According to the World Diabetes Foundation, more than 589 million adults worldwide are currently living with diabetes, a number expected to rise to 853 million by 2050. These projections underline the urgent need to strengthen prevention efforts and improve public awareness of a disease with an increasingly significant global health impact.
Diabetes is generally classified into two main types. Type 1 diabetes is a chronic autoimmune condition in which the immune system destroys the insulin-producing beta cells of the pancreas, requiring lifelong insulin therapy and most often affecting children and young people. Type 2 diabetes, the most common one, is driven by insulin resistance and is typically associated with adulthood — although it is now being diagnosed more frequently among younger age groups.
In the large majority of cases, type 2 diabetes is preceded by prediabetes, also known as impaired fasting glucose, a metabolic condition in which blood sugar levels are higher than normal but not yet high enough for a formal diagnosis. This phase offers a crucial opportunity for prevention. Through targeted lifestyle measures such as a healthy diet, regular physical activity and weight management, glucose levels can be brought back within the normal range, significantly reducing the risk of progression to diabetes. Early detection is therefore essential to limit the long-term health and societal impact of the disease.
This is where the PRAESIIDIUM project (Physics-informed machine learning-based prediction and reversion of impaired fasting glucose management) comes in. Funded by the Horizon Europe programme and recently completed, the project brought together an international consortium of 11 universities, research centres and industry partners, including a SUPSI research group from the Dalle Molle Institute for Artificial Intelligence (IDSIA USI-SUPSI). Its goal was to enhance prediabetes prevention by integrating advanced AI models with established physiological and mathematical knowledge.
A key feature of the project is its use of physics-informed machine learning (PI-ML), an innovative approach that embeds physical and physiological principles directly into machine learning models. “Unlike purely data-driven models, which rely solely on available data, PI-ML integrates biomedical knowledge and validated mathematical models,” explains Laura Azzimonti, senior lecturer and researcher at SUPSI. “This allows the system to build on known biological mechanisms – such as insulin production and glucose absorption – making predictions more consistent with biology and less sensitive to noise in the data.”
Together with her research group, Laura Azzimonti contributed to the development of PI-ML models designed to predict the risk of prediabetes by integrating systems of differential equations into machine learning algorithms. These equations describe the time evolution of key metabolic processes, including blood glucose regulation and the body’s response to insulin.
“This approach ensures that predictions remain consistent with established physiological knowledge, while also adapting to individual variability,” Azzimonti explains. The models combine clinical data with information collected from wearable devices that monitor physical activity, as well as dietary habits. The research team also explored uncertainty quantification techniques to assess the reliability of predictions and applied causal analysis methods to evaluate whether – and to what extent – interventions such as increased physical activity could lead to measurable improvements in metabolic parameters.
The resulting models have been integrated into the PRAESIIDIUM platform, which will be made available to healthcare professionals for research purposes. The platform supports data collection and provides clear, intuitive visualizations of personalized risk assessments.
Overall, the project highlights how explainable artificial intelligence, mathematical modelling and clinical data can be combined to create robust and practical digital tools to support public health. By enabling earlier detection and intervention during reversible stages of the disease, this approach paves the way for more effective prevention strategies and lasting improvements in people’s health and well-being.