Athletic bioimpedance-based equations underestimate fat free mass components in male elite soccer players: development and validation of new soccer-specific predictive models.

Background: Bioelectrical impedance analysis (BIA) is a rapid and user-friendly technique for assessing body composition in sports. Currently, no sport-specific predictive equations are available, and the utilization of generalized formulas can introduce systematic bias. The objectives of this study were as follows: (i) to develop and validate new predictive models for estimating fat-free mass (FFM) components in male elite soccer players; (ii) to evaluate the accuracy of existing predictive equations.

Methods: A total of 102 male elite soccer players (mean age 24.7 ± 5.7 years), participating in the Italian first league, underwent assessments during the first half of the in-season period and were randomly divided into development and validation groups. Bioelectrical resistance (R) and reactance (Xc), representing the bioimpedance components, were measured using a foot-to-hand BIA device at a single frequency of 50 kHz. Dual-energy X-ray absorptiometry was employed to acquire reference data for FFM, lean soft tissue (LST), and appendicular lean soft tissue (ALST). The validation of the newly developed predictive equations was conducted through regression analysis, Bland-Altman tests, and the area under the curves (AUC) of regression receiver operating characteristic (RROC) curves.

Results: Developed models were: FFM = - 7.729 + (body mass × 0.686) + (stature/R × 0.227) + (Xc × 0.086) + (age × 0.058), R = 0.97, Standard error of estimation (SEE) = 1.0 kg; LST = - 8.929 + (body mass × 0.635) + (stature/R × 0.244) + (Xc × 0.093) + (age × 0.048), R = 0.96, SEE = 0.9 kg; ALST = - 24.068 + (body mass × 0.347) + (stature/R × 0.308) + (Xc × 0.152), R = 0.88, SEE = 1.4 kg. Train-test validation, performed on the validation group, revealed that generalized formulas for athletes underestimated all the predicted FFM components (p < 0.01), while the new predictive models showed no mean bias (p > 0.05), with R values ranging from 0.83 to 0.91, and no trend (p > 0.05). The AUC scores of the RROC curves indicated an accuracy of 0.92, 0.92, and 0.74 for FFM, LST, and ALST, respectively.

Conclusions: The utilization of generalized predictive equations leads to an underestimation of FFM and ALST in elite soccer players. The newly developed soccer-specific formulas enable valid estimations of body composition while preserving the portability of a field-based method.