Drug Safety最新文献

Background: Psychedelics are gaining attention for their therapeutic potential in modern and personalized medicine. Online forums such as Erowid provide valuable user insights, but analyses of these experiences using natural language processing (NLP) remain scarce.

Objective: This study aims to utilize NLP, including sentiment and lexicon analysis, to examine user-generated experience reports on psilocybin-containing mushrooms and LSD from the Erowid forum.

Methods: Data from 2188 Erowid users (1161 psilocybin mushrooms and 1027 LSD) was collected via automated web scraping with XPath, CSS selectors, and Selenium WebDriver. The dataset included report titles, substances, and demographics. Sentiment analysis utilized BERT, RoBERTa, and VADER models. Preprocessing involved tokenization, lemmatization, part-of-speech tagging, and stop-word filtering. Lexicon analysis identified themes through recurring n-grams, visualized using Python.

Results: User demographics revealed comparable ages for psilocybin mushrooms (23.8 ± 0.9 years) and LSD users (20.0 ± 0.6 years), with a predominance of male users. The BERT model predominantly labeled experiences as negative (unfavorable), particularly for mushroom users (p = 0.001). VADER indicated more positive experiences for mushroom users (p < 0.001), while RoBERTa mainly classified experiences as negative or neutral. Significant gender differences were found only with VADER, where more male users expressed positive opinions about psilocybin mushrooms (74.09% versus 65.52%, p < 0.021). The VADER model yielded more polarized results, whereas RoBERTa's cautious classifications indicate its suitability for analyzing lengthy and complex psychedelic reports. Further, RoBERTa outperformed other transformer-based models, achieving the highest accuracy. Lexicon analysis revealed emotional, sensory, and temporal themes, with psilocybin reports emphasizing introspection and time dilation phenomenon, while LSD reports highlighted memory issues and cognitive disorientation.

Conclusions: Sentiment analysis showed that VADER produced more polarized results, while RoBERTa offered cautious classifications with the highest accuracy. Lexicon analysis revealed shared themes, with mushroom reports focusing on introspection and time dilation perception, while those of LSD emphasized cognitive disturbances. This study highlights the value of these analyses in understanding psychedelic experiences, informing harm reduction, and guiding policy-making.

Introduction: Recent artificial intelligence (AI) advances can generate human-like responses to a wide range of queries, making them a useful tool for healthcare applications. Therefore, the potential use of large language models (LLMs) in controlled environments regarding efficacy, reproducibility, and operability will be of paramount interest.

Objective: We investigated if and how GPT 3.5 and GPT 4 models can be directly used as a part of a GxP validated system and compared the performance of externally hosted GPT 3.5 and GPT 4 against LLMs, which can be hosted internally. We explored zero-shot LLM performance for named entity recognition (NER) and relation extraction tasks, investigated which LLM has the best zero-shot performance to be used potentially for generating training data proposals, evaluated the LLM performance of seven entities for medical NER in zero-shot experiments, selected one model for further performance improvement (few-shot and fine-tuning: Zephyr-7b-beta), and investigated how smaller open-source LLMs perform in contrast to GPT models and to a small fine-tuned T5 Base.

Methods: We performed reproducibility experiments to evaluate if LLMs can be used in controlled environments and utilized guided generation to use the same prompt across multiple models. Few-shot learning and quantized low rank adapter (QLoRA) fine-tuning were applied to further improve LLM performance.

Results and conclusion: We demonstrated that zero-shot GPT 4 performance is comparable with a fine-tuned T5, and Zephyr performed better than zero-shot GPT 3.5, but the recognition of product combinations such as product event combination was significantly better by using a fine-tuned T5. Although Open AI launched recently GPT versions to improve the generation of consistent output, both GPT variants failed to demonstrate reproducible results. The lack of reproducibility together with limitations of external hosted systems to keep validated systems in a state of control may affect the use of closed and proprietary models in regulated environments. However, due to the good NER performance, we recommend using GPT for creating annotation proposals for training data as a basis for fine-tuning.

Introduction: The COVID-19 pandemic accelerated new vaccine development. Limited safety data necessitated robust global safety surveillance to accurately identify and promptly communicate potential safety issues. The African Union Smart Safety Surveillance (AU-3S) program established the Joint Signal Management (JSM) group to support identification of potential vaccine safety concerns in five pilot countries (Ethiopia, Ghana, Kenya, Nigeria, South Africa), accounting for approximately 35% of the African population.

Objective: Our objective was to provide an overview of the JSM group's role in supporting signal management activities for the AU-3S program during the COVID-19 pandemic.

Methods: Spontaneous, electronically reported COVID-19 vaccine adverse events following immunization (AEFI) from each country's safety data were integrated into the interim Data Integration and Signal Detection system. Statistical disproportionality methods were used to identify and review vaccine-event combinations (VECs) for potential safety concerns. The JSM group-which comprised pharmacovigilance and subject matter experts from National Medicine Regulatory Authorities, Expanded Programs on Immunization, and vaccine safety committees-conducted signal detection activities on cross-country safety data and provided recommendations.

Results: From April 2021 to December 2023, a total of 48,294 spontaneously reported AEFI were analyzed for six COVID-19 vaccines (NRVV Ad [ChAdOx1 nCoV-19]; Ad26.COV2.S; Elasomeran; Tozinameran; Covid-19 vaccine [Vero Cell], Inactivated; NRVV Ad26 [Gam-Covid-Vac]) administered in Ethiopia (34.6%), Nigeria (30.3%), South Africa (16.9%), Ghana (13.5%), and Kenya (4.7%). Overall, 2,742 VECs were validated. A causal association between the COVID-19 vaccines and the reported AEFI cannot be inferred, as data were reported spontaneously. JSM group recommendations included monitoring for further evidence, no immediate action required, engaging marketing authorization holder(s) for additional information, or sensitizing healthcare providers and/or the public about events. Although no new safety signals were identified, nine safety-related recommendations were issued, including patient and healthcare provider education.

Conclusions: The JSM group established a scalable and replicable model for future signal management of other priority health products in low- and middle-income countries, fostering ongoing collaboration and capacity building. Knowledge and experience gained from this pilot initiative will guide stakeholders in future safety surveillance initiatives within the African continent.

Background: On June 8, 2021, a new 20-valent pneumococcal conjugate vaccine (PCV20, PREVNAR 20^®, Pfizer, Inc.) was licensed for use in adults aged ≥ 18 years by the US Food and Drug Administration (FDA).

Objective: To describe reports to the Vaccine Adverse Event Reporting System (VAERS) after administration of the 20-valent pneumococcal conjugate vaccine in adults.

Methods: We searched the VAERS for reports of adverse events involving persons aged ≥ 19 years who received PCV20 during October 20, 2021, through December 31, 2023. Our evaluation included automated analysis of reports, clinical review of serious reports and pre-specified events of special interest, empirical Bayesian data mining to assess for disproportionate reporting, and estimation of reporting rates for reports of Guillain-Barré syndrome (GBS).

Results: The VAERS received 1976 reports after PCV20 administration in persons aged ≥ 19 years (6% of reports involved serious events). The most common adverse events among persons aged 19-64 years (n = 798) were injection-site reactions (231, 29%), pain (134, 17%), erythema (118, 15%), and fever (117, 15%). For persons aged ≥ 65 years (n = 1178), the most common adverse events were injection-site reactions (417, 35%), pain (180, 15%), pain in extremity (162, 14%), and erythema (158, 13%). A data mining alert (EB05 = 3.812) for the MedDRA Preferred Term "Guillain-Barre syndrome" was observed for serious reports. Clinical review verified 11 of 20 GBS reports; 7/11 vaccine recipients were aged ≥ 65 years. Among the 11 verified cases, the median time from vaccination to symptom onset was 14 days. Five persons received another vaccine on the same visit. The reporting rate of GBS after PCV20 receipt was 0.5 cases per million doses distributed. No other safety concern was identified.

Conclusions: During the period of this post-licensure review of PCV20, we found most reports were non-serious and comprised mostly local and systemic (e.g., fever) reactions consistent with prelicensure studies. In serious reports, we also identified a data mining alert for GBS after receipt of PCV20, which Centers for Disease Control and Prevention and the FDA are investigating further. No other new or unexpected safety concern was identified.

Introduction and objective: Proton pump inhibitor (PPI) use in children increases the risk of infections, prompting inquiry into the impact of prenatal PPIs exposure on serious infections in offspring. As a research gap in this area exists, this study aimed to address it by assessing the association between prenatal PPIs exposure and serious infections in infants during their first year of life.

Methods: Using the French health insurance data warehouse (SNDS) (2013-2018), we conducted a retrospective cohort study on singleton, full-term liveborn non-immunocompromised infants, stratified by PPI use during the first three months of life (early-life use). Proton pump inhibitor dispensing in ambulatory care settings during pregnancy defined exposure. Outcomes concerned any serious infections in offspring aged between 3 and 12 months. Adjusted odds ratios (aORs) were estimated using logistic regression with multivariable models to control for potential confounders.

Results: Of the 2,485,545 infants included, 497,060 (23.3%) were prenatally exposed to PPIs and 97,767 (4.6%) had PPI use during the first three months of life. Prenatal PPI exposure was associated with serious infections in offspring (aOR, 1.09 [95% CI, 1.07-1.10]) in infants without early-life PPIs use. No association was found for infants with early-life PPI use (aOR, 1.05 [95% CI, 1.00-1.11]). Gastrointestinal infections were the sole site with persistent significance.

Conclusion: Prenatal PPI exposure is common and is not associated with a major risk of serious infections in infants during their first year. However, even after adjusting for several confounding factors, a weak association remains, especially in infants without early-life PPI use. While offering reassurance, adherence to clinical guidelines is still crucial.