A recent study published in NEJM AI suggests large language models (LLMs) may help clinicians revisit unsolved rare disease genomes and find diagnoses that were previously missed. Researchers say the approach could make genetic interpretation more efficient and improve the diagnostic yield for patients who have gone years without answers.
Methodology
The team conducted a retrospective reanalysis of 376 cases across four cohorts using an LLM-assisted workflow. The model ingested clinician notes, Human Phenotype Ontology terms and filtered variant tables to generate explanation-rich candidate hypotheses for expert review under ACMG and AMP criteria. A diagnosis counted only when a variant was classified as pathogenic or likely pathogenic, confirmed in a Clinical Laboratory Improvement Amendments-certified lab and returned to families.
Results and What They Mean for ECPs
The workflow produced 18 new local diagnoses overall, a 4.8% diagnostic yield. That included 10 of 100 neurodevelopmental cases, four of 61 neuromuscular cases, two of 200 sudden unexpected death in pediatrics cases and two of 15 early psychosis cases. Researchers also found seven “rediscoveries,” where pathogenic findings had been established elsewhere but were not yet available locally. In one case, the model helped flag a 22q11.2 deletion that was later confirmed by whole-genome sequencing.
For eyecare professionals, the broader takeaway is that LLM-assisted reanalysis may help surface genetic explanations that were previously overlooked, especially in complex or unresolved cases. The study also shows the approach can generate testable biological hypotheses, though the authors say prospective multicenter evaluation is still needed before wider use.
Read more of the latest AI in Eye Care news here
Author
