More Than a Gut Feeling: Large Language Model Thematic Analysis of Social Media Posts from Inflammatory Bowel Disease Patient Communities - 15/06/26

Doi : 10.1016/j.clinre.2026.102870

Roy Tal Cohen ^1,^⁎ , Elad Yom Tov ², Daniel Shepshelovich ³

¹ Gray Faculty of Medicine, Tel Aviv University, Tel Aviv, Israel

² Department of Computer Sciences, Faculty of Exact Sciences, Bar Ilan University, Ramat Gan, Israel

³ Department of Medicine, Columbia University Irving Medical Center, New York, NY, USA

^⁎ Corresponding author: Roy Tal Cohen, Division of Gastroenterology, Edith Wolfson Hospital, str. Ha-Lokhamim Street 62, Holon, Israel Division of Gastroenterology Edith Wolfson Hospital str. Ha-Lokhamim Street 62 Holon Israel

Sous presse. Manuscrit accepté. Disponible en ligne depuis le Monday 15 June 2026

Highlights

•	Large-Scale LLM Analysis : Utilized a Large Language Model (GPT-4) to conduct a thematic analysis of over 104,000 authentic patient posts from IBD Reddit communities over a 10-year period.
•	Psychosocial Dominance : Patient discourse is primarily driven by the disease's psychosocial and emotional burden rather than traditional gastrointestinal morbidity.
•	Engagement Drivers : While " Diagnosis struggles " was the most prevalent theme, psychosocial topics like " Relationships " and " Humor & Embarrassment " generated the highest levels of community engagement.
•	Evolving Disease Narrative : A 50% decrease in surgery-related discussions over the past decade reflects a shift in the IBD experience from an acute surgical disease to a chronic psychosocial condition.
•	Demographic Nuances : Significant variations exist in patient concerns across demographics, with female users focusing more on mental health and male users focusing more on medication and surgery.

Le texte complet de cet article est disponible en PDF.

Abstract

Background

: Inflammatory bowel disease (IBD) morbidity extends beyond gastrointestinal symptoms and complications. We aimed to describe IBD patient experience through an analysis of online patient communities.

Methods

: Thematic analysis of 104,123 posts from three IBD patient groups on the social media platform Reddit between 2015-2024 was undertaken using a large language model (LLM) based on the GPT-4 architecture. The Valence Aware Dictionary and Sentiment Reasoner (VADER) tool was used to assess sentiment analysis.

Results

: Of 15 core themes identified by LLM analysis, the most prevalent were 'Diagnosis struggles' (13.1%), 'Peer advice' (12.5%), and 'Mental health & Coping' (11.8%). Results were validated using a natural language processing model and human chart review, with concordance of 88.3% and 87.0%, respectively. The themes that generated the most community engagement were 'Relationships' (15.3 mean comments), 'Humor & Embarrassment' (12.8), and 'Hope & Remission' (12.6). ‘Fatigue & Pain’ was associated with the worst sentiment (-0.44), and 'Hope & Remission' with the best one (+0.43). Posts by self-identified females were more likely to focus on 'Mental Health & Coping' (11.7% vs 10.0%, p=0.003), while self-identified males were more likely to post on 'Medication Experiences' (13.3% vs 11.7%, p=0.003) and 'Surgery' (2.4% vs 1.7%, p=0.003). Over the years, 'Surgery' posts decreased by 50%, while 'Fear of the Future' and 'Diagnosis Struggles' posts increased by 23% and 21%, respectively.

Conclusions

: Patient posting and engagement focus on psychosocial and emotional disease burden, rather than on gastrointestinal morbidity. These results can inform healthcare professionals caring for patients with IBD.

Le texte complet de cet article est disponible en PDF.

Keywords : N/A