Number of accelerometer monitoring days needed for stable group-level estimates of activity

Abstract

To determine the number and distribution of days required to produce stable group-level estimates of a 7 d mean for common accelerometer-derived activity measures. Data from the 2003–2006 NHANES were used in this analysis. The sample included 986 youth (6–19 year) and 2532 adults (⩾20 year) with 7 d of ⩾10 h of wear. Accelerometer measures included minutes of inactive, light physical activity, moderate-to-vigorous physical activity (MVPA); and total activity counts/d. Twenty-five alternative protocols were bootstrapped with 50 000 samples drawn for each protocol. Alternative protocols included: 1–6 random days, Saturday plus 1–5 random weekdays (WD), Sunday plus 1–5 random WD, 1 random weekend day (WE) plus 1–5 WD, and both WE plus 1–4 random WD. Relative difference was calculated between the 7 d mean and alternative protocol mean (((alternative protocol mean – 7 d mean)/7 d mean) * 100). Adult MVPA is used as an example; however, similar trends were observed across age groups and variables except adult inactive time, which was stable across protocols. The 7 d mean for adult MVPA was 44.1(0.9) min d−1. The mean bias for any 1–6 random days ranged from −0.0(0.3) to 0.0(0.2) min d−1 with a relative difference of −0.1 to 0.0%. For protocols with non-random components, bias ranged from −1.4(0.2) to 0.6(0.1) min d−1 with relative difference ranging from −7.2 to 3.1%. Simulation data suggest that stable estimates of group-level means can be obtained from as few as one randomly selected monitoring day from a sampled week. On the other hand, estimates using non-random selection of weekend days may be significantly biased. Purposeful sampling that disproportionally forces inclusion of weekend data in analyses should be discouraged.