Interpretation of Multivariate Association Patterns between Multicollinear Physical Activity Accelerometry Data and Cardiometabolic Health in Children-A Tutorial.

Geir Kåre Resaland,Eivind Aadland,Olav Martin Kvalheim,Lars Bo Andersen

doi:10.3390/metabo9070129

Abstract

Associations between multicollinear accelerometry-derived physical activity (PA) data and cardiometabolic health in children needs to be analyzed using an approach that can handle collinearity among the explanatory variables. The aim of this paper is to provide readers a tutorial overview of interpretation of multivariate pattern analysis models using PA accelerometry data that reveals the associations to cardiometabolic health. A total of 841 children (age 10.2 ± 0.3 years) provided valid data on accelerometry (ActiGraph GT3X+) and six indices of cardiometabolic health that were used to create a composite score. We used a high-resolution PA description including 23 intensity variables covering the intensity spectrum (from 0–99 to ≥10000 counts per minute), and multivariate pattern analysis to analyze data. We report different statistical measures of the multivariate associations between PA and cardiometabolic health and use decentile groups of PA as a basis for discussing the meaning and impact of multicollinearity. We show that for high-resolution accelerometry data; considering all explanatory variables is crucial to obtain a correct interpretation of associations to cardiometabolic health; which is otherwise strongly confounded by multicollinearity in the dataset. Thus; multivariate pattern analysis challenges the traditional interpretation of findings from linear regression models assuming independent explanatory variables

Highlights

There is irrefutable evidence about the favorable influence of physical activity (PA) on cardiometabolic health [1]
We present four different alternatives: (1) Instead of dividing by the residual variance, we divided by the total variance, giving a measure of explained variance of each single explanatory variable, in a multivariate space, related to the target-projected component
Most studies merge all intensities above MPA as MVPA, which gives the same weight to brisk walking and fast running

Summary

Introduction

There is irrefutable evidence about the favorable influence of physical activity (PA) on cardiometabolic health [1]. This evidence is limited by sub-optimal handling of accelerometry data. Such data is commonly reduced to a spectrum of time spent in only few gross intensities (sedentary (SED), light PA (LPA), moderate PA (MPA), vigorous PA (VPA), and/or moderate-to-vigorous PA (MVPA)). Most studies investigating associations between PA and cardiometabolic health have targeted selected parts of this spectrum [1], partly because they rely on statistical methods that cannot handle multicollinear data. Associations with metabolic health for the whole

Objectives

Methods

Discussion

Conclusion