Causal and Federated Multimodal Learning for Cardiovascular Risk Prediction under Heterogeneous Populations
Rohit Kaushik, Eva Kaushik
TL;DR
The paper tackles cardiovascular risk prediction across heterogeneous populations by integrating multi-modal biomedical data (genomics, cardiac MRI, ECG, wearable data, and EHR) within a single, privacy-preserving framework. It introduces a unified architecture combining Cross-Modal Transformer (XMT), Graph Attention Network (GAT), causal latent alignment, and federated learning to achieve accurate, interpretable, and robust predictions. The authors provide a formal mathematical framework for cross-modal representation, graph-based message passing, federated convergence, causal invariance, Bayesian uncertainty, and fairness constraints, along with convergence proofs and uncertainty quantification. Experimental evaluations on UK Biobank, MIT/BIH, and federated multi-hospital EHR data demonstrate state-of-the-art discrimination, strong calibration, minimal fairness gaps across demographic groups, and reliable uncertainty estimates, with promising clinical translation potential and explicit limitations and future directions. Overall, the work advances clinically trustworthy, privacy-respecting multimodal AI for population-level CVD risk prediction and highlights the importance of causal, interpretable, and fair decision-support in healthcare.
Abstract
Cardiovascular disease (CVD) continues to be the major cause of death globally, calling for predictive models that not only handle diverse and high-dimensional biomedical signals but also maintain interpretability and privacy. We create a single multimodal learning framework that integrates cross modal transformers with graph neural networks and causal representation learning to measure personalized CVD risk. The model combines genomic variation, cardiac MRI, ECG waveforms, wearable streams, and structured EHR data to predict risk while also implementing causal invariance constraints across different clinical subpopulations. To maintain transparency, we employ SHAP based feature attribution, counterfactual explanations and causal latent alignment for understandable risk factors. Besides, we position the design in a federated, privacy, preserving optimization protocol and establish rules for convergence, calibration and uncertainty quantification under distributional shift. Experimental studies based on large-scale biobank and multi institutional datasets reveal state discrimination and robustness, exhibiting fair performance across demographic strata and clinically distinct cohorts. This study paves the way for a principled approach to clinically trustworthy, interpretable and privacy respecting CVD prediction at the population level.