Body Adiposity Measured by Bioelectrical Impedance is an Alternative to Dual-energy X-ray Absorptiometry in Black Africans: The Africans in America Study

Abstract

The prevalence of cardiometabolic disease has risen in Africa and parallels the obesity epidemic. To assess cardiometabolic risk, body composition measurements by dual-energy X-ray absorptiometry (DXA) are ideal. In communities with limited resources, alternative measures may be useful but have not been compared extensively in black Africans. Therefore, we sought to identify alternative methods of body composition assessment, such as body adiposity index (BAI) and bioelectrical impedance analysis (BIA), for use in African-born blacks. This was a cross-sectional study of African-born blacks. BAI and five BIA predictive equations (using variations of height, weight, age, sex and impedance) were compared to DXA to estimate percent fat. Participants were 266 African-born blacks (aged 39±10 mean±SD years, BMI 28±4 kg/m², 68% men) living in metropolitan Washington DC. Equivalence (90% CI (−3, 3)), concordance and Bland-Altman analyses (bias <2%, R² closest to zero) compared BAI or BIA predictive equations to DXA as the criterion method. DXA percent fat was 27.2±5.5% and 40.3±6.9% in men and women, respectively. BAI underestimated percent fat in men (Bias: 1.88±4.71, R²=0.25, p<0.001) and women (Bias: 6.47±4.94, R²=0.08, p=0.01). Of the five BIA predictive equations, the equation reported by Sun et al. (2003) had the best agreement with DXA percent fat for men (Bias: −0.91±3.67, R²=0.02, p=0.05) and women (Bias: −0.92±4.02, R²=0.003, p=0.58). Percent fat from the Sun et al. equation best agreed with DXA percent fat. Therefore, BIA with the Sun et al. predictive equation was the best alternative to DXA for body fat assessment in African-born blacks.

Introduction

Obesity is a growing health issue around the world and its prevalence contributes to increasing development of comorbidities such as heart disease and diabetes. It is imperative that disease risk be recognized early so that interventions can be identified to help prevent progression of obesity related comorbidities [1]. Easily obtained assessments of body fatness would be ideal for use within the healthcare setting especially for African-born blacks as comorbid disease has increased greatly in this population in recent years [2].

Traditionally, risk for obesity-related disorders has been assessed in the clinical setting using body mass index (BMI) which is a measurement of weight (kg) divided by height squared (m). This is a simple measurement, using only weight and height making it easy for clinicians to categorize the patient quickly. Though this measurement works well in large populations to predict disease risk, it is not as accurate at the individual level as fat mass varies with factors such as age, sex, physical fitness and race/ethnicity [3, 4]. Therefore, more sophisticated measures are needed in populations such as Africans, who tend to have more dense lean body mass [5, 6].

Body composition measurement by dual-energy X-ray absorptiometry (DXA) is widely used as a criterion method for quantifying percent fat [7]. However, DXA can be impractical for field studies and clinical practices in communities with limited resources due to its expense, equipment size, radiation exposure and need for highly trained technicians [8-10]. Uncomplicated and less expensive methods may be better options. Other methods such as the body adiposity index (BAI) and bioelectrical impedance analysis (BIA) have been proposed as alternatives to DXA in settings where more precise methods such as DXA may not be practical or available [7, 11, 12]. However, noncomplex methods of body composition assessment tend to be developed in populations of European descent or in very small samples of people of other races or ethnicities limiting their generalizability. Since characteristics such as limb-length and lean body mass [6, 13] can vary widely between individuals of different race, ethnicity, sex or age; predictive equations developed in one population subgroup may not be applicable to other population subgroups.

An easier method to assess body fat in field or clinical investigations could be BAI. This equation was developed by Bergman et al. [7] to provide an estimate of percent fat using only the easily obtained measures of hip circumference and height (fat percent = ((hip circumference (cm))/((height (m))^1.5)−18). Though these measurements require training, they are relatively easy to perform and only require the use of a tape measure and a wall stadiometer. BAI was originally validated among Mexican Americans and American-born blacks [7]; however, it has yet to be assessed for use among African-born blacks.

Another uncomplicated, inexpensive measure that can be obtained quickly and easily is BIA. This method uses the electrical impedance of small, alternating currents that are passed through the body and analyzed by an impedance analyzer [14-16]. Impedance is then used in validated predictive equations to approximate fat-free mass, which is ultimately used to calculate percent fat [15].

In the United States (USA), the National Health and Nutrition Examination Survey (NHANES) III equation [17, 18] is widely used with BIA software. The NHANES III equation was developed in a large heterogeneous sample of participants with mixed races within multiple areas of the USA. Many equations have been validated in small populations of African American blacks, [18, 19] or in a combination of black Americans and blacks of African origin from Africa or the Caribbean, [20, 21].

Though many BIA equations have been developed in blacks, these equations may not be applicable in sub-Saharan populations due to the differences in fat free mass between racial and ethnic populations [5, 6]. Many of the available equations examined herein were developed in populations of African Americans [18-22] whose genetic makeup can be quite different from first generation immigrants from sub-Saharan Africa. Previous reports have shown that African Americans have a mixture of primarily European and west African ancestry with some even having Native American roots [23]. Therefore, due to genetic and cultural heterogeneity, body composition in African Americans may be quite different than sub-Saharan blacks.

Therefore, we set out to find an acceptable method to estimate body fatness in African-born blacks. Given the practical limitations of using DXA to assess percent fat in clinical practice and field studies, the aim of this study was to identify alternative methods of body composition assessment for use in African-born blacks in the metropolitan Washington DC area of the USA.

Materials and Methods

Participants

The Africans in America cohort was a cross-sectional study established to evaluate the cardiometabolic health of Africans living in the USA [24-27] (Clinical Trials.gov Identifier: NCT00001853). Recruitment was achieved by newspaper advertisements (43%), previous participant referrals (30%) and flyers (6%) in the Washington, District of Columbia metropolitan area in the United States of America. The remaining 21% of participants in this cohort heard about the study at community events, church meetings and the National Institutes of Health website. The National Institute of Diabetes and Digestive and Kidney Diseases Institutional Review Board approved the study. Informed written consent was obtained prior to enrollment. For purposes of eligibility, pre-enrollment telephone interviews were conducted by the research team using scripted questions. Potential study participants had to report that they were born in sub-Saharan Africa to two black parents who were also born in sub-Saharan Africa. In addition, they had to be 18-65 years old and state that they were healthy and to their knowledge did not have diabetes.

All participants had two outpatient visits at the NIH Clinical Center, located in Bethesda, Maryland. Participants were enrolled between April 2011 and April 2018. At Visit 1, a medical history, physical, bloodwork and electrocardiogram were performed to determine study eligibility. A complete medical history was performed including a review of personal and family illnesses, education and work history, medications, exercise, nutrition, alcohol and smoking. Routine blood tests were performed to confirm the absence of anemia, kidney, liver and thyroid disease. For Visit 2, participants fasted for 12 hours, with plain tap or bottled water encouraged, and came to the Clinical Center at 07:00 hours as described previously [28]. Female patients were required to complete a urine pregnancy test prior to DXA measurement to ensure that participants were not pregnant. All participants underwent anthropometric, BIA and DXA measurements at this visit.

Three hundred African-born blacks were enrolled. For either medical reasons or scheduling challenges 18 enrollees did not proceed to Visit 2. Of the 282 who participated in Visit 2, 16 did not have DXA scans performed or there were technical issues with the scans or BIA. Therefore, this analysis is based on 266 participants.

Measurements

Anthropometry.

Fasting height, weight, and hip circumference were measured using standard NHANES procedures at the second outpatient study visit [29]. Height (centimeters) and weight (kilograms) were measured with participants in minimal clothing without shoes to the nearest tenth of a unit using a wall stadiometer (Seca 242, Hanover, MD, USA) and a calibrated digital scale (Scale-Tronix 5702, Carol Stream, IL, USA), respectively. Hip circumference was measured over the participants underwear in centimeters at the maximal protuberance of the buttocks using a non-stretch tension-sensitive measuring tape (Gulick 2, Gays Mills, WI, USA) by trained nutrition staff. Measurements were done primarily by two registered dietitians who underwent biannual calibration to ensure inter-rater reliability. All measurements were taken in triplicate and the mean of the three measurements was reported.

Body Composition.

Whole body DXA scans were done with a fan beam Hologic Discovery DXA (Bedford, MA, USA) and software versions 13.2 to 13.4.2. Reliability after software version changes was assessed using spine phantoms. Scans were performed per manufacturer’s instructions. Prior to scanning, all participants were asked to remove all jewelry and other items that could interfere with the scan. Lean body mass, fat mass and percentage fat were corrected per the methods of Schoeller et al [30].

Percent fat from BAI was calculated from hip circumference and height (fat percent = ((hip circumference (cm))/((height (m))^1.5)−18) [7].

To assess percent fat using BIA, resistance and reactance (ohms) were measured using a tetrapolar, single frequency (50 kHz) Quantum II analyzer (RJL,Clinton Township, MI, USA). Measurements were taken in the morning following a 12-hour fast and voiding. Participants were instructed to lie supine for 5 minutes on a flat surface without moving, with arms 30 degrees from the body and legs not touching. Electrodes were placed on the right hand/wrist and foot/ankle, per manufacturer instructions. Anthropometric and BIA impedance measures were then applied to the predictive equations outlined in Table 1 to determine fat free mass. The equation recommended for use with the BIA software application as well as equations previously found in a literature search were selected based on the following criteria: conducted in a black population, used a tetrapolar, single frequency (50 kHz) BIA machine and used DXA or total body water as the criterion method. When the criterion method for the predictive equation was total body water, fat free mass was calculated as total body water divided by 0.73 [30, 31]. The fat free mass was then used to calculate percent fat ((body weight – fat free mass)/body weight*100).

Table 1.

Predictive equations used to estimate fat free mass using bioelectrical impedance measurements in healthy, adult African born blacks in the United States.

Reference	Predictive Equation Used to Calculate Fat Free Mass	Criterion Method	Participants (N)
NHANES III [22]	M: −10.678 + 0.262*wt + 0.652*ht2/ Res + 0.015*Res	4-C	Black American Male	114
			White American Male	412
	F: −9.529 + 0.168*wt + 0.696*ht2/Res + 0.016*Res		Black American Female	156
			White American Female	622
Sun [18]	M: 1.2 + 0.45*(ht2/Res) + 0.18(wt)/0.73	D2O	Black American Male	114
			White American Male	412
	F: 3.75 + 0.45*(ht2/Res) + 0.11*wt)/0.73		Black American Female	156
			White American Female	622
METS [20]	12.6 + 0.22*wt + 0.46*(ht2/Res) − 5.7*Sexa	D2O	Black Ghanaian Male	20
			Black South African Male	22
			Black Jamaican Male	23
			Black Seychellois Male	27
			Black American Male	26
			Black Ghanaian Female	24
			Black South African Female	28
			Black Jamaican Female	24
			Black Seychellois Female	27
			Black American Female	23
Zillikens [19]	−9.212 + 0.576*(ht2/Res) + 0.128*Xc + 0.107*wt)/0.73	D2O	Black American Male	43
			Black American Female	45
Luke [21]	1.3 + 0.097*wt + 0.518*(ht2/Res))/0.73	D2O	Black Nigerian Male	137
			Black Jamaican Male	94
			Black American Male	189
			Black Nigerian Female	161
			Black Jamaican Female	146
			Black American Female	327

Abbreviations: M, Males; F, Females; ht, height (centimeters); NHANES, National Health and Nutrition Examination Survey; wt, weight (kilograms); Res, Resistance (ohms); 4-C, four component model; D₂O, deuterated water method; METS, Modeling the Epidemiologic Transition Study; Xc, reactance.

^aSex: 0=Men; 1=Women.

Statistical Analysis

Study data were collected and managed using Research Electronic Data Capture (REDCap), an electronic data capture tool hosted by the National Institute of Diabetes and Digestive and Kidney Diseases [32]. A post-hoc sample size calculation was performed. As the age range of our cohort was primarily between 30 and 50, we performed calculations using the SD for black, non-Hispanic adults obtained from Chumlea et al. [22] for non-Hispanic black adults, where the SD of body fat was 6.7%. We used an acceptable mean difference between the two methods of 3%. Assuming an alpha of 0.01 and a Beta of 0.2, a sample size of 58 subjects was calculated, which is below our studies sample size of 266 participants.

Data are presented as mean±SD unless otherwise stated. All analyses were stratified by sex. Descriptive analyses of participant characteristics and concordance analyses (using Kendall’s tau-b correlation coefficients) were conducted with Statistical Package for the Social Sciences version 25 (SPSS) (IBM Corporation, Chicago, IL, USA). Difference in descriptive data between men and women were analyzed using independent t-tests for years in the USA, which was normally distributed and the Mann Whitney test for all other variables. To interpret effect size of correlation coefficients, Cohen’s effect size conversions of 0.10 as weak or small, 0.30 as moderate and 0.50 or larger as strong associations were utilized [33]. Equivalence tests [34] and Bland-Altman analyses [35] were conducted using GraphPad version 7.2 (GraphPad Software, La Jolla, CA). DXA was used as the criterion method for all analyses. The zone of indifference for the equivalence testing was (±3%) and results are presented as the 90% confidence intervals. Criteria for acceptance based on the Bland Altman analyses was agreement within 3% (based on reports that accuracy of BIA measurements is 3-4% [36]) of the criterion method with an R² closest to zero were determined a priori as an acceptable limit for agreement. P-values <0.05 were considered significant.

Results

As a group, participants had a BMI of 27.7 ± 4.4 kg/m² (range: 19.7-42.4 kg/m²) (Table 2). Men and women were of similar age and had lived in the USA for a similar period of time. However, as expected, they differed in weight, height, BMI and hip circumference. Mean percent fat of men and women, as assessed using DXA, BAI, and the five previously validated BIA equations, are shown in Table 3. Using DXA, the mean percent fat of men was 27.2 ± 5.5% and women was 40.3 ± 6.9%.

Table 2.

Characteristics of 266 adult, healthy African-born Blacks living in the United States

	Men (n = 181) (mean±SD)	Women (n = 85) (mean±SD)	P-Value *
Age (years)	39.8 ± 9.9	39.6 ± 10.3	0.951
Weight (kg)	84.0 ± 13.5	77.6 ± 14.7	0.001
Height (cm)	176.0 ± 7.0	163.5 ± 5.9	0.000
Body Mass Index (kg/m2)	27.1 ± 3.9	29.0 ± 5.1	0.005
Hip Circumference (cm)	100.9 ± 7.5	108.1 ± 11.0	0.000
Years in United States	11.2 ± 9.3	14.6 ± 9.4	0.730

^*All variables were assessed using the Mann Whitney test except for the variable Years in the United States, which was normally distributed and therefore the independent t-test was used. P<0.05 considered significant

Table 3.

Percent body fat estimated by dual-energy x-ray absorptiometry, body adiposity index and five bioelectrical impedance analysis predictive equations with equivalence and concordance between measures in adult, healthy African born blacks (N=266).

	Percent Body Fat (mean ± SD)		Equivalence# with DXA (90% Confidence Interval)		Concordance* with DXA (Tau coefficient)
	Men (n=181)	Women (n=85)	Men (n=181)	Women (n=85)	Men (n=181)	Women (n=85)
DXAa	27.2 ± 5.5	40.3 ± 6.9
BAIb	25.3 ± 3.5	33.8 ± 5.6	(−2.5, −1.3)	(−7.4, −5.6)	0.362	0.515
BIASunc	28.1 ± 5.0	41.2 ± 6.7	(0.5, 1.4)	(0.2, 1.6)	0.506	0.633
BIANHANES III	29.5 ± 5.1	41.4 ± 6.6	(1.9, 2.8)	(0.4, 2.0)	0.517	0.603
BIALuke	36.6 ± 5.7	46.1 ± 6.8	(9.0, 9.9)	(5.1, 6.6)	0.502	0.645
BIAZillikens	28.2 ± 6.2	38.9 ± 8.4	(0.6, 1.5)	(−2.2, −0.6)	0.592	0.696
BIAMETS	28.9 ± 5.1	41.3 ± 5.6	(1.2, 2.2)	(0.3, 1.7)	0.537	0.617

^aDXA=dual-energy x-ray absorptiometry

^bBAI=body adiposity index

^cBIA=bioelectrical impedance analysis

^#The zone of scientific or clinical significance was considered (±3)

^*Kendall’s Tau Correlation Coefficients, all p < 0.0001

Equivalence testing revealed that four BIA equations had 90% confidence intervals within the zone of scientific or clinical indifference (±3). These equations were BIA_Sun, BIA_{NHANES III}, BIA_Zillikens and BIA_METS. BAI was only found to be equivalent in men. Overall, there was medium to strong concordance (using Cohen’s coefficients [33]) between percent fat calculated by DXA and percent fat estimated from each equation (all r > 0.4; p < 0.0001) (Table 3).

Bland-Altman plots were created comparing each predictive equation to DXA to assess agreement between the methods and identify optimal equations for African-born black men and women (Figures 1 and 2).

Figure 1:

Bland-Altman Plots Comparing Percent Body Fat Predicted by Dual-Energy X-Ray Absorptiometry (DXA) and Body Adiposity Index (BAI) and Bioelectrical Impedance (BIA) Predictive Equations Among African Born Black Men in the United States. A – Difference in percent fat between DXA and BAI. B – Difference in percent fat between DXA and BIA using the Sun et al. [18] prediction equation. C – Difference in percent fat between DXA and BIA using the National Health and Nutrition Examination Survey (NHANES) III [22] prediction equation. D – Difference in percent fat between DXA and BIA using the Luke et al. [21] prediction equation. E – Difference in percent fat between DXA and BIA using the Zillikens et al. [19] prediction equation. F – Difference in percent fat between DXA and BIA using the Modeling the Epidemiologic Transition Study (METS) [20] prediction equation.

Figure 2:

Bland-Altman Plots Comparing Percent Body Fat Predicted by Dual-Energy X-Ray Absorptiometry (DXA) and Bioelectrical Impedance (BIA) Predictive Equations Among African Born Black Women in the United States. A – Difference in percent fat between DXA and BAI. B – Difference between DXA percent body fat and BIA using the Sun et al. [18] prediction equation. C – Difference between DXA percent body fat and BIA using the National Health and Nutrition Examination Survey (NHANES) III [22] prediction equation. D – Difference between DXA percent body fat and BIA using the Luke et al. [21] prediction equation. E – Difference between DXA percent body fat and BIA using the Zillikens et al. [19] prediction equation. F – Difference between DXA percent body fat and BIA using the Modeling the Epidemiologic Transition Study (METS) [20] prediction equation.

Figure 1A illustrates that while BAI had an acceptable bias in men (1.88 ± 4.71%), it had a large and significant R². The BIA_NHANESIII, BIA_Luke and BIA_METS equations had a non-significant R² closest to zero (R²= 0.01), (Figures 1C, 1D and 1F). However, the bias of the Luke et al. equation was greater than three percent and was outside of the acceptable limits, as had been determined a priori. BIA_{NHANES III}, BIA_Zillikens, and BIA_METS had biases within 3% with low R² values; however, the BIA_Sun equation was the only equation to have a bias under one percent while also having a non-significant R² that was close to zero (bias: −0.91 ± 3.67, R²=0.02, p=0.05), suggesting the best agreement in predicting percent fat in African-born blacks (Figure 1B).

Among women (Figure 2), all BIA equations had an acceptable level of bias, except for the BIA_Luke equation (Figure 2D). The BAI equation (Figure 2A) did not have an acceptable level of bias and the R² was large; thus, it was not a good predictor of body fat. Of the five remaining BIA equations (Figures 2B, 2C, 2E and 2F), all had an acceptable level of bias; however, since BIA_Sun had the lowest, non-significant R² (R²=0.003, p=0.58), and a bias closest to zero (bias = −0.92 ± 4.02) this equation was therefore determined to be the most suitable option in women.

Discussion

Given the known racial differences in body fat distribution, it is imperative to identify accurate measures of body composition to be used in African-born blacks. However, relatively few simple measures of estimating percent fat have been developed in populations of African-born blacks. Compared to DXA, BIA predictive equations had better agreement than BAI for the prediction of percent fat in African-born blacks. Overall, of the five BIA predictive equations examined, the BIA_Sun equation was the best predictor of percent fat and was found to be equivalent to DXA in both men and women in our population of African-born blacks who were currently living in the Washington, District of Columbia, United States of America metropolitan area. This equation should thus be considered when DXA measurements are unavailable and uncomplicated estimates of percent fat are needed in people of sub-Saharan African descent.

BAI may not have been the best predictor of percent fat in our population of African-born blacks due to racial differences in where body fat is stored, as the equation was developed in Mexican Americans [7]. It was found to be applicable in African Americans; however, African Americans could differ from our population of African-born blacks as they may have a different genetic background and have different environmental influences (i.e., consumption of a Western diet) than the Africans who have lived for generations in Africa [6, 37] Additionally, since BAI relies heavily on hip circumference, body fat may be underestimated in races which store fat preferentially in their abdominal region vs their hips [38]. Our data also agreed with previous reports of BAI by showing that the magnitude of bias varied with the level of percent fat [39, 40]. This indicates that BAI is an unreliable tool for studies requiring greater precision and accuracy especially in groups with extremes in body fatness.

It was surprising that BIA_Luke and BIA_METS did not perform better than equations developed in Americans since those equations were developed in populations of African origin with large sample sizes [20, 21]. Furthermore, the BIA_Luke equation was not a sex-specific equation which may be why it did not perform as well in our population which was stratified by sex. Also, our analyses were all stratified by sex but only three of the BIA predictive equations tested included sex as a variable. Thus, sex specific differences could have affected the biases and 95% COI of our Bland-Altman analyses. BIA_METS also used five populations of African origin, 60% of which were sub-Saharan African, as in our participants. Therefore, differences in the populations may reflect ethnic and cultural differences, which ultimately impacts accuracy of predictive equations. Though BIA_Sun was not specifically developed in African-born blacks, it was developed on the largest number of subjects, which likely improved its predictive value.

A major strength of the present study was our large sample of African-born blacks and analysis of a historically overlooked population subgroup. Previously, only two studies have examined BIA in African-born blacks [20, 21]. Our data add to this limited dataset and further examines African-born blacks living in the Washington, District of Columbia, United States of America metropolitan area. Our participants’ countries of origin were sub-Saharan Africa; therefore, our results may be applicable to blacks living in sub-Saharan Africa. Another strength of our study is that we combined regression with Bland-Altman analyses to facilitate the identification of inconsistencies in error based on body composition. This allowed us to not only examine the amount of difference between our measurements via the bias, but to also examine how the bias changed as the percent fat increased. To account for the subjectivity involved in the analysis of Bland-Altman, we determined the acceptable limits of bias a priori.

Another advantage of this study is that we were able to use DXA as our criterion method and used the same Discovery DXA machine for all participants. Though DXA is considered the most accurate measure of body composition in clinical practice [41], it is not without limitations. This study was conducted between 2011 and 2017, during which there was a change in the DXA software and the reliability between software versions was only assessed for bone mineral density and not body composition. Therefore, changes in software could have impacted our results and thus, this should be looked at more closely with future studies.

Despite the aforementioned advantages of BIA, researchers should be aware that this method is not without limitations, regardless of the equation utilized. Factors such as hydration status and water distribution are known to greatly effect impedance measurements [15]. Thus, hydration status and participant positioning during testing are very important in minimizing the limitations of this method [42]. We were unable to objectively measure hydration status in this population; thus, future studies should obtain objective measures of hydration status. There are also many types of BIA analyzers on the market for use in research and clinical settings, such as multi-frequency analyzers, hand-to-hand and foot-to-foot models. Using the equations discussed within these analyses with analyzers that are not single frequency and quadra-polar would likely yield different results.

Results of our analysis may also be influenced by the demographics of the participants in our population subgroup. Participants’ ages ranged from 20 to 64 years old. As body composition is known to change across the lifespan [43], our results are not generalizable to children or elderly African-born blacks. Lastly, though the measurements were completed by only a few dietitians who participated in bi-annual calibrations, one cannot completely rule out the amount of error introduced by more than one person taking the measurements. Therefore, future studies should determine inter-rater reliability for hip circumference measurements in a subset of participants.

Conclusions

In conclusion, though BAI and all BIA predictive equations were well correlated with Hologic Discovery DXA percent fat in African-born blacks, not all methods performed well when looking at equivalence or agreement with DXA percent fat. Therefore, we suggest using the BIA method with the BIA_Sun predictive equation when assessing percent fat among blacks from sub-Saharan Africa in situations where DXA is not accessible.

Highlights.

• This was a cross-sectional study in African-born black men and women.

• BIA was better than the BAI at estimating body fat

• The BIA predictive equation that performed best was the Sun et al, 2003

• The Sun equation should be used with BIA to assess body fat in African-born blacks.