While the World Health Organization defines pharmacovigilance as the science and activities involved in detecting, assessing, understanding, and preventing adverse effects and any other potential or observed problems associated with drugs, this field has evolved from a passive reporting function to an intelligence-driven process. It now integrates pharmacovigilance into the overall framework of pharmaceutical development and regulation (Silva et al., 2024). Over the past two decades, numerous safety-related cases, such as those involving thiazolidinediones, fluoroquinolones, and biologics, have prompted the development of regulations that incorporate pharmacovigilance findings into the quality assessment process by global regulators, including the USFDA, EMA, and Japan’s Pharmaceuticals and Medical Devices Agency (PMDA) (Badria and Elgazar, 2024).
Quality management in pharmaceuticals encompasses the organizational structure, roles and responsibilities, processes, and resources necessary for the successful implementation of quality assurance and quality control throughout the medicine lifecycle. Traditionally, pharmacovigilance and quality management have been viewed as separate activities within pharmaceutical organizations. However, an integrative approach is gaining prominence due to ICH guidelines such as ICH E2E (Pharmacovigilance Planning) and ICH Q10 (Pharmaceutical Quality System), which explicitly reference the use of pharmacovigilance information for quality decisions (Wasiullah et al., 2025).
With the introduction of spontaneous adverse event reporting systems, such as the FDA Adverse Event Reporting System (FAERS), EudraVigilance, and VigiBase, the volume of regulatory pharmacovigilance data has increased exponentially. Analyzing and interpreting this data, followed by its incorporation into regulatory decision-making—such as labeling updates, REMS changes, and product withdrawals—requires sophisticated approaches, expertise, and organizational capabilities (Nagar et al., 2025).
The period from 2020 to 2025 has been particularly influential in shaping contemporary pharmacovigilance practices, driven by the COVID-19 pandemic, rapid vaccine introduction, health data digitalization, and advancements in AI (Hunsel and Kant, 2025).
Despite years of regulatory progress, pharmacovigilance continues to operate in organizational silos that hinder its integration into quality and regulatory decision-making. First, with hundreds of millions of patient-level reports generated annually by spontaneous reporting systems, the issue of underreporting in pharmacovigilance persists alongside an overwhelming volume of spontaneous reports that exceeds the capacity of safety evaluation teams (Ball and Pan, 2022). Second, the lack of connection between pharmacovigilance outcomes and corrective and preventive actions within quality systems means safety risks are often not addressed in product manufacturing and sourcing processes (Palatty et al., 2024). Third, there is significant variability in how pharmacovigilance data is utilized by regulators across jurisdictions, leading to inconsistencies in decision-making based on this data (Mansuri and Zaman, 2024). Finally, regulations governing the practice of advanced therapy medicinal products (ATMPs), biosimilars, and combinations continue to lag behind scientific advancements in the field, leaving gaps for both the industry and patients (Pizevska et al., 2022).
The study will be guided by five key research questions. Specifically, RQ1 will focus on the signal detection methods used since 2020 and their incorporation into regulatory decision-making. RQ2 will explore the structural connections between quality management systems and pharmacovigilance. RQ3 will identify the role of artificial intelligence in current pharmacovigilance practices and the evidence-based limitations of this approach. RQ4 will examine the impact of harmonization on the integration of pharmacovigilance into quality and decision-making processes. Lastly, RQ5 will assess pharmacovigilance-related challenges specific to biologics and advanced therapy medicinal products (ATMPs).
The aim of the literature review is to provide an evidence-based analysis of academic articles from 2020 to 2025 related to the integration of pharmacovigilance into quality management systems and regulatory decision-making processes.
2. Materials and Methods
2.1 Ethical approval statement
No ethical approval is required for this study.
2.2 Search strategy and databases
The current study employs the principles of narrative synthesis, following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. The search strategy was executed using the PubMed/MEDLINE, Scopus, and Web of Science databases, incorporating MeSH terms and free-text terms such as pharmacovigilance, drug safety signal detection, regulatory decision-making, benefit-risk assessment, pharmacovigilance quality systems, real-world evidence, adverse drug reaction reporting, artificial intelligence in pharmacovigilance, and ICH guidelines (Figure 1).
Figure 1. PRISMA flow diagram illustrating the systematic literature search and study selection process, including identification of records from PubMed, Scopus, and Web of Science, screening, eligibility assessment, and final inclusion of studies (2020–2025).
2.3 Inclusion and exclusion criteria
To be included in this review, each article must have been published in peer-reviewed journals between January 2020 and December 2025, be available in English, and discuss pharmacovigilance methodologies, regulatory integration, or the linkage of quality systems. Studies that did not provide a full text, were gray literature without peer review, or focused solely on non-medicinal products were excluded.
2.4 Study selection and data extraction
After screening more than 200 articles, we selected 15 manuscripts based on their relevance, methodology, and topicality. The chosen articles were coded thematically to identify areas of convergence and divergence in their results. Each publication was then evaluated based on the research questions posed, methodology used, conclusions drawn, potential limitations discussed, and implications for integrating PV quality and regulatory practices. A summary table of all the selected articles is provided below.
2.5 Definition of key terms
Pharmacovigilance is the science of detecting, assessing, understanding, and preventing adverse effects or any other medicine-related problems throughout the life cycle of a medication (Herdeiro et al., 2021). Adverse Drug Reactions (ADRs) are noxious and unintended responses to medications that occur at normal doses in humans when administered for prophylactic, diagnostic, or therapeutic purposes (Aate, 2024). Individual Case Safety Reports (ICSRs) document suspected adverse drug reactions associated with one or more medications (Vemula et al., 2023). Signal detection involves identifying new, potentially causal associations between medications and adverse events within the safety database. Benefit-risk assessment evaluates the positive impacts of a drug compared to its negative aspects (Hammad et al., 2023). Real-world evidence (RWE) refers to data collected on the use, effectiveness, or safety of a product obtained from non-clinical trial environments. Quality Management Systems (QMS) are systematic approaches that document the processes, procedures, and roles necessary to achieve quality targets (Wang et al., 2022). Risk Evaluation and Mitigation Strategies (REMS) are FDA-required programs designed to maximize the benefits of a drug while minimizing its risks (Table 1). Disproportionality analysis is a statistical tool used to identify drug-adverse event pairs that occur more frequently than would be expected by chance. Advanced Therapy Medicinal Products (ATMPs) are innovative medicines that utilize gene, cell, and tissue therapies (Essink et al., 2025).
Table 1. Definition of key terms used in this review.
| Term | Definition |
| Pharmacovigilance (PV) | The science and activities concerned with detecting, assessing, understanding, and preventing adverse effects or other drug-related problems throughout the medicine lifecycle. |
| Adverse Drug Reaction (ADR) | A response to a medicine that is noxious, unintended, and occurs at doses normally used in humans for prophylaxis, diagnosis, or therapy. |
| Individual Case Safety Report (ICSR) | A report describing one or more suspected adverse drug reactions in relation to one or more medicinal products. |
| Signal detection | The process of identifying new, potentially causal associations between a drug and an adverse event from safety databases. |
| Benefit-risk assessment | A structured evaluation of the positive therapeutic effects of a medicine relative to its risks, used in regulatory decision-making. |
| Real-World Evidence (RWE) | Clinical evidence about the usage, benefits, or risks of a product derived from real-world data sources outside clinical trials. |
| Quality Management System (QMS) | A formalized system documenting processes, procedures, and responsibilities for achieving quality policies and objectives. |
| Risk Evaluation and Mitigation Strategy (REMS) | A drug safety program required by the FDA to ensure benefits outweigh risks. |
| Disproportionality analysis | A statistical method to identify drug-adverse event combinations reported more frequently than expected by chance in spontaneous reporting databases. |
| ATMP | Advanced therapy medicinal product, a category of innovative medicines based on genes, tissues, or cells. |
3. Results and Discussion
3.1 Overview of reviewed literature
This study presents an organized overview of 15 sources from 2020 to 2025, categorized into five major themes: methods of signal detection, integration of pharmacovigilance (PV) quality systems, applications of AI technology, harmonization of regulation and benefit-risk assessment, and pharmacovigilance in innovative therapies (Table 2).
Table 2. Summary of reviewed publications (2020–2025).
| Study | Research focus | Technical approach | Key contribution |
| Herdeiro et al. (2021) | Future of pharmacoepidemiology and drug safety | Conceptual and exploratory analysis | Established foundational insights into evolving drug safety practices and future research directions |
| Ball and Pan (2022) | AI in pharmacovigilance readiness | Review of AI applications and limitations | Evaluated the maturity of AI in pharmacovigilance and highlighted readiness gaps |
| Pizevska et al. (2022) | Advanced therapy medicinal products (ATMPs) translation in Europe | Developer-centric qualitative analysis | Identified barriers and enablers in ATMP development and regulatory pathways |
| Wang et al. (2022) | Clinical source data production and quality control in real-world studies | eSource system framework proposal | Proposed a structured system for improving data quality and traceability in clinical research |
| Vemula et al. (2023) | Risk-based safety case follow-up in pharmacovigilance | Risk-based analytical framework | Introduced prioritization strategy for adverse event follow-ups |
| Hammad et al. (2023) | Causality assessment of safety signals | Conceptual expansion of assessment frameworks | Enhanced methodologies for evaluating drug safety signals beyond traditional approaches |
| Silva et al. (2024) | Precision pharmacovigilance in personalized medicine | Data-driven and personalized monitoring approaches | Highlighted integration of personalized medicine with pharmacovigilance systems |
| Badria and Elgazar (2024) | Optimizing pharmacovigilance in an era of accelerating innovation | System-level analysis and innovation strategies | Proposed improvements for pharmacovigilance in rapidly evolving innovation environments |
| Palatty et al. (2024) | Challenges in pharmacovigilance | Review-based analysis | Identified operational and regulatory challenges with mitigation strategies |
| Mansuri and Zaman (2024) | Pharmacovigilance as a global drug safety monitoring tool | Descriptive and systematic review | Reinforced the importance of pharmacovigilance as a global healthcare tool |
| Aate (2024) | Management of adverse drug reactions | Review of clinical practices | Summarized best practices for ADR identification and management |
| Wasiullah et al. (2025) | ICH pharmacovigilance guideline framework | Regulatory framework analysis | Provided structured global standards for pharmacovigilance practices |
| Nagar et al. (2025) | Artificial intelligence in pharmacovigilance and regulatory integration | AI-driven system analysis | Demonstrated how AI enhances drug safety monitoring and regulatory workflows |
| Hunsel and Kant (2025) | Pandemic preparedness in pharmacovigilance | Case study based on COVID-19 learnings | Highlighted lessons for future pharmacovigilance resilience |
| Essink et al. (2025) | Risk minimisation measures for ATMPs in the EU | Cross-sectional study | Provided insights into effectiveness of regulatory risk minimisation strategies |
3.2 Signal detection methodologies (RQ1)
Signal detection is the technical backbone of pharmacovigilance and its most immediate connection to regulatory decision-making. Disproportionality analysis techniques—namely PRR, ROR, and BCPNN—remain essential tools for mining spontaneous report databases. Recent scientific literature, published over the last six months, has further advanced the methodology by enhancing statistical significance, stratification methods, and validation criteria for these algorithms. The application of these methods, particularly in analyzing FAERS and EudraVigilance data for COVID-19 vaccines and treatments, vividly illustrates both their potential and pitfalls.
As noted in many studies, a significant issue related to the disproportional analysis of adverse effects is the bias arising from incomplete reporting. The assumption that a database contains all possible data on exposure and events is flawed; in reality, it is always incomplete due to underreporting, biases, and differences among exposed populations (Ball and Pan, 2022). Suggested solutions include providing incentives for reporting adverse events, implementing mandatory reporting using E2B R3 format, and incorporating additional information sources, such as claims data and electronic health records (EHRs). Research demonstrates that integrating databases with FAERS enhances the ability to confirm signals with increased specificity (Pizevska et al., 2022). This approach has gained traction among regulatory bodies, including the FDA Sentinel System and EMA DARWIN EU initiatives, which aim to mine real-world evidence (RWE).
Patient-reported outcomes are an underutilized source of quality-relevant information. Studies indicate that patients in oncology and dermatology report symptoms more quickly and comprehensively through digital channels than via conventional Individual Case Safety Reports (ICSRs) (Hammad et al., 2023). Integrating these reports into feedback loops necessitates efforts from quality management.
3.3 Artificial intelligence and machine learning in pharmacovigilance (RQ3)
AI has emerged as the fastest-growing frontier of methodologies in pharmacovigilance during the 2020–2025 period. The ability of NLP algorithms to automatically analyze and triage Individual Case Safety Reports (ICSRs) from social media, electronic health records, scientific literature, and other sources has significantly expanded the data horizon in safety surveillance (Hunsel and Kant, 2025). Machine learning models demonstrated high accuracy in distinguishing between serious and non-serious adverse events, as well as in extracting drug-event relationships from unstructured texts, thereby reducing manual workload for pharmacovigilance experts.
All papers under review agree that a major limitation of most AI tools, developed in an academic context, is the lack of prospective regulatory validation. This validation has not been performed due to concerns regarding explainability, reproducibility, and potential biases in the AI algorithms, particularly when models were based predominantly on data collected from Caucasians and written in English (Hunsel and Kant, 2025). The FDA’s 2023 discussion paper on the use of AI in drug development, along with the EMA’s reflection paper on employing machine learning in drug product development, illustrates how regulators are addressing these issues.
Federated learning is a promising technique for training models without accessing centralized patient data, allowing for decentralized AI model training through collaboration among different parties (Vemula et al., 2023). In a feasibility study, federated learning demonstrated compliance with GDPR requirements while still delivering satisfactory AI performance.
3.4 PV-quality system integration (RQ2)
Pharmacovigilance integration with quality management systems is the most challenging area for organizations discussed in this review. Classical models based on ISO 9001 and ICH Q10 Quality Management Systems emphasize process control, deviation management, and CAPAs, but do not explicitly address their association with pharmacovigilance and clinical safety signal detection (Palatty et al., 2024). Some studies have proposed ideas and pilot initiatives to include PSURs and RMP updates in a QMS process, triggering quality investigations if there are any manufacturing correlates to clinical safety signals (Wasiullah et al., 2025).
This example is particularly relevant in the context of biologics manufacture. Monoclonal antibodies and recombinant proteins may undergo subtle modifications and aggregation from batch to batch due to their molecular composition. These changes can first be detected through clinical adverse events and only later through analytical release testing (Palatty et al., 2024). Integrating the PV-QMS model to route adverse event signals of immunogenicity back to the manufacturing process in real time would be a breakthrough. The challenge lies in organizational resistance and the lack of integration between data management systems.
3.5 Regulatory harmonization and benefit-risk frameworks (RQ4)
The regulatory dimension is heavily influenced by the extent to which harmonization can be achieved among key agencies. The ICH E2 guidelines have paved the way for global harmonization of safety reporting standards; however, significant heterogeneity still exists. Studies comparing regulatory decisions made by the FDA, EMA, and PMDA based on the same data have revealed that differences in benefit-risk assessment, population, and culture lead to divergent decisions, despite the use of similar pharmacovigilance data (Badria and Elgazar, 2024). To minimize subjectivity in this process, structured benefit-risk frameworks have been developed, including the Benefit-Risk Action Team (BRAT) framework and the PrOACT-URL model. An assessment of these frameworks, based on 30 regulatory decisions, found that while structured frameworks increased transparency in decision-making, they could not eliminate variations in outcomes between regulatory agencies (Mansuri and Zaman, 2024).
One important downstream function of pharmacovigilance is risk communication, which involves interpreting pharmacovigilance information to inform regulatory action. This topic has gained significant attention among researchers in recent years. Studies analyzing the interpretation of Direct Healthcare Professional Communications (DHPCs) have reported considerable variance in interpretation by healthcare professionals. This suggests that the design of regulatory risk communication should incorporate behavior change strategies to achieve its intended purpose. Additionally, building capacity in pharmacovigilance within low- and middle-income countries is a crucial area that requires intervention. Research has shown a marked increase in Individual Case Safety Reports (ICSRs) following systematic investments in pharmacovigilance capacity-building activities.
3.6 Pharmacovigilance for novel therapeutic modalities (RQ5)
ATMPs, which include gene therapies, cell therapies, and tissue-engineered products, present pharmacovigilance challenges that current strategies do not adequately address. The potential for adverse effects from ATMPs to manifest years or even decades after administration, combined with limited comparative populations for signal evaluation and the need for expert knowledge to interpret immunological mechanisms, complicates the situation. This article highlights suggested adjustments, including extended follow-up registries, individualized longitudinal databases, and adaptable risk management that evolves as more post-authorization safety information becomes available. Additionally, accurately attributing adverse reactions to either the originator or biosimilar drug is complicated by interchangeable policies, automatic substitutions, and insufficient product identification in spontaneous reporting schemes.
The pharmacovigilance response during the COVID-19 vaccination rollout demonstrated that existing structures could be rapidly enhanced, but it also revealed limitations in cross-border data exchange, harmonization of signal evaluation, and timely public communication. The Brighton Collaboration case definitions and innovative approaches like V-safe in the United States significantly contributed to this international pharmacovigilance effort; however, the overall response underscored the consequences of decades of underfunding in this area.
3.7 Strengths and limitations of integrated pharmacovigilance approaches
A comprehensive evaluation of the key strengths, opportunities, limitations, and challenges identified in the reviewed literature on integrated pharmacovigilance frameworks is presented in Table 3.
Table 3. Comparative assessment of integrated pharmacovigilance approaches, strengths and limitations.
| Strengths / opportunities | Limitations / challenges |
| Disproportionality analysis provides scalable, cost-effective safety signal generation from large spontaneous reporting databases. | High false-positive rates and sensitivity to underreporting reduce clinical actionability of generated signals. |
| AI and NLP tools dramatically expand the data horizon of pharmacovigilance beyond ICSRs to social media and EHR data. | Lack of prospective regulatory validation and explainability frameworks limits AI adoption in formal regulatory decisions. |
| Integration of RWE with spontaneous reports enables signal confirmation with higher clinical specificity. | RWE is subject to confounding, selection bias, and variable data quality across jurisdictions. |
| ICH harmonization frameworks (E2E, Q10) provide a shared conceptual basis for PV-quality integration. | Implementation heterogeneity across agencies and countries undermines the uniformity that harmonization is meant to achieve. |
| Structured benefit-risk frameworks improve transparency and reproducibility in regulatory decision-making. | Frameworks require extensive clinical data and may not be feasible for novel products with limited post-market experience. |
| Patient-reported outcomes and digital health platforms provide granular, near-real-time adverse event data. | Data quality, representativeness, and regulatory acceptance of patient-generated data remain open challenges. |
| Extended follow-up registries for ATMPs and biologics enable longitudinal safety characterization. | Registry fatigue, patient attrition, and resource-intensive maintenance reduce long-term data completeness. |
| Federated learning enables privacy-preserving, cross-institutional pharmacovigilance AI development. | Technical complexity and data standardization requirements create barriers to federated learning adoption. |
4. Conclusions
This systematic review highlights the necessity and increasing feasibility of integrating pharmacovigilance into quality management and regulatory decision-making. Advances in artificial intelligence, real-world evidence, and structured benefit-risk frameworks have significantly improved signal detection and safety evaluation capabilities. However, persistent challenges such as underreporting, regulatory fragmentation, and limited validation of AI-based tools continue to hinder full integration. The findings emphasize the need for standardized validation frameworks, stronger alignment between pharmacovigilance and quality systems, and enhanced global harmonization. Future research should focus on developing scalable, regulatory-compliant AI models and integrated decision-making frameworks to strengthen pharmacovigilance across evolving therapeutic landscapes.
