Drug Safety最新文献

{"title":"The WHO-ISoP Pharmacovigilance Curriculum: 2026 Update.","authors":"Jürgen Beckmann, Marco Tuccori, Priya Bahri, Francesco Salvo, Mónica Tarapués, Manal M Younus, Hussain Al Ramimmy, Mayada Alkhakany, Thamir Alshammari, Beate Aurich, Nicole Baker, Madhava Balakrishnan, Ghita Benabdallah, Pernette Bourdillon Esteve, Ian Boyd, Siva Kumar Buddha, Gianmario Candore, Emiliano Cappello, Angela Caro-Rojas, Rebecca Chandler, Jerin Jose Cherian, Karen Cohen, Salvatore Crisafulli, Shelley Deeks, Jean-Christophe Delumeau, Moin Don, Joan Dsouza, Ernest D Dzidzornu, Brian D Edwards, I Ralph Edwards, Jefman Efendi Marzuki Hy, Albert Figueras, Mohammed I Fouda, Ayako Fukushima, Giovanni Furlan, Georgy Genov, Gloria Giraldo, Michael Gold, Tarek Hammad, Kenneth Hartigan-Go, Maria Teresa Herdeiro, Richard Hill, Noha Iessa, Katarina Ilic, Sujeet Jain, EunMi Kim, Larissa Kopp, Veronique Kugener, Smaragda Lamprianou, Navinisha Lee, Marie Lindquist, Michaela Lucas, Viola Macolic Sarinic, Ishak Mansi, Ushma C Mehta, Johanna C Meyer, Joseph Mitchell, Maryam Naeem, Comfort Ogar, Ladi Owolabi, Jayesh Pandit, Lembit Rägo, Emanuel Raschi, Hadir Rostom, Mulugeta Russom, George T Sabblah, Maribel Salas, Souad Skalli, Fumihito Takanashi, Michael Tatley, Gianluca Trifiro, Claudia Vaca, Florence van Hunsel, Eugène van Puijenbroek, Jessica Liliana Vargas-Neri, Ryan J Walker, Xinyu Weng, Ian C K Wong, Qun-Ying Yue, Li Zhang, Omar Aimer, Shanthi N Pal","doi":"10.1007/s40264-026-01717-y","DOIUrl":"https://doi.org/10.1007/s40264-026-01717-y","url":null,"abstract":"<p><p>This article presents a pharmacovigilance (PV) curriculum which was jointly developed by the World Health Organization (WHO) and the International Society of Pharmacovigilance (ISoP). It is an expanded update of a PV curriculum which WHO and ISoP developed in 2014 and responds to major scientific, technological, regulatory, social and public health developments that have reshaped PV over the last 12 years. Topics such as advanced therapies, global health emergencies, antimicrobial resistance, digital health, artificial intelligence, real-world evidence, medical device safety, the environmental impact of medicines, risk communication and patient engagement are addressed. The curriculum was developed through an international collaboration involving 82 contributors from different regions and professional backgrounds. The original three-level hierarchical structure was retained, with each chapter divided into four sections and each section into five or six subsections. The number of chapters increased from 15 to 26, the number of practical tasks from 40 to 125 and the number of references from 260 to 469. Practical tasks and references are provided as Electronic Supplementary Materials. The curriculum provides a comprehensive and structured inventory of contemporary PV and thus represents a reference that lecturers and institutions can use to develop context-specific training programmes. Its structure provides conceptual clarity while allowing adaptation to different settings, learners' roles and prior knowledge, local priorities, available time and resources. It is not intended to prescribe a single standardised course. Instead, it supports the development of introductory, advanced, comprehensive or topic-specific PV education programmes. The update reflects a shared commitment of the WHO and the ISoP to strengthening PV education and capacity across regulatory, academic, healthcare, public health and industrial settings in accordance with the recommendation of the 2026 World Health Assembly to strengthen PV.</p>","PeriodicalId":11382,"journal":{"name":"Drug Safety","volume":" ","pages":""},"PeriodicalIF":5.9,"publicationDate":"2026-09-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149012021","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A Text-Mining Algorithm for Identifying Drug Pairs Potentially Involved in Prescribing Cascades: Development and Internal Validation Within the Prescribing Inappropriateness Assessment (PINA) Digital Tool.","authors":"Massimo Carollo, Anna Forti, Salvatore Crisafulli, Luca L'Abbate, Irene Cristini, Federica Soardo, Marilisa Giustina Stano, David Bellantuono, Davide Benetti, Riccardo Lora, Gianluca Trifirò","doi":"10.1007/s40264-026-01733-y","DOIUrl":"https://doi.org/10.1007/s40264-026-01733-y","url":null,"abstract":"<p><strong>Background: </strong>Prescribing cascades (PCs) occur when an adverse drug reaction (ADR) is intentionally or unintentionally treated with an additional medication, potentially perpetuating inappropriate polypharmacy. Digital tools that systematically support the identification of potential PCs during medication review are lacking. Here, we describe the development of an algorithm designed to identify drug pairs potentially compatible with PCs and evaluate its classification performance.</p><p><strong>Methods: </strong>The algorithm was developed as a component of the Prescribing INappropriateness Assessment (PINA) digital tool, a software platform designed to support clinicians in conducting medication reviews. The algorithm logic identifies potential PCs by detecting intersections between ADRs and therapeutic indications extracted from Italian Summaries of Product Characteristics (SmPCs) and encoded as Medical Dictionary for Regulatory Activities (MedDRA) terms using a natural language processing pipeline. Four algorithm configurations were tested, differing in the MedDRA semantic levels used for matching: (1) Preferred Terms (PTs) only; (2) PTs with semantic extension to High Level Terms (HLTs); (3) PTs with extension to Standardised MedDRA Queries (SMQs); and (4) PTs with combined HLT/SMQ extension. Performance was assessed against a reference standard comprising 69 literature-supported positive drug pairs and 138 randomly generated negative drug pairs in a 1:2 ratio. Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), accuracy, F₁ score, and Youden's index were calculated.</p><p><strong>Results: </strong>All models demonstrated excellent specificity (≥ 99%). Sensitivity increased with semantic expansion: 50.7% for PT-only matching, 73.9% for PT+HLT, 68.1% for PT+SMQ, and 85.5% for PT+HLT/SMQ. The combined HLT/SMQ extension model achieved the highest overall performance, with an accuracy of 94.7%, a PPV of 98.3%, an NPV of 93.2%, an F₁ score of 91.4, and a Youden's index of 84.8. False-positive rates remained minimal across all models.</p><p><strong>Conclusions: </strong>The newly developed algorithm demonstrated high classification performance, particularly when PT-level matching was extended using both HLTs and SMQs. This MedDRA-driven approach based on regulatory data may represent a scalable and reproducible strategy for supporting structured medication review and facilitating the recognition and clinical evaluation of potential PCs.</p>","PeriodicalId":11382,"journal":{"name":"Drug Safety","volume":" ","pages":""},"PeriodicalIF":5.9,"publicationDate":"2026-09-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149008099","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Interpreting Positive and Inverse Disproportionality Signals in Spontaneous Reporting Systems: A Claim-Strength Framework with Direction-Aware Evaluation.","authors":"Yoshihiro Uesawa","doi":"10.1007/s40264-026-01726-x","DOIUrl":"https://doi.org/10.1007/s40264-026-01726-x","url":null,"abstract":"<p><p>Disproportionality analysis of spontaneous reporting systems is usually used to identify positive signals, meaning drug-event pairs reported more often than expected. The same statistical structure also has a lower-reporting side. A reporting odds ratio (ROR) above unity indicates higher-than-expected reporting, whereas an ROR below unity indicates lower-than-expected reporting. Neither result directly estimates incidence, absolute risk, or causality. Existing critiques of inverse disproportionality signals rightly warn against interpreting ROR < 1 as protection, risk reduction, a beneficial reaction, or therapeutic effect. The same caution applies to ROR > 1. Higher-than-expected reporting is not, by itself, causal evidence of harm. This narrative methodological review argues that positive and inverse disproportionality signals should be interpreted according to the same scientific principles. This does not imply symmetric regulatory action thresholds. In this review, \"symmetric scientific standards\" means that claims of the same strength require commensurate evidence in both directions; it does not mean identical case-level information, bias mechanisms, detectability, predictive value, regulatory priority, or follow-up procedures. Instead, it calls for interpretation that is consistent with the strength of the claim. To make this approach practical, the review outlines a symmetry check that asks whether critiques of inverse signals also apply to positive signals and whether they address descriptive reporting or causal interpretation. It also proposes a claim-strength framework in which the required evidence depends on the claim being made, not on the direction of the signal. Under this approach, inverse signals may be reported as lower-than-expected reporting. They may also support hypothesis generation or candidate prioritization for follow-up evaluation when robust to comparator and sensitivity analyses. Claims of protection or risk reduction require external validation.</p>","PeriodicalId":11382,"journal":{"name":"Drug Safety","volume":" ","pages":""},"PeriodicalIF":5.9,"publicationDate":"2026-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148987870","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"The Pitfalls of Disproportionality Analysis: Insights on Their Interwoven Complexity and a Front-End Mitigation Strategy.","authors":"Manfred Hauben, Rave Harpaz, Robert Weber","doi":"10.1007/s40264-026-01707-0","DOIUrl":"https://doi.org/10.1007/s40264-026-01707-0","url":null,"abstract":"","PeriodicalId":11382,"journal":{"name":"Drug Safety","volume":" ","pages":""},"PeriodicalIF":5.9,"publicationDate":"2026-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148988301","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Authors' Reply to John Diaz-Decaro's comment on: \"Maternal and Pediatric Use of Vaccines for Mpox: A Living Systematic Review and Meta-analysis of Safety and Effectiveness\".","authors":"Agustín Ciapponi, Jamile Ballivian, Mabel Berrueta, Juan M Sambade, Noelia Castellana, Ariel Bardach, Martin Brizuela, Julieta Caravario, Daniel Comande, Esteban Couto, Agustina Mazzoni, Edward Parker, Florencia Salva, Katharina Stegelmann, Xu Xiong, Andy Stergachis, Flor M Munoz, Pierre Buekens","doi":"10.1007/s40264-026-01727-w","DOIUrl":"https://doi.org/10.1007/s40264-026-01727-w","url":null,"abstract":"","PeriodicalId":11382,"journal":{"name":"Drug Safety","volume":" ","pages":""},"PeriodicalIF":5.9,"publicationDate":"2026-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148988782","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Comment on \"Maternal and Pediatric Use of Vaccines for Mpox: A Living Systematic Review and Meta-analysis of Safety and Effectiveness\".","authors":"John Diaz-Decaro","doi":"10.1007/s40264-026-01728-9","DOIUrl":"https://doi.org/10.1007/s40264-026-01728-9","url":null,"abstract":"","PeriodicalId":11382,"journal":{"name":"Drug Safety","volume":" ","pages":""},"PeriodicalIF":5.9,"publicationDate":"2026-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148988823","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Non-genetic Risk Factors for Allopurinol-Induced Severe Cutaneous Adverse Reaction (SCAR): A Systematic Review.","authors":"Wei Leik Ng, Prawira Oka, Karleen Chong, Siok Fuang Liong, Rizawati Ramli, Winn Hui Han, Shin Shen Yong, Hooi Min Lim","doi":"10.1007/s40264-026-01729-8","DOIUrl":"https://doi.org/10.1007/s40264-026-01729-8","url":null,"abstract":"<p><strong>Background: </strong>Allopurinol-induced severe cutaneous adverse reactions (SCARs) are rare but potentially life threatening, particularly in Asian populations. While the genetic marker HLA-B*58:01 is a well-established risk factor, non-genetic factors may also contribute. This systematic review synthesizes evidence on associations between non-genetic risk factors and allopurinol-induced SCAR.</p><p><strong>Methods: </strong>We searched MEDLINE, Scopus, Cochrane Library and Web of Science from inception to 29 June 2026 for observational studies examining non-genetic risk factors for SCAR, defined as Stevens-Johnson Syndrome, Toxic Epidermal Necrolysis, Acute Generalised Exanthematous Pustulosis, or Hypersensitivity Syndrome/Drug Reaction with Eosinophilia and Systemic Symptoms. Adults aged ≥ 18 years were included. Pooled odds ratios (ORs) with 95% confidence intervals (CIs) were calculated using a random-effects model; heterogeneity was assessed with I2. Mean differences were calculated for continuous variables. NIH Study Quality Assessment Tool was used for quality assessment of the studies.</p><p><strong>Results: </strong>Twenty-six studies were included. Female sex (20 studies; 3340 SCAR cases, 562,647 controls) was associated with an increased risk of allopurinol-induced SCAR (OR 2.06; 95% confidence interval (CI) 1.25-3.38). Chronic kidney disease (16 studies; 1625 SCAR cases, 553,804 controls) was also significantly associated with SCAR (OR 3.78; 95% CI 1.99-7.17). Five studies (177 SCAR cases, 1368 controls) reported higher allopurinol doses among SCAR cases than tolerant controls (mean difference 19.61 mg; 95% CI 2.97-36.24). No significant associations were observed for age or concomitant diuretic use in the primary meta-analyses. Substantial heterogeneity was observed across studies. Sensitivity analyses demonstrated consistent findings for most factors, although concomitant diuretic use became significantly associated with SCAR after exclusion of non-Asian and zero-event studies.</p><p><strong>Conclusion: </strong>CKD, female sex, and higher allopurinol dose were identified as significant non-genetic risk factors for allopurinol-induced SCAR. These findings support consideration of non-genetic factors alongside pharmacogenomic screening in future risk-stratification strategies. However, substantial heterogeneity and potential publication bias limit the certainty of the available evidence. Well-designed studies evaluating non-genetic predictors as primary outcomes are needed to develop robust integrated risk prediction models for clinical decision making.</p>","PeriodicalId":11382,"journal":{"name":"Drug Safety","volume":" ","pages":""},"PeriodicalIF":5.9,"publicationDate":"2026-09-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148976289","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Sodium Glucose Co-Transporter 2 Inhibitors and Ventricular Arrhythmias in Patients with Type 2 Diabetes: A Systematic Review of Observational Studies.","authors":"Wang-Choi Tang, Wanning Wang, Antonios Douros, Oriana Hoi Yun Yu, Robert W Platt, Kristian B Filion","doi":"10.1007/s40264-026-01715-0","DOIUrl":"https://doi.org/10.1007/s40264-026-01715-0","url":null,"abstract":"<p><strong>Background: </strong>Sodium glucose co-transporter 2 inhibitors (SGLT2i) may exert antiarrhythmic effects, but their association with ventricular arrhythmias remains unclear.</p><p><strong>Objective: </strong>We conducted a systematic review to evaluate the association between SGLT2i use and the risk of ventricular arrhythmias, cardiac arrest, and sudden cardiac death compared with other antidiabetic medications or no SGLT2i use among patients with type 2 diabetes mellitus.</p><p><strong>Methods: </strong>MEDLINE, EMBASE, and CENTRAL were searched for observational studies published between March 2013 and March 2026. Quality was assessed using the Risk of Bias In Non-Randomized Studies of Interventions (ROBINS-I) tool, alongside evaluation of pharmacoepidemiology-specific biases.</p><p><strong>Results: </strong>A total of 17 studies (16 cohort and one nested case-control) were included. Based on ROBINS-I, seven studies had moderate, eight serious, and two critical risks of bias. Eleven studies had at least one pharmacoepidemiology-specific bias. For ventricular arrhythmias, estimates ranged from a protective effect (hazard ratio [HR] 0.20, 95% confidence interval [CI] 0.04-0.97) to a potential increased risk (odds ratio 1.87, 95% CI 0.89-3.95) with SGLT2i use. For cardiac arrest, estimates consistently reported a lower risk with estimates that ranged from HR 0.63 (95% CI 0.59-0.68) to HR 0.85 (95% CI 0.82-0.88). The only study on sudden cardiac death reported a potential risk reduction (HR 0.62, 95% CI 0.38-1.01).</p><p><strong>Conclusions: </strong>While the association between SGLT2i and ventricular arrhythmias remains inconsistent, the use of SGLT2i likely reduces cardiac arrest and may reduce sudden cardiac death, suggesting a possible protective effect on ventricular arrhythmias among patients with type 2 diabetes.</p>","PeriodicalId":11382,"journal":{"name":"Drug Safety","volume":" ","pages":""},"PeriodicalIF":5.9,"publicationDate":"2026-09-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148900159","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Challenges and Insights from the Mepolizumab Pregnancy Exposure Study.","authors":"Keele E Wurst, Michael Schatz, Subramanya Kumar, Megan Delgado, Christina D Chambers","doi":"10.1007/s40264-026-01724-z","DOIUrl":"https://doi.org/10.1007/s40264-026-01724-z","url":null,"abstract":"<p><p>Pregnant women are often excluded from clinical trials owing to uncertainty about fetal safety and adverse pregnancy complications. As such, little is known about the effect of medications on pregnancy at the time of marketing. Post-marketing data are therefore important for establishing the safety of treatments in pregnancy, particularly in chronic diseases requiring long-term therapy. Mepolizumab (NUCALA), a first-in-class biologic treatment for patients with severe eosinophilic asthma, was first approved in the United States (US) and European Union (EU) in 2015 as an add-on maintenance treatment for adult and adolescent patients. At the time of marketing, pregnancy safety data for mepolizumab was limited. To address this gap, a Post-Authorization Safety Study (PASS) was conducted-a pregnancy registry to monitor pregnancies exposed to mepolizumab and evaluate the possible teratogenic effect of this medication compared with women treated with other anti-asthmatics and those without asthma. The Mepolizumab Pregnancy Exposure Study (GSK ID: 200870; EU PAS Number: EUPAS13772), a phase IV, prospective, observational exposure registry, was initiated in 2016 by the Organization of Teratology Information Specialists (OTIS) Research Center. Despite extensive recruitment efforts, enrollment into the mepolizumab-exposed cohort remained low during the 5-year enrollment period. In contrast, robust enrollment in the anti-asthmatic comparison cohort was achieved, suggesting both the ability to recruit pregnant women with asthma and the limited use of mepolizumab in pregnancy. The Mepolizumab Pregnancy Exposure Study was ultimately closed in 2024 due to inadequate recruitment. This commentary presents the challenges experienced in the Mepolizumab Pregnancy Exposure Study and describes a pragmatic decision-making framework aimed at mitigating similar issues in future post-marketing pregnancy safety studies, particularly in the context of low treatment exposure in individual population-level data sources. This framework considers important inputs such as prior experience with treatment in the same therapeutic area, background maternal and fetal risks, and potential treatment utilization in the population of interest. A principal recommendation stemming from this framework is the implementation of enhanced worldwide pharmacovigilance, which offers a more robust and structured approach to data collection than standard pharmacovigilance methods, thereby improving information quality and enabling the capture of important variables and confounders, when the exposure rate is low in a population and registry or other database studies cannot reasonably be conducted. However, given the inherent limitations of pharmacovigilance in generating definitive safety conclusions, enhanced data collection methods serve as a pragmatic and necessary tool for ongoing surveillance, to aid in providing information until other types of study designs becoming feasible.</p>","PeriodicalId":11382,"journal":{"name":"Drug Safety","volume":" ","pages":""},"PeriodicalIF":5.9,"publicationDate":"2026-09-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148886224","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Development and Validation of a Two-Stage NLP System Combining Rule-Based Extraction and LLM-Based Classification for Deprescribing Recommendations in Discharge Summaries.","authors":"Kenji Fujita, Marion Matheson, Bhavya Valecha, Sarah N Hilmer","doi":"10.1007/s40264-026-01716-z","DOIUrl":"https://doi.org/10.1007/s40264-026-01716-z","url":null,"abstract":"<p><strong>Introduction: </strong>Polypharmacy in older adults is associated with increased risks of adverse drug events and functional decline. Discharge summaries often contain deprescribing recommendations, but these are frequently overlooked due to documentation complexity.</p><p><strong>Objective: </strong>The aim of this study was to develop and validate a two-stage hybrid natural language processing (NLP) system combining rule-based extraction and large language model (LLM)-based classification for automated identification of deprescribing recommendations from discharge summaries.</p><p><strong>Methods: </strong>This retrospective cohort study included 850 discharge summaries from patients aged ≥ 65 years with hospitalisation for ≥ 48 h across six public hospitals in New South Wales, Australia. Model 1 (rule-based extraction) identified discharge medications and candidate sentences containing pre-defined deprescribing keywords. Model 2 (LLM-based classification) categorised candidate sentences into five categories, from which a binary outcome (deprescribing recommendation present versus absent) was derived using a pre-specified mapping. Data were split into training (80%) and test (20%) sets. Gold standard classifications were established by independent reviews, followed by adjudication of discrepancies.</p><p><strong>Results: </strong>Model 1 extracted 9631 discharge medications (median 11 per patient) and 1061 candidate sentences from 850 patients (median age 82.8 years). Model 2 achieved an F1 score of 0.91 and accuracy of 0.90 for the binary outcome in the test set (144 medications). Inter-rater reliability showed substantial agreement (Cohen's kappa 0.70). The most frequently identified medications recommended for deprescribing were antibiotics and opioids. The most common misclassification was incorrectly identifying actions completed during hospitalisation as post-discharge recommendations. The combined processing time averaged 12.6 seconds per discharge summary.</p><p><strong>Conclusions: </strong>A two-stage hybrid approach combining rule-based extraction and open-source LLM-based classification could identify deprescribing recommendations from discharge summaries, enabling cost-efficient, privacy-compliant local deployment.</p>","PeriodicalId":11382,"journal":{"name":"Drug Safety","volume":" ","pages":""},"PeriodicalIF":5.9,"publicationDate":"2026-09-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148886263","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}