Evaluation of Accelerometer-Based Walking-Turn Features for Fall-Risk Assessment in Older Adults

Abstract

Falls in older adult populations are a serious health concern, resulting in physical and psychological trauma in addition to increased pressure on healthcare systems. Faller classification and fall risk assessment in elderly populations can facilitate preventative care before a fall occurs. Few research studies in the fall risk assessment field have focused on wearable-sensor-based features obtained during walking-turns. Examining turn based features may improve fall-risk assessment techniques.

Seventy-six older individuals (74.15 ± 7.0 years), categorized as prospective fallers (28 participants) and non-fallers (43 participants), completed a six-minute walk test with accelerometers attached to their lower legs and pelvis. Turn and straight walking sections were segmented from the six-minute walk test, with a feature set extracted for each participant.

This work aimed to determine if significant differences between prospective faller (PF) and non-faller (NF) groups existed for turn or straight walking features. A mixed-design ANOVA with post-hoc analysis showed no significant differences between faller groups for straight-walking features, while five turn based features had significant differences (p <0.05). These five turn based features were minimum of anterior-posterior REOH for right shank, SD of SD anterior left shank acceleration, SD of mean anterior left shank acceleration, maximum of medial-lateral FQFFT for lower back, and SD of maximum anterior left shank acceleration. Turn based features merit further investigation for distinguishing PF and NF.

A novel prospective faller classification method was developed using accelerometer-based features from turns and straight walking. Cross validation was conducted for both turn and straight feature based models to assess classification performance. The best “classifier model – feature selector” combination used turn data, random forest classifier, and select-5-best feature selector (73.4% accuracy, 60.5% sensitivity, 82.0% specificity, 0.44 Matthew’s Correlation Coefficient (MCC)). Using only the most frequently occurring features, a feature subset achieved better classification results, with 77.3% accuracy, 66.1% sensitivity, 84.7% specificity, and 0.52 MCC score (minimum of anterior-posterior ratio of even/odd harmonics for right shank, standard deviation (SD) of anterior left shank acceleration SD, SD of mean anterior left shank acceleration, maximum of medial-lateral first quartile of Fourier transform (FQFFT) for lower back, maximum of anterior-posterior FQFFT for lower back). All classification performance metrics improved when turn data was used for faller classification, compared to straight walking data.