Figure 1.
Degree of decentralization of a federated data platform
(A) A centralized (not federated) data repository where data is pulled manually or automatically into a central data store.
(B) A hub-and-spokes model of federation (a “central access model” as described in Thorogood et al.2), where there is significant central infrastructure that the peers are required to interact with.
(C) A decentralized, peer-to-peer network, where a user sends a request to a peer node—where relevant data may or may not be—and other peers are queried. Results (represented by DNA) are then returned to the user.
(D) Results can be locally processed at each site before being returned; in this case, perturbing results for the purposes of privacy enhanced analytics, in this case local differential privacy.12 We can further categorize well-known health data federations along dimensions of decentralization (central-access to peer-to-peer).
(E) Level of access to data (pre-specified queries to arbitrary workflows).
(F) The range of different kinds of datasets supported by the various platforms.