From Assessment to Engagement: Inertial Gait Analysis and Remote Exercise for Fall-Risk Prevention
Abstract
Introduction
Ageing leads to subtle mobility changes that increase fall risk if not identified early. Most existing solutions intervene only after functional decline becomes evident. This work presents an integrated preventive approach combining clinically accurate gait analysis with remotely supervised exercise programmes, aiming to detect early deterioration and promote safer, more autonomous ageing.
Methods
The system includes: (1) a shoe-mounted inertial sensor (Moverics) enabling rapid gait assessment and extraction of validated fall- and frailty-related metrics; and (2) ViviFil, a mobile app delivering daily exercise adapted to the participant’s functional level, with remote monitoring.
Two clinical studies were conducted. First, gait data from 157 individuals (fallers and non-fallers) from the GSTRIDE database were analysed to evaluate the sensor’s discriminative capacity. Second, the GAIT2CARE dataset included 93 older adults (82±6 years) completing two 8-week multicomponent exercise programmes (on-site and app-guided via ViviFil).
Results
Using inertial gait parameters alone, the sensor achieved 78.2% accuracy classifying fallers versus non-fallers (sensitivity 77%, specificity 79%), outperforming conventional clinical tests (gait speed: 76.4%, TUG: 75.6%). Among participants with remarkable adherence (>70%, n=64), the app-guided programme showed functional improvements of 10.2% and gait improvements of 9.0%, with no significant differences compared to on-site training (p>0.05). However, adherence rates differed substantially (app-guided: 45% vs on-site: 96%), highlighting engagement as a critical factor.
Conclusions
This work demonstrates that combining precise inertial gait evaluation with remote structured exercise is a feasible and effective strategy for fall-risk prevention. The sensor’s discriminative performance, using gait parameters alone, matches conventional clinical assessments while offering rapid, objective measurement. The therapeutic equivalence of ViviFil-guided training is confirmed when adherence is maintained. Future work will explore deep learning approaches to further improve classification accuracy, alongside affective feedback strategies to strengthen sustained engagement.
Comments
Interesting!
Thank you for sharing your interesting research on fall prevention using technology. I am curious about the participants’ experiences of using the technology. Was this something you investigated in your studies?
Thank you for your interest…
Thank you for your interest in our research.
Yes, our studies, in the GAIT2CARE project, specifically examined participants’ experiences with the technology through satisfaction questionnaires administered at weeks 4, 8, and 12. The questionnaires assessed five aspects of the user experience using a 5-point Likert scale (1 = strongly disagree, 5 = strongly agree): (i) experiencing physical problems while using the program/app, (ii) perceived complexity and ease of use, (iii) the suitability of the exercises for different levels of mobility, (iv) overall satisfaction, and (v) perceived usefulness.
Overall, participants reported very positive experiences. Satisfaction scores were consistently high, with most participants scoring between 23 and 25 out of a maximum of 25.
Thank you for your response…
Thank you for your response!
What exciting results you have obtained. Thank you for sharing them. I will definitely read more about the GAIT2CARE project, as it is highly relevant to my own research and area of interest.
Wearable
Was there only one wearable used (i.e., on one foot) and what was the protocol for attaching the sensors i.e,, what foot? And, what were the wearable-based gait metrics?
Thank you for your comment…
Thank you for your comment. Two wearable sensors were used, one attached to each foot. The sensors were secured over the shoelaces using either a dedicated clip or an elastic band, depending on the participant’s footwear.
The wearable-based gait metrics obtained from the sensors were:
In addition to the absolute values of these parameters and statistics (mean, SD, median, CV), gait asymmetry was calculated for each metric by comparing the values obtained from the left and right foot sensors.