ARVO 2026: Scalable pipeline for data extraction from ophthalmic clinical letters

Ariel Ong, an ophthalmology resident and NIH Doctoral Fellow at University College London, presented a scalable data extraction pipeline using large language models (LLMs) to extract valuable information from free-text clinical letters in ophthalmology. The project aimed to unlock hidden data in clinical texts, which are often incomplete in structured data fields. Ong developed a modular, resource-efficient pipeline and introduced a framework for operational considerations. A proof of concept involved extracting nine macular diseases from clinical letters to validate a deep learning algorithm. This method, which could enhance research by reducing bias and improving data utilization, has potential applications in auditing complications, natural history studies, and pharmacovigilance.