AI Model Suggests New Diagnoses for 18 of 376 Unsolved Rare-Disease Cases at Boston Children’s Hospital

Researchers at Boston Children’s Hospital reported that OpenAI’s o3 model identified new diagnoses for 18 of 376 patients whose rare diseases had remained unsolved. The study was published Thursday in NEJM AI. The o3 model, released in April 2025, received doctors’ notes, patient symptoms and lists of candidate genes.

Human clinicians then reviewed the model’s output before recording a final diagnosis. Of the 18 new diagnoses, 10 were rare neurodevelopmental diseases, four were neuromuscular disorders, two involved patients who had died suddenly, and two were early psychosis cases. ” She noted that the center routinely rechecks unsolved genomes against newly published gene-disease links.

Rare diseases affect 30 million people in the United States, according to the study background. One patient, Kyra Benton, began showing symptoms at age nine, including walking on tiptoes and difficulty running normally. Researchers diagnosed her with myofibrillar myopathy, a progressive genetic neuromuscular disorder, just before she turned 20 last year.

Benton told NBC News she is “not all that much in favor of AI,” yet credited the model-assisted review with finally providing an answer. OpenAI’s service terms state that its technology should not be used for self-diagnosis. The hospital study kept final decisions with clinicians.

An OpenAI researcher focused on health applications, Shringarpure, told NBC News that many of the 376 cases had already been examined multiple times by conventional methods.