Wearable sensors use gait parameters to predict fall risk in older adults

Wearable sensors enable precise gait measurements, which in turn could help predict the risk of future falls among older adults, according to a recent study.

“Older adults who have a high risk of falling have gait control deficits and these deficits can be measured by linear and nonlinear variability analysis of walking timeseries,” the researchers said. “The novel contribution of this investigation is identifying the importance of linear and nonlinear gait variables that are sensitive to gait impairments in older adults as a function of fall risk.”

A total of 171 community-dwelling older adults participated in the present analysis and wore an inertial measurement unit (IMU) sensor on their sternum while they performed a 10-minute walking test. Gait parameters, such as gait variability, complexity, and smoothness, were collected and their prospective associations with fall risk were evaluated.

Of the 171 participants, data from 127 were used to train the classification model, of whom 25 (19.7 percent) experienced at least two falls in a year. The model was then blind tested on the remaining 44 participants.

The base random forest model used 58 gait parameters as input variables and classified them as either linear or nonlinear. In the blind test, using the linear gait variables achieved an accuracy of 71.8±7.0 percent for predicting fall risk. The resulting sensitivity and specificity values were 53.3±11.5 percent and 76.6±11.65 percent, respectively. [Sci Rep 2021;11:20976]

In comparison, when using nonlinear gait variables, the base model had an accuracy of 61.4±3.2 percent, with sensitivity and specificity values of 86.7±4.7 percent and 54.9±4.8 percent, respectively.

Since either low accuracy or low sensitivity limited the predictive value of the base model, the researchers applied feature engineering steps to improve the random forest classifier. The resulting model for linear variables had a mean accuracy estimate of 61.4±3.2 percent and specificity of 54.9±4.8 percent, while its sensitivity improved to 86.7±4.7 percent.

In comparison, the nonlinear random forest model after feature engineering yielded a higher accuracy measure of 74.8±5.5 percent, with sensitivity and specificity estimates of 80.8±11.5 percent and 73.4±9.5 percent.

Seeking to further improve the model, the researchers gradually added linear variables to the nonlinear random forest model. They found that the best-performing model could predict fall risk with 81.6±0.7 percent accuracy, 86.7±0.5 percent sensitivity, and 80.3±0.2 percent specificity.

Among the gait parameters assessed, recurrence, complexity, determinism, recurrence, entropy, step time, swing time, smoothness of gait, and overall walking time series complexity emerged as particularly important features for fall prediction.

“Wearable technology allowed us to gather data where it matters the most to answer fall-related questions,” the researchers said, referring to community-based settings rather than in gait laboratories.

“This study opens new prospects of clinical testing using gait stability measures with a wearable sensor that may be relevant for assessing fall risks at home and senior living environments,” they added.