Large language models for scalable standardization in radiation oncology

With the advent of electronic medical records, there has been a sharp increase in the volume and breadth of clinical data collected across healthcare systems.Clinical data is estimated to account for more than 30 % of global data generation and is projected to grow at a compound annual growth rate of 36 % [1].To ensure that this increased data translates into scientific and clinical value, the research community has developed the FAIR principles, a framework requiring that data be Findable, Accessible, Interoperable, and Reusable [2].A persistent barrier to actualizing these principles is the fact that the same clinical concept is often recorded under different names/codes/conventions across institutions, software ecosystems, and countries.This fragmentation is exacerbated by the natural language barrier, limited interoperability between software vendors, and by documentation workflows primarily optimized for local administrative and billing needs rather than multi-center research.Controlled medical terminologies and classification systems have been developed to address these challenges, but achieving consistency in practice remains difficult, particularly for institutions operating outside international collaborative networks.For example, mapping between systems such as the Systematized Nomenclature of Medicine Clinical Terms and International Statistical Classification of Diseases and Related Health Problems nomenclature remains incomplete [3], propagating misclassification into downstream analyses.