OpenAI o3 Helps Diagnose 18 Rare Pediatric Disease Cases

On June 18, OpenAI announced results from a collaboration with Boston Children's Hospital and Harvard Medical School demonstrating that its o3 reasoning model can help physicians diagnose rare genetic diseases in children. The study reanalyzed 376 previously unresolved pediatric cases where patients had undergone genomic sequencing but remained undiagnosed despite extensive specialist review.

The o3 Deep Research model worked by integrating patient clinical symptoms, identified genetic variants, and the published medical literature to generate diagnostic hypotheses. In roughly half of these cases, families had spent years — sometimes over a decade — searching for answers. The AI-driven approach identified 18 new diagnoses that clinicians subsequently confirmed using ACMG/AMP variant classification rules and CLIA-certified laboratory testing.

This matters because roughly half of all rare disease patients never receive a definitive genetic diagnosis, even after extensive testing and multi-specialist review. The economic and emotional costs are enormous — the National Institutes of Health estimates the diagnostic odyssey averages 4-5 years and costs families tens of thousands of dollars.

The clinical workflow is not autonomous diagnosis — clinicians review every AI-generated hypothesis before any diagnostic decision. But the AI dramatically narrows the search space, surfacing connections between genetic variants and rare phenotypes that even expert teams can miss given the exponential growth in genomic literature.

For healthcare systems in MENA where rare disease expertise is concentrated in a handful of centers, AI-assisted genomic interpretation could be especially transformative — extending specialist-level diagnostic capability to regions with limited access to medical genetics.