Gloucester Hospitals NHS Foundation Trust

Long-stay patients (21+ days) represented only 4% of admissions but consumed 34% of all bed days at Gloucester Hospitals NHS Foundation Trust. Prolonged stays cause patient harm including increased mortality, post-discharge complications, and muscle loss — yet most long stays are clinically avoidable with early intervention such as physiotherapy or ward reallocation.

The Accelerated Capability Environment (ACE), commissioned by the NHS AI Lab, built a machine learning model trained on 460,000 anonymised patient records to predict long-stay risk at the point of initial data collection. The model generates a risk score that reception and clinical staff can act on immediately — enabling different ward assignments or early interventions before decline sets in. The tool was subsequently integrated into the Trust's electronic health record system.

The tool detected 66% of long-stay patients in the highest-risk categories, enabling early clinical intervention and improving health outcomes for those patients. Integration into the EHR allowed retrospective validation on new datasets, with accuracy confirmed to remain high across unseen patient populations. The Trust also recognised significant economic benefit, given the high cost of prolonged bed occupancy.

• Early-admission risk scoring can flag avoidable long stays before clinical deterioration occurs, giving clinicians a meaningful intervention window.
• Training on a large anonymised dataset (460,000 records) was key to model accuracy across diverse patient presentations.
• EHR integration — not just a standalone pilot — is necessary for sustained, operational impact at scale.