Copy article

"Pivotal moment" for NHS as drug interaction side effects to be predicted by AI before they reach patients

ended 22. October 2025

A NEW project will work out the side effects from drug interactions before patients get them in a "pivotal moment" for the NHS.

The study will use AI and NHS data to predict side effects from drug combinations before they reach patients, the Medicines and Healthcare products Regulatory Agency (MHRA) announced today. 

This is among three MHRA projects backed by government funding, to modernise how medicines and medical technologies are tested and approved – ensuring faster access for patients, while maintaining the highest safety standards. 

Millions of people in the UK take several medicines every day. In England, around one in seven people (8.4 million) are regularly prescribed five or more medicines. 

While most combinations are safe, some can interact in ways that cause harmful side effects. 

These can mean repeated GP visits, changes to prescriptions, or even hospital stays before treatments are adjusted – adding strain for patients, carers and the NHS.

AI could potentially solve this issue and help doctors better understand how combinations of medicines affect people in real life, improving how treatments are prescribed together so patients get the safest and most effective care, tailored to them, more quickly. 

This personalised approach could help prevent some of the side effects linked to medicines, which are estimated to cause around one in six hospital admissions in England and cost the NHS more than £2 billion every year.

Mitali Deypurkaystha, Human-First AI Strategist & Author at Newcastle upon Tyne-based Impact Icon AI, hailed the announcement.

She added: "This is a pivotal moment for the UK’s life sciences ecosystem. The MHRA’s use of AI to predict drug-interaction risks shows that faster doesn’t have to mean riskier. It’s responsible AI in action. By combining real-world NHS data with human oversight, regulators are proving that AI can enhance, not replace, clinical judgment. 

"But we must remember: AI is only as intelligent as the data it’s fed. It’s easy to be dazzled by the technology. Yet data is the real engine of impact. Without clean, representative, anonymised data, even the smartest AI can misfire or discriminate. In a UK-wide study linking NHS datasets, only 64% of patients had ethnicity recorded, proof that our data foundations still need work. 

“The NHS, like every organisation, must invest in getting its data right if it wants AI that delivers return on investment and truly works for everyone. The prize is huge, preventing side effects that cost over £2 billion a year, but only if the data reflects the nation it serves.”

Simon Jones, Chief Operating Officer at VP MED Ventures, called the project “compelling”.

He continued: “The study is a compelling step forward in using AI to de-risk pharmacovigilance. The significant upside lies in scaling this from post-market analysis to pre-market clinical trial design and real-world evidence generation.

"Furthermore, evolving from a static, anonymised dataset to a dynamic feedback loop where patients directly report outcomes would unlock a powerful new data layer for truly personalised and pre-emptive care. Allowing patients to comment or upload their own side effects and get real time feedback on alternatives to the medicines they are already prescribed could be a game-changer for the NHS.

Colette Mason, Author & AI Solution Architect at London-based Clever Clogs AI, said: "It's welcoming to see AI being used as a tool to speed up analysis rather than simply replacing workers. But here's the question nobody's asking: can clinicians actually see why the AI flagged an interaction, or are we just creating plausible predictions that doctors won't challenge? 

“Real healthcare AI must make clinicians brilliant at their jobs, not turn them into button-clickers who can't override the system when it gets it wrong. And let's be clear: this only works with a robust, explainable and specialised large language model (LLM) that medics can actually interrogate and override, not another black box that makes decisions nobody can justify when things go wrong. We need audit trails that survive litigation, not just algorithms that sound confident.”

3 responses from the Newspage community

Copy all

Star Quote
Copy

Editor
Mitali Deypurkaystha
AI Governance Strategist & CMO at AISensa
This is a pivotal moment for the UK’s life sciences ecosystem. The MHRA’s use of AI to predict drug-interaction risks shows that faster doesn’t have to mean riskier. It’s responsible AI in action. By combining real-world NHS data with human oversight, regulators are proving that AI can enhance, not replace, clinical judgment. But we must remember: AI is only as intelligent as the data it’s fed. It’s easy to be dazzled by the technology. Yet data is the real engine of impact. Without clean, representative, anonymised data, even the smartest AI can misfire or discriminate. In a UK-wide study linking NHS datasets, only 64% of patients had ethnicity recorded, proof that our data foundations still need work. The NHS, like every organisation, must invest in getting its data right if it wants AI that delivers return on investment and truly works for everyone. The prize is huge, preventing side effects that cost over £2 billion a year, but only if the data reflects the nation it serves.
Star Quote
Copy

Simon Jones
Chief Operating Officer at VP MED Ventures
The study is a compelling step forward in using AI to de-risk pharmacovigilance. The significant upside lies in scaling this from post-market analysis to pre-market clinical trial design and real-world evidence generation. Furthermore, evolving from a static, anonymised dataset to a dynamic feedback loop where patients directly report outcomes would unlock a powerful new data layer for truly personalised and pre-emptive care.
Allowing patients to comment or upload their own side effects and get real time feedback on alternatives to the medicines they are already prescribed could be a game-changer for the NHS.
Copy

Colette Mason
AI Ethics Consultant at Clever Clogs AI
It's welcoming to see AI being used as a tool to speed up analysis rather than simply replacing workers. But here's the question nobody's asking: can clinicians actually see why the AI flagged an interaction, or are we just creating plausible predictions that doctors won't challenge? Real healthcare AI must make clinicians brilliant at their jobs, not turn them into button-clickers who can't override the system when it gets it wrong. And let's be clear: this only works with a robust, explainable and specialised LLM that medics can actually interrogate and override, not another black box that makes decisions nobody can justify when things go wrong. We need audit trails that survive litigation, not just algorithms that sound confident.