Statistical analysis

To express precision and repeatability of BIA measurements, the coefficient of variation was calculated and expressed as a percentage, defined as the ratio of the standard deviation to the mean. Studies with a coefficient of variation of more than 5% between the three readings were excluded from data analysis. Descriptive statistics were reported for both normal and overweight children.

We compared our BIA results with two previously published studies by dilution method in children [14, 15]. The mean of two or three BIA readings for each participant was used for data analysis and the coefficient of variation was calculated and reported for each. Demographic data from study participants was reported using frequencies procedures. Scatterplot graphs and best fit regression equations were reported separately for normal and overweight children, as well as gender. Bioimpedance data obtained from normal weight children was compared to two studies using the isotope dilution methodology. Regression equations were calculated for each study and standardized residual values computed. Median differences in standardized residuals between study groups were examined using a non-parametric Kruskal-Wallis procedure, with pair wise comparisons conducted with a non-parametric Mann Whitney U procedure. Accuracy of prediction models between studies were assessed using explained variance (R2), mean square error (MSE), square root of MSE, and average absolute percent error. The study was approved by our Institutional Review Board. All statistical procedures were conducted using NCSS statistical software Version 11.0.