Comparison of Patient Reviews for Submental Liposuction and Kybella Using Deep Learning and Natural Language Processing: Is There a Superior Intervention for Submental Adiposity?

Abstract

Patient-reported outcomes (PROs) are essential in evaluating aesthetic procedures. RealSelf is an online platform for cosmetic and reconstructive feedback. Currently, no validated PRO tools exist to compare Kybella (synthetic deoxycholic acid; AbbVie Pharmaceuticals, North Chicago, IL) with submental liposuction. Robust optimizing Bidirectional Encoder Representation from Transformers, a deep learning natural language processing (NLP) model, can analyze emotions and sentiment in text and was used compare patient-reported experiences for Kybella and submental liposuction on RealSelf. RealSelf was queried for English reviews of Kybella and submental liposuction from 2014 to 2024. Reviews were excluded if they lacked procedural details, focused on the provider, or included adjunct procedures such as a necklift or Botox. Two NLP models analyzed posts: 1 characterized posts as positive or negative (scored 0-1); The latter scored each post from 0 to 1 ∼7 emotions-fear, sadness, anger, disgust, neutral, surprise, and joy. Overall, 1338 posts met the inclusion criteria: 753 for submental liposuction and 585 for Kybella. A greater proportion of liposuction reviews were positive compared with Kybella (83.4% vs 55.2%, P < .00001). Liposuction also had significantly higher mean positive scores (0.73 vs 0.41, P < .0001). Joy was the most common emotion for liposuction reviews (51.3%), whereas Kybella reviews more often reflected sadness (17.9%) and fear (12%), with fewer expressing joy (22.7%). The authors of this study applied NLP to extract structured sentiment from unstructured reviews RealSelf. Our findings suggest that submental liposuction patients report greater satisfaction and more positive emotional experiences than those who receive Kybella.