An Evaluation of Accelerometer-derived Metrics to Assess Daily Behavioral Patterns

Abstract

Introduction The way physical activity (PA) and sedentary behavior (SB) are accumulated throughout the day (i.e., patterns) may be important for health, but identifying measurable and meaningful metrics of behavioral patterns is challenging. This study evaluated accelerometer-derived metrics to determine if they predicted PA and SB patterns and were reliably measured. Methods We defined and measured 55 metrics that describe daily PA and SB using data collected by using the activPal (AP) monitor in four studies. The first two studies were randomized-crossover designs that included recreationally-active participants. Study one experimentally manipulated time spent in moderate-vigorous intensity physical activity (MVPA) and sedentary time and study two held time in exercise constant and manipulated SB. Study three included inactive participants who increased exercise, decreased sedentary time or both. The study conditions induced distinct behavioral patterns, thus we tested whether the new metrics could improve the prediction of an individual’s study condition after adjusting for the overall volume of physical activity or SB using conditional logistic regression. In a fourth study, we measured the three-month reliability for the pattern metrics by calculating intraclass correlations (ICCs) in a community-dwelling sample who wore the AP monitor twice for seven-days. Results In each of the experimental studies, we identified new metrics that could improve the accuracy for predicting condition beyond SB and MVPA volume. In study one, 23 metrics were predictive of a highly active condition and 24 metrics were predictive of a highly sedentary condition in study two. In the fourth study, the median ICC (25–75th percentiles) of the metrics were 0.59 (0.46, 0.65). Conclusions Several new metrics were predictive of patterns of SB, exercise and non-exercise behavior and are moderately reliable over a three-month period. Applying these metrics to determine if daily behavioral patterns are associated with health-outcomes is an important area of future research.

A joint modeling and estimation method for multivariate longitudinal data with mixed types of responses to analyze physical activity data generated by accelerometers

“…Let H(·) be our inverse function of logit link with H(·) = exp(·)∕{1 + exp(·)}, and H ′ (·) and H ′′ (·) be its first and second derivatives. Given specified values of (̂,û i ), the proportional data model in (2) 4). Following the approximation methods discussed in previous studies, 9,[15][16][17] the formulation is displayed as follows with detailed derivations described in the Appendix A.2:…”

“…At visit time j, subject i has 8 observations {Y (1) ij , … , Y (8) ij } as described in Section 2. The continuous variables Y (1) ij and Y (2) ij are generated according to…”

“…We set ( ) 0 = 0.5, ( = 1, 2, … , 8) in 2 × 2 matrix with diagonal elements to be Ψ ( ) [1,1] = Ψ ( ) [2,2] = 0.5 and off-diagonal elements to be Ψ ( ) [1,2] = 0.25, and Ψ (̃) in 2 × 2 matrix with all elements to be 0.1. We also set 2(1) = 2(2) = 1 and 2(3) = 2(4) = 1∕30.…”

“…Recent literature 2 shows that a complete description of an individual's pattern of physical activity requires specification of multiple variables, such as those listed in Figure 1A-H. Among these 8 variables, we note that 2 are continuous measures, 2 are proportions, 2 are counts, and 2 are binary.…”

Deriving objectively-measured sedentary indices from free-living accelerometry data in rural and urban African settings: a cost effective approach

“…To date, the extraction of these objectively-measured sedentary-related variables, has utilized software which requires some level of programming skills (R, SAS, MatLab) [1–3, 5, 9]. More recently, commercial proprietary software such as Actilife [4] and MeterPlus™ [11] provide researchers with the ability to extract these variables without the need for programming skills.…”