Abstract
Federated Learning (FL) offers a promising approach to privacy-preserving clinical risk prediction, yet its deployment is limited by restricted data sharing, client heterogeneity, class imbalance, and the lack of realistic electronic health record (EHR) benchmarks. Synthetic data generation may alleviate data scarcity, but its integration with federated optimization has seen limited systematic study.
We propose SynPre-FL, a unified framework that combines high-fidelity synthetic EHR generation with synthetic-pretrained FL for robust prediction under non-IID conditions. A latent autoencoder-diffusion model generates privacy-preserving synthetic cohorts, which are used to warm-start federated training. This pretraining is followed by heterogeneity-aware optimization using class-balanced local objectives, proximal regularization, and adaptive server aggregation.
Post-hoc calibration and federated-safe explainability support reliable and interpretable risk estimates. Experiments show that the synthetic generator preserves univariate, bivariate, and multivariate structures while protecting against membership-inference and reconstruction attacks. The generated data achieve strong downstream utility under TSTR, TRTS, and model-based evaluations. Across federated settings with 5, 10, and 15 heterogeneous clients, SynPre-FL consistently improves robustness and scalability over baseline methods, especially under severe non-IID fragmentation. Calibration enhances probability reliability, while SHAP analysis yields stable and clinically coherent feature attributions across federation sizes.
Thus, SynPre-FL provides a practical and reproducible framework for combining synthetic data with FL to enable privacy-aware, interpretable, and robust clinical prediction from distributed tabular EHR data.
Blogger's Review: The introduction of SynPre-FL paves a new path for the application of federated learning in healthcare. By generating and optimizing synthetic data, it addresses critical issues of data privacy and imbalance, showcasing significant practical implications. The framework's reproducibility and robustness highlight its broad potential for clinical predictions.