We aimed to develop regression models for estimating oxygen consumption (VO2, ml·kg-1·min-1) during treadmill walking based on accelerations of the upper and lower limbs and walking velocity, quantitatively assess the contribution of each sensor location, and validate the accuracy and practicality of a simplified model. Eighteen healthy adults with regular exercise habits (nine men, nine women) participated in treadmill walking trials at varying speeds (3–6 km·h-1; up to 5.5 km·h-1 for women). Vector magnitude (VM) from triaxial accelerometers attached to both wrists and both ankles was recorded simultaneously with VO2 measurements from a portable breath-by-breath gas analyzer. Multiple regression models were constructed using FootVM (ankle VM), HandVM (wrist VM), and walking velocity as predictors. FootVM alone showed a moderate correlation with VO2 (R2 = 0.464), but adding walking velocity substantially improved the model’s accuracy (Model 2: R2 = 0.810, standard error of estimate = 1.25 ml·kg-1·min-1). Incorporating HandVM yielded only a minimal, non-significant model fit improvement (R2 = 0.815, ΔAIC = +18.4, βstd = −0.06), with no meaningful statistical contribution. Bland–Altman analysis indicated 95% limits of agreement for estimation error within ±2.46 ml·kg-1·min-1, corresponding to < 1 MET (3.5 ml·kg-1·min-1). These findings support the rational selection of a simplified model using only FootVM and walking velocity, which achieved a balance between high accuracy and practicality. The ability to estimate VO2 precisely using only ankle-mounted accelerometers highlights its potential for use in clinical and home-based physical activity assessment.
Nemoto et al. (Thu,) studied this question.