We hypothesized that anthropometric measures of abdominal obesity would have a stronger positive association with nonalcoholic fatty liver disease (NAFLD) measured by noncontrast computed tomography versus general measures of obesity. The Multiethnic Study of Atherosclerosis comprised participants aged 45 to 84 years free of known cardiovascular disease. We studied 4,088 participants with adequate liver and spleen computed tomography imaging and no previous use of oral steroids, class 3 antiarrhythmics, moderately heavy alcohol use, or cirrhosis. Prevalent NAFLD was defined as a liver:spleen Hounsfield attenuation ratio of <1. Multivariable log-linear regression modeled the association of 4 obesity measures—weight, body mass index (BMI), waist circumference, and waist-to-hip ratio—with prevalent NAFLD. Receiver-operator curve analysis compared NAFLD discrimination. Median age was 63 years, and 55% were women. For each obesity measure, adjusted prevalence ratios for NAFLD were fourfold to fivefold greater in the highest versus the lowest quartile (p <0.001). Waist circumference and BMI had the highest prevalence ratios, and waist circumference had the best discrimination, for NAFLD in the total population, although an abnormal BMI categorized subjects with NAFLD as well if not better than waist circumference. In ethnic-specific analysis, whites and Chinese had the strongest association of obesity and NAFLD compared with other ethnicities. In conclusion, although waist circumference provided the best discrimination for NAFLD, BMI may perform similarly well in clinical settings to screen for NAFLD.
Nonalcoholic fatty liver disease (NAFLD) is the most common cause of chronic liver disease in the United States affecting 30% of the general population and up to 80% of those with obesity or diabetes. NAFLD is strongly linked to metabolic risk factors, including diabetes, insulin resistance, and inflammation, and cardiovascular disease is a more common cause of death in patients with NAFLD than liver disease. Some studies have suggested that, similar to the metabolic syndrome, NAFLD is more strongly associated with visceral fat accumulation than other fat distributions. However, this has not been seen in all populations, and many of these studies had limitations such as enrolling relatively small sample sizes, comparing only 2 anthropometric obesity measures, or using less sensitive estimates of NAFLD, such as elevated liver enzymes. The aim of this study was to use the multicenter Multiethnic Study of Atherosclerosis (MESA) cohort to examine the association of 4 anthropometric measures of obesity with NAFLD measured by computed tomography (CT). We hypothesized that measures of abdominal obesity, such as waist circumference or WHR, would be more strongly associated with NAFLD than more general measures of obesity.
Methods
The MESA is a cohort study aiming at investigating the prevalence, correlates, and progression of subclinical cardiovascular disease. Details of its design have been reported. MESA includes 6,814 men and women aged 45 to 84 years, free of clinical cardiovascular disease at baseline (2000 to 2002), recruited from 6 US field centers. Approximately 53% of the cohort is women, 38% white, 12% Chinese, 28% black, and 22% Hispanic.
We excluded all participants with CT imaging that did not extend inferiorly enough to measure attenuation of both the liver and the spleen (n = 2,430). We also excluded participants with a history of moderately heavy alcohol use (>7 drinks per week in women and >14 drinks per week in men, n = 219), self-reported cirrhosis (n = 5), and those using oral steroids (n = 70) and class 3 antiarrhythmics (n = 2) as use of these agents can cause macrovesicular steatosis. Our final study population was 4,088.
Participants completed a self-administered questionnaire on demographics and medical and family histories. Anthropometric measures were performed in light clothing and no shoes. Body mass index (BMI) was calculated as weight (kg) divided by height (m 2 ). Waist circumference was measured horizontally at the level of the umbilicus. Hip girth was measured at the maximum circumference of the buttocks. The ratio of waist circumference to hip girth defined waist-to-hip ratio (WHR). Systolic blood pressure was measured in a seated position 3 times with a Dinamap model Pro 100 automated oscillometric sphygmomanometer (Critikon; Wipro GE Healthcare, Waukesha, Wisconsin); the final 2 measurements’ average was used for analysis. Hypertension was defined by the stages of Joint National Commission on the Prevention, Detection, Evaluation and Treatment of High Blood Pressure VII criteria. Diabetes was defined as a fasting glucose ≥126 mg/dl, self-reported earlier diagnosis, or use of diabetes medication. Blood samples were obtained after a 12-hour fast and were used to measure glucose, lipid profile, C-reactive protein, and interleukin-6.
After providing informed consent, all participants underwent 2 consecutive baseline noncontrast cardiac CT scans, as previously described. Three sites used the Imatron C-150XL CT scanner (GE-Imatron, San Francisco, California) and 3 sites used multidetector CT scanners (4 slices). Each scan was performed from the carina to below the apex of the heart during a breath hold, which, in most cases, contains images of the liver and spleen. Scans were read independently by 2 experienced readers, blinded to demographic data. The liver-to-spleen attenuation ratio was selected as the most stable measure of hepatic fat content, and a liver-to-spleen ratio (LSR) of <1.0 was defined a priori as the cutpoint for NAFLD. The largest scan span was selected for measurement of liver fat. Hepatic and splenic Hounsfield unit attenuation values were measured using regions of interest >100 mm 2 . There were 2 regions of interest in the right liver lobe anteroposteriorly, 1 in the left lobe and 1 in the spleen. Regions of interest with larger areas were used whenever possible. LSR was calculated by taking the mean Hounsfield unit measurement of both right liver lobe regions of interest and dividing it by the spleen Hounsfield unit measurement. MESA reproducibility and variability levels for LSR have been published.
Differences in baseline characteristics between those with and without NAFLD were compared using analysis of variance for continuous variables and the chi-square tests for categorical variables. The Mann-Whitney-Wilcoxon rank-sum was used to compare C-reactive protein and triglycerides. Because the prevalence of NAFLD was >10%, prevalence ratios, rather than odds ratios, were calculated from the regression model y = exp(X T β), assuming Gaussian error and using robust standard error estimates; the exponentiated parameter β is interpreted as the prevalence ratio. The 4 primary predictor variables were the anthropometric obesity measures of weight (lbs), BMI (kg/m 2 ), waist circumference (cm), and WHR; each obesity measure was modeled in a separate regression model. Linear assumptions between predictor and outcome variables were checked. Prevalence ratios were calculated for the highest versus lowest quartile of each obesity measure. An unadjusted model and a model adjusted for age, gender, race-ethnicity, and MESA site were fitted. Receiver-operator curve (ROC) analysis yielded areas under the curve (AUC) to assess the discrimination of LSR <1.0 for each obesity measure. Tests of equality compared the AUCs from models of each obesity measure, and the chi-squared and Bonferroni-corrected p values were calculated.
For regressions that used the highest versus lowest quartile of obesity measure, a p value for linear trend across all quartiles is reported. In race-ethnic strata where highest versus lowest quartile analysis was used, race-ethnicity–specific quartiles of each obesity measure were re-calculated. To perform a “discordance” analysis, we used World Health Organization and Adult Treatment Panel III cutoffs to dichotomize BMI and waist circumference, respectively, as either “abnormal” or “normal” for every subject in the cohort: “abnormal” BMI is defined as BMI ≥30 in both genders, “normal” is BMI <30; “abnormal” waist circumference is defined as >102 cm for men and >88 cm women; “normal” waist circumference is ≤102 cm in men and ≤88 cm in women. We then fitted a regression model comparing groups of subjects with each combination of BMI and waist circumference “normality” for the outcome of LSR <1. A p value ≤0.05 was considered statistically significant for all analyses. All analyses were performed using STATA 10.0 (Stata Co., College Station, Texas).
Results
The prevalence of NAFLD in our sample was 17.3% and was similar in women and men; in whites, Chinese, blacks, and Hispanics, it was 15.2%, 20.2%, 11.2%, and 27.1%, respectively. The range of each obesity measure was weight (85.8 to 314.4), BMI (15.9 to 54.5), waist circumference (61 to 156), and WHR (0.6 to 1.3). NAFLD participants were younger and had higher obesity measures, diabetes, fasting glucose, hypertension and mean systolic blood pressure, triglycerides, CRP and IL-6, lower education level, and mean HDL ( Table 1 ).
| Variable | Non-Alcoholic Fatty Liver Disease | p-value | |
|---|---|---|---|
| NO (n=3,382) | YES (n=706) | ||
| Male | 1,519 (44.9%) | 326 (46.2%) | 0.54 |
| Age (year) mean ± SD | 63.3 ± 10.5 | 61.0 ± 9.6 | <0.01 |
| White | 1,274 (84.8%) | 229 (15.2%) | <0.01 |
| Chinese | 317 (79.9%) | 80 (20.2%) | |
| Black | 1,095 (88.8%) | 138 (11.2%) | |
| Hispanic | 696 (72.8%) | 259 (27.1%) | |
| Education >High School | 2,775 (82.4%) | 525 (74.7%) | <0.01 |
| Weight (lbs), mean ± SD | 170.1 ± 36.6 | 187.3 ± 37.8 | <0.01 |
| BMI (kg/m 2 ), mean ± SD | 28.1 ± 5.2 | 31.1 ± 5.4 | <0.01 |
| Waist Circ. (cm), mean ± SD | 97.3 ± 13.6 | 105.7 ± 13.4 | <0.01 |
| Waist to Hip Ratio, mean ± SD | 0.925 ± 0.078 | 0.965 ± 0.063 | <0.01 |
| Diabetes Mellitus | |||
| Normal | 77.9 (2,624) | 55.6 (392) | <0.01 |
| Impaired Fasting Glucose | 10.7 (361) | 21.8 (154) | |
| Untreated Diabetes | 2.2 (75) | 7.1 (50) | |
| Treated Diabetes | 9.1 (308) | 15.5 (109) | |
| Fasting Glucose (mg/dl), mean ± SD | 95.6 ± 28.3 | 108.5 ± 39.0 | <0.01 |
| Hypertension | |||
| Normal | 1,471 (43.5%) | 24 (34.3%) | <0.01 |
| Prehypertension | 1,035 (30.6%) | 267 (37.8%) | |
| Hypertension Stage 1 | 630 (18.6%) | 130 (18.4%) | |
| Hypertension Stage 2 | 243 (7.2%) | 67 (9.5%) | |
| Smoking | |||
| Never Smoker | 1,744 (51.8%) | 393 (55.9%) | 0.14 |
| Former Smoker | 1,230 (36.5%) | 235 (33.4%) | |
| Current Smoker | 395 (11.7%) | 75 (10.7%) | |
| Systolic blood pressure (mmHg), mean ± SD | 126.7 ± 21.6 | 130.1 ± 20.7 | <0.01 |
| Total Cholesterol (mg/dl), mean ± SD | 193.9 ± 34.9 | 194.5 ± 39.0 | 0.64 |
| LDL Cholesterol (mg/dl), mean ± SD | 117.9 ± 31.2 | 115.9 ± 31.1 | 0.12 |
| HDL Cholesterol (mg/dl), mean ± SD | 51.6 ± 14.8 | 44.5 ± 11.9 | <0.01 |
| Triglycerides median (mg/dl), median (IQR) | 104 (74-151) | 154 (105-211) | <0.01 |
| C-Reactive Protein (mg/L), median (IQR) | 1.8 (0.8-4.1) | 3.0 (1.4-6.5) | <0.01 |
| Interleukin-6 (pg/mL), mean ± SD | 1.53 ± 1.18 | 1.88 ± 1.31 | <0.01 |
For each obesity measure, the NAFLD prevalence ratio ranged from fourfold to fivefold greater in the highest versus the lowest quartile, after adjustment for age, gender, race, and MESA site ( Table 2 ), with waist circumference and BMI demonstrating the strongest prevalence ratios. A significant and graded relation was observed between each increasing quartile of all baseline obesity measures and NAFLD (p <0.001 for linear trend, data not shown). In sensitivity analyses to confirm the overall direction of association, absolute change in continuous LSR was calculated for 1 SD increase in each baseline obesity measure: correlating with a decrease of approximately 0.05 in LSR, for all measures (p <0.001 for all measures). Waist circumference demonstrated the largest decrease in LSR, consistent with the strongest association with NAFLD (data not shown).
| Prevalence Ratio (95% CI) | ||||||||
|---|---|---|---|---|---|---|---|---|
| Weight (lbs) | p-value | BMI | p-value | Waist Circ (cm) | p-value | WHR | p-value | |
| Model 1 | 2.81 (2.27-3.49) | <0.001 | 4.62 (3.54-6.03) | <0.001 | 4.43 (3.42-5.74) | <0.001 | 3.79 (2.96-4.86) | <0.001 |
| Model 2 | 4.36 (3.29-5.79) | <0.001 | 5.69 (4.19-7.73) | <0.001 | 5.82 (4.30-7.88) | <0.001 | 4.41 (3.30-5.88) | <0.001 |
Race-ethnicity–stratified analysis was performed since heterogeneity between obesity measures, and race-ethnicity was tested and found to be significant for obesity measure/ethnicity combinations. In each race-ethnicity stratum, the highest versus lowest quartile of each obesity measure was strongly positively associated with NAFLD ( Table 3 ). In MESA, whites and Chinese demonstrated higher prevalence ratios for NAFLD for nearly every obesity measure (with the exception of Chinese weight) compared with those for blacks and Hispanics.
| Weight (lbs) | p-value | BMI (kg/m 2 ) | p-value | Waist Circ (cm) | p-value | WHR | p-value | |
|---|---|---|---|---|---|---|---|---|
| White | 6.89 (4.16-11.40) | <0.001 | 6.74 (4.11-11.06) | <0.001 | 7.93 (4.72-13.33) | <0.001 | 9.50 (5.50-16.41) | <0.001 |
| Chinese | 4.23 (1.95-9.18) | <0.001 | 6.09 (2.63-14.12) | <0.001 | 4.79 (2.23-10.28) | <0.001 | 5.22 (2.29-11.88) | <0.001 |
| Black | 5.76 (3.08 -10.77) | <0.001 | 4.43 (2.45-8.01) | <0.001 | 4.27 (2.42-7.53) | <0.001 | 3.36 (2.01-5.62) | <0.001 |
| Hispanic | 2.52 (1.77-3.57) | <0.001 | 3.42 (2.29-5.11) | <0.001 | 3.44 (2.31-5.11) | <0.001 | 2.34 (1.56-3.21) | <0.001 |
ROC analysis provided AUC estimates for the ability of each obesity measure to discriminate NAFLD using an adjusted regression model ( Table 4 ). Waist circumference had the highest AUC compared with BMI, WHR, and weight. This difference persisted in male and female strata. A significance test for equality demonstrated that the waist circumference AUC was significantly higher than that all other obesity measures in the total population and in both gender strata ( Tables 4 and 5 ); BMI had the second highest AUC in each of these strata. Among ethnic strata, waist circumference demonstrated the highest AUC in all ethnicities except Hispanics where BMI had a marginally higher AUC (data not shown). Within ethnic strata, however, the AUC of waist circumference and BMI were not statistically different. Waist circumference AUCs were tested for equality between ethnicities and that of whites (0.6979) was significantly higher than that of the other ethnicities; the waist circumference AUC of Chinese was the second highest (0.6868) and was significantly higher than that of blacks (0.6715) but not of Hispanics (0.6803).
| n | AUC | 95% CI | |
|---|---|---|---|
| Weight | 4088 | 0.7124 | (0.691-0.733) |
| BMI | 4088 | 0.7196 | (0.699-0.740) |
| Waist Circumference ∗ | 4088 | 0.7310 | (0.711-0.751) |
| WHR | 4088 | 0.7130 | (0.693-0.733) |
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
