Integrated Algorithms Detect Hospital Bleeding Events

Swiss researchers conducted a multicenter retrospective study of patients aged ≥65 hospitalized Jan 2015–Dec 2016 to develop and validate automated algorithms combining structured-data rule models and an NLP model to detect major bleeding (MB) and clinically relevant nonmajor bleeding (CRNMB) from electronic medical records. The combined SDA+NLP model achieved sensitivity 0.84, positive predictive value 0.51 (F1 0.64) against a 754-record gold standard and reproduced findings on 2021–2022 external data, supporting use for drug-safety surveillance.