OpenAI-assisted review identifies 18 pediatric diagnoses

According to OpenAI and the NEJM AI publication dated June 18, 2026, researchers from Boston Children's Hospital's Manton Center for Orphan Disease Research, Harvard University, and OpenAI reanalyzed 376 previously unsolved pediatric cases using OpenAI's o3 Deep Research reasoning model. Per OpenAI and the NEJM AI paper, the model produced evidence-linked candidate explanations that clinicians reviewed; after additional testing and clinical confirmation, physicians established diagnoses in 18 cases, representing an incremental diagnostic yield of 4.8%. OpenAI states the model did not make diagnoses or clinical decisions.

OpenAI describes the workflow as an AI-assisted research pipeline that ingests de-identified clinical and genomic information, surfaces literature- and evidence-linked variant hypotheses, and produces leads for expert review. The NEJM AI study emphasizes that advances in gene-disease knowledge, case reports, and variant classification can render previously inconclusive tests interpretable on reanalysis, and that linking fragmented clinical records is a practical barrier to interpretation, per OpenAI's summary of the study.

Editorial analysis: For practitioners, the study demonstrates a repeatable example where large reasoning models can accelerate hypothesis generation in complex genomic cases without replacing clinician judgment. The reported 4.8% uplift is concrete evidence that automated reanalysis can add measurable value in cohorts of long-unsolved cases, while still requiring laboratory confirmation and expert adjudication.

Editorial analysis: The reported results are retrospective and depend on expert review and confirmatory testing; they do not establish prospective clinical performance, false-positive rates, or cost-effectiveness in routine practice. OpenAI's public post explicitly states the model produced hypotheses for specialists rather than delivering autonomous diagnoses.

Industry context: Observers should follow independent replication at other centers, prospective validation studies that measure sensitivity and specificity, integration approaches for secure clinical data workflows, and guidance from regulators and professional genetics societies on AI-assisted diagnostics. Media framing as a "total game changer" (New York Post) highlights patient-facing impact, but technical and regulatory work remains before broad clinical adoption.