Validation of anthropometric and bioelectrical impedance equations for the prediction of fat mass amongst South African children

While several prediction equations which combine anthropometric, demographic, and/or bioelectrical impedance (BIA) variables to estimate childhood fat mass (FM) are available, comprehensive comparisons of their performance are lacking. We validated FM estimates for children from a range of published equations against reference-standard deuterium dilution observed FM. This cross-sectional study was based on 323 children (42% male) from South Africa of Black African ethnic origins aged 5 to 8 years with information on age, sex, ethnicity, height, weight, deuterium dilution observed FM, triceps and subscapular skinfold thickness, and BIA observed FM, resistance, and impedance. We extracted all equations from three systematic reviews of childhood FM prediction equations that used the above available predictors and were developed on more than 100 males and females. FM estimates from each equation were calculated and the performance of each, as well as FM reported from the BIA manufacturer software, was compared with deuterium dilution observed FM using statistics of R2, Calibration (slope and calibration-in-the-large), and root mean square error (RMSE). Nineteen equations (1 based on basic anthropometry, 12 on skinfold thickness, 6 on BIA) were validated. R2 and RMSE values ranged between 58.3% (BIA manufacturer equation) and 89.0% (Britz et al. (2017) skinfold thickness equation), and between 1.1 kg (Wendel et al. (2016) skinfold thickness equation) and 3.4 kg (Horlick et al. (2002) BIA equation), respectively. Calibration varied considerably across the equations. From the basic anthropometry, skinfold thickness, and BIA categories, the best performing equations from each category were by: Hudda et al. (2019) (basic anthropometry), Wickramasinghe et al. (2008) (skinfold thickness), and Ramirez et al. (2012) (BIA). The performance of published equations varied considerably upon external validation in this South African childhood population. Notably, the Hudda et al. (2019) equation, which relies solely on readily available information of weight, height, sex, age and ethnicity, produced one of the highest R2 values, was well calibrated, and produced a low RMSE value (1.4 kg). Alternative equations which also performed very well relied on additional measurements of skinfold thickness and/or BIA which require equipment, training, extra costs and additional time to obtain.