Federated Learning for Hospitals: Governance Before Aggregation
Holding
Federated learning reduces the movement of raw data, but it does not remove the need for governance. A hospital cannot outsource accountability to the fact that training is distributed.
Authority
HIPAA security safeguards, AI risk management, and medical device quality principles still apply when protected data remain local. The legal and clinical question is not only where data sit, but what computations occur, what leaves the site, and whether the resulting model is safe for intended use.
Issue
The failure mode is uncontrolled contribution. A site may contribute biased, mislabelled, outdated, or unauthorized data-derived updates. Aggregation may hide the source of error while producing a single global model with broad clinical consequences.
Resolution
The system should define site eligibility, local data checks, secure orchestration, update validation, contribution logging, model acceptance gates, and participant withdrawal rules. Federated infrastructure should support auditability at both the node level and the global model level.
Evidence Package
The record should include data use authorization, site validation checks, training configuration, contribution logs, aggregation records, security review, model evaluation by site segment, and incident handling procedures. Federated learning is defensible only when each participant's role remains visible.