A sensitivity analysis on the variability in accelerometer data processing for monitoring physical activity

Background: Accelerometers are gaining popularity for measuring physical activity, but there are many different ways to process accelerometer data. A sensitivity analysis was conducted to study the effect of varying accelerometer data processing protocols on estimating the association between PA level and socio-demographic characteristics using the National Health and Nutrition Examination Survey (NHANES) accelerometer data. Methods: The NHANES waves 2003-2004 and 2005-2006 accelerometer data (. n=. 14,072) were used to investigate the effect of changing the accelerometer non-wearing time and valid day definitions on the demographic composition of the filtered datasets and the association between physical activity (PA) and socio-demographic characteristics (sex, age, race, educational level, marital status). Results: Under different filtering rules (minimum number of valid day and definition of non-wear time), the demographic characteristics of the final sample varied. The proportion of participants aged 20-29 decreased from 18.9% to 15.8% when the minimum number of valid days required increased from 1 to 4 (. p for trend. <. 0.001), whereas that for aged ≥70 years increased from 18.9% to 20.6% (. p for trend. <. 0.001). Furthermore, with different filters, the effect of these demographic variables and PA varied, with some variables being significant under certain filtering rules but becoming insignificant under some other rules. Conclusions: The sensitivity analysis showed that the significance of the association between socio-demographic variables and PA could be varied with the definition of non-wearing time and minimum number of valid days.