How can AI help the NHS?

We cannot underestimate the level of buzz on AI in health. In the UK, this is evidenced by the volume of projects in hospitals, universities, and the private sector. And there’s an ever growing number of new start-ups. So can AI help the NHS?

While AI is not new, what has changed is the sheer volume of data that can be meaningfully analysed and applied to health settings.

But what AI means for the NHS remains unclear. Work is underway to figure this out. NHS England are developing a terminology and classification document, and a taxonomy of problems to be addressed. They are also considering a Code of Conduct which will lay out the rules of engagement for working with the NHS.

Given my interest in AI in health and care, I jumped at the chance to attend a discussion hosted by AI Med on how AI can help the NHS. The meeting began with a video message from Dr Anthony Chang, the Chair and Founder of AIMed. Dr. Harpreet Sood, the Associate Chief Clinical Information Officer of NHS England, then moderated the panel. It featured Vishal Nangalia, Consultant Anaesthesiologist, Royal Free London NHS Foundation Trust, and Matthieu Komorowski, Locum Consultant in Intensive Care and Anaesthesia, Imperial College Healthcare NHS Trust.

The discussion centred on three areas

• where AI could make a difference;
• addressing some of the practical barriers in accepting AI; and,
• workforce implications.

Where AI could make a difference

What became clear throughout the conversation is that the focus here is not on replacing clinicians, but on augmenting their activities. As one speaker said, robots and algorithms are really rather dumb.

In health, AI offers the opportunity to automate relatively simple tasks  – form filling, interpreting large volumes of scans, identifying people earlier for medical interventions, improving communications. While cost savings may be made, the focus of AI applications must be on improving patient outcomes.

Addressing the practical barriers

We know well that there are real challenges to the adoption and spread of innovations in the NHS. I won’t go into these well rehearsed arguments here.

So what did the speakers think were the specific barriers to AI?

• Procurement – Cycles take roughly 18 months and the system is far from easy to negotiate. So what could be done at the system level while ensuring safety and efficacy?

• Integration of workflow – How do you get people to change the way they’ve always done things? This goes to the heart of the innovation challenge – workforce adaptability and system change.

• Explainability – In developing AI applications, one of the challenges is to do so in a way that the outputs are explainable. In health however, this comes with some additional challenges.

First, in medicine, there’s still a lot we don’t know. For instance, we still don’t understand the full mechanisms of how paracetamol works.

Second, how do we ensure the inclusion of ‘outlier’ data? To explain, adding data to algorithms can create a lot of noise. But some data may be excluded to clean up the ‘noise’ to identify and learn from dominant trends and patterns. But this ‘outlier’ data can represent the experience of real world people. It could be someone with complex health needs and/or disabilities, or a case where disease progression doesn’t fit the norm, or a lack of recognition of diversity [you an listen to a great podcast on this issue here].

• Regulations –  It’s essential to have the right regulatory framework. This touches on a whole host of issues – patient safety, ethical use, explainability. I’ve flagged in a previous post that there is a lot of activity in this space.

• Digitisation and Electronic Health Records – getting this right will be essential.

• Education and Training – under workforce issues

Workforce issues

There remains a skills gap in the NHS when it comes to AI and the required data science skills.

And government recognises this as evidenced but the Topol Review – the interim report recently having been published. The Secretary of State for Health and Care commissioned the review to explore how NHS staff can make the most of technology. The focus is on genomics, digital medicine, and artificial intelligence and robotics.

In addition, there is a real need to have data science as part of training curricula.

It is clear from the discussion that we’ve still got a long way to go in cracking the issue of scaling up innovations, and this remains the case with AI in the NHS.