Can a single lower trunk body-fixed sensor differentiate between level-walking and stair descent and ascent in older adults? Preliminary findings

Stair ascent and descent are common forms of ambulation that may be challenging to detect. Here, we propose the first step towards differentiating between stair negotiation and level-walking using a single body-fixed sensor. Seventeen healthy older adults (age: 79.3 ± 4.2 years, 47% women) wore a body-fixed sensor on the lower-back while performing level-walking and stair negotiation. Measures derived from the 3D acceleration and angular-velocity signals included medians, ranges, step duration, step and stride regularity, filtered vertical to horizontal acceleration ratio (VAF/HAF), and wavelet-based features. Friedman's and Wilcoxon tests compared between conditions. Stepwise-binary logistic-regression evaluated classification accuracy. During level-walking, yaw range was lowest and anterior–posterior and vertical step and stride regularity were highest (p ≤ 0.007). Anterior–posterior step regularity (p = 0.003), VAF/HAF (p = 0.094), and yaw range (p = 0.105) identified level-walking (92.2% accuracy). During stair ascent, roll range, median anterior–posterior acceleration and anterior–posterior wavelet-coefficient were lowest (p ≤ 0.006), while VAF/HAF was highest (p = 0.0029). Anterior posterior wavelet coefficient (p = 0.038) and VAF/HAF (p = 0.018) identified stair ascent (94.3% accuracy). During stair descent, vertical and medio-lateral ranges were highest and medio-lateral stride regularity and VAF/HAF were lowest (p ≤ 0.006). VAF/HAF (p = 0.01), medio-lateral acceleration range (p = 0.069), and medio-lateral stride regularity (p = 0.072) identified stair descent (90.2% accuracy). These findings suggest that a single worn body-fixed sensor can be used to differentiate between level-walking and stair negotiation.