Diet Quality and its Association With Nonalcoholic Fatty Liver Disease and All-cause and Cause-specific Mortality

Eric R. Yoo; Donghee Kim; Luis M. Vazquez-Montesino; Jessica A. Escober; Andrew A. Li; Sean P. Tighe; Christopher T. Fernandes; George Cholankeril; Aijaz Ahmed


Liver International. 2020;40(4):815-824. 

In This Article


Study Population

We performed our analysis using the third National Health and Nutrition Examination Survey (NHANES), conducted from 1988 to 1994 in the United States. NHANES data incorporate a stratified, multistage, clustered probability sampling design to reach a representative sample of the civilian, non-institutionalized population.[26] Of adult participants (20- to 74-year old) in the third NHANES (n = 14 797), we excluded those with significant alcohol consumption (>21 drinks/wk in men and >14 drinks/wk in women), viral hepatitis (positive serum hepatitis B surface antigen or positive serum hepatitis C antibody), iron overload (transferrin saturation ≥50%) and ongoing pregnancy. Out of the remaining 13 176 participants who underwent an examination at a mobile examination centre, we then excluded those with missing data on hepatic ultrasonography, healthy eating index, body mass index (BMI), aminotransferases and mortality. The final study sample consisted of 10 858 adults (Figure 1). The methodology of the NHANES was approved by the institutional review board of the Centers for Disease Control and Prevention. All participants provided written informed consent before participation. Our study received an exemption by the institutional review board, as the data we used was completely de-identified.

Figure 1.

Flow diagram of participants in the study

Clinical and Laboratory Evaluations

The methods used for our study have been previously described.[11,27] Briefly, each participant underwent a detailed interview focusing on a wide array of demographic, lifestyle, and dietary information as well as an anthropometric assessment and comprehensive laboratory evaluation. Hypertension was defined as systolic blood pressure ≥140 mmHg or diastolic blood pressure ≥90 mmHg and/or previous use of antihypertensive medication. Diabetes was defined as physician-diagnosed diabetes, HbA1c ≥ 6.5%, fasting glucose ≥126 mg/dL, and or treatment with a hypoglycaemic agent or insulin. Iron overload was diagnosed using a transferrin level of ≥50%. Current smokers were defined as those with ongoing smoking or those who had smoked at least 100 cigarettes during the past 5 years. Alcohol consumption was calculated using self-reported data on the amount and frequency of alcohol use, as previously described.[28] Physical activity was defined as 'sedentary' if subjects answered 'no' to all questions regarding engagement in any of the following activities over the last month: jogging/running, bicycling, swimming, aerobics, other dancing, calisthenics, garden/yard work, weight lifting or other sports.[29] Insulin resistance was evaluated using the homeostasis model assessment of insulin resistance (HOMA-IR).[30]

Definition of NAFLD

The method used for ultrasonography-diagnosed hepatic steatosis has been previously described.[11,27] Briefly, the NHANES III examination included ultrasonography of the gallbladder as a part of the assessment for digestive diseases in adults 20–74 years of age. Between 2009 and 2010, the archived gallbladder ultrasound video images were reviewed to assess fatty liver by three board-certified radiologists. According to the NHANES III procedures manual, the assessment of hepatic steatosis with ultrasound images ( was performed using the following five criteria: (a) parenchymal brightness, (b) liver to kidney contrast, (c) deep beam attenuation, (d) bright vessel walls and (e) gallbladder wall definition. The ultrasonographic findings based on these five criteria were reported as normal versus mild, moderate or severe hepatic steatosis. For our study, NAFLD was defined as any degree of steatosis – mild to severe – as diagnosed by ultrasonography and without significant alcohol consumption and absence of other causes of chronic liver disease, including viral hepatitis and haemochromatosis.

Dietary Quality

To measure how well American diets conform to the recommended healthful eating patterns, the US Department of Agriculture developed the HEI.[31] Details of the HEI utilized in the NHANES III dataset are available publicly and online ( The HEI score ranges from 0 to 100, which was calculated via summation of 10 equally weighted dietary components scored between 0 and 10, which were derived from a single 24-hour dietary recall. The HEI score was composed of 10 components: grains, vegetables, fruits, dairy products, meats, total fat, saturated fat, total cholesterol, total sodium and dietary variety. A score of 0 is assigned for zero servings and the maximum score is assigned when the recommended servings were consumed. Higher HEI scores represent a healthier and better-quality dietary pattern when compared to lower scores. The NHANES III data use the 1994–1996 version of the HEI.


As part of the study protocol, all participants over 20 years of age in the third NHANES were followed for mortality through 31 December 2015. For decedents, probabilistic matching was conducted with National Death Index records to determine mortality status by date of death. In addition, the cause of death was assessed. The third NHANES-Linked Mortality File uses the Underlying Cause of Death 113 (UCOD_113) code to recode all deaths according to the International Classification of Diseases, 9th Revision (ICD-9) for deaths before 1998 and to ICD-10 for those who died between 1999 and 2015. All-cause mortality and the following two cause-specific mortalities were assessed as cardiovascular disease (UCOD_113 55–64, 70) and malignancy (UCOD_113 19–43).

Statistical Analysis

Given the complex sample design employed by NHANES, we utilized appropriate sampling weights to reconstitute data on a population level for the entire United States.[32] Baseline characteristics were compared using the chi-square test for categorical variables and linear regression for continuous variables. To evaluate the interaction of the HEI and sex on NAFLD or all-cause mortality, two-way interactions between the HEI and sex were tested, and interaction terms with P < .05 were considered statistically significant. There were no significant interactions between the HEI and sex (test of interaction P > .1) on the NAFLD and all-cause mortality. Therefore, we treated the HEI and individual components of the HEI as sex-specific quartiles. In addition, the HEI and individual components of the HEI were standardized to a mean of 0 and a standard deviation (SD) of 1 based on the sex to facilitate comparisons between each index. Multivariable logistic regression models were created to identify independent associations between dietary quality and NAFLD after consideration for potential demographic and clinical confounders. Survival analysis, including all-cause and cause-specific mortality, utilized the Cox proportional hazards regression analysis. Multivariable Cox proportional model was used to investigate the independent association of dietary quality with all-cause mortality and cause-specific mortalities. Analyses were carried out using STATA 15.1 (StataCorp) applying Taylor series linearization.