Accuracy of Anthropometric Equations for Estimating Body Fat in Professional Male Soccer Players Compared with DXA

Background There are several published anthropometric equations to estimate body fat percentage (BF%), and this may prompt uncertainty about their application. Purpose To analyze the accuracy of several anthropometric equations (developed in athletic [AT] and nonathletic [NAT] populations) that estimate BF% comparing them with DXA. Methods We evaluated 131 professional male soccer players (body mass: 73.2 ± 8.0 kg; height: 177.5 ± 5.8 cm; DXA BF% [median, 25th–75th percentile]: 14.0, 11.9–16.4%) aged 18 to 37 years. All subjects were evaluated with anthropometric measurements and a whole body DXA scan. BF% was estimated through 14 AT and 17 NAT anthropometric equations and compared with the measured DXA BF%. Mean differences and 95% limits of agreement were calculated for those anthropometric equations without significant differences with DXA. Results Five AT and seven NAT anthropometric equations did not differ significantly with DXA. From these, Oliver's and Civar's (AT) and Ball's and Wilmore's (NAT) equations showed the highest agreement with DXA. Their 95% limits of agreement ranged from −3.9 to 2.3%, −4.8 to 1.8%, −3.4 to 3.1%, and −3.9 to 3.0%, respectively. Conclusion Oliver's, Ball's, Civar's, and Wilmore's equations were the best to estimate BF% accurately compared with DXA in professional male soccer players.


Introduction
Body fat percentage (BF%) is probably the most evaluated body composition component in sports [1]. In soccer, BF% estimation is part of the habitual assessment [2], because it is related to exercise performance [3,4], may be useful in seasonal monitoring [5,6], and may help compare players in different playing positions [3,[7][8][9] and experience [10,11].
Several methods are available when evaluating BF% [1,12]. Dual-energy X-ray Absorptiometry (DXA) is a laboratory method used to assess body composition, which is increasingly being used to evaluate BF% in soccer players [6,[8][9][10] due to its practicality and accuracy compared with other reference methods [1,13,14].
Conversely, anthropometry, one of the most popular field methods in assessing body composition, is used to estimate BF%, typically with regression equations obtained from other laboratory methods [1,12]. These equations have been developed in athletic and nonathletic populations and they have a too wide variability of estimation when applied to the same subject/population. Similarly, several studies have investigated the accuracy of these equations in estimating 2 Journal of Sports Medicine BF% in specific populations [15][16][17][18][19][20][21]. However, little is known about their accuracy when they are applied on professional male soccer players [16,20]. Therefore, the purpose of this study was to analyze the accuracy of several anthropometric equations that estimate BF% utilizing DXA as the reference method in professional male soccer players. Since there is a great amount of anthropometric equations [1], we decided to analyze some of these separately as follows:  , waist circumference with a metallic tape (to nearest 0.1 cm), and elbow diameter (Campbell 10, Rosscraft, Canada) to nearest 0.05 cm. All measurements were evaluated by trained personnel following standardized procedures. All measurements were evaluated following the ISAK protocol [22,23], but chest and axilla skinfolds were evaluated following Lohman's proposed sites [24]. The intraobserver technical error of measurement in our laboratory is ≤5% for skinfolds and ≤1% for all other measurements. The interobserver technical error of measurement in our laboratory is ≤7.5% for skinfolds and ≤1.5% for all other measurements.

DXA Scanning.
A whole body scan was performed for each subject with a DXA fan beam equipment (Hologic QDR4500 Explorer, Massachusetts, USA). The equipment was calibrated daily following the manufacturers indications. All scans were analyzed with the software for Windows5 Hologic QDR v 12.1 (1986-2002©, Hologic Inc.). The zone of head was excluded from the scan in order to calculate the body fat percentage. The difference between DXA whole body mass and body mass on scale was on average −1.0±0.7 kg. The whole procedure was executed by a certified DXA technician. His technical error of measurement was 0.4 ± 2.9% for body fat, −0.2 ± 1.2% for lean body mass, and 0.0% ± 0.5% for total mass.

Statistical Analysis.
In order to determine which equations estimated the BF% different than the one obtained with DXA, all anthropometrical BF% estimations were tested with the Kruskal-Wallis and Dunn's post hoc test, considering a significance level of < 0.05. Taking only the equations that showed no significant difference with DXA, we carried out a modified Bland-Altman analysis. Mean difference, linear correlation, and 95% limits of agreement for each equation were calculated [51]. Variables with normal distribution (D' Agostino-Pearson test) are expressed as mean ± SD and as median (25th-75th percentile) if they were not normally distributed. All analyses were carried out with GraphPad5 Prism 7 for Windows (GraphPad Software, California, USA).  Table 3.

Body Fat Percentage Estimation with AT Anthropometric
Equations. From the 14 anthropometric equations analyzed, five showed no significant differences compared with DXA (Table 4). From these equations, those with the narrowest limits of agreement with DXA were the proposed by Oliver [32] and Civar [18], followed by Zuti [30], Forsyth (4SKF), and Forsyth (2SKF) [25] equations. These last three equations showed small mean differences with DXA but wider limits of agreement (Table 5). Oliver's equation showed slightly narrower limits of agreement and a smaller mean difference compared with Civar's equation (Table 5).

Body Fat Percentage Estimation with NAT Anthropometric
Equations. From the 17 anthropometric equations analyzed, seven showed no significant differences compared with DXA (Table 6). From these equations, those with the narrowest limits of agreement and smallest mean differences with DXA were Ball's [44] and Wilmore's [43] equations, followed by Durnin-R [34], van der Ploeg [48], Durnin-W [35], Lean [38], and Garcia [49] equations. These last five equations showed wider limits of agreement and/or higher mean differences with DXA (Table 7). Ball's equation showed slightly smaller mean difference and narrower limits of agreement with DXA than Wilmore's equation (Table 7).

Discussion
The purpose of this study was to analyze the accuracy of BF% obtained through several anthropometric equations and that measured with DXA in professional male soccer players. We only found two other studies that addressed the same issue in soccer players [16,20]. In the study of Reilly et al. [16] they evaluated professional male soccer players from the English premier league and settled DXA as the reference method.  [20] studied professional male soccer player from Brazil; however, they compared the body density obtained through anthropometric equations with that evaluated with hydrostatic weighting. They reported that Jackson's equations (using three and seven skinfolds) [36] and Lohman's equation [39] did not differ significantly with the reference method, and Lohman's equation showed the lowest estimated standard error. In the present study, we found these three equations differed significantly with DXA (Table 6).
Other studies have compared anthropometric equations with hydrostatic weighting in different athletic populations [17,18,21]. The study of Sinning et al. [17] reported that Jackson's equation (seven skinfolds) [36] was not different to the reference method and showed a low SEE in collegiate male athletes. Civar et al. [18] reported that the Lohman [39] and Durnin and Womersley (Durnin-W) [35] equations did not differ with the reference method, but Durnin-W equation showed a higher correlation and a lower SEE than Lohman's equation in college males that practiced different sports. Smith and Mansfield [21] evaluated male football players and reported that Jackson's equation (three skinfolds) [36] did not differ significantly with the reference method. Finally, Knechtle et al. [15] employed a different reference method; they studied to ultra-endurance male athletes and evaluated their BF% with bioelectrical impedance. They reported that Ball's equation [44] showed the highest agreement with their reference method.
Different results between previous studies and ours may be explained by the different population in which they were validated (football players, college age men, etc.) and the reference method used.
From the five AT anthropometric equations that did not differ significantly with DXA, four were validated with hydrostatic weighting [18,25,30] and one was validated with DXA [32]. This last one is with highest agreement with DXA in our study (Tables 5 and 7). On the other hand, from the seven NAT anthropometric equations that did not differ significantly with DXA, five were developed with densitometric methods [34,35,38,43,48] and two with DXA [44,49]. Even though these equations were developed with a different body composition method, they were able to estimate BF% with a reasonable degree of agreement with DXA in our sample (Table 7).
Oliver's equation [32] showed the narrowest limits of agreement with DXA in our population (Tables 5 and 7). Oliver's equation was developed in a large sample of college football players. This equation employs the sum of seven Equations are listed from the narrowest to the widest limits of agreement. 2SKF, two skinfolds; 4SKF, four skinfolds; DXA, dual-energy X-ray absorptiometry. * All values were significant ( < 0.001). skinfolds, including limbs and trunk measurements. The accuracy showed with this equation in our study could be explained by the number of skinfold thicknesses employed and the whole body representativeness from these skinfolds and because they also employed DXA as the reference method. Wilmore's equation was developed in male university students (aged 16 to 37 years) [43]. Surprisingly, Although it employs only two skinfolds (abdomen and thigh), it had a higher agreement than those that used more measurements.
In a practical way, Wilmore's equation may be more useful because it accurately estimated BF% using few variables compared to the other equations. From another perspective, we recommend Oliver's equation, as it had higher agreement with DXA, and because it employs seven skinfolds it may offer a better whole body perspective. Equations are listed from the narrowest to the widest limits of agreement. DXA, dual-energy X-ray absorptiometry. * Significant correlation ( < 0.001).
Some of the limitations of this study were as follows: (1) The results of our reference method (DXA) may differ from other reference methods; however several studies have reported that DXA is a practical and accurate method for assessing BF% [1,[12][13][14].
(2) The protocols from the original studies may differ with the one we followed to evaluate the anthropometric measurements; however we tried to keep the anatomical sites as similar as possible.
(3) The instruments we used to evaluate the anthropometric measurements also differed from some of the authors' original studies.
Some of our strengths were as follows: (1) All our personnel (for anthropometric assessments and DXA scanning) are certified to do this kind of evaluations.
(2) This study has a large sample.
(3) We included many AT and NAT anthropometric equations.

Conclusions
In this study, we found that Oliver's (AT), Ball's (NAT), Civar's (AT), and Wilmore's (NAT) equations had the highest agreement with DXA for estimating BF% in professional male soccer players. These equations can be used alternatively to DXA for estimating BF% in a cross-sectional way. It remains to be seen if these equations are useful for monitoring changes in BF% over time. These issues deserve further research.

Disclosure
This work is an extended version of a previous poster presentation.

Conflicts of Interest
The authors declare that they have no conflicts of interest.