AI tool identifies 80% of emergency admissions

Published: 15-Jun-2022

Clinical trial supporting patients with new AI population health tool shows one in four admissions prevented and mortality reduced


An AI tool which was recently tested on NHS Greater Glasgow & Clyde’s patient data has proved it can identify 83% of people who would otherwise need unplanned hospital care.

The tool, called HN Predict, is able to identify patients early enough to prevent many of their visits to A&E and hospitalisations.

In a clinical trial, recently published in the British Journal of General Practice (BJGP), one out of four referrals to hospital identified by the AI tool could have been prevented by supporting these high-risk patients with nurse-led, virtual ward support.

And patients supported through this novel predictive and preventive clinical pathway reported increased quality of life, improved physical health, and an increase in their ability to manage their health conditions themselves.

The same trial also shows an increased survival rate of more than 50%, especially in deprived elderly males over 65 years, which lasted throughout the 24-month trial.

This is one of the first controlled and randomised studies demonstrating a statistically-significant impact on preventing care needs and increasing survival using an AI tool and virtual ward model in the NHS.

It enables us to improve clinical outcomes, the patient experience, and saves the NHS valuable hospital resources all at once

Using predictive models to risk-stratify key population segments is a critical approach to delivering the anticipatory care that is at the heart of the NHS Long Term Plan.

If applied to the NHS as a whole, the technology and related clinical nurse-led services, could prevent 5-7% of all unplanned hospital care and save £2,200 per patient.

It would require the technology to be made available to the 1-5% of patients with the highest risk of clinical crisis and unplanned care.

The technology and its inventor, Dr Joachim Werr; and chief executive, Mark England, have gained national NHS support and endorsement from the NHS Innovation Accelerator and the NHS-led Small Business Research Initiative R&D grant programme.

And the publication in BJGP is the first of several scientific papers now to be published from the large clinical trial which started in 2015.

The trial recruited 1,800 participants at seven acute NHS trusts and followed these individuals for up to five years.

The HN Predict tool is currently being implemented in several integrated care systems in Staffordshire, North Yorkshire, Scotland, and the Republic of Ireland.

Dr Werr said: “HN Predict works by applying machine learning to routinely-collected healthcare data, turning existing information into a powerful source for preventive healthcare.

“The fact that 1% of the population consumes over 50% of NHS hospital bed capacity, and that many of these patients are predictable while still amenable to prevention, represents a fantastic opportunity both for patients and NHS staff.

By focusing on individual people, rather than activity, and applying established machine-learning techniques, it is astounding that we can name the few thousand people who will use unplanned care

“It enables us to improve clinical outcomes, the patient experience, and saves the NHS valuable hospital resources all at once.”

Dr Paddy Hannigan, chairman of the Stafford and Surround CCG and clinical lead for Staffordshire & Stoke-on-Trent ICS Digital Programme, added: “Data can play a huge role for the NHS if it is collected, analysed, and acted upon in the right way.

“The key to our work with HN was the data-driven case finding.

“By using data to better forecast demand and predict outcomes we were able to manage the resources we had accordingly and reduce hospital care costs by 30% per patient.”

And Mark England, chief executive of HN said: “In my career non-elective care has always been managed as random, inexplicable in its cause.

“By focusing on individual people, rather than activity, and applying established machine-learning techniques, it is astounding that we can name the few thousand people who will use unplanned care.

“An emergency admission is a frightening and horrible experience for patients and we need to work systematically to avoid this potential harm.”

You may also like