Cross-validation of bioelectrical impedance analysis of body composition in children and adolescents.

Body composition is an important measure of nutritional status in weight control and exercise programs, not only in adults but also in children and adolescents.[1] Obesity is associated with an increased risk of the development of chronic diseases including ischemic heart disease, hypertension, and diabetes.[2] The prevalence of obesity in young children is low compared with adults but increases throughout childhood and adolescence.[3] Effective intervention requires accurate assessment of body composition in laboratory, clinical, and field settings. With increased emphasis on preventive measures, physical therapists are assuming a primary role in this area.

Body composition is classically divided into two constituent parts[9,10]: the fat component, usually expressed as a percentage of the total body weight (%BF), and the fat-free mass (FFM), usually expressed in kilograms of weight. Historically, the primary laboratory methods of body composition analysis have been hydrostatic weighing (HYDRO), body densitometry (Db), hydrometry-total body water analysis (T W), and K40 spectrometry. The validity of the two-component body composition model (%BF-FFM), upon which these methods are based, assumes chemical maturity of the individual, signified by stable constituents of FFM.

Evidence suggests that the constituents of FFM change in children during normal growth and development, and the relative proportions are different compared with adults.[11-14] Specifically, the FFM of children has less protein and mineral content but contains more water, producing a lower FFM density compared with adults.[11-16] Even though adult body composition prediction equations are known to overpredict percentage of body fat, their use is widespread in the pediatric literature.[11,12,17,18] Potential overprediction errors of 7% to 13%BF have been identified.[12]

Recognizing that the FFM composition is labile during growth and development in children, several investigators have proposed age- and sex-modified two-component (Db)[11,13,19] and multicomponent (Db and TBW[20] and Db, TBW, and TMC[12]) prediction equations. Based on the results of our recent study,[21] Lohman's Siri age-adjusted Db prediction equationsll,[19] proved to be the most accurate two-component model for children and adults.

The lack of laboratory-based measurement capabilities faced by many individuals prompted the development of alternative body composition methods more appropriate to clinical and field testing settings. Bioelectrical impedance analysis (BIA) and anthropometry are two such methods. Again, population specificity is a fundamental issue. A variety of anthropometric techniques (skeletal dimension, circumference, skinfolds) have been used in the development of numerous body composition prediction equations designed for adults[17] as well as children.[20] A major drawback to this approach, however, is poor intertester reliability unless trained personnel are used.[22]

Bioelectrical impedance is a relatively new method of body composition analysis and is attractive in terms of cost, equipment portability, and minimal need for personnel training. The transducer-microprocessor (computer) based design of the BIA system enhances measurement reliability. The method is based on the principle that the resistance of a low-level electrical current applied to the body is inversely related to the TBW and electrolyte distribution. Because the FFM contains almost all the water and electrolytes of the body, its conductivity is greater than that of the body fat. Theoretically, the magnitude of an impedance measurement enables the differentiation of FFM and body fat content. The research literature on BIA is controversial. Several studies have demonstrated excellent test-retest reliability for BIA-obtained measurements, with correlation coefficients ranging from .96 to .999 for resistance measurements.[22-24] Reported criterion (HYDRO) versus BIA-predicted correlation coefficients range from .71 to.93, with standard errors of estimate (SEEs) ranging from 2.7% to 4.7%BF.[22,25,26] Some researchers have shown that the manufacturers' prediction equations tend to systematically overestimate FFM in adults, [23,24] prompting the development of new regression models. The newly reported SEEs for FFM range from 1.97 to 3.03 kg.[22-24]

Improved predictability in body composition has been suggested with the addition of anthropometric data to the BIA measurements.[27,28] Based on this rationale, the manufacturer of the Berkeley Medical Research (BMR) BIA system(*) has incorporated 10 anthropometric measurements into the standard protocol. In a recent study, Oppliger et al[29] found the BMR system to be more accurate than the RJL BIA system' in estimating minimal weight in high school wrestlers (HYDRO versus test method mean differences: BMR=+0.2 kg, RJL= -3.9 kg; total prediction error: BMR= 2.3 kg, RJL=4.4 kg).

Research on the validity of BIA-obtained measurements in children is lacking. Three studies have been published involving criterion-based validation of investigator-generated prediction equations based on combined BIA and anthropometric data (FFM SEES=1.3-3.7 kg).[27,29,30] The adequacy of existing manufacturers' prediction equations has not been adequately investigated.

The purpose of our study was to examine the criterion-based validity of measurements obtained with the manufacturers' (RJL and BMR) standardized BIA testing protocols and supplied prediction equations for the assessment of FFM in children and adolescents. Lohman's Siri age-adjusted Db prediction model was used as the criterion. Between-week test-retest reliability was also investigated. The RJL system was selected because it is one of the most popular BIA systems. Benefits of including anthropometric and multiple impedance measurements were the rationale for studying the BMR system.

Method

Subjects

Forty-seven Caucasian children and adolescents, ranging in age from 8 to 20 years (X = 12.1, SD = 2.3), with no known chronic disease participated in the study. The original sample consisted of 48 subjects. One male subject, however, was dropped from the study because of technical difficulties encountered with the BIA measurements, resulting in deviate measurements (greater than two standard deviations) observed in the screening of the data. We believe that this subject's hand may have been touching his leg during the measurement procedure, producing a short circuit and subsequently very low BIA measurements. The subjects in this study were 23 boys, with ages ranging from 8.5 to 20.3 years (X = 12.7, SD = 2.7), and 24 girls, with ages ranging from 8.8 to 14.8 years (X = 11.6, SD = 1.6). The height and weight of each child were between the 5th and 95th percentiles for age on the National Center for Health Statistics growth chart.[31] A physical examination was administered to each subject by a physician (a pediatric endocrinologist) to determine the subject's maturity level in Tanner stages 1 to 5.[32,33] The Tanner scale of pubertal stage based on the development of genitalia and pubic hair has commonly been used for determining the maturational level of subjects. Stages 4 and 5 were combined because of a limited number of more mature subjects, and members of both groups are considered postpubescent. Height was measured to the nearest 0.5 cm with a stadiometer.[double dagger] Weight was measured to the nearest 0.2 kg with a standard balance-beam medical scale.[double dagger] Descriptive statistics of the entire subject sample are presented according to gender and Tanner stage in Table 1. Written informed parental consent and informed subject assent were obtained prior to participation in the study.

[TABULAR DATA OMITTED]

Design

A repeated-measures design was used in which HYDRO, RJL, and BMR measurements were obtained from every subject. The HYDRO measurements were used to calculate the criterion body composition values based on Lohman's Siri age-adjusted Db prediction equations.[11] Subjects refrained from intake of food and liquids and vigorous exercise for at least 3 hours prior to testing. The BIA tests were conducted in random order, with all measurements taken by the same investigator (Y-TW). For practical reasons (ie, to eliminate the need to change clothes during testing), the HYDRO measurements were taken last. Repeated measurements were taken on a subsample of 14 children 1 week later for reliability analysis.

Hydrostatic Weighing

Hydrostatic weighing was performed in a large Hubbard tank with a dropout bottom. Weight measurement was made with an electronic Genesco load cell (model ASW)[section] that was connected in series between a weighing chair and an electromechanical hoist. The load cell was calibrated by one of the investigators (DHN) to provide measurement resolution of 0.01 V (0.02 kg). Ten trials were performed on each subject, and the average of the three heaviest measurements was used for data analysis according to the protocol of Katch.[34] The HYDRO weight was corrected for gastrointestinal volume (an assumed constant 100 mL)[35] and for individually determined residual lung volume that was measured prior to HYDRO testing by the helium dilution technique using a Gould Godart Pulmonette II testing unit.(//)

Bioelectrical impedance

Measurements

Bioelectric impedance was measured with two commercially available systems, the RJL BIA-103 and the BMR unit, based on the manufacturers' protocols and their most current versions of computer software (RJL-BC II, version 1.1, program[dagger] and BMR-2000 program(*)).

The RJL BIA-103 body composition analyzer was used with four small, disposable, self-adhesive rubber electrocardiographic type electrodes that were attached to the right side of the subject's body. Measurements were made with the child lying quietly on his or her back with arms and legs slightly abducted. The skin sites for electrode application were prepared using alcohol gauze pads to rid the areas of dead skin and oils. The two current-introducing electrodes were placed on the dorsal surface of the right hand, over the distal end of the second and third metacarpals, and on the dorsal surface of the foot, over the distal end of the second and third metatarsals. The recording electrodes were placed on the dorsal surface of the wrist, on a line bisecting the styloid processes of the radius and the ulna, and on the dorsal surface of the ankle, on a line bisecting the medial and lateral malleoli.

The individual computer data entry included the subject's name, age, height, gender, physical activity level, and measured resistance and reactance. Impedance in an alternating circuit is the opposition to current flow. Resistance and reactance contribute to impedance according to the following equation:

Impedance =

(1) [square root of] [Resistance.sup.2] + [Reactance.sup.2]

The physical activity level (ie, inactive, light, moderate, heavy, and vigorous) did not affect the body composition results. It was used in computing the net caloric intake for weight loss or weight gain recommendation. An abbreviated summary report, including FFM and percentage of body fat, was provided after entering the resistance and reactance measurements into the RJL computer software system.

The BMR-2000 BIA measurement system consisted of six nondisposable metal disk electrodes mounted on lightweight plastic attachment clamps. Testing position and skin electrode site preparations were similar to the RJL BIA test. On the right side of the body, one pair of electrodes was placed 2 cm above the styloid processes of the dorsal and ventral aspects of the ulna and radius, and the second pair was placed on the medial and lateral aspects of the shank, which was standardized to 21 cm proximal to the heel by use of a bracket extension attached to the electrode clamp. Two single electrodes on the left side of the body were positioned at the dorsal side of the wrist and lateral side of the shank at sites identical to those on the right limbs. Electrode conductive cream(#) was applied to the metal disk electrodes to optimize the skin electrode interface for current transmission.

According to the manufacturer's protocol, 10 anthropometric measurements (ie, forearm and leg length and girths of the wrist, forearm, biceps muscle, chest, waist, buttocks, thigh, and calf) were taken by a cloth tape measure to the nearest 1 mm on the right side of body. In addition to these physical measurements, the subject's height, weight, age, gender, and level of physical activity were entered into the BMR-2000 system computer. Activity levels (average, intense, and very intense) were used to estimate calorie consumption in an individual and suggest an exercise program as well. Activity levels would not affect the body composition results. With the subject properly hooked up to the system, the computer measurement system automatically took all of the required bioelectrical impedance measurements. The attached printer provided a comprehensive data summary report, including FFM and percentage of fat, but did not provide resistance or reactance values.

Statistical Analysis

The statistical analysis was performed using Statistical Analysis System (SAS)(**) General Linear Model (GLM) program. The .05 alpha level was adopted as the criterion for statistical significance. Descriptive statistics (means and standard deviations) were calculated for all variables. Test-retest group means and intraclass correlation coefficients (ICC, type I),[36] were calculated on the subsample of 14 subjects for between-week reliability analysis. Because of the known gender differences in body composition, the main analysis was performed separately by gender.

The validity analysis was confined to FFM measurements and included a two-way repeated-measures analysis of variance (ANOVA) (factors = test method and maturation level) and Dunnett's pair-wise comparisons between the two BIA test methods (RJL and BMR) versus the HYDRO-determined criterion FFM based on Lohman's Siri age-adjusted Db equation. Pearson Product-moment Correlation Coefficients (r) and coefficients of determination ([r.sup.2]) were also computed between the RJL and BMR versus the criterion. Intraclass correlation coefficients were not computed because equivalence of measurements obtained between methods was not expected. Standard errors of estimate and total errors of prediction (TEs) were calculated to assess random and systematic variations based on the method described by Lohman.[17]

Results

Reliability

Table 2 contains the descriptive data on the reliability sample (n = 14). Test-retest group means and calculated ICCs for body weight, body density (Db), and FFM from Lohman's Siri age-adjusted Db equation, RJL BLA, and BMR BIA methods are presented in Table 3. The mean differences were less than 0.5 kg. The ICCs ranged from .987 to .997 in FFM determinations.

Table 2. Descriptive Subject Data(a) on Sample Used to Examine Reliability (N =
14)
             Age (y)      Height (cm)     Weight (kg)
Gender  N    X     SD     X        SD     X        SD
Boys    8    12.6  1.3    153.6    7.1    47.5     8.0
Girls   6    10.7  2.0    151.0    7.2    43.4     9.7
Total   14   11.8  1.8    152.5    7.0    45.7     8.7
(a) Tanner stages 1-5 represented by both genders.[30,31]
Table 3. Between-Week (Test-Retest) Reliability Data for Basic Measurements
(N = 14)
Variable                     Test   X         SD       ICC(a)
Body weight (kg)               1    45.7      8.7       .999
                               2    45.5      8.7
Body density (Db)(b)           1     1.0475   0.0228    .989
                               2     1.0451   0.0215
Siri age-adjusted Db FFM(c)    1    36.7      5.7       .991
                               2    36.8      5.7
RJL(d) FFM                     1    36.0      6.4       .987
                               2    35.5      6.2
BMR(e) FFM                     1    38.7      7.4       .997
                               2    38.7      7.4
(a) ICC = intraclass correlation coefficient.
(b) Db (in kilograms per liter) determined by hydrostatic weighing.
(c) FFM = fat-free mass (in kilograms).
(d) RJL = RJL Systems bioelectrical impedance analysis method.
(e) BMR = Berkeley Medical Research bioelectrical impedance analysis method.

Validity

The means and standard deviations of FFM for the male and female subjects according to test method and Tanner stage of maturation are graphically presented in Figures 1 and 2, respectively. The FFM systematically increased with level of maturation, and FFM appeared to be greater for boys than for girls. The ANOVA revealed nonsignificant F statistics for test methodxmaturation level interaction. Table 4 contains the final validation statistics for FFM based on the different measurement techniques. The RJL method underestimated FFM, with mean differences of 0. 1 kg in boys and 0.9 kg in girls. The latter difference was statistically significant (P<.05). The BMR method significantly (P<.05) overestimated FFM (3.7 kg) in boys but underestimated FFM (- 0.5 kg) (nonsignificant difference, (P>.05) in girls. High correlation coefficients (r=.97-.99) and coefficients of determination ([r.sup.2] =.93-.98) were obtained for both BIA methods.

[TABULAR DATA OMITTED]

Scatter plots and associated prediction lines for RJL and BMR versus the criterion Lohman's Siri age-adjusted Db equation FFM data for the male and female subjects are presented in Figures 3 and 4, respectively. Each set of data points formed an upward-sloping reasonably tight elliptical pattern, confirming the comparatively small SEE ([is less than or equal to] 2.4 kg) and the relatively high positive correlation coefficient values ([is greater than or equal to] .97) reported in Table 4. As depicted, the prediction lines for both BIA methods paralleled reasonably well the true line of identity. The downward displacement of the BMR prediction line for the male subjects agreed with the significant positive mean differences (criterion versus test method) reported in Table 4. The BMR prediction line for the girls and the RJL prediction lines for the boys and girls appeared to be almost coincident with the identity lines. As shown in Table 4, the highest TE resulted for the boys' BMR data (4.4 kg). The BMR TE for the girls and the RJL TE for both the boys and girls were consistently small ([is less than or equal to] 2.1 kg).

Discussion

Reliability addresses the issue of how consistent test measurements are. Predictability and accuracy, issues related to validity, involve comparing a test method with an acceptable criterion or reference standard. Identification of the best pediatric criterion method of body composition analysis has been a concern. In growing animals and humans, the chemical composition of FFM changes with age as the water content decreases and body solids increase. The application of adult prediction models to the children and adolescents in our study would have resulted in an overprediction of %BF (7%-13%)[12] and a corresponding underprediction in FFM. Theoretically, in children, the more constituents of FFM that can be determined, the more accurately body composition can be estimated. This is the rationale for the three-component (Db, TBW) and four-component (Db, TBW, TMC) prediction models proposed by Slaughter et al[20] and Boileau et al,[12] respectively. In many research and clinical settings, however, only one measurement technique may be available. In this context, it has been shown that Lohman's Siri age-adjusted Db prediction model[19] is an acceptable criterion method. In our study, good agreement was shown between Lohman's Siri age-adjusted Db prediction model and Boileau et al's four-component model (mean differences in FFM [is less than or equal to] 0.6 kg, r [is greater than or equal to] .996, [r.sup.2] [is greater than or equal to] .991 and FFM TE = 0.8 kg between these two models).

Based on the standardized procedures used in this study, between-week test-retest reliability was excellent (ICC [is greater than or equal to] .99) for HYDRO as well as for the RJL and BMR BIA methods. These results are comparable to those of other reports of adults and children (r [is greater than or equal to] .96).[23,26,30]

The correlation coefficient values between the criterion versus the test method indicate the degree of association and the ability of the test to discriminate among individuals in predicting the criterion. The closer the correlation coefficient approximates unity, the stronger the association (ie, a one-to-one relationship). The coefficient of determination reflects the degree of common variability between the criterion and the test method. The closer the coefficient of determination approximates unity (100% agreement), the more confident the predictions. The high correlation coefficient (r [is greater than or equal to] .97) and coefficient of determination ([r.sup.2][is greater than or equal to] .93) values obtained in this study indicate excellent predictability of FFM for both the RJL and BMR BIA systems. Based on multicenter cross-validation analysis of their newly BIA-generated FFM prediction equations, Houtkooper and associates[30] reported coefficients of determination of [is greater than or equal to] .93.

Measurement accuracy reflects the magnitude of the measurement error. Measurement error, in general terms, is simply the absolute difference between a known value (ie, criterion) and the observed value obtained with the method being evaluated. As reflected by our criterion versus BIA method comparisons, the absolute between-method discrepancies were small ([is less than or equal to] 0.9 kg), with the exception of BMR for boys (difference = 3.7 kg). As expected, the general trend, excluding BMR for boys, was for an underprediction of FFM, which was in agreement with the literature.[12] Houtkooper associates[30] have suggested a standard difference of [is less than or equal to] 5% of the measured value of the criterion variable to evaluate the acceptability for cross-validation equations. For our study, this translates into standards of [is less than or equal to] 2.0 kg for boys (criterion FFM = 39.9 kg x 5%) and [is less than or equal to] 1.7 kg for girls (criterion FFM = 33.4 kg x 5%). All of the mean differences met these criteria, with the exception of BMR for boys.

In cross-validation studies, the sources of variation contributing to the measurement error is a primary concern. Technically, measurement error consists of random and systematic factors. As described by Lohman,[17] a strategy for statistically assessing these sources of error is to do linear regression analysis between the criterion versus the test method and compute SEE and TE statistics. Quantitatively, SEE denotes the dispersion of the data points about the derived prediction line and represents the magnitude of the random error. In contrast, TE denotes the dispersion of the data points about the line of identity and collectively represents the random plus systematic error. In essence, TE quantifies the total prediction error. Ideally, the SEE should be as small as possible and the TE should approximate the SEE value. With the exception of BMR for boys, the TE and SEE values were quite similar (SEE versus TE differences [is less than or equal to] 0.3 kg), suggesting minimal systematic error. Applying the [is less than or equal to] 5% standard difference criterion (ie, [is less than or equal to] 2.0 kg for boys and [is less than or equal to] 1.7 kg for girls), SEES for boys with the RJL BIA system and for girls with the RJL and BMR BIA systems met the standard. The RJL TE for boys slightly exceeded the standard (+ 0.1 kg), whereas the RJL TE for girls met the standard (- 0.2 kg). The TE values for the BMR method were larger and exceeded the < 5% standard for both the boys (+ 2.4 kg) and the girls (+ 0.2 kg).

Only two cross-validation studies have been published on body composition in children using Lohman's multiple-component equation as the criterion method.[28,30] Guo et al[28] found SEES of 2.2 to 2.3 kg for predicting FFM by using their BIA prediction equation. Houtkooper et al[30] reported a range of 1.5 to 3.7 kg in SEE when comparing their prediction equations with the criterion method. The subjects in our study were healthy white children and youths ranging in age from 8 to 20 years. The results of this crossvalidation study, therefore, are valid only for subjects meeting this description. The subjects were, however, very similar to the subjects in Houtkooper and associates' multicenter study and Guo and colleagues' study. The results of these studies indicated that BIA can be used to validly measure FFM in children and adolescents. Interstudy comparison suggests that measurements obtained with the RJL unit with the manufacturer-supplied prediction equations appear to have comparable accuracy to those obtained by Houtkooper and associates'[30] new multiple regression equations. The BMR system appeared to work well with girls, but with boys the systematic and random errors of prediction were unacceptable. Based on the sample mean FFM of 39.9 kg, the BMR value could range from 35.5 to 44.3 kg. The BMR system including anthropometric data (girth and limb-length measurements) in the manufacturer's equation did not improve the accuracy in predicting FFM, as suggested by the larger total errors of the BMR values compared with the RJL values. One factor might be differences in age among the samples used to generate the manufacturer's equation, high school wrestlers used in Oppliger and colleagues' study,[29] and the average pediatric sample used in this study.

Conclusions

The adult prediction model overestimates %BF and underestimates the FFM in children and adolescents because of the lower protein and mineral content, more water, and the lower density of FFM. Theoretically, the more constituents of FFM that can be determined, the more accurately the body composition can be estimated. In many research and clinical settings, however, only one measurement technique may be available. Our study demonstrates that Lohman's Siri age-adjusted Db prediction model is an acceptable criterion method compared with the four-component model with Db, TBW, and TMC measured. Using Lohman's Siri age-adjusted Db prediction model as the standard, the RJL BIA estimations of FFM based on the manufacturer's prediction equations are clinically reliable and valid for assessing body composition in children and adolescents. The BMR BIA system was also found to provide reliable measurements. Inclusion of anthropometric measurements in the BMR system does not enhance the predictability of bioelectrical impedance in body composition. The prediction of FFM by the BMR system was acceptable for girls, but significant overprediction of FFM for boys was noted. Refinement of the existing prediction equation is recommended before general application to the pediatric population. Because the subjects in this study were healthy Caucasian children and adolescents with %BF ranging from 6.7 to 37.5, further research is needed to validate the use of the BIA method in extremely lean or obese subjects and in populations of different ethnicity.

(*) Berkley Medical Research, 14269 Doolittle Dr, San Leandro, CA 94577. [dagger] RJL Systems, 9930 Whittier, Detroit, MI 48224. [double dagger] Detecto Scale, Rogan Scale and Equipment Inc, 400 S Devils Glen Rd, Bettendorf, IA 50722. [section] Genesco Technology Corp, 650 Easy St, Simi Valley, CA 93065. (//) SensorMedicas Corp, 1630 S State College Blvd, Anaheim, CA 92806. (#) Hewlett-Packard Inc, 175 Wyman St, Waltham, MA 02154. (**) SAS Institute Inc, Box 8000, Cary, NC 27511.

References

[1] Gilliam TB, MacConnie SE. Coronary heart disease risk in children and their physical activity pattern. In: Boileau RA, ed. Advances in Pediatric Sports Sciences: Biological Issues. Champaign, Ill: Human Kinetics Publishers Inc; 1984; 1: 171-187. [2] US Surgeon General Report on Nutrition and Health, Washington, DC: US Govt Printing Office; 1988:5 (US Dept of Health and Human Services Publication 88-50211). [3] Abraham S. Obese and Overweight Adults in the United States. Washington, DC: US Govt Printing Office; 1983 (National Center for Health Statistics, Vital and Health Statistics Series II, No. 230). [4] Freedman DS, Shear CL, Burke GL, et al. Persistence of juvenile-onset obesity over eight years: the Bogalusa Heart Study, Am J Public Health. 1987;77:588-592. [5] Garn SM, lavelle M Two-decade follow-up of fatness in early childhood. Am J Dis Child. 1985;139:181-185 [6] Gortmaker AL, Dietz WH, Sobol AM, Wehler CA. Increasing pediatric obesity in the United States. Am J Dis Child. 1987; 141:53 5-540. [7] Lohman TG, Going SB, Slaughter MH, Boileau RA. Concept of chemical immaturity in body composition estimates: implications for estimating the prevalence of obesity in childhood and young. Am J Hum Biol. 1989;1:201-204. [8] Zac PM, Harlan WR, Leaverion PE, Cormoni-Huntley J. A longitudinal study of body fatness in childhood and adolescence. J Pediatr. 1979; 95:126-130. [9] Pollock ML, Wilmore JH, Fox SM. Exercise in Health and Disease. Philadelphia, Pa: WB Saunders Co; 1990:319-370. [10] Lukaski HC. Methods for the assessment of human body composition: traditional and new. Am J Clin Nutr. 1987;46:537-556. [11] Lohman TG. Applicability of body composition techniques and constants for children and youth. Exerc Sport Sci Rev. 1986;14:325-357. [12] Boileau RA, Lohman TG, Slaughter MH. Exercise and body composition of children and youth. Scand J Sports Sci. 1985;7:17-27. [13] Westrate JA, Deurrenberg P. Body composition in children: proposal for a method to calculate body fat percentage from total body density or skinfold thickness measurement. Am J Clin Nutr 1989;50:1104-1115 [14] Boileau RA, Lohman TG, Slaughter MH, et al. Hydration of the fat-free body in children during maturation. Hum Biol. 1984;56:651-666. [15] Foman SJ, Haschke F, Ziegler EE, Nelson EN. Body composition of reference children from birth to age 10 years. Am J Clin Nutr. 1982;35:1169-1175. [16] Lohman TG, Slaughter MH, Boileau RA, et al. Bone mineral measurements and their relation to bone density in children, youth and adults. Hum Bioz 1984;56:667. [17] Lohman TG. Skinfolds and body density and their relation to body fatness: a review. Hum Biol 1981;53:181-225. [18] Boileau RA, Wilmore JH, Lohman TG, et al. Estimation of body density from skinfold thickness, body circumferences and skeletal widths in boys 8 to 11 years: comparison of two samples. Hum Biol 1981;53:575-592. [19] Lohman TG. Assessment of body composition in children. Ped Ex Sci. 1989; 1: 19-30. [20] Slaughter MH, Lohman TG, Boileau RA, et al. Skinfold equations for estimation of body fatness in children and youth. Hum Biol. 1988; 60:709-723. [21] Nielsen DH, Cassady SL, Janz KF, et al. Criterion methods of body composition analysis for children and adolescents. Am J Hum Biol. In press. [22] Lukaski HC, Bolonchuk WW, Hall CB, Sidrs WA. Validation of tetrapolar bioelectrical impedance method to assess human body composition. J Appl Physiol 1986;60:1327-1332. [23] Segal KR, Vanloan M, Fitzgerald PI, et al. Lean body mass estimation bv bioelectrical impedance analysis: a four-site cross-validation study. Am j Clin Nutr. 1988;47:7-14. [24] Vanloan M, Mayclin P. Bioelectrical impedance analysis: Is it a reliable estimator of lean body mass and total body water? Hum Biol. 1987;59:299-309. [25] Segal KR, Gutin B, Presta E, et al. Estimation of human body composition by electric impedance methods: a comparative study. J Appl Physiol. 1988;62:529-534 [26] Jackson AS, Pollock ML, Graves JE, Mahar MT. Reliability and validity of bioelectrical impedance in determining body composition. J Appl Physiol, 1988;64:529-534. 27 Houtkooper LB, Lohman TG, Going SB, Hall MC. Validity of bioelectric impedance for body composition assessment in children. J Appl Physiol 1989;66:814-821. [28] Guo S, Roche AF, Houtkooper L. Fat-free mass in children and young adults predicted from bioelectric impedance and anthropometric variables. Am J Clin Nutr. 1989;50:435-443 [29] Oppliger RA, Nielsen DH, Vance CG, Wrestlers' minimal weight: anthropometry, bioimpedance, and hydrostatic weighing compared. Med Sci Sports Exerc. 1991;23:247-253. [30] Houtkooper LB, Going SB, Lohman TG, et al. Bioelectrical impedance estimation of fat-free body mass in children and youth: a cross-validation study. J Appl Physiol. 1992;72:366-373. [31] Hamill PW, Drizd TZ, Johnson CL, et al. NCHS growth curves for children from birth to 18 years: United States Department of Health, Education and Welfare. Vital Health Stat [11]. 1977;165:1-74. [32] Marshall W, Tanner J. Variations in pattern of pubertal changes in girls. Arch Dis Child. 1969;44:291-303. [33] Marshall W, Tanner J. Variations in pattern of pubertal changes in boys. Arch Dis Child. 1969;44:13-23. [34] Katch Fl. Apparent body density and variability during underwater weighing, Res Q. 1968;39:993-999 [35] Burkirk ER. Underwater weighing and body density: a review of procedures. In: Brozek J, Henschel A, eds. Tecbniquesfor Measuring Body Composition. Washington, DC: National Academy Sciences; 1961:90-105 [36] Lahey NIA, Downey RG, Saal FE. Intraclass correlations: There's more there than meets the eye. Psychol Bull. 1983;93:586-595