Development of an accelerometer-based multivariate model to predict free-living energy expenditure in a large military cohort

This study developed a multivariate model to predict free-living energy expenditure (EE) in independent military cohorts. Two hundred and eighty-eight individuals (20.6 +/- 3.9 years, 67.9 +/- 12.0kg, 1.71 +/- 0.10m) from 10 cohorts wore accelerometers during observation periods of 7 or 10 days. Accelerometer counts (PAC) were recorded at 1-minute epochs. Total energy expenditure (TEE) and physical activity energy expenditure (PAEE) were derived using the doubly labelled water technique. Data were reduced to n=155 based on wear-time. Associations between PAC and EE were assessed using allometric modelling. Models were derived using multiple log-linear regression analysis and gender differences assessed using analysis of covariance. In all models PAC, height and body mass were related to TEE (P < 0.01). For models predicting TEE (r 2=0.65, SE=462kcal center dot d1 (13.0%)), PAC explained 4% of the variance. For models predicting PAEE (r 2=0.41, SE=490kcal center dot d1 (32.0%)), PAC accounted for 6% of the variance. Accelerometry increases the accuracy of EE estimation in military populations. However, the unique nature of military life means accurate prediction of individual free-living EE is highly dependent on anthropometric measurements.