Imboden, Mary T.

Inconsistency in the calculation of time spent in moderate to vigorous physical activity (MVPA) limits interstudy comparability and interpretation of surveillance data. This study assesses whether combining multiple individual methods results in a more accurate estimate of MVPA, while considering the influence of device brand and wear location. Participants (n = 30, age = 49.2 ± 19.5 years) wore two accelerometers (GENEActiv, ActiGraph) on each wrist during two laboratory visits. Individual classification methods (11 for left wrist, eight for right wrist) estimated minutes of MVPA using three approaches (cut-point, two-regression, and machine learning), two types of input (count and raw), and five epoch lengths (1, 5, 15, 30, and 60 s). The consensus estimate was calculated as the mean or median (due to skew) across all individual estimates. No individual or consensus estimates were statistically equivalent to direct observation (mean 38.2 min), with 81%–95% of individual methods overestimating MVPA. The best-performing individual methods were raw acceleration cut-points, with a bias of −3.2 to 2.4 min across devices and wrists. Correlation coefficients between individual methods and the criterion were .35–.71 for the left and .12–.67 for the right wrist, compared to .65–.70 and .58–.66 for consensus methods, respectively. Correlations between device brands were .23–.99 for individual methods and .70–.86 for consensus methods, while correlations between locations were .55–.86 and .73–.87, respectively. Better methods are required for estimating MVPA from wrist-worn accelerometers given the consistent overestimation of MVPA observed. While a consensus method for wrist-worn data was not able to fully resolve these issues, it improves interwrist or interbrand comparability.