The UK government has announced three priorities for reforming the UK National Health Service (BBC News):

• move from an analogue to a digital NHS.
• shift care from hospitals to communities.
• pivot to (focus) on prevention instead of treating illness.

I suspect that most health services around the world share these goals (except that some health services are already properly digitised). Stopping someone from getting sick in the first place is far better than letting the person get sick and then treating them, from a health, quality of life, and financial perspective. And too much is done in hospitals, doing more in communities would again have major benefits for health and quality of life as well as saving money.

Can AI help the health service meet the above goals? Unfortunately, the AI in Medicine community mostly focuses on helping hospital doctors do better diagnoses, which does not help with any of the above goals. Of course its useful to have better diagnoses in hospitals, but I have never seen anyone raise this as a priority goal for the NHS or other health services (blog and blog). I appreciate that ML people like to work on diagnosis because its easy to frame this as an ML problem and gets good press coverage, and hospitals usually have much larger and higher-quality data sets than community health centres, but improving diagnosis in hospitals does not help with the above goals.

Digitisation

The first goal is to digitise the NHS, largely in order to make it more efficient. I dont see AI as playing a major role in this, in part because perhaps the biggest problem is change management (people), not technology. In the early 2000s, the NHS spent 10 billion pounds on Connecting for Health digitisation programme. It was never used, and has been described as one of the biggest IT disasters ever, largely because doctors refused to use it (in part because the developers focused on what managers wanted, not what doctors wanted).

So this is a huge challenge, but its a classic software engineering and change management challenge, not an AI one.

Shift care to communities

The second goal is to shift care away from hospitals, and towards communities. GPs, community health workers, and individual patients should do more, and hospitals should do less, This has major financial benefits (hospitals are expensive), and will improve healthcare (problems should be identifies and acted on quicker in a community context) and quality of life (being a hospital in-patient is not a nice experience).

I think the TV show Call the Midwife shows what this looks like – hospitals exist, but most care is in the community.

AI can definitely help here is it focused on supporting GPs, community health workers, etc. Certainly GPs see a lot of potential in AI (I was asked to speak to a group of local GPs next week, but unfortunately had to decline). I have been involved in some GP/community projects over the years (eg, summarising consultations and our current project on supporting people with skin cancer at home), and some of my current PhD students are also working on topics in this area. There is also a lot of potential in using AI on community health centres for screening, testing, etc.

Using AI in the community is more challenging than using it in hospitals, not least because hospitals usually have more and higher-quality data. Ie, while in a hospital context data sets can simply be fed into an AI system, in a community or GP context getting the necessary information from patients (especially if they are elderly, stressed, or not feeling well) can be a major challenge which also needs to be addressed. But we should regard these as challenges, not a blockers, if we want to help the NHS and other health services achieve the goal of shifting care to the community.

Pivot to prevention

Every health service in the world wants to do more to prevent illness. If we can stop people from getting sick in the first place, or detect illness at an early stage where it is easier to treat, this will have a huge impact on health and quality of life, and also cut costs dramatically.

One area where AI has been used is behaviour change, ie encouraging people to have healthier behaviour. This is difficult (one project was involved in failed to have any effect (paper)), especially if the goal is long-term change. But again this should be regarded as a challenge, not a blocker, and I personally am excited about the potential of LLM-powered chatbots in this space.

Work on using AI for diagnosis could have a real impact here if it was oriented towards early detection of problems in the community. I appreciate that many clinicians are sceptical of the Neko Body Scan, but if this kind of thing could be made to work well, and was widely used (by poor people as well as people who can afford to pay hundreds of pounds), it could have a real impact.

Final thoughts

It sometimes seems like 90% of AI/Medicine research is about improving diagnosis in hospitals. This is certainly the impression of media, the public, and even funders (a recent Healthcare Technologies panel in UK refused to consider a proposal from me on using advanced chatbots for behaviour change, because they didnt regard this as fitting the Healthcare Technologies remit). Which is a shame, because while better diagnosis in hospitals is certainly useful, it is not a key priority for the UK NHS, or indeed any other health system which I am aware of.

I think there is a lot of potential to use AI to address the most important challenges facing the NHS and support the kind of health system which is seen as the future in the UK and many other countries. It won’t be easy but we as researchers should welcome new and interesting challenges!