Limits of Inter-Hospital Model Transfer in Delirium Detection

This collaborative effort implemented a natural language processing (NLP) model across seven sites to extract delirium-related features from unstructured clinical notes. While deployment was feasible, performance varied widely due to local documentation styles, screening tools, and note structures. To address these challenges, the group created a standardized ontology, introduced temporal logic for queries, and emphasized site-specific preprocessing, making the system more portable. For pharmacy, this approach enhances identification of delirium cases, especially where medication safety is impacted by high-risk drugs.

Fu S, Kwak MJ, Ahn J, et al. Advancing Delirium Detection through the Open Health Natural Language Processing Consortium and ENACT Network. J Gerontol A Biol Sci Med Sci. Published online September 29, 2025. doi:10.1093/gerona/glaf207