Abstract
BACKGROUND: Use of autologous flaps for postmastectomy reconstruction has increased. Public forums offer large-scale patient narratives that can be analyzed with natural language processing (NLP) models, such as RoBERTa. OBJECTIVES: This study aims to assess patient-reported sentiment and emotions toward autologous reconstruction with deep inferior epigastric perforator (DIEP), transverse rectus abdominis myocutameous (TRAM), and latissimus dorsi (LD) flaps. METHODS: Public reviews from RealSelf (Seattle, WA) referencing DIEP, TRAM, or LD flaps were collected. Two NLP models were applied: (1) a binary sentiment classifier (positive/negative) and (2) an emotion model scoring fear, sadness, anger, disgust, neutral, surprise, and joy; each review was labeled by its highest-scoring emotion. RESULTS: Two hundred and twelve posts were analyzed: 153 DIEP, 20 TRAM, and 39 LD. DIEP had the highest mean positive sentiment (0.627); TRAM had the lowest (0.517). In the emotion model, "joy" was most common for DIEP (68/153; mean 0.389) and TRAM (7/20; mean 0.299). LD posts were most often "neutral" (9/39; mean 0.267). CONCLUSIONS: NLP of public patient reviews suggests DIEP flaps are the most discussed and are associated with more positive sentiment than TRAM or LD reconstructions. Emotion profiles indicate generally joyful narratives for DIEP, neutral for LD, and comparatively less positive for TRAM. This approach complements traditional outcomes research by capturing real-world patient experience at scale.