Let's talk

Could AI transform the NHS?

AI Life sciences
Dr Ben Bray Partner and Evidence Generation Lead

Laura Amin and Dr Ben Bray…in conversation

Laura is an actuary and partner at LCP, and Ben is a medical doctor and partner in the HealthAnalytics department at LCP

This year marks the 75th anniversary of the birth of the NHS. While the NHS is a source of national pride, there has also been much coverage over the last year or so on escalating NHS waiting times, ambulance delays and the missed/late diagnoses following the pandemic.

Does AI have the potential to address some of these pressures? And how is AI already being used within the NHS? What are the current challenges? Where can AI make the biggest difference? How could AI change our experience of seeing our GP?

Does AI have the potential to address NHS pressures?

The government clearly think so as the the Secretary of State for Health and Social Care recently announced that NHS Trusts will be able to bid for cash from a £21 million pot to fund promising AI projects.

Essentially any area where diagnoses rely on images (e.g. radiology and pathology) could potentially harness the power of AI. For example, more than 600,000 chest X-rays are done every month in England, and it is hoped AI tools can support doctors to diagnose the lung cancer earlier, improving patients’ chances of successful treatment. Image interpretation and generation is one of the areas where AI has advanced most quickly in recent years, and the vast majority of the commercially available applications for AI in healthcare currently are for medical imaging software.

How is AI already being used within the NHS?

AI is already being used in the NHS in stroke diagnosis. When someone has a suspected stroke, it is important that they receive a brain scan as quickly as possible to confirm the diagnosis and guide the correct treatment. There are now various AI powered software solutions available that help healthcare professionals interpret the images from the brain scans quickly and accurately – helping patients to be treated more quickly. Stroke diagnosis is likely to be one of the first areas of healthcare where AI becomes part of everyday practice.

What are the challenges?

Despite the recent advances in AI, there are many challenges in making this part of everyday healthcare. Developing AI diagnostic tools which are consistently accurate across different settings and populations is difficult, and lots of testing and evaluation needs to be carried out to demonstrate that these tools are effective and safe.

The quality and volume of data required to train new models is also likely to be a big barrier to the use of AI in diagnostics. Although the NHS collects a huge amount of data, in practice this is often very challenging to use for research or developing new AI systems. For medical imaging AI, high quality labels (i.e. accurate and detailed descriptions about the important findings shown in each image) are needed to develop new AI systems can be very challenging and labour intensive to generate.

Where can AI make the biggest difference to healthcare?

In our view, the largest opportunity for AI at the moment lies in automating the hidden back end of healthcare – logistics, appointment booking and management, hospital bed allocation and resource management, and other aspects of the unseen work of healthcare.

Tools which can automate administrative and operational tasks can free up time for more patient care. The NHS certainly needs any help it can get to tackle these backlogs, with NHS waiting lists now at a record high of 7.4 million people and showing no sign yet of going down. Letting the AI machines do what they do best, and the human healthcare professionals do what they do best, is likely to be the best bet on using AI to save the NHS.

How could AI change my experience of seeing my GP?

For most people, seeing their GP is their most frequent interaction with the NHS. But we think the advent of the AI doctor (‘docbot’) replacing the GP is a long way off.

Beyond the technical issues of building AI models with these capabilities, there are thorny issues about regulation, safety monitoring, patient acceptability, cost and legal liability for medical errors which will need to be resolved. Instead of fully automated docbots, we see that it is far more likely for these types of tools to be used by doctors as a tool as part of a (still human) consultation – to collect or provide information, or help the doctor check that they are on the right track in making a diagnosis or recommending the most appropriate treatment. Think AI assistants or medical co-pilots instead of standalone docbots.

Where to hear more

AI is going to transform many aspects of our lives and healthcare is no different.

In our recent Beyond Curious with LCP podcast, we discussed the ways that AI is currently being used within a healthcare setting and how it could in the future, transform the NHS.

It is, however, very early days for actual real life examples of AI in healthcare and automating the unseen work of making the NHS run is really the big ticket item for AI in healthcare in the next few years.