Differences between percentages of work time in each PA
summary metric category by sensor location (IMUs on the trunk
and each upper arm and the wGT3X-BT on the waist) were examined
using repeated measures one-way analyses of variance
(ANOVA), with device as a fixed effect. A separate analysis was
performed for each PA category, and post-hoc pairwise comparisons
were performed using a Bonferroni correction.
Agreement between the counts/min and METs obtained by the
IMUs and the wGT3X-BT was also assessed using BlandeAltman
analyses (Bland and Altman, 2007). Specifically, for each epoch, we
calculated the average of the IMU and wGT3X-BT measurements
and the difference between the IMU and wGT3X-BT measurements.
Estimates of the mean (fixed) bias of the IMU measurement in
comparison to the wGT3X-BT measurement and upper and lower
limits of agreement were obtained using calculations that considered
both repeated measures and a time-varying “true” value of the
measurement (Bland and Altman, 2007).
Due to the large number of 1-min epochs in the data set (36
participants 720 1-min epochs ¼ 25,920 total epochs), a
“summarized” Bland-Altman plot was also developed to qualitatively
assess inter-method agreement. Specifically, across all
epochs for each participant, we calculated the mean and standard
deviation of the average of the IMU and the wGT3X-BT measurements
and the mean and standard deviation of the differences
between the IMU and the wGT3X-BT measurements. We then
plotted the means of the differences against the means of the
averages and assessed the strength of the linear relationship between
the means of the averages and the means of the differences
using Pearson correlation coefficients. Linear regression lines were
fit to check for proportional bias and hyperbolic, 95% prediction
limits were plotted around the line of best fit