Estimation of visceral fat (validation sample)
Descriptive statistics for anthropometric and body composition for the validation and study samples are presented in Table
1. All subjects were female and over 40 years old at examination. The mean height, weight, BMI, DXA total abdominal fat, blood pressure, cIMT, liver function tests and visceral fat do not statistically differ between the two samples, although the mean age for the validation sample was six years older (p = 6.5 × 10
-11) with a larger waist circumference (p = 2.05 × 10
-6) than the study sample. The study sample prevalence and 95% confidence interval for morbidity related to the quantitative traits presented in Table
1 were as follows: T2D 0.046 (0.04-0.05), HT 0.08 (0.07-0.09), cIMT 0.27 (0.24-0.30), ALT 0.22 (0.21-0.24), ALK 0.27 (0.26-0.29), BIL 0.02 (0.02-0.03) and GGT 0.20 (0.18-0.21).
Table 1
Validation and study sample characteristics
Age (years) | 60.4 | 6.1 | 49.3 | 72.8 | 54.2 | 8.3 | 40.0 | 79.5 |
Weight (kg) | 65.7 | 9.4 | 48.4 | 87.4 | 66.6 | 11.8 | 35.6 | 139.5 |
Height (m) | 1.62 | 0.06 | 1.48 | 1.75 | 1.62 | 0.06 | 1.39 | 1.82 |
Waist circumference (cm) | 88.0 | 9.9 | 66.6 | 111.2 | 81.0 | 11.0 | 55.0 | 134.0 |
Sagittal depth (cm) | 21.8 | 3.2 | 15.9 | 31.1 | - | - | - | - |
Scan difference (years) | 1.3 | 0.8 | 0.2 | 2.5 | - | - | - | - |
BMI (kg/m2) | 25.1 | 3.8 | 19.2 | 33.8 | 25.6 | 4.5 | 15.1 | 51.7 |
Total abdominal fat (kg) | 1.40 | 0.62 | 0.24 | 3.08 | 1.44 | 0.61 | 0.14 | 3.94 |
VAT area (cm2) | 127.8 | 52.1 | 37.7 | 279.5 | 144.6 | 49.6 | 37.7 | 347.4 |
Diastolic BP (mmHG) | 77.3 | 8.3 | 61.0 | 95.5 | 75.9 | 8.9 | 47.5 | 108.0 |
Systolic BP (mmHG) | 122.0 | 11.5 | 92.0 | 151.0 | 123.1 | 14.7 | 86.5 | 189.0 |
cIMT | 0.68 | 0.08 | 0.53 | 0.82 | 0.67 | 0.11 | 0.30 | 1.11 |
ALT | 23.9 | 11.8 | 3.0 | 68.0 | 26.7 | 11.4 | 2.5 | 217.3 |
ALK | 64.8 | 19.0 | 26.0 | 114.0 | 71.2 | 18.2 | 23.5 | 218.9 |
BIL | 9.5 | 3.7 | 5.7 | 23.5 | 8.7 | 3.0 | 1.0 | 30.5 |
GGT | 30.3 | 17.4 | 12.0 | 65.3 | 27.9 | 21.7 | 3.0 | 359.0 |
The Pearson product moment correlation coefficients (
r) between the different adiposity measures and CT measured VAT area for the validation sample are presented in Table
2. CT-VAT area was most strongly correlated with CT-measured body cavity cross sectional area (
r = 0.85), sagittal depth (
r = 0.84) and tape-measured waist circumference (
r = 0.86) and DXA total abdominal fat (
r = 0.79). Consistent with these data, we observed that reported models of visceral fat in the literature, whether linear regressions or derived anthropometric indices, all attempt to capture information about the body cavity volume (or area) in relation to the subcutaneous volume [
8,
10,
11,
15]. We used this insight to guide our choice of linear regression to estimate CT-measured visceral fat.
Table 2
Validation sample (n = 54) correlation coefficients between CT visceral adipose fat (VAT) area, anthropometric and abdominal fat measures
BC |
0.85
| | | | | | | | | | |
Sub. CSA | 0.58 | 0.44 | | | | | | | | | |
Total CSA | 0.81 | 0.80 | 0.90 | | | | | | | | |
SD |
0.84
| 0.80 | 0.84 | 0.96 | | | | | | | |
WC |
0.86
| 0.77 | 0.83 | 0.94 | 0.94 | | | | | | |
TID | 0.66 | 0.72 | 0.46 | 0.67 | 0.61 | 0.65 | | | | | |
TED | 0.66 | 0.56 | 0.89 | 0.87 | 0.82 | 0.85 | 0.62 | | | | |
SFW | 0.43 | 0.26 | 0.83 | 0.69 | 0.66 | 0.67 | 0.17 | 0.88 | | | |
DXA |
0.79
| 0.56 | 0.68 | 0.74 | 0.76 | 0.77 | 0.56 | 0.75 | 0.6 | | |
BMI | 0.71 | 0.60 | 0.80 | 0.84 | 0.83 | 0.82 | 0.57 | 0.85 | 0.72 | 0.79 | |
Weight | 0.67 | 0.64 | 0.77 | 0.84 | 0.76 | 0.81 | 0.69 | 0.88 | 0.68 | 0.69 | 0.86 |
Although the DXA scans were collected between 0.23 – 2.3 years after the CT scans for the validation sample, no evidence was observed for significant change in weight in these individuals nor was change in weight correlated with time lapse between scan dates (data not shown).
Table
3 presents the results for three previously published DXA-based regression models and anthropometric indices for estimating visceral fat, applied to the TwinsUK CT validation sample. The best-replicated regression models included DXA trunk fat and sagittal depth (
R
2 ≈ 0.8), while a combination of DXA and skin fold was less predictive of visceral fat. The best individual indices were functions of sagittal depth (SD), SFW, TID and TED (r ≥ 0.85, equivalent to r
2 ≥ 0.72). We note that the most reproducible indices of visceral fat all relate to body cavity CSA. By assuming body cavity CSA takes the form of an ellipse, we estimated this as body cavity CSA = π x (SD – 2SFW) × TID for our validation sample.
Table 3
Visceral adipose tissue area (VAT area) linear model estimates and correlational indices
A
| | | |
Snijder et al. (2002) [ 11] | DXA trunk fat + sagittal depth | 0.74 | 0.80 |
DXA trunk fat + abdominal circumference | 0.71 | 0.78 |
Treuth et al. (1995) [ 15] | Sagittal depth + age + waist circumference +% DXA trunk fat | 0.81 | 0.79 |
| DXA + skin fold | 0.68 | 0.65 |
B
|
Index
|
Reported
r
|
TwinsUK
|
|
r
|
Abdominal fat mass (kg) | 0.57 | 0.79 |
Thigh fat mass (kg) | 0.06* | - |
Abdominal fat mass/thigh fat mass | 0.75 | - |
Abdominal fat mass/SFW | 0.83 | 0.58 |
TED (cm) | 0.54 | 0.61 |
TID (cm) | 0.9 | 0.61 |
SFW (cm) | −0.23* | 0.28 |
(SD)(TID) | 0.89 | 0.87 |
(SD)(TID)/height | 0.91 | 0.86 |
(SD)(TID)/BMI | 0.66 | 0.49 |
(SD-SFW) | 0.86 | 0.89 |
(SD-SFW)(TID) | 0.92 | 0.79 |
| (SD-SFW)(TID)/height | 0.94 | 0.87 |
In modelling CT-measured VAT area, the best fit and most interpretable model included a combination of measures for DXA abdominal fat, body cavity cross sectional area (estimated using the ellipse formula above) and waist circumference (Table
4, Model 0), which together explained 91% of the variance in CT-measured VAT area (
R
2 = 0.91). However, since sagittal depth was not available for our study sample for which we wished to estimate VAT, we also assessed a model including only DXA total abdominal fat, tape-measured waist circumference and age. For this (Table
4, Model 1) we obtained a model with an
R
2 of 0.83 with the following linear regression equation: VAT area = 10.1(DXA abdominal fat mass) + 40.8(waist circumference) + 1.4(age) using standardised explanatory variables and no intercept term. For this estimate, a Bland-Altman analysis showed no evidence of heteroscedascity across the full range of CT-measured VAT area, with only 2 out of 54 (3.7%) values with a difference outside the 95 limits of agreement (the mean difference ± twice the standard deviation of the difference between the two measures).
Table 4
Linear regression models for CT visceral adipose fat (VAT) area using the validation sample (n = 54)
A
|
Model 0:
| | | | | | | | 0.91 |
Combination of DXA & anthropometric measures | DXA abdominal fat | 20.1 | 3.4 | 5.9 | 2 × 10-9
| 13.2 | 27.0 | |
BC CSA | 32.4 | 4.5 | 7.2 | 4 × 10-13
| 23.2 | 41.6 | |
WC | 11.1 | 5.6 | 2.0 | 2 × 10-2
| -0.3 | 22.4 | |
B
|
Model 1:
| | | | | | | | 0.83 |
Combination of DXA & anthropometric measures | DXA abdominal fat | 10.1 | 4.8 | 2.1 | 0.04 | 0.31 | 19.9 | |
WC | 40.8 | 5.7 | 7.2 | 3 × 10-13
| 29.2 | 52.3 | |
Age | 1.4 | 0.5 | 2.6 | 0.01 | 0.3 | 2.4 | |
C
|
Model 2:
| | | | | | | | 0.86 |
Anthropometric measures only | BC CSA | 25.5 | 5.6 | 4.6 | 2 × 10-6
| 14.1 | 36.8 | |
| | WC | 30.5 | 5.5 | 5.6 | 1 × 10-8
| 19.4 | 41.6 | |
In relation to efforts attempting to estimate visceral fat using only anthropometric measures [
8,
33], we also obtained a highly explanatory model (
R
2 = 0.86) with a linear equation using only two CT measures of body cavity CSA (estimated as an ellipse) and waist circumference (Table
4, Model 2)
. This figure rose to
R
2 = 0.89 using body cavity components SD, SFW and WC as explanatory variables for CT VAT area (data not shown). Body cavity CSA results are presented, as this model is more interpretable, while the model including SD, SFW and WC yields a negative beta coefficient for SFW due to co-linearity. Again, these simple anthropometric models could not be used for the study sample however, as sagittal depth was not recorded along with DXA scan for these subjects.
In addition to these validation models, we indirectly assessed the validity of our study sample estimates of visceral fat by making two observations:
1.
Realised estimates of VAT area (VATModel1 and VATModel2) for the validation sample were equally or more strongly correlated with VATCT (r = 0.89 and r = 0.93) than DXA total abdominal fat (r = 0.88 and r = 0.70, respectively);
2.
Bivariate variance component analysis between VATModel1 and DXA total abdominal fat provided strong statistical evidence (Δχ2 1 = 43.7, p = 4x10-11) for a specific heritable component that was unique to VATModel1 and not shared with DXA total abdominal fat (data not shown).
Visceral fat as a risk factor of morbidity (study sample)
For the study sample we estimated visceral fat as a linear function of DXA total abdominal fat, waist circumference and age (Model 1, Table
4). Although a model including DXA total abdominal fat, waist circumference and estimated body cavity area was a better fit (Model 0, Table
4), we used Model 1 as we did not have a measure of sagittal depth for the study sample subjects required to estimate body cavity area from the DXA images.
The correlations between study sample explanatory variables are presented in Additional file
1: Table S1. As a function of these variables, VAT area is most strongly correlated with waist circumference and DXA total abdominal fat. While the univariate odd ratios for VAT area, DXA abdominal fat and BMI were all significantly associated with each morbidity (Tables
6,
7,
8 and
9), VAT area was most consistently and strongly associated with four morbidity traits - T2D, HT, cIMT and ALT. Residual analyses for these four morbidities were consistent with visceral fat entirely mediating the association with other measures of adiposity (BMI and total abdominal fat), but this was not true for the liver function tests ALK, BIL and GGT (details provided in supplementary Tables online).
Table 6
Type 2 diabetes (T2D) and adiposity
A
| | | | | | | | |
| VAT | 2.17 | 0.18 | 9.5 | <2 × 10-16
| 1.85 | 2.54 | 0.07 |
DXA abdominal fat | 1.86 | 0.13 | 8.6 | <2 × 10-16
| 1.61 | 2.14 | 0.05 |
BMI | 1.66 | 0.12 | 7.2 | 2 × 10-13
| 1.45 | 1.91 | 0.04 |
Age | 1.05 | 0.01 | 4.3 | 8 × 10-6
| 1.03 | 1.07 | 0.02 |
B
| | | | | | | | |
| VAT | 2.08 | 0.18 | 8.5 | <2 × 10-16
| 1.76 | 2.47 | 0.08 |
| Age | 1.02 | 0.01 | 2.0 | 0.05 | 1.00 | 1.05 | |
Table 7
Hypertension and adiposity
A
| | | | | | | | |
VAT | 2.08 | 0.16 | 9.5 | <2 × 10-16
| 1.79 | 2.42 | 0.08 |
DXA abdominal fat | 1.77 | 0.14 | 7.4 | 6 × 10-14
| 1.53 | 2.07 | 0.05 |
BMI | 1.77 | 0.13 | 7.7 | 6 × 10-15
| 1.53 | 2.05 | 0.06 |
Age | 1.07 | 0.01 | 6.4 | 6 × 10-11
| 1.05 | 1.09 | 0.05 |
B
| | | | | | | | |
VAT | 1.90 | 0.17 | 7.4 | 9 × 10-14
| 1.60 | 2.25 | 0.10 |
| Age | 1.04 | 0.01 | 4.0 | 4 × 10-5
| 1.02 | 1.07 | |
Table 8
Sub-clinical atherosclerosis and adiposity
A
| | | | | | | | | | |
| VAT | 1.50 | 0.09 | 6.6 | 2 × 10-11
| 1.33 | 1.69 | 43.8 | 1 | 4 × 10-11
|
| DXA abdominal fat | 1.29 | 0.07 | 4.5 | 4 × 10-6
| 1.16 | 1.45 | 20.0 | 1 | 8 × 10-6
|
| BMI | 1.39 | 0.08 | 5.5 | 2 × 10-8
| 1.23 | 1.55 | 30.5 | 1 | 3 × 10-8
|
| Age | 1.08 | 0.01 | 6.3 | 1 × 10-10
| 1.05 | 1.10 | 40.0 | 1 | 3 × 10-10
|
B
| | | | | | | | | | |
| VAT | 1.36 | 0.10 | 4.4 | 5 × 10-6
| 1.19 | 1.56 | 59.6 | 2 | 1 × 10-13
|
| Age | 1.06 | 0.01 | 5.0 | 3 × 10-7
| 1.04 | 1.09 | | | |
Table 9
Liver function tests (LFTs) and adiposity
ALT (0.22) | | | | | | | | 0.09 |
| VAT | 1.75 | 0.09 | 10.9 | <2 × 10-16
| 1.58 | 1.93 | |
| Age | 1.02 | 0.01 | 2.9 | 0.004 | 1.01 | 1.03 | |
ALK (0.27) | | | | | | | | 0.09 |
| VAT | 1.28 | 0.14 | 2.4 | 0.02 | 1.05 | 1.58 | |
| DXA abdominal fat | 1.26 | 0.13 | 2.3 | 0.02 | 1.03 | 1.54 | |
| Age | 1.06 | 0.01 | 9.8 | <2 × 10-16
| 1.05 | 1.07 | |
BIL* (0.02) | | | | | | | | 0.054 |
| VAT | 0.62 | 0.10 | −3.0 | 0.003 | 0.46 | 0.85 | |
BIL* (0.02) | | | | | | | | 0.049 |
| DXA abdominal fat | 0.67 | 0.10 | −2.7 | 0.01 | 0.50 | 0.90 | |
GGT (0.20) | | | | | | | | 0.06 |
| VAT | 1.25 | 0.14 | 1.9 | 0.05 | 1.00 | 1.56 | |
| DXA abdominal fat | 1.36 | 0.15 | 2.8 | 0.01 | 1.10 | 1.70 | |
| Age | 1.02 | 0.01 | 2.7 | 0.007 | 1.00 | 1.03 | |
For type 2 diabetes, while the univariate ORs for the 3 adiposity measures were all associated with T2D (Table
6A), visceral fat and age provided the best-fit multiple regression model (pseudo-
R
2 = 0.08, Table
6B and LRT, Additional file
1: Table S2), with an adjusted OR of 2.08 (95% CI 1.76 – 2.47) per standard deviation increment in VAT area including age. Removing VAT area from the full model resulted in a significant decline in model fit (LRT χ
2
1 = 14.4, p = 1E-04), while the removal of DXA total abdominal fat and BMI, either individually or together (χ
2
2 = 1.9, p = 0.39) did not reduce the model fit (Additional file
1: Table S2).
Hypertension was equally strongly associated with VAT area, DXA abdominal fat and BMI for univariate analyses (Table
7A), but visceral fat and age provided the best-fit multiple regression model (pseudo-
R
2 = 0.10, Table
7B and LRT, Additional file
1: Table S4), with an adjusted OR of 1.90 (95% CI 1.60 – 2.25) per standard deviation increment in VAT area including age. Removing VAT area from the full model resulted in a nominal decline in model fit (LRT χ
2
1 = 7.1, p = 0.01), whilst the removal of DXA total abdominal fat and BMI, either individually or together (χ
2
2 = 1.99, p = 0.37) did not reduce the model fit (Additional file
1: Table S4).
The prospective analysis of subclinical atherosclerosis had a median follow up time of 9.7 years, during which a total of 221 (27%) individuals were classified as sub-clinically atherosclerotic. Univariate Cox proportional hazard models showed all three measures of adiposity and age at baseline to be associated with incident cIMT (Table
8A), with VAT area (χ
2
1 = 43.8) and age (χ
2
1 = 40) providing the best-fit parsimonious model (Table
8B). DXA and BMI could be dropped from the full model with no nominal (χ
2
2 = 5.7, p = 0.06) deterioration in model fit (Additional file
1: Table S6).
All LFT protein serum levels were positively associated with measures of adiposity (Table
9), except for bilirubin, which was negatively associated. VAT area remained associated with alanine transaminase (ALT) when conditioned upon DXA total abdominal fat and BMI, while VAT area and DXA were still associated with ALK, BIL and GGT conditional upon BMI (Table
9 and related Additional file
1: Tables S8-S15). Removing VAT area from the full model for ALT resulted in a significant decline in model fit (LRT χ
2
1 = 19.6, p = 1E-05), while the removal of DXA total abdominal fat and BMI, either individually or together (χ
2
2 = 1.56, p = 0.46) did not (Additional file
1: Table S8).
The estimated variance inflation factor (VIF) between VAT area and DXA abdominal fat, BMI and age was 8.84. Analyses repeated using subgroups of data and analyses using underlying quantitative traits for HT, ALT, ALK, BIL and GGT all provided qualitatively the same association results (data not shown).