Using AI to help physicians diagnose rare genetic diseases affecting children

Even with genomic sequencing, many rare disease patients remain undiagnosed. Often, their medical data contains clues, but these are difficult to find amidst vast amounts of genetic variants, fragmented records, and evolving scientific literature. This challenge is compounded by the difficulty of linking records across different databases and the continuous advancement of scientific knowledge, making previously inconclusive genetic tests potentially interpretable.

Researchers from Boston Children's Hospital, Harvard University, and OpenAI utilized the OpenAI o3 Deep Research reasoning model to analyze de-identified clinical and genomic information from 376 previously unsolved cases. The AI model surfaced evidence-linked candidate explanations, which, after expert review, additional testing, and clinical confirmation, led to 18 new diagnoses. This represents an additional diagnostic yield of 4.8% after initial specialist analysis.

The study, published in NEJM AI, highlights how an AI-assisted research workflow can help medical experts generate leads when re-examining complex cases. The OpenAI model did not make diagnoses or clinical decisions directly; instead, it provided evidence-linked hypotheses for specialists to review and investigate further. This suggests that expert-led periodic reanalysis could become more scalable as medical knowledge evolves, offering a new pathway to solving long-standing diagnostic mysteries.