Why fixing health inequalities means looking beyond the emergency department
A roundtable of clinicians, policymakers and data scientists asks why deprivation drives ED waiting times – and why the fix lies outside the emergency department.
On 18 June 2026, a workshop I organised at the Nuffield Department of Primary Care and Health Science, jointly with my colleagues, brought together researchers, clinicians and policymakers at the Anna Watts Building in Oxford to look at emergency department waiting times and health inequalities, as part of the NIHR-funded ED-WAITS project (Oxford and Brunel University of London). We ended the afternoon with a roundtable that moved between data, policy and clinical reality, and I wanted to share what was said there, because I’ve been thinking about it ever since.
Who was around the table
I wanted the panel to bring together voices that don’t usually share a table. James Ray, an emergency medicine consultant at Oxford University Hospitals NHS Foundation Trust and an NHS England advisor, sat alongside Alex Novak, a colleague in acute and ambulatory care whose research increasingly focuses on AI diagnostic technology. Janos Suto, a senior civil servant at the Department of Health and Social Care responsible for urgent and emergency care analysis. And then there was Siddarth Arora, associate professor of machine learning at Oxford. I’d hoped that mix of clinical, policy and data perspectives would make for a richer discussion, and it did.
Can you actually see deprivation from inside an emergency department?
The first question from the floor sounded simple: can clinicians actually see the effects of deprivation during a shift? James Ray’s answer was honest. Not really.
“In an emergency department on a given shift, it’s actually really difficult to detect that disparity. The deprivation is not necessarily an intra-departmental issue – it’s more between hospitals.”
You can see anecdotal indicators of social deprivation all around you, he suggested, but how that translates into differential care within a department is genuinely hard to spot in real time.
Alex Novak pushed this further, and in a more structural direction. Drawing on findings presented earlier in the day showing that patients from more deprived areas are significantly more likely to arrive by ambulance rather than via a GP referral or NHS 111, he raised an uncomfortable policy implication. During the pandemic, there was serious interest in the Copenhagen model, which would have fast-tracked patients who had been properly referred into the ED. The logic seemed reasonable: reward the right behaviour, improve flow. But the arrival-mode data complicates this immediately.
“If we end up separating people who use the right route from those who don’t, we could actually end up making the gap even bigger. If we really want the most benefit, we’ve got to think going into the community – we’ve got to think about how we can support primary care to actually look after people with long-term conditions.”
His point was that primary care is where the real lever sits. At the moment, he noted, GPs spend around 75% of their time on admin, and the people living in deprived areas with poorly managed long-term conditions are precisely those who end up in A&E. Trying to fix that inside the ED, he argued, is not only insufficient; it risks making things worse.
Janos Suto added the policy layer. A new Urgent and Emergency Care strategy is currently being developed at DHSC, and he was frank about the complexity of the system it is trying to change.
“No one’s got a monopoly on good ideas. There’s a lot of brilliant thinking out there in trusts, by clinicians, by operational colleagues. If we can promote that dialogue and have those conversations – that’s the win.”
He also acknowledged, with characteristic candour, some of the difficulties in this space. The constantly evolving world of UEC means that too often there are silos: urgent care separated from ambulance, from discharge, from community, each with its own workstream, never properly joined up.
The data is there – but whose job is it to use it?
Siddarth Arora took the conversation in a sharper direction. Before arriving at the workshop, he had asked Claude – Anthropic’s AI assistant – what should be done about deprivation in emergency care. The answer, he reported, was essentially: the data is there, so just use it. He disagreed, and he was precise about why.
“I would still personally not use a deprivation score on a dashboard in the emergency department – and I’m saying this as a non-medic. These scores are really important to understand what’s driving the patient to reach the ED, as opposed to what’s happening within the ED.”
His reasoning went further than just timing. Deprivation scores are population-level constructs based on postcodes. The moment you display that information to a consultant making split-second decisions, you place the responsibility for addressing structural inequality on the one person least able to address it. “I don’t think that’s fair,” he said.
The more fundamental problem, he argued, is one of architecture. GP data, community health data, ED data and hospital data all sit in silos. Joining them up, so that deprivation context arrives before the patient does, at the GP or community level, is what would actually make a difference.
For me, this landed as one of the sharpest points of the afternoon. The data infrastructure question is not just technical; it is a question about where in the system we expect action to happen, and whether that expectation is realistic.
What does good actually look like?
Janos Suto offered what he called his “one thing”: a whole-patient view that stretches beyond the episode of care. He pointed to work on value-based healthcare, some of it in Oxford, more of it in the Netherlands and Singapore, which starts from asking patients and clinicians what actually matters and routinely finds the answers are very different.
“They asked obstetricians what was important, and got very clinical answers. Then they asked the mums, and got lots of answers about incontinence, sexual health, all of those sorts of things, which just hadn’t crossed the clinician’s radar.” The analogy he reached for was simple: every time you buy something from Amazon, you get a follow-up asking how the experience was. What would it mean to build something like that across emergency and community care?
Forecasting, AI and the gap between modelling and reality
The conversation turned to technology, and here Siddarth Arora was in his element. Researchers have published models that can forecast patient length of stay in real time, whereby the forecasts are updated at different stages of the patient journey at the ED, and these models have been shown to perform well across multiple ED sites. There are even apps in development that could route patients between EDs based on expected wait times and travel distance.
James Ray added a second problem: the ground truth. ED coding is often incomplete, operationally driven, and built to explain a clinical decision to the next person in the chain, not to answer research questions. “If you’ve got unsteady data, there’s a real limit on what you’re really going to be able to do with it.”
What is DHSC actually doing about this?
Janos Suto was honest about the moment his department finds itself in: “It’s a time of change – lots going on. Change as usual, as we describe it.” A new UEC strategy is in development, with both a near-term focus on winter pressures and a longer-term ambition.
But he was also frank about what good would require. The system has a habit of organising itself in ways that look rational from the centre but fragment the patient experience at every boundary. “I think if we move the same process from NHS England into DHSC without addressing the silos, we’ll have missed a trick.”
“I fundamentally believe that people are well-intentioned. It’s a very, very complex system, and things will fall down between the gaps – not because anyone intended that, but because the boundaries are where the complexity lives.”
James Ray put it in terms of a principle he keeps coming back to: the continuum of patient need. Every person who arrives at an emergency department has a need; what varies is the timescale. Immediate, within days, within weeks, within months. Understanding that continuum, rather than organising care around the front door of a hospital, is where the real policy work lies.
What does this mean for research?
We ended with more questions than answers, but sharper questions than the ones we started with. A few themes stayed with me, and they are helping shape how I think about the next stage of this work.
The ED is not the right unit of analysis for deprivation. The disparities we see in waiting times are shaped by what happens before the patient arrives: in primary care, in community services, in people’s decisions about whether and how to seek help. Research that focuses only on what happens inside the department will miss most of the picture.
Patient experience and patient outcomes are not the same thing. A patient who waits until 2am for a scan they did not clinically need may leave more satisfied than one who was correctly sent home earlier. Understanding what good looks like, from the patient’s perspective, means asking the right questions, persistently, across the whole journey.
And the technology is further ahead than the adoption. The forecasting models, the routing tools, the length-of-stay predictors: many of these already exist and perform well in research settings. The challenge is not building them. It is working out, together, how to deploy them fairly, communicate them clearly, and make sure they do not simply embed the inequalities they were meant to help address.
More information about the project here https://www.phc.ox.ac.uk/research/groups-and-centres/health-economics-research/ED-WAITS
For more information, please send me an email at catia.nicodemo@phc.ox.ac.uk