Mild low back pain (LBP) in older adults, despite its high prevalence and the need for early intervention, remains difficult to detect using traditional gait assessment methods. This study aimed to identify IMU-derived features sensitive to mild LBP and to develop classification models. Gait data were captured using shank-mounted inertial measurement units (IMUs), with pain levels defined by the Oswestry Disability Index (ODI). Results showed that while total ODI scores exhibited a substantial effect size (Hedges’ g = 2.40), item-level analysis revealed that self-reported walking function remained largely unaffected. Consistent with this preserved subjective perception, traditional IMU features demonstrated limited discriminative power. In contrast, features characterizing stride-to-stride irregularity, nonlinear temporal organization, and bilateral asymmetry captured subtle yet significant pain-related alterations. Based on this optimized feature subset, an artificial neural network model achieved an accuracy of 0.96, precision of 0.91, and specificity of 0.93 through leave-one-out cross-validation. These findings suggest that gait alterations associated with mild LBP are primarily characterized by changes in subtle motor organization rather than macroscopic performance. This study supports the clinical feasibility of low-burden, wearable IMU-based assessments for the early detection and monitoring of functional decline in older adults.
Kim et al. (Mon,) studied this question.