OpenAI o3 model aids rare disease diagnosis at Boston Children’s Hospital

A study published in NEJM AI reports that an OpenAI language model, identified as o3, was used to analyze genome sequences and clinical data for 376 patients with undiagnosed rare diseases at Boston Children’s Hospital. The AI suggested diagnoses for 18 children whose conditions had previously remained unresolved, uncovering new diagnoses in about 5% of cases, particularly among neurodevelopmental and neuromuscular disorders. All AI-generated suggestions were reviewed and confirmed by physicians before being accepted as final diagnoses.

The research highlights the tool’s role as a decision‑support aid that can rapidly cross‑reference genetic variants, symptoms and medical notes, accelerating the diagnostic process for complex genetic conditions. Researchers stress that the system complements, rather than replaces, clinical expertise and that privacy and regulatory considerations remain important for broader clinical adoption.