The Accelerated Capability Environment (ACE) has supported the NHS AI Lab to develop a proof of concept (PoC) which could help identify patients at risk of unnecessary long hospital stays and the associated negative outcomes.
A new tool uses artificial intelligence to assess hospital data and help predict patients at risk of becoming long stayers.
Gloucestershire Hospitals NHS Foundation Trust is the lead acute provider for a population of 660,000. 4 per cent of all admissions to the hospitals stay for 21 days or longer.
Long stayers face an 11 per cent mortality rate (compared to 5 per cent of all admissions); 23 per cent chance of becoming unwell again after being deemed fit for discharge (compared to 1 per cent of all admissions); and for the over 80s, loss of 10 years’ muscle mass in just 10 days in hospital.
The commission asked two key questions, whether systems could be taught to identify people at risk of long staying, and whether patient data already collected had enough predictive power to highlight risk - the answer to both was yes.
ACE worked with Polygeist from its Vivace community to develop a long stay stratification tool, which could use an artificial intelligence (AI) model trained on 460,000 anonymised records to identify those at risk of becoming long stayers from initial patient data collection. They were able to produce an immediate long stayer risk score, which could then be available to all reception and clinical staff to help avoid known risk factors.
In this case, for example, if a doctor suspects a patient to be at a higher risk of becoming a long-stayer, they may opt to avoid catheterisation, admit to a different ward, or immediately refer the patient to a geriatrician or physiotherapist to avoid decline.
The tool detected 66 per cent of long stayers within the highest risk categories. It is estimated that a single-day reduction in average stay yields £1.7m in savings for Gloucester Hospitals alone.
Following successful completion, the PoC was moved into a limited closed Alpha service phase, and integrated with the trust’s electronic health record system via application programming interfaces (APIs), which allowed further testing against Covid-era datasets - anonymously checking patients who had already been discharged to see if the tool could have helped. Here it also remained highly accurate.
Image by Gerd Altmann from Pixabay
