The Accuracy of Smart Devices for Measuring Physical Activity in Daily Life: Validation Study

Laurent Degroote; Ilse De Bourdeaudhuij; Maïté Verloigne; Louise Poppe; Geert Crombez

doi:10.2196/10972

The Accuracy of Smart Devices for Measuring Physical Activity in Daily Life: Validation Study

JMIR Mhealth Uhealth. 2018 Dec 13;6(12):e10972. doi: 10.2196/10972.

Authors

Laurent Degroote^{1

2}, Ilse De Bourdeaudhuij¹, Maïté Verloigne¹, Louise Poppe^{1

2}, Geert Crombez²

Affiliations

¹ Physical Activity & Health, Department of Movement and Sports Sciences, Ghent University, Ghent, Belgium.
² Health Psychology, Department of Experimental Clinical and Health Psychology, Ghent University, Ghent, Belgium.

PMID: 30545810
PMCID: PMC6315254
DOI: 10.2196/10972

Abstract

Background: Wearables for monitoring physical activity (PA) are increasingly popular. These devices are not only used by consumers to monitor their own levels of PA but also by researchers to track the behavior of large samples. Consequently, it is important to explore how accurately PA can be tracked via these devices.

Objectives: The aim of this study was, therefore, to investigate convergent validity of 3 Android Wear smartwatches-Polar M600 (Polar Electro Oy, Kempele, Finland), Huawei Watch (Huawei Technologies Co, Ltd, Shenzhen, Guangdong, China), Asus Zenwatch3 (AsusTek Computer Inc, Taipei, Taiwan)-and Fitbit Charge with an ActiGraph accelerometer for measuring steps and moderate to vigorous physical activity (MVPA) on both a day level and 15-min level.

Methods: A free-living protocol was used in which 36 adults engaged in usual daily activities over 2 days while wearing 2 different wearables on the nondominant wrist and an ActiGraph GT3X+ accelerometer on the hip. Validity was evaluated on both levels by comparing each wearable with the ActiGraph GT3X+ accelerometer using correlations and Bland-Altman plots in IBM SPSS 24.0.

Results: On a day level, all devices showed strong correlations (Spearman r=.757-.892) and good agreement (interclass correlation coefficient, ICC=.695-.885) for measuring steps, whereas moderate correlations (Spearman r=.557-.577) and low agreement (ICC=.377-.660) for measuring MVPA. Bland-Altman revealed a systematic overestimation of the wearables for measuring steps but a variation between over- and undercounting of MVPA. On a 15-min level, all devices showed strong correlations (Spearman r=.752-.917) and good agreement (ICC=.792-.887) for measuring steps, whereas weak correlations (Spearman r=.116-.208) and low agreement (ICC=.461-.577) for measuring MVPA. Bland-Altman revealed a systematic overestimation of the wearables for steps but under- or overestimation for MVPA depending on the device.

Conclusions: In sum, all 4 consumer-level devices can be considered accurate step counters in free-living conditions. This study, however, provides evidence of systematic bias for all devices in measurement of MVPA. The results on a 15-min level also indicate that these devices are not sufficiently accurate to provide correct real-time feedback.

Keywords: accelerometry; fitness trackes; physical activity.

©Laurent Degroote, Ilse De Bourdeaudhuij, Maïté Verloigne, Louise Poppe, Geert Crombez. Originally published in JMIR Mhealth and Uhealth (http://mhealth.jmir.org), 13.12.2018.