Accelerometers have been embraced by the scientific community as valid and reliable physical activity (PA) assessment tools. This is especially true in some sub-populations, such as children, who lack the ability to self- report their behaviors. Tremendous progress has been made in the objective assessment of PA and sedentary behavior (SB). Several measurement challenges remain that limit our ability to identify; dose-response associations with disease and risk factors, prevalence and determinants of these important health behaviors, changes in behavior due to time and intervention efforts. To overcome these challenges, we propose novel calibration procedures that will leverage the richness of three axes of raw acceleration data during free-living calibration activities, from wrist- and hip-worn accelerometers. Consistent with current PA recommendations for youth, the proposed study will focus on calibrating and validating accelerometer output to quantify intensity and duration of PA and identifying specific predominant modes of activity (e.g., sedentary time, locomotion, and fine/gross motor activity). Our primary goal is to follow best-practices data collection and analysis procedures to develop a comprehensive set of accelerometer algorithms that will accurately and precisely estimate minutes of various intensities of PA and mode of activity from hip- and wrist-worn accelerometers in children and adolescents (18 months to 21 years old), using video recorded direct observation during free-living activities as the criterion measure.
Accurate and precise estimates of physical activity, sedentary behavior, and patterns of those behaviors in youth are essential in order to identify changes in those behaviors and to inform related policy decisions. The proposed research will extend previous accelerometer calibration studies by videotaping free-living activities in a large sample across a wide age range of children and adolescents. The raw acceleration signals from both hip- and wrist-worn devices will be analyzed using linear regression and machine learning methods to generate a family of freely available algorithms that produce accurate and precise estimates, reducing inconsistencies and confusion on use of previous accelerometer cutpoints in youth.
