The validity of predicted body fat percent from body mass index and from impedance in samples of five European populations

Objectives: To test and compare the validity of a body mass index (BMI)-based prediction equation and an impedance-based prediction equation for body fat percentage among various European population groups. Design: Cross-sectional observational study. Settings: The study was performed in five different European centres: Maastricht and Wageningen (The Netherlands), Milan and Rome (Italy) and Tampere (Finland), where body composition studies are routinely performed. Subjects: A total of 234 females and 182 males, aged 18-70 y, BMI 17.0-41.9 kg/m2. Methods: The reference method for body fat percentage (BF?F) was either dual-energy X-ray absorptiometry (DXA) or densitometry (underwater weighing). Body fat percentage (BF€was also predicted from BMI, age and sex (BFÆI) or with a hand-held impedance analyser that uses in addition to arm impedance weight, height, age and sex as predictors (BF6P). Results: The overall mean (±s.e.) bias (measured minus predicted) for BFÆI was 0.2±0.3 (NS) and-0.7±0.3 (NS) in females and males, respectively. The bias of BF6P was 0.2±0.2 (NS) and 1.0±0.4 (P<0.01) for females and males, respectively. There were significant differences in biases among the centres. The biases were correlated with level of BF and with age. After correction for differences in age and BF␋etween the centres the bias of BFÆI was not significantly different from zero in each centre and was not different among the centres anymore. The bias of BF6P decreased after correction and was significant from zero and significant from the other centres only in males from Tampere. Generally, individual biases can be high, leading to a considerably misclassification of obesity. The individual misclassification was generally higher with the BMI-based prediction. Conclusions: The prediction formulas give generally good estimates of BF␘n a group level in the five population samples, except for the males from Tampere. More comparative studies should be conducted to get better insight in the generalisation of prediction methods and formulas. Individual results and classifications have to be interpreted with caution