Particle.news

AI Model Helps Clinicians Find 18 New Diagnoses in Boston Children’s Study

Showing that AI can surface testable leads validated by experts and then confirmed in certified clinical labs, the study prompts calls for prospective multicenter trials before clinical rollout.

Overview

  • The New England Journal of Medicine’s NEJM AI published a retrospective study on Thursday reporting that OpenAI’s o3 model helped generate evidence‑linked leads that contributed to 18 confirmed diagnoses from 376 previously unsolved pediatric genomes.
  • Researchers used the model as an assistive reanalysis tool that highlighted candidate gene–disease links while clinicians reviewed results and certified clinical laboratories performed final confirmation.
  • The study yielded about a 4.8% additional diagnostic rate but was limited by its retrospective design and did not measure whether the tool saves time, cuts costs, or improves patient outcomes.
  • Patients described real personal benefits, including one person who received an ultra‑rare myofibrillar myopathy diagnosis after decades without an answer, showing how diagnoses can change care and provide closure.
  • Authors and outside experts stressed strict privacy safeguards and human oversight are required and said the next step is larger, forward‑looking multicenter trials to test real‑time clinical utility and safety.