ABSTRACT
Background
South African children face a double burden of malnutrition from undernutrition and rising obesity. Simple, accurate methods to estimate fat‐free mass, a key health indicator, are needed, as bioelectrical impedance analysis is limited by cost, availability and lack of local validation.
Objectives
To develop and validate prediction equations for fat‐free mass using simple anthropometric measurements in children aged 6–9 years.
Methods
In this cross‐sectional study, anthropometric and bioelectrical impedance data were obtained from 117 children. Bioelectrical impedance‐derived fat‐free mass was used as reference in multivariable regression models. Four equations were externally validated in 75 Black prepubertal children, using dual‐energy X‐ray absorptiometry‐derived fat‐free mass as standard. Relationships, mean differences, and agreement were assessed using Pearson's correlation, independent t‐tests and Bland–Altman plots, respectively.
Results
Fourteen prediction equations, containing five to nine variables, were developed (R 2 range: 0.88–0.92) in the sample of children (51% Black; 55% boys; 7.9 ± 0.8 years). Four equations were strongly correlated with dual‐energy X‐ray absorptiometry‐derived fat‐free mass (r > 0.95; p < 0.001) in the validation sample (8.5 ± 1.3 years), and three yielded estimates with acceptable agreement (mean difference: 0.16–0.94 kg; limit of agreement: ±5 kg).
Conclusion
Fat‐free mass of prepubertal children can be predicted using simple anthropometric measurements, allowing assessment of body composition in low‐resource settings.
Keywords: anthropometry, bioelectrical impedance analysis, body composition, children, dual energy X‐ray absorptiometry, fat‐free mass
Brief Points
- What is already known on this topic
-
○Reducing the double burden of malnutrition requires promoting and monitoring healthy body composition in children.
-
○Fat‐free mass is a key indicator of child health and nutritional status.
-
○Bioelectrical impedance analysis is commonly used to estimate fat‐free mass but lacks validation and accessibility in low‐resource settings.
-
○
- What this paper adds
-
○Simple anthropometric measurements can accurately estimate fat‐free mass of children.
-
○Anthropometric equations are practical for use in low‐resource settings without dual‐energy X‐ray absorptiometry or bioelectrical impedance analysis.
-
○New anthropometric equations to predict fat‐free mass in South African children aged 6–9 years.
-
○
1. Introduction
In sub‐Saharan Africa, there is a continuation of underweight coupled with an overweight/obesity transition in school‐aged children [1]. This double burden of malnutrition is characterised by undernutrition (wasting, stunting and micronutrient deficiency or insufficiency) and overnutrition (overweight and obesity) [2]. The prevalence of overweight among preschool children in Southern Africa is 8.5%, ranging from 4.6% to 13.5% [3], according to the International Obesity Task Force (IOTF) based criteria [4]. However, this estimate is complicated by and differs significantly across different body mass index (BMI) cut‐off references used by the World Health Organization (WHO), Centres for Disease Control, and IOTF [3]. Although BMI‐for‐age z‐scores are an accepted method for identifying malnutrition in children over 5 years of age, they may not be useful indicators of fat mass (FM) [5] and may severely underestimate the prevalence of overweight and obesity compared with body composition assessment using stable isotope dilution techniques [6].
Body composition assessment to evaluate the double burden of malnutrition is important for assessing the risk of non‐communicable diseases (NCDs). Childhood undernutrition hinders the development of lean muscle mass, whereas excessive catch‐up weight or fat gain in early life and beyond is associated with elevated adiposity, especially abdominal adiposity, in later years. Such sub‐optimal body composition affects cardiometabolic homeostasis and increases the risk of NCDs [7]. Strategies to lower the double burden of malnutrition should focus on achieving or maintaining an ideal body composition, making the assessment of body composition an imperative part of monitoring the nutritional status of children [8].
Bioelectrical impedance analysis (BIA) reliably predicts fat‐free mass (FFM) in children when appropriate equations are used, but limited access to equipment in low‐resource and field settings often necessitates simpler anthropometric models [9, 10]. Most existing equations focus on fat mass [11] rather than FFM and are largely impedance‐based [12, 13], with few relying solely on anthropometry. Anthropometry‐based FFM equations for young healthy children are mainly from Brazil [14, 15], the USA [16] and the UK [17], with the latter recently applied across 19 countries [18].
Hand grip strength (HGS) is a strong predictor of fat‐free mass in both healthy and ill children [19]. Higher HGS is linked to lower body fat, greater muscle mass, and higher FFM in sub‐Saharan African [20] and Italian children [21], and recent evidence shows it correlates with fat‐free mass index in children aged 6–10 years [22]. Therefore, our study was designed to derive regression equations for FFM from simple anthropometric measures, including HGS as a functional anthropometric measurement, in South African children aged 6–9 years. In addition, we externally validated selected FFM regression equations against dual‐energy X‐ray absorptiometry (DXA)‐generated body composition.
2. Materials and Methods
2.1. Participants
For equation development 120 Black and White South African children aged 6–9 years were conveniently sampled using a cross‐sectional study in urban Pretoria, Gauteng province, during 2019. Data from three children were excluded due to missing observations. Race was parent‐reported using South African census categories [23]. Including both groups ensured local relevance of the FFM equations. Age was determined using the reported date of birth and date of assessment and reported in completed years. External validation used 2016 data from 75 prepubertal Black children from the same area [24].
2.2. Anthropometrics
Anthropometric data were collected in duplicate by a trained researcher in a private setting. Height (to the nearest 0.1 cm) and weight (to the nearest 100 g) were measured using a wireless Seca 274 stadiometer and medical Body Composition Analyser (mBCA) 514 (Hamburg, Germany), with participants in light clothing and no shoes. Body mass index, weight, and height‐for‐age z‐scores were calculated using the built‐in software from the Seca mBCA 514 and the WHO reference values [25]. Mid‐upper arm (MUAC) and waist circumferences (WC) were measured (to the nearest 0.1 cm) with a Seca 201 tape (0.7 cm width) on bare skin. Waist circumference was measured at the midpoint between the lowest palpable rib and the iliac crest. The MUAC was measured on the posterior upper right arm, bent at 90°, between the acromion process of the shoulder and the tip of the olecranon process. Triceps skinfold (TSF) thickness (0.2 mm) was measured at the same site using a Harpenden calliper (10 g/mm2 measuring pressure) with a precision of 0.2 mm. Handgrip strength (to the nearest 0.1 kg) was assessed three times per hand with a Takei T.K.K.5401 dynamometer [26], using the dominant hand's average for analysis.
2.3. Ethics
This research was approved by the University of Pretoria Faculty of Health Sciences Research Ethics Committee [no. 757/2018; 73/2016]. All procedures involving human participants were conducted in accordance with the University of Pretoria Code of Ethics for Research and the Declaration of Helsinki (1964; as revised in 2013).
2.4. Statistical Analysis
Data were analysed using STATA 14.2. Normality within each group was assessed with the Shapiro–Wilk test. Normally distributed variables were compared using independent‐samples t‐tests, and non‐normal variables with the Mann–Whitney U test. Technical Error of Measurement (TEM; square root of the measurement error variance) for MUAC (0.047; %TEM 0.244), TSF (0.116; %TEM 1.181) and WC (0.116; %TEM 0.202) indicated intra‐observer precision, with a reliability coefficient (R) of 1 for all.
2.5. Development of Fat‐Free Mass (FFMBIA ) Regression Equations
Raw impedance data (reactance [R] and resistance [Xc], at 50 kHz) were obtained using a multi‐frequency Seca mBCA 514. FFM was calculated using the Clasey et al. [27] BIA equation, and FM as body weight minus FFM. The FFM was used as the dependent variable in multivariable linear regression with age, race, dominant hand‐HGS, and anthropometrics (weight, height, MUAC, TSF, WC) as predictors. The sample size of 117 and nine variables for the prediction equation (sample of 13 per variable) is in line with the recommendation of 10–15 participants per variable to be tested in multiple regression analysis [11]. Models with the highest adjusted R 2 were selected as the best fit for predicting FFM, with some models omitting age, race or specific measurements to increase applicability when information or equipment is unavailable.
2.6. External Validation of the New Regression Equations
Data from an independent group of prepubertal Black children (n = 75) were used to externally validate the developed regression equations. Due to the lack of WC measurements from this sample, only four of the developed regression equations could be externally validated.
For validation, DXA‐measured FFM (FFMDXA) was used as the standard. Whole‐body DXA scans were performed using a Hologic Discovery W densitometer (v13.4.2 software) (Hologic Discovery, Hologic Inc., Bedford, MA, USA) on each participant to obtain FFMDXA [24]. We calculated the predicted FFM by applying the available anthropometric data (age, sex, race, height, weight, HGS, TSF and MUAC) from the validation group to the four developed regression equations. Pearson's correlations assessed the linear association between FFMDXA and predicted FFM (FFMBIA), independent t‐tests compared the values, and Bland–Altman plots evaluated their agreement [28]. Regression equations were considered suitable for this population if the mean FFMDXA and FFMBIA were strongly correlated (r > 0.9), statistically similar (p > 0.05), with clinically acceptable limits of agreement (LoA). An upper and lower margin of ±5 kg was considered clinically relevant [29]. The following prediction accuracy classification was used for the agreement between FFM estimates derived from BIA and DXA. Accurate prediction was defined as predicted FFM values falling within ±5% of FFMDXA. Underprediction was defined as predicted FFM values that were more than 5% lower than DXA‐measured FMM, while overprediction was defined as predicted FFM values that were more than 5% higher than DXA‐measured FFM.
3. Results
3.1. Characteristics of the Study Participants
The participant characteristics and anthropometric parameters of the Total development, Black development, and validation group are shown in Table 1. Significant differences in most anthropometric variables were found between the Total development and Black development groups compared to the validation group, except for sex and height‐for‐age z‐scores. Height did differ between the Black development and validation group (p = 0.007), but not between the Total development and validation group (p = 0.072).
TABLE 1.
Description of participants in the development and validation groups.
| Parameter | Total development group (n = 117) | Development group Black (n = 60) | Validation group Black (n = 75) | p a | p b |
|---|---|---|---|---|---|
| Boys/Girls | 64/53 | 30/30 | 35/40 | 0.695 c | 0.853 c |
| Age (years), mean ± SD | 7.9 ± 0.8 | 8.0 ± 0.8 | 8.5 ± 1.3 | 0.035 d | 0.001 d |
| Weight (kg), median (range) | 26.5 (20.0–59.5) | 26.4 (20.0–59.5) | 32.3 (18.9–65.2) | < 0.001 e | < 0.001 e |
| Height (cm), mean ± SD | 130.0 ± 6.4 | 128.2 ± 6.2 | 132.0 ± 9.0 | 0.007 d | 0.072 d |
| Weight‐for‐age z‐score, median (range) | 0.38 (−1.45–4.09) | 0.24 (−1.45–4.09) | 0.95 (−1.71–4.27) | 0.016 e | 0.026 e |
| Height‐for‐age z‐score, median (range) | 0.59 ± 0.94 | 0.18 ± 0.97 | 0.40 ± 0.94 | 0.175 d | 0.177 d |
| BMI‐for‐age z‐score, median (range) | 0.27 (−1.99–4.09) | 0.23 (−1.89–4.09) | 0.66 (−1.21–5.20) | 0.030 e | < 0.001 e |
| MUAC (cm), median (range) | 18.9 (16.1–30.3) | 18.9 (16.1–30.3) | 20.6 (7.5–35.6) | 0.008 e | < 0.001 e |
| Triceps skinfold (mm), median (range) | 8.5 (5.0–34.2) | 9.1 (5.4–34.2) | 11.6 (4.7–50.5) | 0.001 e | < 0.001 e |
| Waist circumference (cm), median (range) | 56.5 (47.6–93.0) | 58.8 (47.6–92.9) | No data | — | |
| Dominant HGS (kg), median (range) | 10.9 (7.4–17.5) | 10.4 (7.5–17.2) | 14.2 (7.0–23.0) | < 0.001 e | < 0.001 e |
| BIA based predicted FFM (FFMBIA) (kg), mean ± SD | 19.7 ± 3.2 | ||||
| Calculated FM (kg), median (range) | 7.7 (4.1–30.5) | ||||
| BIA values | |||||
| Resistance R (Ω), mean ± SD | 845.0 ± 87.2 | 874.3 ± 90.6 | 808.3 ± 98.7 | < 0.001 d | 0.008 d |
| Reactance Xc (Ω), mean ± SD | 68.8 ± 7.8 | 69.0 ± 8.2 | 64.9 ± 8.1 | 0.005 d | 0.001 d |
| Impedance Z (Ω), mean ± SD | 847.8 ± 87.3 | 877.0 ± 90.7 | 811.0 ± 98.9 | < 0.001 d | 0.007 d |
Abbreviations: BIA, bioelectrical impedance analyses; BMI, body mass index; FFM, Fat‐free mass; FFMI, fat‐free mass index; FM, fat mass; HGS, hand grip strength; MUAC, mid‐upper arm circumference.
p‐value between Black children from development and validation groups.
p‐value between Total development and validation groups.
Pearson chi‐square (χ 2) test.
Two‐sample t‐test with equal variances.
Two‐sample Wilcoxon rank‐sum (Mann–Whitney) test.
3.2. Regression Equations for FFM
The mean developmental FFM was 19.7 ± 3.2 kg using the BIA‐based prediction equation of Clasley et al. [27] (Table 1). Table 2 displays the regression equations for the variables age, sex, race, weight, height, HGS, TSF, WC and MUAC, or a combination of these, to determine the best predictor of FFM. Fourteen equations were generated based on the best fit and the availability of measurement equipment for individual variables. The equations are numbered in order of best fit (highest adjusted R 2) and/or equipment requirements.
TABLE 2.
Regression equations for fat‐free mass, ordered by R 2 and equipment requirements.
| Equipment/equation # | Regression equations for fat‐free mass (FFM) (kg) | Multiple R 2 | Adjusted R 2 |
|---|---|---|---|
| Scale, stadiometer, dynamometer, calliper, tape measure | |||
| 1 | −15.43175—0.1378783*age + 0.7738992*sex + 0.9258714*race + 0.1768151*height + 0.4071894*weight + 0.0790531*HGS—0.2395869*TSF—0.1021658*WC + 0.431349*MUAC | 0.9265 | 0.9203 |
| 2 | −14.97579 + 0.7983385*sex + 1.014887*race + 0.1644325*height + 0.4148981*weight + 0.0707649*HGS—0.2378909*TSF—0.0998019WC + 0.417384*MUAC | 0.9258 | 0.9203 |
| 4 | −8.430529—0.3210846*age + 0.4645855*sex + 0.1653742*height + 0.5481861*weight + 0.2006907*HGS—0.1848681*TSF—0.1193784*WC | 0.9042 | 0.8981 |
| Scale, stadiometer, calliper, tape measure | |||
| 3 | −15.76432 + 0.8844663*sex + 1.080319*race + 0.1721481*height + 0.4210491*weight—0.2551277*TSF—0.1040042*WC + 0.4563121*MUAC | 0.9247 | 0.9199 |
| 5 | −14.851 + 0.7000468*sex + 0.1789461*height + 0.4905296*weight—0.2981236*TSF—0.1521489*WC + 0.4589637*MUAC | 0.9026 | 0.8973 |
| Scale, stadiometer, tape measure | |||
| 6 | −15.89051 + 1.426147*sex + 1.279065*race +0.2174473*height + 0.3519496*weight—0.1389884*WC + 0.2157861*MUAC | 0.9006 | 0.8952 |
| 8 | −12.16176 + 1.333068*sex + 1.251838*race + 0.1977733*height + 0.4361496*weight—0.126843*WC | 0.8974 | 0.8928 |
| 9 | −12.14221 + 0.0092767*age + 1.333939*sex + 1.257244*race + 0.1969271*height + 0.4362931*weight—0.1266735*WC | 0.8974 | 0.8918 |
| 12 | −12.60214—0.3239081*age + 1.224318*sex + 0.2455647*height + 0.475061*weight—0.1900691*WC | 0.8712 | 0.8654 |
| 14 | −14.80055 + 1.312353*sex + 0.2350812*height + 0.4226158*weight—0.2050178*WC + 0.1694981*MUAC | 0.8687 | 0.8627 |
| Scale, stadiometer, dynamometer, calliper | |||
| 10 | −15.19542—0.3370645*age + 0.3061147*sex + 0.193865*height + 0.4214121*weight + 0.2203128*HGS—0.2118672*TSF | 0.8973 | 0.8917 |
| Scale, stadiometer, dynamometer, tape measure | |||
| 7 | −20.72913—0.0775705*age + 1.030579*sex + 1.186196*race + 0.228094*height + 0.2147599*weight + 0.1846631*HGS + 0.1159462*MUAC | 0.8996 | 0.8932 |
| 11 | −22.42876—0.4036651*age + 0.7123298*sex + 0.2684586*height + 0.1974033*weight + 0.2960844*HGS + 0.0637366*MUAC | 0.8780 | 0.8713 |
| 13 | −20.82084 + 0.7163502*sex + 0.2323395*height + 0.2310772*weight + 0.2914807*HGS + 0.0114555*MUAC | 0.8712 | 0.8653 |
Note: Age in years; sex: girl = 0, boy = 1; race: White = 1, Black = 2.
Abbreviations: HGS, hand grip strength (kg); MUAC, mid‐upper arm circumference (cm); TSF, triceps skinfold (mm); WC, waist circumference (cm).
The regression equation with best fit to predict FFM included age, sex, race, height, weight, HGS, TSF, WC and MUAC, with an adjusted R 2 of 0.9203 (equation 1), as well as equation 2 that excluded age but still yielded an adjusted R 2 value of 0.9203. Equation 3 excluded age and HGS and yielded an adjusted R 2 of 0.9199. Table 3 presents the p‐values from multivariable regression analyses, showing the statistical significance of selected demographic and anthropometric predictors across 14 equations at the 95% confidence level. Age and HGS were not significant predictors of FFM (p = 0.330 and p = 0.160, respectively). Equations 4 to 14 include different combinations of a few variables to be utilised if some measurement variables were not available.
TABLE 3.
Multivariable regression p‐values for selected variables across 14 regression equations at the 95% confidence level.
| Variables | Equation number | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | |
| Age | 0.330 | — | — | 0.027 | — | — | 0.634 | — | 0.954 | 0.024 | 0.015 | 0.051 | — | — |
| Sex | < 0.001 | < 0.001 | < 0.001 | 0.045 | 0.002 | < 0.001 | < 0.001 | < 0.001 | < 0.001 | 0.183 | 0.004 | < 0.001 | 0.004 | < 0.001 |
| Race | < 0.001 | < 0.001 | < 0.001 | — | — | < 0.001 | < 0.001 | < 0.001 | < 0.001 | — | — | — | — | — |
| Height | < 0.001 | < 0.001 | < 0.001 | < 0.001 | < 0.001 | < 0.001 | < 0.001 | < 0.001 | < 0.001 | < 0.001 | < 0.001 | < 0.001 | < 0.001 | < 0.001 |
| Weight | < 0.001 | < 0.001 | < 0.001 | < 0.001 | < 0.001 | < 0.001 | < 0.001 | < 0.001 | < 0.001 | < 0.001 | 0.003 | < 0.001 | 0.001 | < 0.001 |
| HGS | 0.160 | 0.202 | — | 0.001 | — | — | 0.003 | — | — | < 0.001 | < 0.001 | — | < 0.001 | — |
| TSF | < 0.001 | < 0.001 | < 0.001 | < 0.001 | < 0.001 | — | — | — | — | < 0.001 | — | — | — | — |
| WC | 0.009 | 0.011 | 0.008 | 0.006 | < 0.001 | 0.002 | — | 0.004 | 0.004 | — | — | < 0.001 | — | < 0.001 |
| MUAC | < 0.001 | < 0.001 | < 0.001 | — | < 0.001 | 0.064 | 0.324 | — | — | — | 0.619 | — | 0.929 | 0.200 |
Note: p‐values highlighted in bold were not statistically significant (p > 0.05).
Abbreviations: —, variable not included in equation; HGS, hand grip strength (kg); MUAC, mid‐upper arm circumference (cm); TSF, triceps skinfold (mm); WC, waist circumference (cm).
The simplest equation (equation 8) including sex, race, height, weight, and WC, yielded an adjusted R 2 value of 0.8928. Including age and excluding race in this equation (resulting in Equation 12) yielded an R 2 value of 0.8654. The weakest equation (Equation 14) included sex, height, weight, WC, and MUAC (adjusted R 2 = 0.8627).
3.3. Validation of Equations for Predicting FFM
The agreement of FFM between the four regression equations that contained corresponding anthropometric data in the developmental BIA‐based predictions (FFMBIA) and the validation (FFMDXA) groups is shown in Table 4. The correlations between all four regression equations and FFMDXA were high (r > 0.947) and statistically significant (p < 0.01). However, equation 7 yielded FFM estimates that were significantly different from the FFMDXA (p = 0.0016). The Bland–Altman plots of Equations 10, 11 and 13 indicated no evidence of proportional bias, with random scatter showing consistent differences across measurements (Figure 1). The plots also showed good agreement (LoA within 1 SD of the FFMDXA) [30] and were clinically relevant (LoA ± 5 kg) [29]. Prediction equation 10 had the highest proportion (61%) of accurate estimations of FFM and the lowest mean bias (−0.16 kg). Prediction equations 11 and 13 predicted FFM an average of 0.72–0.94 kg higher than the FFMDXA.
TABLE 4.
Agreement between FFM measured by DXA versus FFM predicted by selected equations (n = 75).
| Origin of FFM | FFM (kg) Mean ± SD | Mean difference (kg) ± SD | 95% CI (kg) | p a | Lower and upper LoA (kg) | Prediction accuracy (%): accurate b ; under c ; over d | r e |
|---|---|---|---|---|---|---|---|
| DXA‐measured | 21.5 ± 4.7 | ||||||
| Predicted by equation 7 | 24.0 ± 5.0 | −2.52 ± 5.0 | −4.07, −0.97 | 0.0016 | −5.36; 0.31 | 17.3; 0.0; 82.7 | 0.956 f |
| Predicted by equation 10 | 21.7 ± 4.8 | −0.16 ± 4.8 | −1.69, 1.36 | 0.8318 | −3.21; 2.88 | 61.3; 14.7; 24.0 | 0.947 f |
| Predicted by equation 11 | 22.2 ± 4.9 | −0.72 ± 4.8 | −2.27, 0.82 | 0.3566 | −3.64; 2.19 | 48.0; 10.7; 41.3 | 0.953 f |
| Predicted by equation 13 | 22.4 ± 5.1 | −0.94 ± 4.9 | −2.51, 0.64 | 0.2424 | −4.02; 2.15 | 46.7; 8.0; 45.3 | 0.952 f |
Abbreviation: LoA, limits of agreement (mean difference ± 1.96 SD).
T‐test between measured and predicted FFM.
Accurate prediction = percentage of FFMBIA values within ±5% of FFMDEXA.
Underprediction = percentage of FFMBIA values within < 5% of FFMDEXA.
Overprediction = percentage of FFMBIA values within > 5% of FFMDEXA.
Pearson correlation between measured and predicted FFM.
Correlation is significant at the 0.01 level.
FIGURE 1.
Bland–Altman plots showing agreement between FFM measured by DXA and predictions from Equations 10, 11 and 13. The solid horizontal line represents the line of perfect agreement (zero difference), the blue dashed line represents the mean bias (average difference), and the red dashed lines represent the 95% limits of agreement (mean ± 1.96 SD).
4. Discussion
This study yielded 14 equations for estimating FFM of children using simple anthropometric measurements. External validation against FFMDXA showed that three equations had good limits of agreement and for one equation more than 60% of predictions were within 5% of the measured value.
The 14 FFM regression equations included all or a combination of the following variables: age, sex, race, weight, height, HGS, TSF, WC and the MUAC. Equations were grouped according to the available equipment and also allowed for the absence of known age and sex information by excluding them from the equations for each respective equipment group. The adjusted R 2 values of the equations ranged from 0.8653 to 0.9203, exceeding the cut‐off of 0.80 or higher, which is considered acceptable for field prediction equations, particularly those using anthropometric or skinfold measurements to estimate body composition [31].
Anthropometry‐based equations that have been developed to predict FFM among young healthy children are limited and include simple anthropometric measurements and variables such as height, weight, age, sex, ethnicity/ancestry [16, 17, 18] and BMI‐for‐age z scores [16], while the criterion methods used varied between DXA [14, 15, 16] and deuterium dilution [17, 18]. Two Brazilian studies used the 3‐compartment model to validate anthropometric models in boys [14] and girls [15] for the simultaneous estimation of lean soft tissue, bone mineral content, and FM, as opposed to FFM. Anthropometric variables included body weight, supra‐iliac skinfold and horizontal abdominal skinfold, while contracted arm circumference was included for girls, and height and peak height velocity (PHV) for boys. Although a mixture of ethnic groups was included, they considered their sample to be uniform, as opposed to our study and other studies [16, 17, 18], in which race/ethnicity was a significant predictor of FFM.
Chronological age was not found to be a significant predictor in our study, while biological age, expressed as years from PHV in the two Brazilian studies [14, 15], and chronological age in the other studies were included as predictor variables [16, 17, 18]. This could be explained by the fact that our study included only prepubertal children aged 6–9 years, with small differences in FFM between age groups before the pubertal growth spurt [32], as opposed to the other studies that included wider age ranges (4–23 years).
Upper‐arm anthropometry (MUAC and TSF) was included as a predictor variable in our study. In only four of the nine equations that included MUAC, it emerged as a significant predictor. In a previous study, MUAC was found to be a moderate correlate of FFM in children, weaker than its link to FM, and once age, sex, and height were accounted for, its FFM prediction value weakened significantly [33]. Two more recent studies have also shown stronger associations of MUAC with fat mass indices than with fat‐free mass indices, indicating that its prediction utility lies primarily in detecting excess adiposity rather than quantifying lean tissue [34, 35]. Triceps skinfold, however, were a significant predictor of FMM in all six equations where it was included. Notably, the negative regression coefficients (−0.18 to −0.30 kg per mm) indicate that higher subcutaneous fat at the triceps is associated with lower fat‐free mass.
Although previous studies have found that HGS is associated with FFM in children [19, 20, 21, 22], in the current study, the inclusion of HGS as a predictor did not improve the prediction ability of FFM estimations. This is the first known study to include HGS as a predictor variable in the development of FFM equations for healthy children.
In the context of measuring body composition in resource‐constrained settings, equations involving anthropometric measurements using basic equipment, such as scales, stadiometers, and tape measures, can be considered. Six equations met this requirement, with equation 6, which includes sex, race, height, weight, WC and MUAC as variables, showing the highest prediction power (R 2 = 0.8952). This equation potentially provides a balance between estimation accuracy and equipment availability in limited‐resource settings.
Among the validated equations, prediction accuracy varied: Equation 10 achieved the highest proportion of accurate estimations (61.3%), whereas Equations 11 and 13 tended to underestimate FFM, with 10.7% and 8.0% underprediction, respectively, and a larger proportion of overestimation (41.3% and 45.3%). Equation 10 included TSF, as opposed to MUAC in Equations 11 and 13, indicating that the inclusion of TSF enhanced the accuracy of fat‐free mass prediction. However, as nearly 40% of predictions from Equation 10 were still inaccurate, individual‐level misclassification remains likely, suggesting that its use is more appropriate for group‐level rather than individual assessments. These findings are consistent with evidence indicating field techniques such as BIA and anthropometry exhibit greater measurement error compared with laboratory reference methods like DXA, limiting their precision for individual clinical applications [36]. When the focus is on describing the body composition of a group or establishing associations with other variables—as in research—such variability tends to be of lesser importance.
The strength of our study is the sample size of 117, providing 13 participants per variable for nine predictors, aligning with the recommendation for multiple regression [11]. Three of the FFM prediction equations were externally validated in an independent sample of Black children. Although the validation groups differed from the development group in most anthropometric characteristics, limiting reproducibility of the validated equations, their generalisability is strengthened. As noted by Ramspek et al. [37], validations in a population that closely resembles the development cohort primarily assess reproducibility, whereas the use of a validation cohort with differing anthropometric profiles provides stronger evidence of generalisability, which is particularly relevant for population‐specific tools in diverse contexts such as South Africa.
One of the limitations of this study was the use of BIA, an indirect method, as a reference method for developing the anthropometric equations. The accuracy of BIA is limited by its underlying biological assumptions, methodological and device variability, particularly in children, and its reliance on population‐specific prediction equations [38]. These factors limit the suitability of BIA as a reference method in this context. Nevertheless, the use of DXA‐derived data for external validation strengthened the validity of the newly developed prediction equations. In addition, although the FFM prediction equations applied in this study were originally developed by Clasey et al. in predominantly White children [27], the equation yielded FFM estimates that were similar to DXA‐derived reference values in South African Black children [24]. This suggests that the prediction equations used were appropriate for the study population and supports the accuracy of BIA‐derived FFM estimates in the study population. It is, however, recommended that a direct reference method, such as DXA or isotope dilution (D2O), be used. Another limitation is that only four of the regression equations could be validated against an independent Black sample group because of the lack of waist circumference measurements for the independent group. Furthermore, the prediction equations developed are limited to a narrow age range of the 6–9‐year‐old prepubertal group, with only two race groups included. However, the developed equations showed acceptable prediction power for field prediction equations to estimate fat‐free mass in this age group.
5. Conclusion
This study developed and validated simple anthropometric equations to predict FFM in prepubertal children, supporting body composition assessment in resource‐limited settings where groups of children are of primary interest. The equations show potential for use in the South African population‐level health and research. Future studies should include more racial and age groups and assess the ability of the equations to track body composition changes over time.
Author Contributions
F.A.M.W., A.P. and Z.W. conceptualised the research. A.P. and L.W. collected the data and Z.W. analysed the data. Z.W. drafted the manuscript. All authors have critically reviewed the manuscript and approved the final version of the manuscript.
Funding
Research reported in this publication was partially supported by the SOUTH AFRICAN SUGAR ASSOCIATION Nutrition research grants program, grant project number 260, funding number (EA/055/19WG/AV), and the Institute of Food, Nutrition and Wellbeing (IFNuW), University of Pretoria.
Conflicts of Interest
The authors declare no conflicts of interest.
Acknowledgements
The authors would like to thank and acknowledge the children, parents, teachers, and principals of the participating schools for their willingness to participate in this study. We thank statistician, Piet Becker, for his contribution to the statistical analysis of the prediction equations.
Data Availability Statement
The data that support the findings of this study are available from the corresponding author upon reasonable request.
References
- 1. Danquah F. I., Ansu‐Mensah M., Bawontuo V., Yeboah M., and Kuupiel D., “Prevalence, Incidence, and Trends of Childhood Overweight/Obesity in Sub‐Saharan Africa: A Systematic Scoping Review,” Archives of Public Health 78 (2020): 109, 10.1186/s13690-020-00491-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. World Health Organization , Double Burden of Malnutrition Policy Brief (World Health Organization, 2017). [Google Scholar]
- 3. Adom T., Kengne A. P., De Villiers A., and Puoane T., “Prevalence of Overweight and Obesity Among African Primary School Learners: A Systematic Review and Meta‐Analysis,” Obesity Science and Practice 5, no. 5 (2019): 487–502, 10.1002/osp4.355. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Cole T. J., Bellizzi M. C., Flegal K. M., and Dietz W. H., “Establishing a Standard Definition for Child Overweight and Obesity Worldwide: International Survey,” BMJ 320, no. 7244 (2000): 1240–1243, 10.1136/bmj.320.7244.1240. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Wilkes M., Thornton J., Horlick M., et al., “Relationship of BMI z Score to Fat Percent and Fat Mass in Multiethnic Prepubertal Children,” Pediatric Obesity 14, no. 1 (2019): e12463, 10.1111/ijpo.12463. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Kigaru D. M. D., Ndung'u Z. W., and Macharia‐Mutie C. W., “Application of Stable Isotope Dilution Techniques to Assess Body Fat and Comparison With WHO BMI‐For‐Age Classification as a Measure of Obesity Among Schoolchildren in Nairobi, Kenya,” Public Health Nutrition 24, no. 12 (2021): 3587–3591, 10.1017/S1368980020001950. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Wells J. C. K., “Using Body Composition Assessment to Evaluate the Double Burden of Malnutrition,” Annals of Nutrition & Metabolism 75, no. 2 (2019): 103–108, 10.1159/000503666. [DOI] [PubMed] [Google Scholar]
- 8. Kulkarni B., “Addressing the Double Burden of Malnutrition in Developing Countries: Need for Strategies to Improve the Lean Body Mass,” Food and Nutrition Bulletin 39, no. 2_suppl (2018): S69–S76, 10.1177/0379572118768572. [DOI] [PubMed] [Google Scholar]
- 9. Kyle U. G., Earthman C. P., Pichard C., and Coss‐Bu J. A., “Body Composition During Growth in Children: Limitations and Perspectives of Bioelectrical Impedance Analysis,” European Journal of Clinical Nutrition 69, no. 12 (2015): 1298–1305, 10.1038/ejcn.2015.86. [DOI] [PubMed] [Google Scholar]
- 10. Shepherd J. A., Heymsfield S. B., Norris S. A., Redman L. M., Ward L. C., and Slater C., “Measuring Body Composition in Low‐Resource Settings Across the Life Course,” Obesity (Silver Spring) 24, no. 5 (2016): 985–988, 10.1002/oby.21491. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Cerqueira M. S., Amorim P. R. S., Encarnação I. G. A., et al., “Equations Based on Anthropometric Measurements for Adipose Tissue, Body Fat, or Body Density Prediction in Children and Adolescents: A Scoping Review,” Eating and Weight Disorders 27 (2022): 2321–2338, 10.1007/s40519-022-01405-7. [DOI] [PubMed] [Google Scholar]
- 12. Silva A. M., Fields D. A., and Sardinha L. B., “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, 10.1155/2013/148696. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Campa F., Coratella G., Cerullo G., et al., “High‐Standard Predictive Equations for Estimating Body Composition Using Bioelectrical Impedance Analysis: A Systematic Review,” Journal of Translational Medicine 22, no. 1 (2024): 515, 10.1186/s12967-024-05272-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Machado D., Oikawa S., and Barbanti V., “The Multicomponent Anthropometric Model for Assessing Body Composition in a Male Pediatric Population: A Simultaneous Prediction of Fat Mass, Bone Mineral Content, and Lean Soft Tissue,” Journal of Obesity 2013 (2013): 428135, 10.1155/2013/428135. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Machado D., Silva A., Gobbo L., Elias P., de Paula F. J. A., and Ramos N., “Anthropometric Multicompartmental Model to Predict Body Composition in Brazilian Girls,” BMC Sports Science, Medicine and Rehabilitation 9 (2017): 23, 10.1186/s13102-017-0088-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Foster B. J., Platt R. W., and Zemel B. S., “Development and Validation of a Predictive Equation for Lean Body Mass in Children and Adolescents,” Annals of Human Biology 39, no. 3 (2012): 171–182, 10.3109/03014460.2012.681800. [DOI] [PubMed] [Google Scholar]
- 17. Hudda M. T., Fewtrell M. S., Haroun D., et al., “Development and Validation of a Prediction Model for Fat Mass in Children and Adolescents: Meta‐Analysis Using Individual Participant Data,” BMJ 366 (2019): l4293, 10.1136/bmj.l4293. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Hudda M. T., Wells J. C. K., Adair L. S., et al., “External Validation of a Prediction Model for Estimating Fat Mass in Children and Adolescents in 19 Countries: Individual Participant Data Meta‐Analysis,” BMJ 378 (2022): e071185, 10.1136/bmj-2022-071185. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. McKirdy S., Nichols B., Williamson S., and Gerasimidis K., “Handgrip Strength as a Surrogate Marker of Lean Mass and Risk of Malnutrition in Paediatric Patients,” Clinical Nutrition 40, no. 9 (2021): 5189–5195, 10.1016/j.clnu.2021.08.005. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Gerber M., Ayekoé S., Bonfoh B., et al., “Is Grip Strength Linked to Body Composition and Cardiovascular Risk Markers in Primary Schoolchildren? Cross‐Sectional Data From Three African Countries,” BMJ Open 12, no. 6 (2022): e052326, 10.1136/bmjopen-2021-052326. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Sartorio A., Lafortuna C. L., Pogliaghi S., and Trecate L., “The Impact of Gender, Body Dimension, and Body Composition on Hand‐Grip Strength in Healthy Children,” Journal of Endocrinological Investigation 25, no. 5 (2002): 431–435, 10.1007/BF03344033. [DOI] [PubMed] [Google Scholar]
- 22. Velazquez‐Villalobos S., LeCheminant J. D., Rasmussen R., and Bellini S. G., “Association Between Handgrip Strength and Muscle Mass in Children 6 to 10 Years Old,” Clinical Nutrition ESPEN 66 (2025): 179–185, 10.1016/j.clnesp.2025.01.048. [DOI] [PubMed] [Google Scholar]
- 23. Statistics South Africa , Census 2022 Statistical Release (Statistics South Africa, 2022), https://census.statssa.gov.za/assets/documents/2022/P03014_Census_2022_Statistical_Release.pdf. [Google Scholar]
- 24. van Zyl A., White Z., Ferreira J., and Wenhold F. A. M., “Developing an Impedance‐Based Equation for Fat‐Free Mass of Black Preadolescent South African Children,” Nutrients 11, no. 9 (2019): 2021, 10.3390/nu11092021. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. de Onis M., Onyango A. W., Borghi E., Siyam A., Nishida C., and Siekmann J., “Development of a WHO Growth Reference for School‐Aged Children and Adolescents,” Bulletin of the World Health Organization 85 (2007): 660–667, 10.2471/BLT.07.043497. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Centers for Disease Control and Prevention , Muscle Strength Procedures Manual (CDC, 2011), 2019, https://search.cdc.gov/search/?query=grip+strength&sitelimit=www.cdc.gov%2Fnchs%2F&utf8=%E2%9C%93&affiliate=cdc‐main. [Google Scholar]
- 27. Clasey J. L., Bradley K. D., Bradley J. W., Long D. E., and Griffith J. R., “A New BIA Equation Estimating the Body Composition of Young Children,” Obesity (Silver Spring) 19, no. 9 (2011): 1813–1817, 10.1038/oby.2011.158. [DOI] [PubMed] [Google Scholar]
- 28. Bland J. M. and Altman D. G., “Statistical Methods for Assessing Agreement Between Two Methods of Clinical Measurement,” Lancet 1, no. 8476 (1986): 307–310, 10.1016/S0140-6736(86)90837-8. [DOI] [PubMed] [Google Scholar]
- 29. Kreissl A., Jorda A., Truschner K., Skacel G., and Greber‐Platzer S., “Clinically Relevant Body Composition Methods for Obese Pediatric Patients,” BMC Pediatrics 19 (2019): 84, 10.1186/s12887-019-1454-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Flegal K. M., Graubard B., and Ioannidis J. P. A., “Use and Reporting of Bland‐Altman Analyses in Studies of Self‐Reported Versus Measured Weight and Height,” International Journal of Obesity 44, no. 6 (2020): 1311–1318, 10.1038/s41366-019-0499-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Heyward V. H. and Wagner D. R., Applied Body Composition Assessment, 2nd ed. (Human Kinetics, 2004). [Google Scholar]
- 32. Gätjens I., Schmidt S. C. E., Plachta‐Danielzik S., Bosy‐Westphal A., and Müller M. J., “Body Composition Characteristics of a Load‐Capacity Model: Age‐Dependent and Sex‐Specific Percentiles in 5‐ to 17‐Year‐Old Children,” Obesity Facts 14, no. 6 (2021): 593–603, 10.1159/000518638. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Chomtho S., Fewtrell M. S., Jaffe A., Williams J. E., and Wells J. C., “Evaluation of Arm Anthropometry for Assessing Pediatric Body Composition: Evidence From Healthy and Sick Children,” Pediatric Research 59, no. 6 (2006): 860–865, 10.1203/01.pdr.0000219395.83159.91. [DOI] [PubMed] [Google Scholar]
- 34. Craig E., Bland R., Ndirangu J., and Reilly J. J., “Use of Mid‐Upper Arm Circumference for Determining Overweight and Overfatness in Children and Adolescents,” Archives of Disease in Childhood 99, no. 8 (2014): 763–766, 10.1136/archdischild-2013-305137. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35. Oriaifo S., Abiodun P., Atimati A. O., and Nwaneri D., “Determination of Overnutrition Using Mid‐Upper Arm Circumference in Comparison With Bioelectrical Impedance Analysis in Children and Adolescents in Benin, Nigeria,” Journal of Health Research 34, no. 1 (2020): 68–78, 10.1108/JHR-02-2019-0030. [DOI] [Google Scholar]
- 36. Orsso C. E., Silva M. I. B., Gonzalez M. C., et al., “Assessment of Body Composition in Pediatric Overweight and Obesity: A Systematic Review of the Reliability and Validity of Common Techniques,” Obesity Reviews 21, no. 8 (2020): e13041, 10.1111/obr.13041. [DOI] [PubMed] [Google Scholar]
- 37. Ramspek C. L., Jager K. J., Dekker F. W., Zoccali C., and van Diepen M., “External Validation of Prognostic Models: What, Why, How, When and Where?,” Clinical Kidney Journal 14, no. 1 (2021): 49–58, 10.1093/ckj/sfaa188. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38. Ward L. C., “Bioelectrical Impedance Analysis for Body Composition Assessment: Reflections on Accuracy, Clinical Utility, and Standardisation,” European Journal of Clinical Nutrition 73, no. 2 (2019): 194–199, 10.1038/s41430-018-0335-3. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
The data that support the findings of this study are available from the corresponding author upon reasonable request.