AI-Assisted Diagnosis Support for Rare Pediatric Genetic Diseases

About half of rare disease patients fail to receive an accurate genetic diagnosis even after extensive testing. This is because it is difficult to review vast genetic variant data, fragmented clinical records, and rapidly evolving medical literature all at once.

Researchers from Boston Children's Hospital, Harvard University, and OpenAI used the OpenAI o3 Deep Research model to re-analyze 376 cases that had previously gone unsolved. The model analyzed clinical information and genomic data to present evidence-based candidate explanations.

After expert review and additional testing, new diagnoses were confirmed in a total of 18 cases, representing an additional diagnostic yield of 4.8% compared to prior specialist analysis. Rather than issuing direct diagnoses, the model served as an explanation-first reasoning layer, connecting clinical features, genetic patterns, variant evidence, and literature for experts to review.

This research shows that as scientific knowledge evolves, previously inconclusive test results may hold new answers. An AI-assisted periodic re-analysis workflow suggests the potential to scale expert-centered diagnostic processes more efficiently.