iCAIRD – a Safe Haven model for NHS, Academia, and Industry partnership in AI.

Objective: The successful creation and evaluation of AI solutions in healthcare depends on two main factors; access to large validated datasets, and the safe, well governed, and anonymised storage of that data. Most epidemiological studies depend on access to such data, however, with the recent advances in computing power and the emergence of AI, access to these datasets is increasingly being discussed. Strictly governed data Safe Havens, coupled with ethical collaboration with NHS, Academia, and Industry, can help advance the use of AI tools within healthcare. Methods: The Industrial Centre for Artificial Intelligence Research in Digital Diagnositcs (iCAIRD) was created following successful bids to the UK Governments Industrial Strategy Challenge Fund in 2018. It has led to a £15m Scotland-wide collaborative effort between the NHS, Universities, and Industry partners, incorporating numerous stakeholders with expertise in health data research. The recent surge in Healthcare AI, and Government-led calls for industry involvement with NHS data, has led to a sense of unease at allowing free access to NHS health data. However, there is significant expertise within the NHS in creating and maintaining data repositories, such a tissue biobanks and data Safe Havens. Within NHS Grampian, there are two main projects within the Scotland-wide iCAIRD collaboration. The first is the clinical evaluation of an industry-created breast screening AI tool, with European regulatory approval for clinical use. This radiology AI exemplar will be evaluated using access to a large validated dataset. The second, more early stage project, is an open competition to interested parties to produce an AI tool that can help identify fractures on plain radiographs. Working closely with both NHS and university Research and Ethics Committees, and with Caldicott Guardians, ensures strict compliance with health data governance. This is a key component in creating a robust environment for meaningful research and evaluation of AI tools in healthcare. Results: The successful creation of a national collaborative, research focused partnership between NHS, Academia, and Industry, centred around effective, well-established, and strictly governed Data Safe havens. Conclusions: The development of iCAIRD has resulted in a robust research network, centred around well-established Data Safe Havens, championing close partnership between the NHS, Academia and Industry, allowing for the development and evaluation of novel AI tools for use within the NHS using large validated datasets.