AIMS
This project aims to develop and test an artificial intelligence (AI) tool to help doctors choose antibiotics for people with serious infections. At the start of treatment, doctors often have to prescribe before laboratory results are available. The tool will use information already held in electronic health records, such as a patient’s health history and previous test results, to estimate which antibiotics are most likely to work. Researchers will compare prescribing patterns with patient outcomes to assess whether the approach could support safer, more targeted treatment and reduce unnecessary use of broad-spectrum antibiotics. Broad-spectrum antibiotics act against many different bacteria but using them when they are not needed can contribute to antibiotic resistance.
BACKGROUND
When someone has a serious infection, doctors need to start treatment quickly. However, different infections can look very similar at first, and antibiotic resistance can mean that the first antibiotic chosen does not work. The project team estimates that this happens in around one in five cases. During the first hour of treatment for suspected sepsis, doctors therefore have to make important decisions with limited information.
Laboratory tests that identify the bacteria and show which antibiotics are likely to work can take two to three days. While waiting, doctors may prescribe broad-spectrum antibiotics to make sure treatment covers several possible causes. These medicines can be lifesaving but continuing them when a narrower antibiotic would work may increase side effects and contribute to antibiotic resistance. Better use of existing patient information could help doctors make a more informed choice from the first dose and review treatment as new results become available.
RESEARCH
The researchers will study antibiotics given to patients who need urgent treatment for serious infections. They will examine how often the chosen antibiotic does not work against the bacteria causing the infection, and how often a broad-spectrum antibiotic is used when a more targeted option may have worked equally well.
Using electronic health record data, the team will develop and test an AI model designed to support clinical judgement, not replace it. The researchers aim to reduce use of broad-spectrum antibiotics by around 40% without increasing the risk that a patient receives an antibiotic that is unlikely to work. They will assess the model’s recommendations against existing clinical decisions and patient outcomes. This early study will show whether the approach is promising enough for further research; it will not introduce the tool into routine patient care.
PATIENT INVOLVEMENT
Patients and members of the public with experience of AI-related health research have contributed to the development of the project. Their involvement has helped the team consider how the research should be explained, what questions may matter to patients and families, and how the proposed use of health data and AI may be viewed by the public. The team will continue to involve public contributors as the research develops.
APPROVAL
This project was supported unanimously by the PIONEER Data Trust Committee.
This work is led by Dr Christopher Green, Infectious Diseases Consultant / Associate Clinical Professor, University Hospitals Birmingham/University of Birmingham.
