Am J Clin Nuir l988;47:7-14. Printed in USA. © 1988 American Society forClinical Nutrition 7

Original Research Communications-general

Lean body mass estimation by bioelectrical impedanceanalysis: a four-site cross-Validation study13

Karen R Segal, EdD; Marta Van Loan, PhD; Patricia I Fitzgerald, PhD;

James A Hodgdon, PhD; and Theodore B Van Itaiie, MD

ABSTRACT This study validated further the bioelectrical impedance analysis (BIA) method

for body composition estimation. At four laboratories densitometrically-determined lean body

mass (LBMd) was compared with BIA in 1567 adults (1069 men, 498 women) aged 17-62 y

and with 3-56% body fat. Equations for predicting LBMd from resistance measured by BIA,

height2, weight, and age were obtained for the men and women. Application of each equationto the data from the other labs yielded small reductions in R values and small increases in SEES.Some regression coefficients differed among labs but these differences were eliminated afteradjustment for differences among labs in the subjects’ body fatness. All data were pooled toderive fatness-specific equations for predicting LBMd: the resulting R values ranged from 0.907to 0.952 with SEES of 1.97-3.03 kg. These results confirm the validity of BIA and indicate that

the precision of predicting LBM from impedance can be enhanced by sex- and fatness-specific

equations. AmfClinNutr l988;47:7-l4.

KEY WORDS Body composition, densitometry, bioelectrical impedance analysis, lean body

mass

Infroduction

The use of bioelectrical impedance analysis (BIA) inbody composition assessment has been investigated re-

cently. BIA, a portable impedance analyzer (RJL Systems,

Detroit, MI), is a localized 50-kHz current-injection

method that yields a measure of total body resistivity.The method is based on the principle that impedance tothe electrical flow of an injected current is related to thevolume ofa conductor (the human body) and the square

of the length of the conductor (height). Hoffer et al (1)demonstrated that total body water (TBW) and lean body

mass (LBM) were strongly correlated with height2/resis-tance, where body resistivity or impedance was measuredwith a tetrapolar electrode configuration.

Several recent studies (2-4) demonstrated strong cor-relations between BIA (either height2/resistance measuredwith BIA or LBM and TBW predicted from BIA with useof equations provided by the manufacturer) and TBWmeasured by isotope dilution and densitometrically de-termined LBM. However, these studies have usually madeuse of small or heterogeneous samples, and the repro-ducibility of the method between laboratories has not beendetermined by means of cross-validation studies.

The purpose of this study is to cross-validate the BIAmethod by comparing the relationship between BIA anddensitometrically determined LBM at four geographical

sites in large samples of men and women who vary widelyin age and body fat content.

Subjects and methods

Subjects

Fifteen hundred sixty-seven subjects (1069 men, 498 women)aged 17-62 y were studied in four laboratories located in fourdifferent cities in the United States: San Francisco, CA (lab A);New York, NY (lab B); Natick, MA (lab C); and San Diego, CA

I From the Division of Pediatric Cardiology (KRS), Mount Sinai

School ofMedicine, New York, NY; the US Dept of Agriculture (MVL),Western Human Nutrition Research Center, San Francisco, CA; the USArmy Research Institute ofEnvironmental Medicine (PIF), Natick, MA;

the Naval Health Research Center(JAH), San Diego, CA; and the ObesityResearch Center (TBVI), St. Luke’s-Roosevelt Hospital Center, New

York, NY.2 Supported in part by grants from the National Institutes of Arthritis,

Diabetes, and Digestive and Kidney Diseases (grant #AM-26687) andby the Naval Medical Research and Development Command (work unit#M0096.OOl-1050).

3 Address reprint requests to Dr Karen R Segal, Annenberg 3-45,Mount Sinai School ofMedicine, Box 1201, 1 Gustave Levy Place, New

York, NY 10029.Receivedianuary 12, 1987.Accepted for publication March 24, 1987.

by on January 11, 2008 w

ww

.ajcn.orgD

ownloaded from

SEGAL ET AL

S Determined from hydrodensitometry. LBM = lean body mass.

(lab D). After all experimental procedures were explained to the

subjects their written informed consent was obtained. The testprotocol was reviewed and approved by the institutional reviewboard at each of the participating institutions. Each subjectcompleted all measurementS on the same morning. Most of thesubjects were studied after an overnight (12 h) fast and thosewho were not tested after an overnight fast were at least 3 hpostabsorptive.

Densitometry

Body fat content and LBM were determined by densitometry.At lab B, body density was determined by hydrostatic weighingin a stainless steel tank in which a swing seat was suspendedfrom a Chatillon 15-kg scale (Chatillon, New York, NY). Atlabs A, C, and D, the underwater weighing systems were mod-ifications ofthe method ofAkers and Buskirk (5), which makesuse of force transducers. The subjects submerged beneath thesurface of the water while expiring maximally and remained asmotionless as possible at the point of maximal expiration for,� 5 5 while underwater weight was recorded. After several prac-

tice trials to familiarize the subjects with the test procedure, 10trials were performed except in lab A where only 4 trials wereperformed. The estimated underwater weight was the highestvalue that was reproduced three times (6). In labs A, B, and Cresidual lung volume was estimated by means of the closed-circuit oxygen dilution method of Wilmore (7) with use of aCollins spirometer (Warren E Collins, Braintree, MA) and aHewlett Packard Model 47302A (Hewlett-Packard, Cupertino,CA) or a Med-Science Model 505D nitrogen analyzer (FiskeMed-Science, St Louis, MO). In lab D, residual volume wasmeasured by the closed-circuit helium dilution method (8) withuse of a Collins Model 3002 modular lung analyzer (Warren ECollins). In labs B, C, and D, two trials were performed whilethe subjects assumed a sitting position that duplicated body po-sition in the tank during underwater weighing. Residual volumeat lab A was measured in the water at the time ofthe underwaterweighing. Body density was calculated from the formula ofGoldman and Buskirk (9) and percent body fat was derivedfrom body density by use ofthe Siri equation (10): percent bodyfat = (4.95/density) - 4.5. LBM is the difference between totalbody weight and fat weight, where fat weight equals total bodyweight multiplied by percent body fat.

Bioe!ectrical impedance analysis

Total body resistivity was measured with a four-terminal por-table impedance analyzer (RJL Systems, Detroit, MI). Mea-surements were made while the subjects lay comfortably on astretcher with the limbs abducted from the body. Current-injectorelectrodes were placed just below the phalangeral-metacarpaljoint in the middle of the dorsal side of the right hand and justbelow the transverse (metatarsal) arch on the superior side ofthe right foot. Detector electrodes were placed on the posteriorside of the right wrist, midline, with the prominent pisiformbone on the medial (fifth phalangeal) side and ventrally acrossthe medial ankle bone ofthe right ankle with the foot semiflexed.Resistance (R) to the flow ofa 50-kHz injected current was mea-sured on a 0- l000-(� scale and reactance (Xc) was measured ona 0-20041 scale. Empirically derived formulas provided by themanufacturer of the instrument were used to calculate es-timated LBM.

Statistical analyses

Multiple regression analyses were applied to the data fromeach laboratory to derive best-fitting multiple regression equa-

tions to predict densitometrically determined LBM (LBMd) for

each sample. Resistance, reactance, height2/resistance, weight,height2, age, and sex (dummy coded with males, = 0, females= 1) were offered as possible predictors. LBMd was used as thedependent variable. The regressions were carried out in stepwisefashion. Before pooling the data from males and females, theequality of the slopes for males and females was tested for sta-tistical significance (1 1).

A quadruple cross-validation of the equations for predictingLBMd from BIA was carried out according to the proceduredescribed by Lord and Novick (12): the best-fitting equationderived from each data set was applied to the other three datasets. For each data set statistical significance ofthe deviation ofthe regression of LBMd on LBM, cross-predicted with use ofthe equations derived from the other three laboratories, fromthe line of identity was tested. This procedure was followed totest the significance of differences in the best-fitting regressionlines among laboratories (1 1).

Additional statistical analyses ofthe data are described in theresults section. The 0.05 level of significance was used for alldata analyses.

Results

The characteristics of the subjects are shown in TableI . The population varied widely with respect to age andbody composition. The same predictor variables were se-lected by the stepwise regression procedure in all four setsof data: R, ht2, wt, age, and sex. The data from the menand women were treated separately because the regressioncoefficients ofthe best-fitting regression lines were signif-icantly different for men and women. The best-fittingequations for each laboratory are shown in Table 2. Re-sistance and height2, individually, were better predictorsof LBMd than the calculated height2/resistance, as deter-mined by greater correlation coefficients and smaller SEES.

The residuals were analyzed and found to be randomly

TABLE 1

Characteristics of the subjects (mean ± SD)

LabA(n = 96)

LabB(n = 99)

LabC(n = 490)

LabD(n = 404)

MenAge 32± 9 26± 8 34± 8 32± 7Weight(kg) 75±12 79±12 79±12 88±13

Height(cm) 178± 8 179± 7 175± 7 179± 7

LBM(kg)* 61± 8 66± 7 62± 8 67± 8Percent fat 18 ± 7 16 ± 8 22 ± 7 23 ± 8

Resistance(Q) 485±63 459±47 442±55 432 ±49

LabA(n = 64)

LabB(n = 81)

LabC(n = 224)

LabD(n = 141)

WomenAge 35± 9 29±10 24± 5 27± 6

Weight(kg) 59± 8 71±23 61± 8 63± 9

Height(cm) 165± 8 165± 7 163± 6 164± 7LBM (kg)* 43 ± 6 48 ± 7 44 ± 6 45 ± 5

Percent fat* 27 ± 8 29 ± 12 28 ± 6 27 ± 8

Resistance(Q) 587±58 551±68 554±62 559±68

by on January 11, 2008 w

ww

.ajcn.orgD

ownloaded from

LEAN BODY MASS ESTIMATION BY IMPEDANCE 9

TABLE 2Best-fitting equations for predicting lean body mass for each lab andall labs pooled

v�nab1e Lab A Lab B Lab C Lab D All labs

Men

Height2 0.00109 0.00124 0.00122 0.00140 0.00132

Resistance -0.01607 -0.06626 -0.03736 -0.06336 -0.04394

Weight 0.41004 0.26261 0.31973 0.26079 0.30520Age -0. 15407 -0.22776 -0.13038 -0. 15634 -0.16760

Intercept 8.14874 41.35041 19.77883 32.29519 22.66827

R 0.9$ I 0.907 0.882 0.896 0.898

SEE 3.28 2.91 3.62 3.49 3.61

v�uiable Lab A Lab B Lab C Lab D All Labs

Women

Height2 0.00112 0.00114 0.000942 0.00103 0.00108

Resistance -0.03797 -0.02502 -0.01410 -0.02578 -0.02090

Weight 0.21110 0.18856 0.31153 0.22280 0.23199

Age -0. 12953 -0.06498 -0. 14505 -0.01802 -0.06777

Intercept 27.16729 19.25955 10.91436 18.29870 14.59453

R 0.891 0.942 0.876 0.861 0.889

SEE 2.51 2.31 2.15 2.42 2.43

distributed and were uncorrelated with the predicted LBMvalues.

Quadruple cross-validation

The quadruple cross-validation of the equations forpredicting LBMd is shown in Table 3. The purpose of thequadruple cross-validation was to determine the repro-

ducibility across laboratories of the relationship betweenLBMd and LBM predicted from BIA and other variables.LBMd was regressed on LBM predicted by each of theequations in order to determine whether the slopes and

intercepts differed from 1 and 0, respectively, indicatingthat the regression lines differed significantly from the line

TABLE 3

ofidentity. Reductions in the correlation coefficients and

increases in the SEEs resulting from application of theequations derived at other laboratories compared with

the best-fitting equations were minimal. However, asshown in Table 3, differences in regression equations werefound among some ofthe laboratories. For the men thesedifferences were attributable to differences among the lab-

oratories in body fat content: The lab C and lab D menwere significantly fatter than the lab A and lab B men.

When adjustment was made statistically for differencesin body fat content among the four labs, differences amongregression equations were eliminated. Specifically, LBMCI

was regressed on body fat and residualized LBMd valueswere obtained. The residualized LBMd, purged of anyrelationship with body fat, was used as the dependentvariable and stepwise regressions were carried out for eachlab using resistance, height2, weight, and age as the in-

dependent variables. The resulting regression equationswere analyzed for statistical differences among labs. For

both men and women (even though the differences amonglabs in the women’s body fat did not achieve statisticalsignificance) no statistical differences among labs in theregression coefficients were obtained when this adjustmentfor body fat was made. This confounding effect of bodyfatness on the prediction of LBMd supports a previousfinding that the error in predicting LBMd from BIA wassignificantly related to obesity (4).

The relationship between LBM predicted from height2,resistance, weight, and age and LBMd for men and women(all labs pooled but separate equations for men andwomen) is shown in Figures 1 and 2. When the data are

expressed as percent body fat, the correlations betweendensitometrically determined percent body fat and pre-

dicted percent body fat are r = 0.809 and r = 0.852 formen and women, respectively, with SEEs of4.44% fat formen and 3.98% fat for women.

Quadruple cross-validation ofequations for predicting lean body mass (LBM): correlations (and SEE) between densitometrically determined LBM(LBMd) and LBM predicted by best-fitting equation from each lab and all labs pooled

LabA LabB LabC LabD

MenLBM (lab A equation) 0.911 (3.23) 0.854 (3.54)t 0.872 (3.75) 0.854 (4.08)ff

LBM (lab B equation) 0.832 (4.34)t1 0.907 (2.86) 0.860 (3.92)tf 0.892 (3.54)LBM (lab C equation) 0.896 (3.47) 0.883 (3. 19) 0.882 (3.61) 0.884 (3.66)

LBM (lab D equation) 0.853 (4.09)ff 0.902 (2.94) 0.871 (3.77)t� 0.896 (3.48)LBM (all labs equation) 0.886 (3.62) 0.893 (3.06) 0.88 1 (3.63) 0.889 (3.58)

LabA L.abB LabC LabD

WomenLBM (lab A equation) 0.891 (2.45) 0.936 (2.38)tf 0.834 (2.45)tt 0.841 (2.55)tf

LBM (lab B equation) 0.878 (2.58) 0.942 (2.27) 0.854 (2.30) 0.856 (2.44)LBM (lab C equation) 0.859 (2.76) 0.932 (2.44)ff 0.876 (2.15) 0.832 (2.62)LBM (lab D equation) 0.868 (2.68) 0.936 (2.38)ff 0.853 (2.32) 0.861 (2.40)LBM (all labs equation) 0.872 (2.64) 0.940 (2.30) 0.866 (2.22) 0.856 (2.44)

S Best-fitting results for each lab are indicated by italics.

t Intercept significantly different from 0; p < 0.05.:l:Slope significantly different from 1; p < 0.05.

by on January 11, 2008 w

ww

.ajcn.orgD

ownloaded from

90

80-

70-

60-

5Q

40

30

000 5

DO DO

C 0 0 D*C*

CC0O* 0* 0 0

O ODD 0 0

*000*CDS 11CS *0000000*C

*0000000000 C *

DDDO000000000* DO

* *A000IHSO*ODCDO*

DSS0N*0000000 C

#{149}000000000000*OC

C CC00000000* CC*A 00*0*0*00*000

A 0*110000000S0* *

*CC*0*00000*0*

*DC00000CXOC000C

CDA C000*000110 CACC0000000COC C A I k A

ACAC0000COC

C ACC*D*AC B=LabBA C *CCC -

CCC -

CCC D�LabDC C * = Multiple data

points

C r.896SEE =3.62kg

I I I I

and vice versa were significantly different from LBMd0 (Table 6).C Figures 3 and 4 show the relationship between LBMd

and LBM predicted with use of the fatness-specific equa-tions. The dispersion ofdata points is considerably smallerthan when the generalized equations were applied (Figs1 and 2). For men, the R value increased from 0.896 to0.938 and the SEE decreased from 3.62 to 2.84 kg with

use of fatness-specific equations. For women, the multiplecorrelation coefficient R increased from 0.889 to 0.930and the SEE decreased from 2.43 to 1.95 kg with use offatness-specific equations. For the men when the data are

expressed as percent body fat, the correlation betweendensitometrically determined and predicted percent bodyfat increases from 0.809 to 0.896 and the SEE decreasesfrom 4.44% fat to 3.35% fat with use ofthe fatness-specific

equations. For the women the correlation between den-sitometrically determined percent body fat and predictedpercent body fat increases from 0.852 to 0.909 and theSEE decreases from 3.98% fat to 3. 18% fat with use of thefatness-specific equations.

The practical application of the fatness-specific equa-tions is questionable since their use depends on priorknowledge of an individual’s body fat content. Further-more, the fatness-specific equations for predicting LBMdare the basis of categorizing the subjects with respect todensitometrically determined percent body fat, and it isimportant to note that percent fat and LBM are not trulyindependent since both are derived from hydrodensitom-

etry. Although for subjects who are obviously lean or obesethere would be no question as to which prediction is mostappropriate, for subjects who are neither clearly lean nor

.. 40 50 60 70 80 90

LBM(BIA) (kg)

(Best fitting equation for all men)

FIG 1. Relationship between densitometrically determined lean bodymass (LBMd) and lean body mass (LBM) predicted from bioelectricalimpedance analysis (BIA) and other indices with use of the best-fitting

equation for the men in all four laboratories.

Fatness-speafic prediction equations

Additional analyses were applied to the data to char-acterize further the relationship between body fatness andthe prediction of LBMd by BIA and to determine theappropriateness of fatness-specific equations. The totalpopulations of men and women were divided randomlyinto two subsets for the purpose of cross-validation andeach subset was divided into normal and obese groupsbased on whether the subjects were less than or greaterthan 20% and 30% body fat for men and women, respec-tively. For each of the two sets of normal and obese menand normal and obese women (total of eight subsets), anequation for predicting LBMd was derived. The same setof variables (height2, resistance, weight, and age) enteredinto all equations except that age did not enter into theequations for the two sets ofnormal women. The fatness-specific equations and the cross-validation ofthese equa-tions between the two randomly generated sets for eachsex and level of fatness are shown in Tables 4 and 5. Useof these fatness-specific equations greatly improved theaccuracy of predicting LBMd: the multiple correlationcoefficients were significantly increased and the SEES weresignificantly reduced. No differences were found in theregression coefficients between set 1 and set 2 of normal

men, obese men, normal women, or obese women, in-dicating the validity and reproducibility of fatness-specificequations (Table 5). However, significant differences wereobserved between the regression equations for the normaland obese subjects for both men and women. Further-more, the mean predicted LBM values obtained by ap-

plying the normal group’s equations to the obese subjects

0’

SI

E0

(I)C

.�

-I

55 -

50 -

45 -

40 -

35 -

30

A = Lab AB Lab BC = Lab C0 = Lab D11 : Multiple data

points

r =889SEE =2.43 kg

ii I I I I I I I I I

30 40 50 60 70

LBM(BIA) (kg)

( Bestfitting equation for all women)

FIG 2. Relationship between densitometrically determined lean bodymass (LBMd) and lean body mass (LBM) predicted from bioelectricalimpedance analysis (BIA) and other indices with use of the best-fitting

equation for the women in all four laboratories.

SEGAL ET AL

0’

SI

E0

SDCSI

�0

-J

65- S S

55

60-S SB

S SA S

CD.

0 C DOD

*C *C* C

O * *D*C*OSD S

#{149}0*0*00*SCSCC *B** IHI C D

*S**C*C*SCDS

C A***IHHI*** S**IHHI***C *

D *0*5*1155CC CD C

C*****000**C

#{149}S*C*0000 SO C

S*S**IIOOS*0C

*005*5000CC*B*DC*CDCC

C *C**CSCC

AACC *SC#{149}CDC C

CCC D

by on January 11, 2008 w

ww

.ajcn.orgD

ownloaded from

LEAN BODY MASS ESTIMATION BY IMPEDANCE 11

TABLE 4

Fatness-specific equations for predicting LBM

Normal Obese

SetI Set2 Setl Setl Set2 SetIVariable (n=244) (n=228) +set2 (n=295) (n=302) +set2

MenHCight� 0.00060171 0.00071366 0.00066360 0.00072092 0.001020 0.00088580Resistance -0.01959 -0.02319 -0.02 1 17 -0.0542 1 -0.08245 -0.02999Weight 0.65940 0.59597 0.62854 0.48291 0.38179 0.42688Age -0. 14244 -0. 10545 -0. 12380 -0.0542 1 -0.08245 -0.07002

Intercept 8.73968 10.64701 9.33285 11.48504 16.69512 14.52435

R 0.948 0.943 0.946 0.937 0.939 0.937SEE 2.50 2.44 2.47 2.97 3.04 3.03

Normal Obese

VariableSeti

(n= 146)Set2

(n= 177)

SetI+set2

SetI(n=99)

Set2(n=76)

Sell+set2

WomenHeight2 0.00060098 0.00066464 0.00064602 0.000955 14 0.00077596 0.00091186

Resistance -0.018 17 -0.01 121 -0.01 397 -0.01420 -0.01 560 -0.01466Weight 0.40328 0.43868 0.42087 0.3 1 134 0.282 16 0.29990Age - - - -0.07187 -0.06215 -0.07012Intercept 15.26646 7.18338 10.43485 7.47371 14.20227 9.37938

R 0.839 0.903 0.907 0.953 0.954 0.952SEE 1.90 2.00 1.97 2.06 1.81 1.97

obese, there is a need for a technique to determine whichprediction equation is most applicable. The validity offatness-specific equations for predicting LBMd requires amethod of categorizing subjects that is objective and in-dependent ofdensitometry. Body mass index (BMI), bodysurface area (BSA), and percent ofdesirable body weightaccording to the 1959 Metropolitan Life Insurance stan-dards (13) were tested as possible classification criteria.Classification of the subjects by these indices did not sig-nificantly improve the prediction ofLBMd: categorizationofthe subjects according to whether they had a BMI less

than or greater than 26, were below or above the 50thpercentile ofBSA for their sex, or were less than or greaterthan 120% of desirable body weight did not lead to sig-nificant increases in the multiple correlation coefficientsor significant reductions in the SEEs. Each of these threecriteria also was tested as a continuous variable: in separateanalyses, each ofthe indices was entered forcibly into theregression equation to predict LBMd at the first step, and

height2, resistance, weight, and age were entered block-wise at the second step (1 1). Separate analyses were carriedout for men and women. The resulting correlations andSEEs were not significantly different from those obtainedby use ofthe generalized equations derived from all menand all women (individually) pooled together (Figs 1 and2), indicating that BSA, BMI, and percent desirable bodyweight did not significantly improve the prediction ofLBMd.

Use ofanthropometry to categorize subjects

To test whether categorization ofsubjects into two levelsof fatness on the basis of anthropometry, which is inde-

pendent of densitometrically determined body fatness,would significantly improve the prediction of LBMd, anindependent group ofsubjects was studied. These subjects

(88 men, 72 women) underwent hydrostatic weighing, BIAmeasurement, and skinfold thickness measurements.Percent body fat was derived from the sum of the biceps,triceps, suprailiac crest, and subscapular skinfolds withuse of the tabled values of Durnin and Womersely (14).The subjects were divided into normal and obese groupsbased on whether anthropometrically determined percentfat was less than or greater than 20% for men and less

than or greater than 30% for women. Compared withdensitometrically determined percent fat, 82 of 88 men(93%) and 66 of 72 women (92%) were correctly catego-rized with respect to their level of fatness. To determinethe effectiveness of this classification, for each subgroup(normal and obese men, normal and obese women) LBMdwas compared with LBM estimated with use of the gen-eralized equations and LBM estimated with use of thefatness-specific equations. These results are shown inTable 7. Use ofthe generalized BIA equations significantly

underestimated LBMd for both the normal men and nor-mal women. However, the question remained as towhether use of the BIA fatness-specific equations signifi-cantly improved the prediction of LBMd over an-thropometry alone. To answer this question, LBM

values (LBMa) were derived from anthropometrically-determined percent body fat: weight - (weight X percentfat). LBMd was regressed on LBMa and LBM determinedwith use of the fatness-specific equations with LBMa en-tered forcibly at the first step. The significance ofthe entryof LBM derived with the fatness-specific BIA equations

by on January 11, 2008 w

ww

.ajcn.orgD

ownloaded from

SEGAL ET AL

p < 0.01 vs LBMd.

TABLES

Cross-validation of fatness-specific equations for predicting lean body mass*(LBM)

Normal

seti

Normal

set2All normal

men

MenNormal men

LBM (normal set I eq) 0.948 (2.50) 0.942 (2.45) -

LBM (normal set 2 eq) 0.947 (2.51) 0.943 (2.44) -

LBM (obese men’s eq) - - 0.939 (2.60)t

Obese

sellObese

set2All obesemen

Obese men

LBM (obese set 1 eq) 0.937 (2.97) 0.934 (.318) -

LBM (obese set 2 eq) 0.933 (3.06) 0.939 (3.05) -

LBM (normal men’s eq) - - 0.932 (3.l3)t

Normal

set 1

Normal

set 2

All normal

women

WomenNormal women

LBM (normal set I eq) 0.916 (1.90) 0.897 (2.05) -

LBM(normalset2eq) 0.911 (1.93) 0.903(2.00) -

LBM (obese women’s UI) - - 0.897 (2.06)t

Obese

set 1

Obese

set 2

Allobese

women

Obese womenLBM (obese set 1 eq) 0.953 (2.05) 0.953 (1.97) -

LBM (obese set 2 eq) 0.952 (2.06) 0.954 (1.81) -

LBM (normal women’s ui) - - 0.938 (2.20)t

11 Best-fining equations for predicting LBM for normal and obese men andwomen denved from set 1 were applied to set 2 and vice versa. Measured LBM

(LBMd) was regressed on LBM predicted with use ofthe equation developed from

the opposite �t and correlations between measured and cross-predicted LBM andSEES are shown. Effect of fitness-specific equations is indicated by regression of

LBMd on LBM cross predicted by applying obese subjects’ equation to normal

subjects and vice versa.

t Slope significantly different from 1, p < 0.05; intercept significantly different

from 0, p< 0.05.

at the second step was then determined. For both menand women the addition ofLBM derived from the fatness-specific BIA equations significantly improved the predic-

tion of LBMd: the change in the correlation coefficients(from R = 0.934 to 0.943 for men and from R = 0.923to 0.952 for women) was highly significant (p < 0.0001)as were the reduction in the SEEs (from 2.53 to 2.22 kg

TABLE 6

for men and from 2.63 to 2.09 kg for women). Thus,anthropometrically determined percent fat can be usedreliably as the criterion for determining which fatness-specific BIA equation to apply. There is obviously no needto prescreen subjects who are extremely lean or extremelyobese. However, for subjects who are neither clearly leannor obese, anthropometry may be useful in determiningthe optimal BIA prediction equation. It is important tonote, however, that the fatness-specific BIA equations sig-nificantly improve the prediction of LBMd over anthro-

pometry alone.

Manufacturer’s prediction equation

Equations are provided with the BIA instrument forthe prediction of LBM: for men LBM = 6.493+ 0.4936(height2/resistance) + 0.332(weight); and forwomen LBM 5.09 1 + 0.6483(height2/resistance)+ 0. 1699(weight). The correlation between LBMd andLBM predicted with use of the manufacturer’s equationwas r = 0.857 (SEE = 3.70 kg) for men and r = 0.800

(SEE = 3.18 kg) for women. However, the mean predictedLBM was significantly greater than LBMd for both men(69.2 ± 8.52 kg predicted LBM vs 64.0 ± 8.2 kg LBMd)and women (47.3 ± 6.2 kg predicted LBM vs 44.6 ± 5.30kg LBMd), indicating that the manufacturer’s equationsystematically overestimated LBMd for both sexes. Thisoverestimation of LBMd by the manufacturer’s equationwas most apparent in the obese men and women (Ta-ble 6).

Discussion

The results ofthis study confirm the validity ofthe BIAmethod for predicting LBM in large heterogeneous sam-ples of men and women. In contrast to previous reports(2-4), height2 and resistance individually rather thanheight2/resistance were selected by the stepwise regressionprocess. This finding was consistent among the four labs.Also, whereas in previous studies (3, 4) the slopes of theprediction equations were not reliably different betweenmen and women, in the present study significant differ-ence in the regression coefficients were found betweenmen and women. It is possible that discrepancies between

Comparison of mean (±SD) densitometrically determined lean

specific equations and RJL System’s equation

body mass (LBMd) with lean body mass (LBM) cross-p redicted with use of fatness-

Normal men Obese men Normal women Obese women

LBMd 64. 12 ± 7.56 63.95 ± 8.63 44.54 ± 4.64 44.71 ± 6.36LBM (manufacturer’s eq) 66.60 ± 7. 18 7 1 . 19 ± 8.95 46.56 ± 5.44 48.87 ± 7.04LBM (obese men’s eq) 57.98 ± 5.90LBM (normal men’s eq) 71.51 ± 10.1 1

LBM (obese women’s eq) 41.38 ± 3.91LBM (normal women’s eq) 49.63 ± 7.79

S

by on January 11, 2008 w

ww

.ajcn.orgD

ownloaded from

90

80 -

70 -

60 -

50 -

40 -

3C

0’

SI

E0

SDCSI0

-J

#{149}5 0

00

a 000 05.5*.B5 0 B

#{149}5#{149}ss* 0*s0 0500*

#{149}**0*550�

5*0*10*00*011

#{149}*0S*00*� 5* 5

*05*000*0*101

*0000000*0*5

*00*5*110*00

000*11*ONBSI

500*011*0*05 0

5*0*0*0*0000 5

B #{149}**0*S*IHI*0**#{149}0*50*0*5* U

U 0000*0II*OSSU

*0*10*000010

0000*5*011.0#{149}B0**0S�

5* **O

0 505 U

UU0

05UI 0

U U

U U

U

ables in the prediction equation) and LBMd and supportsour finding in a previous study (4) that the error in pre-dicting LBMd was significantly correlated with percentbody fat. In both the previous and the present study wehave observed that when generalized equations are appliedto predict LBMd from resistance, LBMd is significantlyoverestimated in obese subjects. The source of this errorremains unclear. It could be the result of the increasedextracellular water volume in obese subjects, which is re-lated to the water content ofadipose tissue. However, thisincreased hydration ought also to affect the calculatedpercent body fat and LBMd since increased hydrationwould not affect underwater weight but would affect

weight on land thus also affecting this validity of densi-tometry in obese subjects. It is therefore important to ac-knowledge that the criterion method in the present study,densitometry, is not error free with an estimated (theo-retically) maximum precision of 2.5% body fat (1 5). Fur-ther quantification ofthe error ofdensitometry in an obesepopulation is needed, however, perhaps leading to therevision ofthe value used for the density ofthe lean body.

The hydration and bone mineral content of the leanbody are also affected by aging (1 5, 16), which may explainwhy age entered into the prediction equations. Althoughinclusion ofage enhanced the precision ofthe prediction

of LBMd, there was no evidence that use ofa populationwhich varied widely with respect to age introduced anysystematic error into the prediction of LDMd: age wasuncorrelated with the residuals; there were no significantinteractions between age and any ofthe other independent

70-

. .

� 60- :

;� 0�S*L.

; cr� 000000�

E ‘-‘ ‘-‘ 00 *000*0000 #{149}0 0000 00N0S�.� *01155 0#{216}�II) �00sI00�,C 0 0*00*01150

�D 00SS000�*0� �0:::�0:0:� 0 < 3Q% fat

m 40 - #{149}00**0000 #{149}>30%fat._j 000*0000 -

0 55*5000 11 - mu�tIpue uato000#{149}0 points

*0 11

#{149}0#{149} r.93030- #{176} SEE=l.95kg

-‘t�__1�/1 I I I I I 1 1 1

LBM(�l�)(kg)

(Predicted from fatness-specific equation)

FIG 4. Relationship between densitometrically determined lean bodymass (LBMd) and lean body mass (LBM) predicted from bioclectricalimpedance analysis (BIA) use of fatness-specific equations in women.

Separate equations were developed for lean and obese women.

30 40 50 60 70

LEAN BODY MASS ESTIMATION BY IMPEDANCE 13

0 < 20% fat

S > 20%fat5 = Multiple data

points

r .938SEE =2.84kg

1,_I I I I I I I I I I I I40 50 60 70 80 90

L BM(BIA) (kg)

Predicted from fatness-specific equation)

FIG 3. Relationship between densitometrically determined lean bodymass (LBMd) and lean body mass (LBM) predicted from bioelectncalimpedance analysis (BIA) use offatness-specific equations in men. Sep-

arate equations were developed for lean and obese men.

the present and previous studies are due to the greaterpower afforded by the large sample size.

The quadruple cross-validation of BIA equations forpredicting LBMd was fairly successful as evidenced bythe relatively small reductions in correlation coefficientsand increases in the SEEs when the equations developedfrom each lab’s data were applied to the data from theother labs (Table 3). However, differences were observedamong some of the regression equations. Subsequentanalyses indicated that these differences could be attrib-

uted in part to differences among labs in the characteristicsof the subjects. But it is possible that factors such as dif-ferences among labs in the hydrodensitometric method-ologies or in the calibration ofthe BIA instruments couldalso account for the differences in regression equations.To determine the variation between labs in the BIA re-sistance measurements, the same set ofprecision resistors(250-, 400-, 500-, and 750 �I) were measured with the BIAinstruments in lab A and lab B. No more than a 2 �difference was found between the two labs for any of thefour resistors.

When separate equations were developed for men belowand above 20% body fat and for women below and above30% body fat, the prediction of LBMd from BIA was sig-nificantly improved. This cut point for categorizing thesubjects as normal or obese was derived empirically byobserving which break point produced the highest mul-tiple correlation coefficients and smallest SEES within bothhalves ofthe data set. The improvement in the predictionof LBMd with use of fatness-specific equations suggeststhat body fatness has an independent effect on the rela-tionship between BIA resistance (and/or the other van-

by on January 11, 2008 w

ww

.ajcn.orgD

ownloaded from

SEGAL ET AL

TABLE 7Comparison ofdensitometrically determined lean body mass (LBMd) (mean ± SD) with lean body mass (LBM) predicted from generalizedequations and fatness-specific equations in an independent sample ofsubjects categorized on the basis of anthropometrically estimated fatness

according to whether they were less or more than 20% body fat (men) and 30% body fat (women)

Men Women

<20%fat >20%fat <30%fat >30%fat

LBMd 65.02 ± 6.13 67.98 ± 7.88 45.27 ± 4.37 52.12 ± 7.38

LBM (generalized eq) 63.53 ± 5.34 68.07 ± 7.84 43.74 ± 3.96 5 1.7 1 ± 7.80LBM (fat-specific eq) 65.09 ± 6.23 67.92 ± 7.63 44.98 ± 4. 16 5 1.61 ± 8.77

variables; and when the total population was grouped intosubject below and above the age of 35 y, the strength ofthe multiple correlation coefficients was not significantlydifferent between the younger and older groups.

In this study we have not emphasized the conversionof the SEE values from LBM to percent-body-fat unitsbecause the data in this and previous studies indicate thatthe BIA method measures hydrated lean tissue and doesnot measure body fat, either quantitatively (kg of fat) orqualitatively (fat as a percent of total body weight).Whereas hydrodensitometry might be considered a qual-itative body composition method because it yields a mea-sure of the proportional contributions of fat and LBMbased on total body density and assumed constants forthe densities of fat and lean, the BIA method appears tobe quantitatively related to lean mass or, more specifically,to hydrated lean tissue.

The results ofthis study indicate that estimation of LBMby bioelectrical impedance measurement is reproducibleacross independent laboratories and that the precision ofLBM prediction is increased by fatness-specific equations.For subjects who are not easily categorized into normalor obese groups, anthropometry may be used as a screen-ing method and accuracy ofbody composition estimationfurther increased by subsequent use of BIA.

BIA is a rapid, noninvasive, and convenient methodfor human body composition estimation that appears tobe appropriate for nutritional assessment of healthy in-dividuals and groups. However, further research is neededto determine the sensitivity of BIA to detect changes inbody composition within individuals that may occur dur-ing nutritional intervention or physical training. The va-

lidity of the method in sick patients, particularly thosewhose state of hydration may be abnormal also requiresfurther study.

The authors wish to acknowledge Mr Thomas Abernathy ofthe De-

partment ofThomathematics, Mount Sinai School ofMedicine, for valu-able assistance with the electronic transfer ofdata among the four labs.

References

1. Hofl’er EC, Meador CK, Simpson DC. Correlation of whole-body

impedance with total body water volume. J Appl Physiol l96927:531-4.

2. Lukaski HC, Johnson PE, Bolonchuk WW, Lykken GI. Assessment

offat-free mass using bioelectrical impedance measurements of thehuman body. Am J Gin Nutr l985;4l:8l0-7.

3. Lukaski HC, Bolonchuk WW, Hall CB, Siders WA. Validation oftetrapolar bioclectrical impedance method to assess human bodycomposition. J Appl Physiol l986;60:l327-32.

4. Segal KR, Gutin B, Presta E, Wang J, Van Itallie TB. Estimationof human body composition by electrical impedance methods: acomparative study. J Appl Physiol l985;58: 1�6�-71.

5. Akers R, Buskirk ER. An underwaterweighing system utilizing “forcecube” transducers. J Appl Physiol 1969;26:649-52.

6. Behnke AR, Wilmore JH. Evaluation ofbody build and composition.Englewood Cliffs. NJ:Prentic Hall, 1974.

7. Wilmore JH. A simplified method for determination of residuallung volume. J Appl Physiol l969,27:96-lOO.

8. Ruppel G. Manual of pulmonary function testing. St Louis: CV

Mosby, 1975.9. Goldman RF, BuSkirk ER. Body volume measurements by under-

water weighing description ofa method. In: Brozek J, Henschel A,

eds. Techniques for measuring body composition. Washington, DC:National Acadamy ofScience, 1961:78-89. -

10. Sin WE. Body composition from fluid spaces and density: analysisofmethods. In: Brozek J, Henschel A, eds. Techniques for measuringbody composition. Washington, DC: National Acadamy of Science,

1961:223-44.1 1. Pedhazur El. Multiple regression in behavioral research. 2nd ed.

New York: Holt, Rinehart, & Winston, 1973.12. Lord FM, Novick MR. Statistical theories of mental test scores.

Reading, MA: Addison-Wesley, 1968.13. Metropolitan Life Insurance Company. New weight standards for

men and women. Stat Bull Metropol Life Insur Co l959;40:l-4.14. Durnin JVGA, Womersley J. Body fat assessed from total body

density and its estimation from skinfold thicknesses: measurementson 481 men and women aged 16 to 72 years. Br J Nutr l974;32:77-92.

15. Lohman TG. Skinfolds and body density and their relation to body

fatness: a review. Hum Biol l98l;53:l81-225.

16. Wedgewood Ri. Inconstancy ofthe lean body mass. Ann NY AcadSci l963;llO:141-l52.

by on January 11, 2008 w

ww

.ajcn.orgD

ownloaded from