Presentation (CTI)

OBJECTIVES: China’s rare disease diagnosis is hindered by long delays, limited primary care expertise, and specialist shortages outside major cities. This study assesses the impact of AI diagnostic tools—disease-specific LLMs—on rare disease care in China using real-world adoption and deployment data.
METHODS: Two data sources were synthesized. First, the 2025 Digital Life Physician (DLP) cross-sectional survey (n=4,500 Chinese physicians) provided real-world adoption metrics for AI tools in rare disease clinical practice, including usage rates, functional preferences, and perceived clinical value. Second, a documented deployment case of a rare disease-specific LLM built on a national rare disease knowledge base (PUMCH Xiehe·Taichu model) was analyzed for clinical architecture, hallucination mitigation mechanisms, and reported outcome metrics including diagnostic timeline compression, differential diagnosis accuracy, and hospital integration pathways across 200+ partner institutions.
RESULTS: Among AI-using physicians (n=4,246), rare disease and cognition-intensive specialties demonstrated the highest AI adoption (gastroenterology 98%, cardiology 95%, neurology 94%). Disease-specific LLMs incorporating explainable symptom-to-differential-diagnosis reasoning chains, multi-dimensional traceable verification, and extreme few-shot training on curated rare disease case repositories reduced average diagnostic timelines from over four years to under four weeks in documented deployments. Platform integration via SaaS or on-premises deployment in 200+ hospitals enabled patient triage, clinical decision support, and rare disease knowledge retrieval at scale. Hallucination rates were maintained below 5% through cross-referencing with authoritative medical literature and confidence scoring.
CONCLUSIONS: Disease-specific LLMs represent a structurally superior approach to rare disease physician support compared to general-purpose AI, demonstrating clinically meaningful reductions in diagnostic delay with compliance-ready traceability. For HEOR frameworks, these tools introduce a new evidence category — AI-enabled diagnostic equity — where value accrues from geographic reach, time-to-diagnosis compression, and standardization of clinical reasoning across tiered hospital systems. Payers and policymakers should incorporate AI-assisted diagnostic efficiency into rare disease access and reimbursement evaluation models.