HARMONY: Hybrid Aggregation and Reputation Mechanism for Cooperative Federated Survival Analysis

Federated Learning (FL) holds promise for digital health by enabling collaborative model training without compromising patient data privacy. However, differences in data completeness, censoring rates, patient populations, institutional reliability, and contribution quality across healthcare institutions remain major challenges. In this paper, we propose HARMONY, a peer-driven reputation mechanism for federated healthcare that employs a hybrid communication model to integrate decentralized peer feedback with clustering-based noise handling to enhance model aggregation. Crucially, HARMONY decouples the federated aggregation and reputation mechanisms by applying differential privacy to client-side model updates before peer evaluation. This protects sensitive information during reputation computation, while unaltered updates are sent to the server for global model updates. Using the Cox Proportional Hazards model for survival analysis across nodes, HARMONY addresses both data heterogeneity and reputation deficit by dynamically adjusting trust scores based on local performance improvements measured via the concordance index. Experimental evaluations on synthetic as well as the SEER dataset show that HARMONY consistently achieves high and stable C-index values, down-weighting noisy clients and outperforming FL methods without a reputation system. Across states, HARMONY improves C-index over no-reputation (FedAvg) by 0.1-4.6% (median +1.3%, IQR 0.5-2.9%) and matches or exceeds TFFL in 9/10 states; under 20% adversarial clients it reduces round-to-round C-index variance by $\sim$35% while adding 12% communication overhead (vs. FedAvg) and 9% client-side compute for peer evaluation.