Preeya Beczek, MD- Regulatory Affairs and Compliance, Beczek.COM; Agnes Cwienczek- Director of Product Management, ArisGlobal
As confidence builds in AI-powered platforms, particularly those harnessing Generative AI to distil insights and repurpose them in new formats, domain leads are eyeing up regulatory impact assessment (a critical first step in product change control) as the obvious next target for an AI-enabled overhaul. Despite the activity’s criticality, regulatory impact assessment adds little immediate value beyond continued quality assurance, compliance, and risk mitigation.
It is also complex and highly resource-intensive, and is becoming increasingly challenging. That’s as regulators progress their efforts to balance speed to market with drug quality and patient safety.
Regulatory Appetites for AI are Increasing
In a pharma industry survey last year,1 regulatory impact assessment was cited as an attractive upcoming use case for AI automation support by senior regulatory professionals. Core AI capabilities are already established in a life sciences regulatory context, where the technology is being used to hone marketing authorization applications and maintain registrations.
This is bringing welcome relief to Regulatory departments beset with multiplying and evolving workloads, entrenched manual processes (e.g., creating documents or summaries from scratch, extracting data manually, uploading agency correspondences), and poor visibility across departmental boundaries (due to siloed systems, duplicated information recording, and disjointed ways of working).
Those charged with performing regulatory impact assessments stand to benefit similarly. Fronting any process involving a product change, the activity bears an inherent time pressure. If an urgent safety change comes in, the associated regulatory impact assessment typically needs to be performed within hours, not days. That’s irrespective of the extensive scouring and reviewing of information this will entail.
Among the immediate considerations are: “What did we present to the authority last time? “What does our label say?”, “How soon must the change be implemented/within what timeframe, and which documentation is required?” All of which requires extensive searching and referencing of diverse and often unconnected sources, including manual lookup of non-indexed (unstructured) data buried in static documents. Investigative work often extends beyond central operations, too, spanning feedback loops from affiliates about the current status and local regulations - information which may be recorded in different languages.
When a product change triggers a regulatory impact assessment, typically this will happen initially at an individual country level. That assessment then has to be repeated to some degree by the local operation, where national licenses are involved. Each country will then decide whether and when they will need to make a change (e.g., reflect it in product labelling), and update their registration/notify the relevant health authority. Is it a case of “do first, then tell”, for instance? And what of the manufacturing sites where the product is held? When will the change be rolled out? Will a grace period be required? How urgent is it: can it wait for the next print run?
Much of this activity will need to take place in parallel, too, to support planning - demand planning, supply planning, materials availability, and so on. And the associated safety/regulatory changes will need to follow this chain of events very promptly. All of this adds to the complexity of product change control and the work it generates. In the 2024 survey, 55% of senior regulatory professionals actively expressed interest in an advanced, AI-enabled technology support for the task, to relieve the intensity of the workload and speed up delivery. As many as 97% agreed that AI-enabled automation would be useful in identifying the direct impact of product changes.
Breaking Down the Problem
A practical approach to applying AI in regulatory impact assessment is to break the end-to-end process down and consider the individual stages where intelligent automation could really make a difference. With so many variables in the make-up, structure, and focus of individual organizations, it is unlikely that one size will fit all. Data- and technology-readiness will have a bearing on what’s possible now and what is likely to deliver the best results. Starting small is advisable, focusing on 1-2 particular product lines, or a specific region or country. The key is to identify a painful problem that needs to be overcome, where AI could present a solution.
As AI assumes the detailed exploratory work, process stakeholders (central regulatory professionals, local regulatory representatives, plus those operating at a manufacturing level, demand, and supply chain level, and in Quality and Safety) can start to align more closely and collaborate more effectively on next actions.
Even just speeding up the review process in the initial assessment (locating and searching all of the information, and determining where efforts need to be concentrated) will empower teams to move more swiftly in determining and executing next moves. The ability to automatically scan the latest regulatory intelligence in different markets, and consult previous Agency exchanges, can then help further expedite next steps – or at least pinpoint where supplementary insights may be needed where the latest local requirements are less clear.
The more embedded AI becomes in the end-to-end process, the more the gains will multiply. Where an AI tool is pulling information from several sources into one place, teams can be ready to review and validate the findings. Generative AI tools can help with structured content authoring, meanwhile, or swiftly bring a document from version zero to a solid first draft, knowing what data to pull in - and where to find it.
Letting Go, While Keeping Control
Where regulatory teams look to AI to take over the administrative heavy lifting, strict governance is essential. Irrespective of how smart and wwell-trainedthe the AI capabilities are, professionals should not be deferring to the technology to make the decisions for them. Cross-functional teams will still need to agree whether and where a change is applicable, whether it needs to be made now, or whether it can be deferred, and when it should be reported to the relevant regulatory body, for instance.
Finally, there should be a well-articulated plan to share the vision and upskill relevant teams. This will be important to ensure that professionals appreciate not only the potential of applying AI-powered automation to regulatory processes, critical documents, and data, but also the need to take great care in the way the technology is applied.
Reference
- Survey: Unsustainable Regulatory Workloads Leave No Choice About AI Adoption, ArisGlobal/Censuswide, November 2024: https://www.arisglobal.com/media/press release/survey-regulatory-workloads-ai-adoption/ (Full survey report at: https://www. arisglobal.com/resources/regulatory-industry-survey/ )
Author Biographies
Preeya Beczek, MD of consulting firm Beczek.COM, is a regulatory affairs and compliance expert with over 26 years’ industry experience.
Agnes Cwienczek is Director of Product Management at ArisGlobal, with a remit including the provision of business process and data management expertise in the areas of Regulatory Information Management, Document Management, Submission Management, and Labelling Management. Agnes previously worked at Merck in Global Regulatory and Quality Assurance, during her two decades at the frontline of regulatory information management.
Subscribe to our e-Newsletters.
Stay up to date with the latest news, articles, and events. Plus, get special
offers from American Pharmaceutical Review delivered to your inbox!
Sign up now!