Therapeutic Advances in Drug Safety

Ramcharran, Darmendra; Painter, Jeffery L.; Kara, Vijay; Glaser, Michael; Vanini, Marco; Chalamalasetti, Venkateswara Rao; Golds, Christopher; Abdelkarim, Azza; Bate, Andrew; Stegmann, Jens-Ulrich

doi: 10.1177/20420986251396023pmid: 41431473

The advent of generative artificial intelligence (GenAI) has introduced both remarkable opportunities and significant challenges in the field of pharmacovigilance (PV). This perspective review reflects on emerging trends, practical use cases, and conceptual frameworks shaping the integration of GenAI in high-risk domains such as drug and vaccine safety monitoring. We draw on current experiments and early real-world applications to examine the potential benefits, inherent risks, and propose a framework for integrating GenAI into PV systems, emphasizing the necessity of rigorous testing, human oversight, and ethical considerations. Our goal is to support PV professionals and stakeholders in navigating this rapidly evolving landscape by identifying promising strategies and implementation pathways.

Hill, Joseph Ben; Simons, Alexis; Wright, Garth; Anderson, Kelly E.

doi: 10.1177/20420986251401515pmid: 41426304

Background:Patients with type 2 diabetes (T2DM) are at risk of developing urinary tract infections (UTIs). Sodium-glucose cotransporter-2 inhibitors (SGLT2i) are a common medication associated with UTIs in these patients. However, emerging data show that other medications may be more frequently prescribed prior to UTI diagnosis.Objectives:Explore the correlation of newly prescribed medications in patients with the diagnosis of T2DM prior to an incidence of UTI and compare it to those without a UTI.Design:This observational case-control study aimed to explore the correlation between the incidence of UTIs in patients with T2DM and new prescription medication fills.Methods:Data were retrieved from national prescription and medical claims database IQVIA PharMetric® Plus for Academics between 2018 to 2021. The exposed cohort included patients with T2DM and an encounter for UTI. The comparator cohort was developed using propensity score matching and consisted of patients with T2DM and a health care encounter, but without a diagnosis of UTI.Results:A total of 31,746 patients met study criteria, with 15,873 in both the exposed and matched comparator cohorts. The medications with the largest percentage point difference were opioids at 3.70 (p-value <0.001), statins at 3.42 (p-value <0.001), amoxicillin at 2.48 (p-value <0.001), metformin at 2.45 (p-value <0.001), and PPIs at 2.19 (p-value <0.001). SGLT2i were the 19th most prescribed medication class.Conclusion:Opioids, statins, amoxicillin, metformin, and PPIs were the top 5 medications prescribed prior to the UTI event based on percentage point difference. SGLT2i were not in the top 10 medications initiated prior to UTI. This adds to existing literature that other new start medications may be correlated with a higher risk of developing a UTI such as opioids and PPIs than SGLT2 inhibitors in patients with T2DM.

Chen, Li-Yu; Liao, Ling-Chiao; Chuang, Chin-Ju; Hsieh, Cheng-Ying; Huang, Yung-Cheng; Yang, Yi-Hsuan; Chang, Wan-Ting; Fu, Shau-Huai; Yang, Feng-Jung; Chang, Kai-Chieh; Chan, Ding-Cheng; Shen, Li-Jiuan; Hsiao, Fei-Yuan; Huang, Yen-Ming; Hsu, Chih-Cheng; Wang, Chen-Yu

Zeng, Huiqiong; Liu, Wei; Li, Aidong; Liu, Hanjiang; Li, Xiaojuan; Lai, Junda; Chen, Miaoqian; Xiong, Gaofeng; Zhang, Ye

doi: 10.1177/20420986251406106pmid: 41426305

Background:Baricitinib is widely used for immune-mediated diseases, yet real-world safety in underrepresented age groups and the temporal dynamics of adverse drug reactions (ADRs) remain insufficiently characterized.Objective:To identify age-stratified ADR signals of baricitinib and to examine potential causal roles of Janus kinase (JAK) 1/2 inhibition in key ADRs.Design:A retrospective pharmacovigilance study integrating disproportionality analysis, Mendelian randomization (MR), and time-to-onset (TTO) assessment.Methods:Baricitinib-associated ADRs reported to the FDA Adverse Event Reporting System (FAERS; Q3-2018 to Q1-2024) were analyzed using Reporting Odds Ratio and Bayesian Confidence Propagation Neural Network, stratified by age (18–65 vs ≥66 years). TTO was modeled to characterize temporal patterns. Two-sample MR using eQTL-based instruments of JAK1/2 expression evaluated causal links with thrombosis and atrial fibrillation (AF).Results:Among 5354 reports, infections were most frequent (28.3%). Thrombotic events (deep vein thrombosis, pulmonary embolism) were more prominent in the elderly (≥66 years), whereas hepatic enzyme elevation and malignancies were more frequent in adults aged 18–65 years. MR suggested that higher JAK2 expression was protective against thrombosis (OR = 0.998, p = 0.028), whereas higher JAK1 expression conferred increased risk of AF (OR = 1.255, p = 0.043). TTO analysis showed that thrombotic ADRs tended to occur early after baricitinib initiation, whereas certain malignancies emerged later.Conclusion:This study highlights distinct age-dependent vulnerabilities to baricitinib-associated ADRs, with genetic evidence suggesting target-specific mechanisms underlying cardiovascular risk. These findings underscore the importance of age-tailored monitoring strategies and proactive pharmacovigilance in clinical practice.

Barma, Sachita; Pathak, Nabin; Dhungana, Shreya; Jha, Prabhat Kumar; Shrestha, Sunil

doi: 10.1177/20420986251398725pmid: 41362372

Background:Drug recalls safeguard patients from potentially harmful existing pharmaceutical products in the market due to safety concerns or manufacturing issues.Objective:To analyze the drug recall patterns, frequencies, and causes based on the publicly available data from the Nepalese Drug Regulatory Authority, Department of Drug Administration (DDA).Design:A 2-year retrospective secondary analysis was conducted based on drug recalls in Nepal, which is available from the official website of DDA, Nepal.Methods:The substandard drug recalls data from April 17th, 2023 to May 14th, 2025 were included in the study from which information were extracted across various domains, including the drug name, dosage forms, manufacturing and expiry date, reasons for the recall, and recall date whereas, inclusion of falsified medicines and information regarding the “special permission,” “news & updates,” and “notices” sections within the DDA website were excluded from the studies to ensure only drug recalls were incorporated.Results:The study showed that 50 recalls were made over the 2 year period. The majority of the recalled drugs were antibiotics (16%). The most common reasons for drug recalls were assay failure (34%) and non-compliance with standards set out by the Indian Pharmacopoeia-2022 (40%), followed by United States Pharmacopoeia-2022 (28%). Oral formulations (74%) were most commonly recalled, out of which tablets (32%), suspensions (16%), and syrups (12%) were recalled in greater frequencies among all the recall notices. Most recalls were made from domestic pharmaceutical companies (84%). The majority of the drugs (34%) were recalled after 15 months of the finished product being in the market, whereas only 26% were recalled within the first 5 months.Conclusion:A robust and continuous evaluation of drug recalls by regulatory authorities can help reduce their frequency, lessen their impact on the healthcare system, and improve overall drug safety.

Roemming, Hans-Joerg; Hauben, Manfred; Wannhoff, Wei; Schaffer, Claudia; Tihaa, Irina; Heitmann, Martin; Mengling, Veit

doi: 10.1177/20420986251386222pmid: 41383438

As per Good Pharmacovigilance Practices, pharmaceutical companies must act on potential adverse reactions to drugs. With significant increases in the number of case reports in recent years, they face pressure to raise the efficiency of their processes while maintaining data integrity and patient safety. The use of Large Language Models (LLMs) in safety case intake provides potential to advance processes without compromising quality. In this perspective review, we highlight the potential benefits of LLMs in case intake workflows, and points to consider relating to the current research landscape, inspired by our proof-of-concept (PoC) study. Benefits include raising the consistency of data extraction, reducing bias, and enhancing efficiency. We reflect on challenges in realizing the potential of this new technology from a practical industry perspective, namely (a) measuring the Return on Investment, (b) early involvement of subject matter experts, (c) handling unclear regulatory expectations, (d) system integration, and (e) organizational readiness. We illustrate the potential and its challenges through the lens of our PoC’s insights as well as through insights from published literature, which allowed us to estimate an efficiency gain from a business process perspective for data extraction and initial case report, demonstrating the technology’s potential and practical applicability in real-world scenarios.

Qiu, Xiaohan; Li, Qinxiao; Rong, Yiyin; Wang, Longyu; Ke, Jiahan; Wang, Min; Zeng, Huasu; Gu, Jun

doi: 10.1177/20420986251401122pmid: 41394295

Background:Cardiovascular adverse events represent critical complications of antineoplastic therapy with profound implications for cancer survivorship and treatment outcomes. Despite the clinical significance, comprehensive pharmacovigilance data characterizing distinctive cardiotoxicity profiles across modern cancer therapeutics remain limited.Objectives:This investigation systematically analyzes cardiotoxicity patterns associated with antineoplastic agents using the FDA Adverse Event Reporting System (FAERS) database to inform evidence-based cardiovascular monitoring strategies.Design:A retrospective pharmacovigilance study utilizing disproportionality analysis and time-to-onset evaluation.Methods:We conducted a comprehensive analysis of FAERS data spanning 2004–2024, employing validated disproportionality metrics including reporting odds ratio (ROR) and proportional reporting ratio (PRR) to detect significant drug-event associations. Advanced time-to-onset analysis revealed temporal patterns of cardiotoxicity development across therapeutic classes. Statistical significance was defined as ROR >1 with 95% confidence intervals excluding 1.0, and PRR >2 with chi-square >4.Results:Analysis of 18,289,374 reports identified 51,402 cases of antineoplastic-related cardiovascular toxicity, demonstrating distinct class-specific patterns. Anthracyclines exhibited profound associations with structural cardiac damage (doxorubicin-cardiomyopathy: ROR = 20.64, 95% CI: 19.87–21.45). Immune checkpoint inhibitors demonstrated unprecedented immune-mediated cardiac inflammation (pembrolizumab-myocarditis: ROR = 245.36, 95% CI: 218.42–275.88). Fluoropyrimidines showed distinctive vasospastic effects (5-fluorouracil-Prinzmetal angina: ROR = 18.27, 95% CI: 14.72–22.69). Critical temporal patterns emerged: fluoropyrimidines caused early-onset cardiotoxicity (median: 11 days, IQR: 4–28), anthracyclines showed intermediate onset (doxorubicin median: 64 days, IQR: 21–156; epirubicin median: 72 days, IQR: 28–168), while mitoxantrone demonstrated delayed effects (median: 457 days, IQR: 182–891). Cardiogenic shock emerged as the most lethal manifestation with a 43.08% mortality rate (95% CI: 40.12–46.14).Conclusion:This landmark pharmacovigilance study reveals previously uncharacterized temporal and mechanistic patterns of antineoplastic cardiotoxicity, providing an essential evidence-based framework for cardiovascular monitoring strategies. The findings highlight critical intervention windows: immediate monitoring for fluoropyrimidines, intermediate surveillance for anthracyclines (2–6 months), and extended follow-up for agents like mitoxantrone (>12 months). These insights support the development of risk-stratified cardio-oncology protocols tailored to specific therapeutic classes.

Kenny, Tiff-Annie

doi: 10.1177/20420986251395925pmid: 41466718

The COVID-19 pandemic has renewed attention to complex chronic health conditions that challenge conventional biomedical paradigms. Syndromes such as postural orthostatic tachycardia syndrome and myalgic encephalomyelitis/chronic fatigue syndrome have gained broader visibility through the lens of Long COVID. As global vaccination campaigns expanded, a subset of individuals began reporting similarly persistent, multisystem symptoms following COVID-19 immunization—informally referred to as post-COVID-19 vaccination syndrome. These presentations, which include dysautonomia, neuropathic pain, post-exertional malaise, and cognitive dysfunction, resemble post-infectious syndromes and may involve shared immune-related mechanisms. Although no causal relationship to vaccination has been established, these cases—together with comparable reports following other vaccines—highlight limitations in current vaccine safety systems for detecting and evaluating complex chronic outcomes. This article introduces the concept of complex chronic adverse events following immunization (CC-AEFIs) as a pragmatic, surveillance-oriented framework to support the systematic identification and investigation of such cases. CC-AEFIs are not syndromic diagnoses but a higher-order category encompassing persistent, multifactorial conditions that may follow immunization yet challenge existing pharmacovigilance definitions and tools. These conditions often involve multiple organ systems, delayed onset, fluctuating trajectories, diagnostic ambiguity, and symptom heterogeneity. Drawing on the author’s lived experience as an affected patient and integrating clinical, regulatory, and experiential evidence, the analysis examines structural and epistemic limitations across the pharmacovigilance continuum—from underrecognition in clinical settings to analytic exclusion and constrained governance. It concludes by proposing reforms to strengthen safety-system responsiveness, including enhanced diagnostic training, longitudinal surveillance, patient-reported outcome integration, and analytic transparency. Addressing these limitations is essential to sustain public trust, ensure equitable care, and uphold the scientific integrity of immunization programs.

Kim, Minjung; Kim, Kyoung Eun; Kwon, Jae-Hee; Han, Ja-Young; Kim, Jae Hyun; Kim, Myeong Gyu

doi: 10.1177/20420986251405082pmid: 41426306

Background:Adverse drug reactions (ADRs) are harmful side effects of medications. Social media provides real-time, patient-generated data, though its unstructured format presents challenges. Natural language processing and transfer learning offer promising solutions.Objective:This study aimed to evaluate whether transformer-based models fine-tuned on a general ADR dataset can effectively classify ADRs from tweets related to glucagon-like peptide-1 (GLP-1) receptor agonists and to benchmark their performance against state-of-the-art large language models (LLMs).Design:This study employed a machine learning approach using transformer-based language models to classify ADRs in social media.Methods:BERT (bidirectional encoder representations from transformers)-base, BERTweet-base, and GPT-2 (Generative Pre-Trained Transformer-2) models were fine-tuned using Sarker and SIDER (Side Effect Resource) datasets for ADR classification. The test dataset comprised 396 tweets mentioning GLP-1 receptor agonists that were categorized as personal experiences. Model performance was primarily evaluated using the F1 score, which was used to select the optimal model. In addition, the fine-tuned transformer models were benchmarked against state-of-the-art LLMs, including ChatGPT 4o, ChatGPT 4o-mini, and Gemini 2.5 Flash.Results:Among 396 tweets, 116 (29.3%) were classified as ADRs and 280 (70.7%) as non-ADRs. Among the transformer-based models, BERTweet-base achieved the highest performance (accuracy: 0.835, F1: 0.729), outperforming both BERT-base (accuracy: 0.826, F1: 0.679) and GPT-2 (accuracy: 0.766, F1: 0.628). Among the LLMs, ChatGPT 4o-mini demonstrated the best results (accuracy: 0.970, F1: 0.948), followed by Gemini 2.5 Flash (accuracy: 0.954, F1: 0.919) and ChatGPT 4o (accuracy: 0.936, F1: 0.895). Overall, LLMs substantially outperformed the fine-tuned transformer-based models.Conclusion:Fine-tuned transformer-based models demonstrated reasonable performance in ADR detection from GLP-1 receptor agonist tweets, with BERTweet-base performing best. However, state-of-the-art LLMs, particularly ChatGPT 4o-mini, substantially outperformed these models, highlighting their potential for pharmacovigilance tasks.

Sha, Yupeng; Yuan, Quan; Qian, Yao; Li, Xiaoming; Niu, Ming; Du, Yi; Liang, Xiaoshuan; Sun, Shanshan; Lu, Yige; Han, Jiguang

doi: 10.1177/20420986251405091pmid: 41409394

Background:There is a rising incidence of neurological adverse events (AEs), such as seizures, associated with novel anticancer agents, warranting investigation. Large-scale studies assessing seizure risk across diverse anticancer drug classes, particularly in breast cancer (BC), remain limited.Objective:This study aimed to systematically evaluate the association between seizures and 14 novel anticancer agents used in BC treatment, compared with traditional chemotherapy, utilizing international pharmacovigilance databases.Design:A large-scale, real-world pharmacovigilance study using data from the US FDA Adverse Event Reporting System (FAERS) and the Canada Vigilance Database (from Q1 2004 to Q1 2025).Methods:Disproportionality analysis was employed to calculate reporting odds ratios (RORs) for identifying significant seizure AE signals. Signals were assessed at both the Standardised MedDRA Query and Preferred Term levels. Pan-cancer transcriptomic data from The Cancer Genome Atlas were integrated to explore biological pathways correlated with drug-induced seizures.Results:Significant and consistent seizure signals were identified for five agents—Lapatinib, Tucatinib, Trastuzumab, Trastuzumab Emtansine (T-DM1), and Atezolizumab—across both databases. In FAERS, over 50% of seizures occurred after 100 days of treatment (median: 68 days); however, fatal cases exhibited a significantly shorter median onset time. Novel agents demonstrated disproportionately higher seizure reporting signals compared to traditional chemotherapy. Pan-cancer analysis revealed negative correlations between seizure RORs and pathways, including asthma and the pentose phosphate pathway.Conclusion:This dual-database pharmacovigilance study identifies potential associations between seizures and five novel BC therapies, underscoring the need for vigilant monitoring during their clinical use.