Consumer-grade wearables showed largely consistent step counts with the research-grade ActiGraph, but demonstrated more limited accuracy for MVPA and PAEE, particularly at higher activity levels, underscoring that device selection must be carefully guided by the specific metric of interest.
Key Findings
Results
Step counts were largely consistent with ActiGraph for Apple Watch and Oura Ring, while Fitbit overestimated step count by 18.00%.
Apple Watch mean percentage difference from ActiGraph was 2.12%, Oura Ring was -6.24%, both within 10% of ActiGraph measurements.
Fitbit overestimated step count by 18.00% compared to ActiGraph.
Correlation coefficients for step counts were strong across all devices (r = 0.84–0.92).
Study enrolled 36 office workers; Apple Watch subgroup n = 21, Fitbit n = 22, Oura Ring n = 5, over a 3-week period.
Results
MVPA showed greater variability across devices, with Apple Watch underestimating by 46.22%, Oura Ring underestimating by 11.64%, and Fitbit showing minimal mean difference of 0.62%.
Apple Watch underestimated MVPA by 46.22% relative to ActiGraph.
Oura Ring underestimated MVPA by 11.64% relative to ActiGraph.
Fitbit showed a minimal mean difference of 0.62% for MVPA.
Bland-Altman analysis revealed proportional bias in Apple Watch's MVPA, with errors increasing at higher activity levels.
Correlation coefficients for MVPA were lower than for step counts across all devices.
Results
PAEE showed the largest discrepancies among all metrics, with Fitbit overestimating by 139.19%, Apple Watch overestimating by 25.91%, and Oura Ring underestimating by 16.87%.
Fitbit overestimated PAEE by 139.19% relative to ActiGraph.
Apple Watch overestimated PAEE by 25.91% relative to ActiGraph.
Oura Ring underestimated PAEE by 16.87% relative to ActiGraph.
Bland-Altman analysis revealed proportional bias in Fitbit's PAEE, with errors increasing at higher activity levels.
Correlation coefficients for PAEE were lower than for step counts across all devices.
Methods
ActiGraph data confirmed comparable physical activity levels across participant subgroups, providing a valid basis for inter-device comparisons.
Thirty-six office workers were enrolled and provided with an ActiGraph GT9X or equivalent research-grade device.
Data collection occurred over a 3-week free-living period in Japan.
Subgroups wearing different consumer devices showed comparable physical activity levels as measured by ActiGraph.
Metrics collected included step count, physical activity energy expenditure (PAEE), and moderate-to-vigorous physical activity (MVPA).
Results
Bland-Altman analysis revealed proportional bias in the Fitbit's PAEE and the Apple Watch's MVPA measurements, with errors increasing at higher activity levels.
Proportional bias indicates that measurement error was not uniform but systematically increased with greater physical activity.
This pattern was specifically identified for Fitbit PAEE and Apple Watch MVPA.
The presence of proportional bias limits the use of these devices for individuals with higher activity levels.
This finding was in addition to the overall mean percentage differences reported for each device-metric combination.
Results
Findings for the Oura Ring should be interpreted with caution due to a small sample size of only five participants.
Only n = 5 participants wore the Oura Ring during the study period.
The authors explicitly state that 'the findings for the Oura Ring should be interpreted with caution due to the small sample size.'
Despite the small sample, Oura Ring step count (-6.24%) and MVPA (-11.64%) differences from ActiGraph were relatively small compared to other devices.
Oura Ring PAEE underestimated ActiGraph by 16.87%.
Miwa T, Mii K, Chatani R, Sugitani Y. (2026). Comparison of consumer-grade wearable devices with a research-grade instrument for measuring physical activity in a free-living setting.. PloS one. https://doi.org/10.1371/journal.pone.0342543