303P A pan-tumor and pan-country approach to LLM-based extraction of systemic therapies from the electronic health record
Natalia Viani, L. Groizard, K. Harrison, Anna M. Schwarz, Nikola Doležalová, F. Stefan, A. Hadjigeorgiou, Megan W. Hildner, H. Gudmundsson, A. Samani, S. Dover, M. Kushnir, B. Adamson, Lauretta J. Ellsworth, Kathi Seidl-Rathkopf · ESMO Real World Data and Digital Oncology · 2025
Systemic therapy data, especially for orally administered drugs, are often incompletely captured in structured electronic health record (EHR) fields, leading to gaps in real-world oncology datasets. Manually extracting this information from unstructured documents is resource-intensive and limits scalability. Large language models (LLMs) offer a promising solution for automated extraction of systemic therapy details from unstructured EHRs. This study evaluates a pan-tumor LLM approach for extracting oral therapies from oncology patient charts in the UK and US.