Estimation of physical activity intensity using triaxial ActiGraph accelerometers in youth populations: Impact of data type, attachment site, and modeling approach, including adaptations of the Sojourn method for varied use in youth

Abstract

<p>Adequate physical activity (PA) in youth is essential for short- and long-term health. However, it is difficult to perform large-scale assessments of youth PA levels and be assured of accurate results. Recent developments in accelerometer-based PA monitoring for adults (e.g. the Sojourn method) have enhanced the accuracy attainable by those devices, but the same has not been accomplished for youth. PURPOSE: The purpose of this study was to develop and cross-validate versions of the established Sojourn method that can be used for research with youth. METHODS: The study involved two phases. In Phase 1, existing ActiGraph monitor data from 54 youth were used to train artificial neural networks that were implanted into customized shells of the existing Sojourn method, and these were compared against indirect calorimetry for validation. In Phase 2, a separate cross validation analysis was conducted on an independent sample of 21 youth that wore Actigraph monitors during a simulated free-living protocol. This phase utilized direct observation as the criterion measure of activity intensity and consisted of one hour of self-directed activity, with limited direction provided outside of the requirement that at least 5 activities be performed within the hour. In both phases, standard processing methods were included to compare the relative utility of the new methods to the established and currently available techniques for both hip and wrist ActiGraph data. Phase 2 additionally compared non-dominant and dominant wrist attachment. Tests of classification accuracy (confusion matrices, sensitivity and specificity, percent accuracy, and kappa statistics) were used in both phases to evaluate the methods. RESULTS: The adapted Sojourn methods achieved accuracies ranging from 53.9% to 73.7% in Phase 1 (kappa scores from 0.24 to 0.44). In Phase 2, the adaptations fell between 38.2% and 60.5% accuracy (kappa scores from 0.06 to 0.41). The adapted Sojourn method using activity counts from hip worn Actigraph monitor showed the strongest overall performance. CONCLUSIONS: The adaptations of the Sojourn method were more accurate than currently available methods for youth, but improvements are still needed, particularly for methods using raw acceleration data from the wrist. Attachment to the dominant or non-dominant wrist is inconsequential.</p>