OpenAI Model Helps Diagnose 18 Rare Diseases in Children

New diagnoses for unsolved cases

Researchers at Boston Children's Hospital have used OpenAI's o3 model to diagnose 18 children with rare diseases whose genetic causes had remained unknown despite previous analysis. The findings, published Thursday in NEJM AI, demonstrate how commercially available AI tools can help clinicians identify disease-causing genetic variants in complex cases.

The research team from the hospital's Manton Center for Orphan Disease Research analyzed genomes from 376 patients who lacked diagnoses across four disease categories. The AI system identified new diagnoses for 10 patients with rare neurodevelopmental diseases, four with neuromuscular disorders, two who had died suddenly, and two with early childhood psychosis.

"It got almost 5% new diagnoses, which doesn't sound like a lot," said Catherine Brownstein, scientific director of the genetic investigations arm at the Manton Center. "But considering how many times these had already been analyzed, that's a huge number, and each one means an answer for a family."

How the system works

The research team provided the o3 model with clinical notes, symptom descriptions, and a filtered list of potentially relevant genes for each case. Human researchers then reviewed all AI outputs before making final diagnoses.

The challenge in genetic diagnosis stems from the complexity of the human genome, which contains roughly 20,000 protein-coding genes. While sequencing a full genome is now routine, connecting specific genetic abnormalities to diseases requires synthesizing scattered research findings that may have been published after a patient's initial evaluation.

"A researcher can only spend so much time on a single case," said Suyash Shringarpure, a technical researcher at OpenAI focused on health applications. "Maybe a case remained unsolved when it came to them first, but a year later a paper was published that clarifies the link between the gene and the disease."

Real impact for patients

Kyra Benton received one of the breakthrough diagnoses. After developing mobility problems at age nine and facing years of worsening health including severe heart issues and a tracheotomy at 13, she had resigned herself to never knowing her diagnosis. Last summer, just before her 20th birthday, researchers informed her that the AI analysis had identified myofibrillar myopathy, a progressive genetic neuromuscular disorder.

The study also revealed seven "rediscoveries" — cases where treatment teams in one location had identified a diagnosis but hadn't shared it with the broader research community. Brownstein emphasized these findings remain valuable for ensuring patients can access new treatments as they become available.

Why it matters

The research demonstrates that off-the-shelf AI systems can augment scarce specialist expertise in rare disease diagnosis. With the Manton Center working with over 3,500 individuals globally affected by rare diseases, and a chronic shortage of geneticists to analyze complex genomic data, AI tools that can process vast amounts of research literature offer a practical way to accelerate diagnoses. Each diagnosis not only provides families with answers but also positions patients to benefit from emerging therapies. The approach could help address significant backlogs of unsolved cases at medical centers nationwide.

The researchers cautioned that large language model results require rigorous human review and that diagnosis is only an early step toward treatment. The tools are designed to assist clinicians, not replace them or be used directly by consumers.

These details were first reported by NBC News.