Comparison of Bioelectrical Impedance and BMI in Predicting Obesity-Related Medical Conditions
See Appendix for list of study centers.
The costs of publication of this article were defrayed, in part, by the payment of page charges. This article must, therefore, be hereby marked “advertisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Department of Nutrition, Harvard School of Public Health, 665 Huntington Avenue, Boston, MA 02115. E-mail: firstname.lastname@example.org
Objective: To determine the relative validity of specific bioelectrical impedance analysis (BIA) prediction equations and BMI as predictors of physiologically relevant general adiposity.
Research Methods and Procedures: Subjects were >12, 000 men and women from the Third National Health and Nutrition Examination Survey population. We examined the correlations between BMI and percentage body fat based on 51 different predictive equations, blood pressure, and blood levels of glucose, high-density lipoprotein cholesterol, and triglycerides, which are known to reflect adiposity, while controlling for other determinants of these physiological measures.
Results: BMI consistently had one of the highest correlations across biological markers, and no BIA-based measure was superior. Percent body fat estimated from BIA was minimally predictive of the physiological markers independent of BMI.
Discussion: These results suggest that BIA is not superior to BMI as a predictor of overall adiposity in a general population.
Bioelectrical impedance analysis (BIA),1 which involves the passage of a weak current across the extremities, has been increasingly used as a measure of body composition. This widespread use of BIA is based on its simplicity and theoretical ability to differentiate between fat mass and fat-free mass (FFM), which anthropomorphic measurements such as height and weight cannot do. Lean body mass (LBM), which consists mainly of ions in aqueous solution, registers a lower impedance, whereas fat mass does not conduct electricity as well and has a higher impedance (1). Thus, bioelectrical impedance measurements most directly reflect FFM. In practice, prediction equations use measured resistance to electricity to calculate FFM, percent body fat (%BF), or total body water (TBW), usually taking into account height, weight, sex, and age. While there has been a general consensus that BIA can be a useful tool in measuring body composition, there is little agreement on the best prediction equations.
Despite the potential value of BIA, body composition prediction equations based on this method rely on assumptions that may limit their usefulness. These assumptions include that the human body is a uniformly shaped cylinder, that the ratio of intra- to extracellular matrix remains constant to provide uniform conductivity, and that all cells are penetrated equally by a 50-kHz (most commonly used) frequency (2). The validity of the equations is further limited by the sample size and representativeness of the sample populations used to develop the prediction equations, which have usually been derived empirically by fitting multiple regression equations using a “gold standard” measure of body fat assessed by another method, such as underwater weighing. Although many prediction equations have been developed, they have rarely been evaluated in direct comparisons with each other or with standard anthropometric measures of adiposity, such as BMI.
The purpose of this study was to evaluate among adult men and women the relative validity of %BF calculated from various prediction equations and adiposity assessed by BMI. Specifically, we compared correlation coefficients between the calculated %BF, odds ratio BMI, and biological variables known to be influenced by adiposity as external validity criteria. For biological variables, we used fasting blood levels of glucose (FBG) and triglycerides (TGs), blood levels of high-density lipoprotein cholesterol (HDL-C), and systolic blood pressure (SBP) measured in the Third National Health and Nutrition Examination Survey (NHANES III) population.
Research Methods and Procedures
All subjects were participants in the NHANES III, conducted between 1988 and 1994 (3). Of the 33, 994 persons 2 months of age and older who were interviewed, we excluded 769 men and 1090 women who were less than 21 years of age, who reported diabetes or diabetes treatment, hypertension, or treatment to lower blood cholesterol, or who were missing BMI or impedance measurements. Because pregnant women and individuals with cardiac pacemakers did not have impedance measurements, they were not included in this analysis. Because some of the survey participants had missing biological measurements, the number of observations varied among the biological measurements. For analysis of blood glucose and TGs, individuals who reported fasting less than 6 hours were excluded. After all exclusions, 5224 men and 5444 women remained in the analysis for FBG, 5941 men and 6051 women for HDL, 6060 men and 6186 women for SBP, and 5275 men and 5479 women for TG.
Measurements of Body Fat
All BIA measurements were taken during the physician's examination portion of the NHANES III survey. For the BIA measurements, patients were instructed to lie on their backs while electrodes were placed on their wrists and ankles. A low-level alternating current (<1 mA) was delivered and measured at distinct frequencies between 5 KHz and 1 MHz.
We used the 1996 review of Houtkooper et al. (4) to identify specific prediction equations, as well as those included in more recent publications (5, 6, 7, 8, 9, 10, 11). Equations that included skinfolds were not used because we were looking at BIA as an alternative method to measure adiposity. Prediction equations were verified by the original articles because several errors were found in the review (see appendix for equations). Published equations were variously developed for predictions of LBM, FFM, TBW, or %BF. Ultimately, 51 separate equations were used in this analysis.
In this study, we converted LBM, FFM, and TBW from the selected equations to %BF to best compare the prediction equations. LBM was converted to FFM by the equation FFM = 0.97 × LBM for men and FFM = 0.92 × LBM for women (12). FFM was converted to %BF by the following equation: %BF = (body mass − FFM)/(body mass). TBW was converted to %BF by the following equation: FFM = TBW/0.73 (4). This was subsequently converted to %BF. BMI was calculated as BMI = weight (kilograms)/height2 (meters squared).
Biological Markers of Adiposity
It has been well documented that blood levels of glucose, TGs, and HDL-C, as well as SBP, reflect adiposity (13). All participants who were examined in the morning were asked to fast for 12 hours before their exams, whereas all those examined in the evening were asked to fast for 6 hours before their exams.
Detailed specimen collection and processing instructions for the NHANES III study are documented on the NHANES III Reference Manuals and Reports CD-ROM (3).
We first calculated the means ± SD and proportions for demographic, anthropometric, and biological variables for NHANES III participants included and excluded from this analysis. We then calculated partial Pearson correlations as measures of association between estimated %BF and each of the selected biological markers and between BMI and each of the biological markers, after adjusting for potentially confounding variables. Specifically, we adjusted both the measures of body fat and the biological markers for age (six categories), current and past smoking status (never, past, time since quitting, current with five categories for amount), height (inches, continuous), ethnicity (non-Hispanic white, non-Hispanic black, Mexican-American, other), alcohol consumption in the past year (four categories), and physical activity (sedentary, lightly active, vigorous) using regression analysis. The primary rationale for controlling for these demographic and behavioral determinants of %BF was that these variables are usually controlled in epidemiological studies investigating the health effects of adiposity, because the usual focus is the effect of adiposity independent of these variables. Also, some prediction equations included terms such as age and sex and might lead to higher correlations with the biological variables for reasons unrelated to adiposity. Because FBG, TG, HDL, SBP, and measures of adiposity were all positively skewed, we used loge-transformed values to improve normality.
In addition to the correlations using loge-transformed and adjusted variables, we calculated Spearman correlations for the same variables using the original variables but did not include them in this paper because the results were nearly identical. We also calculated Pearson correlations between estimated total fat mass and the biological markers. As the results were similar, these correlations were also not presented.
To test for differences in correlation coefficients, we used Hotelling's method, which takes into account the fact that different measures of adiposity are correlated with the same biological marker, and, thus, these correlations are not independent (14). Because of the large data set, all correlations were significantly different from zero, except where noted.
To evaluate the predictive contribution of BMI independent of %BF estimated by BIA, we also examined correlations of BMI with each of the biological markers after adjusting for several of the better BIA prediction equations and the aforementioned covariates. Also, we calculated correlations of %BF estimated from the best BIA prediction equations with each of the biological variables while adjusting for BMI in addition to the previously mentioned covariates. All statistical analyses were conducted with Statistical Analysis System software (version 6.12; SAS Institute Inc., Cary, NC).
The demographic, anthropometric, and biological characteristics of excluded and included participants in NHANES III are shown in Table 1. All characteristics for the two populations were similar except that TGs were somewhat higher in those excluded, and the mean age of excluded men was noticeably higher, largely because of the older ages of men with cardiac pacemakers and missing data for smoking.
Table 1. . Characteristics of men and women participants in the NHANES III
|Age (years)||54.7 ± 22.4*||44.1 ± 18.1||45.8 ± 22.3||43.9 ± 17.5|
| ||n = 769||n = 6260||n = 1090||n = 6365|
|Height (cm)||172.0 ± 7.6||173.8 ± 7.7||158.8 ± 7.9||160.7 ± 7.1|
| ||n = 753||n = 6260||n = 1070||n = 6365|
|Weight (kg)||77.2 ± 17.4||79.2 ± 15.3||67.3 ± 18.2||68.4 ± 16.3|
| ||n = 752||n = 6260||n = 1054||n = 6365|
|BMI (kg/m2)||26.0 ± 5.0||26.2 ± 4.4||26.7 ± 6.6||26.5 ± 6.0|
| ||n = 752||n = 6260||n = 1050||n = 6365|
|Waist circumference (cm)||94.8 ± 13.5||93.3 ± 12.0||90.1 ± 15.2||88.1 ± 14.0|
| ||n = 505||n = 6128||n = 765||n = 6227|
|Waist-to-hip ratio||0.97 ± 0.08||0.95 ± 0.1||0.88 ± 0.08||0.86 ± 0.08|
| ||n = 504||n = 6116||n = 762||n = 6223|
|Current smokers (%)||32.8||32.7||23.7||22.8|
| ||n = 768||n = 6260||n = 1090||n = 6365|
|Blood glucose (mg/dL)||98.6 ± 23.4||95.7 ± 20.0||89.9 ± 19.9||91.3 ± 16.0|
| ||n = 670||n = 5224||n = 950||n = 5444|
|HDL-C (mg/dL)||48.7 ± 15.9||47.5 ± 14.1||57.3 ± 16.0||55.5 ± 15.6|
| ||n = 676||n = 5941||n = 960||n = 6051|
|SBP (mm Hg)||128.6 ± 17.8||125.3 ± 15.8||119.3 ± 19.9||119.1 ± 18.1|
| ||n = 705||n = 6060||n = 1025||n = 6186|
|TG (mg/dL)||165 ± 128||144 ± 111||142 ± 101||119 ± 80|
| ||n = 686||n = 5275||n = 963||n = 5479|
The means of %BF for men calculated using the BIA prediction equations are shown in the second column of Table 2. The mean %BF for men ranged from 4.91% to 94.5%. Some, but not all, of the more extreme values were from equations developed from populations that included children: Boulier (mean %BF = 8.03%) was derived from a population 12 to 71 years of age, Deurenberg1 (mean %BF = 94.5%) was derived from a population of children 11 to 16 years of age, Van Loan1 and Van Loan2 (mean %BFs = 15.6% and 18.4%, respectively) were derived from a population 18 to 32 years of age, Wattanapenpaiboon1 (mean %BF = 18.0%) was derived from a population 26 to 86 years of age, Heitmann2 (mean %BF = 4.91%) was derived from a population 35 to 65 years of age, and Young and Sinha (mean %BF = 57.5%) was derived from a population 8 to 21 years of age. The estimated mean %BF from Heitmann2 was an unreasonable estimate, and, therefore, this prediction equation can probably be discounted along with Deurenberg1.
The Pearson correlation coefficients between estimates of %BF calculated from different BIA equations or BMI vs. the four chosen biological markers (FBG, HDL, SBP, and TG) among men are shown in Table 2. The highest correlations between the estimated %BF and biological markers were with serum levels of TG (positive) and HDL-C (inverse). Because of the large size of the study, many of the correlation coefficients between various estimates of %BF and biological markers were significantly different from each other even though the magnitude of the differences was slight. For FBG, BMI had the highest correlation (r = 0.19). Correlations with the prediction equations ranged from 0.07 to 0.18 (mean r = 0.13). Seven prediction equations gave correlations within 0.02 of the correlation with BMI (Heitmann1, Segal1, Segal5, Segal6, Deurenberg1, Jebb, and Heitmann2), and all other correlations were lower. For HDL among men, the correlations ranged from −0.15 to −0.33 (mean r = −0.24). The correlation with BMI was −0.32, and only Heitmann1, Segal1, Segal5, Segal6, Deurenberg1, and Heitmann2 were within 0.02 of this value. All other correlations were lower. For SBP in men, the correlation with BMI was 0.19, and the range of correlations with BIA estimates of %BF was 0.10 to 0.20 (mean r = 0.15). Fifteen of the %BF prediction equations (Heitmann1, Segal1, Segal2, Segal3, Segal4, Segal5, Segal6, Van Loan and Mayclin, Deurenberg1, Deurenberg4, Gray1, Jebb, Stolarczyk, Young and Sinha, and Heitmann3) produced correlations within 0.02 of the value for BMI, and all other correlations were lower. For TG among men, the correlations ranged from 0.19 to 0.35 (mean r = 0.26), and again BMI was the greatest. Nine %BF equations (Heitmann1, Segal1, Segal5, Segal6, Van Loan and Mayclin, Deurenberg1, Gray1, Stolarczyk, and Heitmann2) gave correlations within 0.02 of this value.
Of all of the prediction equations evaluated among men, only Deurenberg1, Heitmann1, Segal1, Segal5, and Segal6 produced correlations that were higher or within 0.02 of the correlation for BMI for all four of the evaluated variables. Heitmann2 met this criterion three times, whereas Van Loan and Mayclin, Gray1, Jebb, and Stolarczyk met it twice.
The last column in Table 2 shows strong correlations between the %BF estimated by BIA and BMI among men. These correlations ranged from 0.38 to 0.98 (mean r = 0.65), and the %BF estimated by Segal1, Segal5, Segal6, and Deurenberg1 all had correlations with BMI of 0.94 or above.
Table 3 is similar to Table 2 but is for women. The mean %BF ranged from 14.1% to 93.5% among women. Most of the BIA prediction equations that presented unrealistic mean %BF values were described above in connection with Table 2. In addition to these, Rising (mean %BF = 14.1%) was derived from a population 22 to 38 years of age. Wattanapenpaiboon2 (mean %BF = 48.3) was derived from the same population as Wattanapenpaiboon1 (26 to 86 years of age).
For FBG among women, the correlation with BMI was 0.23. Six %BF prediction equations (Heitmann1, Segal5, Segal6, Deurenberg1, Heitmann2, and Young and Sinha) gave correlations within 0.02 of this value, and all others were lower. The range of correlations was 0.05 through 0.23 (mean r = 0.16). For HDL among women, the correlation with BMI was −0.31. Seven BIA prediction equations (Heitmann1, Segal1, Segal5, Segal6, Deurenberg1, Heitmann2, and Young and Sinha) gave correlations within 0.02 of this value, and all others were lower. The range of correlations was −0.09 through −0.31 (mean r = −0.23). For SBP in women, the correlation with BMI was 0.18. The range of correlations with %BF estimated by BIA was 0.07 to 0.18 (mean r = 0.14). Thirteen equations (Heitmann1, Segal1, Segal5, Segal6, Van Loan and Mayclin, Deurenberg1, Gray1, Jebb, Eston2, Heitmann2, Houtkooper3, Young and Sinha, and Heitmann3) were within 0.02 of the BMI correlation, and all other correlations were lower. The correlation between BMI and TG was 0.29. The range of correlations with %BF was 0.14 to 0.30 (mean r = 0.26). Twenty equations (Heitmann1, Segal1, Segal5, Segal6, Van Loan and Mayclin, Deurenberg1, Deurenberg2, Deurenberg3, Gray1, Jebb, Lohman1, Wattanapenpaiboon2, Eston2, Heitmann2, Houtkooper3, Heitmann3, Kushner and Schoeller2, Lukaski and Bolonchuk1, Lukaski and Bolonchuk2, and Chumlea) gave correlations within 0.02 of the correlation with BMI, and all others were lower.
Of all of the prediction equations evaluated among women, Heitmann1, Segal5, Segal6, Deurenberg1, and Heitmann2 produced correlations that were higher or within 0.02 of the correlation for BMI for all four of the evaluated variables. Segal1 and Young and Sinha both met this criterion three times. Altogether, the correlations between BMI and the biological variables were equal to or greater than any of the BIA correlations in five of the eight comparisons among men and women.
As with the men, the last column in Table 3 shows that among women there were generally strong correlations between the %BF estimated by BIA and BMI. These correlations ranged from 0.28 through 0.98 (mean r = 0.74), and the %BF estimated by Heitmann1, Segal5, Segal6, Deurenberg1, and Heitmann2 all gave correlations with BMI of 0.92 or above.
To examine the independent contribution of BIA to prediction of FBG, HDL, SBP, and TG, we calculated Pearson correlation coefficients between %BF (using the six BIA equations with the highest overall correlations) and the four biological variables, adjusting for BMI in addition to the other covariates. These correlation coefficients were small (r ranged from 0.00 to 0.08 in both men and women). We also conducted a similar analysis to evaluate the independent contribution of BMI by calculating Pearson correlation coefficients between BMI and the four biological variables adjusting for %BF estimated by the selected BIA equations in addition to the other covariates. While most correlations were significantly different from zero and slightly larger than for %BF, they were all relatively small (r ranged from 0.00 to 0.08 in men and from 0.02 to 0.12 in women).
The above results suggest that BMI is a better, cheaper, and easier measure of adiposity than is BIA. Of the 51 BIA prediction equations evaluated, <10 seemed to give estimated %BF values that were both plausible and close to BMI in their prediction of FBG, HDL, SBP, and TG. Although their correlations with the biological markers were similar, the populations from which these prediction equations were derived varied dramatically. Deurenberg1 was from a population of 59 boys and 41 girls 11 to 16 years of age (15). Segal1, Segal5, and Segal6 were from a population of 498 women 17 to 62 years of age (5), Heitmann1 and Heitmann2 were from a population of 139 men and women 35 to 65 years of age (16), Gray1 was from a population of 62 women 22 to 74 years of age (17), Jebb was from a population of 205 men and women 16 to 78 years of age (7), Stolarczyk was from a population of 151 Native American women 18 to 60 years of age (18), Van Loan and Mayclin was from a population of 188 men and women 18 to 64 years of age (6), and Young and Sinha was from a population of 100 men and women 8 to 21 years of age (10). The remaining equations gave an estimated %BF that either had low correlations with measured biological markers or was not plausible.
When considering the substantial range of correlations presented in Tables 2 and 3, it would be easy to conclude that several of the BIA equations are considerably better predictors of %BF than others. However, many of these equations were developed from specific populations, and some equations that produced low correlations in the tables may give higher correlations in populations that are more similar to the ones from which they were derived. For example, the Deurenberg1 equation was from a study of children 11 to 16 years of age, and it is possible that it would perform better in a similar group of children.
Conversely, many equations performed well in a population very different from the one from which they were derived. For example, Segal1, Segal5, and Segal6 were all derived from female populations but were three of the best equations among men. Furthermore, Segal6 was derived from an obese population (>30% body fat), whereas Segal5 was derived from a population that excluded these people. We did not evaluate the validity of these equations separately by obese and non-obese strata, because we were attempting to use these equations to determine %BF in an overall population. Segal also derived three BIA prediction equations from a population of 1069 men in this same study, but none was a good predictor of %BF among the NHANES III population. Gray1 was also derived from a female population but performed well in predicting the chosen biological markers among men.
As for the women, both Deurenberg1 and Young and Sinha were derived from a population of children. Although the correlations between the estimated %BF that these equations produced and the biological markers are relatively high, the mean estimated %BF values from these equations are higher than those from most other equations.
The results from the equations listed after Young and Sinha should be interpreted with caution, because these equations are predictors of TBW. To compare these equations with the first 39 equations, we converted TBW to %BF by making the standard assumption that the TBW proportion of FFM is 73%. Although the conversion factor is not subject-specific, a useful prediction equation must use a standard factor for all subjects. A different factor to convert from TBW to %BF could possibly be better than 73%, but this would not alter the correlations appreciably because the same constant would be applied to everyone.
The correlations shown in Tables 2 and 3 suggest that the BIA equations provide little more predictive ability of overall adiposity than the widely used BMI. The fact that the four best BIA equations among men had correlations of 0.93 or greater with BMI shows that these prediction equations are not providing information that differs substantially from BMI. Among women, the four best equations had correlations of 0.90 or above with BMI. Therefore, it seems as though the BIA equations having the highest correlations with the biological markers have these high correlations because they are the most similar to BMI and not because they provide specific independent information on body composition. The lack of independent information derived from the BIA prediction equations is further evidenced by the low correlations that we observed with the biological variables after adjustment for BMI. Similarly, BMI offered little additional prediction of the biological variables after adjustment for %BF calculated from the most predictive BIA equations. That the information provided by the estimated %BF from these equations and BMI is similar should not be surprising because the %BF prediction equations all contain the two variables used in calculating BMI (weight and height).
When interpreting the results of this analysis, the appropriateness of using biological variables that reflect body fat rather than direct measures of %BF as criteria to assess the relative validity of different predictive equations must be considered. In principle, a high correlation with the biological markers does not necessarily signify that an equation is an adequate predictor of %BF. However, the four biological variables are among the most important physiological consequences of obesity, such that the ability to predict them is inherently relevant. This study did not take into account the effects of body fat distribution. Additional measures such as abdominal circumference could add to the prediction of outcome variables but would not necessarily improve assessment of %BF.
Several other studies have examined the relative validity of different measures of body fat compared with BMI as predictors of biological markers that reflect adiposity. A 1992 study by Spiegelman et al. (19) found that BMI was as good a predictor of blood pressure and glucose as was any other measure of body fat in nearly all analyses. The study used densitometry as a gold standard to determine total fat mass and %BF. Spiegelman et al. found that total fat mass generally had a higher correlation than %BF with serum glucose levels and blood pressure. In a study among men, Hunter et al. (20) found that BMI was generally as good as or better at predicting HDL-C levels, serum TGs, and SBP compared with body fat measured by computed tomography. A 1998 study by De Lorenzo et al. (21) found that a fat-mass-based index [(total fat mass/body weight) × 100], rather than BMI, a weight-height—based index, was more strongly associated with circulating coronary risk factors. The results of these studies are generally consistent with our findings that %BF, whether calculated from BIA or other methods, is not clearly better than BMI in predicting biological markers of obesity. Although several of the BIA prediction equations in our study seemed to perform similarly to BMI in predicting HDL, TG, and FBG levels and SBP in adults, none of them could consistently outperform BMI. Also, the four best prediction equations gave mean %BF estimates that were unrealistic. Thus, these equations do not provide superior predictions of the selected biological variables that are known to be affected by adiposity. Rather, using these equations seems to be a good way of predicting BMI.
In summary, in this study of a multiethnic population of men and women 21 to 70 years of age, the %BF calculated from BIA was not superior to BMI as a predictor of biological markers that are known to be affected by adiposity. Although some BIA equations seemed better than others at predicting %BF in a heterogeneous population, none was consistently better than the simple alternative of BMI.
This study was supported by National Cancer Institute Grant 5P01CA055075.
Nonstandard abbreviations: BIA, bioelectrical impedance analysis; FFM, fat-free mass; LBM, lean body mass; %BF, percent body fat; TBW, total body water; FBG, fasting blood glucose; TG, triglyceride; HDL-C, high-density lipoprotein cholesterol; SBP, systolic blood pressure; NHANES III, Third National Health and Nutrition Examination Survey.
Appendix. BIA prediction equations
| Heitmann1||0.279Ht/R + 0.181Wt + 0.231Ht + 0.064(Sex × Wt) − 0.077Age − 14.94; M = 1, F = 0|
| Segal1||0.00108Ht − 0.02090R + 0.23199Wt − 0.06777Age + 14.59453|
| Segal2||0.00132Ht − 0.04394R + 0.30520Wt − 0.16760Age + 22.66827|
| Segal3||0.00066360Ht − 0.02117R + 0.62854Wt − 0.12380Age + 9.33285|
| Segal4||0.00088580Ht − 0.02999R + 0.42688Wt − 0.07002Age + 14.52435|
| Segal5||0.00064602Ht − 0.01397R + 0.42087Wt + 10.43485|
| Segal6||0.00091186Ht − 0.01466R + 0.29990Wt − 0.07012Age + 9.37938|
| Van Loan and Mayclin||0.000985Ht + 0.3736Wt − 0.0238R − 4.2921Sex − 0.1531Age + 17.7868; M = 0, F = 1|
|For FFM|| |
| Boulier||0.40Ht/R + 0.64Wt − 0.16Age + 6.37 − 2.71Sex; M = 1, F = 2|
| Cordain||0.81Ht/R + 6.86|
| Chumlea||−10.678 + 0.262Wt + 0.652Ht/R + 0.015R (M)|
| Chumlea||−9.529 + 0.168Wt + 0.696Ht/R + 0.016R (F)|
| Deurenberg1||0.762Ht/R + 4.20|
| Deurenberg2||0.672 × 10Ht/R + 3.1Sex + 3.9; M = 1, F = 0|
| Deurenberg3||0.406 × 10Ht/R + 0.360Wt + 5.58Ht + 0.56Sex − 6.48|
| Deurenberg4||0.340 × 10Ht/R + 15.34Ht + 0.273Wt − 0.127Age + 4.56Sex − 12.44|
| Eston1||0.52Ht/R + 0.28Wt + 3.25|
| Gray1||0.00151Ht − 0.0344R + 0.140Wt − 0.158Age + 20.387|
| Gray2||0.00139Ht − 0.0801R + 0.187Wt + 39.830|
| Houtkooper1||0.58Ht/R + 0.24Wt + 2.69|
| Houtkooper2||0.61Ht/R + 0.25Wt + 1.31|
| Jebb||0.348613Ht/R + 0.168998Wt + 13.96674|
| Lohman1||0.475Ht/R + 0.295Wt + 5.49|
| Lohman2||0.485Ht/R + 0.338Wt + 5.32|
| Lohman3||0.62Ht/R + 0.21Wt + 0.10Xc + 4.2|
| Lukaski1||0.821Ht/R + 4.917|
| Lukaski2||0.827Ht/R + 5.21|
| Lukaski3||0.756Ht/R + 0.110Wt + 0.107Xc − 5.463|
| Rising||0.34Ht/R + 0.33Wt − 0.14Age + 6.18Sex + 13.74|
| Roubenoff||0.734Ht/R + 0.116Wt + 0.096Xc + 0.984Sex − 4.03; M = 1, F = 0|
| Stolarczyk||0.001254Ht − 0.04904R + 0.1555Wt + 0.1417Xc − 0.0833Age + 20.05|
| Van Loan1||0.50Ht/R + 0.37Wt + 1.93Sex + 3.12; M = 1, F = −1|
| Van Loan2||0.51Ht/R + 0.33Wt + 1.69Sex + 3.66; M = 1, F = −1|
| Van Loan3||0.53Ht/R + 0.29Wt + 1.38Sex + 4.40; M = 1, F = −1|
| Wattanapenpaiboon1||0.4936Ht/R + 0.332Wt + 6.493|
| Wattanapenpaiboon2||0.6483Ht/R + 0.1699Wt + 5.091|
|For %BF|| |
| Eston2||−1.00Ht/R + 1.03Wt + 12.04|
| Heitmann2||−0.283Ht/R − 0.222Ht + 0.804Wt − 0.283(Sex × Wt) + 18.71|
| Houtkooper3||−1.11Ht/R + 1.04Wt + 15.16|
| Young and Sinha||0.3981Ht/R + 0.3066Wt + 0.0953(Ht − 100) + 0.7414|
|For TBW|| |
| Danford1||0.65Ht/R + 0.71|
| Danford2||0.45Ht/R + 0.11Wt + 1.84|
| Davies||0.60Ht/R + 0.50|
| Fjeld1||0.67Ht/R + 0.48|
| Fjeld2||0.18Ht/R + 0.39Wt + 0.76|
| Heitmann3||0.240Ht/R + 0.172Wt + 0.040(Sex × Wt) + 0.165Ht − 17.58|
| Kushner1||0.593Ht/R + 0.065Wt + 0.04|
| Kushner2||0.593Ht/R + 0.065Wt + 0.04|
| Kushner and Schoeller1||0.5561Ht/R + 0.0955Wt + 1.726|
| Kushner and Schoeller2||0.382Ht/R + 0.105Wt + 8.315|
| Kushner and Schoeller3||0.396Ht/R + 0.143Wt + 8.399|
| Lukaski and Bolonchuk1||0.372Ht/R + 3.05Sex + 0.142Wt − 0.069Age + 4.98; M = 1, F = 0|
| Lukaski and Bolonchuk2||0.374Ht/R + 0.151Wt − 0.083Age + 2.94Sex + 4.65; M = 1, F = 0|