Sentiment Analysis of Autologous Breast Reconstruction Using Natural Language Processing and Deep Learning.

Area of Science:

• Plastic Surgery
• Natural Language Processing
• Patient-Reported Outcomes

Background:

• Autologous flaps are increasingly used for postmastectomy reconstruction.
• Public forums provide extensive patient narratives suitable for Natural Language Processing (NLP) analysis.
• RoBERTa models can analyze patient-generated text data.

Purpose of the Study:

• To evaluate patient sentiment and emotions regarding autologous reconstruction using deep inferior epigastric perforator (DIEP), transverse rectus abdominis myocutaneous (TRAM), and latissimus dorsi (LD) flaps.
• To compare patient-reported experiences across different flap types.

Main Methods:

• Collected patient reviews from RealSelf referencing DIEP, TRAM, or LD flaps.
• Applied two NLP models: a binary sentiment classifier (positive/negative) and an emotion classifier (fear, sadness, anger, disgust, neutral, surprise, joy).
• Labeled each review with its highest-scoring emotion.

Main Results:

• Analyzed 212 posts: 153 DIEP, 20 TRAM, 39 LD.
• DIEP flaps showed the highest mean positive sentiment (0.627), while TRAM flaps had the lowest (0.517).
• Joy was the most frequent emotion for DIEP (68/153) and TRAM (7/20) reconstructions. LD flap reviews were most often "neutral" (9/39).

Conclusions:

• NLP analysis of patient reviews indicates DIEP flaps are most discussed and associated with greater positive sentiment than TRAM or LD.
• Emotion profiles reveal predominantly joyful narratives for DIEP, neutral for LD, and less positive sentiment for TRAM.
• This NLP approach captures large-scale, real-world patient experiences, complementing traditional outcomes research.