Aims
In this project, we plan to use to PIONEER data to:
- Investigate whether a new approach to data analysis can be used to support clinical decision-making in healthcare.
- Investigate how this new approach can be used to help improve treatment of patients with community acquired pneumonia (CAP).
Scientific rationale
Chest infections (pneumonia) are a very common reason for people to attend hospital, especially in the elderly population. The severity of the disease ranges from very mild (patients treated with oral antibiotics at home) to very severe (patients requiring admission to intensive care). In hospitalised patients, up to 15% die within 30-days.
For clinicians treating this condition in the hospital setting, care of these patients can be challenging. Key decisions include the type of antibiotics (oral or intravenous), the appropriate place of care (home, hospital or intensive care), and when it is appropriate to stop antibiotics.
Artificial intelligence is the development and use of computer systems to perform tasks that require human level intelligence, such as understanding images, recognising speech and decision-making. Machine learning is a branch of artificial intelligence where systems learn to perform such tasks without being told (programmed) exactly how.
A particular type of machine learning is reinforcement learning which looks at scenarios where multiple decisions need to be made and where decisions may affect other decisions. The aim is to train systems to make the best possible decisions. Decision-making in healthcare is often complex, and reinforcement learning is extremely attractive as a strategy to help clinicians make the best decisions to improve patient care. For various reasons, however, it is not currently known how well reinforcement learning methods can be used in healthcare
Duration
24 months.
Methodology
Using the anonymised PIONEER dataset, we will identify adults that attended the emergency department with a diagnosis of community-acquired pneumonia.
Using these data, we will attempt to see whether an algorithm can be developed that supports clinical decision-making in patients with community-acquired pneumonia. We will then test how well the algorithm works using the PIONEER data. Our algorithm will cover key clinical decisions including how antibiotics are prescribed and the patient’s place of care.
Public health impact
This project will provide important new information about how it might be possible to use reinforcement learning in healthcare by examining its potential use in a complex and common health condition. This project may help us improve treatment for patients with community-acquired pneumonia. More broadly, it may pave the way for researchers to look at how reinforcement learning might be used in other conditions to support clinical decision-making and ensure patients receive the best evidence-based care.
Approval
This project was supported unanimously by the PIONEER Data Trust Committee.
