How Many Accelerometer Days are Needed for Stable Population and Individual Weekly Activity Estimates?

Abstract

PURPOSE: To determine number and distribution of monitoring days required to produce stable individual and population level estimates of a true 7 day mean for common accelerometer parameters. METHODS: Data for n=2532 adults (age≥20yrs) participating in the 2003-6 NHANES activity monitor protocol with ≥10 hrs wear for 7 days were analyzed. Accelerometer parameters included: 1) Moderate-to-vigorous physical activity (MVPA) minutes (≥2020 cnt/min); 2) bouted MVPA minutes; 3) sedentary minutes (<100 cnt/min); 4) total counts; and 5) average counts/min wear. Both random and systematic day deletion schemes were examined, including keeping 1 to 6 days at random, keeping Saturday + 1 to 5 random weekdays, keeping Sunday + 1 to 5 random weekdays, keeping 1 weekend day at random + 1 to 5 weekdays, and keeping both weekend days + 1 to 4 random weekdays. The above 25 deletion schemes were bootstrapped with 250 independent samples drawn from the master 7 day dataset for each deletion scheme. For each of n=250 bootstraps per deletion scheme, individual and group level means for the 5 outcomes were calculated and saved. To compare group level outcomes, percent difference was calculated between the true 7 day mean and the mean from each deletion scheme [((true mean - deletion mean)/ true mean)*100]. RESULTS: MVPA is used as an example for presentation. Similar trends were observed across variables except sedentary minutes, which remained relatively stable across all deletion schemes. Population MVPA for any 1 to 6 random days ranged from 20.0±0.3 to 20.0±0.0 with a mean bias ranging from -0.1±0.8% to 0.2±1.0%. Population MVPA for the 19 deletions with non random components ranged from 18.6±0.1 to 20.6±0.0 with a mean bias ranging from -3.0±0.0% to 7.2±0.5%. Similar to population results, individual bias is minimized for any 1 to 6 random days and observed to increase for deletions with non-random components. CONCLUSIONS: Simulation data suggest that stable estimates of population means can be obtained from a single randomly selected day of monitoring from a sampled week. However, bias is introduced in both population and individual estimates based on non-random selection of weekend days. Purposeful sampling of adults which forces inclusion of weekend data in analyses should be discouraged.