Disclosure: None reported. The authors declare no competing interests.
Funding agencies: This study was supported by the Alexandra Health Enabling Grant, FY 2011. The sponsors had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, or approval of the manuscript.
Author contributions: BCC Lam had full access to all of the published data in the study and takes responsibility for the integrity of the data and the accuracy of data analysis. Study concept and design: BCC Lam, CY Ho, JIE Bosco, MTK Wong. Acquisition of data: BCC Lam, CY Ho, JIE Bosco. Analysis and interpretation of data: BCC Lam, SC Lim, E Shum, GCH Koh, C Chen. Drafting of the manuscript: BCC Lam. Critical revision of the manuscript for important intellectual content: SC Lim, E Shum, GCH Koh, MTK Wong, C Chen. Study supervision: BCC Lam, MTK Wong. All authors were involved in writing the article and had final approval of the submitted and published versions.
A recently developed parameter, the body adiposity index (BAI)—a composite index based on hip circumference and height—estimates the percentage (%) body adiposity indirectly. The BAI was compared with dual energy X-ray absorptiometer (DEXA)-derived % adiposity to validate the BAI in the local Chinese population.
Design and Methods
105 Chinese were recruited and % adiposity estimated by BAI was compared with that derived from DEXA using the Bland Altman plot. A correlation study comparing the BAI with body mass index (BMI) was also done.
BAI underestimated DEXA-derived % adiposity by a mean of 5.77% with 95% limits of agreement of ±8.4%. When stratified by gender, BMI correlated with DEXA-derived % adiposity better than BAI (r = 0.81 vs. 0.74 for males, P = 0.088, and r = 0.87 vs. 0.82 for females, P = 0.087). Hip circumference and waist circumference also correlated better with the BMI than BAI (r = 0.94 vs. 0.71 for hip circumference, P < 0.001, and r = 0.93 vs. 0.50 for waist circumference, P < 0.001, respectively).
The BAI underestimates DEXA-derived % adiposity in a Chinese population in Singapore and is unlikely to be a better overall index of adiposity than the established BMI.
Obesity is a major and growing problem throughout the world . It is characterized by a relative excess of adipose tissue mass, and is currently defined as a body mass index (BMI) ≥30 kg m−2. However, the BMI is not a very accurate measure of adiposity in individual patients . It is unable to differentiate between lean mass and fat mass, and hence it is limited by differences in body fatness for a given BMI across age, gender, and ethnicity . This can lead to individuals being misdiagnosed as having inappropriate body fat due to variation in muscle mass, while others with significant adiposity are being overlooked.
The more accurate methods of quantifying body fat involve underwater weighing (in a four-compartment criterion model) and dual-energy X-ray absorption (DEXA) . However, these methods are technologically complex and are too costly and time consuming to be applied routinely in clinical settings, where follow-up measurements are often required. Surrogate methods such as impedance analysis and skin fold thickness can also be used, but can be very inaccurate [5, 6]. Recently, a better index of body adiposity has been proposed , and currently being evaluated. According to Bergman et al. , this new parameter, the body adiposity index (BAI), is “a direct estimate of % body fat” and is calculated from hip circumference and height only. This new parameter is developed from a population study of Mexican Americans, and is validated in a separate study of African Americans. For the BAI to be used as widely as the BMI, it is necessary to validate it in the other major ethnic groups, and examine its usefulness as a predictor of health outcomes.
The primary aim of this study is to compare the BAI with DEXA-derived % adiposity, so as to determine the validity of the BAI in Chinese subjects, one of the major ethnic groups in Singapore. The secondary aim of this study is to study the relationships between the BAI and the BMI and the other anthropometric measurements like waist circumference, hip circumference, weight, and height with reference to DEXA-derived % adiposity and compare them with other ethnicities that have already been reported.
The subjects recruited were a mixture of healthy volunteers and patients who attended the outpatient clinic of the Department of Family and Community Medicine in Khoo Teck Puat Hospital (KTPH), Singapore. This clinic provides health screening, weight management, and chronic disease management services in a hospital setting. Interested patients were identified by their attending doctor at this clinic, and subsequently referred for eligibility assessment. Healthy volunteers were mainly from staff of KTPH (outside of the department) who had previously indicated an interest in participating in research during the hospital's annual health screening for staff. All interested subjects were assessed for their eligibility (age >21-years old, and both parents of Chinese ethnicity), and subjects were excluded if they were pregnant, breastfeeding, or has a cancer diagnosis within the past 5 years. All subjects provided written and informed consent, and ethics approval was obtained from the National Healthcare Group (NHG) Domain Specific Review Board (DSRB). (NHG DSRB Ref: 2011/01813)
The subjects, wearing minimal clothing, were weighed to the nearest 0.1 kg using a strain gauge scale, which was also equipped with a stadiometer (TBF-215, Tanita, Tokyo, Japan). Their height was measured with the subject standing barefoot and to the nearest 1 cm. BMI was calculated in the standard way: weight in kg divided by square of height in meters. Waist and hip circumferences were measured to the nearest 0.1 cm using a flexible metric measuring tape while the subject was in a standing position. Waist circumference was measured as the horizontal distance around the abdomen at the level of the superior border of the iliac crests, in accordance with the National Institutes of Health (NIH) protocol . The mean of three readings was recorded. Hip circumference was measured over non restrictive underwear or light-weight shorts, at the level of the maximum extension of the buttocks posteriorly in a horizontal plane. Similarly, the mean of three readings was recorded. BAI was calculated as proposed by the original authors: hip circumference in cm divided by (height in m)1.5 – 18 or BAI = (hip/height1.5) – 18.
All measurements as described above were taken by a single observer for all the subjects, to negate any inter observer variability.
Whole body composition measurements were performed using the Hologic QDR Discovery Wi dual energy X-ray absorptiometer (DEXA) in the auto whole body scan mode, using software version 13.3 (Hologic, Bedford, MA). DEXA has been shown to be precise , requiring only one measurement. The scan was conducted by a trained technician who was assessed to be competent. The algorithm for whole body fat calculation was standardized for the same software version, and calculations were done automatically by the system software after each scan was completed. Quality assurance via calibration with a lumbar spine phantom was performed daily during the study period.
All measurements were taken on the same day, at the same visit. The subjects were not allowed to eat or drink between the measurements to minimize variations in body compositions during measurements.
The validation of the BAI was done by comparing the BAI with DEXA-derived % adiposity, as what the original authors had done in their article, by using the Bland Altman plot . Pearson's correlation analysis (r) was done to investigate the relationships between the BAI with the BMI and other anthropometric measurements like waist circumference, hip circumference, weight and height with reference to DEXA-derived % adiposity, with significance based on a two-tailed test. Test of significance between dependent correlation coefficients was performed using T test and respective P value was reported [11, 12]. In addition, to further examine the differences between the BAI and the BMI when compared with DEXA-derived % adiposity, scatter grams were plotted. Simple linear regression was used to obtain the unadjusted coefficient of determination (R2) and multiple linear regression adjusting for gender and age was used when obtaining the model's R2. Two independent samples t test was used in the comparison of anthropometric measurements between males and females. The determination of the sample size for this study was based on the recommendation by Bland, one of the creators of the Bland Altman plot, who stated that a sample size of at least 100 is recommended for a method comparison study (single measurement for both methods) . All statistical analyses were performed with the use of R software, version 2.14.1 (R Development Core Team).
A total of 105 subjects were recruited, with almost equal distribution of genders, 53 males, and 52 females. 77 (73.3%) of the subjects were patients from the clinic, while the rest were healthy volunteers. Characteristics of the subjects are shown in Table 1. The subjects' ages ranged from 21 to 65, with a wide range in weight (35.8 to 167.1 kg) and BMI (17.0 to 55.2 kg m−2). Comparing the means in males and females, males tend to have higher height (P < 0.001), higher weight (P < 0.001), higher body mass index (P = 0.001), lower DEXA% (P < 0.001), larger waist (P < 0.001), larger hip (P = 0.008), and lower body adiposity index (P = 0.001). Based on the histograms and quantile–quantile plot, there was no major skewness of the variables listed in Table 1.
Table 1. Characteristics of study subjects
Mean (standard deviation)
Total (n = 105)
Male (n = 53)
Female (n = 52)
Body mass index (kg m−2)
Dual energy X-ray absorptiometry (DEXA) adiposity (%)
Body adiposity index (BAI) (%)
Validation of BAI
In this population of Chinese subjects, the BAI tended to underestimate DEXA-derived % adiposity by a mean of 5.77% (95% confidence interval: 4.94-6.60%), and the 95% limits of agreement were ±8.4% which means that the difference between the BAI and DEXA-derived % adiposity could range from +2.61% to −14.15% (Figure 1). The difference between BAI and DEXA derived % adiposity (i.e., BAI minus DEXA) was positive for lower mean BAI and DEXA values but negative for higher mean BAI and DEXA values, suggesting that the underestimation of DEXA derived % adiposity by BAI was greater at higher body % adiposities.
BAI, BMI, waist circumference, hip circumference, height, weight and % adiposity
Tables 2 and 3 shows the correlations between BAI, BMI, waist circumference, hip circumference, height, weight, and % adiposity. Without adjusting for gender, the BAI had a higher correlation with DEXA-derived % adiposity than the BMI (r = 0.81 for BAI vs. r = 0.55 for BMI, P < 0.001). However, upon stratification by gender, this relationship is lost and would even appear that BMI correlated with DEXA-derived % adiposity better than the BAI, although this was borderline significant in our sample size of 53 males and 52 females (r = 0.81 vs. 0.74 for males, P = 0.088, and r = 0.87 vs. 0.82 for females, P = 0.087). Among all the primary anthropometric measurements (i.e., height, weight, waist circumference, and hip circumference), hip circumference was most strongly correlated with DEXA-derived % adiposity when males and females were considered together (r = 0.59) (Table 2).Upon stratification by gender, hip circumference continued to be the most strongly correlated with DEXA-derived % adiposity in females (r = 0.86), while waist circumference was the most strongly correlated with DEXA-derived % adiposity in males (r = 0.85). We also found that both hip circumference and waist circumference were more correlated to the BMI rather than the BAI (r = 0.94 vs. 0.71 for hip circumference, P < 0.001 and r = 0.93 vs. 0.50 for waist circumference, P < 0.001 respectively) (Table 3). The higher correlation between hip or waist circumference and BMI than BAI persisted even after stratification by gender. Finally, we found that there was a positive correlation between hip circumference and height (r = 0.44, P < 0.01), the two parameters used in the BAI.
Table 2. Pearson correlation coefficients between anthropometric characteristics and DEXA-derived % adiposity
Dual energy X-ray absorptiometry (DEXA) adiposity (%)
The effect of gender on BAI vs. BMI in assessing adiposity
Figure 2 shows the scatter plot diagrams of the BAI and BMI when compared with DEXA-derived % adiposity, stratified by gender. When DEXA-derived % adiposity was correlated with BAI, the points for both genders grouped along a common line with the males more proximal to the origin and females distal to the origin (consistent with a lower % adiposity in males compared to females) (Figure 2a), explaining the higher correlation coefficient between DEXA-derived % adiposity and BAI when both genders are considered together (Table 2). In contrast, when DEXA-derived % adiposity was correlated with BMI, the points for each gender grouped into two distinct clusters with the female cluster demonstrating higher body adiposity than the male cluster (Figure 2b), explaining why the poorer correlation coefficient between DEXA-derived % adiposity and BMI when both genders were considered together (Table 2). But, when stratified by gender, the correlation between DEXA-derived % adiposity and BMI improved considerably, and would even appear that BMI now correlated with DEXA-derived % adiposity better than the BAI, although this was borderline significant in our sample size of 53 males and 52 females. These observations are supported by R2 from multiple linear regression for DEXA-derived % adiposity with BMI and BAI (Table 4). In the univariate model, 26.3% of variation in DEXA-derived % adiposity can be explained by BMI. Upon further adjustment of gender, both BMI and gender explained 76.0% (substantial increase from 26.3%) of the variation in DEXA-derived % adiposity. This is in contrast to the BAI, where further adjustment of gender did not substantially increase the explained amount of variation in DEXA-derived % adiposity (69.8%, increased from 65.0%). Finally, the inclusion of age when BMI or BAI and gender are in the model did not explain additional variation in DEXA-derived % adiposity (Table 4).
Table 4. R2 values of unadjusted and adjusted linear regression models of dual energy X-ray absorptiometry (DEXA)-derived adiposity
R2 of model
Adjusted for gender
Adjusted for gender and age
Body mass index (BMI) (kg m−2)
Body adiposity index (BAI) (%)
We believe that this is the first study to specifically address the BAI for the Chinese population in Singapore and we found that the BAI underestimates DEXA-derived % adiposity by a mean of 5.77%. Although Bergman et al. , original creators of the BAI, managed to validate it in a separate study of African American, the BAI has not been validated in other ethnic groups so far. Freedman et al.  found that the BAI overestimated % adiposity for men and underestimated % adiposity for the women in a study of 1,151 adults with various ethnicities (White 37%, Black 27%, Hispanic 25%, Asians 8%, and Others 3%). Similar findings have been reported in Johnson et al. who studied 623 European Americans , and Schulze et al. who studied 360 White Europeans (but using Magnetic Resonance Tomography instead of DEXA) .
The 95% limits of agreement between the BAI and DEXA-derived % adiposity in our study was ±8.4% which is similar to those reported by Freedman et al. (±12.3%) , Schultz et al. (≈±10%, from their Bland Altman plot)  and Bergman et al (≈±10%, from their Bland Altman plot) . These 95% limits of agreements may be considered clinically unacceptable, and hence, there is poor agreement between the two methods of measuring % adiposity. In addition, the level of agreement can vary with gender and level of body fatness, as shown in Freedman et al.  and Bergman et al. . From our own Bland Altman plot (Figure 1), we also observe that the underestimation of DEXA derived % adiposity by BAI was greater at higher body % adiposities. Hence, as a tool for the measurement of % adiposity, the BAI does not perform well enough to be able to replace the DEXA scan.
As with Freedman et al.  and Schulze et al. , we also observe that when stratified by gender, BAI no longer correlated with % adiposity better than the BMI, and the initially poor performance of the BMI in predicting DEXA-derived % adiposity compared to the BAI was due to gender, as illustrated by our scatter plot diagrams of the BAI and BMI when compared with DEXA-derived % adiposity (Figure 2) and R2 from multiple linear regression for DEXA-derived % adiposity with BMI and BAI (Table 4). Looking back at the original derivation of the BAI by Bergman et al. , hip circumference was chosen as one of the principal anthropometric measure on which the BAI was based upon because it was the most positively correlated with DEXA-derived % adiposity in the study population where males and females were considered together. However, from our own data, we found that while hip circumference continued to be the most positively correlated of all primary anthropometric measurements with DEXA-derived % adiposity in females, waist circumference became the primary anthropometric measure most positively correlated with DEXA-derived % adiposity in males. This is similar with findings in Freedman et al.  and Schulze et al. , and could account for the observation that BAI correlated with DEXA-derived % adiposity better only in analyses that combined males and females together.
It was also noted in our study population that there was a positive correlation between height and hip circumference, suggesting that they are not independent variables for our population. This is in contrast to Bergman et al.  whose original ‘derivation’ population showed that there were no correlation between height and hip circumference which led to them being considered as ideal parameters to base their design of the BAI. This could be another reason why the BAI did not perform as well in our study population.
Lastly, it is well known that waist circumference is an effective method to assess central obesity with excellent correlation with abdominal imaging and high association with cardiovascular risk and mortality . Our observation that waist circumference was better correlated with BMI than BAI suggests that the BMI is a better overall index of adiposity compared to the BAI which in itself is unable to distinguish the distribution of adiposity. The better correlation between waist circumference and BMI than BAI persisted even after stratification by gender. Hence, based on this and the preceding discussion on the performance of the BAI when compared to the BMI, the BAI is unlikely to be superior to the established BMI as an overall index of adiposity.
Our study has important limitations. First, our study population may not be entirely representative of the general adult Chinese population in Singapore. Second, while the sample size is believed to be adequate for an assessment of the level of agreement between the BAI and DEXA-derived % adiposity as a whole, the small number limits any conclusions on the level of agreement between the BAI and DEXA-derived % adiposity when stratified by gender and level of fatness. Lastly, DEXA estimates of % adiposity have been known to vary between manufacturers and across models . In our case, although the manufacturer of our DEXA (Hologic) was the same as the one used in Bergman et al.'s study, the model with its attending scan mode and software was different.
In summary, although the BAI was originally designed in view of the limitations of the BMI, namely being unable to differentiate between lean and fat masses, the results of this study shows that the BAI has its own set of limitations as well. Therefore we conclude that the BAI underestimates DEXA-derived % adiposity, with a wide 95% limits of agreement, in a Chinese population in Singapore and is unlikely to be a better overall index of adiposity than the established BMI.
We acknowledge the support of the Clincal Research Unit, Khoo Teck Puat Hospital, Singapore, in particular for the preliminary statistical analysis, which was provided by Ng Tze Pin, Department of Psychological Medicine, National University of Singapore. We thank Senior Staff Nurse Ng Poh Hiang, and Staff Nurse B.L Shorna Latha, Health For Life Clinic, Khoo Teck Puat Hospital, Singapore, for their invaluable contributions to the conducting of this study.