Beyond Billing Codes: An Agentic Large Language Model Pipeline for Extracting Glaucoma Procedural Data from Free-Text Clinical Notes

Claude Opus 4.6 most accurately extracted glaucoma prior and future procedure data from free-text EHR notes, reaching 0.996 accuracy for prior procedures and 0.978 for future procedures.

• Why it matters: Structured EHR fields miss much glaucoma procedural detail, creating a gap for research and clinical characterization of procedure burden beyond billing codes.
• What they did: In a cross-sectional study of 1,250 subjects from the Bascom Palmer Ophthalmic Repository, randomly sampled 2014–2024 notes were labeled by glaucoma specialists and processed by a three-step agentic LLM workflow (extraction, validation, conditional revision) with text standardization.
• The result: Prior and future procedures were identified in 53.3% and 7.7% of eyes, and only 25.1% of adjudicated prior procedures matched CPT-coded EHR data, enabling scalable recovery of largely unstructured procedure information.

AI summary of the abstract · may contain errors · not medical advice · read the paper.