Guideline-grounded large language models for extracting genome-informed clinical recommendations from…

AI / informaticsJournal of the American Medical Informatics Association : JAMIA· Oct 2026

Guideline-grounded large language models for extracting genome-informed clinical recommendations from electronic health records.

Yi Xin, Kyle W Davis, Wu-Chen Su … Wei-Qi Wei

Large Language Models in CareAI / informaticsClinical practice

Guideline-grounded large language models improve extraction of genome-informed clinical recommendations.

Guideline-grounded large language models for extracting genome-informed clinical recommendations from electronic health records.

Yi Xin … Wei-Qi Wei

Journal of the American Medical Informatics Association · Oct 2026

Background

Large language models (large language models) offer a promising approach to automated extraction, but their performance in genome-informed clinical contexts remains incompletely characterized.

Purpose

To evaluate guideline-grounded large language models approaches for extracting genome-informed clinical recommendations from electronic health record notes in the eMERGE study.

Methods

Three large language models (GPT-4o, LLaMA-3.3-70B, and LLaMA-3.1-8B) were evaluated using baseline prompting, guideline-aware prompting, and retrieval-augmented generation (retrieval-augmented generation), validated against manual chart review (N = 18 participants; N = 34 documents).

n = 18 participants

Results

F1 = 0.81

LLaMA-3.1-8B with retrieval-augmented generation achieved the highest performance in extracting recommendations

n = 18 participants

“

Conclusion

Domain-grounded large language models approaches can support scalable extraction of genome-informed clinical recommendations from electronic health record data. Larger studies are needed to assess generalizability in real-world workflows.

Guideline-grounded large language models for extracting genome-informed clinical recommendations from electronic health records.

Guideline-grounded large language models for extracting genome-informed clinical recommendations from electronic health records.

Related papers