The present study estimated the long-term reproducibility of accelerometer-based measures over 6 months in adults and the implications for statistical power, and attenuation in regression coefficients for future activity–disease studies.
We used data from 914 adults in the Interactive Diet and Activity Tracking in AARP study. Participants wore an activPAL 3 (AP) and an ActiGraph GT3X (AG) twice, 6 months apart. AP measures included time spent sitting or lying, standing, and stepping
, whereas AG measures included time spent in sedentary
, light, and moderate-to-vigorous physical activity
(PA). Reproducibility of each metric and implications for epidemiological studies were determined based on intraclass correlation coefficients (ICC; 95% confidence interval).
The ICC values for AP estimates were 0.58 (95% confidence interval, 0.53–0.63) for sitting, 0.62 (0.57–0.67) for standing, and 0.57 (0.51–0.62) for stepping
. The ICC values for AG were 0.56 (0.50–0.61) for sedentary
, 0.54 (0.49–0.60) for light PA, and 0.58 (0.52–0.63) for moderate-to-vigorous PA. Modeling showed that increasing the number of replicate administrations to two or three resulted in the most noticeable increases in ICC values, statistical power, and reductions in attenuation coefficients. For example, administering the AP twice reduced within-subject variability by half and resulted in an increase in the ICC associated with sitting time from 0.58 to 0.74. Similar comparisons for AG and measure of sedentary
time resulted in an increase in ICC values from 0.56 to 0.72. Increasing the number of replicate administrations from one to two reduced the attenuation in activity–outcome associations from 40% to 25%.
Accelerometer-based classifications of activity are moderately stable over time, but there is considerable within-subject variability that needs to be considered when estimating usual activity in future studies.