Related Experiment Video
Updated: Jan 17, 2026
07:24
Home-Based Monitor for Gait and Activity Analysis
Published on: August 8, 2019
7.2K
Machine Learning Approach for Frailty Detection in Long-Term Care Using Accelerometer-Measured Gait and Daily
Xiaoping Zheng1, Ziwei Zeng1, Kimberley S van Schooten2,3
1Department of Sports Science and Physical Education, The Chinese University of Hong Kong, Hong Kong, China (Hong Kong).
JMIR Aging
|September 15, 2025
Summary
Machine learning effectively identifies frailty in long-term care using wearable sensor data. Dynamic gait analysis, including variability and asymmetry, offers sensitive frailty indicators for improved detection and management.
Area of Science:
- Gerontology
- Biomedical Engineering
- Artificial Intelligence in Healthcare
Background:
- Frailty is prevalent in over 50% of long-term care (LTC) residents, necessitating early detection for potential reversibility.
- Machine learning (ML) shows promise for frailty detection in community settings, but its application in LTC requires further investigation.
- Dynamic gait characteristics may provide more sensitive frailty indicators than traditional measures like gait speed.
Purpose of the Study:
- To assess the efficacy of ML models in identifying frailty among LTC residents.
- To utilize gait and daily physical activity data from a single accelerometer for frailty detection.
- To explore the potential of dynamic gait outcomes as sensitive frailty markers in LTC.
Main Methods:
- A cross-sectional analysis of 51 LTC residents using baseline data from a randomized controlled trial.
- Frailty status assessed via the FRAIL-NH scale.
- Gait data collected during a 5-meter walk and daily physical activity monitored for approximately one week using a 3D accelerometer.
- 34 dynamic and spatial-temporal gait outcomes, 3 physical activity variables, and 6 demographic characteristics extracted.
- Five ML models trained using leave-one-out cross-validation; performance evaluated by accuracy and AUC.
- Explainable AI (XAI) techniques employed for outcome interpretability.
Main Results:
- The extreme gradient boosting model achieved the highest performance with 86.3% accuracy and an AUC of 0.92.
- XAI analysis indicated that frail individuals exhibited more variable, complex, and asymmetric gait patterns.
- Key gait indicators for frailty included higher stride length variability, increased sample entropy, and a higher gait symmetry score.
Conclusions:
- Dynamic gait outcomes, particularly variability and asymmetry, are more sensitive indicators of frailty in LTC settings than traditional spatial-temporal measures.
- ML models trained on accelerometer data show significant potential for accurate frailty detection in LTC.
- These findings can inform enhanced strategies for frailty detection and management in long-term care.
