Integration of Physiological and Accelerometer Data to Improve Physical Activity Assessment

Abstract

Accurate measurement of physical activity (PA) is a prerequisite to determine dose-response relationships between activity and health. The combination of HR and accelerometers (ACC) holds promise for improving the accuracy of PA assessment, but it is unclear how currently proposed modeling techniques compare and to what extent different levels of individual calibration (IC) of HR influence monitoring accuracy. A total of 10 men and women (25.8 +/- 3.4 yr, 1.70 +/- 0.1 m, 71.7 +/- 11.8 kg, 24.4 +/- 5.0 kg.m-2) were recruited for this study, in which IC of HR to PA energy expenditure (PAEE) during both arm crank and treadmill activity were available. Participants completed 6 h of free-living activity, during which PAEE (obtained with indirect calorimetry), HR, hip ACC, arm ACC, and leg ACC were collected. PAEE was then modeled from two different methods of combining HR and ACC (arm-leg HR+M and branched model), both with IC and group-level calibration (GC) of HR, and also from hip ACC estimates alone. Estimates of PAEE were compared with criterion values for PAEE. Combined estimates of PAEE from the arm-leg HR+M and the branched model were similar when IC was used (R2 = 0.81, SEE = 0.55 METs and R2 = 0.75, SEE = 0.61 METs, respectively). When using GC, all estimates of PAEE had larger error, but the performance of the branched model suffered less than the arm-leg HR+M model (R2 = 0.75, SEE = 0.67 METs and R2 = 0.67, SEE = 0.88 METs, respectively). Both combination modeling techniques were more precise than single-measure hip ACC estimates (R2 = 0.41, SEE = 0.96 METs). The combination of HR and ACC improves the accuracy of PAEE estimates and could be applied in large-scale epidemiological studies.