<< Back to the abstract archive
Understand and Define Breast Implant Illness: A Novel Approach Using Natural Language Processing Technology and Social Media
Krasniak PJ, Nguyen MH, Dey V, Ning X, Lee CN
Ohio State University
2020-02-15
Presenter: Peter Krasniak
Affidavit:
I certify that this abstract has not been published in any scientific journal or previously presented at a major meeting.
Director Name: Gregory Pearson
Author Category: Resident Plastic Surgery
Presentation Category: Clinical
Abstract Category: Aesthetics
PURPOSE: Public alarm is growing over a constellation of symptoms called "breast implant illness" (BII), described mostly on social media. The FDA has called for definitive research on the condition, but a satisfactory disease definition does not yet exist, partially due to a perceived stigma decreasing patient communication with their plastic surgeons.
METHOD: We derived a dataset from social media posts under the Instagram hashtag "#Breastimplantillness". Data processing consisted of extracting medical terms from the text, with grouping of similar terms under one unified term and assigned code (Concept Unique Identifier - CUI). Topic modeling programming called Latent Dirichlet Allocation (LDA) sought connections between these CUI codes based on probabilistic modeling of word distribution across the entire dataset to highlight topics hidden within the data. We then identified topics focused on symptoms to report.
RESULTS: A total of 29,011 posts with 3,116,966 words were in the dataset. The frequencies of most common mentions related to illness and symptoms were: pain (3349 mentions), autoimmune (1,609), depression (807), tired and fatigue (637), vision issues (414), ALCL (374), anxiety (370), swelling (343), and infections (337). The analysis also showed that "implant removal" and "capsulectomy" frequently appeared in the posts (313 and 1,421 times respectively).
CONCLUSIONS: We successfully used social media as a massive patient-reported dataset to identify most commonly mentioned symptoms related to BII. This patient-derived definition provides a starting point for future studies investigating incidence and etiology and enables a framework for in-office discussions using patient-centered language.
