AI project to predict health outcomes paused over GP data concern
- 4 June 2025

-
NHS England has paused development of a generative AI model capable of predicting future health outcomes after doctors raised concerns about the use of GP patient data
-
The model, called Foresight, is trained on de-identified NHS data from 57 million people in England
-
The British Medical Association and the Royal College of General Practitioners joint GP IT committee said that it had not been made aware that GP data collected for COVID-19 research had been used to train the AI model
NHS England has paused development of a generative AI model capable of predicting future health outcomes after doctors raised concerns about the use of GP patient data.
The model, called Foresight, trained on de-identified NHS data from 57 million people in England, was being tested by researchers at University College London and Kingās College London as part of a national pilot exploring how AI could support earlier intervention and more personalised care.
However the British Medical Association (BMA) and the Royal College of General Practitioners joint GP IT committee said that it had not been made aware that GP data collected for Covid-19 research had been used to train the AI model.
Dr David Wrigley, BMA England GP committee deputy chair, said: “For GPs, our focus is always on maintaining our patientsā trust in how their confidential data is handled.
“We were not aware that GP data, collected for Covid-19 research, was being used to train an AI model, Foresight.
“As such, we are unclear as to whether the correct processes were followed to ensure that data was shared in line with patientsā expectations and established governance processes.
“We have raised our concerns with NHS England through the joint GP IT committee and appreciate their verbal commitment to improve on these processes going forward.
“The committee has asked NHS England to refer itself to the Information Commissioner so the full circumstances can be understood, and to pause ongoing processing of data in this model, as a precaution, while the facts can be established.”
Wrigley added that patients “shouldnāt have to worry that what they tell their GP will get fed to AI models without the full range of safeguards in place to dictate how that data is shared”.
An NHSE spokesperson told Digital Health News: “Maintaining patient privacy is central to this project and we are grateful to the joint GP IT committee for raising its concerns and meeting with us to discuss the strict governance and controls in place to ensure patientsā data remains secure.
“The committee has asked us to confirm that GDPR [General Data Protection Regulation] principles are being upheld at all times and we have agreed to pause the project while our data protection officer now conducts a review and recommends whether any further action is needed.”
They added that the project has been conducted under a data sharing agreement between NHSE and a consortium of researchers, brought together by the British Heart Foundation Data Science Centre and this agreement was reviewed by the GP Profession Advisory Group.
Also they said that the research has been conducted by approved researchers within NHSE’s Secure Data Environment (SDE) which cannot be accessed from outside that environment.
Results of the research can only leave the SDE after being checked by a member of NHS staff to ensure that they could not be used to identify a patient, NHSE said.
2 Comments
As a patient I don’t recall being asked if I consented to my personal health data being used for this research. How can I check if my personal health data was used? Would my GP know?
You can opt out of sharing your information held by your GP, by completing a form at your surgery. You can opt out nationally with NHS England. Great. However these only apply to identifiable data⦠If the records have been anonymised/de-identified⦠then NHS England considers the data can be shared⦠even if individuals have āopted outā. People requesting such data from NHS England are required to detail their intended use of the datasets. Thereās considerable scrutiny of such applications and you are required to detail why you need each and every field. You are not allowed to request data for one purpose, and then decide to do something else. You certainly may not attempt to link de-identified data with any identifiable data that you happen to have. Anyone working in this field knows these controls and why they exist. Challenge with AI and the way this technology works is that if you go and decide to share healthcare data with a large language model, who and what is going to prevent it from attempting to link data that it shouldnāt? Once it has the information, if someone later decides it shouldnāt have it, how will it be removed?