Validity of Body Adiposity Index in Predicting Body Fat in aSample of Brazilian WomenMatheus Cerqueira1,2, Paulo Amorim1, Filipe Magalh~aes3, Eliane Castro1, Frederico Franco2, Sylvia Franceschini4,Lizbeth Cerqueira4, Jo~ao Marins1 and Leonice Doimo1

Objective: This study aimed to verify the validity of BAI in predicting %BF in a sample of Brazilian

women

Design and Methods: A total of 102 women (average age 60.3 6 9.8) were assessed. To determine per-

centage body fat (% BF), dual-energy X-ray absorptiometry (DXA) was used as the “gold standard.” To

evaluate the association between body adiposity index (BAI) and % BF assessed by DXA, we used Pear-

son’s correlation coefficient. Paired sample t-test was used to test differences in mean % BF between

BAI and DXA. To evaluate the concordance between % BF measured by DXA and estimated by BAI, we

used the Lin’s concordance correlation coefficient and the agreement analysis of Bland-Altman.

Results: The correlation between % BF obtained by DXA and that estimated by BAI was r 5 0.65, P <

0.001. Paired t-test showed significant mean difference between methods (P < 0.0001). Lin’s concordance

correlation coefficient was C_b 5 0.73, which is classified as poor, while the Bland-Altman plots showed

BAI underestimating % BF in relation to the used criterion measure in a large portion of the sample.

Conclusions: Results of the present study show that BAI presented low agreement with % BF measured

by DXA, which is not recommended for % BF prediction in this studied sample.

Obesity (2013) 21, E696–E699. doi:10.1002/oby.20543

IntroductionAnthropometric methods of body composition assessment using

weight measurements, height and body circumferences (1) have

been used as an alternative to laboratory methods, which are costly

and require sophisticated equipment and skilled professionals, mak-

ing them impractical for evaluating large population groups (2).

Anthropometric methods are practical, quick and inexpensive and

can be easily applied to large samples (3). However, despite show-

ing an association with body fat, they do not allow the breakdown

of the constituent elements of body composition (4).

Because of this limitation, a new index (5) was recently proposed

based on measurements of height and hip circumference for estimat-

ing body fat in a simple and easy way. This index, called body adi-

posity index (BAI), was developed with data from Mexican Ameri-

can adults of both sexes, from 18 to 67 years old and tested in an

African American sample. The method proved to be valid for esti-

mating percentage body fat (% BF) in these populations, representing

an evolution in anthropometric methods using simple height and hip

circumference measures to estimate body composition does a two-

compartment segmentation of the body in fat mass and fat free mass.

BAI has been analyzed in comparison with other methods and as to

its validity in determining % BF. Some studies have indicated that

BAI is not a more accurate measure of adiposity than is body mass

index (BMI) (6-10), waist circumference, or hip circumference

(11,12), as was also observed that the BAI was a better predictor of

adiposity than BMI (13-15).

In relation to the analysis of % BF determined by BAI, the compari-

son a group of women with familial partial lipodystrophy and healthy

women with the same age and BMI, lipodystrophic women had sig-

nificant differences in fat content and distribution as evaluated by

dual-energy X-ray absorptiometry (DXA), and BAI was able to catch

differences in the total of body fat content between groups, as well as

DXA (14). Was also found that BAI estimates % BF with high accu-

racy in nondialyzed chronic kidney disease patients (16). Opposite

results shown that BAI does not provide valid estimates of % BF for

a Caucasian, European population (17), European American adults

(15) and in athletic women (18), determining that this method should

not be used for predicting individual % BF.

Although some studies verify the validity of the BAI, was observed

great lack of standardization in the criteria used by the studies, as

1 Department of Physical Education, Federal University of Vicosa, Minas Gerais, Brazil. Correspondence: Matheus Cerqueira(matheus.cerqueira@ifsudestemg.edu.br) 2 Federal Institute for Education,Sciences and Technology, Rio Pomba, Minas Gerais, Brazil 3 Iguacu University,Rio de Janeiro, Brazil 4 Department of Nutrition and Health, Federal University of Vicosa, Minas Gerais, Brazil

Disclosure: The authors declared no conflict of interest.

Author contributions: MC, PA, and LD conceived and carried out experiments. FM, EC, and FF conceived experiments and analysed data. SF, LC, and JM carried out

experiments. All authors were involved in writing the article and had final approval of the submitted and published versions.

Received: 13 March 2013; Accepted: 1 May 2013; Published online 26 June 2013. doi:10.1002/oby.20543

E696 Obesity | VOLUME 21 | NUMBER 12 | DECEMBER 2013 www.obesityjournal.org

Original ArticleOBESITY BIOLOGY AND INTEGRATED PHYSIOLOGY

Obesity

well as controversial results. The authors (5) considered that theresults of the BAI could be extrapolated to other populations in Cen-tral and South America; however, they suggested further researchwas needed to validate and confirm the results.

Thus, this study aimed to verify the validity of BAI in predicting %

BF in a sample of Brazilian women, using dual-energy X-ray

absorptiometry (DXA) as the reference method.

MethodsThis was a cross-sectional analytical study carried out on 102

women with a mean age of 60.3 6 9.8 ranging from 35 to 83 years

old. They participated in an exercise project open to the public of a

federal university in Brazil, and the baseline data was used in this

study. With regard to skin color, the study included whites, inter-

mediates and blacks. However, because of high individual ancestral

variability, Brazilian’s have a singular proportion of Amerindian,

European and African ancestries in their genome, and cannot predict

ethnic group of persons from their skin color (19).

The project was approved by the Ethics Committee for Research of

Vicosa Federal University, Minas Gerais State, Brazil, and informed

written consent was obtained from all participants.

Anthropometric dataThe hip circumference measurement was taken at the point of the

largest circumference of the buttocks, with a flexible tape, with a

precision of 0.1 cm. Body height was measured to the nearest 0.5

cm and weight was measured in individuals wearing light clothes to

the nearest 0.1 kg on a balance beam scale. The mean of the three

measures were recorded. All measurements were performed at the

Human Performance Laboratory at the Federal University of Vicosa.

With these anthropometric measurements, BMI and BAI were calcu-

lated by the following equations: BMI 5 weight (kg)/height2 (m),

BAI 5 (hip circumference (cm)/height (m)1.5) 2 18.

Body composition by DXAMeasurements of total body composition were determined by the

dual-energy X-ray absorptiometry (DXA) method. The tests were

carried out by a qualified and experienced technician in medical

radiology using a densitometer (GE Healthcare Lunar Prodigy

Advance DXA System, software version13.31). To ensure data

quality the equipment has been calibrated daily using a known

calibration standard and weekly using a step-wedge phantom, fol-

lowing manufacturer instructions.

Statistical analysisDescriptive statistics were calculated for age, height, hip circumfer-

ence, weight, BMI, % BF measured by DXA and estimated by BAI,

and are expressed as mean 6 standard deviation and minimum e

maximum values. To evaluate the association between BAI and %

BF assessed by DXA, we used Pearson’s correlation coefficient.

Paired sample t-test was used to test differences in mean % BF

between BAI and DXA. Lin’s concordance correlation coefficient

(20) was used to assess the reproducibility between BAI and DXA.

Mcbride (21) classifies the strength of agreement as poor (<0.90),

moderate (0.90-0.95), substantial (0.95-0.99), and almost perfect

(>0.99). The plot of the differences between DXA and BAI was

showed by the Bland–Altman procedure (22). Analyses were per-

formed using a statistical software (MedCalc version 11.5.1, Maria-

kerke, Belgium) and we adopted a significance level of P < 0.05.

ResultsAge and body characteristics of the studied sample are in Table 1.

The correlation analysis of % BF determined by DXA and BAI was

significant (r 5 0.65; P < 0.001). Paired t-test showed significant

mean difference between methods [t(101) 5 27.07; P < 0.0001].

The strength of correlation between the % BF measured by DXA

and that estimated by BAI in accordance with Lin’s concordance

correlation coefficient was C_b 5 0.73. The plot of the differences

between DXA and BAI showed by the Bland-Altman procedure (22)

can be seen in Figure. 1

The mean (6SD) % BF of the DXA method was 36.91 (66.21),

compared with 33.61 (63.77) for the BAI. The bias (SD) of the

BAI is 3.29 6 4.71% BF (95% CI 5 2.37-4.22), indicating that the

TABLE 1 DXA measurements of % BF and anthropometriccharacteristics of participants in the study

Variable Mean (SD) Min-Max

Age 60.3 (9.8) 35.0-83.0

Height (m) 1.55 (5.4) 1.44-1.66

Weight (kg) 64.5 (8.4) 45.1-95.0

BMI (kg/m2) 26.9 (3.1) 20.1-36.8

% BF (DXA, %) 36.9 (6.2) 22.5-48.8

% BF (BAI, %) 33.6 (3.8) 25.9-42.9

Hip (cm) 99.3 (6.6) 84.0-124.0

FIGURE 1 Agreement limits of Bland–Altman between the percentage of body fatmeasured by DXA and that estimated by BAI.

Original Article ObesityOBESITY BIOLOGY AND INTEGRATED PHYSIOLOGY

www.obesityjournal.org Obesity | VOLUME 21 | NUMBER 12 | DECEMBER 2013 E697

BAI measured lower % BF than the DXA. The lower limit of agree-

ment 25.93 (95% CI 5 27.52 to 24.34) and the upper limit of

agreement of 12.53 (95% CI 5 10.04-14.12), representing a confi-

dence limit of the 18.4% BF [12.53 % BF 2 (25.93 %BF)]. The

percentage error is 50.01% (confidence limit of 18.46 divided by

mean % BF value of 36.91 multiplied by 100).

The plot suggests that differences between the two methods exhibit

a regular pattern obvious (proportional error), with overestimation of

% BF by the BAI in subjects with lower % BF and underestimation

in subjects with higher % BF (Figure 1).

DiscussionThe used sample in this study presented values for weight and height

measurements similar to those of the Brazilian population (23).

In this study, the correlation between % BF determined by DXA

and that estimated by BAI was significant (r 5 0.65), although

lower than the observed one in the reference article (5) (r 5 0.79).

Paired t-test showed a significant mean difference between methods,

with % BF measured by BAI showing values lower than determined

by DXA. The result of Lin’s concordance correlation coefficient

was classified as poor (21). The bias of the BAI 3.29 6 4.71 % BF,

represents a low accuracy while the limits of agreement show a low

precision of the method. The percentage error of 50% determinated

for the BAI is the highest value determined by the error-gram of the

Critchley and Critchley (24), who is a enabling one to graphically

determine the limits of agreement between two techniques. This

value of 50% is determined by a limit of agreement of 30% and a

limit of error of the test method of 40%, excessively high values of

acceptable limits of agreement between two methods.

Bland-Altman plots showed a tendency to BAI to overestimated adi-

posity in subjects with lower BF% and underestimated it in obese

subjects in relation to the used criterion measure (Figure 1), as

observed in other populations (15,17). This fact is an aggravating

error underestimation of BAI, it generates large percentage of false

negatives (really obese individuals, but classified as eutrophic by

BAI), failing to correctly identify cases of obesity.

This underestimation when applying BAI may be because of

changes in body fat accumulation with age because it was higher in

this study compared to the original population from which the equa-

tion was derived. To allow comparison’s we have analyzed a subset

of our sample with the same age range of Bergman’s study. How-

ever, the results were similar to the whole sample analyzes. This

can be explained by the average age still be much higher than the

validation sample BAI and much of this group have passed through

menopause. With aging occur changes in body composition, a

decrease in muscle mass and changes in the pattern of body fat

accumulation, with more deposit in the upper body (25). In women,

besides age, menopause is another factor that alters the pattern of

body fat distribution. In postmenopausal American white women

significant mean difference between measurements showed that BAI

underestimated % BF measured by DXA (26). Even without show-

ing differences in total weight, postmenopausal women have a

higher amount of total body fat, trunk fat, less fat in the lower

limbs, and reduced muscle mass in the lower limbs than premeno-

pausal women (27). Thus, as the present study used individuals with

a higher age than that which gave rise to the BAI equation, it is pos-

sible that older women may have reduced hip circumferences as a

result of reduced muscle mass and fat in the lower limbs. Yet, they

had a higher total body fat content, thus changing the relationship

between circumference and body fat. However, we have no data

about menopausal status to determine its influence in our results,

this aspect has to be assumed as a study limitation.

As to the difference between the sexes, women have higher levels of

% BF than men and differences in the distribution pattern of body fat

(25), whereas for height, men have higher mean values than women

(28). From this information, one can assume that a method that aims

to estimate % BF based on anthropometric measurements of hip cir-

cumference and height must propose gender-specific equations. These

statements are confirmed by the results of studies that analyzed the

BAI. Barreira et al. (29) observed differences between BAI estima-

tions and % BF measured by DXA among sexes and races. In this

study, both white and African American women showed lower % BF

estimated by BAI than those determined by DXA, as observed in this

study. For men of the same race, the BAI overestimated % BF. John-

son et al. (15) also observed similar results for European American

men and women. The inclusion of both sexes can distort correlations

between anthropometric measurements versus % BF, as observed by

Vinknes et al. (17). The proposition of a general equation for both

sexes may incur estimate errors of % BF and its wide application

may generate spurious results.

Ethnicity is another factor that greatly influences the shape and

body composition of an individual. African American women have a

higher amount of body fat when compared to Mexican American

and European American women (27). Hispanic individuals of both

sexes with similar ages have been observed to have a lower height

and weight than white and black individuals, with these women dis-

playing a higher waist-hip index than the other two groups (30).

These differences in the anthropometrics profile and body composi-

tion among ethnic groups can change the relationship between

anthropometric measurements and % BF, invalidating the equation

in other populations. The extrapolation of an equation for estimating

% BF based on measurements of body circumference and height for

the Brazilian population should be viewed with caution because it is

composed of a mixture of Amerindians, Europeans, and Africans,

one of the most heterogeneous populations in the world and confer-

ring their peculiar characteristics (31).

Based on these results, BAI was not an appropriate method for

assessing % BF. The method showed poor concordance, low accu-

racy and precision, was statistically lower than that measured with

DXA and underestimated the % BF especially in cases of obesity.

For all the discussed factors, the proposal of an equation for % BF

determination based on anthropometric measurements should be spe-

cific according to sex, age, and ethnicity, enabling their appropriate-

ness and validity in populations with similar characteristics to that

which gave rise to the equation.O

VC 2013 The Obesity Society

References1. Stevens J, McClain JE, Truesdale KP. Selection of measures in epidemiologic

studies of the consequences of obesity. Int J Obes 2008;32 Suppl 3:S60-S66.

2. Li LM, Lei SF, Chen XD, et al. Anthropometric indices as the predictors of trunkobesity in Chinese young adults: receiver operating characteristic analyses. AnnHum Biol 2008;3:342-348.

Obesity Validity of Adiposity Index in Brazilian Women Cerqueira et al.

E698 Obesity | VOLUME 21 | NUMBER 12 | DECEMBER 2013 www.obesityjournal.org

3. Whitlock G, Lewington S, Sherliker P, et al. Body-mass index and cause-specificmortality in 900 000 adults: collaborative analyses of 57 prospective studies. Lancet2009;9669:1083-1096.

4. Flegal KM, Shepherd JA, Looker AC, et al. Comparisons of percentage body fat,body mass index, waist circumference, and waist-stature ratio in adults. Am J ClinNutr 2009;2:500-508.

5. Bergman RN, Stefanovski D, Buchanan TA, et al. A better index of body adiposity.Obesity 2011;5:1083-1089.

6. Geliebter A, Atalaye D, Flancbaum L, Gibson CD. Comparison of body adiposityindex (BAI) and body mass index (BMI) with estimations of % body fat inclinically severe obese women. Obesity 2013;21:493-498.

7. Appelhans BM, Kazlauskaite R, Karavolos K, et al. How well does the bodyadiposity index capture adiposity change in midlife women?: the SWAN fatpatterning study. Am J Hum Biol 2012;24:866-869.

8. L�opez AA, Cespedes ML, Vicente T, et al. Body adiposity index utilization in aSpanish Mediterranean population: comparison with the body mass index. PLoSOne 2012;7:e35281.

9. Schulze MB, Thorand B, Fritsche A, et al. Body adiposity index, body fat contentand incidence of type 2 diabetes. Diabetologia 2012;55:1660-1667.

10. Suchanek P, Kralova Lesna I, Mengerova O, Mrazkova J, Lanska V, Stavek P.Which index best correlates with body fat mass: BAI, BMI, waist or WHR? NeuroEndocrinol Lett 2012;33 Suppl 2:78-82.

11. Freedman DS, Thornton JC, Pi-Sunyer FX, et al. The body adiposity index (hipcircumference 4 height(1.5)) is not a more accurate measure of adiposity than isBMI, waist circumference, or hip circumference. Obesity 2012;20:2438-2444.

12. Freedman DS, Blanck HM, Dietz WH, Dasmahapatra P, Srinivasan SR, BerensonGS. Is the body adiposity index (hip circumference/height1�5) more strongly relatedto skinfold thicknesses and risk factor levels than is BMI? The Bogalusa HeartStudy. Br J Nutr 2013;109:338-345.

13. Sun G, Cahill F, Gulliver W, et al. Concordance of BAI and BMI with DXA in theNewfoundland Population. Obesity 2013;21:499-503.

14. Godoy-Matos AF, Moreira RO, Valerio CM, Mory PB, Moises RS. A new methodfor body fat evaluation, body adiposity index, is useful in women with familialpartial lipodystrophy. Obesity 2012;20:440-443.

15. Johnson W, Chumlea WC, Czerwinski SA, Demerath EW. Concordance of therecently published body adiposity index with measured body fat percent in European-American Adults. Obesity 2012;20:900-903.

16. Silva MI, Vale BS, Lemos CC, Torres MR, Bregman R. Body adiposity indexassess body fat with high accuracy in nondialyzed chronic kidney disease patients.Obesity 2013;21:546-552.

17. Vinknes KJ, Elshorbagy AK, Drevon CA, et al. Evaluation of the body adiposityindex in a Caucasian population: the Hordaland health study. Am J Epidemiol 2013;177:586-592.

18. Esco MR. The accuracy of the Body Adiposity Index for predicting body fatpercentage in collegiate female athletes. J Strength Cond Res 2012 Sep 17 [Epubahead of print].

19. Pena SD, Bastos-Rodrigues L, Pimenta JR, Bydlowski SP. DNA tests probe thegenomic ancestry of Brazilians. Braz J Med Biol Res 2009;42:870-876.

20. Lin LI. A concordance correlation coefficient to evaluate reproducibility. Biometrics1989;1:255-268.

21. McBride GB. A proposal for strength-of-agreement criteria for Lin’s ConcordanceCorrelation Coefficient, 2005. NIWA Client Report: HAM2005-062. http://www.niwa.co.nz.

22. Bland JM, Altman DG. Statistical methods for assessing agreement between twomethods of clinical measurement. Lancet 1986;8476:307-310.

23. Instituto Brasileiro de Geografia e Estat�ıstica (IBGE). Pesquisa de OrcamentosFamiliares 2008-2009: antropometria e estado nutricional de criancas, adolescentese adultos no Brasil. Rio de Janeiro, Brasil: IBGE; 2010.

24. Critchley LA, Critchley JA. A meta-analysis of studies using bias and precisionstatistics to compare cardiac output measurement techniques. J Clin Monit Comput1999;15:85-91.

25. Borrud LG, Flegal KM, Looker AC, Everhart JE, Harris TB, Shepherd JA. Bodycomposition data for individuals 8 years of age and older: U.S. population, 1999-2004. Vital Health Stat 2010;250:1-87.

26. Lemacks JL, Liu PY, Shin H, Ralston PA, Ilich JZ. Validation of body adiposityindex as a measure of obesity in overweight and obese postmenopausal whitewomen and its comparison with body mass index. Menopause 2012;19:1277-1279.

27. Douchi T, Yonehara Y, Kawamura Y, Kuwahata A, Kuwahata T, Iwamoto I.Difference in segmental lean and fat mass components between pre- andpostmenopausal women. Menopause 2007;5:875-878.

28. Sorkin JD, Muller DC, Andres R. Longitudinal change in height of men andwomen: implications for interpretation of the body mass index: the BaltimoreLongitudinal Study of Aging. Am J Epidemiol 1999;9:969-977.

29. Barreira TV, Harrington DM, Staiano AE, Heymsfield SB, Katzmarzyk PT. Bodyadiposity index, body mass index, and body fat in white and black adults. JAMA2011;8:828-830.

30. Carroll JF, Chiapa AL, Rodriquez M, et al. Visceral fat, waist circumference, andBMI: impact of race/ethnicity. Obesity 2008;3:600-607.

31. Pimenta JR, Zuccherato LW, Debes AA, et al. Color and genomic ancestry inBrazilians: a study with forensic microsatellites. Hum Hered 2006;4:190-195.

Original Article ObesityOBESITY BIOLOGY AND INTEGRATED PHYSIOLOGY

www.obesityjournal.org Obesity | VOLUME 21 | NUMBER 12 | DECEMBER 2013 E699