Researchers at Boston Children’s Hospital and Harvard University say OpenAI’s o3 model helped diagnose 18 children with rare diseases that had gone unidentified for years. The study, published in NEJM AI, has fueled interest in how advanced artificial intelligence could support difficult medical cases.
The research reexamined 376 complex pediatric cases in which advanced genetic tests and expert opinions had still not produced a clear answer. By combining genetic data, clinical symptoms and recent medical literature, the model found patterns and suggested diagnoses across a range of conditions, including neurodevelopmental disorders, rare neuromuscular diseases and even cases of sudden death in children.
The researchers stressed that AI is not replacing doctors but acting as a decision-support tool that expands human analysis. Every diagnosis suggested by the system was reviewed and confirmed by medical teams. The project is part of a broader effort to integrate AI into Boston Children’s Hospital’s health system.
Since the partnership with OpenAI began in early 2025, backed by about $50 million in investment, the researchers say they have reached more than 40 new diagnoses for diseases once considered unsolvable. The hospital also says more than a third of its staff now use AI tools daily, saving about 60,000 work hours, valued at more than $7 million. The findings matter because around 300 million people worldwide live with rare diseases, often after years of frustrating diagnostic searches. Dr. John Brownstein, the hospital’s chief innovation officer, said, “What was once unimaginable is becoming reality that gives hope to many families.”
