The BEACON project aims to improve how we identify children in hospital who are at risk of becoming seriously unwell, using data from their routine observations, such as heart rate, breathing rate and blood pressure. We are developing and testing ways to combine this information to predict deterioration and give healthcare professionals an earlier and more accurate warning when a child may need urgent attention. Rather than simply building a complex computer model, BEACON investigates which specific improvements make predictions better, with the goal of developing a reliable, understandable tool that can be used in everyday hospital care.

When someone comes to hospital with an infection, one of the most important questions is whether bacteria have spread into their blood. This is called “bacteraemia”. Bacteraemia can make people much more unwell and can increase the risk of dying, so finding it early can change treatment and save lives. The usual test is a “blood culture”, where a blood sample is checked for bacteria. The problem is that blood cultures take a long time to give a result, anywhere from 24 hours to 5 days. Many of the most important decisions about a patient’s care are made in the first few hours, long before the result is back: whether to start antibiotics, whether the patient needs to stay in hospital or can safely go home, whether to seek advice from infection specialists, and whether further tests are needed. 

Not everyone with an infection has the same chance of having bacteraemia. The risk can change depending on how severe the infection is, the person’s age and other health problems, and where the infection is in the body. We can combine all this information, which is often available within a few hours, into a risk prediction model that estimates a person’s chance of having bacteraemia. This could help doctors identify patients who are likely to have bacteraemia early, before blood culture results are available, and use this to guide decisions about antibiotics, admission or discharge, specialist advice and further testing, including who most needs blood cultures taken. 

The aim of this project is to use existing hospital data at University College London Hospitals (UCLH) to build a prediction model. This model would estimate a person’s risk of bacteraemia early in their hospital visit. The data source is large: during a recent UCLH evaluation, more than 40,000 patients had blood cultures taken within 24 hours of emergency hospital attendances over a 5‑year period.