The practical implementation of AI in health and care is still a relatively new space, yet to establish clear best practice for implementing AI across organisations and managing change. This uncertainty is challenging for leaders who are expected to keep up with the rapid pace of technical innovation.
The NHS AI Lab
The NHS AI Lab was set up to accelerate the safe and effective adoption of AI in health and care. Its mission is to create a sustainable health and care system which achieves better outcomes, equality and fairness for all. It does this by supporting AI technologies that have the potential to improve the quality of health and care services, while building a robust ethical and regulatory framework to ensure patient and citizen safety.
Working with NHSX on the NHS AI Lab and the AI in Health and Care Award, we had the opportunity to speak with many specialists and people across the UK about their experiences implementing AI, the lessons they’ve learned and the change AI has supported in organisations.
Some initial challenges experienced by early AI adopters will be familiar to anyone who has experienced digital change in health and care. The complexity and unfamiliarity of AI can add additional barriers to effective adoption. Despite this, many have seen widespread positive impacts for their efforts, from improving care to saving money and assisting in Covid-19 responses.
By sharing the lessons learned from these organisations, we hope to highlight crucial factors which will help smooth the experience journey for organisations thinking of implementing AI in the future.
As part of its work, the NHS AI Lab provides resources for those looking to understand, develop or implement AI, including case studies, guidance and research. The NHS AI Lab is committed to bringing people in the NHS and care together to share lessons and help organisations learn from one another. It has recently launched the AI Virtual Hub on NHS Futures to provide a space for collaboration and best practice sharing for those interested in the development and adoption of AI in health and care.
AI in health and care is rarely just ‘plug in and play’
Even ‘off-the-shelf’, pre-developed and pre-validated AI solutions require a level of iteration and development before they can be used. AI models may need to be trained on data that’s specific to an organisation or adjusted to make sure it’s compatible with existing systems and workflows.
"AI companies have the tech know-how to develop the algorithms but don’t have the understanding of clinical pathways that’s necessary to actually implement it. [They’re] different in every organisation"
CEO, NHS Trust
Leaders need to understand these development requirements upfront and that their AI solution may not be ready to implement straight away.
"The integration cost of AI into your existing system or infrastructure is often a lot higher than people expect. Many organisations are built around enterprise systems like Microsoft, whereas AI products are often built on Linux, which can make them very hard to support."
Clinical Director of Data Science
Even if AI has been implemented successfully in one pathway, it shouldn’t be ‘lifted and shifted’ into other areas without additional evaluation and development. A slightly different pathway or context could radically change the efficacy and value of the AI.
"You can’t just cut and paste the solution. The context can radically shape the risk"
The quantity (and quality) of data can make or break AI
Data access and management is a common source of frustration and delays in getting started. AI requires a lot of data and many organisations need to change their ways of working to access the data they need. The complex world of data sharing agreements and governance, paired with a fear of GDPR, can be daunting. It’s crucial to begin data gathering sooner rather than later.
"We’ve got acute data, social care data, but can’t get GP data. We’d need individual agreements with 40–50 GP practices. There’s lots of useful information they collect that we don’t have access to yet."
Product manager, local authority
Organisations starting their AI journeys need to know the current state of their existing data because the process of ‘cleaning’ it takes both time and skill. It could be as simple as consolidating multiple datasets for consistent formatting or an extensive process of digitising patient records, reformatting years worth of retrospective data and collecting new data.
"We often start projects, people say they have clean data, no GDPR issues, then it takes three months to work through all that. Those things are generally underestimated."
Director, National Centre for Data Innovation
The challenge for organisations to access the type of data they need highlights the importance of initiatives like the National Covid-19 Chest Imaging Database, which is helping coordinate data collection, support the development of technology and together, helping create the best care for patients.
It all comes back to changing hearts and minds
As is common with the adoption of digital technologies, getting the technical aspects right is only one part of the change. Engagement from staff and patients is essential and building trust around AI is a crucial part of this.
"Getting people to change how they work to use new features is difficult. It’s easy to put a system in, but they’ll default to using paper."
The narrative around the change is the biggest facilitator to engaging people. Acknowledging challenges upfront, setting out the potential future benefits and outlining exactly what’s needed and why is the best way to change hearts and minds. We can help further demonstrate value by involving staff who speak both the clinical and technical ‘languages’ and can be useful translators for the core message of benefits and change.
"What I’d want to know is how these systems behave. When as a radiologist can I trust it, and when should I even trust it over my judgement? That’s a really difficult thing to work out."
The work is never done
AI in health and care is an ongoing commitment that requires updating, tinkering and iteration, improving based on the ongoing feedback from users. It’s also essential to factor in adjustments to the AI model that may need to be made over time, such as software or hardware updates.
"A provider might buy a breast screening AI to replace the first reader. And after five years they go through an equipment refresh, meaning lots of new equipment. Their AI vendor might not support new hardware yet, especially if the new hardware hasn’t been around long enough for lots of data to be available to train the algorithm. So what does that [provider] do while their AI doesn’t work? They’ll have to revert back to humans reading the first images. In those scenarios that’s a real challenge."
Head of Scientific Computing
This process requires both ongoing time and skilled resource which should be factored in from the beginning of implementation.
Embedding AI in our organisations
The benefits of implementing AI in the health and care sector have the potential to be life-changing for both patients and staff. Organisations which are prepared to examine and adapt their mindsets and ways of working, alongside adopting the tech, are set to be the most successful.
AI solutions and technologies are no longer parts of a story in future decades. They’re very real and viable options to embed in our organisations that can help alter citizens’ lives for the better.
Our recent insights
Andy Callow, CDIO of Kettering General and Northampton General NHS Trusts, on using digital as a tool to improve patient and staff experience.