Main

Application of BI in pediatric populations is of great interest as it is a technique which is particularly suitable for the study of children, and there is increasing recognition of the importance of body composition in this age group(1, 2). Validity of BI relative to reference methods of measuring body composition is reportedly high(3, 4) in adults, although the point is disputed(5, 6). The contradictions in the literature arise in part from the wide range of impedance software and hardware used, the lack of cross-validity for some prediction equations(7, 8), and large biases between estimates derived from different commercial software(8). The principle of using BI for estimation of total body water and/or FFM in children is established(9, 10), but a number of doubts remain. Errors in estimation of FFM may be unacceptably high, and the technique may not offer improved estimation relative to anthropometry(10). In addition, users of BI in childhood are currently faced with a choice of at least four published, empirically derived, equations for prediction of FFM in addition to a range of commercial software and the rapid emergence of new equations(11). The cross-validity (general applicability) of the four currently published equations is unknown, but is in question because work on application of BI in adults has shown that cross-validity cannot be taken for granted(7, 8). The aim of the present study was therefore to test the ability of these four published equations to accurately estimate FFM in a group of prepubertal children in the UK, using hydrodensitometry with a variable density model(12) as the reference method.

METHODS

Subjects. The study sample consisted of 98 healthy, self-selected, prepubertal children, 64 boys (mean age 9.3 y, SD 1.7) and 34 girls (mean age 8.9 y, SD 1.6). Physical characteristics of subjects are given in Table 1. All subjects were in good health, and all but two were white. The sample was heterogenous with respect to body fatness(Table 1) and not greatly dissimilar from Scottish reference values for body mass index: in the girls, median body mass index centile relative to Scottish girls was 50, and in the boys, the median body mass index centile was 56(13).

Table 1 Characteristics of subjects
Full size table

Informed consent was obtained from all subjects and their parents before undertaking the study, and approval was obtained from the local Ethics Committee.

Anthropometry, measurement of body density, and estimation of reference FFM. Body weight was measured to 0.1 kg in swimsuit/swimming trunks using a standard beam balance. Height was measured to 0.5 cm using a Holtain wall-mounted stadiometer. In 57 of the boys and 24 of the girls, body density was measured in triplicate by hydrodensitometry, as previously described(14). In 7 of the 64 boys and 10 of the 34 girls, density measurements could not be performed adequately, and their data were restricted to comparisons between prediction equations. They were not included in the comparisons between the prediction and reference method. Measured body density was converted to FFM using the model described by Westrate and Deurenberg(12) which assumes changing FFM density during childhood. Density-derived estimates of FFM were then used as the reference method against which the impedance-derived estimates were validated.

BI measurements. All subjects abstained from eating, drinking, and vigorous exercise for at least 3 h before BI measurement, and measurements were carried out after voiding and 5 min of bed rest. In all 98 children, whole body electrical impedance was measured on the right-hand side of the body with the child supine using standard 800 mAmp and 50 kHz current. The Bodystat 500 system (Bodystat Limited, Isle of Man, UK) was used for BI measurements, because its technical validity had been established by us and others(15) before the study, and measured impedance values can be incorporated into prediction equations separately from the manufacturers software. In 10 of the children BI measurements were repeated within 3 d of the initial measurements (under the same conditions) for the purpose of assessing measurement precision.

The four equations tested were as follows.

  1. 1

    Schaefer et al.(10): FFM (kg) = 0.15 + 0.65 (RI) + 0.68 (age, years)

  2. 2

    Houtkooper et al.(16): FFM (kg) = 0.61 (RI) + 0.25 (weight, kg) + 1.31

  3. 3

    Deurenberg et al.(17): FFM (kg) = 0.406 (RI) + (0.36 weight, kg) + 5.58 height (M) + 0.56 gender - 6.48 where gender is male = 1; female = 0.

  4. 4

    Cordain et al.(18): FFM (kg) = 6.86 + 0.81 (RI).

Statistical analysis. Agreement between estimates of FFM derived from hydrodensitometry and those derived from the four prediction equations tested was determined according to the method of Bland and Altman(19) by calculation of biases and limits of agreement. Standard 95% CI for the average difference between the reference method and the prediction are also presented. Relationships between bias in estimation of FFM and other variables (age, fatness, FFM) were assessed by the method of Bland and Altman(19).

RESULTS

Measurement precision. In the 10 children included in the assessment of precision, mean (SD) FFM was 26.4 (5.4) kg. The 95% CI for the difference in test-retest estimates of FFM was -0.5 to +0.4 kg, indicating that, on average, the paired difference between measurements was not significantly different from zero.

Comparison of four equations with densitometry. Some distinct biases in prediction were apparent, and limits of agreement relative to hydrodensitometry were wide (Table 2). FFM (densitometry) was systematically overestimated by the equation of Cordain et al.(95% CI +2.1 to +3.3 kg) and systematically underestimated by the equations of Deurenberg et al. (95% CI -1.9 to -2.9 kg) and Schaefer et al. (95% CI -1.4 to -2.5 kg). The equation of Houtkooper et al. predicted FFM with negligible bias (95% CI -0.2 to +0.8 kg) and narrower limits of agreement relative to densitometry than the other three equations.

Table 2 Biases and limits of agreement* (in FFM, kg) between four predictions of FFM and densitometry
Full size table

For the equations of Houtkooper et al. (r = 0.07), Deurenberg et al. (r = 0.09), and Cordain et al.(r = 0.20; NS) there was little evidence of a relationship between bias and FFM size. For the equation of Schaefer et al. (r= 0.27, p < 0.05) there was such an association. There were no statistically significant relationships between errors in estimation and age or gender for any of the four equations. For the equations of Schaeferet al. (r = 0.00) and Cordain et al.(r = 0.06), there was no evidence of an association between bias and body fatness (as a percentage of body weight). Error in FFM estimation was significantly correlated with body fat percentage for the equations of Houtkooper et al. (r = 0.39) and Deurenberg et al. (r = 0.41; Fig. 1).

Figure 1
figure 1

Error in estimation of FFM (kg) in relation to body fatness. Density-derived estimate minus estimates from equations of:(a) Houtkooper et al.(16)(r = 0.39, p < 0.05); (b) Deurenberget al.(17) (r = 0.41, p< 0.05); (c) Cordain et al.(18)(r = 0.06, NS); and (d) Schaefer et al.(10) (r = 0.00, NS).

Full size image

Differences between prediction equations.Table 2 provides a comparison of estimates of FFM of prediction equations relative to each other and confirms that estimates tended to be highest by the equation of Cordain et al.(18), followed by, in descending order, Houtkooperet al.(16), Schaefer et al.(10), and Deurenberg et al.(17). Because body composition is often considered as percentage of body fat, for additional information the group mean estimates of body composition expressed as body fatness (percent of body weight) are presented in Table 3. Errors are in the opposite direction to those for FFM. This analysis confirms the hypothesis that there are large differences in estimates of FFM and body fat percentage associated with choice of prediction equation.

Table 3 Group mean (SD) estimates of body fat (% of body weight)* and 95% CI relative to densitometry
Full size table

DISCUSSION

BI is now widely used as a method of measuring body composition in children and in a variety of fields: epidemiology, clinical nutrition, and applied physiology. The present study, in common with other published work, suggests that the technique has reasonable precision under standardized conditions(10, 20, 25). However, establishing the validity of BI in childhood is more complicated. The reference method most commonly used in adults, densitometry, is difficult to apply in children. From a practical point of view, measurement of body density is not possible in all children, particularly in younger prepubertal children. Even with self-selection of subjects, we were operating at the limit of the age range over which accurate hydrodensitometry is possible. In our experience recruiting subjects under the age of 7 y was not productive because of their inability to perform the underwater weighing procedure adequately.

The “chemical immaturity” of children presents theoretical problems. FFM does not have constant composition in childhood, but shows systematic variation during development(21, 22), and there is probably relatively high inter-individual variability in FFM composition in children of similar age(23).“Variable density models” represent one attempt(2, 12) to incorporate this variability in FFM and are necessary if densitometry is used as a reference method in children. This was the approach taken in the present study with the caveats that 1) apparent differences between reference and alternative methods are increased by “error” in the reference method(24),2) multicomponent models may supersede densitometry as reference methods for pediatric use, and 3) any reference methods provide tests of relative rather than absolute validity because there is no“gold standard”(2).

The cross-validity of some published BI equations in adults is poor(7, 8), and differences in estimates by different published or commercial equations probably derive largely from the equations themselves rather than from technical error(8). Occasional reports in children have also indicated poor cross-validity of some published equations in ethnic groups(25) or populations(10, 26) other than that for which the equations were derived, or in children of different ages(17) or in disease states(20). In view of these doubts over the general applicability of published pediatric equations, the proliferation of new equations(11), and the widespread use of BI, a test of cross-validity was justified and should be considered desirable by critical users of the method. The present study suggests caution in the use of equations derived from the literature for prediction of FFM from impedance in prepubertal children. Choice of equation will be a major determinant of the estimate obtained. Most investigators would agree that systematic errors in estimation of >2 kg FFM with a mean of FFM 25 kg are undesirable(Table 2). Systematic differences of >4 kg FFM between prediction equations (Table 2) also underscore the need to use equations which are valid in the population in which the technique is being applied. In the present study commercial pediatric software for estimation of FFM from BI was not tested, because in the adult literature commercial software generally predicts body composition less accurately than prediction equations based on published studies(8), and because the origin of commercial software is usually unknown (in contrast to published studies).

The present study suggests that the equation of Houtkooper et al.(16) has a high degree of cross-validity and is suitable for use in Caucasian prepubertal children, below the age range used to derive the original equation. The equation of Houtkooper et al.(16) not only predicted FFM with negligible bias, but individual errors in estimation were considerably smaller than for the other equations currently available (Table 2) and must be considered acceptable in view of the errors inherent in the reference method(2, 24). Error arising out of the use of the Houtkooper et al.(16) equation was higher in boys than girls, and was related to the fatness of the child(Fig. 1), so there may still be some scope for adjustment of this equation to improve prediction. The other equations tested, although apparently valid in the populations in which they were derived, predicted FFM with statistically significant systematic errors and with larger random errors than did the Houtkooper et al. equation (Table 2). A detailed discussion of the source of differences between different pediatric prediction equations is beyond the scope of this report. However, differences in BI hardware as well as software(15), in premeasurement criteria(9, 17) in the reference method used, and in body composition (or even body shape) of children, between studies probably contributes to the failure of prediction equations to transfer between pediatric populations. It is also important to note that ethnicity may also be relevant(2, 16, 25) to failure of cross-validity and that disease states, particularly when associated with disturbance of water balance or distribution of body water, have a tendency to invalidate prediction equations derived from healthy populations. Finally, the present study is restricted largely to consideration of estimates of absolute body composition; the estimation of changes in body composition using BI is a complex and separate question(27).

In conclusion, BI is a simple technique that generates reasonably reproducible results in children. Choice of a prediction equation for particular purposes must be made carefully, and if validation of equations is not possible, the equation of Houtkooper et al.(16) might be the most appropriate of those currently available for estimating the FFM of prepubertal Caucasian children.