2.3.1. Calibration Study

The calibration study was performed in the biomechanical–physiological lab at the Center for Health and Performance, Gothenburg, Sweden. The participants were asked to refrain from food intake and strenuous PA 3–4 h prior to measurements. Body weight and height were measured and one Axivity AX3 triaxial accelerometer (Axivity Ltd., Newcastle upon Tyne, UK) was attached over the right hip in an elastic belt around the waist and another at the middle of the front side of the right thigh using medical tape. The accelerometers were set to record data at a sample rate of 100 Hz and an acceleration range of ±8 g and the raw triaxial data was extracted with the OmGUI software (Axivity Ltd., Newcastle upon Tyne, UK). Oxygen consumption was measured using the stationary metabolic system Oxycon Pro (Jaeger, BD Corporation, Franklin Lakes, NJ, USA) where the participants breath through a face-mask with a turbine flow meter and expired air-sampler. Oxycon Pro has shown high accuracy when evaluated with the Douglas bag method [22]. The activities included consisted of seated rest in an armchair for 20 min to determine resting energy expenditure (REE) followed by 4 min standing, walking at 3, 4, 5 and 6 km·h−<sup>1</sup> and running at 8 and 10 km·h−<sup>1</sup> on a treadmill during 4 min at each speed without interruption, in order to reach steady-state [23]. VO2 data was collected during all 20 min and respective 4 min activities.

The triaxial acceleration data were processed using our new algorithm to produce a measure of PA intensity expressed in mg, including a band-pass filter with the cut-point range of 0.29–10 Hz that was shown to include all relevant information to assess activity intensity and to minimize inclusion of noise [17,18]. However, acceleration data was down-sampled to 30 Hz and truncated to ±6 g before processing to match the corresponding ranges of the ActiGraph GT3X+ accelerometer used in the free-living measurements. VO2 data from the 20 min of resting was filtered with a moving average filter with a window size of 2 min and the minimum value was considered the individual REE. One minute of data (accelerometer, VO2) captured between duration 2:45 and 3:45 at standing and at each treadmill speed was used for calibration. VO2net was calculated by subtracting VO2stand from VO2gross divided by body weight (mL·min−1·kg−1), while MET-values were calculated by the quotient of the VO2gross and VO2rest.

In each age-group, the relationships between Speedabs and accelerometer output (mg), Speedabs and VO2net and Speedabs and MET were investigated, as well as the relationship between Speedeq and VO2net and Speedeq and MET. The Speedeq was calculated as Speedeq <sup>=</sup> V2·g−1·h−<sup>1</sup> (V <sup>=</sup> Speedabs (m·s−1), g = gravity (9.81 m·s−2), h = body height (m)), and represent the Speedabs performed with similar kinematical effort in individuals of different body size [8,9,15]. Smoothing splines were fitted to the data from each age group, with the intercept forced to zero VO2net or 1 MET, and zero acceleration. Based on previous biomechanical and physiological research, we made the following assumptions:


MET cut-points at 1.5, 3.0, 6.0 and 9.0 were implemented to represent LPA, MPA, VPA and very vigorous PA (VVPA), respectively, which is in line with previous literature [12]. To calibrate accelerometer data to VO2net, the first step was to determine the VO2net corresponding to 3.0, 6.0 and 9.0 METs in adults by linear regression. The next step was to apply the same VO2net cut-points (and therefore the same effort) to all three age-groups to calibrate the corresponding accelerometer mg cut-points based on fitted smoothing splines. For comparison, accelerometer mg cut-points were also determined using the traditional 1.5, 3.0, 6.0 and 9.0 METs in all three age-groups.
