594 Am iC/in Nuir 199 1;53:594-8. Printed in USA. © 1991 American Society for Clinical Nutrition

Original Research Communications-general

Body composition in Pima Indians:validation of bioelectrical resistance1’2

Russell Rising, Boyd Swinburn, Karen Larson, and Eric Ravussin

ABSTRACF To assess the validity of bioelectrical resistance

(BR) in an obese population, body composition was determined

by both hydrostatic weighing and by BR in 156 Pima Indian

volunteers representing a wide range of body weight (46.1-202.6

kg) and body composition (1 1-52% fat). A predictive equation

was derived by use of data on height, BR, weight, age, and sex

from 130 randomly selected volunteers and was applied to theremaining 26 volunteers. When compared with the manufac-

turer’s software, the new equation increased correlations withhydrostatic weighing for predicting percent body fat and fat-free

mass (FFM) from 0.70 to 0.92 and 0.79 to 0.97, respectively.

The manufacturer’s software underestimated FFM by 5.3 ± 8.6

kg (P < 0.05) when compared with FFM derived from hydro-

static weighing whereas the new equation improved the accuracy

to -0. 1 ± 3.3 kg (NS). There were no significant effects of fluid

intake (700 mL) or breakfast consumption on body composition

as determined by BR. BR represents a simple and accurate way

to assess body composition in Pima Indians with our newly de-

rived equation. Am J C/in Nuir 199 1;53:594-8.

KEY WORDS Bioelectrical impedance, obesity, fat-free

mass, fat mass

Introduction

The Pima Indians are a relatively genetically homogeneous

group of people living southeast of Phoenix, AZ. They have thehighest incidence of non-insulin-dependent diabetes in the world

and a very high prevalence of obesity (1). In metabolic studies

an accurate assessment of body composition to estimate fat-free

mass (FFM) and fat mass is important so that results can beexpressed in terms ofthe volunteer’s metabolic size. Bioelectrical-resistance (BR) analysis (2-5) offers an inexpensive, rapid, and

safe method for estimating body composition without the needfor a hydrostatic weighing facility.

In general, studies validating BR analysis for determining bodycomposition used sample populations of various ages and in-

cluding many racial groups (6, 7). These sample populations did

not include extremely obese individuals (> 50% fat) or individ-

uals from Native American Indian populations. Total body waterwas found to be similar in young black and white men (8) al-though blacks had a higher body density, possibly because ofgreater bone mineral content in their FFM. In people with var-ious degrees of obesity, fat-specific regression equations (9) were

derived by use of anthropometry to classify men and women

into categories of> 20% and < 30% fat, respectively. These fat-

specific equations (9) increased correlations between FFM de-

termined by hydrostatic weighing and FFM predicted by resis-

tance from 0.90 to 0.94 and from 0.89 to 0.93 for men andwomen, respectively.

The degree ofobesity observed in the Pima Indian community

is far greater than that measured in the populations previously

used to validate the BR method. To validate this method inPima Indians, FFM was estimated by both hydrostatic weighing

and BR in 156 Pima Indians representing a wide range of bodyweight (46. 1-202.6 kg) and body composition (1 1-52% body

fat). A new equation was derived in 130 Pima Indian volunteersand was tested on the remaining 26 volunteers. To test its validity

in field conditions, the effect offood and fluid ingestion on bodycomposition by BR was also assessed in another 1 7 Pima Indian

volunteers.

Methods

Subjects

One hundred fifty-six Pima Indian volunteers (92 males, 64females) were admitted to the Clinical Diabetes and Nutrition

Section ofthe National Institutes of Health in Phoenix, AZ, forvarious studies of carbohydrate and energy metabolism over a3-y period. All volunteers were fed a weight-maintenance diet

(50% ofenergy as carbohydrate, 30% as fat, and 20% as protein)

during their stay in the metabolic ward. Upon admission, allvolunteers were determined to be in good health by means of

medical history, physical examination, electrocardiogram, blood

screening, and urine testing. Volunteers were screened for dia-

betes mellitus by an oral-glucose-tolerance test, according to the

procedure of the National Diabetes Data Group (10), but were

not excluded from the study if they were diabetic. None was

taking any medication or had clinical evidence of illness apartfrom obesity.

Body composition of each volunteer was determined by hy-

drostatic weighing (1 1), and percent body fat was calculated ac-

cording to the equation of Keys and Brozek (12). FFM was

I From the Clinical Diabetes and Nutrition Section, National Instituteof Diabetes and Digestive and Kidney Diseases, National Institutes ofHealth, Phoenix, AZ.2 Address reprint requests to R Rising, National Institutes of Health,4212 North 16th Street, Room 541-A, Phoenix, AZ 85016.

Received April 27, 1990.Accepted for publication July 18, 1990.

by on Decem

ber 24, 2007 w

ww

.ajcn.orgD

ownloaded from

BIOELECTRICAL RESISTANCE IN PIMA INDIANS 595

C Results are pooled from five repeated measurements.

t Significantly greater than the value before fluid, P < 0.05 (paired I test).

TABLE IPhysical characteristics of 156 Pima Indians in whom a prediction

equation to estimate body composition by bioelectrical resistance was

derived and tested

Predictive equation(n = 130)

Testing of equation(n = 26)

Sex ratio (M:F) 74:56 18:8

Age(y) 30 ± 8 32 ±10Weight (kg) 99.3 ± 26.7 106.8 ± 24.8Height (cm) 166 ± 9 168 ± 9

Fat, hydrostatic (%) 34 ± 9 34 ± 8Fat, resistance (%) 38 ± 9 39 ± 9FFM, hydrostatic (kg) 63.5 ± 13.4 69.3 ± 13.5FFM, resistance (kg) 58.6 ± 1 1.7 64.0 ± 12.9

S � � SD.

calculated from the difference between the volunteers’ bodyweight and their fat mass. BR was measured between 0700 and

0800 after an 1 1-12-h fast. For the resistance measurement vol-

unteers were instructed to lie flat with their hands at their sidesand with their thighs maintained apart. Electrodes (Conmed

Corp, Utica, NY) were placed on the subjects and resistance

measurements were obtained with a bioelectrical impedance an-

alyzer (model BIA-l03, RJL Systems, Inc, Detroit) accordingto instructions provided by the instrument manufacturer. Percent

body fat was calculated by using the manufacturer’s software(Bodycomp II, version 1. 1, RJL Systems, Inc). The instrumentwas calibrated periodically with a 50041 resistor provided by the

manufacturer, and measurements were continued as long as the

instrument readings remained within 2 �2 of the calibrated re-

sistor. The experimental protocol was approved by the NIDDK

Clinical Research Subpanel and the Tribal Council of the Gila

River Indian Community. Written informed consent was ob-

tamed from the volunteers. All physical characteristics of the

volunteers are expressed as mean ± SD.

TABLE 2

Predictive equation

Ofthe 156 volunteers whose characteristics are shown in Table

1 ( 123 nondiabetic, 33 diabetic), 1 30 (74 males, 56 females)were randomly selected (13) for the derivation ofa new predictive

equation, whereas 26 (18 males, 8 females) were utilized fortesting the equation. Percent body fat was calculated by use of

the new prediction equation by subtracting FFM from the vol-

unteer’s body weight to determine fat mass, then dividing fat

mass by total body weight, multiplied by 100. There were nosignificant differences in age, height, body weight, and bodycomposition between the subjects used to derive the new equa-

tion and those utilized for testing this equation.

Effect offluid intake

To test the effect of hydration, resistance measurements wereobtained according to the above protocol on another eight malePima Indian volunteers (aged 35 ± 14 y, with 25 ± 10% bodyfat by hydrostatic weighing) 1 h before and 1 h after consuming

700 mL of fluid (water or diet soda). Percent body fat was cal-culated by use of the new prediction equation. The volunteers’physical characteristics are presented in Table 2. To test thereproducibility of the resistance measurements, this procedure

was repeated for 5 consecutive days. The mean CVs of the per-cent body fat determinations over the 5-d period were 3.6% and

3.8%, before and after fluid intake, respectively. Because of thesmall variability ofthe repeated measurements, data were pooledfor the 5 d (Table 2).

Effect ofbreakfast ingestion

Resistance measurements were also obtained on another nine

Pima Indian volunteers (eight males, one female) (aged 30 ± 10

y, with 28 ± 9% body fat by hydrostatic weighing) 1 h beforeand 1 h after eating breakfast. Their physical characteristics are

shown in Table 3. The measurements were repeated for 2 con-secutive days. Percent body fat was calculated by use ofthe newprediction equation, according to the above procedure. Because

the mean CVs ofthe percent body fat were only 2. 1% and 2.8%

Physical characteristics and mean changes in resistance (AR) and percent body fat (iWat) for eight volunteers before and after 700 mL of fluidintake

Subject Height Weight

Fluid status

i�R �Fat

Before After

R Fat R Fat

cm kg �l % Q % (1 %

I

2

3

4

56

7

8

i±SD

177

169

165

170

173170

163

172

170±4

122.6

131.6

73.5

80.5

91.676.8

52.7

105.0

91.7±26.5

380

393

490

449

482

503

604

477

472±70

41

39

18

15

25

15

3

34

24± 13

392

402

511

460

489

493

616

481

481±70

42

39

19

16

2515

3

34

24±13

+12

+9

+21

+11

+7-10

+12

+4

8±9t

+1

0

+1

+1

00

0

0

0±1

by on Decem

ber 24, 2007 w

ww

.ajcn.orgD

ownloaded from

40

30

20

10

.%

+ 0.33 X (body weight) - 0.14 X (age) + 6.18 X (sex)

where height is expressed in cm, resistance in �, weight in kg,

and age in y; a value of 1 is given for males and 0 for females.In the 26 volunteers for whom the above new equation was

applied, the coefficient of correlation and SEE between percent

body fat from resistance and percent body fat from hydrostatic

TABLE 4Coefficient and SEE for the determinants of FFM (kg), by bioelectricalresistance, in I 30 Pima Indians

10 20 30 40 500 60

% Body fat by hydrostatic weighing

FIG 1. Variation of percent body fat, by bioclectrical resistance (by use

of the manufacturer’s software), compared with percent body fat, byhydrostatic weighing, for 156 Pima Indians (P < 0.0001), SEE 5.04,

r = 0.83). Points plotted along the line of identity.

r SEE F P

Intercept(kg) 13.74 1.52 - 0.0001Ht2/R (cm2/tI) 0.34 0.05 42.8 0.0001Body weight (kg) 0.33 0.02 324.8 0.0001Age (y) -0.14 0.03 18.3 0.0001

Sex (0 or I ) 6. 18 0.9 1 44.4 0.0001

596 RISING ET AL

TABLE 3Physical characteristics and mean changes in resistance (SR) and percent body fat (�Fat) for nine volunteers before and after breakfast5

Breakfast consumption

Before After

Subject Height Weight R Fat Weight R Fat �R �Fat

cm kg % kg t� % t� %

1 178 118.4 500 41 118.8 490 41 -10 0

2 177 123.2 382 42 123.8 377 42 -5 03 168 59.6 588 7 59.3 609 8 +21 +14 164 85.2 499 22 85.4 500 22 +2 0

5 172 118.7 435 41 118.3 429 40 -6 -16 159 86.6 606 35 87.0 612 35 +6 0

7 176 92.0 560 27 92.4 530 26 -30 -1

8 170 116.7 384 39 117.4 385 39 +1 0

9 173 138.1 439 53 138.6 427 53 -12 0

i±SD 171 ±6 104.3±24.7 488±84 34± 14 104.5±24.9 484±88 34± 13 -4± 14 0± 1

S Results are pooled from two repeated measurements.

before and after the standardized breakfast, respectively, mean

results over the 2 d are presented in Table 3.

Data analysis

All data were analyzed by use ofthe StatisticalAnalysis Systempackage (SAS Institute, Cary, NC). Stepwise multiple-regression

analysis (forward selection technique) was used to derive the

best predictive equation with FFM, determined by hydrostatic

weighing as the dependent variable, and height (cm), body weight

(kg), height/resistance (cm2/�2), reactance (t�1), body mass index

(kg/m2), age (y), body weight#{176}”, and sex (class variable; male =

1 , female = 0) were entered as independent variables. Indepen-dent variables were eliminated if they did not meet the signifi-

cance level of P < 0. 1 5 for entry into the model.To test the new prediction equation, percent fat and FFM

values from both the manufacturer’s software and from the new

prediction equation were regressed on percent fat and FFM,

� 60Ca

.� 5�

a)

0U

Ua)a)0

.�

>�

.0

0

>,�00

determined by hydrostatic weighing. Statistical significance of

differences between pooled means before and after hydration

and breakfast, respectively, were assessed by paired I test.

Results

Figure 1 shows the relationship between percent body fat (by

resistance) calculated by use ofthe manufacturer’s software, and

percent body fat (by hydrostatic weighing) in the 1 56 volunteers

used to derive and test a new predictive equation. At any given

percent body fat measured by hydrostatic weighing, there is alarge range ofpercent body fat estimated from BR. In addition,

BR overestimated percent body fat. From the multiple-regression

analysis with use of FFM (by hydrostatic weighing) as the de-

pendent variable, Ht2/R, body weight, age, and sex were signif-

icant determinants of FFM in Pima Indians. The following pre-

diction equation provides the best estimate of FFM (Table 4):

FFM (kg) = 13.74 + 0.34 X (Ht2/R)

by on Decem

ber 24, 2007 w

ww

.ajcn.orgD

ownloaded from

BIOELECTRICAL RESISTANCE IN PIMA INDIANS

100

90

80

70

60

50

40

30

597

.

#{149}S

C

0

0

a.w5)

5)L

U0

C0

C

0

0

0�

wa)z

0’.�

C)C0

U)

Ina)

0UL

C)a)a)0

.0

>‘.0

InU)

0

Ea)a)L

0

1(

0 �0 50 60 70 80 90100

S

S

S

40 50 60 70 80 90 100

Fat-free moss by hydrostatic weighing (kg)

C0

a

U)

a)

Ua

C0

C0

0

L4J

a)z

a)0C

0

C’,

‘I)

a)

00

0

a)

a)0

n>‘

0

>��00

In

10 20 30 40 50

% Body fat by hydrostatic weighing

weighing increased from 0.70 (SEE 6.89) when the manufac-

turer’s software was used, to 0.92 (SEE 3.22) when the new

equation was used (Fig 2). Similar improvements in coefficients

of correlation with the new equation were found when FFM

(predicted from resistance or determined by hydrostatic weigh-

ing) were compared (Fig 3). Correlation coefficients for percent

body fat and FFM, between resistance (new equation) and hy-

drostatic weighing, were not equal (0.92 and 0.97, respectively)

because the variables used for the comparison (kg FFM and

percent body fat) were not directly proportional to each other.

For the 26 test volunteers, the new prediction equation im-

proved the mean accuracy of predicting FFM from -5.3 ± 8.6

kg (P < 0.05 for difference from 0) to -0. 1 ± 3.3 kg (NS from

0) of the mean value (69.3 ± 13.5 kg) obtained by hydrostatic

weighing. A similar magnitude of improvement was found for

percent body fat with values decreasing from an overestimate

of 5 ± 7% (P < 0.05 for comparison with 0) to 0 ± 3% (NS for

comparison with 0) of the mean value (34 ± 8%) obtained with

hydrostatic weighing.

In the eight subjects who participated in the fluid intake study,

the individual CV for five repeated resistance measurements

varied from 2.0% to 4.3% before fluid intake and from 2.8% to

5.1% after fluid intake. Mean resistance increased slightly from

472 ± 70 121 h before fluid intake to 481 ± 70 12 (P < 0.05) 1

h after fluid intake (Table 2). This increase did not result in a

FIG 2. Percent body fat, by bioelectrical resistance (by use of the man-

ufacturer’s software, upper panel) and by the new prediction equation(lower panel) vs percent body fat, by hydrostatic weighing, in 26 Pima

Indians (P < 0.0001, SEE 6.89 (r = 0.70) and 3.22 (r = 0.92) for theupper and lower panels, respectively). Points plotted along the line ofidentity.

FIG 3. Fat-free mass (kg), by bloelectrical resistance, (by use ofthe man-ufacturer’s software) (upper panel) and the new prediction equation (lower

panel) vs fat-free mass (kg), by hydrostatic weighing, in 26 Pima Indians

(P < 0.0001 ), SEE 8. 10 (r = 0.79) and 3.30 (r = 0.97) for the upper andlower panels, respectively). Points plotted along the line of identity.

significant change in percent body fat when estimated by resis-

tance.

For each volunteer in the breakfast study, mean resistance,

and therefore estimate of percent body fat, did not change inresponse to breakfast consumption (Table 3).

Discussion

BR, a simple and inexpensive method, does not accurately

estimate body composition in Pima Indians when the equations

contained in the manufacturer’s software are used. Use of an

equation specific for Pima Indians improved the average accu-

racy of estimating FFM and percent body fat to within -0.1

± 3.3 kg and 0 ± 3%, respectively, of the values obtained by

hydrostatic weighing. This new equation will provide a reason-

ably accurate determination of body composition in this pop-

ulation.

Hydrostatic weighing is the most widely used standard for60 determining body composition in humanc and is considered to

be the reference standard against which new methods are tested

(1 1). We found hydrostatic weighing to be satisfactory (CV 7%)

(14). Therefore, the large variation in percent body fat estimated

by BR at any given percent body fat, derived from hydrostatic

weighing, indicates that the method is not suitable in Pima In-

dians when the manufacturer’s software is used. Indeed, other

equations have been derived from data on groups of subjects

by on Decem

ber 24, 2007 w

ww

.ajcn.orgD

ownloaded from

598 RISING ET AL

from other races (6-8), but more importantly, not representing

the degree of obesity found in Pima Indians. Therefore, a newequation needed to be derived for Pima Indians to easily estimatebody composition in field conditions. We found that such a new

equation can greatly improve the measurement of body corn-

position in the Pima Indian population.

The improvement in the accuracy ofdetermining body corn-

position with BR in Pima Indians, by use ofan equation derived

in this population, confirms the results ofother studies of bodycomposition in other populations (8, 9). Young black men had

higher body densities than did young white men with the sametotal body water content (8). Also, the accuracy of predictingFFM from resistance analysis was improved when subjects were

grouped as normal or obese depending on their degree of obes-

ity (9).

There was no effect of fluid intake or breakfast ingestion on

body composition estimated by BR. However, the small but

significant increase in BR observed 1 h after fluid intake was

unexpected because an increase in the conductive medium (wa-ter) theoretically should have reduced the resistance to current

flow (5). It is possible that remeasuring resistance only 1 h afterfluid ingestion does not allow enough time for complete fluid

absorption and equilibration of the body water pool to occur.Other investigators did not find a decrease in resistance in vol-unteers rehydrated after an 18-h fast (15). On the other hand,resistance did change according to water balance when measured

daily in obese woman who fasted for 14 d with limited fluidintake, and were refed for an additional 5 d (16). Thus, majorshifts in electrolyte and fluid balance may influence the results.

As implied by Segal et al (9), use of Ht2/R in our prediction

equation may mask small changes in percent body fat with small

changes in resistance. Another prediction equation (not shown)was derived with a separate coefficient for resistance to determine

ifthis type ofequation would show changes in percent body fatin response to fluid or breakfast intake. This latter equation did

not show any effect of fluid or breakfast intake on body corn-position. Also, coefficients were very similar to those presentedby Segal ci al (9).

In conclusion, BR is an accurate and reproducible method

for determining body composition in Pima Indians of both sexes

when a new predictive equation was used. Fluid or food ingestion

shortly before resistance measurements does not affect the out-come of body composition determinations.

We thank Carol Lamkin and the nursing staff of the Clinical Diabetes

and Nutrition Section for their professional care of the volunteers andfor their assistance in performing the resistance measurements; TomAnderson for his help with the hydrostatic weighing; RJL Systems In-

corporated, Detroit, for providing the equipment for the study; and most

importantly, the volunteers.

References

1. Knowler WC, Peuitt DJ, Savage PJ, Bennet PH. Diabetes incidence

in Pima Indians: contribution ofobesity and parental diabetes. AmJ Epidemiol l98l;l 13:144-56.

2. Nyboer J, Bango S. Barnett A, Halsey RH. Radiocardiograms-theelectrical impedance changes of the heart in relation to electrocar-

diograms and heart sounds. J Clin Invest I940;l9:963.3. Thomasett A. Bioelectrical properties of tissue impedance. Lyon

Med l962;207: 107-18.

4. Hoffer EC, Meador CK, Simpson DC. Correlation of whole body

impedance with total body water volume. J Appi Physiol l969;27:

531-4.

5. Nyboeri, Liedtke Ri, Reid KA, Gessert WA. Nontraumatic electrical

detection of total body water and density in man. Med Jadertina

l983;l5:38l-4.

6. Segal KR, Gutin B, Presta E, Wang J, Van Italic TB. Estimationof human body composition by electrical impedance methods: a

comparative study. J Appl Physiol l985;58: 1565-71.7. Lukaski HC, Johnson PE, Bolonchuk WW, Lykken 01. Assessment

offat free mass using bio-electrical impedance measurements of thehuman body. Am J Clin Nutr 1985;41:810-7.

8. Schutte JE, Townsend EJ, HuggJ, Shoup RF, Malina RM, Blomqvist

CG. Density oflean body mass is greater in blacks than in whites.J AppI Physiol l984;56: 1647-9.

9. Segal KR, Loan MV, Fitzgerald P1, Hodgdon JA, Van Itallie TB.

Lean body mass estimation by bioelectrical impedance analysis: a

four-site cross-validation study. Am J Clin Nutr 1988;47:7-l4.10. National Diabetes Data Group. Classification and diagnosis of di-

abetes mellitus and other categories ofglucose intolerance. Diabetes

l979;28: 1039-57.1 1. Goldman RF, Buskirk ER. A method of underwater weighing and

the determination of body density. In: Brozek J, Herschel A, eds.Techniques for measuring body composition. Washington, DC: Na-

tional Academy of Sciences, 196 1:78-89.12. Keys A, Brozek J. Body fat in adult man. Physiol Rev l953;33:245-

325.13. Fishman GS, Moore LR. A statistical evaluation of multiplicate

congruential generators with modulus (2�’-1). J Am Stat Assocl982;77: 129-36.

14. Bogardus C, Lillioja S. Mott DM, Hollenbeck C, Reaven G. Rela-

tionship between degree of obesity and in vivo insulin action inman. Am J Physiol l985;248 (Endocrinol Metab 1 l):E286-9 1.

15. Stump CS, Houtkooper LB, Hewitt Mi, Going SB, Lohman TG.Bioelectric impedance variability with dehydration and exercise. MedSci Sports Exerc l987;20:S82 (abstr).

16. Gray DS. Changes in bioelectricalimpedance during fasting. Am JClin Nutr l988;48:l 184-7.

by on Decem

ber 24, 2007 w

ww

.ajcn.orgD

ownloaded from