Table 1 Advantages and disadvantages of alternative approaches to data sharing

The ‘Beacon’ model

• A common web service allows researchers to discover data relevant to their research without the data holder storing the data outside the host institution

• Comparatively easy to implement

All those of share-on-request systems, including:

Being piloted by Global Alliance for Genomics and Health for genomics data https://genomicsandhealth.org/work-products-demonstration-projects/beacon-project-0

• Can improve discoverability of clinical datasets which cannot be openly shared

• Lack of data preservation guarantees

• No independent governance of data requests

• No common system for citing datasets

The ‘Federation’ model

• Separate, locally controlled data resources share a common index and data transfer protocols

• Improved data preservation over the Beacon model

• Data preservation relies on multiple partner nodes.

Global Alzheimer’s Association Interactive Network (GAAIN) [40] http://www.gaain.org/

• Easier for institutions and ethical committees to accept because data holder does not give up control of the data to an independent repository

• No independent governance of data requests

• Terms for anonymous peer review of data, if permitted, would likely need to be negotiated with each node independently

• Linking with the literature possible if stable data identifiers are used across the whole network

The ‘Iron-safe’ model

• Data stored in a hardened, centralised resource and analysis conducted within the confines of the system

• Appropriate for highly sensitive data collected in the course of clinical care

• Access barriers may be prohibitive

Planned for 100,000 English Genomes system (http://www.genomicsengland.co.uk/the-100000-genomes-project/data/)

• Anonymous peer review of data generally impossible

• A centralised resource can, in principle, provide an independent system for vetting and providing access, helping avoid the creep of biasing access requirements like co-authorship

• Difficult to link data with literature in a robust manner, if the index of the data resource is also protected

Also, similar to the Clinical Study Data Request (CSDR) model

• Data export from the system is limited and tightly controlled