A comparison of analytical approaches to investigate associations for accelerometry-derived physical activity spectra with health and developmental outcomes in children

Abstract

The use of high-resolution physical activity intensity spectra obtained from accelerometry can improve knowledge of associations with health and development beyond the use of traditional summary measures of intensity. The aim of the present study was to compare three different approaches for determining associations for spectrum descriptors of physical activity (the intensity gradient, principal component analysis, and multivariate pattern analysis) with relevant outcomes in children. We used two datasets including physical activity spectrum data (ActiGraph GT3X+) and 1) a cardiometabolic health outcome in 841 schoolchildren and 2) a motor skill outcome in 1081 preschool children. We compared variance explained (R2) and associations with the outcomes for the intensity gradient (slope) across the physical activity spectra, a two-component principal component model describing the physical activity variables, and multivariate pattern analysis using the intensity spectra as the explanatory data matrices. Results were broadly similar for all analytical approaches. Multivariate pattern analysis explained the most variance in both datasets, likely resulting from use of more of the information available from the intensity spectra. Yet, volume and intensity dimensions of physical activity are not easily disentangled and their relative importance may be interpreted differently using different methodology.