# Are We Missing the Environmental Factors in AI-Based Fall Risk Models?: A Systematic Review

## TL;DR

This paper reviews AI models for predicting falls in older adults and finds that environmental factors are often overlooked despite their importance in fall prevention.

## Contribution

The study systematically evaluates how environmental factors are integrated into AI-based fall risk models and highlights their underutilization.

## Key findings

- Only nine studies incorporated environmental factors into AI-based fall risk models.
- Environmental features improved model discrimination with AUC-ROC scores between 0.67 and 0.76.
- Environmental data was inconsistently represented across studies.

## Abstract

Falls commonly occur in home environments where environmental conditions can contribute to fall risk. Identification and mitigation of environmental hazards are critical components of fall prevention. However, artificial intelligence (AI)-based fall prediction models have largely focused on individual-level predictors, with limited attention to home environmental hazards despite their modifiable role in fall risk.

To systematically review how environmental factors are incorporated into existing AI-based fall risk prediction models and summarize reported AI approaches and model performance among community-dwelling older adults.

This systematic review followed PRISMA guidelines. Six electronic databases (PubMed, Embase, CINAHL, Cochrane Library, Web of Science, and Scopus) were searched from inception through December 2025. Eligible studies applied AI-based models to predict falls among older adults in community settings and incorporated environmental factors as model inputs.

Of more than 10,000 records identified, nine studies met final inclusion criteria. Six used supervised machine learning with structured data, while three employed computer vision or robotics-based approaches. Environmental factors were heterogeneously represented, ranging from checklist-based indicators to sensor- and vision-derived measures. When included, environmental features contributed meaningful information by improving discrimination or identifying actionable home hazards (AUC-ROC ranged from 0.67 to 0.76).

Environmental factors remain underemphasized in AI-based fall prediction models. Greater integration of standardized and context-aware environmental information may enhance the relevance and preventive utility of AI-based fall risk prediction in community settings.