Ariel Ong unveils ARVO 2026 clinical-letter pipeline

Ariel Ong used her ARVO 2026 presentation to describe a problem common across hospital records: ophthalmic clinic letters often contain clinically useful details in free text that are difficult to query at scale. Ong, a fellow at University College London, said her team built a scalable extraction pipeline using large language models to turn those letters into structured fields that can be used for patient management and research. (ophthalmologytimes.com) Ophthalmology Times reported the presentation in an article and short video published on May 5. ARVO 2026 was held in Denver from May 3 through May 7. (ophthalmologytimes.com) ### What problem was Ong trying to solve? Ong said the core issue was the inaccessibility of free-text clinical letter data sets. (ophthalmologytimes.com) In ophthalmology, clinic correspondence can contain diagnoses, disease status and management details, but those details are often trapped in narrative text rather than stored in fields that can be easily searched or analyzed. University College London's profile for Ariel Yuhan Ong says her research focuses on validating and deploying AI systems for clinical decision-making and scientific discovery in retinal disease, while also building the data infrastructure needed to support that work. That description aligns with the presentation's focus on converting unstructured records into usable research inputs. (ophthalmologytimes.com) ### What did the pipeline actually do with the letters? Ophthalmology Times said Ong presented a poster examining a scalable extraction pipeline that used large language models to process ophthalmic clinical letters. The aim was to convert free-text records into extractable structured data, allowing information in letters to be used more systematically. (ophthalmologytimes.com) The publication said the work was framed around data extraction rather than automated diagnosis. Ong said the significance of the pipeline was tied to managing patients and making more of the record available for downstream analysis. (profiles.ucl.ac.uk) ### How large was the data set in the ARVO presentation? Ophthalmology Times said the analysis covered 85,000 patients. In that group, structured diagnostic data were available for only about 25% of patients before the extraction approach was applied. The same report said that without the pipeline, up to two-thirds of patient records would have been excluded from the analyses. (ophthalmologytimes.com) That figure gives the clearest measure of the data loss the project was designed to address. ### What did Ong say the added data could be used for? Ophthalmology Times said the added structured data increased the proportion of records available to validate a deep learning algorithm for macular disease detection. (ophthalmologytimes.com) That places the project in a workflow where letters are not only clinical documents but also a source of labels and context for model development and validation. Moorfields BRC, in a separate conference roundup, said Ong developed a scalable process using large language models to digest free-text letters written by clinicians and make previously inaccessible records usable for research at scale. (ophthalmologytimes.com) That description matches the use case reported by Ophthalmology Times. ### Where did this appear, and what comes next? Ophthalmology Times published the article “ARVO 2026: Scalable pipeline for data extraction from ophthalmic clinical letters” on May 5 and also posted a short video version of the report. The article credits Ariel Ong and Martin David Harp. (ophthalmologytimes.com) ARVO said its 2026 annual meeting in Denver ran from May 3 to May 7 and included artificial intelligence sessions focused on data science, real-world implementation and AI-aided research. Ong’s poster appeared within that broader meeting agenda. (arvo.org) (ophthalmologytimes.com) (moorfieldsbrc.nihr.ac.uk)