The framework outlines 35 consensus-based requirements covering clinical performance, patient safety, regulation, data governance, cybersecurity, environmental sustainability, value for money and integration with existing NHS systems.
Diabetic eye screening is one of the NHS's largest national screening programmes, serving almost four million people with diabetes in England. It aims to detect diabetic retinopathy – a complication of diabetes and one of the leading causes of preventable sight loss among working-age adults – before symptoms develop, enabling timely treatment and reducing the risk of vision impairment. The service faces growing demand as diabetes prevalence increases, alongside wider workforce pressures affecting the NHS.
While many AI systems have demonstrated strong diagnostic performance in research settings, relatively few have been successfully implemented at scale. According to the researchers, this is often because technology developers are unclear about the broader requirements needed for adoption in routine care.
Clear roadmap for developers and suppliers of AI screening tools
The new target product profile aims to address that challenge by providing a clear roadmap for developers and suppliers, while supporting policymakers and healthcare organisations in evaluating potential technologies.
The framework identifies a preferred role for AI as a first-pass screening tool, automatically analysing retinal images and identifying those that require further review by specialist graders.
The recommendations were developed using a modified Delphi consensus process involving a wide range of stakeholders, including people living with diabetes, healthcare professionals, healthcare managers, regulators, policymakers and AI developers.
Professor Alastair Denniston, Professor of Regulatory Science and Innovation at the University of Birmingham, Honorary Consultant Ophthalmologist at University Hospitals Birmingham NHS Foundation Trust, and co-lead for the NIHR BRC: Birmingham’s Data, Diagnostics, and Decision Tools research theme, said: "Artificial intelligence has huge potential to improve care and help healthcare systems manage growing demand, but successful implementation requires much more than demonstrating high diagnostic accuracy.
"This work brings together the perspectives of patients, clinicians, regulators, policymakers and industry to define what good looks like for AI adoption in diabetic eye screening. By providing a clear and transparent framework, we can help ensure future technologies are developed around the needs of patients and the NHS from the outset.”
Dr Kevin Dunbar, Deputy Director of Public Health, Vaccination and Screening at NHS England, said: "AI has real potential to help the NHS protect more people’s sight as demand for diabetic eye screening grows.
"This framework will help ensure future technology is safe, effective and delivers high-quality screening for patients wherever they live."
The researchers believe the approach used in this study could also provide a model for future AI implementation across other areas of healthcare.
Lead author Trystan Macdonald, Clinical Research Fellow at the University of Birmingham, said: "Healthcare organisations often face a difficult challenge when assessing new AI technologies. Strong technical performance alone does not guarantee that a product will be suitable for routine care.
"Our framework helps bridge that gap by setting out the wider requirements that innovators need to consider if they want their technologies to be successfully adopted in practice. This framework could help accelerate the safe adoption of healthcare AI by providing greater clarity for developers, evaluators, regulators and healthcare providers.”
