Hypoglicaemic events risk prediction for type 1 diabetes subjects during physical activity

The management of chronic conditions (CCs) represents an increasing burden on health care systems in Europe and worldwide. They are on the rise, and represent 86% of all deaths in Europe. Health care systems also have to deal with multimorbidity, due to both lifestyle choices and the inherent health issues of an aging population. As a consequence, patients often have to manage a certain CC (including medication, lab tests and clinical examinations) while at the same time implementing advice on behavioural lifestyle interventions for other CCs, e.g., to stop smoking, increase physical activity, eat healthy food, etc.

Many risk factors are closely related to lifestyle, and changing the lifestyle may significantly influence the outcome of CCs. In fact, there is now increasing discussion among health care professionals as to whether lifestyle medicine should be identified as a new medical specialty.
The main objective of the project is to develop a technical prototype of a comprehensive AI-based system to provide person-centred integrated early risk prediction for multiple CCs. The main components of the system will be in place on a central server which individual citizens and patients can access on their smartphone via the WARIFA app. The integrated risk prediction enables the system to provide and improve access to preventive care within the healthcare system.
The included CCs are melanoma, cardiovascular diseases (CVD), type 1 diabetes (T1D) and chronic respiratory diseases.

Diabetes is a metabolic chronic disease characterised by high value of glucose in the blood affecting about 61 million individuals worldwide (https://diabetesatlas.org/). In T1D, also known as insulin-dependent diabetes, insulin is not produced by the pancreas, which requires an intensive insulin administration treatment.

Among the many lifestyle modifications that can be implemented, physical activity represents a major benefit for T1D subjects both from a physical and psychological point of view. However, risk of dysglycemia is higher during and after physical activity and an optimal management of blood glucose level is crucial to reduce this risk. Type of exercise, insulin management, carbohydrate intake and many other factors need to be balanced to perform at the best while reducing risk of hypo- or hyperglycemia events.

The aim of the project is to provide a personalised device to help preventing dysglycemia during physical activity. To do that, an algorithm will be trained using several data sources coming from the users (e.g. carbohydrate intake, insulin intake), from sensors (e.g. CGM), and from trackers (e.g. heart rate during activity).
The specific methodology chosen to build the algorithm is the Bayesian Belief Network (BBN). This type of model is based on directed acyclic graphs where each node represents a risk factor and each directed edge between a pair of node represents the causal relationship between that nodes. Many of the classical approaches consider risk factors as independent to one another, which is often not the case. Furthermore, BBN allow to incorporate in a rigorous way previous medical expert knowledge in the modelling process.