A Preliminary Approach to Fall Risk Assessment in Aged Care Facilities Using Generative AI Technologies

Falls among elderly residents in residential aged care facilities (RACFs) are a critical public health concern, often resulting in severe injuries, reduced mobility, diminished quality of life, and increased healthcare costs. Traditional machine learning methods struggle to accurately identify fall risk factors within unstructured electronic health record (EHR) data due to the lack of labelled datasets. This study develops methods for extracting fall risk factors from unstructured EHRs in RACFs, leveraging large language models (LLMs) and the retrieval-augmented generation (RAG) technique. Utilising an established knowledge base from RAG documents, LLMs with tailored prompts were employed to extract fall risk factors from the collected 1,854 nursing notes in the EHR. Expert feedback was used to evaluate and refine the extracted factors. The paired t-test suggested that the RAG method significantly improved risk factor extraction, achieving a precision of 97.3%, a recall of 89.9%, and an F1 score of 93.3%. Integrating LLMs with literature-based RAG effectively extracts fall risk factors from unstructured EHRs in RACFs, addressing challenges associated with unlabelled datasets. This approach enhances the identification of risk factors, informing targeted fall prevention strategies.