Signal Management Isn’t Broken; It Simply Belongs to a Previous Era

The Gap Between Ambition and Architecture 

Although regulatory ambition has advanced, the operational infrastructure has failed to keep up. The European Medicines Agency’s stated aim is that pharmacovigilance for key new medicines should support real-time regulatory decision-making by 2030; GVP Module IX on signal management identifies early detection and prompt evaluation as central objectives; and the FDA’s Sentinel Initiative is working toward an active surveillance infrastructure with comparable intent.  

The dominant conversation in drug safety IT, however, remains focused on incremental process gains in the ICSR space, such as faster MedDRA coding, streamlined intake, and reduced clerical errors. Although these are worthy improvements, they don’t fundamentally address the case processing bottleneck. 

Agentic AI offers to enable a step change here—not in terms of faster sequential processing, but rather enabling near-instantaneous end-to-end case handling. This would see AI agents manage data collection, coding, quality review, and preparation for signal detection, freeing up human experts to focus on the interpretive and decision-making stages where their judgment is crucial. As a result of this shift, the current generous SUSAR timelines could become redundant, along with monthly line listing review for signal detection. For products launched with limited patient exposure data, early signal detection is critical, therefore same day turnaround would be extremely valuable as an operating standard rather than a theoretical ambition. 

The Signal-to-Noise Problem Is Getting Worse

The imperative to address the case processing bottleneck is becoming increasingly acute. On top of rising signal volumes, the data environment itself has become far more complex, and the signal-to-noise problem has worsened proportionally. Internal datasets are now supplemented by real-world evidence, electronic health records, published literature, and regulatory repositories such as FAERS and EudraVigilance—none of which sequential workflows were designed to cross-analyze rapidly or coherently.  

The ICSR duplication problem alone is significant and growing. TransCelerate BioPharma research across seven major pharmaceutical companies 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. Meanwhile, external factors—media attention on a particular drug class, public concern about a newly authorized product—can generate reporting surges with no underlying change in product risk, as illustrated by the documented spike in GLP-1 adverse event reports as public interest in the product class soared. In situations like this, genuine signals risk being obscured by a level of noise that no human-curated workflow can efficiently filter at scale. 

AI-powered signal prioritization offers a genuine answer to this problem, not by replacing expert analysis but by making it more effective. Trained AI models are highly effective at distinguishing statistical patterns from reporting artifacts, surface cases most likely to represent true safety signals, and direct expert attention accordingly—so that safety scientists spend their time evaluating pre-screened, contextually-enriched signals rather than endlessly searching for them. 

Human Input 

Every conversation about increased automation in signal management inevitably comes round to considerations of how much humans should be involved. The real question here is about best use of experts’ time.  

With the introduction of agentic AI, we can move beyond automation of discrete or rule-governed tasks, and into agentic workflows that are able to act more holistically and autonomously. This means that human expertise can be focused on interpretation, decision-making, regulatory dialogue, and communication with healthcare professionals.  

Five Years Is a Credible Horizon 

Automated ICSR processing pipelines, agentic coding, continuous quality review, cross-domain data integration, and AI-powered signal evaluation are capabilities now within reach of drug safety functions. This emerging opportunity now needs to be matched by organizational readiness: the confidence to commit to using the technology at scale, underpinned by a governance infrastructure that is sufficiently robust to withstand regulatory scrutiny of the outputs. Certainly, regulators’ ambitions around real-time pharmacovigilance will not be delivered by process optimization alone. There needs to be decisive infrastructure investment and organizational commitment, to unlock real change.  

The cost of doing nothing will only grow, meanwhile. Before long, companies will be called on to explain at inspection why monthly listing cycles remain standard practice—when the regulatory and patient expectation is for continuous monitoring. Organizations that commit to real transformation now could be operating with near-instantaneous case processing as a baseline within five years, and although this isn’t a pre-requisite for transformed signal management, signal detection will by then be running on real-time data as a matter of course.  

For those prepared to make the leap, the rewards extend well beyond operational efficiency: faster detection, stronger patient protection, and a safety function whose human expertise is directed where it can make the greatest difference. 
 

Lucinda Smith is ArisGlobal’s chief safety product officer. She previously more than two decades working in frontline scientific and strategic pharmacovigilance and drug safety roles at a major pharma brand. She can be reached at [email protected].