The NHS can get value from artificial intelligence (AI) applied to big data, but there is much to do on data quality basics, said a group of doctors and policy chiefs at a Westminster Health Forum seminar last week.
The event was chaired by Chris Ruane, Welsh Labour MP and vice-chair of the all-party parliamentary group for data analytics.
In his opening remarks, Ruane said: “We’ve seen the corruption, or alleged corruption, of big data in recent years and there is a big fear out there among people that, with every click, the analysts are getting further into their brains. Big data with big psychology could end up with Big Brother.”
Ruane added later that data visualisation software is a great boon to “enable the ordinary person to understand”, adding: “There is massive potential out there [for data analytics]. But there is an issue of public and professional trust. If you are going to take out 40% of the jobs in the National Health Service, you need to explain that the new technology will bring along other jobs, better jobs.
“There is a job of work to do in political leadership here, winning hearts and minds. Data visualisation can help there, raising this to the level of art, as with David McCandless in his book Data is beautiful.”
Simon Eccles, chief clinical officer for health and care at the Department of Health and Social Care, and a practising accident and emergency consultant at St Thomas’ hospital in London, said in his address: “The time is now right [for applying digital technology in the NHS]. Public expectations are different to the time of the National Programme for IT in the NHS [NPfIT].”
He cited the smartphone explosion since those baleful days – extensively chronicled by Computer Weekly at the time – and described the “tech vision” of new health secretary Matt Hancock as “genuinely impressive”.
Going down a rabbit hole
Eccles said the NPfIT lesson was that government cannot drive digital change from the centre, but also that it delayed digital in the NHS. “Unless you are improving care, you are going down a rabbit hole,” he said.
“You can’t do everything with an app, but a lot can be done on an app,” he said. The aim now, he said, is that “every citizen who needs a care plan has one co-produced with a clinician and that it is digitally accessible”.
Eccles added: “I think it is really important to make the NHS transparent to the citizen. We are effectively going to translate the Bible into English.
“We in the NHS are not terribly digital. We are doing things the way we have always done them. The model of care is the 1948 model: the GP in the surgery, the community practice and district nurse in the car, the hospital with consultants and junior doctors in white coats.
“My great aunt ,who trained before the NHS, would recognise every element of our organisational structure. We need to digitise the NHS alongside changing our care processes. We have been reasonably good at doing each, but not joining them together.”
As regards AI, Eccles said: “We don't use AI much in healthcare, and we should.”
Read more on the NHS, AI and big data
- NHS data not fit for AI, Lords select committee told.
- If we don’t digitise the NHS, we are condemning people to die, says NHS England’s operational director.
- Data could be a huge advantage for the NHS, but data quality and accessibility is still a big challenge, says NHS Digital CEO Sarah Wilkinson as she promises clear guidance on standards and interoperability.
He gave the example of demand prediction. “Amazon knows what we are going to do,” he said. “We should, but we don't. We treat the surge in admissions on a Monday as if it were a surprise. Every Monday! It is quite daft.”
But Eccles said he was wary of the suggestion that AI can replace doctors. “Augmented analytics can support us to make decisions, but better organisational structures are needed too,” he said.
Charles Gutteridge, chief clinical information officer at Barts NHS Trust, said the fundamental problem with data in the health service is that: “We accrete a carapace of data as we go from birth to death, some of which is easily available, but some of which is locked away in silos and hard to reach. And releasing that data in a governed way is important to getting future efficiencies.
“There is a big issue for doctors and nurses, however, and that is that humans are much smarter than computers.”
Gutteridge added: “We are not there yet with NLP [natural language processing] being usable on narrative text.”
Daniel Ray, director of data at NHS Digital, said there is “still a lot of paper in the health service” that is effectively unanalysable. But his organisation is building a new data services platform to capture big data and do more by way automated analysis, deploying machine learning algorithms, he said, with the “seamless flow of data around the NHS” being the goal.
Code of conduct for the NHS
Ray drew attention to an NHS Digital AI code of conduct for the NHS, summed up in 10 principles, and published on the gov.uk website. He asked for feedback to develop the code, the principles of which include: “being transparent to the limitations of the data used and algorithms deployed”; “make security integral to the design”; and “show what type of algorithm you are building, the evidence base for choosing that algorithm, how you plan to monitor its performance on an ongoing basis and how you are validating performance of the algorithm”.
Ray said the NHS is roughly on a par with the US in the application of AI to healthcare. “People imagine that in other parts of the world, people are significantly ahead, and there are robots walking around treating patients,” he said. But in terms of predictive analytics, chatbots and healthcare trackers, the UK is at a similar stage to the US, he added.
Guy Boersma, managing director of the Kent, Surrey, Sussex Academic Health Network, gave an example of a chatbot called Oly in use at Alder Hey Children’s Hospital in Liverpool. He said a clinician told him one child asked Oly if they would wake up from their operation – a question that that particular healthcare worker had never heard asked of a human medic.
Boersma said, citing the classic Gartner hype cycle, that chatbots, predictive analytics and speech recognition were all in use in the health service, demonstrating substance.
Adam Steventon, director of data analytics at The Health Foundation charity, urged forum attendees to “think about raising the floor as well as ceiling” when it comes to data analysis. He said his organisation estimates there are some 10,000 people analysing data in the NHS in the UK, but who are not properly supported with career development and are isolated.
Most of their time is spent “feeding the beast” with data reports, he said, rather than performing higher-value analysis.
But Steventon cautioned against “technology for its own sake”, adding: “It is better to focus on [data analytics] technology that addresses multiple health conditions, and that improves healthcare services. Public trust lost is hard to regain, as care.data showed.”
NHS England scrapped the controversial Care.data project in July 2016 after the national data guardian for health and care, Fiona Caldicott, asked the government to consider its future.
“Analysts have been stuck in the basement for too long,” said Steventon. “They have a vital role to play in supporting innovative health services and supporting new forms of data.”