Boston Children's Hospital AI model o3 identifies 18 rare pediatric diagnoses

A study published in the New England Journal of Medicine’s AI journal (NEJM AI) reported that the OpenAI‑based o3 model helped resolve previously undiagnosed rare diseases in children. Researchers at Boston Children’s Hospital fed the model genomic sequences, clinical notes and patient records from 376 children whose conditions had resisted standard testing. The system generated diagnostic suggestions that were later confirmed by clinical geneticists, leading to new diagnoses in nearly 5 % of the cases – 18 children in total. The confirmed conditions included neurodevelopmental disorders, neuromuscular diseases, early‑onset psychosis and two post‑mortem diagnoses. The model is presented as a decision‑support tool, not a replacement for physicians, and all suggestions were validated with laboratory testing. One highlighted case involved Kyra Benton, who received a diagnosis of myofibrillar myopathy after two decades of uncertainty. The findings illustrate how AI can accelerate the interpretation of complex genomic data, offering families long‑awaited answers while raising questions about integration, safety and governance in routine clinical practice.