HDR UK Gateway
HDR Gateway logo

Bookmarks

Develop a machine learning algorithm to predict pressure injuries using de-identified patient data from iCARE

Safe People

Organisation name

Imperial College London

Organisation sector

Academic Institute

Applicant name(s)

Mikael Sodergren

Funders/ Sponsors

Murray A J Hudson

DEA accredited researcher?

Unknown

Sub-licence arrangements (if any)?

No

Safe Projects

Project ID

NIBDAPC_2024_0037

Lay summary

Pressure ulcers, also known as bedsores, are a common but preventable condition that can affect people who are immobile, such as patients in hospitals. They occur when prolonged pressure on the skin cuts off blood flow, leading to painful sores that can become infected. Pressure ulcers impact around 700,000 people in the UK each year, including 20% of hospitalised patients. They can lead to serious health issues, extended hospital stays, and high treatment costs. For the NHS, managing pressure ulcers costs an estimated £1.4 to £2.1 billion each year. While most pressure ulcers are preventable with simple interventions, such as regularly repositioning patients or using specialised mattresses, identifying those most at risk remains challenging. Current assessment tools often rely on manual checklists, which can miss important risk factors. For example, these tools might overlook pre-existing health conditions that affect a patient's movement or sensation. They can also miss changes in a patient's health that occur during their hospital stay, such as low blood pressure or anaemia. This project aims to create a more accurate, data-driven system for predicting who is at high risk of developing pressure ulcers. We will use the iCARE database and by applying machine learning—a type of artificial intelligence—to analyse this data, we will develop a scoring system that categorises patients by risk level (low, moderate, high) for pressure ulcers. This tool could help healthcare teams quickly and accurately identify high-risk patients, allowing for early, targeted preventive measures. The goal is to reduce the number of patients who develop pressure ulcers, improve patient comfort and outcomes, and decrease treatment costs for the NHS. This approach could also pave the way for using data-driven tools to improve other areas of patient care.

Public benefit statement

This research proposal addresses a critical gap in pressure injury prevention by developing a novel prediction tool using machine learning. This project was designed is alignment with priorities established by leading organisations such as the James Lind Alliance, European Pressure Ulcer Advisory Panel, National Pressure Injury Advisory Panel and Pan Pacific Pressure Injury Alliance, and the American College of Physicians. Prevention has been identified as the highest care goal for patients who have experienced pressure ulcers. In this context, the James Lind Alliance Priority Setting Partnership has highlighted the following as a specific research priority: ‘Is using a pressure ulcer risk rating scale better than clinical judgement in preventing pressure ulcers and is there a best scale?’ This priority is based upon previous systematic reviews of the literature which have highlighted that data on the effectiveness of existing tools to help predict and prevent pressure injury is uncertain. The significant clinical need for this research is underscored by the high prevalence and associated costs of pressure injuries. In the UK alone, 700,000 individuals are affected annually, incurring a daily cost of £3.8 million to the NHS. Current risk assessment tools have limitations in predictive accuracy, highlighting the need for innovative solutions. The proposed project, informed by data from iCARE would hope to provide a more accurate tool, which not only can help prediction of pressure ulcers in a research setting, but lead to reduction in pressure ulcers through integration into clinical care. The complications of pressure injuries are costly. These include increased risk of mortality, with 4.2% of stays where pressure injury is the primary diagnosis and 11.6% of stays with secondary pressure injuries ending in death. Pressure injuries can also result in pain, local or systemic infection, psychological stress, and reduced quality of life. The effect on psychological health and quality of life extends to family and caregivers due increased dependence on their help. This results in prolonged hospital stays (5-8 days per pressure injury) and higher readmission rates, leading to increased healthcare costs. Treating pressure injury costs the NHS more than £3.8 million every day. In 2004 the estimated annual cost of pressure injury care in the UK was £1.4-2.1 billion per year, and the mean cost per patient of treatment for a grade IV pressure injury was calculated to be £10,551 a year. It is therefore likely that current costs to the NHS are higher. Pressure injuries are avoidable causes of harm and care to prevent new pressure injuries or progression typically involve inexpensive changes to delivery of care compared to the cost of the cost of treatment. Despite the development of many manual risk assessments, there is minimal high-quality evidence of their effectiveness, particularly in changing outcomes. The improvement in the availability and quality of artificial intelligence algorithms informed by data from iCARE in disease diagnosis and risk assessment may lead to a step-change in improvement for many conditions, including pressure injuries. Through developing improved measures to assess the risk of pressure injury, we therefore hope this can lead to earlier intervention, reducing its incidence. This would lead to individual benefits for affected patients, but also considering the substantial healthcare costs associated with pressure injuries should also be a key priority of the wider public.

Request category type

Public Health Research

Other approval committees

Project start date

16/01/2025

Latest approval date

16/12/2024

Safe Data

Dataset(s) name

ICHT iCARE Data Model

Data sensitivity level

De-Personalised

Common Law Duty of Confidentiality

Not applicable

National data opt-out applied?

Not applicable

Request frequency

One-off

Release/Access date

16/01/2025

Safe Setting

Access type

TRE

Safe Outputs

Link to research outputs