SMARTCHANGE

ABSTRACT

Non-communicable diseases (NCDs) are the leading cause of death and healthcare expense. Common risk factors for many of them are obesity and low physical fitness resulting from an unhealthy lifestyle. Targeting children and youth for lifestyle interventions has been suggested because (1) early precursors of most NCDs are already present at this age, (2) childhood and adolescence are critical periods for the acquisition of healthy lifestyle habits, and (3) unhealthy lifestyle in this age group is prevalent.

We propose to develop long-term risk-prediction models for cardiovascular and metabolic disease for people aged 5–19. We have already identified 15 datasets with data on behaviour, fitness, biomarkers and actual NCDs spanning various ages. We will develop machine-learning methods that can train models on such heterogeneous datasets, enabling the prediction of risk for people of various ages for whom different data is available. We will employ federated learning for data privacy, carefully curate and balance the data to ensure it is bias-free and representative of the target group, and employ methods for explanation and visualisation of the data, models and predictions. Participatory design involving explanation of the AI will be used to design two applications: one for health professionals and the other for citizens. Both will show the risks broken down by risk factors, and the recommended behaviour changes to reduce them, in a manner appropriate for each user group. The developed solution will be validated in a large proof-of-concept study in four countries involving different health settings (family, school, primary care, integrated care …).

To facilitate practical use of the developed solution, we will prepare recommendations for their implementation, and a realistic exploitation plan. These activities will be supported by dissemination and communication activities specifically tailored to the target groups (e.g., involving science museums).

AIM (WHAT)

Non-communicable diseases (NCDs) are currently the leading cause of death and healthcare expenses, with the total costs expected to rise to more than 12 trillion euros by 2030. In the European region alone, up to 90% of deaths are caused by NCDs. These figures have shed light on the need for a cost-effective and resource-efficient long-term management of NCDs. The SmartCHANGE project will address this challenge by focusing on childhood and adolescence as ideal periods to implement risk-lowering strategies by modifying behaviour in a timely manner. Concretely, using exisiting datasets and machine-learning methods to train models, the SmartCHANGE project will develop trustworthy, AI-based decision-support tools that can assess the risk of NCDs in children and young people, thus fostering the adoption of risk-lowering strategies at an early stage.

METHODOLOGY (HOW)

In order to achieve its aim, the SmartCHANGE project will build one unified model that spans different variables (e.g. physical activity, fitness, biomarkers for NCDs) across multiple ages. The project will rely on deep neural networks, which can better utilise all the available data. The consortium will also ensure data privacy by using federated learning to train risk-prediction models, which means that no data will leave trusted local storage (e.g., hospital). Furthermore, the data used to train the models will be representative of the target population, so that the models will perform robustly on previously unseen data. The models and their predictions will also be explainable, i.e., they will be explained to health experts and other stakeholders. The SmartCHANGE project will rely extensively on the engagement of different users (e.g. health professionals, public health policy makers, educators, children and families) so as to collect data on their needs and preferences. Following the development of the SmartCHANGE system, the partners will carry out evaluations through proof-of-concept studies.

IMPACT (WHY)

SmartCHANGE will contribute to raising awareness about the importance of healthy lifestyle and will encourage citizens to better manage their health. Furthermore, the project will allow health professionals to improve the detection of high-risk individuals at an early age and engage them in personalised therapeutic lifestyle programs recommended by the AI models. Through SmartCHANGE solutions, citizens will benefit from optimized health care measures superior to the standard-of-care being offered, validated and positive outcomes provided by accurate risk prediction models, trustworthy federated learning approaches, and explainable models.