Body adiposity index and other indexes of body composition in the SAPHIR study: Association with cardiovascular risk factors†‡
Disclosure: The authors declared no conflict of interest.
See the online ICMJE Conflict of Interest Forms for this article.
The accuracy of anthropometric surrogate markers such as the body adiposity index (BAI) and other common indexes like the body mass index (BMI), waist-to-hip ratio (WHR) and waist-to-height ratio (WHtR) to predict metabolic sequelae is essential for its use in clinical practice.
Design and Methods:
Thus, we evaluated the strength of BAI and other indexes to relate with anthropometric parameters, adipocytokines, blood lipids, parameters of glucose-homeostasis and blood pressure in 1,770 patients from the Salzburg Atherosclerosis Prevention Program in Subjects at High Individual Risk (SAPHIR) study in a crosssectional design.
Measurements were BAI, BMI, WHR, WHtR, abdominal subcutaneous and visceral adipose tissue (aSAT and VAT), total body adipose tissue mass, body weight, waist- and hip circumference (WC and HC), leptin, adiponectin, high-density lipoprotein-cholesterol (HDL-C), low-density lipoprotein-cholesterol (LDL-C), triglycerides (TG), fasting plasma glucose, fasting plasma insulin, the homeostasis model assessment of insulin resistance (HOMAIR), systolic and diastolic blood pressure.
Results and Conclusions:
BAI was significantly associated with leptin and HC. We conclude that BAI was the best calculator for leptin. BAI was inferior to BMI to predict anthropometric parameters other than HC, adiponectin, blood lipids, parameters of glucose homeostasis, and blood pressure in this cross-sectional study.
Obesity caused by genetical as well as environmental factors is linked to a variety of metabolic and hormonal dysfunctions such as type II diabetes, insulin resistance, dyslipidemia and hypertension. These comorbidities are capable of major endothelial damage. In 1961 the Framingham heart study evaluated the cardiovascular risk (CVR) (1), which was extended by overweight and obesity subsequently (2). Several studies illustrated a linkage between the accumulation of body adipose tissue, metabolic adverse events and an elevated risk for cardiovascular disease (3, 4, 5, 6).
Numerous methods are known for an exact determination of body composition like magnetic resonance imaging, computed tomography, dual-energy X-ray absorption, isotopic measurement of body water, whole body plethysmography, bioelectrical impedance analysis and underwater weighting. Despite their benefit in accuracy, their use in clinical practice is limited due to a financial and time expenditure. Thus, several so called “surrogate indexes” are in use to easily calculate an approximate amount of body adipose tissue in both men and women. Among others, these indexes include the BMI, waist-to-hip ratio (WHR) and the waist-to-height ratio (WHtR). The BMI is the most widely used index for the estimation of adiposity in clinical practice and surrogates of total body adipose tissue mass.
BMI as a single predictor of CVR was challenged as BMI does neither differentiate between adipose tissue and muscle nor between body adipose tissue compartments, among which visceral adipose tissue (VAT) is rather associated with CVR than subcutaneous adipose tissue (7, 8). Several studies evaluated BMI and other surrogate indexes and compared them against each other in their ability to estimate adiposity and its associated cardiovascular risk (CVR). BMI was criticized for its single estimate of the degree of body adipose tissue, as it was shown that indexes which estimate body adipose tissue distribution were higher correlated with CVR than the single degree of adiposity (9, 10). The INTERHEART study stated that obesity based on WHR would be a better estimate of CVR than BMI (10). WHtR was identified as a proficient predictor of adipose tissue distribution, metabolic risk, and CVR superior to BMI, though to a small degree (11, 12). Despite findings in the INTERHEARD study, results of the DETECT study reconfirmed BMI to be the best predictor for hypertension and close to WHtR for other risk factors of CVR (13). Taking these results together, there is a good evidence that obesity should be evaluated for its body distribution rather than single calculating the amount of total body adipose tissue mass.
Recently the “body adiposity index (BAI)” was described and validated (14). BAI calculates the percentage of body adipose tissue in both men and women of different ethnicity without numerical correction. Bergman et al. developed the BAI using a population study of 1,700 Mexican American. BAI was subsequently validated against dual-energy X-ray absorption measurements of % body adipose tissue as a gold standard in a cross-sectional study population of 223 African Americans (14). BAI was not validated in its ability to predict cardiovascular and metabolic risk factors associated with adiposity. The purpose of was study is to evaluate BAI in its accuracy to estimate anthropometric variables and adipocytokines, blood lipids, variables of glucose homeostasis and blood pressure, to illustrate the correlated CVR as well as to compare its predictive capacity of these variables with BMI, WHR, and WHtR in a cross-sectional population of the Salzburg Atherosclerosis Prevention Program in Subjects at High Individual Risk (SAPHIR) study.
Methods and Procedures
This investigation was based on the SAPHIR-population, a population-based prospective study that investigates the involvement of genetic factors controlling plasma lipid transport and carbohydrate metabolism in the progression of atherosclerotic vascular disease and initiates appropriate interventions in subjects at high CVR. Unrelated male and female subjects between 40 and 70 years of age, who live in the greater Salzburg region and responded to invitations by their family or workplace physician or to announcement in the local press were included in the study. At baseline, all study participants were subjected to a screening program that included a detailed personal and family history, a physical examination and measurement of various biochemical parameters, among other more-specialized procedures. The percentage of subjects with a BMI >30 kg/m2 in this sample was comparable to the reported prevalence of obesity in the same geographic region (15). Both the case-control and the cross-sectional populations comprised only white Europeans, mainly of Bavarian and Austrian German descent. All study subjects provided informed consent, and the study protocol was approved by the local ethics committee.
In total, 1,770 subjects from SAPHIR study were included into our calculations. Among these, 331 subjects were obese, according to a BMI ≥30 kg/m2 (16). For measuring leptin, patients were selected for the following exclusion criteria: lipid-lowering drugs and comorbidities associated to obesity. In total, 390 patients were eligible.
Bioelectrical impedance analysis analyses (lean mass, total body adipose tissue mass) were determined by impedance analysis using InBody 3.0 Body composition Analyser from Biospace Europe (Deitzenbach, Germany) with an integrated scale using software Lookin Body Version 1, Body composition Analysis Management System.
Body height was measured to the nearest 0.1 cm; body weight was measured to the nearest 0. 1 kilogram by using an electronic scale.
BAI was calculated following the formula ((hip circumference)/((height)1.5)-18)), which refers to Bergman et al. (14). BMI was calculated dividing body weight by body height in meters squared. Waist circumference (WC) and hip circumference (HC) were measured using a tapeline at the level midway between the lateral lower rib margin and iliac crest as well as at the levels of trochanters. WHR was calculated as WC divided by HC. WHtR was calculated by dividing WC by height in cm.
All measured variables were classified into one of five predefined variable groups: Group 1 includes abdominal subcutaneous adipose tissue (aSAT), VAT, total body adipose tissue mass, body weight, WC and HC and therefore covered anthropometric measurements. Group 2 includes the adipocytokines leptin and adiponectin, thus covering adipocytokines representative of body adipose tissue mass and CVR. Group 3 includes high-density-lipoprotein-cholesterol (HDL-C), low-density-lipoprotein-cholesterol (LDL-C), and triglycerides (TG) to cover blood lipids. Group 4 was build of variables of glucose homeostasis including plasma fasting glucose, plasma insulin and homeostasis model assessment of insulin resistance (HOMAIR). Group 5 includes the clinical parameters systolic blood pressure and diastolic blood pressure.
ASAT and VAT were assessed by computed tomography scan using a single cross-section at position L4/5 (Picker CTMXTWIN; Picker International, Cleveland, OH).
Diastolic and systolic blood pressure levels were measured in triplicates in a resting state twice and calculated as an average of both measurements, the first being taken after a 5-min rest. After the computer tomography was computed within 30-40 min, the second blood pressure measurement was taken after an additional 5-min rest.
Blood samples were drawn in the morning fasting state from the antecubital vein into EDTA and serum tubes (1.6 mg/ml) and centrifuged immediately after collection at 3,000 rpm for 10 min at 4°C. Plasma and serum samples were stored at –80°C until being used for assaying.
For the evaluation of the correlation of the surrogate indexes, the predefined variable groups (anthropometrical variables, adipocytokines, blood lipids, variables of glucose-homeostasis and blood pressure) were used. For consistency reasons, the negative values of adiponectin and HDL-C were used. Then, the same direction of correlation coefficients could have been expected for all pair wise correlations. Pearson correlation coefficients were first calculated for all possible pairs of the surrogate indexes with the mentioned variables.
For each variable, the differences between two dependent correlations sharing this one variable were tested using Williams's test (17). This includes, for example, all pair wise comparisons of the correlations of aSAT with BAI, aSAT with BMI, aSAT with WHR and aSAT with WHtR. Therefore, for each parameter, 6 tests were calculated, leading to 96 tests altogether. Thus, a Bonferroni corrected P value of smaller than 0.05/96 = 0.00052 was considered to be significant. The Williams test accounts for the correlations between the surrogate indexes, i.e., a small difference of correlations between highly correlated markers (as BMI and WHtR) may be more statistically relevant than a higher difference between only moderately correlated markers (as BAI and BMI).
To get a combined measure for correlation coefficients within each of the five predefined variable groups a fixed effect inverse variance model of the individual Pearson correlation coefficients was calculated after converting into Fishers z-scale (18). The summary effect and its confidence intervals were then retransformed into the original scale yielding a combined measure of correlation for each parameter group. This method is known from meta-analyses and is basically a weighted mean of estimates.
Statistical tests comparing the correlations within variable groups were based on the comparison of the complete correlation matrices (17). Altogether, 25 comparisons of correlation matrices were performed, leading to a significance level of 0.05/25 = 0.002.
We summarized operating characteristics of surrogate indexes in terms of their sensitivity and specificity for specified cutoff values. Receiver operating characteristic curves were constructed for the estimation of every test-variable by each surrogate index. Higher measured values in BAI, BMI, WHtR, and WHR were considered indicative for a metabolic dysfunction. For each specified cutoff value, referring to one of the test-variables, a positive surrogate index test was defined as surrogate index higher than the cutoff value. Wilcoxon-rank-sum-test was computed to test for the null hypothesis that the Area under the receiver operating characteristic-curve (AUC) is 50%, vs. the alternative that it exceeds 50%. U-statistics were performed to statistically compare surrogate-index specific AUCs between each other, according to Delong et al. (19). A post-hoc corrected P value of 0.05/96 = 0.00052 was considered statistically significant for the comparison of AUCs (Bonferroni post-hoc correction for 96 tests).
All analyses were performed using “R” (R Development Core Team, Wilcoxon), the “R”-incorporate-package “psych” was used for testing correlation coefficients and correlation matrices (W. Revelle (2011) psych: Procedures for Personality and Psychological Research Northwestern University, Evanston, http://personality-project.org/r/psych.manual.pdf, 1.1.12).
Baseline values of the cross-sectional population are given in Table 1. Detailed and overall combined estimates for Pearson correlation coefficients of surrogate indexes and variables of each predefined variable group are shown in Table 2. The surrogate index associated with the highest correlative strength to the variable in the same row is highlighted. Two or more highlighted indices are possible, if the dependent correlations did not differ according to Williams test. All results of the correlation differences between the surrogate indexes and the overall combined estimates are shown in the Supplementary Tables S1 and S2 online. Data obtained by receiver operating characteristic curves (ROC) are shown in Table 3. Each first row indicates the area under the ROC curve (AUC) according to one of the surrogate indexes. Each second row indicates the best cutoff value for the specific test. The surrogate index associated with the highest AUC for the variable in the same row is highlighted. Two or more highlighted indices are possible, if the AUCs did not differ according to U-statistics. P values indicate comparisons between highlighted and unlighted areas.
Table 1. Demographics and clinical data of the cross-sectional SAPHIR-population (n = 1,770)
Table 2. Detailed and overall combined correlations between surrogate indexes and variables (n = 1,770)
Table 3. Receiver operator characteristics indicating estimative capacity of surrogate indexes for different test variables
Variable group 1
Regarding anthropometric measurements, correlation coefficient for BAI to aSAT was smaller than for BMI, but higher than WHR. BMI reached the highest correlation coefficient with aSAT compared to other indexes. BAI was significantly less accurate than BMI, WHtR, and WHR in estimating VAT, whereas BMI and WHtR both were superior to BAI and WHR. BAI was inferior to BMI, which was the best surrogate index for total body adipose tissue mass. BAI was inferior to BMI, WHtR, and WHR in its correlation with body weight, whereas BMI was associated with body weight to the highest correlation coefficient. BAI was inferior to BMI, WHtR, and WHR in its correlation to WC, whereas WHR significantly reached the highest correlation coefficient for WC. BAI, BMI, and WHtR were significantly more accurate in its correlation to HC, compared to WHR. BMI reached the highest overall combined estimate for variable group 1, followed by WHtR, BAI, and WHR. Identical results were found for the distinction of variable-measurements indicative for a metabolic dysfunction via ROC curves for each surrogate index. Every surrogate-index-specific AUC significantly differed from 50%, according to Mann–Whitney U-test.
Variable group 2
Concerning measurements of plasma adipocytokine concentrations, correlation coefficient for BAI to leptin was higher than for BMI, WHtR, and WHR. BAI was inferior to BMI, WHtR, and WHR in its estimation of adiponectin plasma concentrations. WHR reached the highest overall combined estimate for variable group 2. Identical results were found for the distinction of variable-measurements indicative for a metabolic dysfunction via ROC curves, except for distinction of adiponectin. WHR was found to have the highest AUC for adiponectin. The AUC for WHR determining leptin was not significantly different to 50%, according to Mann–Whitney U-test.
Variable group 3
Concerning measurements of blood lipids, correlation coefficient for BAI to HDL-C was smaller than for BMI, WHtR, and WHR. WHR significantly reached the highest correlation coefficient with HDL-C compared to BAI, BMI, and WHtR. BAI, WHtR, and WHR were statistically superior to BMI in their estimation of LDL-C. BAI was inferior to BMI, WHtR and WHR in estimating TG. Identical results were found for the distinction of variable-measurements indicative for a metabolic dysfunction via ROC curves, except for distinction of HDL-C and LDL-C. BAI was found to have the lowest AUC for HDL-C, whereas for LDL-C there was no significant difference in AUC between each surrogate index. WHR was found to have the highest AUC for adiponectin. The AUC for BAI determining HDL-C, LDL-C and TG was not significantly different to 50%, according to Mann–Whitney U-test.
Variable group 4
Concerning measurements of glucose homeostasis, BAI was inferior to BMI and WHtR in its estimation of plasma insulin concentrations. BMI and WHtR were significantly superior to BAI and WHR in their correlation with plasma insulin concentrations. Similar results were found for the estimation of plasma fasting glucose, HOMAIR, and the overall combined estimate for variable group 4. Identical results were found for the distinction of variable-measurements indicative for a metabolic dysfunction via ROC curves for each surrogate index. Every surrogate-index-specific AUC significantly differed from 50%, according to Mann–Whitney U-test.
Variable group 5
Concerning measurements of blood pressure levels, BAI was inferior to BMI and WHtR but superior to WHR in its estimation of systolic blood pressure levels. For diastolic blood pressure levels, BAI was inferior to BMI and WHtR. Similar results were found for diastolic blood pressure levels and the overall combined estimate for variable group 5. Identical results were found for the distinction of variable-measurements indicative for a metabolic dysfunction via ROC curves for each surrogate index. Every surrogate-index-specific AUC significantly differed from 50%, according to Mann–Whitney U-test.
Indexes to predict body composition and risk factors of associated comorbidities are widely used in clinical practice. Calculating adipose tissue distribution rather than the single amount of total body adipose tissue mass is thought to be more precise in its prediction of the CVR associated with adiposity. There is good evidence that the CVR increases by more VAT (20, 21). Beyond that, aSAT and VAT are known to show functional differences: A former investigation discovered twice as many macrophages in visceral compared to subcutaneous adipose tissue. These cell-accumulations were significantly associated with a higher incidence for hepatic fibro-inflammatory lesions in obese subjects (22). Measurements of interleukin-6 demonstrated significantly higher concentrations in plasma obtained from the portal vein compared to peripheral venous plasma samples in obese subjects. This indicates VAT as an important source of interleukin-6; beyond that, interleukin-6 concentrations were significantly correlated to C-reactive protein-concentrations. Therefore, VAT accumulation seems to be accompanied by systemic inflammation (23). Higher mRNA concentrations for angiotensinogen were reported for visceral compared to abdominal subcutaneous tissue. Angiotensinogen precedes angiotensin II, which is involved in the pathophysiologic mechanism of hypertension as well as in adipocyte differentiation. This makes VAT a likely conjuncture for hypertension (24).
In this study, BAI could give valuable information about aSAT and the total body adipose tissue masses, whereas BMI was even more precise. Beyond that, BAI was very accurate in estimating plasma concentrations of leptin. Adipocytes secrete leptin, a circulating peptide hormone involved in the regulation of food intake and metabolism. Increasing plasma concentrations of leptin are highly correlated with increasing BMI and body adipose tissue result in an increase of plasma leptin concentrations (25). In addition, leptin plays a role in myocardial metabolism, cardiomyocyte hypertrophy and remodeling as well as in inflammation and immunomodulation (26, 27, 28). Moreover, heart failure results in higher plasma leptin concentrations (29). This led to the suggestion that plasma leptin may be the mechanism linking obesity and heart failure (25, 26), although the mechanism underlying this process is not fully elucidated. In contrast, former prospective studies disagree with the role of leptin as a mediator of heart failure (30, 31).
VAT is closely associated with coronary heart disease (32) and is thought to be a major contributor to vascular risk and the development of type 2 diabetes mellitus. BAI was inferior to BMI and WHtR in its calculation of visceral body adipose tissue mass.
Blood lipids and blood pressure as part of the metabolic syndrome are both known for their contribution to cardiovascular disease and the development of type 2 diabetes mellitus. A study of the Framingham population illustrated an elevated risk for both diseases in patients with metabolic syndrome (33), while another study highlighted the association between dyslipidemia and hypertension with insulin resistance (34). Insulin resistance is known to be associated with risk factors of cardiovascular disease (35, 36), the metabolic syndrome and is thought to arise by central accumulation of adipose tissue (37). In this study, BAI was inferior to BMI, WHtR, and WHR in its estimation of HDL-C and TG. For parameters of glucose homeostasis, BAI was inferior to BMI and WHtR. BAI´s correlation coefficient and AUC was beneath BMI and WHtR in its estimation of systolic and diastolic blood pressure levels.
Findings of the DETECT study showed BMI to be a very precise predictor of total body adipose tissue mass as well as its associated risk for CVR (13). Another former study illustrated a good correlation of VAT mass with BMI in obese women (38). These findings were confirmed in our study population. The combined analysis of correlative accuracy illustrated that BMI was the best predictor of aSAT and total body adipose tissue mass and that BMI was statistically similar to WHtR in its estimation of VAT. Besides that BMI was beneath the accuracy of WHR to predict adiponectin. Concerning parameters of glucose homeostasis and blood pressure, BMI and WHtR were superior to BAI and WHR. Both of them reached small to medium scaled correlation coefficients.
WHtR has been reported to be the best predictor for CVR factors, type 2 diabetes mellitus, hypertension and dyslipidemia in central obesity by several studies. In our study population combined analysis of correlative accuracy illustrated WHtR to be more precise than BMI in estimating VAT mass. Moreover, WhtR gave valuable information about aSAT, total body adipose tissue mass, body weight, WC and HC.
The Dallas Heart study illustrated that WHR was stronger associated with the risk of myocardial infarction and atherosclerosis than BMI (10, 39) and was suggested to be the best measurement of adiposity as it differentiates between central and peripheral body adipose tissue distribution (40). In our study population, WHR was not able to give more precise estimates for anthropometrical measurements than BAI, BMI, or WHtR. WHR was comparable in it estimation of adiponectin, LDL-C, and TG and was the best calculator for HDL-C. Parameters of glucose homeostasis and blood pressure levels were hardly estimated by WHR.
Taking the present observations together, we conclude, that BAI and WHtR were comparable in estimating aSAT and total body adipose tissue mass, whereas BMI was the most accurate surrogate index for both variables. BAI was the best predictor for leptin in our cross-sectional population. The advantage of BAI is that neither a scale nor a gender-specific calculation is needed, which makes this surrogate index very convenient for clinical use. Concerning the estimation of visceral body adipose tissue, WHtR was preverable to other indexes in our study population, while the prediction of glucose homeostasis was equally accurate for BMI and WHtR, although to a limited extend. Neither BAI, BMI, WHtR nor WHR had a high predictive value for both, blood lipids or blood pressure levels in our cross-sectional population.
The expert technical assistance of Karin Salzmann gratefully acknowledged. This study was supported by the Jubiläumsfond der österreichischen Nationalbank (Grant No. 13211) to C.F.E.