Drug Safety最新文献

Background: It is unknown whether nR value [(ALT or AST/ULN) ÷ (AP/ULN)] ≥ 5 is better than alkaline phosphatase less than twice the upper limit of normal (AP < 2x ULN) in identifying hepatocellular drug-induced liver injury (HC DILI) consistent with Hy's law in clinical trials.

Objective: We aimed to compare nR value ≥ 5 and AP < 2x ULN in clinical trial DILI cases with ALT or AST ≥ 3x ULN and total bilirubin (TB) > 2x ULN.

Methods: We retrospectively categorized clinical trial, DILI cases from July 2020 to April 2024 with ALT or AST ≥ 3x ULN and jaundice as meeting nR value ≥ 5, AP < 2x ULN, both, or neither. We determined positive predictive values (PPVs) and sensitivities for HC DILI-related fatality (death or liver transplant) and acute liver failure (ALF).

Results: Of 1314 liver injuries across 73 drug applications, 294 (22%) were attributed to DILI; 55 had ALT or AST ≥ 3x ULN and TB > 2x ULN. We excluded three cases (Gilbert's, high baseline enzymes, hepatitis B reactivation). Of 52 remaining, 16 (31%) met nR ≥ 5, five (10%) AP < 2x ULN, 21 (40%) both, and 10 (19%) neither. There were four DILI fatalities. Excluding one cholestatic fatality, nR ≥ 5 and AP < 2x ULN had PPVs for HC DILI fatality of 8 and 4%, respectively; sensitivities were 100 and 33%, respectively. One patient survived HC DILI-related ALF. Including this ALF case with the fatalities, nR ≥ 5 and AP < 2x ULN had PPVs of 11 and 4%, respectively; sensitivities were 100 and 25%, respectively. All fatalities and ALF cases were due to different drugs.

Conclusion: While the number of cases with the most severe DILI outcomes was small, particularly those that resulted in fatalities or ALF, nR ≥ 5 better approximated Hy's Law and was more sensitive than AP < 2x ULN in detecting fatalities and ALF.

Introduction: In drug-safety monitoring systems, adverse events (AEs) associated with the use of medical products often consist of complex patterns of clinical events. Network analysis (NA) was used for pattern recognition and characterizing the Vaccine Adverse Event Reporting System (VAERS), but limited applications of NA to the US Food and Drug Administration (FDA) Adverse Event Reporting System (FAERS) left its network description incomplete.

Methods: In this analysis, the network properties of FAERS were characterized and leveraged to facilitate pattern discovery. Reported AE information in FAERS is represented using preferred terms (PTs) in Medical Dictionary for Regulatory Activities terminology. The FAERS subsets were analyzed with drugs and PTs as nodes and interconnections as edges. Global characteristics, like the scale-free nature of the distribution, were examined to explore theoretical and structural considerations. Metrics that assess connectivity and edge weighting algorithms based on report co-occurrence or clustering were applied.

Results: Serious AE reports from 2016 to 2023 (2,062,099) were represented as a network of 20,965 nodes (16,847 PTs and 4116 drugs) with more than four million interconnections. Characteristics of FAERS subnetworks were determined with heavy-tailed degree distributions, high local clustering, and low diameters. Complexities related to structural and evolutionary characteristics were revealed as the log-normal model fits the degree distribution better than the power law.

Conclusions: Network-based techniques identified clinically relevant patterns and clustering patterns representative of known adverse drug reactions. Comparisons to VAERS reveal similarities in networks of AE reporting systems. This initial systematic application of NA to FAERS describes the overall network characteristics of the FAERS database and provides insight into the use of network applications in drug safety research.

Introduction: Dementia is the most prevalent neurodegenerative disorder. Several studies have demonstrated an association between anticholinergic drug use and an increased risk of cognitive and physical impairment. However, anticholinergic drugs are commonly prescribed for various clinical conditions, and their cumulative effects, referred to as anticholinergic burden, can contribute to cognitive decline and dementia. Although the causal relationship remains inconclusive, a higher anticholinergic burden is linked to a greater risk of cognitive deterioration.

Objective: This study aims to assess the compliance of patients aged ≥ 65 years with the STOPP-START and Beers criteria concerning the concurrent use of medications with high anticholinergic potency in situations where their use is not clinically justified.

Methods: This observational descriptive study was conducted using data from the Spanish Database for Pharmacoepidemiological Research (BIFAP). The study population comprised male and female patients aged ≥ 65 years.

Results: Of the 81,405 patients who developed dementia during the study period, 46.7% had been exposed to multiple anticholinergic drugs. Among them, 81.1% used these drugs sequentially, while 18.9% used two or more simultaneously. The absolute risk of developing dementia was 6.5% in patients who met the STOPP-START and BEERS criteria, compared to 13.4% in those who did not.

Conclusion: Although a high anticholinergic burden is a risk factor for cognitive decline, the unjustified concurrent use of multiple anticholinergic drugs remains uncommon among the elderly population in Spain.

Introduction: Adverse drug reactions (ADRs), including those resulting from drug interactions, remain a leading cause of morbidity and mortality. Structured product labels (SPLs) serve as a primary source for drug safety information. Having machine-readable product labels, including adverse reactions (ARs) and drug interactions, readily available would allow researchers to streamline medication safety studies. However, extracting this information is complex and requires the use of natural language processing (NLP) methods.

Objective: In this study, we explored the application of generative language models in the extraction of drug safety information from SPLs.

Methods: We compared multiple generative LLMs (GPT, Llama, and Mixtral) to two baseline methods in the task of extracting adverse reactions (ARs) from SPLs. We explored various factors, such as prompting strategies and term complexity, that impact the performance of these models in the extraction of ARs. Finally, we explored the generative models' capacity to extract drug interactions from a separate section of SPLs without additional fine-tuning or training, demonstrating their flexibility and adaptability for information retrieval.

Results: We found that generative language models, specifically GPT-4, are able to match or exceed the performance of previous state-of-the-art models without additional training or fine-tuning. Additionally, we found that the specific SPL section, surrounding context, and complexity of the AR term impacted the extraction performance. Finally, we demonstrated the generalizability of these models by applying them to a separate task of extracting drug names from the drug interaction section where curated training data are not available.

Conclusion: Generative language models demonstrate significant potential for automating drug safety information extraction from SPLs, offering a promising avenue for improving post-market surveillance and reducing ADRs. Future work should focus on refining prompting strategies and expanding the models' capabilities to handle increasingly complex and nuanced drug safety information.

Disproportionality analysis is used by many pharmacovigilance organizations for detecting and assessing signals of potential adverse drug reactions. However, its goal is often misunderstood and the approach misapplied, leading to erroneous conclusions due to neglected violated assumptions. In this paper we illustrate how simplistic use and interpretation of disproportionality analysis can lead to incorrect conclusions. Using VigiBase, the WHO global database of adverse event reports, and the Information Component disproportionality metric, we provide selected examples to highlight common sources of error that can introduce spurious disproportionalities or lead to missing important signals: confounding (by age, sex, indication, comedication), effect modification (by age), notoriety bias, masking, misclassification (by miscoding, incomplete or imprecise event retrieval), neglecting report utility, and violated independence assumption. Additionally, we present how sophisticated analyses may introduce new biases or amplify existing ones, such as collider bias or masking amplification. Due to its pitfalls, disproportionality analysis plays a supportive rather than decisive role in signal detection and assessment. Careful design and interpretation of disproportionality analysis, with appropriate subgrouping and clinical assessment, are essential. While subgrouping can mitigate some pitfalls, it reduces sample size and may introduce or amplify existing biases and needs to be used with care. Further development of tools to detect and mitigate biases in disproportionality analyses, and to assess their risk of bias, is needed.

Medical devices are indispensable in modern healthcare. They enable the prevention, diagnosis, and treatment of diseases while enhancing patient outcomes. However, the increasing complexity of these devices, particularly those incorporating advanced technologies such as artificial intelligence (AI) introduces new challenges to their safe use. The vulnerabilities of medical devices can lead to adverse events ranging from minor complications to severe injuries or fatalities, and there is an increasing health risk to those devices that are interconnected to electronic health management systems and internet protocols. Despite efforts by regulatory authorities such as the US Food and Drug Administration (FDA), European Medicines Agency (EMA), and Health Canada, disparities in reporting systems and monitoring practices persist globally, hindering effective safety oversight. This paper explores the current landscape of medical device safety, focusing on regulatory frameworks, reporting systems, and the challenges posed by fragmented data collection and underreporting. It highlights the critical role of postmarket surveillance (PMS) in identifying risks and ensuring device performance in real-world settings. The integration of emerging technologies, such as AI for predictive safety and blockchain for traceability, offers promising solutions to enhance monitoring and mitigate risks early in the device lifecycle. In addition, the paper examines harmonization efforts led by organizations such as The International Medical Device Regulators Forum (IMDRF), the International Society of Pharmacovigilance (ISoP) and the World Health Organazition (WHO), which aim to standardize reporting practices and improve global collaboration. Key recommendations include leveraging real-world data, enhancing cybersecurity measures, and fostering international cooperation to streamline regulatory processes. By addressing these challenges and embracing innovation, stakeholders can ensure that medical devices continue to advance healthcare while maintaining the highest safety standards. Such collective efforts are essential for safeguarding patient trust and improving global health outcomes.

Background: Adverse event reporting systems are an important source of safety signals for drug use in pregnancy, but their usefulness in the identification of potential drug-drug interactions (DDIs) remains unclear.

Objective: Our objective was to explore the reliability of signal detection for pharmacokinetic DDIs during pregnancy in adverse event reporting systems, focusing on potential interactions between antipsychotics (APs) or antidepressants (ADs) and drugs modifying cytochrome P450 (CYP450) activity, increasing the occurrence of gestational diabetes mellitus (GDM).

Methods: Reports related to the use of drugs during pregnancy were identified in VigiBase, the World Health Organization (WHO) global database of adverse event reports. Potential interacting drugs were selected based on WHO Drug Standardised Drug Groupings for CYP450 isoenzymes involved in the metabolic pathway of the AP or AD of interest. We conducted statistical interaction analysis using the omega disproportionality measure and including concomitant medication to identify potential DDIs, followed by a case series review for supporting evidence. Evaluation was subjective by author consensus.

Results: Of the 30 drug-drug-event combinations considered, statistical signals emerged for escitalopram, citalopram, and sertraline and the simultaneous use of CYP2D6 inhibitors with a higher relative reporting rate of GDM. However, case series review of reports did not support the existence of these DDIs because of uncertainties regarding the actual timing of medication use reported as concomitant.

Conclusion: Statistical signals of DDIs between ADs and potential interacting drugs during pregnancy were identified but not pursued further after case reviews. Uncertainty around medication use and event timing affected the reliability of the outcomes. These findings highlight the need to validate signals using detailed report data and stress the importance of accurate medication reporting.