Validation of two portable bioelectrical
impedance analyses for the assessment of
body composition in school age children
Li-Wen Lee1,2, Yu-San Liao1,2, Hsueh-Kuan Lu3, Pei-Lin Hsiao1, Yu-Yawn Chen4,5, Ching-Chi Chi6,7,8☯*, Kuen-Chang Hsieh9,10☯*
1 Department of Diagnostic Radiology, Chang Gung Memorial Hospital, Chiayi, Taiwan, 2 Department of Nursing, Chang Gung University of Science and Technology, Chiayi Campus, Chiayi, Taiwan, 3 Sport Science Research Center, National Taiwan University of Sport, Taichung, Taiwan, 4 Department of Physical Education, National Taiwan University of Sport, Taichung, Taiwan, 5 Department of Cosmetic Application and Management, St. Mary’s Junior College of Medicine, Nursing and Management, Yilan, Taiwan, 6 Department of Dermatology, Chang Gung Memorial Hospital, Linkou, Taiwan, 7 Department of Dermatology, Chang Gung Memorial Hospital, Chiayi, Taiwan, 8 College of Medicine, Chang Gung University, Taoyuan, Taiwan, 9 Office of Physical Education and Sport, National Chung Hsing University, Taichung, Taiwan, 10 Research Center, Charder Electronic Co, Ltd, Taichung, Taiwan
☯These authors contributed equally to this work.
*chingchi@cgmh.org.tw(CCC);abaqus0927@yahoo.com.tw(KCH)
Abstract
Background
Bioelectrical impedance analysis (BIA) is a convenient and child-friendly method for longitu-dinal analysis of changes in body composition. However, most validation studies of BIA have been performed on adult Caucasians. The present cross-sectional study investigated the validity of two portable BIA devices, the Inbody 230 (BIA8MF) and the Tanita BC-418 (BIA8SF), in healthy Taiwanese children.
Methods
Children aged 7–12 years (72 boys and 78 girls) were recruited. Body composition was measured by the BIA8SFand the BIA8MF. Dual X-ray absorptiometry (DXA) was used as the reference method.
Results
There were strong linear correlations in body composition measurements between the BIA8SFand DXA and between the BIA8MFand DXA. Both BIAs underestimated fat mass (FM) and percentage body fat (%BF) relative to DXA in both genders The degree of agree-ment in lean body mass (LBM), FM, and %BF estimates was higher between BIA8MFand DXA than between BIA8SFand DXA. The Lin’s concordance correlation coefficient (ρc) for LBM8MFmet the criteria of substantial to perfect agreement whereas theρcfor FM8MFmet the criteria of fair to substantial agreement. Bland-Altman analysis showed a clinically acceptable agreement between LBM measures by BIA8MFand DXA. The limit of agreement a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS
Citation: Lee L-W, Liao Y-S, Lu H-K, Hsiao P-L, Chen Y-Y, Chi C-C, et al. (2017) Validation of two portable bioelectrical impedance analyses for the assessment of body composition in school age children. PLoS ONE 12(2): e0171568. doi:10.1371/ journal.pone.0171568
Editor: Francesco Cappello, University of Palermo, ITALY
Received: August 31, 2016 Accepted: January 21, 2017 Published: February 3, 2017
Copyright:© 2017 Lee et al. This is an open access article distributed under the terms of theCreative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability Statement: All relevant data are within the paper.
Funding: This study was funded by a single institution, the Chang Gung Memorial Hospital (grant number CMRPG6D0353, CCC and CMRPG6C0052, LWL). The funder had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. One of the authors (KCH) is an employee of Charder Electronic Co, Ltd. This company did not provide KCH financial support in executing this study. Nor
in %BF estimation by BIA and DXA were wide and the errors were clinically important. For the estimation of ALM, BIA8SFand BIA8MFboth provided poor accuracy.
Conclusions
For all children, LBM measures were precise and accurate using the BIA8MFwhereas clini-cally significant errors occurred in FM and %BF estimates. Both BIAs underestimated FM and %BF in children. Thus, the body composition results obtained using the inbuilt equa-tions of the BIA8SFand BIA8MFshould be interpreted with caution, and high quality validation studies for specific subgroups of children are required prior to field research.
Introduction
Growth monitoring is important for early detection of health and nutritional problems during child development. Growth charts of length-for-age, weight-for-age, and BMI-for-age are cur-rently used to assess physical growth in children. These charts can provide a general clinical overview of the health and nutritional status of children. However, body composition under-goes dynamic changes throughout growth and development, and current growth charts only provide proxy measures for changes in body composition.
The techniques most commonly used to assess body composition in children are underwa-ter weighing, isotope dilution, dual-energy X-ray absorptiometry (DXA), air-displaced plethysmography, and bioelectrical impedance analysis (BIA). Among these techniques, BIA employs portable equipment and is a safe, convenient, and child-friendly method that is suit-able for measurement and tracking of body composition changes in children [1].
The two common BIA techniques are the whole-body and segmental modes, in which a current passes from hand-to-foot, foot-to-foot, or hand-to-hand, with subjects either in the supine position or standing [2]. Whole-body BIA employs four electrodes attached to different sides of the body for measurement of electrical resistance. Body composition parameters, such as fat free mass (FFM), lean body mass (LBM), fat mass (FM), and percentage body fat (%BF), are then calculated using specific equations based on recorded impedance, height, age, sex, anthropometric index, and other factors [3]. Multi-segmental BIA employs eight electrodes to calculate whole-body and regional body composition, and can provide information on the spa-tial distribution of different components of body composition and their changes over time [4]. Therefore, multi-segmental BIA is theoretically superior to classical BIA for studies of pediatric body composition. Moreover, multi-segmental BIA can provide an estimate of appendicular lean mass (ALM), which constitutes the majority of skeletal muscle mass (SM) and thus can be used as a proxy for SM [5,6].
Multi-segmental BIA is available in single-frequency and multi-frequency modes. Single-frequency BIA generally employs a 50 kHz current that passes through extracellular and intra-cellular fluids for estimation of total body water [7]. The multi-frequency method uses multiple frequencies to differentiate intracellular from extracellular fluid, and, therefore, provides a bet-ter estimation of total body wabet-ter than the single frequency method [7]. However, there is con-troversy concerning whether multi-frequency BIA provides more accurate estimates of body composition in children compared with the single frequency method [3,8,9].
Previous BIA validation studies were conducted predominantly in adult Caucasians [7]. Pietrobelli et al. [10] demonstrated that appendicular electrical resistance had a strong positive correlation with ALM in white healthy adults, and could be used to estimate the lean mass of
did the company have any additional role in the research funding, study design, data collection and analysis, decision to publish, or preparation of the manuscript. The specific roles of the author are articulated in the Author Contributions section. Competing Interests: One of the authors (KCH) is employed by Charder Electronic Co, Ltd. This does not alter our adherence to all the PLoS ONE policies on sharing data and materials. There are no patents, products in development nor marketed products to be declared. The other authors declare no conflict of interest.
the limbs. However, children are not simply “miniature adults”, thus, equations established for adults may not be applicable to children. Therefore, it is necessary to investigate the reliability and validity of different BIA devices before initiation of field studies on pediatric body composition.
This cross-sectional study of healthy Taiwanese children (age 7–12 years) examined the accuracy and validity of two portable multi-segmental BIA devices by comparing their results with those from DXA measurements.
Materials and methods
Study design
This cross-sectional study was approved by local Institutional Review Board of the Chang Gung Memorial Hospital (103-1027A3), and written informed consent was provided by the subjects and their parents. Subjects were recruitedvia hospital advertisements and word-of-mouth from February to December, 2015. All subjects were healthy Taiwanese children aged 7–12 years-old. None of the subjects were pregnant, had amputations, implants, or chronic ill-nesses, or were prescribed regular medication.
Participants were instructed to eat breakfast on the study day and then fasted completely for at least 2 h before reporting to the Chang Gung Memorial Hospital (Chiayi branch) between 8:30–11:00 am. Vigorous activities and alcohol were avoided for a minimum of 48 h before the study day. Girls were not given appointments during their menstrual cycle. On arrival, participants were asked to void and change into a hospital gown. All measurements including body weight, height, BIA, and DXA were completed on the same morning, with a total study time of approximately one hour. One measurement per subject was performed using each instrument. Body height (cm) and weight (kg) were measured with subjects wear-ing no shoes uswear-ing a digital scale (Super-View, HW-3050, Taipei, Taiwan).
Bioelectrical impedance analysis (BIA)
All BIA measurements were made by trained research assistants. Subjects were measured wear-ing hospital gowns (< 0.2 kg) and weight adjustment for clothwear-ing was not applied. A swear-ingle-fre- single-fre-quency (50 kHz, 500μA) BIA device (Tanita BC-418, Tanita Corp., Tokyo, Japan), referred to as BIA8SF, was used to estimate LBM8SF, ALM8SF, FM8SF, and %BF8SF[11]. This method allows
bioelectricity impedance measurement of the whole body and each part (right leg, left leg, right arm and left arm). The age limits for the BIA8SFare 7–99 years. After the sex, age and height
information had been entered into the BIA8SF, subjects were asked to stand in a stable position
with bare feet. Their toes and heels were placed in contact with the anterior and posterior elec-trodes of the weighting platform, respectively. The measurements began when the grips were grasped by both hands. With BIA8SF, electric current was supplied from the toe tips of both feet
and the fingertips of both hands, and the voltage was measured on the heel of both feet and the thenar area of both hands. Finally, the inbuilt equation was used to convert the input imped-ance to body composition estimates. Test-retest reliability for whole body LBM and %BF esti-mates by BIA8SFwere both 0.99 (n = 5) using the intra-class correlation coefficient (ICC).
A multi-frequency (20 kHz and 100 kHz) BIA device using eight-point tactile electrode sys-tem (Inbody 230, Biospace Corp., Seoul, Korea), referred to as BIA8MF, was used to measure
LBM8MF, ALM8MF, FM8MF, and %BF8MF[12]. The BIA8MFis suitable for individuals aged
3–99 years-old according to the manufacturer. The BIA8MFproduces 10 impedance values by
using two different frequencies to measurement the five segments of the body (right leg, left leg, right arm, left arm and the trunk). The measurement procedure for BIA8MFwas similar to
handle for BIA8MF. Body composition estimates were calculated by using the manufacturer’s
software (Lookin’Body 120, Biospace Corp., Seoul, Korea). Test-retest reliability for whole body LBM and %BF estimates by BIA8MFwere both 0.99 (n = 5) using ICC.
Dual-energy X-ray absorptiometry (DXA)
DXA is the reference method for assessment of body composition. Whole body DXA was per-formed using a fan-beam system (Delphi A, QDR series, Hologic, Bedford, MA, USA) config-ured with software version 12.5. The scanner was equipped with switched pulse dual-energy x-ray tube, operating at 100 kVp and 140 kVp. Thein vivo precision of the scanner for whole body measurement was 1.0%, according to the product specification. The scanner was cali-brated daily with the Hologic spine and body composition step phantoms before scanning the subject. Then, subjects were instructed to lie supine on the scanning bed. The DXA operator manually assisted subjects to position within the scanning zone with their head, neck and torso parallel to the long-axis of the scanning bed; arms at their sides; palms down; legs inter-nally rotated about 25˚ until the toes touched; and feet fixed together using strapping tape. Subjects were instructed to remain still and breathe normally during the scan. All DXA scans were analyzed by the same operator who followed the manufacturer’s instructions and used the pediatric mode and standardized cutoff for regional measurements [13]. The subregions were defined as the head, trunk, right arm, left arm, right leg, left leg. DXA measured regional and whole body composition, including LBMDXA, ALMDXA, FMDXA, and %BFDXA.
Statistical analysis
The statistical software package SPSS version 17.0 (SPSS Inc., Chicago, IL, USA) was used for data analysis. All data are reported as means± SDs. Analysis of variation (ANOVA) with Stu-dent’s independentt-test (two-sided) was applied for analysis of repeated measurements to compare the different testing methods. The statistical significance level was set atα = 0.05. Pearson’s product moment correlation and ordinary least products regression analysis were used to examine the relationship between the BIA and DXA and to determine the proportional bias and fixed bias [14]. The correlation coefficient (r) and determination coefficient (r2) from linear regression analysis were used to define the strength of linear association. The standard error of the estimate (SEE), a measure of the accuracy of predictions made with a linear regres-sion, was used to assess the statistical conformity of the two BIA methods.
To assess the degree of agreement between BIA and DXA measurements, three statistical techniques were used: the ICC, Lin’s concordance correlation (CCC) and Bland-Altman plot. The ICC coefficient (r1) (with two-way random and single measure) was used to assess the
agreement between BIA and DXA methods [15]. Anr1value 0.8 was considered a strong level
of agreement. The CCC coefficient (ρc) was used to assess how close the data from BIA and
DXA methods was about the line of best fit and also how far that line was from the 45-degree line through the origin [16]. Theρcand a concordance scale used including ratings of almost
perfect:ρc> 0.99; substantial: 0.99 ρc> 0.95; fair: 0.95 ρc 0.9; poor:ρc< 0.9) were used
to assess the concordance of the two BIA methods [17]. Bland-Altman plot with a regression analysis using ordinary least squares regression was used to display the difference between a pair of measurements against the mean of the pair [18]. Limits of agreement (LOA) were used to assess the agreement between two readings obtained by BIA and DXA on the same variable.
Results
A total of 150 children (72 boys and 78 girls) with a mean age of 9.3± 1.5 years were enrolled. Subject demographics and body composition estimates are shown inTable 1. There were no
significant differences in age, height or weight between boys and girls. However, the boys had significantly higher BMI compared with the girls (18.3± 4.3 in boys and 17.1 ± 3.0 in girls, p = 0.038). Based on DXA results, FM and %BF showed no significant difference between boys and girls whereas the boys had significantly higher LBM and ALM than the girls. For both boys and girls, all body composition results by BIA8MFand BIA8SFwere significantly different
from the results by DXA (P < 0.001,Table 1), except for LBM by BIA
8MF.
Table 2shows the Pearson product moment correlations coefficient (r) and the regression equation used to predict DXA results from BIA readings. There were strong linear correlations between the two BIA methods and DXA in the measurement of LBM, ALM, FM, and BF% (r 0.9 for all comparisons). However, there was a proportional bias and/or a fixed bias for each BIA measurement, except for LBM8MF. The scatter plots of body composition data by
BIA and DXA methods showed BIA underestimated FM and %FM relative to DXA in both genders (Figs1and2).
Pearson correlation was used to quantify the strength of linear association between two methods of measuring the same variable, and it should not be used to assess agreement between methods. Therefore, the agreement of BIA8SFand BIA8MFwith DXA was further
examined using three statistical techniques: ICC, CCC and Bland-Altman plot (Table 3). In general, an ICC value (r1) 0.8 is considered a strong level of agreement. This study showed
Table 1. Anthropometric characteristics and body composition measurements of Taiwanese children (age 6 to 12 years) determined by DXA (ref-erence method), BIA8MF, and BIA8MF.
Boys (n = 72) Girls (n = 78) Total (n = 150)
Mean SD Range Mean SD Range Mean SD Range
Age (years) 9.4 1.6 7.1–12.7 9.2 1.5 7.1–12.1 9.3 1.5 7.1–12.7 Height (cm) 138.0 11.0 114.7–164.9 137.5 11.3 112.2–159.1 137.7 11.1 112.2–164.9 Weight (kg) 35.6 11.9 19.2–73.1 33.0 9.5 19.3–60.4 34.2 10.8 19.2–73.1 BMI 18.3* 4.3 13.4–30.0 17.1 3.0 12.3–26.6 17.7 3.7 12.2–30.0 LBM (kg) DXA 24.3 5.7 15.2–40.6 22.4 5.5 13.6–38.1 23.3 5.7 13.6–40.6 BIA8MF 24.1 5.7 14.9–39.0 22.8 5.6 13.9–38.6 23.4 5.7 13.9–39.0 BIA8SF 26.4** 5.3 17.2–39.6 24.8** 5.4 16.1–40.0 25.6** 5.4 16.1–40.0 FM (kg) DXA 10.9 7.6 3.6–35.7 10.2 4.9 4.3–24.7 10.6 6.3 3.6–35.7 BIA8MF 9.6** 7.2 2.8–34.6 8.5** 4.4 3.0–21.8 9.1** 5.9 2.8–34.6 BIA8SF 7.9** 7.6 1.1–35.2 6.9** 4.2 2.0–21.0 7.4** 6.1 1.1–35.2 %BF (%) DXA 27.3 10.3 13.4–48.2 29.2 7.1 17.7–47.6 28.3 8.8 13.4–48.2 BIA8MF 24.3** 10.5 11.6–47.2 24.7** 7.1 14.4–42.9 24.5** 8.8 11.6–47.2 BIA8SF 18.5** 12.6 4.7–48.0 19.5** 6.7 9.3–36.9 19.0** 10.0 4.7–48.0 ALM (kg) DXA 10.4 2.9 5.3–19.0 9.4 2.6 5.3–16.5 9.9 2.8 5.3–19.0 BIA8MF 13.3** 3.7 7.5–22.8 12.4** 3.5 6.8–22.5 12.9** 3.6 6.8–22.8 BIA8SF 12.3** 3.5 6.9–22.3 10.8** 2.5 7.0–18.4 11.5** 3.1 6.9–22.3
Abbreviations: ALM, appendicular lean mass; BIA8SF, Tanita BC-418; BIA8MF, Inbody 230; BMI, body mass index; DXA, dual-energy X-ray absorptiometry;
FM, fat mass; LBM, lean body mass; SD, standard deviation; %BF: percent body fat. *P<0.05, by repeated-measures ANOVA with Student’s independent t-test; **P<0.01, by repeated-measures ANOVA with Student’s independent t-test doi:10.1371/journal.pone.0171568.t001
that all BIA parameters hadr1 0.9 except for LBM8SFin boys, which was 0.887, indicating a
strong agreement between the measures by BIA and DXA.
In general, the CCC values (ρc) for LBM, FM, and %BF were higher between BIA8MFand
DXA than between BIA8SFand DXA (Table 3), indicating a better agreement between BIA8MF
and DXA measures. In both sexes, theρcvalues for LBM, FM, and %BF were 0.9 between
BIA8MFand DXA, except for %BF8MFin girls (Table 2). Theρcfor LBM8MFmet the criteria for
substantial to perfect agreement (ρc> 0.95) whereas theρcfor FM8MFmet the criteria for fair
to substantial agreement (0.99 >ρc 0.9). For the %BF estimations, only theρcvalues
obtained by BIA8MFin the boys (ρc= 0.936) met the criteria for fair agreement with DXA and
the rest of the %BF estimations showed poor agreement (Table 3).
Bland-Altman plots were used to determine bias and LOA between BIA and DXA methods in boys (Fig 3) and girls (Fig 4). The LOAs were greater for the BIA8SFand DXA measurements
than for the BIA8MFand DXA measurements, except for the ALM measures in girls (Table 3).
Similar to the results by CCC, Bland-Altman analysis showed a good and clinically acceptable agreement between LBM measures by BIA8MFand DXA (LOA = -1.82 to 1.52 kg in boys and
LOA = -0.88 to1.63 kg in girls,Table 3).
In the human body, the FM is the total body weight minus LBM. Indeed, the LOAs of FM measures by BIA8MFand DXA (-3.21 to 0.55 kg in boys and -3.30 to -0.10 kg in girls,Table 3)
showed similar ranges to that of LBM but with different plus-minus sign (negative values in FM). In this study, the mean FM was about half of the LBM in children (Table 1) and thus, the degree of error was larger in FM estimation by BIA8MFand DXA compared with that in LBM.
Regarding %BF estimation, BIA8SFmeasurements underestimated %BF by 8.82% in boys
and 9.72% in girls, whereas the BIA8MFmeasurements underestimated %BF by 3.00% in boys
Table 2. Correlation of body composition estimates using Pearson product moment correlation and ordinary least products regression.
Method r a 95% CI b 95% CI Fixed bias Proportional bias SEE
Boys (n = 72) LBM8SF 0.971 -3.533 -5.188, -1.877 1.053 0.991, 1.115 Yes No 1.368 LBM8MF 0.989 0.354 -0.509, 1.217 0.991 0.957, 1.026 No No 0.839 FM8SF 0.986 3.248 2.813, 3.683 0.974 0.934, 1.014 Yes No 1.283 FM8MF 0.993 0.854 0.508, 1.200 1.050 1.020, 1.078 Yes Yes 0.876 %BF8SF 0.949 12.962 11.586, 14.339 0.776 0.715, 0.838 Yes Yes 3.285 %BF8MF 0.976 3.880 2.530, 5.229 0.964 0.913, 1.014 Yes No 2.256
ALM8SF 0.922 1.115 0.265, 2.178 0.748 0.673, 0.823 Yes Yes 1.116
ALM8MF 0.970 0.287 -0.337, 0.912 0.758 0.713, 0.804 No Yes 0.698 Girls (n = 78) LBM8SF 0.982 -2.354 -3.469, -1.329 0.996 0.952, 1.042 Yes No 1.043 LBM8MF 0.994 0.213 -0.373, 0.800 0.972 0.947, 1.002 No No 0.616 FM8SF 0.976 2.375 1.911, 2.840 1.132 1.074, 1.189 Yes Yes 1.064 FM8MF 0.991 0.822 0.492, 1.153 1.102 1.068, 1.137 Yes Yes 0.666 %BF8SF 0.897 10.609 8.407, 12.810 0.954 0.847, 1.061 Yes No 3.141 %BF8MF 0.925 5.336 3.638, 7.434 0.984 0.892, 1.077 Yes No 2.707
ALM8SF 0.956 -1.408 -2.190, -0.627 0.920 0.848, 0.989 Yes Yes 0.783
ALM8MF 0.974 0.383 -0.110, 0.876 0.727 0.668, 0.765 No Yes 0.596
Abbreviations: r, Pearson product moment correlation coefficient; a, b, coefficients in ordinary least products regression model: E(A) = a + b(B); a, (y axis) intercept; b, slope; fixed bias, if 95% confidence interval (CI) for a does not include 0; proportional bias, if 95% confidence interval (CI) for b does not include 1; SEE, standard error of the estimate.
and 4.48% in girls (Figs3dand4d). The LOAs in %BF estimation between BIA8MFand DXA
were clinically important. Even worse, there were larger LOAs in %BF estimation by BIA8SF
and DXA (-17.46 to -0.19% in boys and -15.99 to -3.45% in girls,Table 3).
Theρcvalue for ALM estimated by BIA8SFwas 0.770 in boys and 0.828 in girls, and theρc
value for ALM estimated by BIA8MFwas 0.671 in boys and 0.635 in girls, all of which were
con-sidered poor agreement (Table 3). In agreement with CCC, Bland-Altman analysis showed a poor agreement with clinically importance between ALM estimations by BIA and DXA in both genders (Table 3).
Discussion
This study compared the estimates of body composition obtained from multi-segment BIA8SF
and BIA8MFwith DXA measurements in primary school children from Taiwan. Pearson
prod-uct moment correlation was used to test the linear association whereas ICC, CCC and Bland-Altman Plot were used to test agreement between BIA and DXA results. So far, there is still a debate about which method is the best for assessing agreement between two instruments. The ICC and CCC are scaled agreement indices depending on the measurement range, and there-fore they are easy to summarize but hard to interpret [19]. In contrast, bias and LOAs (Bland-Altman plot) are unscaled indices based on the original unit and interpretation of the
Fig 1. Correlation between dual-energy X-ray absorptiometry results and estimates of body composition in boys obtained with either BIA8SFor BIA8MF. (a) LBM: BIA8SF: r2= 0.940, BIA8MF: r2= 0.979 (b) ALM: BIA8SF: r2= 0.858, BIA8MF: r2= 0.944 (c) FM: BIA8SF: r2
= 0.940, BIA8MF: r2= 0.979 (d) %BF: BIA8SF: r2= 0.898BIA8MF: r2= 0.951.
agreement requires prior knowledge of the measurement variables [20]. Since these methods all have some disadvantages, we have used more than one statistical method to assess agree-ment between two instruagree-ments in this study.
The LBM estimates by BIA8MFand DXA were in high agreement for both genders using all
statistical methods in this study. Therefore, BIA8MFand DXA were interchangeable test
meth-ods for the measurement of LBM in children. However, the FM estimates showed fair to sub-stantial agreement between BIA8MFand DXA by CCC but clinically important differences by
Bland-Altman plots. One possible explanation for the discrepancy in the degree of agreement may due to the fact that CCC was scaled relative to the between-subject variability and the large FM range in our subjects produced a relatively highρcvalue. In contrast, Bland-Altman
analysis was not dependent on between-subject variability such that it was easier to identify the error between the two methods.
Except for LBM estimates, the remainder of the BIA measurements showed strong linear correlated (but with clinically significant errors) with the gold standard method, DXA. Talma et al. [21] reported similar findings in a review article. Most previous BIA validation studies reported high precision using the BIA models but did not use a reference method to measure the accuracy of BIA estimates [22]. In addition to linear regression and ICC, we also per-formed Bland-Altman analysis and determined CCC to rigorously assess the statistical
Fig 2. Correlation between dual-energy X-ray absorptiometry results and estimates of body composition in girls obtained with BIA8SFor BIA8MF. (a) LBM: BIA8SF: r
2 = 0.964, BIA8MF: r 2 = 0.987 (b) ALM: BIA8SF: r 2 = 0.915, BIA8MF: r 2 = 0.951 (c) FM: BIA8SF: r 2 = 0.953, BIA8MF: r2= 0.981 (d) %BF: BIA8SF: r2= 0.802, BIA8MF: r2= 0.964.
consistency of body composition estimates from BIA relative to DXA. Our results indicated clinically important errors in FM and %BF estimated by both BIA devices which may limit their applicability to body composition measurements at an individual level in children, even though ther and r1values were high between both BIA methods and DXA. It is worth noting
that although ICC is a popular test to compare the results between two methods, there is still a debate about the use of ICC in assessment agreement [23,24].
We also compared both BIA8SFand BIA8MFmodels in children with a wide range of body
fat composition, using DXA as the gold standard. Although the estimates from both BIA devices and DXA showed strong linear correlations, the correlation coefficients and agree-ments were higher for BIA8MFcompared with BIA8SF. In general, the BIA devices (especially
the BIA8SF) overestimated LBM and underestimated FM. In addition, the LOAs were larger
and the biases were greater for BIA8SFmeasurements compared with BIA8MFmeasurements,
except for ALM in girls. The CCC analysis also indicated better agreements in measurements of LBM, FM, and %BF for the BIA8MFin both sexes. These results confirm the findings of
Kriemler et al. [25] that BIA8MFis superior to BIA8SFin pediatric body composition analysis.
In our study, both BIA8SFand BIA8MFunderestimated FM and %BF in children who had
large or small amounts of body fat. Additionally, BIA8SFhad a fixed bias or proportional bias
in all components of body composition. Talma et al. [21], in their systematic review, indicated that BIA provided inconsistent results, depending on the reference method used. A literature review of validation studies for the Tanita BC-418 system in children also showed inconsistent results similar to our findings, whereas other studies had results which contradicted our find-ings. For example, Pietrobelli et al. [26] showed a perfect linear correlation between body com-position parameters measured by the Tanita BC-418 system and DXA in subjects aged 6–64 years. However, they did not perform agreement analysis, and had a small sample size and
Table 3. Agreement between bioelectrical impedance analysis and dual-energy X-ray absorptiometry.
Method Bland-Altman Plot CCC (ρc) ICC (r1)
Bias Limit of agreement Function p
Boys (n = 72) LBM8SF 2.12 -0.65 to 4.90 y = 0.082 x + 4.208 0.005 0.900 0.887 LBM8MF -0.15 -1.82 to 1.52 y = -0.002 x−0.096 0.902 0.989 0.943 FM8SF -3.05 -5.63 to -0.47 y = 0.012 x−3.156 0.571 0.911 0.973 FM8MF -1.35 -3.21 to 0.55 y = -0.055 x−0.763 <0.0001 0.975 0.977 %BF8SF -8.82% -17.46 to -0.19% y = 0.205 x−13.526 <0.0001 0.717 0.992 %BF8MF -3.00% -7.54 to 1.55% y = 0.013 x−3.335 0.620 0.936 0.986 ALM8SF 1.87 -0.97 to 4.71 y = 0.216 x−0.585 <0.0001 0.770 0.989 ALM8MF 2.93 0.69 to 5.17 y = 0.248 x + 0.016 <0.0001 0.671 0.972 Girls (n = 78) LBM8SF 2.44 0.37 to 4.52 y = -0.015 x + 2.790 0.500 0.890 0.990 LBM8MF 0.37 -0.88 to 1.63 y = 0.018 x−0.076 0.126 0.991 0.994 FM8SF -3.29 -5.68 to -0.90 y = -0.150 x−2.008 <0.0001 0.763 0.992 FM8MF -1.70 -3.30 to -0.10 y = -0.107 x−0.691 <0.0001 0.923 0.970 %BF8SF -9.72% -15.99 to -3.45% y = -0.065 x−8.149 0.229 0.445 0.989 %BF8MF -4.48% -8.50 to -0.46% y = -0.001 x−4.512 0.969 0.798 0.979 ALM8SF 1.42 -0.14 to 2.97 y = -0.045 x + 1.877 0.192 0.828 0.981 ALM8MF 3.01 0.74 to 5.28 y = 0.295 x−0.208 <0.0001 0.635 0.953
Abbreviation: CCC, Lin’s concordance correlation coefficient;ρc, CCC coefficient; ICC, intra-class correlation; r1, ICC coefficient
wide age range. Some studies showed that the Tanita BC-418 underestimated FM in obese children compared with other reference methods [27,28]. Shaikh et al. [28] reported a strong linear correlation between FM determined by the Tanita BC-418MA and DXA in obese boys aged 11.0± 0.53 years; however, the BIA system underestimated %BF, and the LOA in %BF was -3.8 to 15.4%. Haroun et al. [27] examined obese subjects (between 5–22 years of age) and found that the Tanita BC-418 underestimated FM by 3.5 kg in males and 3.6 kg in females, compared with the isotope dilution method. In contrast, Prins et al. [29] showed the Tanita BC-418MA system overestimated %BF in normal-weight Gambian children aged 5–16 years relative to the isotope dilution method.
We found that LBM estimates between BIA (BIA8SFand BIA8MF) and DXA were in fair to
substantial agreement whereas ALM estimates between BIA and DXA showed poor agree-ment. Few previous studies have used eight-electrode multi-frequency BIA devices (i.e. the Inbody-230) for estimates of body composition in children. Kriemler et al. [25] used a different BIA8MFdevice (Inbody 3.0, Biospace, Seoul, Korea) in 6 years-old and found no fixed bias or
proportional bias in FFM or ALM relative to measurements from DXA. Jensky-Squires et al. [30] used the Inbody-320 (Biospace, Seoul, Korea) to estimate %BF in children between 10–17 years of age relative to underwater weighing, and found significant differences in girls but not boys. Lim et al. [31] used the Inbody 720 (Biospace, Seoul, Korea) to estimate FFM, FM, and %
Fig 3. Bland-Altman plots with linear regression analysis of dual-energy X-ray absorptiometry results vs. BIA8SFand BIA8MF estimates of body composition in boys.
BF in healthy children between 6–18 years of age and reported a high precision relative to DXA results. In their study, the LOA in %BF was -2.2± 6.1%, which was far less than ours.
BIA is primarily designed to estimate FFM, and the FFM prediction equations were devel-oped using a reference method, such as DXA and/or isotope dilution. Variables in the regres-sion equations may include height, weight, age, sex, race, and other factors [7]. Therefore, the established FFM equations may not applicable to all pediatric populations such as our pediatric populations [32,33]. Body hydration status can also influence FFM calculation from BIA mea-surements. Most BIA prediction equations assume that the FFM consists of 73% water. How-ever, although the water content of FFM is about 73% in adults, it is greater in children [22]. Therefore, a BIA prediction equation developed for adults could overestimate FFM in chil-dren. Moreover, hydration status changes as a child develops [34]. Therefore, an equation developed for school-aged children may not be accurate for adolescents. These major limita-tions of the BIA method remain unresolved.
Conclusion
For all children, LBM measures using the BIA8MFwere precise and accurate whereas clinically
significant errors occurred in both FM and %BF estimates. The BIA8SFand BIA8MFboth
underestimated FM and %BF in children. For the estimates of ALM, both BIA devices showed
Fig 4. Bland-Altman plots with linear regression analysis of dual-energy X-ray absorptiometry results vs. BIA8SFand BIA8MF estimates of body composition in girls.
poor agreement with DXA. Thus, the body composition results obtained using the inbuilt equations of the BIA8SFand BIA8MFshould be interpreted with caution, and high quality
vali-dation studies in specific subgroups children are required prior to field research.
Author contributions
Conceptualization: LWL YSL HKL PLH YYC CCC KCH. Formal analysis: LWL KCH.
Funding acquisition: LWL CCC. Investigation: LWL YSL PLH. Methodology: LWL YSL.
Project administration: LWL CCC.
Resources: LWL YSL HKL PLH YYC CCC KCH. Supervision: CCC KCH.
Validation: LWL YSL HKL PLH YYC CCC KCH. Visualization: LWL KCH.
Writing – original draft: LWL KCH.
Writing – review & editing: LWL YSL HKL PLH YYC CCC KCH.
References
1. Wells JC, Fewtrell MS. Measuring body composition. Archives of disease in childhood. 2006; 91 (7):612–7. Epub 2006/06/23. doi:10.1136/adc.2005.085522PMID:16790722
2. Khalil SF, Mohktar MS, Ibrahim F. The theory and fundamentals of bioimpedance analysis in clinical status monitoring and diagnosis of diseases. Sensors (Basel). 2014; 14(6):10895–928. Epub 2014/06/ 21.
3. Kushner RF. Bioelectrical impedance analysis: a review of principles and applications. Journal of the American College of Nutrition. 1992; 11(2):199–209. Epub 1992/04/01. PMID:1578098
4. Organ LW, Bradham GB, Gore DT, Lozier SL. Segmental bioelectrical impedance analysis: theory and application of a new technique. J Appl Physiol (1985). 1994; 77(1):98–112. Epub 1994/07/01.
5. Kim J, Shen W, Gallagher D, Jones A Jr., Wang Z, Wang J, et al. Total-body skeletal muscle mass: esti-mation by dual-energy X-ray absorptiometry in children and adolescents. The American journal of clini-cal nutrition. 2006; 84(5):1014–20. Epub 2006/11/10. PMID:17093152
6. Kim J, Wang Z, Heymsfield SB, Baumgartner RN, Gallagher D. Total-body skeletal muscle mass: esti-mation by a new dual-energy X-ray absorptiometry method. The American journal of clinical nutrition. 2002; 76(2):378–83. Epub 2002/07/30. PMID:12145010
7. Kyle UG, Bosaeus I, De Lorenzo AD, Deurenberg P, Elia M, Gomez JM, et al. Bioelectrical impedance analysis—part I: review of principles and methods. Clin Nutr. 2004; 23(5):1226–43. Epub 2004/09/24. doi:10.1016/j.clnu.2004.06.004PMID:15380917
8. Gudivaka R, Schoeller DA, Kushner RF, Bolt MJ. Single- and multifrequency models for bioelectrical impedance analysis of body water compartments. J Appl Physiol (1985). 1999; 87(3):1087–96. Epub 1999/09/14.
9. Dittmar M, Reber H. New equations for estimating body cell mass from bioimpedance parallel models in healthy older Germans. American journal of physiology Endocrinology and metabolism. 2001; 281(5): E1005–14. Epub 2001/10/12. PMID:11595657
10. Pietrobelli A, Morini P, Battistini N, Chiumello G, Nunez C, Heymsfield SB. Appendicular skeletal mus-cle mass: prediction from multiple frequency segmental bioimpedance analysis. European journal of clinical nutrition. 1998; 52(7):507–11. Epub 1998/07/31. PMID:9683333
11. Body Composition Analyzer BC-418 Instruction Manual: Tanita corporation; [cited 2017 Jan 9].http:// www.tanita.com/es/bc-418/.
12. InBody User’s Manual: Biospace Co., Ltd; 1996 [cited 2017 Jan 9].https://cdn.shopify.com/s/files/1/ 0832/8945/files/InBody230_User_sManual.pdf?7633250021353425932.
13. Centers for Disease Control and Prevention. National Health and Nutrition Examination Survey (NHANES). Body Composition Procedures Manual 2013 [cited 2017 Jan 9].https://www.cdc.gov/nchs/ data/nhanes/nhanes_13_14/2013_Body_Composition_DXA.pdf.
14. Ludbrook J. Statistical techniques for comparing measurers and methods of measurement: a critical review. Clinical and experimental pharmacology & physiology. 2002; 29(7):527–36. Epub 2002/06/13. 15. Shrout PE, Fleiss JL. Intraclass correlations: uses in assessing rater reliability. Psychological bulletin.
1979; 86(2):420–8. Epub 1979/03/01. PMID:18839484
16. Lin LI. A concordance correlation coefficient to evaluate reproducibility. Biometrics. 1989; 45(1):255– 68. Epub 1989/03/01. PMID:2720055
17. McBride GB. A proposal for strength-of-agreement criteria for Lin’s Concordance Correlation Coeffi-cient. 2005 NIWA Client Report: HAM2005-062.
18. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet. 1986; 1(8476):307–10. Epub 1986/02/08. PMID:2868172
19. Barnhart HX, Haber MJ, Lin LI. An overview on assessing agreement with continuous measurements. Journal of biopharmaceutical statistics. 2007; 17(4):529–69. Epub 2007/07/07. doi:10.1080/ 10543400701376480PMID:17613641
20. Giavarina D. Understanding Bland Altman analysis. Biochemia medica. 2015; 25(2):141–51. Epub 2015/06/26. doi:10.11613/BM.2015.015PMID:26110027
21. Talma H, Chinapaw MJ, Bakker B, HiraSing RA, Terwee CB, Altenburg TM. Bioelectrical impedance analysis to estimate body composition in children and adolescents: a systematic review and evidence appraisal of validity, responsiveness, reliability and measurement error. Obesity reviews: an official jour-nal of the Internatiojour-nal Association for the Study of Obesity. 2013; 14(11):895–905. Epub 2013/07/16. 22. Silva AM, Fields DA, Sardinha LB. A PRISMA-driven systematic review of predictive equations for
assessing fat and fat-free mass in healthy children and adolescents using multicomponent molecular models as the reference method. Journal of obesity. 2013; 2013:148696. Epub 2013/07/12. doi:10. 1155/2013/148696PMID:23844282
23. Bland JM, Altman DG. A note on the use of the intraclass correlation coefficient in the evaluation of agreement between two methods of measurement. Computers in biology and medicine. 1990; 20 (5):337–40. Epub 1990/01/01. PMID:2257734
24. Zaki R, Bulgiba A, Ismail R, Ismail NA. Statistical methods used to test for agreement of medical instru-ments measuring continuous variables in method comparison studies: a systematic review. PloS one. 2012; 7(5):e37908. Epub 2012/06/05. doi:10.1371/journal.pone.0037908PMID:22662248
25. Kriemler S, Puder J, Zahner L, Roth R, Braun-Fahrlander C, Bedogni G. Cross-validation of bioelectrical impedance analysis for the assessment of body composition in a representative sample of 6- to 13-year-old children. European journal of clinical nutrition. 2009; 63(5):619–26. Epub 2008/02/21. doi:10. 1038/ejcn.2008.19PMID:18285806
26. Pietrobelli A, Rubiano F, St-Onge MP, Heymsfield SB. New bioimpedance analysis system: improved phenotyping with whole-body analysis. European journal of clinical nutrition. 2004; 58(11):1479–84. Epub 2004/05/13. doi:10.1038/sj.ejcn.1601993PMID:15138459
27. Haroun D, Croker H, Viner RM, Williams JE, Darch TS, Fewtrell MS, et al. Validation of BIA in obese children and adolescents and re-evaluation in a longitudinal study. Obesity (Silver Spring). 2009; 17 (12):2245–50. Epub 2009/04/18.
28. Shaikh MG, Crabtree NJ, Shaw NJ, Kirk JM. Body fat estimation using bioelectrical impedance. Hor-mone research. 2007; 68(1):8–10. Epub 2007/01/11. doi:10.1159/000098481PMID:17213729
29. Prins M, Hawkesworth S, Wright A, Fulford AJ, Jarjou LM, Prentice AM, et al. Use of bioelectrical imped-ance analysis to assess body composition in rural Gambian children. European journal of clinical nutri-tion. 2008; 62(9):1065–74. Epub 2007/07/12. doi:10.1038/sj.ejcn.1602830PMID:17622262
30. Jensky-Squires NE, Dieli-Conwright CM, Rossuello A, Erceg DN, McCauley S, Schroeder ET. Validity and reliability of body composition analysers in children and adults. The British journal of nutrition. 2008; 100(4):859–65. Epub 2008/03/19. doi:10.1017/S0007114508925460PMID:18346304
31. Lim JS, Hwang JS, Lee JA, Kim DH, Park KD, Jeong JS, et al. Cross-calibration of multi-frequency bio-electrical impedance analysis with eight-point tactile electrodes and dual-energy X-ray absorptiometry for assessment of body composition in healthy children aged 6–18 years. Pediatrics international: offi-cial journal of the Japan Pediatric Society. 2009; 51(2):263–8. Epub 2009/05/02.
32. Lewy VD, Danadian K, Arslanian S. Determination of body composition in African-American children: validation of bioelectrical impedence with dual energy X-ray absorptiometry. Journal of pediatric endo-crinology & metabolism: JPEM. 1999; 12(3):443–8. Epub 2000/05/23.
33. Clasey JL, Bradley KD, Bradley JW, Long DE, Griffith JR. A new BIA equation estimating the body com-position of young children. Obesity (Silver Spring). 2011; 19(9):1813–7. Epub 2011/06/18.
34. Wells JC, Fewtrell MS, Davies PS, Williams JE, Coward WA, Cole TJ. Prediction of total body water in infants and children. Archives of disease in childhood. 2005; 90(9):965–71. Epub 2005/08/23. doi:10. 1136/adc.2004.067538PMID:16113134
