Federated learning limits long-term clinical predictions from distributed data
Understanding Trajectory Heterogeneity in Federated World Model Learning
Machine LearningComputation and Language
Summary
Predicting patient health over time requires seeing long sequences of medical events. The authors study how federated learning—where data stays on many separate computers—struggles to use long patient records due to ownership and participation limitations. They find that only a small fraction of long time-window data is fully accessible to each site, leading to higher prediction errors when splitting data more finely by disease severity. Different federated algorithms behave quite differently in their updates but offer limited improvements in long-term prediction accuracy.
What this means in practice
- •For hospital data teams: Improve understanding of federated training limits for clinical prediction models using patient time series data split by disease severity.
- •For healthcare ai developers: Design federated learning pipelines that realistically account for limited temporal coverage when building predictive models from longitudinal patient data.
Authors
Yipan Wei, Zhaokun Yan, Ziming Hong, Jiaqi Wu, Lixu Wang
Abstract
World models learn state evolution from trajectories, making access to temporal context a central training requirement. Federated learning can use distributed records, while ownership boundaries within a trajectory restrict the examples each client can construct. Our study benchmarks this cross-time setting through hourly action-conditioned clinical prediction on eight MIMIC-IV disease cohorts, comprising 40.87 million transition memberships. We specify severity-based client ownership, patient-separated construction, local history and future-window rules, and paired rollout evaluation from one to 32 hours. A matrix of ten federated algorithms covers 32 disease--partition configurations under five rounds of ten-percent participation. Three findings emerge from existing results and training logs. First, client ownership and participation jointly restrict long-window coverage: only 7.55\%--21.36\% of pooled-available 32-step windows have a locally complete anchor visited during training, averaged across diseases. Second, finer severity partitions accompany higher FedAvg error in 15 of 16 paired comparisons, while algorithm gains are small and horizon-dependent: FedProx reduces mean error by 0.56\%, with no consistent improvement at 32 steps. Third, algorithm labels conceal distinct update behavior, including inactive extrapolation and orders-of-magnitude differences in update scale. Cached-update performance also varies strongly across trajectory partitions under the same benchmark protocol. These results establish temporal access, participation coverage, optimization behavior, and horizon-resolved prediction as complementary dimensions for evaluating federated clinical world models.