Published in:
Open Access 01-12-2023 | Research
Predictors of fall risk in older adults using the G-STRIDE inertial sensor: an observational multicenter case–control study
Published in: BMC Geriatrics | Issue 1/2023
Login to get accessAbstract
Background
There are a lot of tools to use for fall assessment, but there is not yet one that predicts the risk of falls in the elderly. This study aims to evaluate the use of the G-STRIDE prototype in the analysis of fall risk, defining the cut-off points to predict the risk of falling and developing a predictive model that allows discriminating between subjects with and without fall risks and those at risk of future falls.
Methods
An observational, multicenter case–control study was conducted with older people coming from two different public hospitals and three different nursing homes. We gathered clinical variables ( Short Physical Performance Battery (SPPB), Standardized Frailty Criteria, Speed 4 m walk, Falls Efficacy Scale-International (FES-I), Time-Up Go Test, and Global Deterioration Scale (GDS)) and measured gait kinematics using an inertial measure unit (IMU). We performed a logistic regression model using a training set of observations (70% of the participants) to predict the probability of falls.
Results
A total of 163 participants were included, 86 people with gait and balance disorders or falls and 77 without falls; 67,8% were females, with a mean age of 82,63 ± 6,01 years. G-STRIDE made it possible to measure gait parameters under normal living conditions. There are 46 cut-off values of conventional clinical parameters and those estimated with the G-STRIDE solution. A logistic regression mixed model, with four conventional and 2 kinematic variables allows us to identify people at risk of falls showing good predictive value with AUC of 77,6% (sensitivity 0,773 y specificity 0,780). In addition, we could predict the fallers in the test group (30% observations not in the model) with similar performance to conventional methods.
Conclusions
The G-STRIDE IMU device allows to predict the risk of falls using a mixed model with an accuracy of 0,776 with similar performance to conventional model. This approach allows better precision, low cost and less infrastructures for an early intervention and prevention of future falls.