health
Using AI to help physicians diagnose rare genetic diseases affecting children
In an NEJM AI study, experts used an OpenAI reasoning model to reanalyze 376 previously unsolved cases and surface leads for 18 diagnoses.
TL;DR
- An AI reasoning model, OpenAI o3 Deep Research, was used to reanalyze 376 previously unsolved rare disease cases.
- The AI model surfaced evidence-linked candidate explanations, leading to the establishment of diagnoses in 18 cases, an additional diagnostic yield of 4.8%.
- The AI assisted researchers by synthesizing clinical features, inheritance patterns, variant evidence, and scientific literature into justifications for human review.
- Diagnoses were only confirmed after expert review, classification of variants as pathogenic or likely pathogenic, and confirmation by a CLIA-certified laboratory.
- The study shows AI can help revisit difficult cases as scientific knowledge advances and new gene-disease relationships are discovered.
- The AI model did not diagnose patients or make clinical decisions; it provided hypotheses for specialists to investigate.
- The AI's self-reported confidence scores were helpful in guiding expert reviewers but were not used as a substitute for evidence or clinical adjudication.
- The research team also identified a possible novel mechanistic explanation for vitiligo and potential phenotype expansions for certain genetic disorders.
- The study emphasizes that AI is a tool to assist, not replace, expert clinicians and established diagnostic processes.
- Future research will focus on developing AI copilots to help clinical teams analyze rare disease cases more quickly and consistently.