Harnessing AI and social media to understand real-world patient experiences in SLE

To apply large language models (LLMs) to Reddit posts referencing SLE to identify patient-expressed unmet medical needs, symptom experiences and healthcare challenges, demonstrating how artificial intelligence-enabled social media listening complements traditional patient-experience research.

We extracted 4633 posts from 10 SLE-related Reddit communities using the public Reddit Application Programming Interface (October-November 2025). After removing duplicates, promotional content and posts with insufficient information, 2603 posts remained. We developed a thematic codebook through manual review of 300 posts and refined it iteratively. Two state-of-the-art LLMs (Gemini 3.0 and GPT-5.2) were evaluated for automated thematic labelling using per cent agreement, Cohen's κ and performance against a human-annotated reference set (n=100). The best-performing model was then used to quantify theme prevalence across the full dataset, followed by qualitative review of representative narratives.

GPT-5.2 demonstrated higher performance (F1=0.844) than Gemini 3.0 (F1=0.811), with substantial intermodel agreement across main themes (mean κ=0.71). Posts reflected multidimensional experiences. The most frequent subtheme was Advice Seeking (84.1%), followed by Emotional Coping (55.6%). Common symptom-related themes included Pain (37.2%), Other Symptom Presentations (37.6%), Fatigue (24.7%) and Acute or Worsening Flares (30.2%). Diagnostic uncertainty was prominent, including testing and results confusion (24.0%) and emotional impact of uncertainty and disease progression (33.0%). Qualitative review highlighted emotional and social burden, reliance on peer communities for advice and shared experiences, and difficulty managing complex treatment regimens.

LLM-enabled social media listening offers a scalable method for synthesising large volumes of unstructured patient narratives, providing timely insights into lived experiences and unmet needs among individuals discussing lupus online. Findings align with established qualitative literature while highlighting persistent gaps in patient education, communication and care coordination. This analytical framework can be extended to other disease areas to support patient-centred care, measurement development and evidence generation relevant to therapeutic and health outcomes research.