Multimodal Causal Health Modeling

Causal modeling plays a crucial role in understanding complex health phenomena, where multiple factors interact in dynamic and often non-linear ways. Our research focuses on developing causal artificial intelligence methods that move beyond correlation-based analysis toward a causal understanding. By explicitly modeling cause-and-effect relationships across clinical, lifestyle, biological, and environmental domains, we aim to uncover the mechanisms that drive changes in health and behavior. These models enable rigorous “what-if” analyses, allowing the simulation of interventions and supporting the development of personalized prevention and treatment strategies.

 

Obesity and related metabolic disorders represent a key use case for this framework. In this context, we integrate multimodal data—including nutrition, lifestyle, and clinical information—to better understand the drivers of weight dynamics and metabolic risk. While this approach enables the identification of key influencing factors, a major open challenge remains the development of temporal and dynamic causal models that can capture how these relationships evolve over time and respond to interventions.

To address scalability, our work incorporates federated learning, enabling causal analysis across large international cohorts while preserving data privacy. This allows the development of robust and generalizable models from heterogeneous datasets, supporting personalized and evidence-based interventions across diverse populations.

This research is supported in part by the European Commission and the Swiss Confederation - Swiss State Secretariat for Education, Research and Innovation (SERI) within the BETTER4U project (Grant 101080117), and the University of Bern within the https://strada.unibe.ch/.