Federal Government Bets $35 Million on AI to End Diagnostic Odysseys for 350 Million Rare Disease Patients

The Advanced Research Projects Agency for Health (ARPA-H) has awarded up to $35 million to the University of North Carolina School of Medicine and Emory University to address diagnostic delays affecting an estimated 350 million people worldwide living with rare diseases. The four-and-a-half-year initiative aims to create the world's largest rare disease data resource and deploy AI-powered clinical decision support tools that could transform how non-specialist physicians identify these conditions.

Addressing the Rare Disease Diagnostic Gap

Despite affecting as many as 1 in 10 Americans across more than 10,000 rare disease types, these conditions remain difficult to diagnose in typical healthcare settings. The project, funded through ARPA-H's Rare Disease AI/ML for Precision Integrated Diagnostics (RAPID) program, will focus on approximately 2,700 of the more prevalent and potentially treatable rare diseases—conditions where earlier diagnosis can lead to actionable interventions.

Led by Melissa Haendel, PhD, FACMI, Sarah Graham Kenan Distinguished Professor in the Department of Genetics at UNC School of Medicine, and Richard Moffitt, PhD, Associate Professor in the Department of Hematology and Medical Oncology at Emory University School of Medicine, the initiative will aggregate data from health records, insurance claims, medical imaging, video technologies, patient surveys, patient registries, and other real-world sources. The data will be de-identified with tiered secure access—some datasets publicly available, others requiring data use agreements—to protect patient privacy while enabling research applications.

Building Infrastructure for AI Diagnostics

The consortium includes academic institutions such as Johns Hopkins University, Queen Mary University of London, University of California at San Francisco, and University of Iowa, alongside industry partners including Across Healthcare, Combined Brain, DartNet, Datavant, Mendelian.co, and Truveta. Patient advocacy organizations including EB Research, Global Genes, and the National Organization for Rare Diseases (NORD) are participating, as are technology companies OpenAI, Anthropic, Amazon Web Services, and Google.

The primary objective is to train AI models on these datasets to create clinical decision support tools deployable in less specialized diagnostic settings. By enabling non-specialist clinicians to access AI-powered diagnostic assistance, the project could reduce the multi-year diagnostic delays that rare disease patients currently endure, potentially preventing inappropriate treatments and disease progression.

Implications Beyond Diagnosis

The initiative addresses challenges extending beyond individual patient care. For pharmaceutical companies, identifying eligible patients for rare disease clinical trials represents a significant bottleneck in drug development. The project's patient identification algorithms could accelerate clinical trial enrollment and expand the viability of rare disease therapeutics by improving recruitment efficiency.

Qualified researchers and companies will be able to apply for tiered data access to build and validate their own algorithms, positioning the resource as foundational infrastructure for the rare disease AI ecosystem. The NIH's All of Us Center for Linkage and Acquisition of Data and the Monarch Initiative, both co-led by Haendel, are also involved in the effort, suggesting alignment with broader federal precision medicine initiatives.

By standardizing rare disease data at scale and making it accessible to the AI development community, this ARPA-H investment could catalyze development of diagnostic tools that bring specialized medical knowledge to diverse healthcare settings—potentially reducing the diagnostic delays that have long characterized the rare disease patient experience.

Sources

1. UNC Awarded Up to $35 Million to Lead Landmark Initiative to Build World's Largest Data Resource for Rare Disease AI — Newswise