A resident doctor reviews a patient late in the day. The presence of an infection is uncertain. The guidelines are long and complex, and time is limited. The consultant wants a decision. The patient is expecting treatment.
Does the doctor prescribe antibiotics or not?
This is the reality of antimicrobial prescribing in hospitals. Decisions are often made under pressure, shaped not only by clinical evidence but by time constraints, hierarchy and patient expectations. These decisions matter. Every unnecessary or inappropriate prescription contributes, in small but cumulative ways, to a much larger global challenge: antimicrobial resistance (AMR).
AMR occurs when microbes such as bacteria, viruses, fungi and parasites no longer respond to the drugs used to treat them. It has been described as a “silent pandemic” because it builds gradually – in infections that take longer to treat, in extended hospital stays and in the slow narrowing of effective treatment options.
The scale of the challenge is stark. Global estimates suggest that AMR was associated with 4.95 million deaths in 2019.
Research led by Dr William Waldock, Clinical Research Fellow at our NIHR Northwest London Patient Safety Research Collaboration and supported by the Fleming Initiative and published in Nature npj responds to this wider challenge by exploring how antimicrobial resistance can be better measured across healthcare systems and addressed in clinical decision-making.
AMR is not just a scientific challenge; it is also behavioural and systemic. The knowledge needed to prescribe antibiotics correctly already exists in clinical guidance, alongside diagnostic information. Yet across hospitals and community care settings, this guidance can be difficult to use, inconsistently applied or overridden altogether.
In Dr Waldock’s two recent studies, the researchers set out to address this challenge from two complementary angles: how antimicrobial resistance is measured within healthcare systems and how clinicians can be better supported to make prescribing decisions in real time.
Measuring the problem: why antimicrobial resistance is hard to track
If AMR is such a significant global threat, why is it so difficult to control? Part of the answer lies in how it is measured.
Traditionally, AMR has been tracked through surveillance data – monitoring which bacteria are resistant to which drugs and where those patterns are emerging. While this provides an important picture, it does not always capture how resistance is experienced within healthcare institutions.
In the first study – Development of the antimicrobial resistance burden score through a modified eDelphi – the researchers highlight this gap and propose a new approach.
The study shows that relying on isolated indicators can be misleading. A hospital with high antibiotic use may still be practising strong stewardship, while another with lower reported resistance rates may reflect limited diagnostic capacity or incomplete reporting.
The AMR Burden Score brings these different measures together into one structured view, combining resistance patterns, prescribing practices and clinical outcomes. This allows healthcare organisations to better understand their AMR burden and assess whether interventions are making a meaningful difference over time.
In doing so, it provides a more integrated and interpretable picture of AMR within healthcare systems, allowing patterns to be tracked over time and the impact of interventions to be more clearly assessed.
The decision problem: why prescribing is so difficult
If measuring AMR is one part of the challenge, the other lies in how prescribing decisions are made.
Antibiotic prescribing is rarely straightforward. Clinicians often need to make decisions quickly, sometimes with incomplete information, balancing the risks of under-treating infection against those of unnecessary antibiotic use. While guidelines exist to support these decisions, they are not always easy to apply in busy clinical environments.
The research highlights how this complexity plays out in everyday care, where time pressure, workflow design and differences in clinical judgement all influence how guidance is applied.
As Dr Waldock explains:
“In-hospital antimicrobial prescribing is frequently driven by the urgency of a deteriorating patient. While diagnostics are vital, clinical reality often demands pre-emptive intervention before full data is available. Junior doctors may occasionally have senior guidance, but more often, they must navigate these high-stakes initial prescriptions alone. In such moments, objective and accessible resources to guide the first response are invaluable.”
In this context, prescribing decisions are not made in isolation. They are shaped by interactions between clinicians, patient expectations and wider system pressures. This can create situations where decisions that seem appropriate for one patient in the moment may conflict with longer-term public health interests.
As Dr Waldock puts it:
“This reflects what economists call the ‘tragedy of the commons’ – where individual decisions, whether driven by clinical caution, patient expectation or organisational pressure, can work against the long-term interests of the wider population.”
This helps explain why improving antimicrobial use is not simply a matter of producing better guidance. The challenge is not only what clinicians should do but also how they are supported to make decisions in complex clinical environments.
It is this gap between knowledge and its application that the second study seeks to address.
Supporting better decisions: the role of AI in prescribing
If AMR is shaped by everyday prescribing decisions, the next challenge is how those decisions can be better supported.
In the second study – Enhancing quality of antimicrobial prescribing through ‘Ask Eolas’ (language model): a user-testing and simulation evaluation – the researchers explored how AI can support prescribing decisions in clinical settings.
Ask Eolas is an AI-supported clinical decision tool designed to help clinicians access the right antimicrobial guidance more quickly and accurately. The tool retrieves and summarises reliable clinical guidance while providing clear links back to source material, allowing clinicians to verify its recommendations.
On the value of Ask Eolas in clinical settings, Dr Waldock notes:
“Ask Eolas appears to make the prescribing process much safer and more reliable. Unlike previous tools that could feel like a ‘black box’, this technology is transparent about why it is making a suggestion. This gives clinicians more peace of mind and makes their daily workflow feel much smoother.”
In a structured simulation study, Ask Eolas outperformed both traditional PDF guidelines and existing digital tools. Participants using the system achieved fully accurate prescribing decisions across the study scenarios.
Clinicians also reported higher confidence and lower cognitive workload when using the tool, describing it as clearer and easier to use than traditional guidance formats.
This highlights that improving antimicrobial use is not only about providing the right information, but about presenting it in a way that can be used effectively under pressure.
While these findings are based on a controlled simulation, they provide early evidence that carefully designed AI tools could support safer and more consistent prescribing in clinical settings.
Looking ahead: towards more responsive healthcare systems
Together, these two studies outline a more connected approach to antimicrobial stewardship, where better data and better decision-making reinforce one another.
The AMR Burden Score provides a more complete view of how AMR is developing within healthcare systems, while Ask Eolas supports clinicians to make more accurate, evidence-based prescribing decisions in real time.
This reflects a wider transformation in healthcare. Rather than relying solely on static guidance and retrospective review, there is growing interest in more responsive systems that can support clinical decisions as they are made.
Looking ahead, Dr Waldock points to a move towards a more “agentic” hospital:
“The agentic hospital is a shift from doctors using tools to doctors leading a team of ‘agents’ in the delivery of healthcare. Instead of a computer just holding your medical records, it’s now an ‘agent’ that supports your care: spotting risks before they happen, coordinating your tests instantly and handling the paperwork so your doctor can spend their time focusing entirely on you.”
There is still more to do. Both studies highlight the need for further validation, real-world testing and continued collaboration across healthcare systems. They also demonstrate what is possible when research is grounded in real clinical challenges and designed with end users in mind – central to the Fleming Initiative’s work in harnessing technology for real-world impact.
AMR may be shaped by decisions made every day. Strengthening how those decisions are supported in clinical settings will be central to any meaningful response.