People who stay in hospital for 21 days or longer are defined as long stayers; this not only means hospital beds are occupied longer but it also leads to negative outcomes for the patient including increased mortality rate, becoming unwell again after discharge, and loss of muscle mass particularly in older populations. However, most long stays are avoidable where there is no medical reason for bed rest, and where interventions such as walking or physiotherapy are known to reduce overall stay length. This project used artificial intelligence to identify patients at risk of becoming long stayers so that clinicians could adjust treatment plans to prevent it happening.

What happened?

One type of artificial intelligence is machine learning, where a model is fed data and trained to recognise patterns so that it can make predictions about new data it is given.

The Accelerated Capability Environment (ACE) developed a tool which trained an artificial intelligence model to identify those at risk of becoming long stayer based on their initial patient data collection. The model saw patterns or links between the initial data and long stay outcomes, and could calculate a risk score that could be used during that initial data collection to predict whether someone would become a long stay patient.

This risk score could then be used by reception and clinical staff to make different decisions based on known risk factors. For example, a doctor could choose a different intervention or admit to a different ward to avoid decline if they know a patient is at risk of becoming a long stayer.

What were the benefits?

At Gloucester Hospitals NHS Foundation Trust, where ACE ran a pilot, 4% of patients had been long stayers, comprising 34% of all bed stays. The tool was able to detect 66% of long stay patients in the highest risk categories, which not only improved health outcomes of the identified patients, but also offered huge economic benefits for the Trust as bed stays cost a lot of money for them.

The Trust decided to integrate the tool into its electronic health record system. Once installed, the tool was able to anonymously test new datasets and patients to see if it could have helped in those past situations; it was found to remain highly accurate at predicting long stays in those situations too.

What type of data was involved?

The AI model was trained using 460,000 anonymised patient records at Gloucester Hospitals NHS Foundation Trust.

Who funded and collaborated on this work?

The Accelerated Capability Environment is a unit within the Home Office who tackle public safety and security challenged arising from evolving digital and data technology. ACE is delivered by Vivace, a community of multi-disciplinary organizations and experts from industry and academia focused on innovation.

They were commissioned by the NHS AI Lab, a programme which brings together government, health and care providers, academics and technology companies to address the barriers to developing and deploying AI systems in healthcare.

Where can I go for more information?

Using AI to help identify long stay patients and improve outcomes

About ACE

About Vivace