The prevalence of childhood obesity is increasing globally(1). While BMI is commonly used to define obesity, it is unable to differentiate between fat and muscle mass, leading to calls to measure body composition specifically(2). While several tools are available to assess body composition in infancy, it is unclear if they are directly comparable. Among a subset of healthy infants born to mothers participating in a randomised controlled trial of a preconception and antenatal nutritional supplement(3), measurements were made at ages 6 weeks (n = 58) and 6 months (n = 70) using air displacement plethysmography (ADP), whole-body dual-energy X-ray absorptiometry (DXA), and bioelectrical impedance spectroscopy (BIS). Estimates of percentage fat mass (%FM) were compared using Cohen’s kappa statistic (κ) and Bland-Altman analysis (4,5). There was none to weak agreement when comparing tertiles of %FM (κ = 0.15–0.59). When comparing absolute values, the bias (i.e., mean difference) was smallest when comparing BIS to ADP at 6 weeks (+1.7%). A similar bias was observed at 6 months when comparing DXA to ADP (+1.8%). However, when comparing BIA to DXA at both ages, biases were much larger (+7.6% and +4.7% at 6 weeks and 6 months, respectively). Furthermore, there was wide interindividual variance (limits of agreement [LOA] i.e., ± 1.96 SD) for each comparison. At 6 weeks, LOA ranged from ± 4.8 to ± 6.5% for BIA vs. DXA and BIA vs. ADP, respectively. At 6 months, LOA were even wider, ranging from ± 7.3 to ± 8.1% (DXA vs. ADP and BIA vs. DXA, respectively). Proportional biases were apparent when comparing BIS to the other tools at both ages, with BIS generally overestimating %FM more among infants with low adiposity. In addition to differences according to tool type, within-tool factors impacted body composition estimation. For ADP measurements, the choice of FFM density reference (Fomon vs. Butte) had minimal impact; however, choice of DXA software version (GE Lunar enCORE basic vs. enhanced) and BIS analysis approach (empirical equation vs. mixture theory prediction) led to very different estimates of body composition. In conclusion, when comparing body composition assessment tools in infancy, there was limited agreement between three commonly used tools. Therefore, researchers and clinicians must be cautious when conducting longitudinal analyses or when comparing findings across studies, as estimates are not comparable across tools.

Air displacement plethysmography (ADP) has been considered as the ‘standard’ method to determine body fat in children due to superior validity and reliability compared with bioelectrical impedance analysis (BIA). However, ADP and BIA are often used interchangeably despite few studies comparing measures of percentage body fat by ADP (%FMADP) with BIA (%FMBIA) in children with and without obesity. The objective of this study was to measure concurrent validity and reliability of %FMADP and %FMBIA in 6-to-12-year-old boys with and without obesity. Seventy-one boys (twenty-five with obesity) underwent body composition assessment. Ten boys participated in intra-day reliability analysis. %FMADP was estimated by Bodpod using sex- and age-specific equations of body density. %FMBIA was estimated by a multi-frequency, hand-to-foot device using child-specific equations based on impedance. Validity was assessed by t tests, correlation coefficients and limits of agreement (LoA); and reliability by technical error of measurement (TEM) and intraclass correlation coefficients (ICC). Compared with %FMADP, %FMBIA was significantly underestimated in the cohort (–3·4 ± 5·6 %; effect size = 0·42) and in both boys with obesity (–5·2 ± 5·5 %; ES = 0·90) and without obesity (–2·4 ± 5·5 %; ES = 0·52). A strong, significant positive correlation was found between %FMADP and %FMBIA (r = 0·80). Across the cohort, LoA were 22·3 %, and no proportional bias was detected. For reliability, TEM were 0·65 % and 0·55 %, and ICC were 0·93 and 0·95 for %FMBIA and %FMADP, respectively. Whilst both %FMADP and %FMBIA are highly reliable methods, considerable differences indicated that the devices cannot be used interchangeably in boys age 6-to-12 years.

The purpose of this investigation was to expand upon the limited existing research examining the test–retest reliability, cross-sectional validity and longitudinal validity of a sample of bioelectrical impedance analysis (BIA) devices as compared with a laboratory four-compartment (4C) model. Seventy-three healthy participants aged 19–50 years were assessed by each of fifteen BIA devices, with resulting body fat percentage estimates compared with a 4C model utilising air displacement plethysmography, dual-energy X-ray absorptiometry and bioimpedance spectroscopy. A subset of thirty-seven participants returned for a second visit 12–16 weeks later and were included in an analysis of longitudinal validity. The sample of devices included fourteen consumer-grade and one research-grade model in a variety of configurations: hand-to-hand, foot-to-foot and bilateral hand-to-foot (octapolar). BIA devices demonstrated high reliability, with precision error ranging from 0·0 to 0·49 %. Cross-sectional validity varied, with constant error relative to the 4C model ranging from −3·5 (sd 4·1) % to 11·7 (sd 4·7) %, standard error of the estimate values of 3·1–7·5 % and Lin’s concordance correlation coefficients (CCC) of 0·48–0·94. For longitudinal validity, constant error ranged from −0·4 (sd 2·1) % to 1·3 (sd 2·7) %, with standard error of the estimate values of 1·7–2·6 % and Lin’s CCC of 0·37–0·78. While performance varied widely across the sample investigated, select models of BIA devices (particularly octapolar and select foot-to-foot devices) may hold potential utility for the tracking of body composition over time, particularly in contexts in which the purchase or use of a research-grade device is infeasible.

Early life exposures and growth patterns may affect long-term risk of chronic non-communicable diseases (NCD). We followed up in adolescence two Zambian cohorts (n 322) recruited in infancy to investigate how two early exposures – maternal HIV exposure without HIV infection (HEU) and early growth profile – were associated with later anthropometry, body composition, blood lipids, Hb and HbA1c, blood pressure and grip strength. Although in analyses controlled for age and sex, HEU children were thinner, but not shorter, than HIV-unexposed, uninfected (HUU) children, with further control for socio-demographic factors, these differences were not significant. HEU children had higher HDL-cholesterol than HUU children and marginally lower HbA1c but no other biochemical or clinical differences. We identified three early growth profiles – adequate growth, declining and malnourished – which tracked into adolescence when differences in anthropometry and body fat were still seen. In adolescence, the early malnourished group, compared with the adequate group, had lower blood TAG and higher HDL, lower grip strength (difference: −1·87 kg, 95 % CI −3·47, −0·27; P = 0·02) and higher HbA1c (difference: 0·5 %, 95 % CI 0·2, 0·9; P = 0·005). Lower grip strength and higher HbA1c suggest the early malnourished children could be at increased risk of NCD in later life. Including early growth profile in analyses of HIV exposure reduced the associations between HIV and outcomes. The results suggest that perinatal HIV exposure may have no long-term effects unless accompanied by poor early growth. Reducing the risk of young child malnutrition may lessen children’s risk of later NCD.

Few investigations have evaluated the validity of current body composition technology among racially and ethnically diverse populations. This study assessed the validity of common body composition methods in a multi-ethnic sample stratified by race and ethnicity. One hundred and ten individuals (55 % female, age: 26·5 (sd 6·9) years) identifying as Asian, African American/Black, Caucasian/White, Hispanic, Multi-racial and Native American were enrolled. Seven body composition models (dual-energy X-ray absorptiometry (DXA), air displacement plethysmography (ADP), two bioelectrical impedance devices (BIS, IB) and three multi-compartment models) were evaluated against a four-compartment criterion model by assessing total error (TE) and standard error of the estimate. For the total sample, measures of % fat and fat-free mass (FFM) from multi-compartment models were all excellent to ideal (% fat: TE = 0·94–2·37 %; FFM: TE = 0·72–1·78 kg) compared with the criterion. % fat measures were very good to excellent for DXA, ADP and IB (TE = 2·52–2·89 %) and fairly good for BIS (TE = 4·12 %). For FFM, single device estimates were good (BIS; TE = 3·12 kg) to ideal (DXA, ADP, IB; TE = 1·21–2·15 kg). Results did not vary meaningfully between each race and ethnicity, except BIS was not valid for African American/Black, Caucasian/White and Multi-racial participants for % fat (TE = 4·3–4·9 %). The multi-compartment models evaluated can be utilised in a multi-ethnic sample and in each individual race and ethnicity to obtain highly valid results for % fat and FFM. Estimates from DXA, ADP and IB were also valid. The BIS may demonstrate greater TE for all racial and ethnic cohorts and results should be interpreted cautiously.

The present study reports the validity of multiple assessment methods for tracking changes in body composition over time and quantifies the influence of unstandardised pre-assessment procedures. Resistance-trained males underwent 6 weeks of structured resistance training alongside a hyperenergetic diet, with four total body composition evaluations. Pre-intervention, body composition was estimated in standardised (i.e. overnight fasted and rested) and unstandardised (i.e. no control over pre-assessment activities) conditions within a single day. The same assessments were repeated post-intervention, and body composition changes were estimated from all possible combinations of pre-intervention and post-intervention data. Assessment methods included dual-energy X-ray absorptiometry (DXA), air displacement plethysmography, three-dimensional optical imaging, single- and multi-frequency bioelectrical impedance analysis, bioimpedance spectroscopy and multi-component models. Data were analysed using equivalence testing, Bland–Altman analysis, Friedman tests and validity metrics. Most methods demonstrated meaningful errors when unstandardised conditions were present pre- and/or post-intervention, resulting in blunted or exaggerated changes relative to true body composition changes. However, some methods – particularly DXA and select digital anthropometry techniques – were more robust to a lack of standardisation. In standardised conditions, methods exhibiting the highest overall agreement with the four-component model were other multi-component models, select bioimpedance technologies, DXA and select digital anthropometry techniques. Although specific methods varied, the present study broadly demonstrates the importance of controlling and documenting standardisation procedures prior to body composition assessments across distinct assessment technologies, particularly for longitudinal investigations. Additionally, there are meaningful differences in the ability of common methods to track longitudinal body composition changes.

This study reports the validity of body fat percentage (BF%) estimates from several commonly employed techniques as compared with a five-component (5C) model criterion. Healthy adults (n 170) were assessed by dual-energy X-ray absorptiometry (DXA), air displacement plethysmography (ADP), multiple bioimpedance techniques and optical scanning. Output was also used to produce a criterion 5C model, multiple variants of three- and four-component models (3C; 4C) and anthropometry-based BF% estimates. Linear regression, Bland–Altman analysis and equivalence testing were performed alongside evaluation of the constant error (CE), total error (TE), se of the estimate (SEE) and coefficient of determination (R2). The major findings were (1) differences between 5C, 4C and 3C models utilising the same body volume (BV) and total body water (TBW) estimates are negligible (CE ≤ 0·2 %; SEE < 0·5 %; TE ≤ 0·5 %; R2 1·00; 95 % limits of agreement (LOA) ≤ 0·9 %); (2) moderate errors from alternate TBW or BV estimates in multi-component models were observed (CE ≤ 1·3 %; SEE ≤ 2·1 %; TE ≤ 2·2 %; R2 ≥ 0·95; 95 % LOA ≤ 4·2 %); (3) small differences between alternate DXA (i.e. tissue v. region) and ADP (i.e. Siri v. Brozek equations) estimates were observed, and both techniques generally performed well (CE < 3·0 %; SEE ≤ 2·3 %; TE ≤ 3·6 %; R2 ≥ 0·88; 95 % LOA ≤ 4·8 %); (4) bioimpedance technologies performed well but exhibited larger individual-level errors (CE < 1·0 %; SEE ≤ 3·1 %; TE ≤ 3·3 %; R2 ≥ 0·94; 95 % LOA ≤ 6·2 %) and (5) anthropometric equations generally performed poorly (CE 0·6– 5·7 %; SEE ≤ 5·1 %; TE ≤ 7·4 %; R2 ≥ 0·67; 95 % LOA ≤ 10·6 %). Collectively, the data presented in this manuscript can aid researchers and clinicians in selecting an appropriate body composition assessment method and understanding the associated errors when compared with a reference multi-component model.

Measurement of body composition is increasingly important in research and clinical settings but is difficult in very young children. Bioelectrical impedance analysis (BIA) and air displacement plethysmography (ADP) are well-established but require specialist equipment so are not always feasible. Our aim was to determine if anthropometry and skinfold thickness measurements can be used as a substitute for BIA or ADP for assessing body composition in very young New Zealand children. We used three multi-ethnic cohorts: 217 children at a mean age of 24·2 months with skinfold and BIA measurements; seventy-nine infants at a mean age of 20·9 weeks and seventy-three infants at a mean age of 16·2 weeks, both with skinfold and ADP measurements. We used Bland–Altman plots to compare fat and fat-free mass calculated using all potentially relevant equations with measurements using BIA or ADP. We also calculated the proportion of children in the same tertile for measured fat or fat-free mass and tertiles (i) calculated using each equation, (ii) each absolute skinfold, and (iii) sum of skinfold thicknesses. We found that even for the best equation for each cohort, the 95 % limits of agreement with standard measures were wide (25–200 % of the mean) and the proportion of children whose standard measures fell in the same tertile as the skinfold estimates was ≤69 %. We conclude that none of the available published skinfold thickness equations provides good prediction of body composition in multi-ethnic cohorts of very young New Zealand children with different birth history and growth patterns.

The purpose of the present study was: (1) to develop a new dual-energy X-ray absorptiometry (DXA)-derived body volume (BV) equation with the GE-Lunar prodigy while utilising underwater weighing (UWW) as a criterion and (2) to cross-validate the novel DXA-derived BV equation (4C-DXANickerson), Wilson DXA-derived BV equation (4C-DXAWilson) and air displacement plethysmography (ADP)-derived BV (4C-ADP) in Hispanic adults. A total of 191 Hispanic adults (18–45 years) participated in the present study. The development sample consisted of 120 females and males (50 % females), whereas the cross-validation sample comprised of forty-one females and thirty males (n 71). Criterion body fat percentage (BF %) and fat-free mass (FFM) were determined using a four-compartment (4C) model with UWW as a criterion for BV (4C-UWW). 4C-DXANickerson, 4C-DXAWilson and 4C-ADP were compared against 4C-UWW in the cross-validation sample. 4C-DXANickerson, 4C-DXAWilson and 4C-ADP all produced similar validity statistics when compared with 4C-UWW in Hispanic males (all P > 0·05). 4C-DXANickerson also yielded similar BF % and FFM values as 4C-UWW when evaluating the mean differences (constant error (CE)) in Hispanic females (CE = –0·79 % and 0·38 kg; P = 0·060 and 0·174, respectively). However, 4C-DXAWilson produced significantly different BF % and FFM values (CE = 3·22 % and –2·20 kg, respectively; both P < 0·001). Additionally, 4C-DXAWilson yielded significant proportional bias when estimating BF % (P < 0·001), whereas 4C-ADP produced significant proportional bias for BF % and FFM (both P < 0·05) when evaluated in Hispanic females. The present study findings demonstrate that 4C-DXANickerson is a valid measure of BV in Hispanics and is recommended for use in clinics, where DXA is the main body composition assessment technique.

Lipid-based nutrient supplements (LNS) may be beneficial for malnourished HIV-infected patients starting antiretroviral therapy (ART). We assessed the effect of adding vitamins and minerals to LNS on body composition and handgrip strength during ART initiation. ART-eligible HIV-infected patients with BMI <18·5 kg/m2 were randomised to LNS or LNS with added high-dose vitamins and minerals (LNS-VM) from referral for ART to 6 weeks post-ART and followed up until 12 weeks. Body composition by bioelectrical impedance analysis (BIA), deuterium (2H) diluted water (D2O) and air displacement plethysmography (ADP), and handgrip strength were determined at baseline and at 6 and 12 weeks post-ART, and effects of LNS-VM v. LNS at 6 and 12 weeks investigated. BIA data were available for 1461, D2O data for 479, ADP data for 498 and handgrip strength data for 1752 patients. Fat mass tended to be lower, and fat-free mass correspondingly higher, by BIA than by ADP or D2O. At 6 weeks post-ART, LNS-VM led to a higher regain of BIA-assessed fat mass (0·4 (95 % CI 0·05, 0·8) kg), but not fat-free mass, and a borderline significant increase in handgrip strength (0·72 (95 % CI −0·03, 1·5) kg). These effects were not sustained at 12 weeks. Similar effects as for BIA were seen using ADP or D2O but no differences reached statistical significance. In conclusion, LNS-VM led to a higher regain of fat mass at 6 weeks and to a borderline significant beneficial effect on handgrip strength. Further research is needed to determine appropriate timing and supplement composition to optimise nutritional interventions in malnourished HIV patients.

Newborn adiposity, a nutritional measure of the maternal–fetal intra-uterine environment, is representative of future metabolic health. An anthropometric model using weight, length and flank skinfold to estimate neonatal fat mass has been used in numerous epidemiological studies. Air displacement plethysmography (ADP), a non-invasive technology to measure body composition, is impractical for large epidemiological studies. The study objective was to determine the consistency of the original anthropometric fat mass estimation equation with ADP. Full-term neonates were studied at 12–72 h of life with weight, length, head circumference, flank skinfold thickness and ADP measurements. Statistical analyses evaluated three models to predict neonatal fat mass. Lin’s concordance correlation coefficient, mean prediction error and root mean squared error between the predicted and observed ADP fat mass values were used to evaluate the models, where ADP was considered the gold standard. A multi-ethnic cohort of 468 neonates were studied. Models (M) for predicting fat mass were developed using 349 neonates from site 1, then independently evaluated in 119 neonates from site 2. M0 was the original anthropometric model, M1 used the same variables as M0 but with updated parameters and M2 additionally included head circumference. In the independent validation cohort, Lin’s concordance correlation estimates demonstrated reasonable accuracy (model 0: 0·843, 1: 0·732, 2: 0·747). Mean prediction error and root mean squared error in the independent validation was much smaller for M0 compared with M1 and M2. The original anthropometric model to estimate neonatal fat mass is reasonable for predicting ADP, thus we advocate its continued use in epidemiological studies.

Air displacement plethysmography utilises a two-component model to assess body composition, which relies on assumptions regarding the density of fat-free mass (FFM). To date, there is no evidence as to whether Lohman’s or Wells et al.’s FFM density values are more accurate in young children. Therefore, the aims of this study were to compare total body fat percentage (TBF%) assessed using the BodPod with both Lohman’s and Wells et al.’s FFM density values with TBF% from the three-component (3C) model in forty healthy Swedish children aged 5·5 years. Average TBF% calculated using Lohman’s FFM density values underestimated TBF% in comparison with the corresponding value assessed using the 3C model (22·2 (sd 5·7) and 25·1 (sd 5·5) %, respectively; P<0·001). No statistically significant difference was observed between TBF% assessed using Wells et al.’s FFM density values and the 3C model (24·9 (sd 5·5) and 25·1 (sd 5·5) %, respectively; P=0·614). The Bland and Altman plots for TBF% using both Lohman’s and Wells et al.’s FFM density values did not show any bias across the range of body fatness (Lohman: r 0·056, P=0·733 and Wells et al.: r −0·006, P=0·970). These results indicate that Wells et al.’s FFM density values should be used when assessing body composition with the paediatric option for BodPod in 5-year-old children. However, future studies are needed to confirm these results in other populations, including a wider age range of children.

This study examines the associations between maternal Traditional dietary pattern adherence and HIV/treatment with neonatal size and adiposity in urban, black South Africans, as well as how specific maternal factors – that is BMI and gestational weight gain (GWG) – may influence these associations. Multiple linear regression models were used to examine associations among maternal Traditional diet pattern adherence (pattern score), HIV/treatment status (three groups: HIV negative, HIV positive (antenatal antiretroviral treatment (ART) initiation), HIV positive (pre-pregnancy ART initiation)), BMI and GWG (kg/week), and newborn (1) weight:length ratio (WLR, kg/m) in 393 mother–neonate pairs, and (2) Peapod estimated fat mass index (FMI, kg/m3) in a 171-pair subsample. In fully adjusted models, maternal obesity and GWG were associated with 0·25 kg/m (P=0·008) and 0·48 kg/m (P=0·002) higher newborn WLR, whereas Traditional diet pattern score was associated with lower newborn WLR (−0·04 kg/m per +1 sd; P=0·033). In addition, Traditional diet pattern score was associated with 0·13 kg/m3 (P=0·027) and 0·32 kg/m3 (P=0·005) lower FMI in the total sample and in newborns of normal-weight women, respectively. HIV-positive (pre-pregnancy ART) v. HIV-negative (ref) status was associated with 1·11 kg/m3 (P=0·002) higher newborn FMI. Promotion of a Traditional dietary pattern, alongside a healthy maternal pre-conception weight, in South African women may reduce newborn adiposity and metabolic risk profiles. In HIV-positive women, targeted monitoring and management strategies are necessary to limit treatment-associated effects on in utero fat deposition.

Two, three and four compartment (2C, 3C and 4C) models of body composition are popular methods to measure fat mass (FM) and fat-free mass (FFM) in athletes. However, the impact of food and fluid intake on measurement error has not been established. The purpose of this study was to evaluate standardised (overnight fasted, rested and hydrated) v. non-standardised (afternoon and non-fasted) presentation on technical and biological error on surface anthropometry (SA), 2C, 3C and 4C models. In thirty-two athletic males, measures of SA, dual-energy X-ray absorptiometry (DXA), bioelectrical impedance spectroscopy (BIS) and air displacement plethysmography (BOD POD) were taken to establish 2C, 3C and 4C models. Tests were conducted after an overnight fast (duplicate), about 7 h later after ad libitum food and fluid intake, and repeated 24 h later before and after ingestion of a specified meal. Magnitudes of changes in the mean and typical errors of measurement were determined. Mean change scores for non-standardised presentation and post meal tests for FM were substantially large in BIS, SA, 3C and 4C models. For FFM, mean change scores for non-standardised conditions produced large changes for BIS, 3C and 4C models, small for DXA, trivial for BOD POD and SA. Models that included a total body water (TBW) value from BIS (3C and 4C) were more sensitive to TBW changes in non-standardised conditions than 2C models. Biological error is minimised in all models with standardised presentation but DXA and BOD POD are acceptable if acute food and fluid intake remains below 500 g.

The aim of this review is to extend present concepts of body composition and to integrate it into physiology. In vivo body composition analysis (BCA) has a sound theoretical and methodological basis. Present methods used for BCA are reliable and valid. Individual data on body components, organs and tissues are included into different models, e.g. a 2-, 3-, 4- or multi-component model. Today the so-called 4-compartment model as well as whole body MRI (or computed tomography) scans are considered as gold standards of BCA. In practice the use of the appropriate method depends on the question of interest and the accuracy needed to address it. Body composition data are descriptive and used for normative analyses (e.g. generating normal values, centiles and cut offs). Advanced models of BCA go beyond description and normative approaches. The concept of functional body composition (FBC) takes into account the relationships between individual body components, organs and tissues and related metabolic and physical functions. FBC can be further extended to the model of healthy body composition (HBC) based on horizontal (i.e. structural) and vertical (e.g. metabolism and its neuroendocrine control) relationships between individual components as well as between component and body functions using mathematical modelling with a hierarchical multi-level multi-scale approach at the software level. HBC integrates into whole body systems of cardiovascular, respiratory, hepatic and renal functions. To conclude BCA is a prerequisite for detailed phenotyping of individuals providing a sound basis for in depth biomedical research and clinical decision making.

Assessment of body fat (BF) in pregnant women is important when investigating the relationship between maternal nutrition and offspring health. Convenient and accurate body composition methods applicable during pregnancy are therefore needed. Air displacement plethysmography, as applied in Bod Pod, represents such a method since it can assess body volume (BV) which, in combination with body weight, can be used to calculate body density and body composition. However, BV must be corrected for the thoracic gas volume (TGV) of the subject. In non-pregnant women, TGV may be predicted using equations, based on height and age. It is unknown, however, whether these equations are valid during pregnancy. Thus, we measured the TGV of women in gestational week 32 (n 27) by means of plethysmography and predicted their TGV using equations established for non-pregnant women. Body weight and BV of the women was measured using Bod Pod. Predicted TGV was significantly (P = 0·033) higher than measured TGV by 6 % on average. Calculations in hypothetical women showed that this overestimation tended to be more pronounced in women with small TGV than in women with large TGV. The overestimation of TGV resulted in a small but significant (P = 0·043) overestimation of BF, equivalent to only 0·5 % BF, on average. A Bland–Altman analysis showed that the limits of agreement were narrow (from − 1·9 to 2·9 % BF). Thus, although predicted TGV was biased and too high, the effect on BF was marginal and probably unimportant in many situations.

The objectives of the present study were to compare body composition assessed by dual-energy X-ray absorptiometry (DXA), air displacement plethysmography (ADP) and bioelectrical impedance analysis (BIA) in severely obese Caucasian children and adolescents and to develop and validate new equations for predicting body composition from BIA using DXA as the reference method. Body composition was assessed in fifty-eight obese children and adolescents (BMI 34·4 (sd 4·9) kg/m2) aged 10–17 years by DXA, ADP and BIA. ADP body fat content was estimated from body density using equations devised by Siri (ADPSiri) and Lohman (ADPLohman). In the whole sample, the Bland–Altman test showed that ADPSiri and ADPLohman underestimated percentage fat mass (%FM) by 2·1 (sd 3·4) and by 3·8 (sd 3·3) percent units (P < 0·001), respectively, compared to DXA. In addition, compared to DXA, BIA underestimated %FM by 5·8 (sd 4·6) percent units in the whole group (P < 0·001). A new prediction equation (FFM (kg) = 0·87 × (stature2/body impedance)+3·1) was developed on the pooled sample and cross-validated on an external group of sixty-one obese children and adolescents. The difference between predicted and measured FFM in the external group was − 1·6 (sd 2·9) kg (P < 0·001) and FFM was predicted accurately (error < 5 %) in 75 % of subjects. In conclusion, DXA, ADP and the BIA are not interchangeable for the assessment of %FM in severely obese children and adolescents. The new prediction equation offers an alternative approach to DXA for the estimation of body composition in severely obese children and adolescents.