Imperial College London

Academic Institute

Murray A J Hudson

Unknown

No

NIBDAPC_2025_0052

Hospital falls are the most common type of patient safety incident in the NHS, affecting over 247,000 patients each year in England alone. When someone falls in hospital, they face serious risks including broken bones, head injuries, longer hospital stays, and loss of independence. These falls cost the NHS £630 million annually and can have devastating effects on patients and their families. While most hospital falls are preventable with simple interventions like regular check-ins, helping patients move safely, and modifying their environment, identifying which patients are most at risk remains challenging. Current assessment tools, such as the Purpose T or Waterlow, rely on manual checklists that often miss important risk factors and can be inaccurate. These tools might overlook subtle changes in a patient's condition during their hospital stay that could indicate increased fall risk. This project aims to create a more accurate, data-driven system for predicting which adults (≥18 years old) who are admitted to hospital are at risk of experiencing a fall during the admission. We will use the iCARE database, which contains detailed information about patients' medical conditions, medications, test results, and hospital care. By applying machine learning to analyse this data, we will develop a scoring system that categorises patients by risk level (low, moderate, high) for falls. As a study that looks back at previous patients this risk will not be communicated to the affected patients. To help inform this analysis, we shall seek to understand the accuracy of how falls are recorded in the iCARE database through matching electronic healthcare records with pseudonymised data from mandatory reports clinical staff must complete after a patient has an in-hospital fall. This tool could help clinical teams quickly and accurately identify high-risk patients, allowing for early, targeted preventive measures such as enhanced monitoring, mobility assistance, and environmental modifications. The goal is to reduce the number of patients who fall in hospital, prevent injuries and complications, improve patient comfort and outcomes, and decrease treatment costs for the NHS.

This research proposal addresses a critical gap in patient safety by developing a novel prediction tool using machine learning for in-hospital falls prevention. This project has been designed in alignment with priorities established by leading organisations such as the James Lind Alliance, National Institue for Health and Care Excellence (NICE), World Health Organisation (WHO), and the NHS Patient Safety Strategy. The identification and prioritisation of this research topic has been developed in line with the James Lind Alliance Multi-morbidity in Later Life Priority Setting Partnership, which involved 354 participants including patients, carers, and healthcare professionals. This partnership specifically identified falls prevention as a top research priority, with participants emphasising the need for better prediction and prevention approaches to reduce the devastating impact of falls on patients and families. The significant clinical need for this research is underscored by the enormous scale and impact of in-hospital falls. In England alone, over 247,000 inpatient falls occur annually, making them the most frequently reported patient safety incident in NHS healthcare. These falls cost the NHS £630 million annually and result in over 4 million additional bed days. The human impact is equally devastating: falls contribute to 130 annual deaths, over 2,000 hip fractures requiring surgical intervention, and significant long-term disability affecting patient independence and quality of life. Current risk assessment tools, such as the Purpose T or Waterlow, have fundamental limitations that compromise patient safety. NICE Guidelines explicitly recommend against using numeric risk scoring tools due to insufficient evidence of clinical utility. Systematic reviews reveal these tools achieve inadequate discriminative ability, with most tools showing area under the cure values below 0.7 and poor positive predictive values of only 2-62%. This means most patients identified as "high-risk" will not actually fall, whilst many who do fall are missed entirely. The complications of in-hospital falls are profound and costly. More than 90% of hip fractures result from falls, with one-fourth of elderly hip fracture patients dying within six months. Hip fracture survivors experience 10-15% decrease in life expectancy and significant decline in quality of life. Fear of falling affects patient willingness to engage in rehabilitation, creating cycles of deconditioning that increase future fall risk. If the algorithm is successful in determining which individuals are at risk of falls, successful deployment of this algorithm and appropriate management may help reduce the likelihood of an individual having a fall. This would therefore have profound effects on quality of life, as well as potentially quantity of life. As improving the accuracy of prediction compared to current clinical assessment is an important metric, this project will also incorporate an investigation on the accuracy of coded information that will inform the algorithm. This shall be undertaken oby matching cases to those reported through Datix and diagnostic coding. Pseudonymised data will then be imported into the iCARE environment to enable assessment of the accuracy of the coded data that will be used to inform the model. Importantly this should not overlap with the currently ongoing insightFall study but hopefully in time be complementary in improving patient stratification and outcomes.

Public Health Research

03/12/2025

Not applicable

Not applicable

One-off

TRE