Identification of Risk Factors for Glaucoma Progression in Free-Text Clinical Notes Using a Local Large Language Model
Translational Vision Science & Technology · 19.05.2026 · 15 (5) · 13
DOI: 10.1167/tvst.15.5.13 ↗Kurz gefasst
Fachartikel in der Zeitschrift Translational Vision Science & Technology, erschienen am 19.05.2026. Verfasst von Anshul Bhatnagar und 7 weiteren Autor:innen. Beteiligte Einrichtung: Bascom Palmer Eye Institute, University of Miami, Miami, FL, USA. Hauptthemen laut MEDLINE: Large Language Models, Electronic Health Records, Glaucoma. Die Arbeit wurde bisher 4-mal zitiert (Crossref) und nennt 42 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
Purpose
To evaluate the performance of a large language model (LLM) in identifying medication non-adherence, visit non-adherence, and family history of glaucoma (FHoG) in clinical notes from the electronic health record (EHR).
Methods
We extracted clinical notes of 1250 glaucoma-related encounters between 2014 and 2024 and structured EHR family history field data from the Bascom Palmer Ophthalmic Repository; of these, 125 randomly selected notes (10%) were used for prompt development and were excluded from analysis. Two fellowship-trained glaucoma specialists labeled notes for evidence of non-adherence and FHoG, which was used as the reference standard. We utilized MedGemma-27B-text-it, a specialized medical LLM, to identify medication non-adherence, visit non-adherence, and FHoG. We calculated accuracy, sensitivity, and specificity of LLM performance for each task, the Jaccard index for FHoG, and the mean squared error (MSE) of number of family members with glaucoma.
Results
Prevalence of medication non-adherence, visit non-adherence, and FHoG were 7.3%, 4.7%, and 29.2%, respectively. LLM accuracy was 0.91 (sensitivity, 0.96; specificity, 0.91) for medication non-adherence and 0.96 (sensitivity, 0.97; specificity, 0.94) for visit non-adherence. For FHoG, LLM accuracy was 0.98 (sensitivity, 0.99; specificity, 0.99) with a Jaccard index of 0.99, whereas EHR family history field accuracy and Jaccard index were 0.49 and 0.75, respectively. LLM and EHR MSE in quantifying the number of relatives with glaucoma were 0.05 ± 0.56 and 0.85 ± 1.80, respectively (P < 0.001).
Conclusions
LLMs identified non-adherence to medication and visit schedules, as well as degree of FHoG, in clinical notes with high accuracy. Translational Relevance: Local LLM pipelines can enable large-scale research into glaucoma risk factors that are unavailable in discrete EHR fields.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Autor:innen und Institutionen
- Bascom Palmer Eye Institute, University of Miami, Miami, FL, USA.
- Department of Ophthalmology and Visual Sciences, Escola Paulista de Medicina Universidade Federal de São Paulo, São Paulo, Brazil.
Zitieren
APA
Bhatnagar, A., Scherer, R., Samico, G. A., Muralidhar, R., Gutkind, N. E., Palazoni, V., Medeiros, F. A., & Swaminathan, S. S. (2026). Identification of Risk Factors for Glaucoma Progression in Free-Text Clinical Notes Using a Local Large Language Model. Translational Vision Science & Technology, 15(5), 13. https://doi.org/10.1167/tvst.15.5.13
Vancouver
Bhatnagar A, Scherer R, Samico GA, Muralidhar R, Gutkind NE, Palazoni V, et al. Identification of Risk Factors for Glaucoma Progression in Free-Text Clinical Notes Using a Local Large Language Model. Translational Vision Science & Technology. 2026;15(5):13. doi:10.1167/tvst.15.5.13
BibTeX
@article{Bhatnagar2026,
author = {Bhatnagar, Anshul and Scherer, Rafael and Samico, Gustavo A and Muralidhar, Rohit and Gutkind, Naomi E and Palazoni, Vitoria and Medeiros, Felipe A and Swaminathan, Swarup S},
title = {Identification of Risk Factors for Glaucoma Progression in Free-Text Clinical Notes Using a Local Large Language Model},
journal = {Translational Vision Science & Technology},
year = {2026},
volume = {15},
number = {5},
pages = {13},
doi = {10.1167/tvst.15.5.13}
}
EUVIMED