Public Health Nutrition: 26(12), 2936–2944 doi:10.

1017/S1368980023001970

Dietary inflammatory potential is associated with higher odds

of hepatic steatosis in US adults: a cross-sectional study
Hu Yang1,2, Tengfei Zhang1, Wen Song3, Zhaohong Peng3, Yu Zhu1, Yong Huang1,
Xiude Li1, Zhuang Zhang1, Min Tang4 and Wanshui Yang1,*
1
Department of Nutrition, School of Public Health, Anhui Medical University, 81 Meishan Road, Hefei, Anhui 230032,
People’s Republic of China: 2Taizhou Central Hospital (Taizhou University Hospital), Taizhou, Zhejiang, People’s
Republic of China: 3Department of Interventional Radiology, the First Affiliated Hospital of Anhui Medical University,
Hefei, Anhui, People’s Republic of China: 4Department of Gastroenterology and Hepatology and Clinical Nutrition,
the Fourth Affiliated Hospital of Anhui Medical University, Hefei, Anhui, People’s Republic of China

Submitted 23 May 2022: Final revision received 1 August 2023: Accepted 5 September 2023: First published online 9 October 2023

Abstract
Objective: Inflammation plays a critical role in the progression of chronic liver
diseases, and diet can modulate inflammation. Whether an inflammatory dietary
pattern is associated with higher risk of hepatic steatosis or fibrosis remains
unclear. We aimed to investigate the associations between inflammatory dietary
pattern and the odds of hepatic steatosis and fibrosis.
Design: In this nationwide cross-sectional study, diet was measured using two 24-h
dietary recalls. Empirical dietary inflammatory pattern (EDIP) score was derived to
assess the inflammatory potential of usual diet, which has been validated to highly
predict inflammation markers in the study population. Controlled attenuation
parameter (CAP) and liver stiffness measurement (LSM) were derived from
FibroScan to define steatosis and fibrosis, respectively.
Setting: US National Health and Nutrition Examination Survey.
Participants: 4171 participants aged ≥18 years.
Results: A total of 1436 participants were diagnosed with S1 steatosis
(CAP ≥ 274 dB/m), 255 with advanced fibrosis (LSM ≥ 9·7 kPa). Compared with
those in the lowest tertile of EDIP-adherence scores, participants in the highest
tertile had 74 % higher odds of steatosis (OR: 1·74, 95 % CI (1·26, 2·41)). Such
positive association persisted among never drinkers, or participants who were free
of hepatitis B and/or C. Similarly, EDIP was positively associated with CAP in
multivariate linear model (P < 0·001). We found a non-significant association of
Keywords
EDIP score with advanced fibrosis or LSM (P = 0·837). Cross-sectional study
Conclusions: Our findings suggest that a diet score that is associated with Diet
inflammatory markers is associated with hepatic steatosis. Reducing or avoiding Inflammation
pro-inflammatory diets intake might be an attractive strategy for fatty liver disease Hepatic steatosis
prevention. Controlled attenuation parameter

Non-alcoholic fatty liver disease (NAFLD) imposes an pathogenesis(3). Circulating concentrations of inflammation
enormous burden on health care systems and affects markers, such as IL-4, IL-6, C-reactive protein (CRP) and
approximately 25 % of the population worldwide and tumour necrosis factor-α receptor 2 (TNFα-R2), have been
30 % of people in the USA(1). To date, due to the lack of shown to be associated with NAFLD in prior studies(1,4,5).
approved drug therapy, lifestyle modification to achieve Moreover, in previous studies, lifestyles including diets can
weight loss remains an optimal intervention for patients modulate inflammation(6–10). For instance, Mediterranean-
with NAFLD(1,2). Accumulating evidence indicates that type diets have anti-inflammatory properties and are
chronic inflammation contributes substantially to NAFLD effective in decreasing the risk of NAFLD and slowing its
progression(11). Thus, we hypothesised that higher inflam-
Hu Yang, Tengfei Zhang and Wen Song contributed equally as co-first authors matory potential of diet might be associated with increased
for this article. risk of hepatic steatosis or fibrosis.

*Corresponding author: Email wanshuiyang@[Link]

© The Author(s), 2023. Published by Cambridge University Press on behalf of The Nutrition Society. This is an Open Access article, distributed under
the terms of the Creative Commons Attribution licence ([Link] which permits unrestricted re-use,
distribution and reproduction, provided the original article is properly cited.
Inflammatory diet and hepatic steatosis 2937
Recently, we re-derived and validated an empirical Assessments of diet and empirical dietary
dietary inflammatory pattern (EDIP) in the US National inflammatory pattern score
Health and Nutrition Examination Survey (NHANES)(12), Dietary information was collected using two 24-h dietary
which is originally developed by Tabung et al. in three recalls by skilled investigators. We used multiple-pass
Harvard cohorts(13). EDIP is a hypothesis-driven a posteriori method to enhance complete and accurate data collection
dietary pattern and was derived by entering thirty-nine and decrease respondent burden(24). Dietary sampling
predefined food groups into the reduced rank regression weights were used to overcome the limitations including
(RRR) followed by stepwise linear regression, which was the dietary interview-specific nonresponse, day of the
highly predictive of concentration of two plasma inflamma- week for dietary recalls, unequal probability of selection
tion markers including CRP and leucocytes count. EDIP has and oversampling(24,25).
been suggested to be associated with higher risk of several The development and validation of EDIP scores have
chronic diseases including CVD(14), cancers(15–17) and type 2 been described previously(12). In short, thirty-nine pre-
diabetes(18). Moreover, we previously showed that EDIP is defined food groups(26) were entered into RRR model
positively associated with risk of total and cancer-specific followed by stepwise linear regression analysis to identify a
mortality(12), and hepatocellular carcinoma(16). However, to dietary pattern most predictive of two inflammation
our knowledge, there have been no epidemiological studies markers (i.e. CRP and leucocytes count). RRR can identify
regarding the association between EDIP and hepatic linear functions of predictors (i.e. food groups) that
steatosis and fibrosis to date, although few studies(19–22) simultaneously explain as much response variation of
have investigated hepatic steatosis in relation to a literature- inflammation markers as possible. The first factor (i.e. the
derived dietary inflammatory index (DII), which is an RRR dietary pattern) identified by RRR then underwent
a priori dietary pattern (i.e. its development is based on the further data reduction by stepwise linear regression to
peer-reviewed articles on the association between dietary identify the most important component food groups of
factors and inflammation). Given that DII is mainly nutrient- the RRR dietary pattern, with the RRR dietary pattern as
based (i.e. thirty-eight of its forty-five components are the dependent variable, the thirty-nine food groups as
nutrients), findings from DII studies could be difficult to be independent variables, and a significance level of P = 0·01
translated readily into public health practice. for entry into, and retention in the model. A total of twenty-
Herein, to add more evidence, we investigated the five food components were included in EDIP score (see
cross-sectional association between adherence to EDIP online Supplemental Table 1). We used the regression
and odds of hepatic steatosis and fibrosis in a US coefficients in the final stepwise linear regression model as
nationwide sample. weights to calculate the EDIP scores. Higher EDIP scores
(more positive) denote more pro-inflammatory potential
of diets, while lower (more negative) scores indicate
anti-inflammatory potential of diets.
Methods
In validation study of our prior publication(12), EDIP
have shown a high ability to predict inflammatory markers
Study population
(i.e. plasma high-sensitivity CRP and leucocytes count) in
This study used data from the 2017–2018 cycle of the US
NHANES 2015–2018.
NHANES, in which hepatic transient elastography (TE) was
performed for the first time in the survey. NHANES is a
continuous cross-sectional survey conducted in the USA by Assessments of covariates
the National Center for Health Statistics of the Centers for Information on demographic and lifestyle factors, including
Disease Control and Prevention. The survey aimed to age, sex, race/ethnicity, educational level, income, smok-
assess the health of a representative sample of about 5000 ing and physical activity, were collected by standardised
persons each year in the USA. Details of NHANES study questionnaires during household interview. Information
design, study protocol and data collection approaches have on alcohol intake, body weight and height was obtained
previously been reported(23). from participants who received physical examinations in
The flow chart of how we selected the study population the NHANES Mobile Examination Center. Individuals who
was shown in Supplemental Fig. 1. Individuals aged 18 had smoked at least 100 cigarettes in life were defined as
years or older were included. We excluded participants if ever smokers, and never smokers were defined as those
they: (i) had missing data on diet (n 873); (ii) had who did not have cigarettes consumption before the time of
implausible energy intake(16) (<600 or >3500 kcal/d for the interview. BMI was calculated as weight in kilograms
women; <800 or >4200 kcal/d for men, n 221); (iii) did not divided by the square of the height in meters (kg/m2). The
receive TE detection (n 264) or (iv) had invalid TE detection ratio of family income to poverty that accounts for family
results (n 327). A total of 4171 participants were finally size and annual inflation and was calculated by dividing
included in the analysis. family income by the poverty thresholds. The poverty
2938 H Yang et al.
thresholds were defined as the dollar amounts set by the in the models were as follows: Model 1 was adjusted for age
U.S. government to indicate the least amount of income a (18–39, 40–59 and ≥60). Model 2 was further adjusted
person or family needs to meet their basic needs, which for sex (male and female), smoking status (never smokers
were used to estimate the population’s income and poverty and ever smokers), race/ethnicity (non-Hispanic white,
levels and related information. Physical activity was non-Hispanic black and other races), education (less than
assessed by Global Physical Activity Questionnaire, which high school, high school diploma and more than high
has been shown to have good reliability and validity in school), family income to poverty ratio (<1·30, 1·30–3·49
multiple populations(27). Individuals who performed <8·3, and ≥3·50), marital status (never married, married and
8·3 to 16·7 and >16·7 metabolic equivalents of tasks hours widowed/divorced), physical activity (low level, moderate
of physical activity/week were classified as low, moderate level and high level), total energy intake (tertiles),
and high levels according to the 2018 Physical Activity HBV (positive and negative) and HCV (positive and
Guidelines for Americans(28). Hepatitis B virus (HBV) negative) infection. Of note, we did not adjust for alcohol
infection was defined as positive hepatitis B surface antigen consumption in our main analyses because wine and beer
(HBsAg), while both positive hepatitis C antibody and are components in EDIP. For covariates with missing
RNA indicated hepatitis C virus (HCV) infection. Diabetes values, a separate missing indicator variable was created
was diagnosed if there were: (i) a self-reported history of and included in the models. We presented OR by tertile
diabetes; (ii) a fasting plasma glucose level of more than categories and per 1-SD increase of EDIP scores. Linear
126 mg/dl; (iii) a random glucose level of more than 200 trends across increasing categories of EDIP scores were
mg/dl and (iv) a HbA1c level of more than 6·5 % for tested by entering EDIP scores as a continuous variable in
participants. the models, and P values for trend were calculated using a
Wald test. We also used restricted cubic spline to identify
the dose–response relationship between EDIP and hepatic
Ascertainments of hepatic steatosis and fibrosis
steatosis.
Vibration-controlled TE using the FibroScan® model
Allowing for the potential intermediate role of BMI and
502 V2 Touch equipped with a medium (M) or extra-large
diabetes in the association of EDIP and chronic liver
(XL) wand (probe) was performed by technicians after
disease(16), we did not adjust for BMI and diabetes in the
a 2-d training program with an expert technician. Hepatic
main analyses but additionally adjusted for these two
steatosis and fibrosis were measured using controlled
factors in the sensitivity analyses. To reduce measurement
attenuation parameter (CAP) and liver stiffness measurement
error and reflect dietary composition, we adjusted the EDIP
(LSM), respectively. In accordance with previous studies(29–31),
scores for total energy intake using the nutrient residual
we used cut-off values of median CAP ≥ 274 for S1 steatosis,
method(32). Considering that HBV and HCV infections are
CAP ≥ 290 for S2 steatosis, LSM ≥ 8·2 kPa for significant
important risk factors for liver diseases, we repeated
fibrosis and LSM ≥ 9·7 kPa for advanced fibrosis. TE
analysis within individuals who are free of hepatitis B
examinations were considered as reliable only when more
and/or C. Likewise, we investigated EDIP after removing
than 10 LSMs were obtained after a fasting time no less than
their alcohol components (i.e. beer and wine) in relation to
3 h, with an interquartile range (IQR) to median ratio < 30 %.
the prevalence of hepatic steatosis, with further adjust-
By comparing CAP measurement for the detection of
ments for alcohol drinking status (never, low to moderate
steatosis against biopsy, the area under the receiver
and heavy drinking), although beer and wine are included
operating characteristic curves was 0·87 (95 % CI (0·82,
in the construct of EDIP. In addition, considering
0·92)) with a sensitivity and specificity of both 90 % for
the possible etiological differences, we investigated the
S ≥ S1 among patients with NAFLD(29). Similarly, when
association of EDIP with non-alcoholic fatty liver and other
using LSM to define patients with fibrosis, the area under
steatosis separately. The non-alcoholic fatty liver diseases
the receiver operating characteristic curves was 0·80 (95 %
were defined if individuals: (i) were detected as steatosis
CI (0·75, 0·84)) for advanced fibrosis (F ≥ F3), with the
through TE test; (ii) did not have significant alcohol
corresponding sensitivity of 71 % and specificity of 75 %(29).
consumption (>2 drinks/d for women and >3 drinks/d for
men); (iii) were free of hepatitis B and/or C infection and
Statistical analysis (iv) did not take steatogenic medications (i.e. amiodarone,
The prevalence of hepatic steatosis and fibrosis was valproate, methotrexate, tamoxifen and corticosteroid) for
standardised based on 2020 US population(31). Multiple at least 3 months or more before study enrollment(33,34).
linear regression was performed to evaluate the percentage Previous studies suggested that several factors including
change and 95 % CI for the associations of the EDIP scores age, sex, race, smoking status, drinking status, marital
with continuous CAP and LSM. Both CAP and LSM were status, BMI and diabetes could modify the associations
natural logarithms transformed in the models given the between inflammatory dietary pattern and chronic liver
deviation from normal distribution. We used multiple diseases(16,35–37). Therefore, we stratified analyses accord-
logistic regression to estimate the OR and 95 % CI for EDIP ing to these factors and tested the potential interactions. Wald
scores in relation to S1 and S2 steatosis. Covariates adjusted test was used to check whether the cross-product terms
Inflammatory diet and hepatic steatosis 2939
Table 1 Age-adjusted characteristics of participants according to tertiles of EDIP scores in NHANES 2017–2018*

EDIP scores

Characteristics Tertile 1 (n 1390) Tertile 2 (n 1391) Tertile 3 (n 1390)

Mean IQR Mean IQR Mean IQR
EDIP scores† −0·21 −0·31, −0·15 −0·03 −0·07, 0·00 0·09 0·06, 0·15
Mean SD Mean SD Mean SD
Age (years)† 48·1 17·7 48·3 18·6 51·8 18·2
BMI (kg/m2) 28·8 6·8 29·6 7·1 29·9 7·2
METS-h/week|| 78·1 124·9 63·7 103·7 78·8 124·7
Energy (kcal/d) 2295 708 1858 640 1804 686
CAP (dB/m) 259·2 61·8 263·7 61·7 268·2 64·3
LSM (kPa) 5·6 4·1 5·6 3·8 6·0 5·3
n % n % n %
Male (%) 749 53·9 618 44·4 666 47·9
Drinking status (%)
Never 143 10·3 160 11·5 154 11·1
Low to moderate‡ 1049 75·5 1085 78·0 1090 78·4
Heavy 198 14·1 146 10·5 146 10·5
Ever smokers (%)§ 489 35·2 522 37·5 674 48·5
Race (%)
Non-Hispanic white 407 29·3 408 29·3 645 46·4
Non-Hispanic black 363 26·1 374 26·9 214 15·4
Other 620 44·6 609 43·7 531 38·2
Education (%)
Less than high school 202 14·5 256 18·4 317 22·8
High school diploma 278 20·0 345 24·8 413 29·7
More than high school 910 65·5 790 56·8 660 47·5
Family income to poverty ratio (%)
<1·30 318 22·9 417 30·0 453 32·6
1·30–3·49 539 38·8 551 39·6 580 41·7
≥3·50 533 38·3 423 30·4 357 25·7
Marital status (%)
Never married 374 26·9 391 28·1 370 26·6
Married 774 55·7 762 54·8 733 52·7
Widowed or divorced 242 17·4 238 17·1 287 20·7
Physical activity (%)
Low level 434 31·2 505 36·3 510 36·7
Moderate level 145 10·4 145 10·4 114 8·2
High level 811 58·4 741 53·3 766 55·1
Diabetes (%) 232 16·7 273 19·6 278 20·0
HBV infection (%) 10 0·75 2 0·15 6 0·41
HCV infection (%) 14 1·04 22 1·58 41 2·96

EDIP, empirical dietary inflammatory pattern; HBV, hepatitis B Virus; HCV, hepatitis C Virus; NHANES, US National Health and Nutrition Examination Survey.
*Continuous variables were presented as means (SD) if they were normally distributed, otherwise median (IQR) estimate was used; All the variables were standardised to the age
distribution of the study population except for EDIP scores and age; Notably, the summing proportions for some categories are not 100 % because of missing values or rounding.
†Value was not age adjusted.
‡Individuals who never drank in the last year but reported a history of alcohol drinking previously were also assigned in this category.
§Individuals who had smoked at least 100 cigarettes in life.
||Individuals who performed <8·3, 8·3 to 16·7, >16·7 METS-hours of physical activity/week were classified as low, medium, and high levels.

between these variables and exposures were statistically 5·6 % (255 cases) for advanced fibrosis. The median (IQR)
significant. We used the Bonferroni correction to define of EDIP scores for the total population was −0·03 (IQR:
the statistical significance as P < 0·0031 (0·05/(2 outcomes x 8 −0·15 to 0·06), ranged from a median of −0·21 (IQR: −0·31
groups)) for subgroup analysis to account for multiple to −0·15) in the lowest tertile to 0·09 (IQR: 0·06 to 0·15) in
comparisons. All statistical tests were two-tailed and the highest tertile. Participants with higher EDIP score were
performed using SAS version 9.4 (SAS Institute). older, had higher BMI, were more likely to be ever smokers
and non-Hispanic whites, were less educated, were more
likely to be married, had lower ratio of family income to
Results poverty, were less physically active and were more likely to
have a history of diabetes and hepatitis C (Table 1).
Characteristics of participants
A total of 4171 participants aged 18 years or older (mean
age, 49·4 years; SD, 18·3 years) were included in our study. EDIP, hepatic steatosis and fibrosis
The age-standardised prevalence was 42·5 % (1806 cases) After adjusting for age, sex, and other covariates (Table 2),
for S1 steatosis, 33·8 % (1436 cases) for S2 steatosis and EDIP score was positively associated with CAP, with the
2940 H Yang et al.
Table 2 Percentage change (%) and 95 % CI for the associations of the empirical dietary inflammatory pattern with controlled attenuation
parameter (CAP) and liver stiffness measurement (LSM) in NHANES (2017–2018)*

Tertile 2 Tertile 3 Per 1-SD

Tertile 1 % 95 % CI % 95 % CI % 95 % CI Pvalue†
CAP (dB/m)
No. of participants 1390 1391 1390
Model 1 Reference 2·6 −0·2, 5·6 5·7 2·2, 9·2 2·5 1·2, 3·8 <0·001
Model 2 Reference 4·2 1·6, 6·9 7·4 4·1, 10·9 3·1 2·0, 4·3 <0·001
LSM (kPa)
Model 1 Reference 1·7 −4·9, 8·7 5·0 −0·4, 10·7 0·6 −2·1, 3·4 0·640
Model 2 Reference 3·6 −2·8, 10·5 5·4 −0·4, 11·5 0·3 −2·9, 3·6 0·837

EDIP, empirical dietary inflammatory pattern; HBV, hepatitis B Virus; HCV, hepatitis C Virus; NHANES, US National Health and Nutrition Examination Survey.
*Model 1 was adjusted for age; Model 2 was further adjusted for sex, smoking status, race, education, family income to poverty ratio, marital status, physical activity, total
energy, HBV, and HCV.
†Linear trends across increasing categories of EDIP scores were tested by entering EDIP scores as a continuous variable into the models, and P values for trend were
calculated using a Wald test.

Table 3 Odds ratios and 95 % confidence intervals for hepatic steatosis according to tertiles of EDIP scores in NHANES 2017–2018*

Tertile 2 Tertile 3 Per 1-SD

Tertile 1 OR 95 % CI OR 95 % CI OR 95 % CI Ptrend‡
Steatosis† (≥S1)
No. of cases 565 596 645
Model 1 Reference 1·21 0·95, 1·55 1·47 1·09, 1·98 1·23 1·10, 1·38 <0·001
Model 2 Reference 1·36 1·09, 1·69 1·74 1·26, 2·41 1·33 1·16, 1·53 <0·001
Steatosis (≥S2)
No. of cases 438 468 530
Model 1 Reference 1·20 0·91, 1·57 1·52 1·16, 1·99 1·24 1·10, 1·39 <0·001
Model 2 Reference 1·33 1·04, 1·69 1·74 1·29, 2·34 1·32 1·14, 1·51 <0·001

EDIP, empirical dietary inflammatory pattern; HBV, hepatitis B Virus; HCV, hepatitis C Virus; NHANES, US National Health and Nutrition Examination Survey.
*Model 1 was adjusted for age; Model 2 was further adjusted for sex, smoking status, race, education, family income to poverty ratio, marital status, physical activity, total
energy, HBV, and HCV.
†CAP values ≥ 274 dB/m and 290 dB/m were considered indicative of S1 and S2 steatosis, respectively.
‡Linear trends across increasing categories of EDIP scores were tested by entering EDIP scores as a continuous variable into the models, and P values for trend were
calculated using a Wald test.

percentage difference of 7·4 % (95 % CI (4·1, 10·9), (LSM ≥ 8·2 kPa), advanced fibrosis (LSM ≥ 9·7 kPa) or
P < 0·001) in participants with the highest tertile of EDIP cirrhosis (LSM ≥ 13·6 kPa, data not shown).
scores, compared with those in the lowest tertile. This
positive association was partly attenuated but remained
statistically significant with further adjustments for BMI Sensitivity and subgroup analyses
and diabetes (percentage difference: 3·5 %, 95 % CI (1·7, In sensitivity analysis, when repeated analysis using the
5·4), P < 0·001). We found a non-significant association energy-adjusted EDIP scores, the results were similar to
between EDIP score and LSM (percentage difference: those in the main analysis (see online Supplemental
5·4 %, 95 % CI (−0·4 , 11·5), P = 0·837). Table 2). Likewise, the results were not essentially changed
Similarly, participants with higher EDIP score had higher among individuals who were free of hepatitis B and/or C
odds of hepatic steatosis with OR (comparing extreme (Fig. 2). After removing alcohol components in EDIP and
tertile) of 1·74 (95 % CI (1·26, 2·41), Ptrend < 0·001, Table 3). further adjusted for alcohol intake in the models, the results
When we additionally controlling BMI and diabetes, the were essentially unchanged (see online Supplemental
magnitude of the positive association between EDIP and Table 3). When examining non-alcoholic fatty liver
steatosis was partly attenuated (OR: 1·35, 95 % CI (1·04, and other steatosis separately, we did not find the
1·74), Ptrend = 0·001). Similar association was observed with significant heterogeneity on the associations of EDIP with
the cut-off value of median CAP of no less than 290 dB/m odds of non-alcoholic fatty liver and other steatosis
(OR: 1·34, 95 % CI (1·06, 1·70), Ptrend = 0·004). Restricted (Pheterogeneity = 0·477) (see online Supplemental Table 4).
cubic spline analysis did not support the non-linear In subgroup analysis (Fig. 2), there was no differential
association between EDIP and steatosis (P for linearity = association between EDIP and odds of hepatic steatosis
0·002, Fig. 1). We did not find any significant association when stratified by age, sex, race/ethnicity, smoking
between EDIP score and odds of significant fibrosis status, drinking status, marital status, BMI or diabetes
Inflammatory diet and hepatic steatosis 2941

Fig. 1 Association between empirical dietary inflammatory pattern scores and hepatic steatosis (≥S1) in NHANES (2017–2018)*.
*Model was adjusted for age, sex, smoking status, race, education, family income to poverty ratio, marital status, physical activity, total
energy, HBV, HCV, BMI and diabetes except for variables examined in the figure. Notably, the restricted multivariable cubic spline
analysis showed significantly linear association between empirical dietary inflammatory pattern scores and hepatic steatosis (≥S1)
(P for linearity = 0·002 and P for non-linearity = 0·157). Reference levels were set to the median EDIP value. Solid lines indicate OR,
and dashed lines depict 95 % CI. EDIP, empirical dietary inflammatory pattern; HBV, hepatitis B Virus; HCV, hepatitis C Virus;
NHANES, US National Health and Nutrition Examination Survey

(all the P values for interaction were greater than association between DII and fatty liver diseases or their
Bonferroni-corrected statistical significance of 0·0031). parameters, which all used cross-sectional design(19,21,22).
The EDIP and DII both evaluate the inflammation potential
of diet, while the two dietary patterns differ in several
Discussion aspects. The EDIP is a hypothesis-driven a posteriori
pattern (i.e. its development is based on RRR to identify
In this nationwide cross-sectional study among US adults, food groups predictive of inflammation biomarkers) and is
we examined associations between EDIP and odds of based exclusively on food groups. The DII is an a priori
hepatic steatosis and fibrosis. We found that persons with pattern (i.e. its development is based on the 1943 peer-
higher EDIP scores (i.e. consuming a pro-inflammatory reviewed articles on the association between dietary factors
diet) had a higher prevalence of hepatic steatosis. This and inflammation) and is mainly nutrient-based. Different
positive association remained among individuals who were from our study, previous DII studies on fatty liver diseases
free of hepatitis B and/or C and persisted regardless of used different approaches for outcome ascertainment,
alcohol drinking status. EDIP seemed not to be associated including Fatty Liver Index(19,21,22), the aspartate transami-
with fibrosis, as indicated by LSM. nase to alanine transaminase ratio(19) or fibrosis-4 score(19),
Previous studies have reported that several nutrients with the exception of 2 studies(19,43). In the current study,
and foods, such as fructose(38,39), soft drinks(40) and red we were able to derive CAP and LSM through TE
meat(40), have been associated with high risk for NAFLD. (FibroScan®) to define hepatic steatosis and fibrosis with
However, diets are complex combinations of nutrients and higher sensitivity and specificity(29,44). However, our study
foods, which may interact mutually(41,42). Thus, dietary together with previous DII studies(19,21,22,43) consistently
patterns considering multiple dietary factors may provide a support that pro-inflammatory diets are associated with
more comprehensive assessment of diet and may thus be higher risk of fatty liver diseases.
more predictive of diet–disease associations compared In line with our study and previous DII studies(19,21,22,43),
with the approach of using single nutrients or foods. a randomised controlled trial(20) among younger adults
This is the first observational study to investigate the with obesity showed the effectiveness of an energy-
association of EDIP score with hepatic steatosis and fibrosis reduced anti-inflammatory diet with significant improve-
among the US adults, though few studies have assessed the ment of liver parameters, including Fatty Liver Index, liver
2942 H Yang et al.
Subgroup OR of Steatosis OR and 95% CI P for Interaction

Age, years 0·833

< 60 (n 2675) 1·21(1·06-1·37)
≥ 60 (n 1496) 1·30(0·96-1·74)
Sex 0·453
male (n 2038) 1·19(0·99-1·42)
female (n 2133) 1·17(1·00-1·38)
Race/ethnicity 0·238
non-Hispanic white (n 1460) 1·30(1·00-1·69)
other (n 2711) 1·07(0·96-1·21)
Smoking status 0·574
never smokers (n 2485) 1·32(1·09-1·61)
ever smokers (n 1686) 1·12(0·98-1·27)
Drinking status 0·196
never drinkers (n 444) 1·65(0·97-2·81)
ever drinkers (n 3616) 1·16(1·05-1·29)
Marital status 0·514
Married (n 2154) 1·14(0·99-1·32)
Other (n 1806) 1·24(1·04-1·49)
BMI 0·433
< 30 (n 2481) 1·15(1·01-1·31)
≥ 30 (n 1666) 1·28(0·99-1·66)
Diabetes 0·379
no (n 3384) 1·15(1·04-1·27)
yes (n 786) 1·70(1·20-2·39)
HBV and HCV infection
HBV negative (n 3976) 1·18(1·06-1·32)
HCV negative (n 3850) 1·19(1·07-1·32)
Both negative (n 3832) 1·19(1·07-1·32)

0·5 1·0 1·5 2·0 2·5 3·0

Fig. 2 Subgroup analysis on the association of EDIP scores (per 1-SD increase) with hepatic steatosis (≥S1) in NHANES
(2017–2018)*. *Model was adjusted for age, sex, smoking status, race, education, family income-poverty ratio, marital status,
physical activity, total energy, HBV, HCV, BMI and diabetes except for variables examined in the figure. EDIP, empirical dietary
inflammatory pattern; HBV, hepatitis B Virus; HCV, hepatitis C Virus; NHANES, US National Health and Nutrition Examination Survey

fat score and fibrosis-4. Consistently, in two Harvard resistance(24,48,49). Meantime, the elevated levels of inflam-
cohorts, the Nurses’ Health Study and the Health matory markers (i.e. CRP, IL-6 and TNFα) are observed
Professionals Follow-up Study of 119 316 participants with among individuals with NAFLD(50,51). Thus, one possible
142 incident hepatocellular carcinoma cases, we found a mechanism is that diet can modulate inflammation and
positive association between EDIP score and risk of mediate insulin resistance, which in turn leads to hepatic
hepatocellular carcinoma(16). These findings further sup- steatosis. However, we did not find any significant
port that a diet score that is associated with inflammatory association between EDIP-adherence score and the like-
markers is associated with hepatic steatosis. lihood of fibrosis (data not shown). One possible reason is
The association between greater adherence to pro- that coffee is included as a component in EDIP (see online
inflammatory diet and higher odds of hepatic steatosis has Supplemental Table 1), whereas coffee could induce UDP
its biological plausibility. It is accepted that insulin resistance glucuronosyltransferases, which may contribute to the
is a crucial pathophysiological factor in the development of protective, antioxidant effects in the progression of hepatic
NAFLD(45,46), since the decreased insulin sensitivity of fibrosis(52–54). Alternatively, the lack of an association
adipocyte causes an increased hepatic-free fatty acid flux between EDIP and fibrosis may be due to the insufficient
creating favourable conditions for the progression of hepatic power caused by limited cases of fibrosis in the present study.
steatosis(46,47). Moreover, inflammation cytokines, such as IL Strengths of our study include the use of validated food-
and TNFα, may disrupt insulin action and mediate insulin based EDIP scores, a large nationally representative sample
Inflammatory diet and hepatic steatosis 2943
of US adults and valid TE detection to measure hepatic content: All authors. Statistical analysis: H.Y. and Y.Z.
steatosis and fibrosis. However, our study has several Administrative, technical or material support: W.Y. Study
limitations. First, dietary information was measured by supervision: W.Y.
24-h recalls, which may limit the ability to capture habitual
diets of individuals. To overcome this limitation, we used
several methods, such as multiple-pass method and dietary Ethics of human subject participation
sampling weights(24,25). We also used energy-adjusted EDIP
in the models(32) and yielded similar results. Second, we are This study was conducted according to the guidelines laid
unable to completely rule out residual or unmeasured down in the Declaration of Helsinki, and all procedures
confounders (e.g. the use of anti-inflammation drugs). involving research study participants were approved by the
Third, the cross-sectional design in the current study does (NCHS Research Ethics Review Board; Protocol #2011-17;
not allow the determination of causality. Protocol #2018-01). Written informed consent was
In conclusion, findings from our study indicate that a obtained from all subjects.
diet score that is associated with inflammatory markers is
associated with hepatic steatosis. Interventions to reduce
the adverse effect of pro-inflammatory diet may reduce the Supplementary material
likelihood of hepatic steatosis among US adults. However,
our results should be interpreted with caution, given For supplementary material accompanying this paper visit
the measurement of diet using 24-h recalls and the cross- [Link]
sectional design in the current study. More prospective
cohort studies and clinical trials are needed to validate our
findings. References

1. Rinella ME (2015) Nonalcoholic fatty liver disease:

a systematic review. JAMA 313, 2263–2273.
2. Hallsworth K & Adams LA (2019) Lifestyle modification in
Acknowledgements NAFLD/NASH: facts and figures. JHEP Rep 1, 468–479.
3. Friedman SL, Neuschwander-Tetri BA, Rinella M et al. (2018)
Mechanisms of NAFLD development and therapeutic
We thank Meiling Li and Xueke Zeng for their help with the strategies. Nat Med 24, 908–922.
final English language version. 4. Cobbina E & Akhlaghi F (2017) Non-alcoholic fatty liver
disease (NAFLD) – pathogenesis, classification, and effect on
drug metabolizing enzymes and transporters. Drug Metab
Rev 49, 197–211.
Financial support 5. Duarte N, Coelho IC, Patarrao RS et al. (2015) How
inflammation impinges on NAFLD: a role for Kupffer cells.
This work was supported by the National Natural Science Biomed Res Int 2015, 984578.
6. Dawson DR 3rd, Branch-Mays G, Gonzalez OA et al. (2014)
Foundation of China (WY, grant number 82073651); Anhui
Dietary modulation of the inflammatory cascade.
Provincial Natural Science Foundation (WY, grant number Periodontol 2000 64, 161–197.
2008085MH262), (ZZ, grant number 2108085QH357); 7. Garcia-Arellano A, Ramallal R, Ruiz-Canela M et al. (2015)
Anhui Provincial Education Department (WY, grant Dietary inflammatory index and incidence of cardiovascular
disease in the PREDIMED study. Nutrients 7, 4124–4138.
number gxyqZD2021099) and Anhui Medical University 8. Due A, Toubro S, Stender S et al. (2005) The effect of diets
(WY, grant numbers 2021xkjT007, XJ201935), (MT, grant high in protein or carbohydrate on inflammatory markers in
numbers 2020lcxk033, 2019xkj161). overweight subjects. Diabetes Obes Metab 7, 223–229.
9. Asgharpour A, Cazanave SC, Pacana T et al. (2016) A diet-
induced animal model of non-alcoholic fatty liver disease
and hepatocellular cancer. J Hepatol 65, 579–588.
Conflicts of interest 10. Kerr J, Anderson C & Lippman SM (2017) Physical activity,
sedentary behaviour, diet, and cancer: an update and
There are no conflicts of interest. emerging new evidence. Lancet Oncol 18, e457–e471.
11. Ryan MC, Itsiopoulos C, Thodis T et al. (2013) The
Mediterranean diet improves hepatic steatosis and insulin
sensitivity in individuals with non-alcoholic fatty liver
Authorship disease. J Hepatol 59, 138–143.
12. Li X, Chen B, Zhang J et al. (2022) Association of dietary
inflammatory potential with risk of overall and cause-specific
H.Y., T.Z. and W.S. had full access to all the data in the mortality. Br J Nutr 127, 1878–1887.
study and take responsibility for the integrity of the data 13. Tabung FK, Smith-Warner SA, Chavarro JE et al. (2016)
and the accuracy of the data analysis. Study concept and Development and validation of an empirical index of dietary
inflammatory potential. J Nutr 146, 1560–1570.
design: W.Y. Acquisition, analysis, or interpretation of
14. Li J, Lee DH, Hu J et al. (2020) Dietary inflammatory potential
data: All authors. Drafting of the manuscript: H.Y. and T.Z. and risk of cardiovascular disease among men and women in
Critical revision of the manuscript for important intellectual the U.S. J Am Coll Cardiol 76, 2181–2193.
2944 H Yang et al.
15. Tabung FK, Liu L, Wang W et al. (2018) Association of dietary are associated with low risk of NAFLD among the US
inflammatory potential with colorectal cancer risk in men and population. Hepatology 75, 1491–1506.
women. JAMA Oncol 4, 366–373. 34. Kim D, Konyn P, Cholankeril G et al. (2022) Physical
16. Yang W, Sui J, Zhao L et al. (2021) Association of activity is associated with nonalcoholic fatty liver disease and
inflammatory and insulinemic potential of diet and lifestyle significant fibrosis measured by fibroscan. Clin Gastroenterol
with risk of hepatocellular carcinoma. Cancer Epidemiol Hepatol 20, e1438–e1455.
Biomarkers Prev 30, 789–796. 35. Long L, Liu X, Petrick J et al. (2023) Dietary inflammatory
17. Tabung FK, Huang T, Giovannucci EL et al. (2017) and insulinemic potential, risk of hepatocellular carcinoma,
The inflammatory potential of diet and ovarian cancer risk: and chronic liver disease mortality. JNCI Cancer Spectr 7,
results from two prospective cohort studies. Br J Cancer 117, pkad023.
907–911. 36. Shivappa N, Hebert JR, Polesel J et al. (2016) Inflammatory
18. Lee DH, Li J, Li Y et al. (2020) Dietary inflammatory and potential of diet and risk for hepatocellular cancer in a case-
insulinemic potential and risk of type 2 diabetes: results from control study from Italy. Br J Nutr 115, 324–331.
three prospective U.S. cohort studies. Diabetes Care 43, 37. Zhang S, Meng G, Zhang Q et al. (2022) Inflammatory
2675–2683. potential of diet and risk of nonalcoholic fatty liver disease:
19. Ramirez-Velez R, Garcia-Hermoso A, Izquierdo M et al. a prospective cohort study. Eur J Clin Nutr 76, 1125–1132.
(2022) The dietary inflammatory index and hepatic health in 38. Noakes TD & Windt J (2017) Evidence that supports the
the US adult population. J Hum Nutr Diet 35, 968–979. prescription of low-carbohydrate high-fat diets: a narrative
20. Kendel Jovanovic G, Mrakovcic-Sutic I, Pavicic Zezelj S et al. review. Br J Sports Med 51, 133–139.
(2021) Metabolic and hepatic effects of energy-reduced 39. Vos MB & Lavine JE (2013) Dietary fructose in nonalcoholic
anti-inflammatory diet in younger adults with obesity. Can J fatty liver disease. Hepatology 57, 2525–2531.
Gastroenterol Hepatol 2021, 6649142. 40. Zelber-Sagi S, Nitzan-Kaluski D, Goldsmith R et al. (2007)
21. Cantero I, Abete I, Babio N et al. (2018) Dietary inflammatory Long term nutritional intake and the risk for non-alcoholic
index and liver status in subjects with different adiposity fatty liver disease (NAFLD): a population based study.
levels within the PREDIMED trial. Clin Nutr 37, 1736–1743. J Hepatol 47, 711–717.
22. Mazidi M, Shivappa N, Wirth MD et al. (2019) Diet with 41. Hu FB (2002) Dietary pattern analysis: a new direction in
greater inflammatory potential is associated with higher nutritional epidemiology. Curr Opin Lipidol 13, 3–9.
prevalence of fatty liver among US adults. Eur J Clin Nutr 73, 42. Tapsell LC, Neale EP, Satija A et al. (2016) Foods, nutrients,
1653–1656. and dietary patterns: interconnections and implications for
23. Centers for Disease Control and Prevention (2020) National dietary guidelines. Adv Nutr 7, 445–454.
Health and Nutrition Examination Survey (NHANES) About 43. Vahid F, Shivappa N, Hekmatdoost A et al. (2018)
the National Health and Nutrition Examination Survey. Association of pro-inflammatory dietary intake and
[Link] non-alcoholic fatty liver disease: findings from Iranian
(accessed September 2021). case-control study. Int J Vitam Nutr Res 88, 144–150.
24. Ahluwalia N, Dwyer J, Terry A et al. (2016) Update on 44. Chang TY, Chang SH, Lin YH et al. (2021) Utility of
NHANES dietary data: focus on collection, release, analytical quantitative ultrasound in community screening for hepatic
considerations, and uses to inform public policy. Adv Nutr 7, steatosis. Ultrasonics 111, 106329.
121–134. 45. Marchesini G & Forlani G (2002) NASH: from liver
25. Tooze JA, Midthune D, Dodd KW et al. (2006) A new diseases to metabolic disorders and back to clinical
statistical method for estimating the usual intake of hepatology. Hepatology 35, 497–499.
episodically consumed foods with application to their 46. Petta S, Gastaldelli A, Rebelos E et al. (2016)
distribution. J Am Diet Assoc 106, 1575–1587. Pathophysiology of non alcoholic fatty liver disease. Int J
26. Hu FB, Rimm E, Smith-Warner SA et al. (1999) Mol Sci 17, 2082.
Reproducibility and validity of dietary patterns assessed 47. Tilg H & Moschen AR (2008) Insulin resistance, inflamma-
with a food-frequency questionnaire. Am J Clin Nutr 69, tion, and non-alcoholic fatty liver disease. Trends Endocrinol
243–249. Metab 19, 371–379.
27. Bull FC, Maslin TS & Armstrong T (2009) Global physical 48. Taniguchi CM, Emanuelli B & Kahn CR (2006) Critical nodes
activity questionnaire (GPAQ): nine country reliability and in signalling pathways: insights into insulin action. Nat Rev
validity study. J Phys Act Health 6, 790–804. Mol Cell Biol 7, 85–96.
28. US Department of Health and Human Services Physical 49. White MF (2002) IRS proteins and the common
Activity Guidelines for Americans (2018). [Link] path to diabetes. Am J Physiol Endocrinol Metab 283,
paguidelines/second-edition/pdf/Physical_Activity_Guidelines_ E413–422.
2nd_edition.pdf (accessed September 2021). 50. Lizardi-Cervera J, Chavez-Tapia NC, Perez-Bautista O et al.
29. Eddowes PJ, Sasso M, Allison M et al. (2019) Accuracy of (2007) Association among C-reactive protein, Fatty
fibroscan controlled attenuation parameter and liver stiffness liver disease, and cardiovascular risk. Dig Dis Sci 52,
measurement in assessing steatosis and fibrosis in patients 2375–2379.
with nonalcoholic fatty liver disease. Gastroenterology 156, 51. Utzschneider KM & Kahn SE (2006) Review: the role of
1717–1730. insulin resistance in nonalcoholic fatty liver disease. J Clin
30. Ciardullo S, Monti T & Perseghin G (2021) High prevalence of Endocrinol Metab 91, 4753–4761.
advanced liver fibrosis assessed by transient elastography 52. Molloy JW, Calcagno CJ, Williams CD et al. (2012)
among U.S. adults with type 2 diabetes. Diabetes Care 44, Association of coffee and caffeine consumption with fatty
519–525. liver disease, nonalcoholic steatohepatitis, and degree of
31. Zhu Y, Peng Z, Lu Y et al. (2022) Higher dietary insulinaemic hepatic fibrosis. Hepatology 55, 429–436.
potential is associated with increased risk of liver steatosis 53. Kalthoff S, Ehmer U, Freiberg N et al. (2010) Coffee
and fibrosis. Liver Int 42, 69–79. induces expression of glucuronosyltransferases by the aryl
32. Willett WC, Howe GR & Kushi LH (1997) Adjustment for total hydrocarbon receptor and Nrf2 in liver and stomach.
energy intake in epidemiologic studies. Am J Clin Nutr 65, Gastroenterology 139, 1699–1710.
1220S–1228S. 54. Attar BM & Van Thiel DH (2013) Current concepts and
33. Vilar-Gomez E, Nephew LD, Vuppalanchi R et al. (2022) management approaches in nonalcoholic fatty liver disease.
High-Quality diet, physical activity, and college education ScientificWorldJournal 2013, 481893.