Accurate assessments of both physical activity and sedentary behaviors are crucial to understand the health consequences of movement patterns and to track changes over time and in response to interventions.
Purpose: The study evaluates the validity of an integrative, machine learning method for processing activity monitor data in relation to a portable metabolic analyzer (Oxycon mobile [OM]) and direct observation (DO).
Methods: Forty-nine adults (age 18-40 yr) each completed 5-min bouts of 15 activities ranging from sedentary to vigorous intensity in a laboratory setting while wearing ActiGraph (AG) on the hip, activPAL on the thigh, and OM. Estimates of energy expenditure (EE) and categorization of activity intensity were obtained from the AG processed with Lyden's sojourn (SOJ) method and from our new sojourns including posture (SIP) method, which integrates output from the AG and activPAL. Classification accuracy and estimates of EE were then compared with criterion measures (OM and DO) using confusion matrices and comparisons of the mean absolute error of log-transformed data (MAE ln Q).
Results: The SIP method had a higher overall classification agreement (79%, 95% CI = 75%-82%) than the SOJ (56%, 95% CI = 52%-59%) based on DO. Compared with OM, estimates of EE from SIP had lower mean absolute error of log-transformed data than SOJ for light-intensity (0.21 vs 0.27), moderate-intensity (0.33 vs 0.42), and vigorous-intensity (0.16 vs 0.35) activities.
Conclusions: The SIP method was superior to SOJ for distinguishing between sedentary and light activities as well as estimating EE at higher intensities. Thus, SIP is recommended for research in which accuracy of measurement across the full range of activity intensities is of interest.