Machine Learning and Deep Learning Models for Predicting Future Falls in Community-Dwelling Older Adults: Systematic...

Area of Science:

• Gerontology
• Artificial Intelligence
• Biostatistics

Background:

• Machine learning (ML) and deep learning (DL) show promise for fall risk prediction.
• Prior reviews often focused on in-hospital falls or real-time detection, leaving the performance of ML-DL models for community-dwelling older adults unclear.

Purpose of the Study:

• To review ML-DL studies predicting future falls in community-dwelling older adults.
• To meta-analyze the discrimination performance of these models where feasible.

Main Methods:

• Searched six databases for longitudinal studies developing or validating ML-DL models for fall prediction in adults aged ≥60.
• Excluded studies on real-time detection, simulated falls, or inpatient settings.
• Assessed risk of bias using PROBAST and meta-analyzed areas under the curve (AUCs) using random-effects models.

Main Results:

• Included 28 studies, with 18 focusing on general older adults; prediction horizons varied from 3 months to 7 years.
• ML was used in 82.1% of studies, DL in 17.9%, with text, sensor, image, and multimodal data as input.
• Pooled AUC was 0.79 (95% CI 0.69-0.87) with high heterogeneity; models had a high risk of bias and limited external validation.

Conclusions:

• ML-DL models show potential for identifying community-dwelling older adults at elevated future fall risk.
• Wide prediction intervals and high risk of bias suggest potential overestimation of real-world performance.
• Emphasizes the need for rigorous validation and context-specific implementation for proactive fall prevention.