STARTER-KIT: A technical, procedural and ethical template for accelerating the start-up of AI multicentre studies requiring data re-use and/or sharing in clinical settings

Background Modern radiotherapy research paradigms often require large or varied datasets that can be assembled only via collaboration between multiple centres and by combining multiple data holdings.

Typical use cases include development of artificial intelligence (AI) algorithms; tackling rare cancers or subtypes where the number of cases registered at a single centre is small; gathering a complete, demographically unbiased cross-section of the population; and industry-academia co-production partnerships.

There is a great appetite within the research community for findable, accessible, interoperable and reusable (FAIR) data, which has the potential to catalyse a host of new projects in diverse areas.

Yet such retrospective data are often “siloed” within institutions or have access restricted for a variety of complicated ethical, procedural and technical reasons.

Equally, prospective studies are slowed down by the need to interact with multiple entities in a complex ecosystem of rules and governing authorities.

As a result, new multicentre studies involving data sharing take too long to “get off the ground” and repeatedly “reinvent the wheel”.

Aims STARTER-KIT will bring together complementary research and expertise, developed separately as part of the CRUK-funded RadNet, ART-NET and NCITA consortia and made available via newly designated CRUK Centres, to create an infrastructure template and worked examples for facilitating multicentre collaborative data analytics projects.

Methods Across a range of data-sharing problems, four interconnected work packages will implement a common methodology, viz.: Step 1: Gather input from stakeholders; Step 2: Develop technical solutions; Step 3: Present solutions and seek consensus for future joint work.

WP1 will examine the ethics landscape of creating FAIR data, considering particularly HRA databases and "micro-approvals" and the extent to which these can be federated, as well as issues of data quality and how to make this "findable". WP2 considers technical aspects of data access and federation.

WP3 will develop methods of federated identity management and maintenance of catalogues for tools and data, and will map the outputs of WP1 into software to create an electronic approvals workflow tool.

WP4 will organise engagement activities with stakeholders (patients and the general public, alongside AI researchers and industry partners), demonstrate the software created in WP2 and WP3 and build consensus on its ethical use.

How the results of this research will be used The deliverable of a complete starter-kit will significantly reduce the resource requirement for the setup of new studies and increase our ability to launch new initiatives in a timely fashion.