Developing and Applying Predictive Models to Identify Patients at Increased Risk of Falling in Domiciliary Care

OBJECTIVES: Falls among older adults are a significant concern, causing over 200,000 hospitalisations annually in the UK and costing the NHS an estimated £630 million each year. This study evaluates Cera Care's capability to develop and implement an AI-powered tool to forecast falls among older adults with high accuracy, thereby enhancing patient safety, improving health outcomes, and reducing healthcare costs. This initiative aims to provide a proactive approach to fall prevention using real-time data from in-person interviews, frequent daily visits, and medication history, enhanced by personalised questionnaires and carer assessments.

METHODS: Cera developed a Machine Learning (ML) based model to predict falls seven days in advance. The model uses features related to patient demographics, health conditions, medications, lifestyle along with individual components from Fall Risk Assessment Tool (FRAT). FRAT comprises 9 features: seven selected from a literature review and two new features identified through in-house experiments. Each of them was assigned a score based on predefined criteria. Features used in the ML model and FRAT were designed by extracting data points from sources like Cera digital care plans, daily visit reports, and electronic medication records (eMars).

RESULTS: The machine learning model achieved 83% accuracy and 61% recall in forecasting falls within the next seven days. This predictive technology identifies a user's likelihood of falling, allowing improved prioritisation of high-risk patients to receive a preventive intervention.

CONCLUSIONS: Cera's AI-driven approach significantly enhances the quality and safety of care for older adults by enabling timely interventions to predict and prevent falls. By raising awareness of AI benefits in healthcare, Cera aims to encourage broader adoption of similar technologies across national and private health and social care providers, promoting predictive, preventive, and patient-centred care through personalised, condition specific predict and prevent initiatives.