Reimagining Signal Management: How Autonomous Pharmacovigilance Could Render Reporting Timelines a Thing of the Past

Much of the current conversation about artificial intelligence (AI) in drug safety revolves around incremental gains: streamlining intake forms, accelerating Medical Dictionary for Regulatory Activities (MedDRA) coding, or reducing clerical error rates in adverse-event records. These are meaningful improvements, but they fall short of addressing the more fundamental question: what becomes possible when safety case processing itself is no longer a bottleneck?

The answer is a qualitative shift in signal management—from a periodic, retrospective activity to a continuous and near real-time discipline. If structured, quality-reviewed case data were available to signal detection teams within hours of receipt rather than days, thanks to the use of an AI signals “agent” to perform much of the data collection and preparation and analysis, the downstream implications would be profound. Routine monthly listing cycles would become redundant. The seven-day Suspected Unexpected Serious Adverse Reaction (SUSAR) timeline, currently regarded as a regulatory floor rather than a performance aspiration, would be not only comfortably achievable but easy to surpass. Meanwhile safety scientists would be freed to focus their time on interpretation and decision-making.

Regulatory Direction of Travel

This ambition for collapsed timescales reflects a direction of travel already charted by the major regulatory bodies. The European Medicines Agency (EMA) articulated some years ago that, by 2030, pharmacovigilance (PV) for key new medicines should support real-time decision-making by regulators—transitioning the discipline from reactive case collection toward proactive, continuous safety monitoring.2 GVP Module IX on signal management reinforces this expectation, identifying early detection and prompt evaluation as central objectives.3

In the United States, the FDA’s Sentinel Initiative has pursued comparable goals, seeking to build active surveillance infrastructure capable of complementing—and ultimately outpacing— the inherent limitations of spontaneous adverse-event reporting.4

The technologies required to meet these expectations, most notably agentic AI, are not decades away. Automated case intake, AI-assisted coding and medical review, and continuous quality monitoring are either available today or in active development. What’s also needed now is the willingness to connect these components at scale, and to invest in the governance infrastructure that would allow companies to harness these capabilities with confidence.

Once near-instantaneous case processing is possible as standard, the impact on reporting timelines would be material. Current regulations require that fatal or life-threatening unexpected reactions reach health authorities within seven calendar days—a window calibrated around manual workflows. Automated processing could conceivably make a same-day turnaroundachievable, meaning earlier regulatory visibility, earlier patient protection, and a stronger overall safety picture.

Volume, Noise, and the Limits of Sequential Analysis

Greater data availability does not, on its own, translate into sharper signal intelligence. In many respects, the growth of data sources has made the analytical task harder, not easier. Internal dataset—clinical trial records, exposure histories, safety databases—are now supplemented by real-world evidence, electronic health records, published literature, and regulatory repositories such as FDA’s Adverse Event Report System (FAERS) and the European Union’s EudraVigilance. Current sequential workflows were not designed to cross-analyze these sources rapidly or coherently.

The duplication problem alone is instructive. Research across seven major pharmaceutical companies by TransCelerate BioPharma identified a mean of three submissions per case version across 2.5 million case versions, with a meaningful fraction reaching ten or more health authority recipients.5 Meanwhile, external factors—heightened media attention on a particular drug class, public concern about a newly authorized vaccine—can generate surges in reporting volumes with no corresponding change in underlying product risk. The documented spike in GLP-1 receptor agonist adverse event reports, coinciding with the surge in public and media interest in that class, is an illustration of this effect. Genuine signals risk being buried beneath an expanding layer of noise.

AI offers a route through this problem—not by replacing expert analysis, but by making this more effective. Trained models can distinguish genuine statistical patterns from reporting artefacts, surface the subset of cases most likely to represent true safety signals, and focus expert attention on the highest-priority items. The result is a safety scientist whose time is spent evaluating pre-prioritized, contextually enriched signals rather than working through undifferentiated volumes.

Reconsidering the Role of Human Review

Concerns about reducing human involvement in case processing are understandable. They typically rest, however, on an implicit assumption that manual performance is both reliable and consistent—an assumption that does not withstand close scrutiny.

That routine inspection findings relating to individual case study report (ICSR) quality exist at all is because human-dependent workflows are difficult to standardize at scale. Quality control sampling at 5% or 10% of case volume provides no guarantee of systematic quality. And the well-documented effects of fatigue and interpretive inconsistency become more pronounced as processing volumes increase. An automated system offering 100% quality coverage—with a governance layer that detects model drift and enables systematic retraining—represents a substantive improvement on that baseline. When quality problems do arise, retraining a model is considerably more tractable than retraining hundreds of case processors. AI-powered processing is also inherently more auditable: each decision is logged, every deviation traceable.

It is worth recalling that the industry has navigated analogous transitions before. When PV functions began outsourcing case processing to contract research organizations in the late 1990s and early 2000s, the concerns raised closely mirror those being voiced today about AI. How would sponsors maintain oversight? How would quality be assured at a distance? Those concerns were resolved through governance frameworks, contractual controls, and accumulated experience.6 The transition is now so established that it is barely remarked upon. The structural challenge of transitioning from outsourced human processing to AI is not categorically different. Even where organizations may have been hesitant about AI use out of a fear of inspections, inspections can be navigated very successfully with the right governance and oversight.

The goal, in any case, is not autonomous pharmacovigilance in some absolute sense—a system operating entirely without human input. It is augmented pharmacovigilance: AI handling the high-volume, rule-governed tasks for which it is well suited, while human expertise is concentrated on the interpretive, contextual, and communicative dimensions that genuinely require it. Signal validation, regulatory dialogue, benefit-risk assessment, and engagement with healthcare professionals all remain firmly in the domain of the safety scientist. The question is whether those scientists should be spending most of their time on data preparation.

A Credible Near-Term Horizon

The building blocks for the transition described here are already in place or in active development. Automated ICSR processing pipelines, agentic AI coding assistants, continuous quality review layers, cross-domain data integration platforms, and AI-assisted signal prioritization tools represent, collectively, a substantive and deployable capability set. What is required to complete the picture is organizational readiness—the confidence to deploy these tools at scale, and governance frameworks robust enough to support regulatory scrutiny of the outputs.

Organizations that make this commitment could, within 5 years, be operating with near-instantaneous case processing as a baseline capability. Signal detection, in such an environment, would run on current data as a matter of course rather than against periodic snapshots. Safety scientists would allocate the majority of their working time to decision-making rather than data assembly.

With near-real-time reporting, the 7-day SUSAR timeline would shift from being the standard to being obsolete. This is of particular significance for products approved after exposure in small patient populations, where the safety database at launch may be limited and early post-market signal detection is correspondingly more critical. In such cases, the speed of detection directly determines the speed of mitigation. Where discontinuation of development is warranted, earlier signals would mean earlier intervention to protect patients.

EMA’s stated ambition for real-time PV by 2030 will not be delivered by incremental process optimization. It will require the kind of infrastructure investment and organizational commitment described here. Organizations that move early stand to gain not only in operational efficiency, but in the quality and timeliness of their safety data—a tangible asset in building and maintaining regulatory trust. Explaining at inspection why monthly listing cycles are still in use, in a landscape that has moved to real-time monitoring, will become an increasingly uncomfortable position to defend.

Embracing Change

The infrastructure to transform PV signal management is within reach today, and the regulatory direction is clear. Companies now just need to be willing to embrace the change now possible. For those prepared to move beyond piecemeal automation and invest in ICSR transformation, with a positive impact on overall signal management, the rewards are substantial. These range from faster detection and stronger patient protection, to a safety function whose human expertise is directed where it can have the greatest impact.

References

1. ICH. Guideline E2A: Clinical Safety Data Management — Definitions and Standards for Expedited Reporting. 1994. Accessed June 15, 2026. https://www.ich.org/page/safety-guidelines
2. EMA. Signal management. Updated July 2025, reflecting Implementing Regulation (EU) 2025/1466. Accessed June 15, 2026. https://www.ema.europa.eu/en/human-regulatory-overview/post-authorisation/pharmacovigilance-post-authorisation/signal-management
3. European Commission. Implementing Regulation (EU) 2025/1466 of 22 July 2025, amending Implementing Regulation (EU) No 520/2012 as regards pharmacovigilance requirements. Official Journal of the European Union, July 23, 2025. Accessed June 15, 2026. https://eur-lex.europa.eu/eli/reg_impl/2025/1466/oj/eng
4. Gavin KM, Sundermann ML, Wieland A. Leveraging real-world data for safety signal detection and risk management in pre- and post-market settings. Frontiers in Drug Safety and Regulation. October 2025. https://www.frontiersin.org/journals/drug-safety-and-regulation/articles/10.3389/fdsfr.2025.1626822/full
5. van Stekelenborg J et al. Individual case safety report replication: An analysis of case reporting transmission networks. Drug Safety. 2023;46(1):39–52. Epub 2022 Dec 24. Doi: 10.1007/s40264-022-01251-7. https://pmc.ncbi.nlm.nih.gov/articles/PMC9870831/
6. 6. Smith L, et al. Might we come together on a paradigm shift to manage ICSRs with a decentralized data model? Drug Safety. 2025;48(8):843–853. Epub 2025 March 8. doi: 10.1007/s40264-025-01539-4

About the Author

Lucinda Smith is chief safety product officer at ArisGlobal. Prior to joining the company, she spent more than two decades in frontline scientific and strategic PV and drug safety roles at a major pharmaceutical company.