Harnessing Patient-Generated Data for Rare Disease Knowledge Enrichment: A Pilot Study

Patients with rare diseases often endure a 4-5-year diagnostic odyssey. Patient-generated data on social media remains an untapped resource for understanding rare diseases. This study characterizes patient descriptions of rare disease from social media and systematically compares patient-reported symptoms to literature-documented ones. 166 unique rare diseases were identified from 229 patient posts. A hybrid human-AI framework using GPT-4 with full manual verification was used to extract and standardize patient-reported symptoms, which were then compared against symptoms documented in PubMed. The most frequent patient-reported symptoms were motor impairment, ataxia, pain, and fatigue. A low average overlap was observed between patient-reported symptoms and clinical literature. Patient narratives emphasized quality-of-life impacts (e.g., fatigue, anxiety), whereas clinical literature focused on diagnostic markers (e.g., dysarthria, dysphagia). Patient-generated data reveals underreported symptoms that are particularly meaningful to patients and complement clinical findings. Harnessing these real-world insights holds immense potential to shorten the diagnostic odyssey, inform patient-centered care, and guide future research priorities for rare diseases.