters