Article

How automation, analytics, and AI can assist safety signal management

Lin Li
Stephen Sun

Discover how the collaboration between safety, statistics, and AI experts are helping to solve the growing challenges of pharmacovigilance signal management

Safety signal management is a core responsibility of pharmacovigilance organizations and a critical component of protecting patient safety across the product lifecycle.^1,2

Whether for drugs or biologics, the objective remains consistent: to identify, assess, and act on emerging safety concerns as early as possible while maintaining scientific rigor and regulatory compliance.

In recent years, however, the complexity of achieving this objective has increased substantially. Growing data volumes, increasingly complex products, and heightened regulatory expectations have placed significant strain on traditional signal management approaches, making optimization an operational and scientific necessity.

Growing data volume and complexity

Current pharmacovigilance systems must process safety information from an expanding range of sources, including spontaneous patient and provider adverse event reports, clinical trials and post-authorization safety studies, scientific literature, patient support programs, field representatives, and real-world data streams. These data sources vary widely in structure, quality, and clinical context. Reports can be incomplete, unstructured, or duplicated across regions and databases, creating noise and inefficiency, making it difficult to assess real safety concerns. For biologics, these signal detection challenges are further compounded by immunogenicity, class effects, manufacturing variability, and delayed onset adverse events. As a result, safety teams now spend a significant portion of their time on manual triage, reconciliation, and data cleaning activities.³

Automation and artificial intelligence offer practical solutions at the data intake and processing level.^3,4 Natural language processing can extract clinically relevant information from unstructured narratives, while automated de-duplication reduces redundancy and improves data quality.⁶ Risk-based prioritization tools can help identify cases that warrant earlier review based on seriousness, novelty, or potential impact.⁷ These technologies do not replace human expertise; rather, they can focus constrained resources on higher-value activities, such as allowing safety professionals to focus on meaningful signal interpretation and decision-making.

Limitations of traditional signal detection methods

Traditional statistical approaches such as proportional reporting ratios, reporting odds ratios, and empirical Bayesian metrics remain widely used in signal detection. These methods serve an important role as screening tools, helping to identify potential associations between products and adverse events. However, they were not designed to function as standalone decision-making mechanisms and have well-recognized limitations. Disproportionality analyses often evaluate data in isolation, without formally incorporating prior knowledge such as known class effects, biological plausibility, or historical safety experience. They are sensitive to reporting bias, changes in exposure, and external influences such as media attention or regulatory actions. As a result, these methods may generate false positives or fail to detect emerging risks, particularly in complex or low-exposure settings.⁸

In practice, signal assessment remains heavily dependent on expert judgment, leading to variability across teams and organizations. Different reviewers may interpret the same data differently, influenced by experience, therapeutic area familiarity, or organizational risk tolerance. Traditional approaches also struggle to integrate non-clinical data, mechanistic insights, and real-world evidence in a systematic way, making it difficult to appropriately weight different data sources. This multi-factorial scenario complicates the ability to provide transparent and consistent signal decisions.

Advanced analytics and artificial intelligence can strengthen signal evaluation by providing more structured and context-aware insights.⁹ Bayesian approaches allow prior knowledge to be formally combined with emerging data, supporting more nuanced interpretation.¹⁰Longitudinal and multivariate analyses enable assessment of trends over time and across patient subgroups. AI-based pattern recognition can highlight emerging issues across products, indications, or therapeutic classes. When used as decision-support tools, these approaches enhance consistency and transparency while preserving the central role of expert medical judgment.

Sustainability, workforce constraints, and regulatory expectations

Signal management needs to operate within the realities of limited resources, regulatory scrutiny, and organizational risk tolerance. Case volumes continue to rise, yet organizations face constraints in hiring, training, and retaining experienced safety professionals. Effective signal management requires a combination of clinical expertise, statistical understanding, and regulatory knowledge—skills that take years to develop.

At the same time, regulators expect continuous safety surveillance, timely escalation of potential risks, and clear documentation of decision-making processes. The US Food and Drug Administration has sought to make it easier for stakeholders to search for and organize data on adverse events with the FDA Adverse Event Reporting System (FAERS) Public Dashboard.¹¹

Under traditional operating models, highly trained professionals spend a substantial amount of time on manual tracking, periodic reviews, and document preparation. Knowledge is often held by individuals rather than embedded in systems, increasing vulnerability to turnover and inconsistency. These pressures contribute to burnout, variability in signal assessment, and potential inspection findings.

AI-enabled operations would offer a more sustainable approach. “Human-in-the-loop" systems allow automation to manage routine tasks such as signal tracking, workflow management, and documentation generation, while preserving expert oversight for interpretation and decisions. Automated triggers can support timely review of emerging risks, and AI systems can learn from prior signal decisions and regulatory outcomes, helping to institutionalize knowledge over time. When implemented within robust governance, quality frameworks, and transparent documentation, these models support scalable, inspection-ready signal management.

Conclusion

The foreseeable near future of drug and biologic safety signal management is an augmented scenario rather than a fully automated system. Automation, advanced analytics, and artificial intelligence, when applied responsibly and transparently, enable pharmacovigilance teams to manage growing complexity without compromising scientific rigor or regulatory compliance.¹² Even with such toolbox awareness, real-world operational implementation will need to be bespoke and customized to accommodate for company plans, investments, risk tolerance, and portfolio size. By reducing administrative burdens, improving analytical consistency, and strengthening documentation, these technologies allow safety professionals to focus on their highest-value role: applying expert judgment to protect patients and support informed benefit–risk decisions throughout the product lifecycle.

*Sources suite ci-dessous

About the authors:

Lin Li, Ph.D., is Head of Clinical Statistics and Predictive AI at Cencora. He provides data-driven and tailored solutions that integrate biostatistics, bioinformatics, computer science, and biology to tackle challenges in discovery and clinical development.

Stephen Sun, M.D., MPH, is Head of Pharmacovigilance and the Practice Area Lead for Benefit-Risk Management for Cencora. He has worked in generics, branded, and OTC products as a sponsor overseeing clinical, medical affairs, and drug safety. He has also served as a medical officer in risk management and controlled substances at the US FDA.

Clause de non responsabilité :
Les informations fournies dans cet article ne constituent pas des conseils juridiques. Cencora, Inc. encourage vivement les lecteurs à consulter les informations disponibles relatives aux sujets abordés et à s’appuyer sur leur propre expérience et expertise pour prendre des décisions à ce sujet.

Entrez en contact avec notre équipe

Notre équipe d’experts en valeur de premier plan se consacre à la transformation des données probantes, des informations sur les politiques et des renseignements sur le marché en stratégies efficaces d’accès au marché mondial. Laissez-nous vous aider à naviguer en toute confiance dans les méandres complexes des soins de santé d’aujourd’hui. N’hésitez pas à nous contacter pour découvrir comment nous pouvons vous aider à atteindre vos objectifs.

Nous contacter

Sources:

^1.ICH guideline E2C (R2) on periodic benefit-risk evaluation report, EMA, 2013. https://www.ema.europa.eu/en/documents/regulatory-procedural-guideline/international-conference-harmonisation-technical-requirements-registration-pharmaceuticals-human-use-guideline-e2c-r2-periodic-benefit-risk-evaluation-report-step-5_en.pdf
^2.Best Practices for FDA Staff in the Postmarketing Safety Surveillance of Human Drug and Biological Products, FDA. https://www.fda.gov/media/130216/download
^3.Practical Aspects of Signal Detection in Pharmacovigilance, CIOMS Working Group VIII, 2010. https://cioms.ch/wp-content/uploads/2018/03/WG8-Signal-Detection.pdf
^4.Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products, FDA. https://www.fda.gov/media/184830/download
^5.Reflection paper on the use of Artificial Intelligence (AI) in the medicinal product lifecycle, EMA, 2024. Reflection paper on the use of Artificial Intelligence (AI) in the medicinal product lifecycle_240903
^6.Artificial intelligence in pharmacovigilance, CIOMS Working Group report, 2025. https://cioms.ch/wp-content/uploads/2022/05/CIOMS-WG-XIV_Draft-report-for-Public-Consultation_1May2025.pdf
^7.Supporting Pharmacovigilance Signal Validation and Prioritization with Analyses of Routinely Collected Health Data: Lessons Learned from an EHDEN Network Study, Drug Saf., Oct 2023. https://pmc.ncbi.nlm.nih.gov/articles/PMC10684396/
^8.Exploring the complexities of disproportionality analysis in pharmacovigilance: reflections on the READUS-PV guideline and a call to action, Front Pharmacol. 2025. https://www.frontiersin.org/journals/pharmacology/articles/10.3389/fphar.2025.1573353/full
^9.Artificial Intelligence: Applications in Pharmacovigilance Signal Management, Pharmaceut Med. 2025. https://pubmed.ncbi.nlm.nih.gov/40257538/
^10.A Bayesian method to detect drug‐drug interaction using external information for spontaneous reporting system. Statistics in Medicine, 43(18), pp.3353-3363. https://pubmed.ncbi.nlm.nih.gov/38840316/
^11.FDA Adverse Event Reporting System (FAERS) Public Dashboard. https://www.fda.gov/drugs/fdas-adverse-event-reporting-system-faers/fda-adverse-event-reporting-system-faers-public-dashboard
^12.Best Practices for FDA Staff in the Postmarketing Safety Surveillance of Human Drug and Biological Products. https://www.fda.gov/media/130216/download

Ressources connexes

Webinaire

Exploiter l’IA pour révolutionner la gestion des signaux de pharmacovigilance

Livre blanc

Libérez le potentiel de l’externalisation de la pharmacovigilance

Article

Gérer les défis de la pharmacovigilance sur les marchés locaux