Accelerometer non-wear time was estimated on the basis of the standard deviation and the value range of each accelerometer axis, calculated for consecutive blocks of 30 minutes. A block was classified as non-wear time if the standard deviation was less than 3.0 mg (1 mg = 0.00981 m·s−2) for at least two out of the three axes or if the value range, for at least two out of three axes, was less than 50 mg. Thresholds were based on lab experiments in which thirty GENEA accelerometers were left motionless on a flat, stable surface for 30 minutes, showing that the standard deviation of an acceleration signal (which has some inherent noise) is 2.6 mg during non-motion. Therefore, the 3.0 mg threshold will allow a maximal increase of 0.4 mg in the standard deviation, which when expressed in angular orientation of the acceleration sensor corresponds to a standard deviation of 1.6 degrees [ ]. Phan et al. showed that the acceleration of the chest in a resting person resulting from the breathing movement alone has an amplitude of 10 mg, while the vibrations resulting from the heart beat have a peak-to-peak amplitude of 80 mg [16] (link). Therefore, even the tiniest wrist movements are likely to be picked up by the method as described above.
Participants for whom more than 50% of the wrist data was classified as non-wear were excluded from further analyses (two pregnant women from the Swedish study). For the remainder of the participants, non-wear time segments were labelled as missing.
Next, a simple summary measure was derived from the raw acceleration signals, involving a filtering stage to extract the accelerations related to body movement using a fourth-order Butterworth band pass filter (ω0: 0.2–15 Hz), followed by the calculation of the vector magnitude ( ). The resulting signal was then averaged over intervals of one second. Three basic approaches for the imputation of movement during non-wear segments were evaluated: i) no imputation, with non-wear time regarded as no movement (Acc0); ii) imputation of non-wear time by the average movement during wear time for that participant (Acc1); and iii) imputation of non-wear time using the available wear time data at similar times on other days for each participant (Acc2). Here, Acc2 is assumed to be best capable of dealing with large periods of missing data as it takes into account the 24-hour cycle of human behaviour. Finally, the average was calculated for each participant. All signal processing was done in R (http://cran.r-project.org ) using package Signal.
Participants for whom more than 50% of the wrist data was classified as non-wear were excluded from further analyses (two pregnant women from the Swedish study). For the remainder of the participants, non-wear time segments were labelled as missing.
Next, a simple summary measure was derived from the raw acceleration signals, involving a filtering stage to extract the accelerations related to body movement using a fourth-order Butterworth band pass filter (ω0: 0.2–15 Hz), followed by the calculation of the vector magnitude ( ). The resulting signal was then averaged over intervals of one second. Three basic approaches for the imputation of movement during non-wear segments were evaluated: i) no imputation, with non-wear time regarded as no movement (Acc0); ii) imputation of non-wear time by the average movement during wear time for that participant (Acc1); and iii) imputation of non-wear time using the available wear time data at similar times on other days for each participant (Acc2). Here, Acc2 is assumed to be best capable of dealing with large periods of missing data as it takes into account the 24-hour cycle of human behaviour. Finally, the average was calculated for each participant. All signal processing was done in R (
Full text: Click here