Journal Club: De-identifying sensitive health information across electronic health record text

The growing secondary use of unstructured Electronic Health Record (EHR) data for artificial intelligence (AI) research and development has highlighted the need for robust privacy-preserving mechanisms. De-identification is removing or replacing sensitive health information (SHI), remains essential for regulatory and ethical compliance. Despite substantial advances in automated de-identification, critical research gaps exist. This Journal Club webinar examines two complementary studies recently published in BMJ Health & Care Informatics (2026) from the SREDH Consortium that address these questions. The first study, provides the most comprehensive benchmark to date of LLM-based de-identification methods for EHR notes, evaluating eight LLMs across five heterogeneous international datasets. Fine-tuned models achieved strict F1 scores of up to 0.9447, significantly outperforming traditional rule-based approaches (F1 = 0.8172), while also revealing critical performance variability across data sources. Speakers: Jitendra Jonnagaddala, MD,MIS,PhD,FACHI,FIAHSI Senior Research Fellow Discipline of General Practice Prof. Dr. Shabbir Syed Abdul, Professor of Artificial Intelligence & Digital Health, Graduate Institute of Biomedical Informatics