What can UK AI Healthcare Regulation Learn from The Pharmacovigilance Approach?
Pharmacovigilance has long operated on the principle that regulatory approval is not the end of safety assessment, but the beginning of continuous monitoring. Now, the UK Government is moving towards applying a similar philosophy to AI in healthcare.
The UK Government has announced it will accept all 44 recommendations made by the National Commission into the Regulation of AI in Healthcare, signalling a shift in how AI may be regulated across the NHS. Their response outlines ambitions to replace traditional one-off assessments with a flexible framework that continuously monitors AI systems throughout their lifecycle.
The independent NHS doctor-led commission developed its recommendations through consultation with over 12,000 patients, clinicians and members of the public. It argues that existing regulatory approaches must evolve to address fluctuating AI performance after deployment, making static assessments inadequate.
The proposed lifecycle-based framework would emphasise evaluating AI systems in real-world clinical environments, rather than relying predominantly on pre-deployment assessments. The MHRA is already exploring this approach through its AI Airlock regulatory sandbox, focusing on post-market surveillance of AI-enabled medical devices.
These recommendations recognise that AI systems are dynamic. Performance can change as models are updated, clinical contexts evolve and new data becomes available. Consequently, demonstrating safety before deployment does not guarantee continued reliability. For PV professionals, this should sound familiar.
Medicines undergo rigorous pre-market evaluation, but monitoring does not stop there. Safety profiles continue to be assessed through real-world evidence, adverse event reporting and ongoing benefit-risk analysis. Regulators increasingly recognise that AI may require comparable post-deployment surveillance.
“I particularly welcome the commitment to monitoring AI throughout its use in practice, strengthening patient involvement in decision-making and ensuring there are clear routes to redress when things go wrong”
Henrietta Hughes, Patient Safety Commissioner for England
The third phase of the MHRA’s AI Airlock will investigate lifecycle regulation of AI-enabled medical devices to inform future guidance. Further commitments include guidance on managing changes to AI devices and exploring staged authorisation pathways permitting supervised clinical deployment while additional evidence is gathered.
These PV parallels invite an intriguing possibility: rather than AI simply transforming pharmacovigilance, could pharmacovigilance’s established safety principles help shape AI regulation? Concepts such as signal detection, continuous benefit-risk assessment and systematic safety reporting could offer valuable lessons for identifying emerging problems in AI systems.
For an industry accustomed to asking whether AI can improve pharmaceutical safety, it is worth asking whether the principles of pharmaceutical safety can improve AI.
