Skip to main content
Physician using AI tool analyzing genetic test results for diagnosing rare pediatric genetic diseases, improving accuracy by

Editorial illustration for AI helps physicians diagnose rare pediatric genetic diseases, 4.8% rate

AI helps physicians diagnose rare pediatric genetic...

AI helps physicians diagnose rare pediatric genetic diseases, 4.8% rate

2 min read

Why does this matter? Rare pediatric genetic disorders often slip through even the most thorough genomic workups. Roughly half of affected children remain without a clear diagnosis after extensive testing and specialist review, their records scattered across thousands of possible variants and an ever‑growing body of literature.

While the problem is well known, a new study published June 18, 2026, in NEJM AI offers a concrete data point. Researchers from Boston Children’s Hospital’s Manton Center for Orphan Disease Research, Harvard University, and OpenAI applied OpenAI’s o3 Deep Research reasoning model to de‑identified clinical and genomic data from 376 cases that had previously been deemed unsolvable. The AI surfaced evidence‑linked candidate explanations, which clinicians then vetted through additional testing and standard clinical confirmation.

The result? Diagnoses in 18 patients—a 4.8 % diagnostic yield on top of prior specialist analysis. The workflow suggests that periodic, AI‑assisted reanalysis could become a scalable complement to existing expert review, especially as new gene‑disease links emerge.

After the model surfaced candidates and experts completed review and clinical confirmation, physicians established diagnoses in 4.8% of the cases. That rate is modest but meaningful in this population because previous expert reviews had not resolved the cases. Similar reanalysis studies report single-digit gains in heavily reviewed cases; higher yields usually come from studies containing new cases or well-known disorders awaiting genetic confirmation.

Of the 18 diagnoses, 7 were rediscoveries: diagnoses established outside the local research workflow but absent from the record the team reviewed. In several cases, the variants were already listed as pathogenic or likely pathogenic in public databases, highlighting the operational challenge of synthesizing information across data sources.

Why this matters

The NEJM study shows an OpenAI reasoning model can pull diagnostic leads from 376 unsolved pediatric cases, ultimately confirming 18 new diagnoses. That translates to a 4.8% hit rate—modest, yet it represents outcomes that prior expert review missed. For developers, the result underscores that even a small lift in yield may justify building systems that can comb through thousands or millions of data points, something human reviewers cannot feasibly do.

However, the modest percentage also reminds us that AI assistance is not a substitute for clinical expertise; the model’s suggestions still required expert validation and confirmation. Researchers should note that the study does not reveal how the model performed across different disease categories or whether the approach scales to larger cohorts, leaving open questions about broader applicability. Founders might see a niche for tools that streamline reanalysis pipelines, but they must temper expectations with the reality that diagnostic impact remains limited and contingent on rigorous integration into clinical workflows.

We remain cautious.

Further Reading