Drug Safety最新文献

Introduction: Improving adverse events following immunisation (AEFI) detection is vital for vaccine safety surveillance, as an early safety signal can help minimize risks. In February 2022, the World Health Organization reported a preliminary signal on sudden sensorineural hearing loss (SSNHL) following coronavirus disease 2019 (COVID-19) vaccination, 54 million persons in France received at least one dose, covering 78.8% of the population within a year.

Objective: The primary objective of this study was to identify a method of disproportionality analysis capable to detect a safety signal for hearing impairment (HI) as early as possible during the initial phases of the COVID-19 vaccination campaign. Secondly, we described all cases of SSNHL reported during vaccine booster campaigns in France.

Methods: Data from January 2011 to February 2022 were extracted from the French pharmacovigilance database. Cases were all spontaneous reports of AEFI for elasomeran and tozinameran, while non-cases were AEFI reported for other vaccines. Disproportionality analysis for HI was performed monthly during 2021, to estimate a reporting odds ratio (ROR). Four different methods were used for ROR estimation. Furthermore, we reviewed cases of SSNHL following messenger RNA COVID-19 vaccinations reported during booster campaigns, from 2 February 2022 to 1 March 2023, based on a comprehensive medical evaluation.

Results: Using a standard methodology, we identified a signal on 31 July 2021 (ROR 1.50, 95% confidence interval [CI] [1.06-2.18]). Multivariate analysis adjusted for sex, age, ototoxic drugs and excluding reference reports of common AEFI for vaccines allowed us to detect the HI signal as early as 31 March 2021 (ROR 2.67, 95% CI [1.36-5.57]). The SSNHL reporting rate was estimated to be 0.83/1,000,000 doses for tozinameran and 4.3/1,000,000 for elasomeran during the booster campaigns.

Conclusion: Using a well-structured disproportionality analysis could have enhanced early detection of safety signals and contribute to risk minimizing measures. According to descriptive data, HI following mRNA COVID-19 vaccines remains rare.

HIV-prevention efforts focusing on women of child-bearing potential are needed to end the HIV epidemic in the African region. The use of antiretroviral drugs as pre-exposure prophylaxis (PrEP) is a critical HIV prevention tool. However, safety data on new antiretrovirals during pregnancy are often limited because pregnant people are excluded from drug development studies. Calls from communities, healthcare professionals, and regulators to improve the information supporting decision-making around the use of medical products during pregnancy have been increasing. Post-marketing safety surveillance is an essential tool for detecting adverse outcomes and evaluating real-world, longer-term effects of drugs. Detecting and evaluating uncommon pregnancy outcomes requires large sample sizes, highlighting the benefits of and need for safety surveillance. Surveillance systems vary widely across Africa, and the need for enhanced surveillance of PrEP use during pregnancy highlights the limitations of current regulatory and surveillance systems. Challenges include weak regulation and insufficient resources. Pooling of resources and regulatory harmonization could address resource challenges. The African Medicines Agency, as a specialized agency of the African Union, has the potential to improve African medical product regulation, including post-marketing safety surveillance. This can strengthen regulation and ensure that market authorization holders meet their responsibility to invest in post-marketing surveillance systems, such as pregnancy registries. At the same time, independent post-marketing studies are needed to ensure generation of essential safety data. The Forum for Collaborative Research has initiated a project to facilitate interactions between regulators in Africa, the USA, and Europe, as well as other stakeholders, and to work toward consensus on safety data generation from PrEP during pregnancy before and after marketing authorization.

Background: Previous meta-epidemiological surveys have found considerable misinterpretation of results of disproportionality analyses. We aim to explore the relationship between the strength of causal statements used in title and abstract conclusions of pharmacovigilance disproportionality analyses and the strength of causal language used in citing studies.

Methods: On March 30, 2022, we selected the 30 disproportionality studies with the highest Altmetric Attention Scores. For each article, we extracted all citing studies using the Dimension database (n = 1434). In parallel, two authors assessed the strength of causal statements in the title and abstract conclusions of source articles and in the paragraph of citing studies. Based on previous studies, the strength of causal language was quantified based on a four-level scale (1-appropriate interpretation; 2-ambiguous interpretation; 3-conditionally causal; 4-unconditionally causal). Discrepancies were solved by discussion until consensus among the team. We assessed the association between the strength of causal statements in source articles and citing studies, separately for the title and abstract conclusions, through multinomial regression models.

Results: Overall, 27% (n = 8) of source studies used unconditionally causal statements in their title, 30% (n = 9) in their abstract conclusion, and 17% (n = 5) in both. Only 20% (n = 6) used appropriate statements in their title and in their abstract's conclusions. Among the 622 citing studies analyzed, 285 (45.8%) used unconditionally causal statements when referring to the findings from disproportionality analysis, and only 164 (26.4%) used appropriate language. Multinomial models found that the strength of causal statements in citing studies was positively associated with the strength of causal language used in abstract conclusions of source articles (Likelihood Ratio Test (LogLRT) p < 0.00001) but not in the titles. In particular, among studies citing source articles with appropriate interpretation, 30.2% (95% confidence interval [CI] 22.8-37.6) contained unconditionally causal statements in their abstract conclusions, versus 56.4% (95% CI 48.7-64.2) for studies citing source articles with unconditionally causal statements.

Conclusions: Nearly half of the studies citing pharmacovigilance disproportionality analyses results used causal claims, particularly when the causal language used in the source article was stronger. There is a need for higher caution when writing, interpreting, and citing disproportionality studies.

Introduction: Transformer-based large language models (LLMs) have transformed the field of natural language processing and led to significant advancements in various text processing tasks. However, the applicability of these LLMs in identifying related drug-adverse event (AE) pairs within clinical context may be limited by the prevalent use of non-standard sentence structures and grammar.

Method: Nine transformer-based LLMs pre-trained on biomedical domain corpora are fine-tuned on annotated data (n = 5088) to classify drug-AE pairs in unstructured discharge summaries as causally related or unrelated. These LLMs are then validated on text segments from deidentified hospital discharge summaries from Singapore (n = 1647). To assess generalisability, the models are validated on annotated segments (n = 4418) from the Medical Information Mart for Intensive Care (MIMIC-III) database. Performance of LLMs in identifying related drug-AE pairs is then compared against a prior benchmark set by traditional machine learning models on the same data.

Results: Using an LLM-Bidirectional long short-term memory (LLM-BiLSTM) architecture, transformer-based LLMs improve F1 score as compared to prior benchmark with BioM-ELECTRA-Large-BiLSTM showing an average F1 score improvement of 16.1% (increase from 0.64 to 0.74). Applying additional rules on the LLM-based predictions, like ignoring drug-AE pairs when the AE is a known indication of the drug, results in a further reduction in false positive rates with precision increases of up to 5.6% (0.04 increment).

Conclusion: Transformer-based LLMs outperform traditional machine learning methods in identifying causally related drug-AE pairs embedded within unstructured discharge summaries. Nonetheless the improvement in performance with rules indicates that LLMs still possess some degree of imperfection for this causal relation detection task.

Background: In Switzerland, consumers are exposed to drugs with pharmacogenetic (PGx) recommendations in 78% of cases. Pre-emptive PGx testing for seven drugs (abacavir, carbamazepine, 6-mercaptopurine, azathioprine, 5-fluorouracil, capecitabine, and irinotecan) has been covered by basic health insurance since 2017. PGx testing for other drugs is only covered if it is reactive and prescribed by a clinical pharmacologist. No data are yet available on the implementation of PGx testing in the outpatient setting.

Aim: The objective of this study was to determine the prevalence of ambulatory PGx testing in the Swiss population, to characterize PGx-tested individuals, and to identify the most commonly used drugs before and after PGx testing.

Methods: We assessed the prevalence of PGx testing in Switzerland and characterized individuals who underwent PGx testing between 2017 and 2021 using claims data from a large health insurance company.

Results: Of 894,748 individuals registered for the entire study period, only 817 (0.09%) underwent PGx testing. Those who underwent PGx testing were more frequently female and claimed more drugs and PGx drugs than those who did not undergo PGx testing. The drugs used before and after PGx testing differed, and fewer drugs with reimbursement for pre-emptive PGx testing were included before PGx testing.

Conclusion: In Switzerland, personalized pharmacotherapy has the potential to be improved, as only 0.09% of the studied population underwent PGx testing, despite 77.4% claiming PGx drugs.

Background: Drug adverse events (AEs) represent a significant public health concern. US Food and Drug Administration (FDA) drug labeling documents are an essential resource for studying drug safety such as assessing a drug's likelihood to cause certain organ toxicities; however, the manual extraction of AEs is labor-intensive, requires specialized expertise, and is challenging to maintain, due to frequent updates of the labeling documents.

Objective: To automate the extraction of AE data from FDA drug labeling documents, we developed a workflow based on AskFDALabel, a large language model (LLM)-powered framework, and its demonstration in drug safety studies.

Methods: This framework incorporates a retrieval-augmented generation (RAG) component based on FDALabel to enhance standard LLM inference. Key steps include (1) selection of a task-specific template, (2) FDALabel database querying, and (3) content preparation for LLM processing. We evaluated the performance of the framework in three benchmark experiments, including drug-induced liver injury (DILI) classification, drug-induced cardiotoxicity (DICT) classification, and AE term recognition.

Results: AskFDALabel achieved F1-scores of 0.978 for DILI, 0.931 for DICT, and 0.911 for AE annotation, outperforming other traditional methods. It also provided cited labeling content and detailed explanations, facilitating manual verification.

Conclusion: AskFDALabel exhibited high consistency with human AE annotation, particularly in classifying and profiling DILI and DICT. Thus, it can significantly enhance the efficiency and accuracy of AE annotation, with promising potential for advanced AE surveillance and drug safety research.

The Medical Dictionary for Regulatory Activities (MedDRA) was developed in the mid-to-late 1990s to address the shortcomings of other medical dictionaries used for coding adverse events. Since that time, MedDRA has become the required coding dictionary for major regulatory authorities involved with the International Council for Harmonisation. Due to the increased specificity and significant increase in terminology over time, several approaches have been developed to aggregate terms for the purposes of signal detection and labeling. We present the approaches taken and suggested to date to aggregate preferred terms into meaningful medical concepts. We discuss the pros and cons of different methods in which to group terms, and illustrate that analyses performed for MedDRA preferred terms can also be conducted using aggregated terms. However, some features of adverse events available at the preferred term level, such as severity and relationship to study therapy, require additional consideration for analysis. In the last 25 years, the pendulum for medical coding is swinging in the other direction. Faced with a deluge of preferred terms, users of MedDRA are developing new ways in which to collapse terms into medical concepts. The ability to identify safety concerns and communicate important data in drug labels effectively and consistently are at risk, particularly with the introduction of new aggregations.

Introduction: Despite the increased scrutiny on vaccine safety following the coronavirus disease 2019 (COVID-19) pandemic, Morocco's Centre of Antipoison and Pharmacovigilance (CAPM) remained concerned that the pharmacovigilance system in Morocco was insufficiently implemented, including limited adverse event (AE) reporting, poor data use, and inconsistent training nationwide.

Objectives: We sought to assess the status of pharmacovigilance activities (PAs) prior to formally institutionalizing them across university hospital centers (UHCs), given their position as the main providers of healthcare in Morocco and key sources for reporting serious AEs.

Methods: We assessed seven UHCs (housing 30 hospitals) in 2023 using a structured questionnaire with pharmacovigilance focal points developed from the World Health Organization's indicators of pharmacovigilance and the Global Benchmarking Tool. Data were grouped into 28 PAs and scored from 0 (not implemented) to 3 (fully implemented). We then calculated an implementation rate for each site on the basis of percent of PAs fully implemented (≥ 70%, well established; > 40% to < 70%, partially implemented; and ≤ 40%, not implemented). A desk review was also performed at the sites. Using the results of the assessment, three working groups of pharmacovigilance stakeholders developed recommendations to be formally adopted by UHCs.

Results: Basic elements of pharmacovigilance (notification forms and VigiFlow^® or Excel databases) were present at all the UHCs assessed. In total, 14 hospitals (47%) had well-established PAs, including advanced activities such as signal detection of adverse events following the use of medicines and vaccines, as well as causality assessment; 9 hospitals (30%) were partially implementing pharmacovigilance, and 7 hospitals (23%) had no established activities or very basic activities. Within four UHCs, activities had not been implemented at the same level from one hospital to another and vaccine vigilance was largely deprioritized. The working groups made recommendations for improving collaboration, standardizing procedures, and outlining a new organizational structure for pharmacovigilance, which was institutionalized by a formal agreement among UHCs in July 2023.

Conclusions: The assessment revealed a subgroup of centers with well-established AE signal detection and causality assessment abilities, which could play a leading role in the country. After the site assessment, our collaborative approach of bringing together focal points to identify next steps and generate buy-in helped to formalize pharmacovigilance across centers.