Bipolar bioelectrical impedance : lack of accuracy in estimating relative body fat in men Impedância bioelétrica bipolar : falta acuracidade para estimar a gordura relativa em homens

Abstract – Bioelectrical impedance (BI) is commonly used to estimate body composition in different populations. The aim of this study was to determine the accuracy of bipolar BI (Omron BF 300) in estimating relative body fat (%F) by comparing it with dual-energy X-ray absorp-tiometry (DEXA). %F was measured in 66 men (18 to 33 years) with a mean body weight of 73.7 ± 8.4 kg, height of 175.1 ± 6.5 cm, body mass index of 23.9 ± 2.2 kg/m 2 , %F BI 15.1 ± 4.4%, and %F DEXA 17.7 ± 5.1%. A paired t-test, correlation test, standard error of the estimate (SEE), and residue analysis were used to evaluate the accuracy of the method. BI significantly (p 3.5%F) in 42% of the sample. In subjects with higher fat accumulation (%F >20%), the Omron BF BI model produced greater bias in the %F measurement. Therefore, in the case of adult men, the Omron BF 300 BI model does not agree with the DEXA measurement for the estimation of %F in adult men. In addition, the higher the level of fat accumulation, the higher the error in estimated %F.


INTRODUCTION
The scientific community has shown great interest in the quantification of body composition, particularly body fat 1,2 .Excessive accumulation of body fat is a risk factor for the occurrence of morbidities and mortality at different ages 3,4 , whereas low levels are associated with disorders such as anorexia and bulimia, which show a significant prevalence among young adults 5 and adolescents 6 .
The fact that relative body fat (%F) is associated with non-transmissible chronic diseases has encouraged the search for an accurate estimation technique.Dual-energy X-ray absorptiometry (DXA) is considered to be the gold standard, mainly because it also measures bone density 7 .However, due to its laboratory characteristic and high cost, DXA is limited to large--scale screening.In contrast, bioelectrical impedance (BI) analysis has been used for different purposes in Brazil over the last years [8][9][10] , especially because of the easy handling of the equipment, rapid data collection, and low cost.BI is based on the concept that an electric current passes more easily through hydrated fat-free tissue and extracellular water than fat because of the higher electrolyte content of the former, reducing electric resistance or impedance 11 .Thus, resistance to the electric current is inversely proportional to fat content.The traditional BI method is based on the estimation of electric resistance using four or eight electrodes (tetrapolar or octapolar BI, respectively) to transmit the electric current through the body.The resistance observed permits to estimate %F and other body composition components using regression equations.The tetra-and octapolar devices are more expensive, a fact limiting their use in population studies.However, there are simpler devices containing only two electrodes (bipolar BI) coupled to the apparatus itself, which are based on the same principle as the other methods.
Despite its practicality of data collection, the accuracy of bipolar BI devices is still questionable since these devices provide relative and absolute body fat values rather than resistance values.Therefore, the objectives of the present study were a) to compare the accuracy of bipolar BI analysis (Omron BF 300) and DXA for the estimation of %F in adult men, and b) to evaluate the influence of excess body fat on the estimation of %F by bipolar BI compared to DXA.

Study design, sampling, and ethical aspects
The present descriptive, cross-validation study compared bipolar BI analysis and DXA (reference method) for the estimation of %F.The study population consisted of adult men.
The sample was recruited by the distribution of posters and leaflets in places such as clubs, gyms and universities.The sample consisted of 74 adult men ranging in age from 18 to 33 years.Eight subjects were excluded because they did not follow the recommendations for BI analysis.The final sample consisted of 66 subjects.All participants were volunteers and signed a free informed consent form that guaranteed anonymity and confidentiality of the individual data.The subjects had the right to withdraw from the study at any time without penalty as established by the ethical guidelines of Resolution 196/96 of the National Health Council.The study was approved by the Ethics Committee of Universidade Católica de Brasília (process 36/2007).

Data collection
All procedures were conducted at the same place at a controlled ambient temperature (20 o to 22 o C).For each subject, all measurements were obtained on the same day.The DXA exams were performed and analyzed by an examiner with 3-year experience.The individual error was <1.6% (coefficient of variation).
Body weight and height were measured with a balance to the nearest 100 g coupled to a stadiometer (scale of 0.5 cm) 12 .
Measurement of %F by BI: The subjects were asked to adhere to the following protocol: 4-h fast, no consumption of coffee or alcoholic beverages 24 h prior to the test, no strenuous physical activity, and no use of diuretics.The subjects were instructed to empty their bladder and bowel before the measurement.The Omron BF 300 device was used, which is equipped with software that provides relative and absolute body fat values.However, the manufacturer does not inform the regression equation used for their estimation.Measurements were obtained according to the manual of the apparatus.The subject remained standing with the elbows extended in front of the body and holding the device until BI was estimated and %F and fat mass were calculated.
Measurement of %F by DXA: The Lunar DPX-IQ TM (4.7e software) apparatus was used and the procedures were performed according to the manual of the device.The equations used by the software to estimate body composition components including %F are also unknown.Whole-body scan-ning was performed with the subject remaining still in the supine position and the elbows and knees extended.Two calibrations of the apparatus, one weekly and one daily, were performed to ensure the quality of the measurements.A phantom test was performed for weekly calibration and quality assurance testing was used for daily calibration.The calibrations were performed according to the manual of the equipment.
To evaluate the effect of excess adiposity on the estimation of %F by BI, the sample was stratified into two groups: 1) %F <20%, and 2) %F ≥20%.The cut-off values established are based on evidence that indicates risk factors associated with excessive body fat accumulation 13 .

Statistical analysis
The data showed a normal distribution (Kolmogorov-Smirnov test).Descriptive statistics was used for characterization of the sample.The criteria proposed by Lohman 14 were used to test the accuracy of BI compared to DXA: Pearson's linear correlation coefficient (r) >0.79; paired t-test (t) where t calculated < t tabulated , and a standard error of the estimate (SEE) <3.5% for the prediction of %F.The formulas cited in another study 9 with a similar design were used for the calculation of the constant error and SEE.Residual scores were analyzed as proposed by Bland and Altmann 15 .A level of significance of p<0.05 was adopted.The calculations were performed using a licensed copy of the Statistical Package for the Social Sciences (SPSS), version 14.

RESULTS
The sample consisted of adult men ranging in age from 18 to 33 years.Table 1 shows the descriptive characteristics of the sample.Mean %F measured by DXA was 17.7% (range: 8.3 to 28.9%), indicating the presence of subjects with adequate %F and overweight and obese subjects 13,16 .These findings demonstrate the external validity of the study since the sample was heterogenous in terms of %F as observed in the general population, permitting generalization of the results.This is supported by the variability in body mass index (Table 1), with the observation of subjects who belong to different categories of nutritional status 17 .
Comparison of the two methods (Table 2) indicates that BI tends to significantly underestimate (p<0.05)%F, on average 2.6% in relation to DXA.The correlation between methods ranged from moderate to high (r=0.769;p<0.05).The SEE (3.3%) was below the criterion established 14 .Residue analysis (Figure 1) confirmed the tendency of BI (Omron BF 300) to underestimate %F when compared to DXA.There was a higher concentration of residues below point zero, with a mean error of -2.6% (95%CI: -9.1;3.9).As can be seen in Figure 2, the SEE for %F was £3.5% in only 58% of the sample studied, whereas 42% presented an SEE above the cut-off considered to be acceptable (3.5%) 14 .The mean error and standard deviation obtained for the two groups classified according to %F are shown in Figure 3. Mean errors were significantly higher in the group with %F ≥20% (-4.47%) compared to the group with %F <20% (-1.48%) (p<0.05).Mean errors significantly differed in the group with %F ≥20% (%F BI 18.6 ± 3.5% vs %F DXA 23.1 ± 2.4%) (p<0.01),but were similar in the group with %F <20% (%F BI 12.95 ± 3.38% vs %F DXA 14.4 ± 3.12%).

DISCUSSION
Despite the excellent applicability of BI measures in field studies 18,19 , the validity and accuracy of the equipment and of the regression equations used to estimate %F should be determined for a given population.The Omron BF 300 BI device underestimated %F when compared to DXA in the sample studied (Table 2).Bipolar BI has also been shown to underestimate %F by 5.75% in adult women (20-40 years) 20 , by 4.5% in Indians 21 , and by 11% in youngsters 22 when DXA was used as the reference method.The same trend has been reported in another study 23 , but no difference (p>0.05) was observed between the Omron BF 300 BI device and the anthropometric equation of Durnin and Womersley 24 .These findings indicate the lack of accuracy of this method/device for the estimation of %F.
Although the correlation between methods was moderate and the SEE was acceptable 14 , indicating agreement between the two techniques, residue analysis (Figure 1) confirmed the tendency of BI to underestimate %F when compared to DXA.In this respect, the results showed that, if the Omron BF 300 BI model was applied to estimate %F in adult men, the SEE was ≤ 3.5% in only 58% of the subjects.Therefore, the use of this method/device may lead to incorrect diagnoses.As a consequence, possible interventions may also not be the most adequate.A study comparing bipolar and tetrapolar BI analysis revealed no differences between the two models and there was also no influence of the amount of body fat on the estimation of %F 25 .However, these findings should be interpreted with caution since no reference method was used for comparison of the two models.
The constant error and residual scores indicate that a subject may present %F of 16%, whereas, in fact, it is much higher than 20%.As a consequence, the subject is erroneously classified as overweight, although he should be in the normal/ adequate group.
The BI device used in the present study can yield a large number of false-positive and, especially, false-negative results.Users should therefore be aware of this fact in order to correctly interpret the results, taking into account an error in %F quantification of at least 3.5% as recommended 14 .Once demonstrated that the Omron BF 300 device tends to underestimate %F, an SEE of ±3.5% should be added to the value estimated in order to obtain a more reliable diagnosis.The advantage of this BI device seems to be its low cost and practicality, even permitting self-evaluation by the subject.
With respect to the effect of excess adiposity on the estimation errors of bipolar BI compared to DXA, these errors tend to be higher in men with higher adiposity.These results agree with Glaner 9 , who tested the validity of BI equations by comparing a tetrapolar model with DXA and observed that few of the equations developed in other countries are valid for the Brazilian population.In addition, this author 9 showed that the only equation valid for adult men is that specific for subjects with elevated adiposity (%F ≥ 20%).This fact demonstrates that BI models should include the degree of adiposity of the subjects to increase the accuracy of estimation.
The Omron BF 300 BI device was not valid for the estimation of %F in men.However, other models (Omron BF 550) have shown good agreement with DXA measurements in women 26 , and no difference in %F estimates was observed between the Omron BF 306 BI and DXA 22 .
One limitation of the present study is that DXA was used as the gold standard for the quantification of %F, although the manufacturer of these devices only recommends their use for the quantification of bone density 27 .This may increase the possibility of errors and reduce the validity of the results.However, in vitro experiments found no differences (p ≥ 0.05) in the estimates of different body components 28,29 , a fact justifying the use of this method in the present study.Another limitation is the use of only one bipolar BI model.As a consequence, the present results can only be extrapolated if the same device is used.

CONCLUSIONS
The present results demonstrate that the Omron BF 300 BI model is not accurate for the estimation of %F in adult men, with the estimation error increasing with increasing body fat.

Figure 1 .
Figure 1.Comparison of residual scores of relative body fat (%F) measured by dual-energy X-ray absorptiometry (DXA) and bioelectrical impedance (BI) (%F BI -%F DXA ) with mean %F DXA and %F BI .The dotted lines indicate the 95% confidence interval.The continuous line corresponds to the mean error.

Figure 2 .
Figure 2. Analysis of the standard error of the estimate (SEE) for agreement between relative body fat estimation by dual-energy X-ray absorptiometry and bioelectrical impedance (BI).

Table 1 .
Descriptive analysis of anthropometric variables and relative body fat in the sample studied (n = 66).

Table 2 .
Statistical comparison of relative body fat (%F) obtained by dual-energy X-ray absorptiometry (DXA) and bioelectrical impedance (BI) in adult men.t: t-test; r: Pearson's correlation coefficient; CE: constant error (%F BI -%F DXA ); SEE: standard error of the estimate; a p<0.05.