Concordance of BAI and BMI with DXA in the Newfoundland Population

Authors


  • Disclosure: The authors declared no conflict of interest.

  • Funding agencies: Supported by the Canadian Institutes for Health Research (CIHR) and the Canada Foundation for Innovation (CFI).

Abstract

Background:

Body adiposity index (BAI), indirect method proposed to predict adiposity, was developed using Mexican Americans and very little data are available regarding its validation in Caucasian populations to date.

Objective:

The study objectives were to validate the BAI with dual-energy X-ray absorptiometry (DXA) body fat percentage (%BF), taking into consideration the gender and adiposity status.

Design and Methods:

A total of 2,601 subjects (Male 662, Female 1939) from our Complex Diseases in the Newfoundland population: Environment and Genetics (CODING) study participated in this investigation. Pearson correlations, with the entire cohort along with men and women separately, were used to compare the correlation of both BAI and BMI with %BF. Additionally, the concordance between BAI and BMI with %BF were also performed among normal-weight (NW), overweight (OW), and obese (OB) groups. Adiposity status was determined by the Bray Criteria according to DXA %BF.

Results:

BAI performs better than BMI in our Caucasian population by: (1) reflecting the gender difference in total %BF between women and men, (2) correlating better with DXA %BF than BMI when women and men are combined, and (3) performing better in NW and OW subjects for both the sexes. However, BAI performs less effectively than BMI in OB men and women.

Conclusion:

In summary, the BAI method is a better estimate of adiposity than BMI in non-OB subjects in our Caucasian population. A measurement sensitive to the changes in adiposity for both men and women is suggested to be incorporated into the present BAI equation to increase accuracy.

Introduction

With the rapid increase of obesity prevalence worldwide, attempts to develop a simple and low cost method to estimate adiposity more accurately than body mass index (BMI) are being made (1-3). The BMI ([weight, kg]/[height, m]2) was designed more than 100 years ago (4); due to its low cost and availability, it is the most commonly accepted method for estimating human adiposity by health professionals and the world-wide general population. In addition, BMI is currently the method utilized for the World Health Organization's annual reports regarding obesity. However, although BMI is the most cost effective method to evaluate adiposity, it has been shown to have significant weaknesses in accurately determining adiposity. One such weakness of the BMI is that there are no sex-specific criteria even though gender differences for body fat percentage (%BF) have been well documented (5-7). BMI reflects neither the clear accumulation of body fat with age nor the interindividual difference of %BF among people within the same BMI category (8). According to dual energy X-ray absorptiometry (DXA) %BF measurements, BMI scores can misclassify people by one or even two obesity categories (8, 9). Lastly, due to the fact that BMI was originally developed in Caucasian populations, the accuracy within various ethnic populations is problematic (10).

A study by Bergman et al. (1) recently proposed a new method to determine %BF, which has been labeled the body adiposity index (BAI). The BAI is calculated from two anthropometric measurements, height and hip circumference measurements in centimeters. The BAI equation {Hip, cm/([height, cm]1.5) – 18} was derived from the concordance between the %BF of an Mexican American population with resultants from proposed formulas utilizing various anthropometric measurements. Bergman et al. claim that their BAI equation can estimate percent body fat more reliably than BMI. However, like the BMI equation, it is unknown if BAI reflects the gender differences of both height and hip circumference measurements (11, 12) along with various dynamic anatomical measurements utilized to evaluate adiposity. Our laboratory and others have shown that equations such as BMI, which ignore gender differences for %BF and the accumulation of body fat with age will, inaccurately estimate relative body fat and adiposity status (5, 8, 13, 14). In addition to gender, obesity status can be a critical factor affecting the accuracy of methods that depend on bone structure measurements because there is little change of the skeletal measurement with changes in body weight or adiposity. Presently, there is very little data available regarding the influence of these two important factors on the performance of BAI.

Although Bergman et al. (1) presuppose that BAI could be a better adiposity status measure than BMI, only one study has attempted to validate the BAI equation on a Caucasian population. The letter to the editor by Barreira et al. (15) was the first published material regarding the validation of the BAI equation on a Caucasian population against relative body fat. They found that the correlation between %BF with BAI was very similar to that with BMI (r = 0.82 and r = 0.83, respectively) in women. However, the correlation of %BF with BAI was not as similar BMI (r = 0.75 and r = 0.81) for men. Barreira et al. conclude that BAI equation was an effective method for predicting %BF for a Caucasian population, although further studies are required to support these findings. Therefore, the objectives of this study were to further investigate the performance of BAI according to sexes and obesity statuses (defined by %BF) in our Caucasian population.

Methods

A total of 2,660 subjects were recruited from an ongoing large-scale nutritional genetics study of human complex diseases called the Complex Diseases in the Newfoundland population: Environment and Genetics (CODING) study (8, 16, 17). As BMI and %BF criteria are designed for individuals ≥ 20 years, we excluded all participants below this age, leaving us with a cohort of 2,601 (1,939 females, 663 men). Each individual completed a screening questionnaire that included information regarding physical characteristics, dietary habits, and physical activity levels. Dietary information was obtained from each participant completing the Willett Food Frequency Questionnaire (FFQ), which is a semiquantitative method for the assessment of dietary intake patterns. The Willett FFQ is the most widely used dietary intake questionnaire for the study of nutritional information at the population level (18). Physical activity was measured using the ARIC-Baecke Questionnaire, which consists of a Work Index, Sports Index, and Leisure Time Activity Index (19). The primary method of subject recruitment was the use of posters and handouts. This literature was distributed throughout public facilities (offices, hospitals, and gyms) in the city of St. John's, Newfoundland. Inclusion criteria in the present study were as follows: (i) more than 19 years of age; (ii) at least third generation Newfoundlander; and (iii) healthy, without any serious metabolic, cardiovascular, or endocrine disease. This study was approved by The Human Investigations Committee for the Faculty of Medicine, Memorial University of Newfoundland and Labrador, St John's, NL, Canada. All subjects provided written informed consent. Anthropometrics, body composition, and biochemical measurements were collected following a 12-h fasting period.

Measurements of BAI, BMI, and %BF

Subjects were weighed to the nearest 0.1 kg in standardized clothing as previously described by us (Health O Meter, Bridgeview, IL) (8, 16, 17, 20). Height was measured using a fixed stadiometer (nearest 0.1 cm). Hip circumference was measured as the largest circumference between the waist and thighs. Waist circumference was measured as the horizontal distance around the abdomen at the level of the umbilicus. Hip and waist measurements were recorded to the nearest 0.1 cm using a flexible metric measuring tape while the participant was in a standing position. BAI was calculated based on the equation reported in the Bergman paper (1). BMI was calculated as weight in kg divided by participants' height in m2. Whole body composition measurements including fat mass, lean body mass, and bone mineral densities were measured using DXA Lunar Prodigy (GE Medical Systems, Madison, WI). DXA can produce an accurate measurement of adipose tissue within the body with a low margin of error. For this reason, DXA is considered to be one of the most accurate measurements of adiposity and is commonly used as a standard compared to less-accurate field methods such as BMI. DXA measurements were performed on subjects following the removal of all metal accessories, while lying in a supine position as previously described by us (8, 16, 17, 20). %BF is determined as a ratio of fat mass over total body mass (including bone mineral densities) through the manufacturer's DXA software. Quality assurance was performed on our DXA scanner daily and the typical coefficient of variation was 1.3% during the study period.

Statistical analysis

All data are reported as mean ± SD. The gender differences of variables measured were determined by an independent t test. Pearson correlation analysis was performed to compare the concordance between both BAI and BMI with %BF (%BF) measured by DXA, taking in consideration both gender and adiposity status. Adiposity status (Normal-weight, overweight, or obese) was determined by the Bray criteria (age and gender specific) (21) according to %BF measured by DXA. The Lin's concordance correlation method was applied for both BAI and BMI with %BF, although the correlation coefficients were very poor for both. The Lin's concordance correlation coefficient represents the strength-of-agreement between two continuous variables, which if is lower than 0.90 is considered to be poor. Both of the BAI versus %BF and BMI versus %BF concordance correlation coefficients were much lower than the lowest acceptable coefficient. Pearson correlation analysis was used as the main validation method in this study. SPSS version 17.0 (SPSS, Chicago, IL) was used for all the analyses. Statistical analyses were two-sided and a P value <0.05 was considered to be statistically significant.

Results

Body composition characteristics of subjects in this study are shown in Table 1. Women on average were 3.7 years older than men. As most population studies have shown, women had lower body weight, height, and waist circumference but higher BMI (27.2 men vs. 26.3 women) and hip circumference than men. DXA measurements on body composition revealed an average of 12.8% higher percent body fat in women than men. Interestingly, BAI estimation only demonstrates an average of 6.4% difference between women and men. Pearson correlations of both BAI and BMI with adiposity measurements within the entire study and each gender are shown in Table 2. In the entire cohort, the correlation coefficient between BMI and %BF (r = 0.56) was found to be lower than the correlation coefficient between BAI and %BF (r = 0.78). When analyses were performed according to gender, the Pearson correlation coefficients of BAI with %BF were slightly lower than those of BMI with %BF. Pearson correlations of both BAI and BMI with adiposity measurements among normal-weight (NW), overweight (OW), and obese (OB) groups within the entire study and each gender are shown in Table 3. NW or OW men and women had BAI versus %BF correlation coefficients higher than the BMI versus %BF correlation coefficients. However, BMI versus %BF performed better than BAI versus %BF in OB males and females.

Table 1. Composition characteristics of subjects
 Entire cohort (n = 2,601)Male (n = 662)Female (n = 1,939)P
  • All values are mean ± SD. Gender differences were analyzed by an independent t-test. Significance level for t-tests were set to P ≤ 0.05.

  • a

    Variable significantly greater in women.

  • b

    Variable significantly greater in men.

Agea42.34 ± 13.139.55 ± 14.443.30 ± 12.40.000
Weight (kgb73.28 ± 15.584.70 ± 14.369.38 ± 13.90.000
Height (cm)b165.98 ± 8.7176.37 ± 6.6162.43 ± 6.00.000
Waist (cm)b91.35 ± 14.396.09 ± 13.389.73 ± 14.20.000
Hip (cm)a100.34 ± 11.799.14 ± 9.9100.74 ± 12.20.002
BMI (kg/m2)b26.55 ± 5.027.23 ± 4.326.31 ± 5.20.000
BAIa29.16 ± 6.624.41 ± 4.630.78 ± 6.40.000
Body fat (%)a34.06 ± 9.624.52 ± 7.937.32 ± 7.70.000
Table 2. Pearson correlation between both BAI and BMI with body composition measurements within the entire cohort
 Entire CohortMaleFemale
 BMIBAIBMIBAIBMIBAI
 rPrPrPrPrPrP
  1. Pearson correlation analysis was used to determine the relationships between both BAI and BMI with body composition measurements. Significance levels were set to P ≤ 0.05.

Age0.190.000.370.000.250.000.420.000.190.000.340.00
Weight (kg)0.860.000.350.000.890.000.550.000.930.000.680.00
Height (cm)−0.04NS−0.550.00−0.110.00−0.410.00−0.140.00−0.390.00
Waist (cm)0.830.000.620.000.810.000.750.000.840.000.800.00
Hip (cm)0.830.000.840.000.790.000.860.000.850.000.900.00
Body Fat (%)0.560.000.780.000.700.000.670.000.760.000.740.00
Table 3. Pearson correlation between both BAI and BMI with body composition measurements among NW, OW, OB men and women
 Entire CohortMaleFemale
 Normal weight (n = 849)Normal weight (n = 244)Normal weight (n = 605)
 BMIBAIBMIBAIBMIBAI
 rPrPrPrPrPrP
Age0.190.000.500.000.340.000.490.000.220.000.490.00
Weight (kg)0.780.00−0.170.000.810.000.250.000.820.000.260.00
Height (cm)0.150.00−0.630.00−0.130.05−0.450.00−0.150.00−0.490.00
Waist (cm)0.700.000.280.000.720.000.610.000.630.000.540.00
Hip (cm)0.570.000.650.000.550.000.760.000.590.000.810.00
Body Fat (%)0.05NS0.650.000.460.000.510.000.510.000.510.00
 Overweight (n = 776)Overweight (n = 177)Overweight (n = 599)
Age0.250.000.410.000.180.010.500.000.290.000.470.00
Weight (kg)0.780.00−0.150.000.810.000.160.040.820.000.290.00
Height (cm)0.080.02−0.670.00−0.10NS−0.460.00−0.190.00−0.550.00
Waist (cm)0.700.000.320.000.560.000.600.000.690.000.620.00
Hip (cm)0.590.000.690.000.470.000.770.000.650.000.800.00
Body Fat (%)−0.090.010.620.000.180.020.420.000.430.000.490.00
 Obese (n = 976)Obese (n = 241)Obese (n = 735)
  1. Pearson correlation analysis was used to determine the relationships between both BAI and BMI with body composition measurements. Subjects were classified on the basis of percentage body fat as NW, OW, or OB according to criteria recommended by Bray (21). Significance levels were set to P ≤ 0.05.

Age0.100.000.290.000.150.020.340.000.080.030.220.00
Weight (kg)0.800.000.200.000.860.000.430.000.890.000.540.00
Height (cm)−0.080.01−0.620.00−0.10NS−0.490.00−0.100.01−0.410.00
Waist (cm)0.730.000.490.000.710.000.610.000.760.000.720.00
Hip (cm)0.790.000.800.000.790.000.810.000.790.000.850.00
Body Fat (%)0.340.000.680.000.510.000.400.000.580.000.540.00

Discussion

To our knowledge, this is the first study to attempt to validate the accuracy of the Body Adiposity Index against DXA measured percentage body fat among NW, OW, and OB Caucasian men and women. Simple adiposity assessment methods, such as BMI, are usually used either to evaluate obesity status at an individual level in clinics and health clubs or to estimate the adiposity of a sample population within a research investigation. In the entire cohort of this study, in which male and female subjects of all ages were mixed, the correlation coefficient between BAI and %BF was indeed much higher than that between BMI and %BF. This is generally consistent with the finding obtained in the Mexican American and African American populations, but the correlation coefficient between BAI and %BF in our CODING study was not as high as that shown in these two populations (1). This result is more than likely due to the differences among the physical characteristics of these populations. The data from our large Caucasian population demonstrate that the BAI method is more strongly associated with DXA (the current gold standard for adiposity measurement) %BF than BMI.

Although our data showed that the BAI performs better in estimating adiposity compared with BMI in a large population-based study (men and women combined), what is of concern is that both of the BAI equation variables (hip circumference and height measurements) are essentially bone structure dependent. In addition, data from the adult population of our CODING study show that women are generally shorter in height, have smaller waist, and larger hip circumferences than men. Therefore, longitudinal studies on measurement sensitive to the changes in adiposity similar for both men and women are warranted to aid in the development of an equation to accurately estimate adiposity. Our measurements of percent body fat in the CODING study indicate a large gender difference of adiposity. In the CODING study, women on an average had 12.8% more total body fat than men. Due to the findings from our population studies, we believe it is also important to further evaluate the efficiency of the BAI to predict %BF in women and men separately.

To have an equation that can accurately estimate adiposity for both men and women in the general population would be ideal due to the inherent gender differences. However, considering that hip circumference, the primary measure in the BAI equation, is larger in females than in men (8), it could potentially reflect gender differences to some degree. Conversely BAI adiposity scores are lower for men than women and more closely represent the %BF. Although the estimation of gender difference from the BAI measure is about half the real range measured by DXA, this ability of BAI to properly reflect the gender differences in adiposity is certainly an advantage over BMI.

Finally, we were also interested in investigating the potential influence of adiposity on the accuracy of BAI and the comparison with BMI through DXA as a gold criterion. Therefore, the question that we wished to address was whether or not the two skeletal system-dependent variables would correctly reflect adiposity within various obesity statuses and if gender could be an important factor under these conditions. We found that the BAI method performed better than BMI among normal weight, OW, and OB groups when both women and men were combined. However, when women and men were analyzed separately the BAI only remained more strongly associated for the normal and OW subjects. The BAI versus %BF correlation coefficients for OB men and women were less than that of those for the BMI versus %BF. Our findings indicate that caution should be taken when BAI is used to measure adiposity when OB women and men were evaluated separately. This finding indicates the weakness of BAI as a new evaluation method of adiposity in OB subjects. This weakness could be caused by the lack of measurement necessary to reflect adiposity and that the change in adiposity does not rely on height and hip circumference alone. The poor ability to evaluate adiposity in OB women and men is a problem that should be addressed in future study.

In summary, this study attempted to validate the accuracy by which BAI estimates adiposity among more than 2,600 Caucasian men and women (from the CODING study) along with various adiposity statuses defined by percentage body fat measured by DXA. Our results indicate that the BAI method is a better estimate of adiposity than BMI in non-OB Caucasian subjects. BAI can reflect the gender difference in total %BF between men and women. The %BF evaluated by BAI correlates better than the BMI in NW and OW men and women. However, BAI was less associated with percent body fat than that of BMI in OB men and women. Therefore, caution should be taken when BAI is used to evaluate adiposity in OB individuals. We would suggest that a measurement sensitive to the changes in adiposity for both men and women be incorporated into the present BAI equation to increase accuracy.

Acknowledgements

We greatly appreciate the contributions of the volunteers to this study. We would like to acknowledge Danny Wadden, Alison Jenkins, and Lauren Jones for efforts regarding subject recruitment and data collection. The authors' responsibilities were as follows; Guang Sun was responsible for the study design and writing the manuscript; Farrell Cahill was responsible for the study design, data collection, statistical analysis, and the revision of the manuscript; Wayne Gulliver was responsible for revision of the manuscript; Hong Wei Zhang assisted with data collection.

Ancillary