Video-Based Fall Risk Assessment Using Multimodal Large Language Models in Home Health Care: A Proof-of-Concept...

Area of Science:

• Artificial Intelligence
• Gerontology
• Health Informatics

Background:

• Falls are a major cause of injury and death in home health care (HHC).
• Traditional fall risk assessments lack a comprehensive view of contributing factors.
• There is a need for advanced methods to improve fall risk evaluation in HHC settings.

Purpose of the Study:

• To introduce and evaluate a novel approach for fall risk assessment using multimodal large language models (MLLMs).
• To leverage video data and structured prompts for a more dynamic and comprehensive fall risk evaluation.
• To explore the feasibility of video-based fall prevention strategies in HHC.

Main Methods:

• Utilized the LLaVA-NeXTVideo-7B-hf multimodal large language model (MLLM).
• Analyzed simulated in-home patient video data, processed into 24 frames.
• Developed and tested standardized prompts based on 12 visually observable fall risk factors.

Main Results:

• The MLLM achieved 85.71% accuracy with concise prompts for simple risk factors.
• 100% accuracy was reached with elaborated prompts for complex risk factors.
• The model encountered limitations with factors requiring clinical judgment or limited visual data.

Conclusions:

• MLLMs show significant potential to augment fall risk assessment in HHC when guided by effective prompts.
• This video-based approach can complement, not replace, traditional clinical evaluations.
• This study establishes a foundation for future research in video-based fall risk analysis in HHC.