Using AI to help physicians diagnose rare genetic diseases affecting children

What happened

OpenAI and clinical experts published a June 18, 2026 NEJM AI study where an o3 Deep Research workflow generated hypotheses linking phenotypes, variants, and literature. After ACMG/AMP review and CLIA confirmation, diagnoses were established in 18 previously unsolved pediatric cases.

Context and impact

Pediatric rare-disease diagnostics has long struggled with knowledge accumulating faster than experts can reanalyze old cases. The study suggests LLM-assisted periodic reanalysis could be a scalable complement to the clinical process, not a replacement for the physician.

Details

• +4.8% additional diagnostic yield over expert analysis
• Diagnoses spanned neurodevelopmental, neuromuscular, sudden-death and early-psychosis groups
• The model never diagnosed any patient alone — only generated leads for human review
• Clinical confirmation through CLIA labs and ACMG/AMP rules