Abstract 4371038: Exploring Real-World Barriers to Statin Use Through Reddit Conversations Using LLM with Expert-in-the-Loop

Introduction: Statin nonadherence remains a challenge—over 50% discontinue within one year. Clinical barriers include side effects and poor communication, but patient perspectives on social media are underexplored. Hypothesis: Analyzing Reddit posts via LLM with expert-in-the-loop will reveal concerns and barriers missing from clinical data. Methods: We collected Reddit submissions/comments containing 17 statin-related keywords (“statin,” “atorvastatin,” “rosuvastatin,” “simvastatin,” etc.) from January 1, 2022 to May 1, 2025 across 438 subreddits. After text cleaning (concatenating title/body; removing URLs, markdown, and non-ASCII characters), deduplication, bot filtering, and a personal-experience filter, 5,618 documents remained. Metadata enrichment added upvote and engagement metrics. We randomly sampled 1,000 documents for detailed GPT-4.1-based analysis. Two rounds of expert-in-the-loop prompt refinement (20 documents per round) optimized extraction of twelve domains: primary theme; specific statins mentioned; sentiment toward statins, physicians, and pharmaceutical companies; reported side effects; decision factors; and adherence issues. Results: Among the 1,000 sampled documents, 43% expressed negative sentiment versus 26% positive and 31% neutral/mixed. Side effect concerns (19%) and doctor–patient communication issues (17%) were the most frequent themes. Reported side effects included vision changes, dizziness, muscle pain, and cognitive dysfunction. Nearly 25% of patients described feeling rushed by providers or inadequately informed about side effect risks. Alternative therapies—such as supplements and dietary interventions—were discussed in 14% of documents. Rosuvastatin (12%) and atorvastatin (9%) were the most frequently mentioned statins. Conclusions: Reddit discussions expose gaps between clinical guidelines and patient experiences. The predominance of negative sentiment and emphasis on side effects and communication breakdowns suggest that clinical approaches inadequately address patient needs. Enhanced patient–provider dialogue, proactive side effect management, and shared decision-making strategies could improve adherence. LLM-based social media analysis with expert-in-the-loop refinement provides real-world insights that complement traditional adherence research and can inform targeted interventions to reduce statin discontinuation.