4.2 Technical implementation

Confidence that the implementation of artificial intelligence (AI) technologies is supported by appropriate information technology infrastructures and data governance.

The adoption of artificial intelligence (AI) requires integration of these technologies with existing information technology (IT) infrastructures or the development of new IT infrastructures to support data storage, security, and information provision. It also requires the adaptation or development of information governance (IG) arrangements on data security, privacy, and clinical safety.

A key insight from the interviews conducted for this research is that IT and IG processes are a major factor in healthcare workers’ confidence in AI technologies. Securing this level of confidence will require reaching internal agreements on the value of the data, the consent protocols, and the control, storage and use of the data.

Interviewees for this research noted that the broader technical challenges that relate to change and digital transformation in health and care settings complicate the adoption of AI technologies, including:

• issues with hardware and software interoperability
• time-consuming and impersonal processes for communication with IT support (for example, IT support being outsourced or located at separate sites)
• unclear structures and responsibilities in IT and IG teams
• miscommunications due to technical language and abbreviations

Of these challenges, interoperability (compatibility and ease of integration with existing infrastructures) is particularly important. Interviewees suggested that interoperability is a frequent barrier to deploying new technologies in their settings, both from the perspective of users and industry innovators. For example, health and care settings can use various software and hardware infrastructures (including record systems, pathology systems, radiology systems and patient communication tools) that often require separate access details. AI technologies that operate as separate applications would likely frustrate workers and limit their uptake.

More importantly, a site’s ability to adopt AI technologies can rely on its capacity and resources to support related IT and IG arrangements and other requirements (including commercial agreements and Data Protection Impact Assessments).

Discrepancies between clinical departments at each site can also influence the ability to adopt AI; for example, some departments maintain separate IT infrastructures to other departments (like cardiology and radiology) that impact how data is collated and stored. These discrepancies can result in inconsistent and incomplete data sets. Developing and maintaining comprehensive and representative data at a scale required by AI technologies remains a significant limitation of the healthcare system. 

Interviewees cautioned that, although infrastructure and data-related challenges are not specific to AI, they must be resolved to achieve broad deployment of, and support confidence in, these technologies in health and care settings.

Interviewees suggested that the adoption of AI technologies should be integrated into broader digital transformation systems and involve coordinated multi-disciplinary teams across clinical, technical, and administrative roles, potentially across different settings (as noted in section 4.1 and detailed in the second report). One interviewee suggested streamlining approvals for related infrastructures through centralised networks or clusters (for example, through the Integrated Stroke Networks for radiology-related technologies).

This connection between the adoption of AI technologies and broader efforts to implement change and digital transformation in healthcare settings confirms the importance of incorporating change management skills and enhancing digital literacy amongst the workforce. HEE is currently undergoing an extensive Digital Readiness programme to enable staff to identify their digital readiness and meet their training needs.

Interviewees for this research noted that the coronavirus (COVID-19) pandemic has assisted in changing attitudes towards developing new infrastructures and being open to changing the status quo in health and care settings. Many healthcare workers have been ‘forced’ to reconsider existing IT and IG infrastructures to address the growing backlog of required services, essentially embracing the importance and urgency of digital health and digital transformations.

The development and dissemination of AI-related resources and guidelines could also assist health and care settings to overcome potential challenges related to their IT and IG systems. Existing resources, like the NHS’s Interoperability Toolkit,⁶⁷ can provide a blueprint for AI-specific guidelines. Interviewees for this research spoke also of the need for standardised information on related terminologies (including on the efficiency, safety, and performance of different AI technologies) and common approaches to IT, IG, data security, and patient privacy. 

Sharing of knowledge and experiences relating to the challenges of implementing AI in healthcare settings (for example through creating communities of practice) could also be beneficial (as noted also in section 4.1). A few industry innovators have developed their own initiatives by providing ‘start-up packs’ to guide the technical deployment of their technologies within specific settings, and by coordinating peer support groups amongst their deployment sites

References

⁶⁷ NHS. Interoperability Toolkit - NHS Digital. https://digital.nhs.uk/services/interoperability-toolkit. Published 2021. Accessed March 7, 2022.