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

Area of Science:

• Gerontology
• Health Informatics
• Artificial Intelligence in Healthcare

Background:

• Falls in residential aged care facilities (RACFs) are a significant public health issue.
• Unstructured electronic health record (EHR) data presents challenges for identifying fall risk factors using traditional methods.
• Lack of labelled datasets hinders accurate fall risk assessment in RACFs.

Purpose of the Study:

• To develop and evaluate methods for extracting fall risk factors from unstructured EHRs in RACFs.
• To leverage large language models (LLMs) and retrieval-augmented generation (RAG) for improved risk factor identification.
• To address the limitations of unlabelled datasets in fall risk assessment.

Main Methods:

• Utilized LLMs with RAG and a knowledge base to extract fall risk factors from 1,854 nursing notes.
• Employed tailored prompts for LLMs to process unstructured EHR data.
• Incorporated expert feedback for refinement and evaluation of extracted factors.

Main Results:

• The RAG method significantly improved fall risk factor extraction.
• Achieved high performance metrics: 97.3% precision, 89.9% recall, and 93.3% F1 score.
• Demonstrated the effectiveness of integrating LLMs with RAG for EHR data analysis.

Conclusions:

• LLMs combined with literature-based RAG offer an effective solution for extracting fall risk factors from unstructured EHRs in RACFs.
• This approach enhances the identification of risk factors, supporting targeted fall prevention.
• Overcomes challenges posed by unlabelled datasets in healthcare data analysis.