26-27 SAFETY - SUSAR coated

The conversation about AI in pharmacovigilance has typically focused on automating individual steps in the case processing workflow.

A greater opportunity lies in broader, more autonomous signal management, which could transform the speed of ICSR processing from days to minutes.

For all the attention on AI’s potential to streamline operational processes in pharmacovigilance, such as case processing, the more transformative opportunity is in making signal management more autonomous.

If safety case processing could be automated to the point of near instant completion, with structured data available to signal detection teams as soon as a case is received rather than days later, the entire downstream workflow would change.

Signal detection would move from periodic to continuous. The seven day Suspected Unexpected Serious Adverse Reaction reporting timeline would feel archaic.

Meanwhile safety scientists, freed from the routine burden of data management via a signals agent that could remove much of the collation and analysis, would be able to devote their time to signal validation.

The technology components to make this possible – automated case intake, AI-assisted coding and medical review, continuous quality review – are in some cases already available or in active development.

All that is needed now is the organisational confidence to deploy them at scale.

Regulators have already set expectations around continuous benefit-risk monitoring.

While there have been advances in applying AI to safety signal work, including smarter disproportionality analysis, natural language processing of literature and machine learning across large data sets, real transformation in turnaround speed demands greater operationalisation of AI.

If cases could be processed automatically – extracted, coded, medically reviewed, quality reviewed and entered into a safety database within hours of receipt – signal detection teams would no longer work from snapshots. They would have access to current, structured data as standard.

The monthly listing cycle would become unnecessary. With cases reaching health authority databases more promptly, the regulatory picture would improve in parallel.

Greater data availability does not automatically translate into better signal intelligence.

The proliferation of sources has made the analytical task harder. Internal data sets are supplemented by external data from real-world evidence, literature, electronic health records and regulatory databases such as FAERS and EudraVigilance.

Current sequential, manual workflows were not designed for cross-analysing these data sets rapidly. More sources mean more noise, more duplication and greater potential for distortion.

The scale of ICSR replication alone is notable. A study of seven major pharma companies by TransCelerate BioPharma found a mean of three submissions per case version across 2.5 million case versions, with a significant proportion reaching ten or more health authority recipients.

Media attention also affects reporting rates, as seen in the surge in GLP-1 receptor agonist reports and the spike in COVID vaccine reports during 2021. Increasingly, the signals that matter are being buried.

This is where next-level AI comes in, alleviating pressure on human teams via assisted filtering. Trained models can distinguish genuine patterns from reporting artefacts, surface the cases most likely to represent true signals and direct expert attention accordingly.

The safety scientist’s job becomes one of evaluating signals that have already been prioritised and contextualised.

The building blocks for this next level of improvement – automated ICSR processing, agentic AI capabilities including intelligent coding agents, continuous quality review layers, cross-domain data integration and AI-assisted signal prioritisation – exist now or are in active development.
All that is needed is the willingness to connect these capabilities, supported by governance frameworks that allow organisations to stand behind the output with regulatory confidence.

Within a couple of years, organisations could be operating with near instant case processing.

Signal detection could run on current data rather than periodic listings, and safety scientists could spend most of their time on analysis and decision-making rather than data management.

The seven-day SUSAR timeline, currently a regulatory baseline rather than an ambition, would become a ceiling. For products reaching market after exposure in only a few hundred patients, earlier detection means faster mitigation and earlier decisions on discontinuation where necessary.

EMA’s 2030 vision for PV sets an expectation around real-time decision-making.

Organisations that get ahead will not only be more efficient, they will detect safety risks earlier, protect more patients and cultivate regulatory trust, while avoiding the need to explain at inspection why they are still running monthly listings in a real time world.

The building blocks for this next level of improvement – automated ICSR processing, agentic AI capabilities including intelligent coding agents, continuous quality review layers, cross-domain data integration and AI-assisted signal prioritisation – exist now or are in active development.

All that is needed is the willingness to connect these capabilities, supported by governance frameworks that allow organisations to stand behind the output with regulatory confidence.

Within a couple of years, organisations could be operating with near instant case processing.

Signal detection could run on current data rather than periodic listings, and safety scientists could spend most of their time on analysis and decision-making rather than data management.

The seven-day SUSAR timeline, currently a regulatory baseline rather than an ambition, would become a ceiling. For products reaching market after exposure in only a few hundred patients, earlier detection means faster mitigation and earlier decisions on discontinuation where necessary.

EMA’s 2030 vision for PV sets an expectation around real-time decision-making.

Organisations that get ahead will not only be more efficient, they will detect safety risks earlier, protect more patients and cultivate regulatory trust, while avoiding the need to explain at inspection why they are still running monthly listings in a real time world.