The Accident & Emergency department at Queen’s Hospital in Romford sees 240,000 attendees every year, leading to more than 50,000 admissions. With resources increasingly stretched, the department identified a need for an early-warning prediction algorithm that can flag up the severity of presenting cases and can help with diagnosis and chances of admission. Starting from an initial Data Study Group challenge, this collaboration is developing an analytics dashboard that can be used to assist A&E nurses with quantitative insights to help with effective triaging of patients.

Explaining the science

The triage process in A&E involves a lot of uncertainty and it relies heavily on the previous experience of the nurse. Some patients are clearly ill and some patients are clearly OK, but there is large uncertainty for a substantial number of patients. Machine learning can help to summarise clinical information into risk prediction scores that can help to reduce this uncertainty. However, care has to be taken when working with clinical data extracted from routine electronic health records. In particular, it is critical to design algorithms that can account for the biases that can be reflected in such datasets (for example, due to high patient demand at specific hours of the day). 

The researchers on this project therefore started assessing the presence of such biases and their effect in downstream machine learning algorithms. The team is also exploring a range of statistical and machine learning methods to evaluate the trade-off between how much predictive power it's possible to get from a method and how interpretable the method is (i.e. how easy is to explain and understand how an algorithm has made a specific decision). 

Project aims

The project is a collaboration between the Turing, Queen’s Hospital (Barking, Havering and Redbridge University Hospitals NHS Trust), the University of Warwick and the University College London Institute of Digital Health.

The project is developing a risk prediction tool based not only on clinical variables, but also on the free-text notes that nurses type into the A&E department’s electronic system. The project is currently benchmarking methods, developments and refinements to inform the design of a subsequent digital dashboard.

The dashboard will aim to summarise a lot of information that is routinely collected and interpret it efficiently, making it as intuitive to use as possible for A&E nurses. The methods the researchers think will work best will be trialled with hospital staff who will provide their feedback, and subsequent adjustments to the back engine and the interface of the dashboard will be made accordingly.

Offline evaluation with nurses interacting with the dashboard after they’ve seen patients will be conducted and compared with real interactions with patients. In the long term, the team aims to conduct a randomised trial to assess the utility of the dashboard.

The project is working in direct collaboration with nurses at Queen's Hospital A&E to gain their feedback. Image courtesy of Queen's Hospital



Dr Nik Haliasos, Digital Transformation Lead and Consultant Neurosurgeon at Queen's acts as the main liaison between the Turing and the hospital. Of the potential application and impact of this project's work he says: “The big impact in all of this is that it involves many, many people: A&E is a front door for all healthcare services in the country. Effective intervention of data science right at the beginning of a patient’s healthcare journey could make a huge difference. However, the ultimate decision about diagnosis will always lie with a clinician."

Recent updates


Project received a Health Foundation award to increase analytical capabilities within the NHS, helping to bring data science techniques into working practice in the NHS.


Dr Nik Haliasos

Digital Transformation Lead and Consultant Neurosurgeon, Queen's Hospital

Researchers and collaborators

Amber Gibney

Data Scientist, Barking, Havering and Redbridge University Hospitals NHS Trust

Contact info

[email protected]