Abstract
The SmartCHANGE project, a Horizon-Europe Research & Innovation project (2023–2027), aims to develop AI-driven decision-support tools to identify and reduce the long-term risk of non-communicable chronic diseases (NCDs) in children and youth. NCDs represent a significant public health burden in developed countries, with risk factors like obesity and low physical fitness often established during childhood. Current risk-assessment tools primarily target adults, leading to missed opportunities for early detection. SmartCHANGE addresses this gap by creating trustworthy AI tools that assess risk factors in youth and promote optimized strategies for risk reduction. This paper outlines the project's methodology, including the use of massive longitudinal datasets, federated learning, and explainable AI to drive meaningful behavior change.
Introduction: Shifting the Paradigm of Prevention
Non-communicable diseases (NCDs)—including cardiovascular disease, type 2 diabetes, and obesity-related conditions—account for over 70% of mortality in Europe. While these diseases typically manifest in adulthood, the physiological groundwork is often laid during childhood and adolescence through lifestyle habits and early biomarkers.
The SmartCHANGE project tackles two primary challenges in pediatric health: the lack of youth-specific risk calculators and the difficulty of accessing sensitive health data across international borders. By building a framework that integrates heterogeneous data (from wearables to clinical records) into a unified model, the project aims to move medical intervention from reactive adult care to proactive, personalized prevention for youth.
Methodology: Building Trustworthy AI
The authors propose a multi-layered technical approach to ensure the AI tools are both accurate and clinically acceptable.
Federated Learning and Privacy
To solve the privacy issues inherent in pediatric health data, SmartCHANGE employs Federated Learning (FL). Instead of moving sensitive patient data to a central server, the model "travels" to the data. Local institutions train the model on their own datasets and only share encrypted weight updates, which are then aggregated into a global model. This ensures data remains localized and secure while benefiting from a massive, diverse training pool.
Explainable AI (XAI) and Counterfactuals
For AI to be useful for clinicians and parents, it must provide more than just a risk score. SmartCHANGE integrates XAI to generate "counterfactual explanations"—essentially "what-if" scenarios. For example, the system might demonstrate how increasing a child's physical activity by 20 minutes a day could drop their long-term risk of metabolic syndrome by a specific percentage. This makes the AI's output actionable and transparent.
Data Infrastructure and Harmonization
The project utilizes a massive pool of approximately 200,000 subjects across 15 datasets. Key benchmarks include the SLOfit system (Slovenia), which monitors physical development for 13 consecutive years, and the Cardiovascular Risk in Young Finns Study (YFS), a landmark 42-year longitudinal study linking childhood factors to adult outcomes.
To handle data from diverse sources, the project maps all inputs to the HL7 Fast Healthcare Interoperability Resources (FHIR) standard. This ensures semantic consistency across the pipeline, allowing the unified deep learning architecture to process variables across the entire life course.
Key Insights and Future Directions
SmartCHANGE moves beyond simple correlation by attempting to map the causal chain from childhood behaviors to adult NCDs. A core tenet of the project is participatory design; by involving children, parents, and health professionals in the design of mobile and web apps, the team ensures the "behavior change" strategies are practically achievable.
As the project is ongoing, current results focus on the successful establishment of the federated infrastructure and the launch of four pilot studies in Portugal, Slovenia, Finland, and the Netherlands. Future work will focus on refining the counterfactual generator and measuring intermediary outcomes such as changes in health literacy and short-term behavior modification, providing a proof-of-concept for long-term NCD reduction.
Summary updated