This study, presented by Gaspar et al., explores the application of natural language processing (NLP) models for detecting bleeding-related adverse drug events (ADEs) among older patients receiving antithrombotic therapies. ADE detection is a major safety concern in hospital settings, as rule-based systems within electronic medical records often fail to capture nuanced clinical information. To address this, clinicians manually annotated 404 discharge summaries into categories of “no bleeding,” “clinically significant bleeding,” “severe bleeding,” and “history of bleeding.” Using these annotations, researchers trained an NLP model and compared its performance against a conventional ICD-10–based algorithm. The NLP model outperformed the rule-based approach across all evaluation metrics, achieving an accuracy of 0.81 and an F1 score of 0.80. It demonstrated especially high accuracy in detecting severe (0.92) and clinically significant (0.87) bleeding events. The model also achieved an AUC of 0.91 for distinguishing relevant from irrelevant sentences and 0.94 for differentiating clinically significant bleeding from severe bleeding. In contrast, the ICD-10 method had high accuracy (0.94) for clinically significant bleeding but a much lower recall (0.03) for severe events such as hemorrhagic shock. The authors conclude that NLP-based models enable more comprehensive and sensitive detection of bleeding ADEs, particularly in high-risk cases often missed by ICD-10–based methods.