Capturing Free-living Physical Activity: Impacts Of Data Collection And Processing Decisions On Intensity Gradient

Device-based physical behavior measurement is important for surveillance and health research. Novel approaches to analyzing accelerometer data, like intensity gradient, capture the entire activity profile and may enhance comparability across studies and populations. However, it is unclear how data collection and processing decisions impact intensity gradient. PURPOSE: To assess if intensity gradient is affected by device type, location, and analytic decisions (epoch, metric, wear time algorithm). METHODS: Undergraduates (N=70) wore triaxial accelerometers for four waking days. All participants wore the same model at the hip and each wrist, and a subset (n=56) wore an accelerometer with a different dynamic range on one wrist. Mean Euclidean norm minus one (ENMO) and intensity gradient using ENMO were calculated for each wear location and device, three epoch lengths (1, 5, 60-s), and two wear time algorithms (counts, raw). Intensity gradient was also calculated for three other metrics (mean amplitude deviation, raw acceleration and count vector magnitude). Equivalence testing compared intensity gradient and mean ENMO (within 10% bounds) by location, device type, epoch length, wear time algorithm, and metric (intensity gradient only). RESULTS: Intensity gradient was equivalent (p<.05) across all wear locations, device types, wear time algorithms, and 1-s and 5-s epochs but not metrics (Table 1). ENMO was equivalent (p<.05) for device types, wear time algorithms, epochs, and wrists (Table 1). CONCLUSION: While intensity gradient and ENMO were invariant to device type, wear time algorithm, wrist location, and 1-s and 5-s epochs, only intensity gradient was equivalent between all wear locations, and only ENMO was equivalent across all epochs. Caution should be used when comparing across studies using different processing techniques. Future research should establish intensity gradient processing steps that improve comparability across epochs and metrics.