Makale
AI in the spotlight: Takeaways from the World Patient Safety Congress
-
Lin Li, Ph.D.
-
José Miguel Rivas Romero
-
Stephen Sun, MD, MPH
The industry has reached a point where understanding of artificial intelligence (AI) and its potential applications in pharmacovigilance has matured considerably. That was clear at both the World Patient Safety Congress held in Amsterdam from 22 to 24 September and the congress held in Boston from 13 to 15 October. For PV leaders globally, the implication of these developments is how PV workflows themselves should evolve to make the best use of AI and the importance of not only implementing but also understanding the technology.
Managing the risks of AI
Speakers at both meetings emphasized the importance of knowing AI’s limitations and understanding the risks and how to control them. The objective of all the stakeholders was to demonstrate the value from implementation of AI.
At one session in Boston, discussion emphasized focused governance, continuous AI use case assessment, and risk management. This is important because highly regulated environments require validation of outputs, human oversight, documented controls, review of where AI is appropriate and where it is not, and assessment of the model or workflow risks. A key takeaway from the session was that AI value is inseparable from governance in PV and safety settings.
At one session in Boston, discussion emphasized focused governance, continuous AI use case assessment, and risk management. This is important because highly regulated environments require validation of outputs, human oversight, documented controls, review of where AI is appropriate and where it is not, and assessment of the model or workflow risks. A key takeaway from the session was that AI value is inseparable from governance in PV and safety settings.
Oversight, change management and monitoring
There was also a strong message at both meetings that innovation cannot be separated from change management. Any new system implementation needs to be supported by a robust change management plan, which applies equally to broader organizational shifts such as centralized models. Successful adoption depends on clear roles, stakeholder readiness, training, workflow design, and disciplined execution.
Health authorities are also testing AI, and regulators are encouraging companies to continue to explore its use. However, regulators expressed a clear expectation for structured evaluation protocols, strong validation approaches, and continuous monitoring over time — even when early data looks encouraging.
Health authorities are also testing AI, and regulators are encouraging companies to continue to explore its use. However, regulators expressed a clear expectation for structured evaluation protocols, strong validation approaches, and continuous monitoring over time — even when early data looks encouraging.
The recommendation from regulators is to start small and scale up. The preferred approach is to begin with targeted, manageable use cases, demonstrate value, and refine governance and workflows before expanding more broadly.
Another important discussion point at the Amsterdam meeting was the role of agentic AI in adverse event processing. While there is clear interest in the potential of these models, the emphasis was firmly on oversight and escalation. The concept discussed was one of agent supervisors embedded within the organization to oversee agentic squads, with mandatory handoffs to humans whenever ambiguity arises — especially in the context of serious adverse events.
Another important discussion point at the Amsterdam meeting was the role of agentic AI in adverse event processing. While there is clear interest in the potential of these models, the emphasis was firmly on oversight and escalation. The concept discussed was one of agent supervisors embedded within the organization to oversee agentic squads, with mandatory handoffs to humans whenever ambiguity arises — especially in the context of serious adverse events.
How to make AI a reality in PV
Key themes in the Boston meeting explored how best to apply AI for successful outcomes. Speakers highlighted the importance of standardization, noting that local variation creates inefficiency, quality risk, and implementation friction. If the process is fragmented, AI tends to amplify inconsistency rather than solve it.
Successful adoption should also be tied to addressing operational pain points, such as intake automation, use of advanced technologies such as natural language processing (NLP) and optical character recognition (OCR) for document handling, case triage, AI-powered translation, medical review support, and summarization and follow-up.
Successful adoption should also be tied to addressing operational pain points, such as intake automation, use of advanced technologies such as natural language processing (NLP) and optical character recognition (OCR) for document handling, case triage, AI-powered translation, medical review support, and summarization and follow-up.
.jpg?h=768&iar=0&w=1365&hash=E0A6FCA47BA854226B601A260EE129BE)
Jobs now and in the future
One key area of focus is how AI will impact jobs, including its potential to create new roles focused on managing and coordinating AI agents, while enabling employees to shift to more strategic roles.
The role of pharmacovigilance leaders and the impact technology has on jobs was also discussed at the Amsterdam meeting. A key consideration is the increasing complexity of maintaining oversight of a company’s global PV activities.
Heads of PV, the EU qualified person for pharmacovigilance (QPPV) and other PV functions must demonstrate to inspectors and auditors that processes are being managed as intended and that the company is inspection ready.
The role of pharmacovigilance leaders and the impact technology has on jobs was also discussed at the Amsterdam meeting. A key consideration is the increasing complexity of maintaining oversight of a company’s global PV activities.
Heads of PV, the EU qualified person for pharmacovigilance (QPPV) and other PV functions must demonstrate to inspectors and auditors that processes are being managed as intended and that the company is inspection ready.
However, many companies lack dashboards that would enable more complete oversight. Discussions looked at the potential to leverage technology that can demonstrate, at a glance, that a company is operating in a validated environment and that can check for any gaps in the PV system would help to address this complexity.
It was also made clear that even in more centralized or technology-enabled models, local teams remain essential for building relationships with regulators, maintaining strong networks in-country, and collecting relevant insights from local regulations. This will require clearer ownership of processes and a thoughtful definition of where local expertise adds the greatest value.
The importance of human expertise was also emphasized at the Boston meeting, with a key takeaway being that, while automation accelerates work, experts remain accountable for clinical and regulatory judgment.
It was also made clear that even in more centralized or technology-enabled models, local teams remain essential for building relationships with regulators, maintaining strong networks in-country, and collecting relevant insights from local regulations. This will require clearer ownership of processes and a thoughtful definition of where local expertise adds the greatest value.
The importance of human expertise was also emphasized at the Boston meeting, with a key takeaway being that, while automation accelerates work, experts remain accountable for clinical and regulatory judgment.
Conclusion: Looking ahead to AI potential
Key takeaways from the meetings are that as companies gain experience with AI, they can assess where it can be leveraged, how information should flow, where statistical methods and human expertise are important, and how these components should work together.
Another important direction is the use of AI-enabled proactive and predictive methods to move toward anticipating emerging issues and potential future outcomes earlier, giving teams more opportunity to evaluate the evidence, understand possible implications, and prepare for decisions.
In our experience, achieving this will require more than advances in AI models alone. It may involve bringing together new or broader sources of data, statistical and predictive methodologies, relevant scientific and clinical expertise, and new ways of collaborating.
Equally important is determining how different types of evidence can appropriately support predictions and how uncertainty around those predictions should inform decisions. This connects closely with the predictive AI work we have been developing, where the objective is to use data and analytical methods to help anticipate outcomes and provide meaningful decision support. It is also possible that what is talked about today as AI may simply be automation by next year’s conference.
Another important direction is the use of AI-enabled proactive and predictive methods to move toward anticipating emerging issues and potential future outcomes earlier, giving teams more opportunity to evaluate the evidence, understand possible implications, and prepare for decisions.
In our experience, achieving this will require more than advances in AI models alone. It may involve bringing together new or broader sources of data, statistical and predictive methodologies, relevant scientific and clinical expertise, and new ways of collaborating.
Equally important is determining how different types of evidence can appropriately support predictions and how uncertainty around those predictions should inform decisions. This connects closely with the predictive AI work we have been developing, where the objective is to use data and analytical methods to help anticipate outcomes and provide meaningful decision support. It is also possible that what is talked about today as AI may simply be automation by next year’s conference.
Yasal Uyarı:
Bu makalede verilen bilgiler yasal tavsiye niteliğinde değildir. Cencora, Inc., okuyucuları tartışılan konularla ilgili mevcut bilgileri gözden geçirmeye ve bunlarla ilgili kararlar alırken kendi deneyim ve uzmanlıklarına güvenmeye şiddetle teşvik eder.
