Could automating tasks liberate general practice? Interview with Dr Matt Willis

What did you set out to explore in studying automation in primary care?

Automation and artificial intelligence (AI) are being widely discussed in many different sectors. There’s a lot of uninformed opinion and the debate is often polarised, with utopian and dystopian visions of what technology will do to employment. 

Our research builds on the 2013 Frey and Osborne report from the US, which found that almost half of occupations, across all sectors, have aspects of the job that are vulnerable to automation. Work that involves routine, repetitive tasks is most suited to automation, but way down at the bottom of the list are occupations in health care; work that relies on creative thinking, thinking on your feet and reading social cues is harder to automate. We wanted to challenge that a little and look at the variety of tasks in health care in more detail. 

The main driver is whether we can provide some relief for primary care. NHS primary care is struggling and technology is often talked about as a solution. We want to be able to provide better evidence to inform the conversation about automation in health care, looking at what is possible with current technology. 

How did you gain insight into the potential for automating tasks?

We started by trying to identify what tasks in general practice computers are good at, and what they have trouble with. 

First, we used data from Occupation Net (O∗NET), which is a large database covering thousands of tasks for hundreds of different occupations. It holds granular data on the skills, knowledge and abilities required to do certain tasks. These data informed our analysis and we used them to develop a scale of automatability. 

I then conducted nearly 350 hours of ethnographic observation in primary care, gathering detailed data about tasks performed by all staff members, to really understand who does what and the ways things are done. 

We used these three tools – the observation data, O∗NET and the automatability scale – to inform a machine learning model that enables us to apply the scale to my observations. From this we can draw different insights about what can and cannot be automated. 

What is the potential for automation in primary care? 

There are seven areas that seem to have high potential for automation. It’s important to keep in mind that we are only talking about the technical capability to automate these tasks. This isn’t addressing the social and organisational challenges of doing so. 

The first is letter writing, which is really the backbone of communication in the NHS. I was surprised at how much letter writing and what I call ‘letter work’ happens. It’s a large part of the work in primary care as it is how information is transmitted, not just from primary care to the patient, but within primary care, and from primary care to other places. Then there are phone calls, and paperwork, including reporting, error checking and auditing, and a lot of that work has the potential to be automated. 

Further areas are clinical documentation, e-referrals and bookings, communication between primary and secondary care and communication inside the practice between staff. 

Could you give us an example of where automation could really add value in primary care?

People work with letters all day. In some practices, they have an inbox of letters, and there are always hundreds to be dealt with. Staff have to apply read codes (a way of categorising patient records) to those letters so whenever they're not busy, that's what they're doing. 

You could think of a letter as a digital object. They are really notifications of things happening in patients’ care – notifications of treatments or referrals. They don’t have to be letters. Staff are spending time doing things with these letters, when you could use existing technology to automate much of that work. 

We hear a lot about clinical automation, AI and medical imaging, and those things are helpful, but they are a very small part of the overall work of a health system. There are 2.5 administrators and support staff for every health care professional, and clinical staff spend a significant amount of time on administrative tasks too. Removing a lot of that work could really free people up to see more patients, or perform other valuable tasks.

What kind of tasks can’t be automated?

Lots of things can’t be automated. For example, relationship building with patients, consoling and empathising, helping patients to navigate the health system and helping make sense of the patient’s context. 

Anything that requires fine motor skills is also difficult to automate, like applying labels to small bottles or patient examinations that require touch. 

Another thing to remember is that even if a particular occupation has a lot of tasks that could be automated, the people doing those jobs will have knowledge and experience of working in the NHS, or working at a particular practice and you want to retain that knowledge. You can't programme a machine with that kind of experience.

What does the potential of automation mean for the future of primary care?

The people who know the future of primary care best are the people who are working in primary care. Practices have interesting and innovative ideas, but they don't have the breathing space to act on them. 

What you don't want is to apply a technology to automate something, and then have people just monitoring a machine. Automation should be about shifting and rethinking the work that people do. Does it allow people more room to innovate and do new tasks that have more value? 

There’s no silver bullet, but I think the most important thing that automation can do is to provide relief to the general practice workload. If that happens, then practices are free to create the future of primary care. 

For more information and publications, visit the Future of Healthcare website.

This content originally featured in our email newsletter, which explores perspectives and expert opinion on a different health or health care topic each month.