Causal and Federated Multimodal Learning for Cardiovascular Risk Prediction under Heterogeneous Populations
This work addresses the need for interpretable and privacy-preserving predictive models for cardiovascular disease in heterogeneous populations, representing an incremental advancement by combining existing techniques like transformers and causal learning.
The researchers tackled cardiovascular disease risk prediction by developing a multimodal learning framework that integrates diverse biomedical data with causal and federated learning, achieving fair performance across demographic and clinical cohorts.
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.