Posted on 8 October 2024
By 2040, it is estimated that more than half a billion people worldwide will be living with Type 2 Diabetes (T2D).
The team from the Centre for Assuring Autonomy at the University of York developed an AI-based clinical decision-support model based on 1.4 million records of diabetes patient data from the Bradford region.
It’s hoped that the research will lead to AI being used to help those with Type 2 Diabetes (T2D) avoid the worst outcomes. The AI support-model, by drawing on data about a patient, can help clinicians in choosing a personalised pathway, such as a particular diet, exercise regime, or a more robust medical approach.
Highest occurrence
The Bradford and Craven district has one of the highest occurrence rates for diabetes in the UK and the highest across the Yorkshire and the Humber region.
Understanding who might be at risk from comorbidities associated with T2D, like high blood pressure or heart attacks, is a critical part of ensuring patients receive the correct care.
Working in partnership with Bradford Teaching Hospitals NHS Trust, and the Trust’s Head of Clinical AI, Professor Tom Lawton, researcher, Berk Ozturk, from the University of York, used data from the Connected Bradford Data Warehouse, an accelerator set up to link health, education, social care, environmental and other local government data to drive learning health systems, prevention and population health management.
Lower the risk
As the dataset spanned four decades this enabled the research team to build a comprehensive picture of health characteristics and risk factors. However, such a large dataset comes with additional challenges, including how to eliminate biases in healthcare data, which include ensuring there is balance across key demographics, like gender, age and ethnicity.
This helped lower the risk of incorrect diagnosis around the comorbidities of T2D, leading to more accurate and personalised care.
Dr Philippa Ryan, from the University’s Centre for Assuring Autonomy, developed a method which identified and highlighted potential biases in the dataset used. She said: “The increasing use of AI to support clinical decision making can make a positive step forward in developing targeted and effective care, but it’s crucial that the training data an AI system uses is as balanced and equitable as possible.”
AI safety
Researchers believe that the system also has the potential to deliver significant positive results in those of South Asian heritage. The results from the research indicate that earlier diagnosis and treatment is critical to tackling the poor outcomes for those with diabetes from this community.
Berk Ozturk said: “While AI-based applications in healthcare have shown promising results, especially for managing Type 2 Diabetes, ensuring the safety of AI solutions in healthcare is just as important as developing them.
“Our research addresses the challenge of integrating safety assurance into AI models, demonstrating how AI systems designed for critical healthcare applications can be both effective and safe.”
Global scale
By 2040, it is estimated that more than half a billion people worldwide will be living with T2D, highlighting the increasing risk of serious health complications related to this condition. Effective management of T2D is therefore crucial to reducing these risks on a global scale.
Professor Tom Lawton, Head of Clinical AI at Bradford Teaching Hospitals, said: “Diabetes is a huge and growing problem across the UK, not least in Bradford which has amongst the highest prevalence of diabetes in the UK.
“Using AI to deliver personalised and bespoke care has the potential to transform lives, but it’s important that safety, both from a physical and ethical perspective are considered carefully before introducing any kind of AI.
“The research conducted by the Centre for Assuring Autonomy in the use of AI in healthcare is key to helping the safe rollout of such technologies.”
