Utility of Specific Bioelectrical Impedance Vector Analysis for the assessment of body composition in children

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.


Introduction
There is increasing interest in the assessment of body composition in children, for several reasons. Body composition measurements could inform clinical diagnosis, improve routine management, help determine nutritional and fluid requirements as well as some therapeutic doses, and assess longer-term cardio-metabolic risk [1]. However, obtaining accurate measurements is challenging in sick or malnourished children, who often cannot comply with demanding protocols. For many decades, routine body composition assessment was restricted to simple anthropometry, such as body mass index (BMI), skinfold thicknesses and body circumferences. Recently, more sophisticated methods have become available, including air-displacement plethysmography, dual-energy X-ray absorptiometry and isotope dilution [2,3], but remain restricted to specialist research centres. There is still a need for simpler methods that can be widely used, especially in community studies [2]. In this context, bioelectrical impedance analysis (BIA) has long attracted interest. The method involves passing a very small imperceptible current between different parts of the body and measuring the resistance of the tissues. Raw data are quick and easy to collect, though individuals must stay still in standardised position and relaxed state for a few seconds.
Conventionally, whole-body impedance (Z) is measured between the wrist and the ankle. The height (H)-adjusted impedance index (H 2 /Z) is then a strong predictor of either total body water (TBW) or fat-free mass (FFM) [2]. Resistance is more often measured in place of Z; though similar in magnitude, it is more closely related to TBW (see below). However, associations of H 2 /Z(or R) with TBW or FFM vary between populations and population-specific calibration studies are recommended. Even then, subtle variation in traits such as body proportions (e.g., the ratio of limb to trunk length), maturation state and ethnic ancestry result in significant random error in individuals [4][5][6][7][8]. Since FM is typically the smaller component of weight, random error on FFM is relatively greater when propagated to FM. Several solutions to this predicament have been suggested, such as incorporating more predictors in calibration equations, conducting segmental analysis or using different bioelectrical frequencies to predict different fluid compartments, but none of these approaches has substantially reduced predictive error [9][10][11].
An alternative approach, developed by Piccoli et al and known as bioelectrical impedance vector analysis (BIVA), divides Z into its components, resistance (R) and reactance (Xc), normalizing each for H [12]. Based on bioelectrical theory, R is expected to correlate negatively with body fluids, whereas Xc should correlate positively with body cell mass [13]. On this basis, if the data are plotted on 'R/H-Xc/H' graphs, data from a population will form an ellipse as expected for bivariate data, where one diagonal axis represents variability in hydration, and the orthogonal axis variability in body cell mass, closely correlated with FFM [13]. A key advantage of Piccoli's approach is that no predictive equations are required, however the resulting data are both qualitative and semiquantitative, and still require some form of processing to aid interpretation. For example, BIVA traits vary with age, which may in part by due to changes in body size [14]. As one solution to this, we previously published BIVA reference data from children and adolescents aged 5 to 20 years, allowing the use of age-and sex-standardized z-scores [15].
Recently, studies of disease states have supported several assumptions of BIVA theory [16,17], however in healthy children the findings were only partially supportive [15].
While BIVA variables correlated as expected with hydration, they did not correlate with FFM.
A new variant known as specific BIVA (BIVA specific ), aims to improve the correlation with body composition by adjusting R and Xc not only for height, but also for body cross-sectional area [18,19]. This addresses fundamental bioelectrical theory, since according to Ohm's law, the resistance of a conductor (e.g., a cylinder) to a current varies proportionally with its length but inversely with its cross-sectional area [18,19]. Research in adults has shown that BIVA specific measurements correlate strongly with % fat as opposed to FFM [18].
We therefore aimed to evaluate specific BIVA data in a large dataset of body composition in children from a wide range of nutritional status, which we previously analysed for our assessment of conventional BIVA. We also conducted exploratory analyses, combining both conventional BIA and specific BIVA approaches to see if they contributed independently to the prediction of body composition.

Methods
The data has been described in detail previously [15], and came from two prior studies conducted by our group, both approved by the Ethical Committee of UCL Institute of Child Health and Great Ormond Street Hospital. Informed consent was obtained from all participants and/or their parents as appropriate. For this analysis, we included children of European ancestry only, as ethnicity has been associated with variability in fat and lean distribution [6,20] and the loci of BIVA ellipses [21].

Participants
Most individuals were from a study of healthy children/adolescents aged 4 to 20 years, with no BMI exclusion criteria except that they could not be attending a weight loss clinic, and they must not have any disease that might affect growth and development. Baseline data from obese children aged 7 to 14 years participating in weight loss intervention studies were also used. In combination, the two datasets cover a wide range of BMI [22,23].

Data collection
Anthropometry and body composition were measured as described previously, with FFM and FM calculated using the 4-component model [24]. Single-frequency BIA was conducted at 50 kHz (Quadscan 4000 instrumentation; Bodystat, UK). This frequency is proposed to maximise signal-to-noise ratio and minimise frequency-dependent errors and variability of electric flow paths [25,26], though the optimal frequency also varies between individuals and by age [27]. Participants lay supine on a non-conducting couch. Disposable EKG-style Ag/AgCl gel electrodes were attached in standard tetrapolar manner to left hand and foot [28]. Z, R, Xc and PA were recorded in duplicate, and the average used in analyses. The device was regularly calibrated, and on all occasions the device was within the manufacturer's specifications.
As usual in the conventional BIVA (BIVA conventional ) approach, R and Xc were standardised for height (H) and expressed as R/H and Xc/H in ohm (Ω)/m [12]. Prior to analysis, we excluded individuals with PA>8.0 (values in healthy people range between 5° and 7°, hence allowing for measurement error, values above 8° were considered implausible; n=14 excluded) [29], as well as those with poor repeatability (exclusion criteria were duplicates >0.5 for PA, and ≥6.0 for R/H and Xc/H; n=25 excluded). In those data retained for analysis, technical error of the mean calculated using the formula of Ulijaszek and Kerr [30] was 1.9 ohm for Z and R and 1.1 ohm for Xc.
The new data incorporated in this analysis were body girths, which were available for all but one of the subjects in the previous analysis. Girths were measured for the mid-upper arm, waist and maximal calf, using a non-stretchable tape. Technical error of the mean for girth data in our research centre is 0.2 cm for waist and 0.1 cm for arm and calf girth. The raw data are available as Supplementary online Data.

Data processing and statistical analysis
Three age groupings were derived, broadly representing pre-pubertal (4-9 years), pubertal (10-14 years) and post-pubertal (15-20 years) individuals. Assessment of nutritional status was based on UK BMI z-scores, using cut-offs of <-2 to define thinness, >1 to define overweight and >2 to define obesity.
For conventional BIA, the impedance index was calculated as the square of height divided by Z (HT 2 /Z) in cm 2 /ohms. For specific BIVA (BIVA specific ), following the approach of Buffa and colleagues [18], cross-sectional areas (A) of the arm, torso and leg were first calculated as follows: These were then summed, again as recommended [18], to give a whole body area correction factor as follows: We obtained a value for length from height, again as recommended [18]: In BIVA specific , R and Xc are then multiplied by the correction factor, A/L, to give R specific and Xc specific respectively.
Graphs were plotted with sex-specific LOESS lines, fitted with 75% span, for BIVA conventional and BIVA specific outcomes against age, and for FFM and fat mass against BIVA conventional and BIVA specific parameters. Correlations between conventional or specific BIVA parameters and body composition outcomes were calculated. Multiple regression models were used to test independent associations of conventional BIA and BIVA conventional and BIVA specific parameters with body composition outcomes, adjusting for age and sex (males coded 1, females coded 2).
We developed a series of multiple regression models, intended to reveal the differing associations of anthropometry and different BIA approaches with the two body composition outcomes. We evaluated the different models in terms of the proportion of variance in the outcome explained (adjusted r 2 value, which aids comparison across models) and the standard error of the estimate (SEE) in individuals. To aid comparisons across models, we also provide the t-statistic for each individual predictor, and the overall F-statistic of each model.
Baseline models included only age and sex, in order to help understand how the addition of any further anthropometric or BOA parameters improved the accuracy of predicting outcomes. We then developed models that introduced only anthropometric parameters (either weight, or the three girths). Subsequent models introduced either conventional BIA parameters (HT 2 /Z), or conventional BIVA parameters (R/H, Xc/H), or specific BIVA parameters (R specific , Xc specific ). Finally, we developed 'combined models' incorporating both conventional BIA and specific BIVA parameters, as well as testing the addition of weight. All graphs and analyses were run in IBM SPSS Statistics, version 24.

Results
After data cleaning, full data were available for 281 individuals, comprising 130 boys and 151 girls. The average age was 11.8 (SD 3.7) years, range 4.2 to 19.9 years. There was no difference in average age between the sexes.

Discussion
The conventional approach to predicting body composition from BIA relies on the close association of height-adjusted resistance or impedance with body components that conduct electricity, the most obvious of which are TBW or FFM. These associations are strong in any given population, though the slope varies between populations according to age, maturation state and ethnicity [6,8]. Using carefully designed calibration studies, where children are sampled in equal numbers across a wide range of nutritional status, the SEE in individuals can be reduced to 1 kg of FFM [31], meaning that predicted values lie within ±2kg of the 'true' value. In our study, the best SEE value using conventional BIA was slightly greater (2.46 kg of FFM), but this is due in part to our including a very wide range of both age and BMI, and we expect that better accuracy could be obtained using our approach in more homogeneous populations.
Our aim was to see if we could improve the prediction of FFM, using approaches based on BIVA. Expressing BIA data as height-adjusted vectors, using the BIVA conventional approach, did not show strong associations with FFM, while for FM the approach was no better than that using conventional BIA. Indeed, no simple BIA model performed better for FM than a model containing only age, sex and three body girths. It has previously been recognised that BIVA conventional parameters remain strongly associated with body shape, as impedance of body components is influenced by both cylinder length and cross-sectional area [18,19]. We therefore tested a new variant of BIA [18,19], which adjusts impedance for body crosssectional area as well as length, with the aim of addressing more effectively variability in body morphology.
We first showed that two girths (arm and waist) were significant predictors of FM but not FFM, whereas the reverse scenario was apparent for calf girth. This supports the notion that incorporating upper body girths into regression models addresses variability in fatness. This is similar to data from adults, where for example girths tend to show higher correlations with FM than skeletal muscle mass, especially in females, though the arm is an exception to this pattern [32]. Consistent with that, we found overall that BIVA specific parameters were more accurate predictors of FM than BIVA conventional parameters, though this pattern was Our study shows for the first time in healthy children covering a wide range of nutritional status that BIVA specific performs more poorly than conventional BIA in predicting FFM, but performs substantially better in predicting FM. This suggests that correcting bioelectrical data for cylinder cross-sectional area as well as height improves the correlation with body composition outcomes, as shown previously for body fatness in adults [18]. Nonetheless, this approach still could not match the accuracy of conventional BIA for predicting FFM, and it was only by combining the two approaches that the prediction of both tissue masses improved.
A combined approach to BIA would be easy to operationalize without greatly complicating the quick and simple protocol for data collection, even in young patients, which is a key strength of BIA as a technique. Simply by adding in the 3 additional girths, both BIVA conventional and BIVA specific parameters could be obtained, and the same data also allow other analytical approaches to be used, including the assessment of phase angle as a proxy for cellular health [29], and Piccoli's graphical approach which provides information on hydration status [13,33].
The strengths of our study include the high-quality measurements of body composition obtained using the four-component model, the relatively large sample size, and the wide range of body composition and BMI assessed. Restricting the analysis to children of European ancestry avoided the potential complication of ethnic variability in shape.
However, there are also some limitations to our analysis. At this stage, we do not know how much ethnicity might influence BIVA specific , as has already been assessed for conventional BIA in the pediatric age range [6][7][8]. Second, we have not considered children aged under 5 years, though this population has particular need of simpler protocols. Third, we have not yet addressed patients, in whom the need for simpler protocols is again particularly important. Moreover, we do not know how perturbations of body composition associated with illness (dehydration, wasting, oedema) might affect the ability of BIVA specific parameters to index body fatness. Further work in younger children and patients is therefore required to fully appreciate the potential of this BIA variant for clinical assessment.
In summary, our analysis shows for the first time that conventional BIA and the new BIVA specific approach make independent and additive contributions to predicting body composition variability in healthy children and adolescents across wide range of age and BMI. Further work may extend this approach to patients, and may potentially improve the accuracy of BIA for predicting body composition variability in those most in need of such measurements.

Conflict of interest
The Quadscan BIA instrument used in this study was received gratis from Bodystat. The manufacturer had no input into the design, conduct or analysis of the study. Other authors declare no conflicts of interest.

CRediT authorship contribution statement
The body composition studies that provided data were conceived and conducted by JCW, MF and JEW. JW conceived the analysis plan, derived the specific BIVA variables and ran the statistical analyses. All authors discussed the analyses and contributed to writing the manuscript.    Z -impedance, R -resistance, Xc -reactance, H -height R specific and X specific -R and Xc adjusted for both height and cross-sectional area SE -standard error, SEE -standard error of the estimate