177MO Multi-center validation of an artificial intelligence electronic health records extraction pipeline
Kevin Zarca, L. Zullo, Virginie Levrat, J. Bennouna, S. Schneider, O. Mercier, Emmanuelle Mougenot, E. Bergot, Cécile Dujon, N. Cloarec, Clarisse Audigier-Valette, Antonio Nuccio, C.F.A. Helissey, A. Carpentier, A. Djarallah, P.A. Rolland, F. Barlesi, Franck Le Ouay, B. Besse, Mihaela D. Aldea · ESMO Real World Data and Digital Oncology · 2025
Manual abstraction of real-world oncology data from unstructured Electronic Health Records (EHRs) is slow, error-prone, and heterogeneous across institutions. While large language models (LLMs) can scale information extraction, rigorous validation in multicenter settings remains limited.