Researchers from Boston Children’s Hospital, Harvard University, and OpenAI used the OpenAI o3 Deep Research model to analyze 376 previously unsolved cases of rare genetic diseases. The model provided evidence-linked candidate explanations, which, after expert review and clinical confirmation, led to 18 new diagnoses. This study, published on June 18, 2026, in NEJM AI, highlights how AI can assist experts in revisiting difficult cases.
The model was trained to connect clinical features, inheritance patterns, variant evidence, and scientific literature into justifications for human review. Researchers used the same ACMG/AMP framework that clinical labs use to classify genetic variants. Each candidate diagnosis required at least two team members to review, with disagreements resolved by consensus. A finding was only considered a diagnosis after qualified experts reviewed the evidence, the variant was classified as pathogenic or likely pathogenic, and a CLIA-certified laboratory confirmed it.
The study showed that even old cases can yield new diagnoses as scientific knowledge evolves. Researchers noted that the model did not make clinical decisions but instead generated hypotheses for specialists to investigate. The model’s self-reported confidence scores correlated with correct diagnoses, though they were not used as a substitute for clinical judgment. The workflow was first tested on cases with established diagnoses, achieving high accuracy in recovering correct gene and variant information.
Source: openai
