Abstract
Background
Mechanistic studies and short-term randomized trials suggest higher intakes of dietary flavonoids may protect against nonalcoholic fatty liver disease (NAFLD).
Objectives
We aimed to perform the first population-based study with long-term follow-up on flavonoid consumption, incident NAFLD, and validated NAFLD biomarkers.
Methods
In a prospective study, we assessed the associations between flavonoid intake based on ≥2 24-h dietary assessments and NAFLD risk among 121,064 adults aged 40–69 y by multivariable Cox regression analyses. We further assessed the associations between flavonoid intake and magnetic resonance imaging-derived liver fat (a subset of n = 11,435) and liver-corrected T1 values (cT1; a subset of n = 9570), a marker of steatosis, more sensitive to inflammatory pathology.
Results
Over 10 y of follow-up, 1081 cases of NAFLD were identified. Participants in the highest quartile (Q4) of the flavodiet score reflecting the consumption of foods high in flavonoids, had a 19% lower risk of NAFLD compared to the lowest quartile (Q1) [hazard ratio (HR) (95% confidence interval (CI): 0.81 (0.67, 0.97), P-trend = 0.02)]. Moreover, participants in the Q4 of the flavodiet score had lower liver fat and cT1 values than those in Q1 (liver fat: relative difference Q1 compared with Q4: –5.28%, P-trend = <0.001; cT1: relative difference Q1 compared with Q4: –1.73%, P-trend = <0.001). When compared to low intakes, high intakes of apples and tea were associated with lower NAFLD risk [apples: HR (95% CI): 0.78 (0.67, 0.92), P-trend = <0.01; tea: HR (95% CI): 0.86 (0.72, 1.02), P-trend = 0.03)]. Additionally, when compared to low intakes, high apple, tea, and dark chocolate intakes were significantly associated with lower liver fat values, whereas high tea and red pepper intakes were significantly associated with lower cT1 values.
Conclusions
The consumption of flavonoid-rich foods was associated with a reduced risk of NAFLD among middle-aged adults.
Keywords: flavonoids, flavonoid-rich foods, nonalcoholic fatty liver disease, prospective, liver fat, liver-corrected T1
Introduction
Over the past 30 y, nonalcoholic fatty liver disease (NAFLD) prevalence has increased by over 50% globally, with recent estimates suggesting over 32% of the global adult population is now affected [1,2]. NAFLD is defined as lipid accumulation in the liver of >5% in the absence of excessive alcohol consumption or other cause [3]. NAFLD can progress to inflammatory nonalcoholic steatohepatitis, fibrosis, and cirrhosis, the result of which is a loss of viable hepatic tissue and eventual liver failure [[3], [4], [5]]. The etiology of NAFLD is encompassed by a multi-hit hypothesis with a number of individual underlying factors, such as obesity and a sedentary lifestyle increasing risk, which, when combined, lead to the development of NAFLD [6]. Principal treatment options are cardiovascular disease risk reduction, mainly through weight loss [5], which may not be appropriate for all individuals. Preventative and treatment strategies should, therefore, focus on nutritional and lifestyle interventions, which include the optimal dietary components to mitigate NAFLD and its progression.
A healthful plant-based dietary pattern characterized by high intakes of fruits, vegetables, and whole grains has been shown to be beneficial for a range of cardiometabolic health outcomes [[7], [8], [9]]. Plant foods are rich in flavonoids, a heterogeneous group of bioactive compounds [10,11]. Flavonoid compounds can be broadly classified into 7 subgroups: anthocyanins, proanthocyanidins, flavones, flavanones, flavonols, flavan-3-ol monomers, and flavan-3-ol polymers [12]. Flavonoids are abundant in foods such as tea, red wine, apples, berries, grapes, and cocoa [13,14]. In population-based studies, a higher intake of several flavonoid subclasses has been associated with better cardiometabolic health, including lower risks of cardiovascular disease and type 2 diabetes mellitus[15,16]. Additionally, evidence from randomized controlled trials (RCTs) has demonstrated improvements in biomarkers of cardiometabolic health following consumption of flavonoid-rich foods [11,[17], [18], [19]].
For NAFLD, current evidence suggests a Mediterranean dietary pattern may reduce risk [20,21], and preliminary research has demonstrated the potential beneficial effects of flavonoid-rich dietary patterns in the development of NAFLD [22,23]. One previous prospective study in a Chinese population observed a 29% risk reduction for NAFLD progression in those with higher total flavonoid intakes [24], and further, it is possible that increased flavonoid intakes may partially explain the benefits conferred by a Mediterranean diet. However, to date, limited large-scale population-based research has been conducted on flavonoid-rich foods in relation to NAFLD. Here we aimed to assess the association between a flavonoid-rich diet with incident NAFLD and imaging-derived biomarkers of liver fat and steatosis in a large population-based prospective cohort study, the UK Biobank.
Methods
Study population
The UK Biobank is a prospective cohort in the United Kingdom. Between 2006 and 2010, over 500,000 participants aged 40–69 y were recruited from study centers in England, Scotland, and Wales. Comprehensive baseline assessments were taken, including sociodemographic data, health status, and dietary intake (via food frequency questionnaire). Additional and more detailed dietary data were collected between 2009 and 2012 through multiple Oxford WebQ 24-h dietary assessments (≤5). At follow-up, additional markers were measured, such as magnetic resonance imaging-based biomarkers of body composition (Supplementary Figure 1). The study was approved by the National Health Service Northwest Multicentre Research Ethics Committee (Reference 11/NW/0382), and all participants provided written informed consent at recruitment. A more detailed description of the study protocols can be found elsewhere [25].
The analysis is in 2 parts. Firstly, we present associations between 3 indices of flavonoid intake, a flavodiet score (FDS), flavonoid-rich foods, and flavonoid subclasses, and incident NAFLD. Secondly, we present associations between the same flavonoid indices and imaging biomarkers of NAFLD collected at imaging follow-up 1 (dates of first imaging collection ranged from 2014 to 2020 in our cohort) (Supplementary Figure 1). All dietary data were derived from the Oxford WebQ 24-h dietary assessments. Participants who withdrew consent during follow-up had completed <2 24-h dietary assessments (n = 375,497), had implausible energy intakes [>17,573 or <3347 kilojoule (KJ) for males and >14,644 or <2092 KJ for females] (n = 4818), had a prevalent diagnosis of conflicting conditions (n = 3935) (Supplementary Table 1) from International Classification of Disease 10th edition (ICD-10) inpatient diagnosis codes, or prevalent NAFLD (n = 274) (ascertained by hospital inpatient record diagnosis before baseline) were excluded from all analyses (Supplementary Figure 2).
Assessment of flavonoid exposures
Further details of the Oxford WebQ diet recall questionnaire, from which flavonoid indices were derived, can be found in Supplementary Methods 1. The primary exposure was the FDS, originally described in our previous work [26]. Briefly, the FDS is an additive score based on the sum of portions of flavonoid-rich foods consumed per day. Although the initial FDS was constructed in United States cohorts, we have recently established a modified FDS based on the top 3 highest contributing foods to total flavonoid and flavonoid subclass intake in the UK Biobank [27]. The components of the FDS are measured in servings per day, and the score is calculated as the sum of total intake for all included foods. Foods in the FDS are tea (black and green, capped at 4 servings per day), apples, berries, red wine, grapes, sweet peppers, onions, dark chocolate, oranges (including satsumas), and grapefruit. In ancillary analyses, we further analyzed associations between individual food constituents of the FDS in relation to incident NAFLD and imaging-derived biomarkers of liver fat and steatosis. In addition, flavonoid intake was analyzed at the subclass level (details on how subclass intakes were calculated can be found in Supplementary Methods 2).
Ascertainment of outcomes
Incident NAFLD was ascertained from ICD-10 data linked to hospital inpatient records. A diagnosis of any of the following codes after baseline was considered incident NAFLD: K74.0, K74.1, K74.2, K74.6, K75.8, K75.9, and K76.0. Participants were entered into the study from their last dietary assessment and followed up until the earliest censoring event (NAFLD diagnosis, competing diagnosis, death, loss to follow-up, or study censoring). Study censoring was based on the most up-to-date censoring dates from the hospital episode statistics for England, the Scottish morbidity records, and the patient episode database for Wales (31 October 2022, 31 August 2022, and 31 May 2022, respectively).
Imaging data was collected at the second follow-up assessment from 2014 onwards. Liver fat was measured as proton density fat fraction (PDFF) using Siemens 1.5T MAGNETOM Aera, which was used to calculate the PDFF map of the liver. Image analysis was performed using LiverMultiscan software, and PDFF maps were produced using a 3-point DIXON technique by trained image analysts. Liver-corrected T1 (cT1) values were also taken using Siemens 1.5T MAGNETOM Aera. Additional details on the exact protocol can be found for PDFF and cT1 values in Wilman et al. [28] and Mojtahed et al. [29], respectively.
Assessment of covariates
For this study, covariates were ascertained using the sociodemographic and anthropometric data collected through questionnaires at baseline assessment between 2006 and 2010. The following covariates for the present study were selected by literature search: sex (male or female), age, physical activity (metabolic equivalent scores, classified into “low,” “medium,” “high,” and “missing”), education (“low,” medium,” and “high”), deprivation (Townsend deprivation index), smoking status (“never,” “previous,” and “current”), prevalent type 2 diabetes mellitus [coded using ICD-10 E11 codes (“no” or “yes”)], total energy intake (KJ/d), fiber intake (g/d), number of 24-h dietary assessments completed, nonred wine alcohol intake (g/d, modeled in quartiles), coffee intake (3 groups set to “low,” “moderate,” and “high”), and BMI (in kg/m2). Further detail on each covariate and their classification, including how we handled missing data can be found in Supplementary Methods 3.
Statistical analyses
For prospective analyses on incident NAFLD, Cox proportional hazard regression models were utilized to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) of NAFLD risk. Age was used as the underlying time scale. For analyses on flavonoid exposures and imaging-derived liver fat and cT1, multivariable linear regression was used to estimate adjusted mean values in imaging parameters using the least squares means statistic. For all liver fat analysis, PDFF was transformed via natural logarithm to maintain a normal distribution. In all analyses, the FDS and flavonoid subclasses were modeled in quartiles to rank individuals’ intakes. In analyses on single flavonoid-rich foods, foods were categorized into 3 groups (0 recorded consumption, >0 and ≤ median, > median), except for grapefruit and dark chocolate, which were categorized into low and high consumers due to low frequency of consumption of these foods. As the dietary assessment in the UK Biobank captures the previous 24 h of dietary intake, infrequent consumption of flavonoid-rich foods may be recorded as 0 consumption. Thus, those with 0 recorded consumption are likely to be a combination of true 0 intake and those with low intakes. Therefore, throughout the text we refer to the 0 recorded consumption and low intake group as “low intake.” The median number of servings (in the case of the FDS and flavonoid-rich foods) and milligrams (in the case of the flavonoid subclasses) per day, constituting the respective intake groups, are shown in the results tables. Linear trend tests were carried out, modeling the categories of consumption as continuous variables in the multivariable regression models. All analyses were adjusted for baseline covariate data (see below).
All main regression analyses were conducted utilizing 2 principal models. Model 1 was adjusted for age (prospective analyses on incident NAFLD were also stratified by 5-y integers of age) and sex (male or female). Model 2 consisted of model 1 plus all other covariates listed above. For flavonoid-rich food-based analysis, a third model is presented which is model 2, with additional adjustments for the other major flavonoid-containing foods in the FDS (sum of portions).
We additionally performed sensitivity analyses to better understand specific factors driving any noted associations. First, we used an FDS excluding red wine to better account for the potential impact of immortal time bias and alcohol-specific reporting biases on our results. Immortal time bias relates to the lower likelihood of those with high red wine intake to be diagnosed with NAFLD, given that low total alcohol intake is a diagnostic criterion for NAFLD. Reporting alcohol intake has also been shown to be prone to different types of biases, such as reverse causation and social desirability bias [30]. These biases may, in part, favor inverse associations between red wine intake and NAFLD. Additionally, alcohol consumption is not recommended for the prevention of liver diseases [31]. Next, as tea is highly consumed in the United Kingdom and contributes 67.5% of total flavonoid intake in the UK Biobank, we used an FDS excluding tea to better ascertain if tea intake may be driving associations between the FDS and liver outcomes. In analyses on single foods, we used 2 nonflavonoid-containing negative controls [32]. These foods present similar foods (in composition and consumption pattern) that do not contain flavonoids, such as to analyze whether specific associations are present for the flavonoid-containing foods. We utilized coffee as a negative control for tea and white wine as a negative control for red wine. Models were altered to replicate conditions for comparison analysis (see legend in tables). We performed likelihood ratio tests to assess interactions between the FDS, key covariates, and NAFLD risk, and lastly the proportionality assumption for the Cox models was tested by assessing correlations between the Schoenfeld residuals and follow-up time.
All statistical analyses were conducted in R version 4.3.2 (R Foundation for Statistical Computing) [33]. Associations were considered statistically significant with a 2-sided P value <0.05.
Results
Characteristics of the study population
The mean (SD) age of the participants was 59.0 (7.9) y, 55.9% of participants were female, 63.0% had attained a “high” educational status, and the mean (SD) BMI was 26.7 (4.6) (Table 1). When stratified across quartiles of the FDS, participants in the lowest quartile were younger [mean (SD), quantile (Q1) = 57.6 (8.2) y compared with Q4 = 60.1 (7.4) y], had lower educational attainment (percentage in highest category: Q1 = 60.8% compared with Q4 = 65.6%), and were more likely to be a current smoker (Q1 = 8.3% compared with Q4 = 5.3%) (Table 1). The characteristics of the imaging cohorts and across the red wine-excluded FDS were similar (Supplementary Tables 2–4). Mean intakes for the flavonoid subclasses can be found in Supplementary Table 5.
TABLE 1.
Participant characteristics across quartiles of the flavodiet score.
| Q1 |
Q2 |
Q3 |
Q4 |
Overall |
|
|---|---|---|---|---|---|
| (N = 30,266) | (N = 30,266) | (N = 30,266) | (N = 30,266) | (N = 121,064) | |
| Sex | |||||
| Female, (%) | 16,819 (55.6) | 16,952 (56.0) | 16,810 (55.5) | 17,144 (56.6) | 67,725 (55.9) |
| Age at baseline (years) | |||||
| Mean (SD) | 57.6 (8.2) | 58.8 (7.9) | 59.4 (7.7) | 60.1 (7.4) | 59.0 (7.9) |
| Education | |||||
| Low, (%) | 6394 (21.1) | 5995 (19.8) | 6373 (21.1) | 5504 (18.2) | 24,266 (20.0) |
| Medium, (%) | 5475 (18.1) | 5018 (16.6) | 5160 (17.0) | 4906 (16.2) | 20,559 (17.0) |
| High, (%) | 18,397 (60.8) | 19,253 (63.6) | 18,733 (61.9) | 19,856 (65.6) | 76,239 (63.0) |
| Smoking status | |||||
| Never, (%) | 17,595 (58.1) | 17,610 (58.2) | 17,697 (58.5) | 17,090 (56.5) | 69,992 (57.8) |
| Previous, (%) | 10,174 (33.6) | 10,804 (35.7) | 10,651 (35.2) | 11,579 (38.3) | 43,208 (35.7) |
| Current, (%) | 2497 (8.3) | 1852 (6.1) | 1918 (6.3) | 1597 (5.3) | 7864 (6.5) |
| Deprivation index | |||||
| Mean (SD) | –1.4 (3.0) | –1.6 (2.8) | –1.8 (2.8) | –1.9 (2.7) | –1.7 (2.8) |
| Physical activity | |||||
| Low, (%) | 5421 (17.9) | 4540 (15.0) | 4568 (15.1) | 3753 (12.4) | 18,282 (15.1) |
| Moderate, (%) | 13,436 (44.4) | 14,171 (46.8) | 13,813 (45.6) | 14,001 (46.3) | 55,421 (45.8) |
| High, (%) | 6442 (21.3) | 6878 (22.7) | 6839 (22.6) | 7932 (26.2) | 28,091 (23.2) |
| Missing, (%) | 4967 (16.4) | 4677 (15.5) | 5046 (16.7) | 4580 (15.1) | 19,270 (15.9) |
| BMI (kg/m2) | |||||
| Mean (SD) | 27.2 (5.0) | 26.6 (4.5) | 26.5 (4.4) | 26.4 (4.3) | 26.7 (4.6) |
| Alcohol intake (g/d) | |||||
| Mean (SD) | 15.0 (18.4) | 16.8 (18.7) | 16.3 (18.9) | 20.2 (21.9) | 17.1 (19.7) |
| Energy intake (KJ/d) | |||||
| Mean (SD) | 8221.3 (2085.4) | 8484.1 (2013.4) | 8690.1 (2032.4) | 9041.3 (2095.0) | 8609.2 (2078.6) |
| Fiber Intake (g/d) | |||||
| Mean (SD) | 15.8 (5.2) | 17.3 (5.3) | 18.0 (5.4) | 20.6 (6.2) | 17.9 (5.8) |
| FDS (servings/d) | |||||
| Mean (SD) | 1.3 (0.7) | 3.2 (0.4) | 4.5 (0.4) | 6.3 (1.0) | 3.8 (1.9) |
| Prevalent T2DM, (%) | 582 (1.9) | 515 (1.7) | 516 (1.7) | 512 (1.7) | 2125 (1.8) |
| Incident NAFLD, (%) | 308 (1.0) | 279 (0.9) | 268 (0.9) | 226 (0.7) | 1081 (0.9) |
Abbreviations: BMI, body mass index; FDS, flavodiet score; KJ, kilojoule; NAFLD, nonalcoholic fatty liver disease; Q1–Q4, quantile; SD, standard deviation; T2DM, type 2 diabetes mellitus.
Incident NAFLD
Over a mean follow-up period of 10 y, 1081 cases of NAFLD were identified. The FDS demonstrated a linear inverse association with NAFLD risk in model 2, with the highest quartile of the FDS being associated with a statistically significant 19% lower risk for NAFLD compared to the lowest [HR (95% CI): 0.81 (0.67, 0.97), P-trend = 0.02) (Table 2)]. In food-based analyses, model 2 demonstrated high apple intake was associated with a 22% lower risk of NAFLD compared to low intakes [HR (95% CI): 0.78 (0.67, 0.92), P-trend = <0.01; Table 2)], whereas high tea intake was associated with a 14% lower risk of developing NAFLD compared to low intake [HR (95% CI): 0.86 (0.72, 1.02), P-trend = 0.03; Table 2)]. Associations between tea and apple intake and incident NAFLD did not change measurably following further adjustment with intakes of other flavonoid-rich foods. In model 2, no significant associations with red wine, berries, or less commonly consumed flavonoid-rich foods were observed (Supplementary Table 6). When looking at flavonoid subclasses, we observed that the highest quartile of proanthocyanidin, theaflavin and thearubigin, flavonol and flavan-3-ol intake was significantly associated with a lower risk of NAFLD in model 2 when compared to the lowest quartile of intake [proanthocyanidins: HR (95% CI): 0.77 (0.63, 0.93), P-trend = <0.01; theaflavins and thearubigins: HR (95% CI): 0.78 (0.65, 0.94), P-trend = 0.01; flavonols: HR (95% CI): 0.78 (0.64, 0.95), P-trend = <0.01; flavan-3-ols: HR (95% CI): 0.74 (0.61, 0.90), P-trend = <0.001) (Table 2). In model 2, no significant associations were noted for the other subclasses.
TABLE 2.
Hazard ratios and 95% confidence intervals for associations between main flavonoid exposures and incident nonalcoholic fatty liver disease.
| Exposure |
P-trend | |||||
|---|---|---|---|---|---|---|
| Diet score | Q1 | Q2 | Q3 | Q4 | ||
| Flavodiet score | n/cases | 30,266/308 | 30,266/279 | 30,266/268 | 30,266/226 | |
| Intake (servings/d) | 1.5 (0.0–2.4) | 3.2 (2.4–3.9) | 4.5 (3.9–5.1) | 6.0 (5.1–19.2) | ||
| Model 1 | 1 | 0.87 (0.74, 1.02) | 0.83 (0.70, 0.97) | 0.68 (0.57, 0.81) | <0.001 | |
| Model 2 | 1 | 1.00 (0.84, 1.18) | 0.92 (0.78, 1.10) | 0.81 (0.67, 0.97) | 0.02 | |
| Subclasses | ||||||
| Anthocyanins | n/cases | 30,266/319 | 30,266/268 | 30,266/250 | 30,266/244 | |
| Intake (mg/d) | 3.2 (0.0–7.2) | 13.4 (7.2–20.6) | 29.1 (20.6–40.4) | 58.4 (40.4–307.2) | ||
| Model 1 | 1 | 0.82 (0.69, 0.96) | 0.75 (0.63, 0.88) | 0.72 (0.61, 0.86) | <0.001 | |
| Model 2 | 1 | 0.98 (0.83, 1.16) | 1.01 (0.85, 1.20) | 0.98 (0.82, 1.18) | 0.92 | |
| Proanthocyanidins | n/cases | 30,266/340 | 30,266/258 | 30,266/255 | 30,266/228 | |
| Intake (mg/d) | 162.5 (0.0–230.5) | 285.7 (230.5–336.4) | 387.5 (336.4–447.3) | 536.2 (447.3–2338.3) | ||
| Model 1 | 1 | 0.73 (0.62, 0.85) | 0.71 (0.60, 0.83) | 0.62 (0.53, 0.74) | <0.001 | |
| Model 2 | 1 | 0.85 (0.72, 1.01) | 0.83 (0.70, 0.99) | 0.77 (0.63, 0.93) | <0.01 | |
| Theaflavins and thearubigins | n/cases | 30,266/288 | 30,266/283 | 30,266/273 | 30,266/237 | |
| Intake (mg/d) | 0.00 (0.00–83.02) | 226.62 (83.02–332.08) | 456.61 (332.08–581.14) | 747.18 (581.14–1100.19) | ||
| Model 1 | 1 | 0.94 (0.80, 1.11) | 0.90 (0.76, 1.06) | 0.78 (0.66, 0.93) | <0.01 | |
| Model 2 | 1 | 1.08 (0.91, 1.27) | 1.02 (0.86, 1.22) | 0.78 (0.65, 0.94) | 0.01 | |
| Flavonols | n/cases | 30,266/307 | 30,266/290 | 30,266/250 | 30,266/234 | |
| Intake (mg/d) | 13.4 (0.0–20.3) | 26.3 (20.3–31.8) | 37.3 (31.8–43.5) | 51.5 (43.5–161.6) | ||
| Model 1 | 1 | 0.91 (0.77, 1.07) | 0.77 (0.66, 0.92) | 0.72 (0.61, 0.85) | <0.001 | |
| Model 2 | 1 | 1.03 (0.88, 1.22) | 0.89 (0.74, 1.06) | 0.78 (0.64, 0.95) | <0.01 | |
| Flavones | n/cases | 30,266/332 | 30,266/250 | 30,266/237 | 30,266/262 | |
| Intake (mg/d) | 0.3 (0.0–0.5) | 0.7 (0.5–0.9) | 1.1 (0.9–1.4) | 2.0 (1.4–15.6) | ||
| Model 1 | 1 | 0.74 (0.63, 0.87) | 0.69 (0.59, 0.82) | 0.77 (0.66, 0.91) | <0.01 | |
| Model 2 | 1 | 0.89 (0.76, 1.06) | 0.89 (0.74, 1.06) | 1.03 (0.86, 1.24) | 0.89 | |
| Flavan-3-ols | n/cases | 30,266/303 | 30,266/297 | 30,266/264 | 30,266/217 | |
| Intake (mg/d) | 38.9 (0.0–85.3) | 128.1 (85.3–166.1) | 203.6 (166.1–247.1) | 307.0 (247.1–1903.7) | ||
| Model 1 | 1 | 0.94 (0.80, 1.11) | 0.83 (0.70, 0.98) | 0.68 (0.57, 0.81) | <0.001 | |
| Model 2 | 1 | 1.08 (0.92, 1.28) | 0.94 (0.79, 1.12) | 0.74 (0.61, 0.90) | <0.01 | |
| Flavanones |
n/cases | 30,266/289 | 30,266/282 | 30,266/258 | 30,266/252 | |
| Intake (mg/d) | 1.6 (0.0–5.3) | 10.8 (5.3–17.9) | 25.7 (17.9–35.7) | 50.2 (35.7–376.7) | ||
| Model 1 | 1 | 0.96 (0.81, 1.13) | 0.86 (0.72, 1.01) | 0.82 (0.69, 0.97) | <0.01 | |
| Model 2 |
1 |
1.09 (0.92, 1.29) |
1.02 (0.86, 1.21) |
1.01 (0.84, 1.20) |
0.90 |
|
| Flavonoid-rich foods |
Low intake |
Moderate intake |
High intake |
|||
| Tea | n/cases | 22,981/226 | 49,630/461 | 48,453/394 | ||
| Intake (servings/d) | 0.0 (0.0–0.0) | 1.7 (0.2–2.7) | 4.0 (2.8–10.5) | |||
| Model 1 | 1 | 0.91 (0.77, 1.06) | 0.78 (0.67, 0.92) | <0.01 | ||
| Model 2 | 1 | 1.06 (0.90, 1.25) | 0.86 (0.72, 1.02) | 0.03 | ||
| Model 3 | 1 | 1.07 (0.90, 1.25) | 0.86 (0.72, 1.02) | 0.03 | ||
| Berries | n/cases | 77,445/715 | 30,161/257 | 13,458/109 | ||
| Intake (servings/d) | 0.00 (0.00–0.00) | 0.33 (0.10–0.50) | 1.00 (0.60–4.00) | |||
| Model 1 | 1 | 0.92 (0.79, 1.06) | 0.89 (0.73, 1.09) | 0.16 | ||
| Model 2 | 1 | 1.06 (0.92, 1.23) | 1.00 (0.81, 1.24) | 0.70 | ||
| Model 3 | 1 | 1.07 (0.92, 1.24) | 1.00 (0.81, 1.23) | 0.70 | ||
| Apple | n/cases | 53,683/540 | 35,057/308 | 32,324/233 | ||
| Intake (servings/d) | 0.00 (0.00–0.00) | 0.33 (0.10–0.50) | 1.00 (0.60–4.00) | |||
| Model 1 | 1 | 0.86 (0.75, 0.99) | 0.69 (0.59, 0.80) | <0.001 | ||
| Model 3 | 1 | 0.94 (0.81, 1.08) | 0.78 (0.67, 0.92) | <0.01 | ||
| Model 3 | 1 | 0.94 (0.82, 1.09) | 0.79 (0.67, 0.93) | <0.01 | ||
| Red wine | n/cases | 76,567/769 | 23,517/155 | 20,980/157 | ||
| Intake (servings/d) | 0.0 (0.0–0.0) | 0.5 (0.1–0.8) | 1.3 (0.8–6.0) | |||
| Model 1 | 1 | 0.63 (0.53, 0.75) | 0.70 (0.59, 0.83) | <0.001 | ||
| Model 2 | 1 | 0.89 (0.73, 1.07) | 0.90 (0.75, 1.09) | 0.21 | ||
| Model 3 | 1 | 0.89 (0.73, 1.07) | 0.90 (0.74, 1.08) | 0.20 | ||
Abbreviations: BMI, body mass index; Q1–Q4, quantile.
Cox proportional hazard models were used to obtain hazard ratio and 95% confidence interval estimates across categories of flavonoid exposures. P-trend values were obtained by modeling our exposures as continuous in the same test.
Model 1: age as the underlying time variable. Adjusted for sex and further stratified by age (5-y integers).
Model 2: model 1, and further adjusted for sex, education, smoking, deprivation, BMI, physical activity, prevalent type 2 diabetes mellitus, number of dietary assessments completed, total energy intake, fiber intake, total coffee intake, nonred wine alcohol intake.
Model 3: model 2 plus summed intakes of other flavonoid-rich foods (servings/day).
Intakes are reported as median (range).
Imaging biomarkers
Participants in Q4 of the FDS had significantly lower liver fat percentages compared to participants in Q1 in model 2 (relative difference Q1 compared with Q4: –5.28%, P-trend = <0.001) (Table 3). A similar association was noted for liver cT1 (relative difference Q1 compared with Q4: –1.73%, P-trend = <0.001) (Table 4). In food-based analyses, model 2 demonstrated participants with a high intake of tea and apples had a lower liver fat content compared to those with low intakes (tea: relative difference low compared with high = –6.60%, P-trend = <0.001); apple: relative difference low compared with high = –4.38%, P-trend = <0.001). However, a high berry and red wine intake was not associated with lower liver fat values when compared to a low intake (Table 3). With regards to cT1 values, in model 2, a high intake of both tea and red wine was inversely associated with cT1 values when compared to a low intake (tea: relative difference low compared with high = –1.63%, P-trend = <0.001; red wine: relative difference low compared with high = –1.05%, P-trend = <0.001) (Table 4).
TABLE 3.
Associations between main flavonoid exposures and liver fat.
| Exposure |
Relative difference (highest vs. lowest category) (%) | P-Trend | |||||
|---|---|---|---|---|---|---|---|
| Diet score | Q1 | Q2 | Q3 | Q4 | |||
| Flavodiet score | Intake (servings/d) | 1.5 (0.0–2.5) | 3.2 (2.5–4.0) | 4.5 (4.0–5.1) | 6.0 (5.1–12.2) | ||
| Model 1 | 5.01 (4.91, 5.11) | 4.59 (4.50, 4.69) | 4.57 (4.48, 4.67) | 4.39 (4.30, 4.48) | –12.38 | <0.001 | |
| Model 2 | 5.11 (4.85, 5.38) | 4.91 (4.66, 5.17) | 4.92 (4.67, 5.18) | 4.84 (4.60, 5.10) | –5.28 | <0.001 | |
| Subclasses | |||||||
| Anthocyanins | Intake (mg/d) | 3.7 (0.0–8.0) | 14.3 (8.0–21.6) | 30.0 (21.6–41.0) | 58.5 (41.0–307.2) | ||
| Model 1 | 4.81 (4.71, 4.91) | 4.55 (4.46, 4.64) | 4.51 (4.42, 4.60) | 4.68 (4.59, 4.78) | –2.7 | 0.06 | |
| Model 2 | 4.88 (4.63, 5.13) | 4.89 (4.65, 5.15) | 4.92 (4.68, 5.18) | 5.08 (4.82, 5.35) | 4.1 | <0.01 | |
| Proanthocyanidins | Intake (mg/d) | 168.1 (0.0–236.2) | 289.8 (236.2–342.1) | 393.1 (342.1–451.9) | 539.9 (451.9–2338.3) | ||
| Model 1 | 5.12 (5.02, 5.22) | 4.65 (4.56, 4.75) | 4.48 (4.39, 4.57) | 4.33 (4.25, 4.42) | –15.43 | <0.001 | |
| Model 2 | 5.16 (4.90, 5.43) | 4.95 (4.70, 5.21) | 4.82 (4.58, 5.08) | 4.80 (4.56, 5.06) | –6.98 | <0.001 | |
| Theaflavins and thearubigins | Intake (mg/d) | 0.0 (0.0–83.0) | 249.1 (83.0–333.2) | 462.0 (333.2–581.1) | 747.2 (581.1–1100.2) | ||
| Model 1 | 4.88 (4.78, 4.98) | 4.65 (4.56, 4.75) | 4.58 (4.48, 4.67) | 4.45 (4.36, 4.54) | –8.81 | <0.001 | |
| Model 2 | 5.11 (4.86, 5.38) | 5.03 (4.78, 5.30) | 4.95 (4.71, 5.22) | 4.73 (4.49, 4.98) | –7.44 | <0.001 | |
| Flavones | Intake (mg/d) | 0.3 (0.0–0.5) | 0.7 (0.5–0.9) | 1.1 (0.9–1.5) | 2.0 (1.5–8.5) | ||
| Model 1 | 4.87 (4.77, 4.97) | 4.69 (4.59, 4.78) | 4.53 (4.44, 4.62) | 4.47 (4.38, 4.56) | –8.21 | <0.001 | |
| Model 2 | 4.91 (4.66, 5.17) | 4.94 (4.69, 5.20) | 4.89 (4.64, 5.15) | 5.02 (4.77, 5.29) | 2.24 | 0.22 | |
| Flavanones | Intake (mg/d) | 2.0 (0.0–6.3) | 12.3 (6.30–18.8) | 26.6 (18.9–36.8) | 51.3 (36.8–182.4) | ||
| Model 1 | 4.79 (4.70, 4.89) | 4.68 (4.59, 4.78) | 4.59 (4.50, 4.68) | 4.49 (4.40, 4.58) | –6.26 | <0.001 | |
| Model 2 | 4.94 (4.69, 5.20) | 4.96 (4.71, 5.22) | 4.94 (4.69, 5.20) | 4.92 (4.67, 5.17) | –0.4 | 0.62 | |
| Flavan-3-ols | Intake (mg/d) | 39.8 (0.0–87.2) | 131.3 (87.3–168.4) | 205.1 (168.5–248.3) | 306.9 (248.3–1903.7) | ||
| Model 1 | 5.01 (4.91, 5.11) | 4.60 (4.51, 4.70) | 4.52 (4.43, 4.62) | 4.43 (4.34, 4.52) | –11.58 | <0.001 | |
| Model 2 | 5.15 (4.89, 5.42) | 4.95 (4.70, 5.21) | 4.85 (4.61, 5.11) | 4.79 (4.55, 5.04) | –6.99 | <0.001 | |
| Flavonols |
Intake (mg/d) | 13.8 (0.3–20.6) | 26.7 (20.6–32.2) | 37.6 (32.2–43.7) | 51.3 (43.8–125.2) | ||
| Model 1 | 5.04 (4.94, 5.15) | 4.59 (4.50, 4.69) | 4.53 (4.44, 4.63) | 4.40 (4.31, 4.49) | –12.7 | <0.001 | |
| Model 2 |
5.17 (4.91, 5.44) |
4.92 (4.68, 5.19) |
4.89 (4.65, 5.15) |
4.78 (4.54, 5.04) |
–7.54 |
<0.001 |
|
| Flavonoid-rich foods |
Low intake |
Moderate intake |
High intake |
||||
| Tea | Intake (servings/d) | 0.0 (0.0–0.0) | 1.7 (0.2–2.8) | 4.0 (2.8–10.0) | |||
| Model 1 | 4.96 (4.84, 5.07) | 4.63 (4.56, 4.71) | 4.50 (4.42, 4.57) | –9.27 | <0.001 | ||
| Model 2 | 5.15 (4.88, 5.43) | 5.00 (4.75, 5.25) | 4.81 (4.57, 5.06) | –6.6 | <0.001 | ||
| Model 3 | 5.14 (4.88, 5.42) | 4.99 (4.75, 5.25) | 4.81 (4.58, 5.06) | –6.42 | <0.001 | ||
| Berries | Intake (servings/d) | 0.0 (0.0–0.0) | 0.3 (0.1–0.5) | 1.0 (0.6–3.5) | |||
| Model 1 | 4.69 (4.63, 4.75) | 4.55 (4.46, 4.64) | 4.54 (4.40, 4.68) | –3.2 | <0.01 | ||
| Model 2 | 4.92 (4.68, 5.17) | 4.96 (4.71, 5.22) | 5.01 (4.74, 5.30) | 1.83 | 0.19 | ||
| Model 3 | 4.93 (4.69, 5.18) | 4.97 (4.72, 5.23) | 5.02 (4.75, 5.30) | 1.83 | 0.21 | ||
| Apple | Intake (servings/d) | 0.0 (0.0–0.0) | 0.3 (0.1–0.5) | 1.0 (0.6–4.0) | |||
| Model 1 | 4.88 (4.80, 4.95) | 4.62 (4.53, 4.71) | 4.31 (4.22, 4.39) | –11.68 | <0.001 | ||
| Model 2 | 5.02 (4.77, 5.27) | 4.96 (4.71, 5.22) | 4.80 (4.57, 5.06) | –4.38 | <0.001 | ||
| Model 3 | 5.02 (4.78, 5.28) | 4.96 (4.72, 5.23) | 4.81 (4.57, 5.06) | –4.18 | <0.01 | ||
| Red wine | Intake (servings/d) | 0.0 (0.0–0.0) | 0.5 (0.1–0.8) | 1.3 (0.8–6.0) | |||
| Model 1 | 4.70 (4.63, 4.76) | 4.39 (4.30, 4.49) | 4.74 (4.63, 4.86) | 0.85 | 0.38 | ||
| Model 2 | 4.92 (4.68, 5.17) | 4.89 (4.63, 5.15) | 5.06 (4.80, 5.34) | 2.85 | 0.08 | ||
| Model 3 | 4.93 (4.70, 5.18) | 4.89 (4.64, 5.16) | 5.06 (4.80, 5.34) | 2.64 | 0.10 | ||
Abbreviations: BMI, body mass index; Q1–Q4, quantile.
Adjusted mean values for biomarker data were obtained via multivariable linear regression using the least squares mean statistic. P-trend values were obtained by modeling our exposures as continuous in the same test.
Model 1: age and sex.
Model 2: model 1, and further adjusted for sex, education, smoking, deprivation, BMI, physical activity, prevalent type 2 diabetes mellitus, number of dietary assessments completed, total energy intake, fiber intake, total coffee intake, nonred wine alcohol intake.
Model 3: model 2 plus summed intakes of other flavonoid-rich foods (servings/day).
Intakes are reported as median (range).
TABLE 4.
Associations between main flavonoid exposures and liver-corrected T1.
| Exposure |
Relative difference (highest vs. lowest category) (%) | P-trend | |||||
|---|---|---|---|---|---|---|---|
| Diet score | Q1 | Q2 | Q3 | Q4 | |||
| Flavodiet score | Intake (servings/d) | 1.5 (0.0–2.5) | 3.2 (2.5–4.0) | 4.5 (4.0–5.1) | 6.0 (5.1–12.2) | ||
| Model 1 | 708.62 (706.5, 710.75) | 697.43 (695.31, 699.55) | 695.85 (693.73, 697.98) | 692.89 (690.77, 695.02) | –2.22 | <0.001 | |
| Model 2 | 713.71 (707.90, 719.52) | 705.62 (699.85, 711.38) | 704.37 (698.59, 710.16) | 701.37 (695.60, 707.14) | –1.73 | <0.001 | |
| Subclasses | |||||||
| Anthocyanins | Intake (mg/d) | 3.7 (0.0–8.0) | 14.3 (8.0–21.7) | 30.3 (21.7–41.2) | 58.8 (41.2–307.2) | ||
| Model 1 | 702.14 (700.00, 704.29) | 701.21 (699.08, 703.34) | 697.44 (695.30, 699.57) | 694.01 (691.87, 696.15) | –1.16 | <0.001 | |
| Model 2 | 706.56 (700.80, 712.33) | 708.63 (702.81, 714.44) | 706.13 (700.32, 711.94) | 703.00 (697.19, 708.82) | –0.5 | 0.01 | |
| Proanthocyanidins | Intake (mg/d) | 165.9 (0.0–234.4) | 289.6 (234.4–341.8) | 393.2 (341.9–451.8) | 541.4 (451.9–2338.3) | ||
| Model 1 | 707.24 (705.11, 709.36) | 698.19 (696.07, 700.32) | 695.52 (693.39, 697.64) | 693.89 (691.76, 696.01) | –1.89 | <0.001 | |
| Model 2 | 711.34 (705.55, 717.13) | 705.98 (700.18, 711.78) | 703.36 (697.58, 709.13) | 702.51 (696.68, 708.34) | –1.24 | <0.001 | |
| Theaflavins and thearubigins | Intake (mg/d) | 0.0 (0.0–83.0) | 222.1 (83.0–332.1) | 456.6 (332.1–581.1) | 747.2 (581.1–1100.2) | ||
| Model 1 | 707.82 (705.69, 709.94) | 697.61 (695.48, 699.73) | 694.83 (692.70, 696.95) | 694.61 (692.48, 696.73) | –1.87 | <0.001 | |
| Model 2 | 713.33 (707.54, 719.13) | 706.49 (700.70, 712.29) | 703.11 (697.34, 708.88) | 701.86 (696.08, 707.63) | –1.61 | <0.001 | |
| Flavones | Intake (mg/d) | 0.3 (0.0–0.5) | 0.7 (0.5–0.9) | 1.1 (0.9–1.5) | 2.0 (1.5–7.8) | ||
| Model 1 | 704.24 (702.10, 706.38) | 697.87 (695.74, 700.00) | 696.67 (694.54, 698.80) | 695.96 (693.82, 698.10) | –1.18 | <0.001 | |
| Model 2 | 709.66 (703.84, 715.47) | 705.08 (699.30, 710.85) | 704.76 (698.98, 710.55) | 704.47 (698.63, 710.32) | –0.73 | <0.01 | |
| Flavanones | Intake (mg/d) | 2.0 (0.0–6.4) | 12.6 (6.4–19.2) | 26.8 (19.2–36.9) | 51.2 (36.9–182.4) | ||
| Model 1 | 700.89 (698.75, 703.03) | 699.38 (697.24, 701.51) | 696.31 (694.18, 698.44) | 698.28 (696.15, 700.41) | –0.37 | 0.03 | |
| Model 2 | 706.09 (700.30, 711.87) | 706.96 (701.17, 712.74) | 704.29 (698.47, 710.12) | 706.36 (700.58, 712.13) | 0.04 | 0.68 | |
| Flavan-3-ols | Intake (mg/d) | 38.8 (0.0–85.8) | 129.6 (85.8–167.6) | 204.4 (167.7–247.3) | 307.7 (247.3–1903.7) | ||
| Model 1 | 708.56 (706.43, 710.68) | 697.57 (695.45, 699.69) | 695.02 (692.89, 697.14) | 693.67 (691.55, 695.80) | –2.1 | <0.001 | |
| Model 2 | 713.21 (707.43, 718.99) | 705.67 (699.89, 711.44) | 702.96 (697.16, 708.75) | 701.37 (695.58, 707.15) | –1.66 | <0.001 | |
| Flavonols |
Intake (mg/d) | 13.6 (0.4–20.5) | 26.5 (20.5–32.0) | 37.6 (32.0–43.5) | 51.2 (43.5–125.2) | ||
| Model 1 | 709.03 (706.91, 711.15) | 696.84 (694.72, 698.96) | 696.29 (694.17, 698.41) | 692.68 (690.56, 694.80) | –2.31 | <0.001 | |
| Model 2 |
714.16 (708.37, 719.96) |
705.02 (699.21, 710.82) |
704.76 (699.00, 710.52) |
700.16 (694.37, 705.94) |
–1.96 |
<0.001 |
|
| Flavonoid-rich foods |
Low intake |
Moderate intake |
High intake |
||||
| Tea | Intake (servings/d) | 0.0 (0.0–0.0) | 1.7 (0.2–2.7) | 4.0 (2.8–10.0) | |||
| Model 1 | 708.94 (706.50, 711.38) | 698.04 (696.38, 699.69) | 694.57 (692.89, 696.25) | –2.03 | <0.001 | ||
| Model 2 | 713.93 (708.01, 719.86) | 706.68 (701.04, 712.33) | 702.32 (696.68, 707.96) | –1.63 | <0.001 | ||
| Model 3 | 714.08 (708.15, 720.00) | 706.74 (701.10, 712.38) | 702.22 (696.58, 707.86) | –1.66 | <0.001 | ||
| Berries | Intake (servings/d) | 0.0 (0.0–0.0) | 0.3 (0.1–0.5) | 1.0 (0.6–3.0) | |||
| Model 1 | 699.96 (698.61, 701.32) | 696.31 (694.22, 698.40) | 697.32 (694.11, 700.53) | –0.38 | 0.01 | ||
| Model 2 | 706.47 (700.90, 712.04) | 704.73 (698.93, 710.54) | 706.02 (699.79, 712.26) | –0.06 | 0.41 | ||
| Model 3 | 706.91 (701.36, 712.45) | 705.06 (699.27, 710.84) | 706.34 (700.13, 712.55) | –0.08 | 0.35 | ||
| Apple | Intake (servings/d) | 0.0 (0.0–0.0) | 0.3 (0.1–0.5) | 1.0 (0.6–4.0) | |||
| Model 1 | 699.59 (697.96, 701.22) | 699.17 (697.20, 701.13) | 696.87 (694.84, 698.90) | –0.39 | 0.05 | ||
| Model 2 | 706.05 (700.42, 711.68) | 706.95 (701.16, 712.74) | 705.42 (699.65, 711.18) | –0.09 | 0.72 | ||
| Model 3 | 706.24 (700.63, 711.84) | 707.32 (701.55, 713.09) | 706.03 (700.29, 711.78) | –0.03 | 0.97 | ||
| Red wine | Intake (servings/d) | 0.0 (0.0–0.0) | 0.5 (0.1–0.8) | 1.3 (0.8–6.0) | |||
| Model 1 | 702.44 (701.06, 703.82) | 694.44 (692.15, 696.73) | 691.66 (689.16, 694.16) | –1.53 | <0.001 | ||
| Model 2 | 707.92 (702.33, 713.51) | 704.89 (699.00, 710.79) | 700.47 (694.50, 706.44) | –1.05 | <0.001 | ||
| Model 3 | 708.30 (702.72, 713.87) | 705.16 (699.28, 711.04) | 700.53 (694.58, 706.49) | –1.1 | <0.001 | ||
Abbreviations: BMI, body mass index; Q1–Q4, quantile .
Adjusted mean values for biomarker data were obtained via multivariable linear regression using the least squares mean statistic. P-trend values were obtained by modeling our exposures as continuous in the same test.
Model 1: age and sex.
Model 2: model 1, and further adjusted for sex, education, smoking, deprivation, BMI, physical activity, prevalent type 2 diabetes mellitus, number of dietary assessments completed, total energy intake, fiber intake, total coffee intake, nonred wine alcohol intake.
Model 3: model 2 plus summed intakes of other flavonoid-rich foods (servings/day).
Intakes are reported as median (range).
Associations for less commonly consumed flavonoid-rich foods can be found in Supplementary Table 7. In model 2, when compared to low intakes, high sweet pepper intakes were associated with a lower cT1 value (relative difference low compared with high = –0.51%, P-trend = 0.01), whereas high grape intakes were associated with higher cT1 values when adjusted for other flavonoid-rich food intakes (relative difference low compared with high = +0.59%, P-trend = <0.01). Additionally, in model 2, high intakes of dark chocolate were associated with a lower liver fat when compared to low intakes (relative difference low compared with high = –3.63%, P-trend = <0.01), whereas a high intake of onions and grapes were associated with higher liver fat when compared to low intakes (onion: relative difference low compared with high = +4.53%, P-trend = <0.001, grape: relative difference low compared with high = +2.85, P-trend = 0.04). Further adjustment for other flavonoid-rich foods in the FDS did not significantly alter these associations (TABLE 3, TABLE 4, Supplementary Table 7). In model 2, the highest quartile of intake of the same subclasses associated with incident NAFLD were associated with lower liver fat values when compared to the lowest quartile (relative difference Q1 compared with Q4; proanthocyanidins: –6.98%, P-trend = <0.001; theaflavins and thearubigins: –7.44%, P-trend = <0.001; flavonols: –7.54%, P-trend = <0.001; flavan-3-ols: –6.99%, P-trend = <0.001) and cT1 values (relative difference Q1 compared with Q4; proanthocyanidins: –1.24%, P-trend = <0.001; theaflavins and thearubigins: –1.61%, P-trend = <0.001; flavonols: –1.96%, P-trend = <0.001; flavan-3-ols: –1.66%, P-trend = <0.001). Additionally, the highest quartile of anthocyanin intake was positively associated with liver fat when compared to the lowest quartile (relative difference Q1 compared with Q4; anthocyanins: +4.10%, P-trend = <0.01, whereas for cT1, inverse associations for higher intakes of anthocyanins and flavones were observed in the same comparison (relative difference Q1 compared with Q4; anthocyanins: –0.50%, P-trend = <0.01; flavones: –0.73%, P-trend = <0.01) (TABLE 3, TABLE 4).
Incident NAFLD - sensitivity analyses
The exclusion of red wine from the FDS did not materially change our main results, with an HR: 0.79; 95% CI: 0.66, 0.96, P-trend = 0.03 for participants in the highest compared to the lowest quartile of the modified FDS (Supplementary Table 8). The exclusion of tea from the FDS attenuated the association [HR (95% CI): 0.93 (0.78, 1.12), P-trend = 0.25)].
Using coffee intake as a negative control for tea intake, we observed an inverse association of coffee intake with incident NAFLD in model 2 when comparing high to low intakes [HR (95% CI): 0.71 (0.59, 0.84), P-trend = <0.001] (Supplementary Table 8). Using white wine as a nonflavonoid-containing control for red wine, we observed a similar null association with NAFLD risk in model 2 when comparing high to low intakes [HR (95% CI): 0.95 (0.78, 1.17), P-trend = 0.67] (Supplementary Table 8).
Analyses of Schoenfeld residuals indicated that these were not independent of follow-up time with regard to the FDS (P = 0.03). Thus, we stratified our main analyses by follow-up duration using 3 categories (≤4 y, 5–8 y, and 9–12 y). We observed similar associations between the FDS and NAFLD risk across the first and second 4 y of follow-up comparing Q4 of the FDS to Q1 [follow-up years 0–4, HR (95% CI): 0.62 (0.40, 0.95), P-trend = 0.048; follow-up years 4–8, HR (95% CI): 0.73 (0.54, 0.99), P-trend = 0.046] in model 2 (Supplementary Table 9). However, restricting the analysis to years 9–12 did result in an attenuation of the association between the FDS and NAFLD risk [HR (95% CI): 1.06 (0.78, 1.42), P-trend = 0.97)] in model 2 (Supplementary Table 9). We further carried out sensitivity analyses excluding cases diagnosed in calendar years 2020–2022 due to potential detection bias related to the COVID-19 pandemic (Supplementary Table 10). These analyses showed similar associations compared to our main analyses without indication of nonproportionality (Supplementary Table 9).
Likelihood ratio tests for statistical interaction did not indicate heterogeneity in the association between the FDS and NAFLD across subgroups of the population (Supplementary Table 11).
Imaging Biomarkers - sensitivity analyses
The highest quartile of the red wine-excluded FDS was also inversely associated with both liver fat and cT1 values in model 2 when compared to the lowest quartile (liver fat: relative difference Q1 compared with Q4 = –5.88%, P-trend = <0.001; cT1: relative difference Q1 compared with Q4 = –1.39%, P-trend = <0.001). Exclusion of tea from the FDS resulted in attenuation of the inverse association with liver fat, but not cT1 (liver fat: relative difference Q1 compared with Q4 = +1.83%, P-trend = 0.06; cT1: relative difference Q1 compared with Q4 = –0.67%, P-trend = 0.01) (Supplementary Table 12).
Using white wine as a negative control resulted in similar associations to red wine for both liver fat and cT1 (Supplementary Table 12), with high intakes of white wine showing no significant association with liver fat but a significant inverse association with cT1 in model 2 when compared to low intakes (liver fat: relative difference Q1 compared with Q3 = +1.01%, P-trend = 0.46; cT1: relative difference Q1 compared with Q3 = –1.41%, P-trend = <0.001). Using coffee as a negative control for tea intake, we observed significant inverse associations for coffee intake with liver fat, but not cT1 when comparing high to low intakes (liver fat: relative difference Q1 compared with Q3 = –8.25%, P-trend = <0.001; cT1: relative difference Q1 compared with Q3 = –0.28%, P-trend = 0.25).
Discussion
We have shown that a higher FDS and intake of several specific flavonoid-rich foods, including apples and tea, was associated with a lower risk of developing NAFLD and lower imaging-derived liver fat and cT1 values. Furthermore, these associations remained statistically significant following adjustment for established risk factors of NAFLD.
Few prospective studies have investigated the associations between flavonoid intakes and NAFLD specifically. Zhong et al. [24] observed a 29% lower risk of NAFLD progression for higher total flavonoid intake over a 3-y follow-up in an aging cohort from China. Interestingly, the flavan-3-ol subclass showed a strong inverse association with NAFLD progression in the study by Zhong et al. [24], which is in line with our finding of an inverse association between flavan-3-ol intake and NAFLD risk. Additionally, a recent systematic review and meta-analysis of RCTs demonstrated reductions in biomarkers of liver dysfunction from 12 RCTs of flavonoid supplementation [34]. Although the results by Zhong et al. [24] and Li et al. [34] are in line with ours in that they point to the beneficial effects of flavonoids across different phases of NAFLD development, our study was the first to assess the associations between a flavonoid-rich diet and both our incident NAFLD and validated biomarkers of NAFLD in a prospective cohort. Furthermore, our study adds to this body of research with a larger sample size, longer follow-up time, and a dietary pattern score demonstrating relevance to food-based public health guidance.
Mechanistically, NAFLD is thought to be driven by a multi-hit hypothesis, with key “hits” considered to be insulin resistance, impaired lipid metabolism (resulting in lipotoxicity), and inflammation [6]. Insulin resistance upregulates de-novo lipogenesis in hepatocytes and suppresses hormone-sensitive lipase-driven lipolysis in adipose tissue, resulting in increased nonesterified fatty acid delivery to the liver [6]. This lipid accumulation, alongside oxidative stress and inflammatory signaling, is thought to be a key driver of NAFLD progression [3]. Both mechanistic and RCT data have shown flavonoid-rich foods and isolated flavonoid compounds attenuate inflammation [35,36], reduce insulin resistance [36,37], and modulate lipid metabolism, with a recent meta-analysis showing reductions in triglycerides and low-density lipoprotein (LDL) levels across various RCTs examining the impact of various flavonoid compounds including anthocyanins, and hesperidin on NAFLD [34]. Indeed, the impact of flavonoids, in particular flavan-3-ols and anthocyanins, on lipid metabolism has been established in other RCTs relating to cardiometabolic disease [18, 37]. Mechanistic insight may be inferred from our negative controls. In the study, we observed an association between both tea and coffee intake and lower liver fat, but only tea intake demonstrated a significant association for cT1. The impact of coffee and, specifically, caffeine on liver fat is well established [38], thus partially explaining the shared association of both tea and coffee with liver fat. However, cT1 may provide additional accuracy in measuring hepatic inflammation [39]. As such, this finding may suggest an additional impact of flavonoids from tea consumption on inflammatory pathways in the liver.
We observed unexpected positive associations between onions, grapes, anthocyanins, and liver fat. Onions are consumed in a variety of dishes, including those high in fat, calories, and meat products, and thus, positive associations may be driven by other dietary factors associated with onion consumption. One contributor to anthocyanin intake is grapes, which have a high sugar content. Red wine is another main source of anthocyanins, and alcohol may increase steatosis risk [40]. However, we acknowledge that the observed associations were weak in magnitude, and we cannot rule out chance findings.
A key strength of our study was the integration of prospective analyses on incident NAFLD and objective imaging-derived biomarkers of liver health in addition to the large sample size and significant follow-up period. However, limitations also exist. Although we have attempted to improve the accuracy of dietary data by restricting the analysis to study participants with >2 24-h assessments, intake data for more infrequently consumed flavonoid-rich foods may lack precision. Although the FDS shows very good reproducibility over time [27], regression dilution due to a lack of repeated dietary assessments may be 1 explanation for the lack of association between the FDS and NAFLD in the last third of follow-up in our study. Nevertheless, despite excluding implausible energy intakes, we cannot rule out that food-specific recall bias and under/over-reporting may have affected the accuracy of the dietary assessments. The exact calculation of flavonoid subclass intakes was limited in our study because the dietary assessment tool combines some individual foods into groups without precise information on processing, leading to the use of mean values for some food groups (for example, berries). Additionally, Household-level data on processing and cooking was not available, and cooking methods may alter flavonoid concentrations in food. Thus, when assigning flavonoid values to foods where the cooking method was not known, the most common cooking method or raw values were used. As such, our ranking of flavonoid intakes may have been diluted due to a lack of precision. Additionally, NAFLD diagnosis in our study was based on ICD-10 hospital records, not considering primary care diagnosis. Beyond the source of diagnoses, detection bias may have affected our results in the context of the COVID-19 pandemic, with a relatively low number of NAFLD cases from 2020. The latter may be another reason for the nonproportional association between the FDS and NAFLD in our study, in addition to regression dilution. The UK Biobank is also made up of a predominantly White British population of relatively high socioeconomic status, which may limit the generalizability of our findings. Lastly, despite adjustment for a broad range of potential confounders, we cannot rule out residual confounding.
In conclusion, we demonstrate in a large United Kingdom-based prospective cohort that a flavonoid-rich diet is associated with a lower risk of NAFLD and lower imaging-derived biomarkers of NAFLD. This study adds to the growing body of literature suggesting a possible protective effect of flavonoids on NAFLD and its progression.
Author contributions
The authors’ responsibilities were as follows– WB, TK, AC: design and concept; WB, AST: data extraction and cleaning; WB, TK, AC: analyzed and interpreted data; WB, TK, AC, AJ: drafted manuscript; AJ, NPB, AT-R, AST: provided critical review of the manuscript; WB, TK, AC: guarantors of the work; and all authors: read and approved the final manuscript.
Funding
This research was conducted using UK Biobank-funded and sourced data (application 64426). The UK Biobank was established by the Wellcome Trust, the Medical Research Council, the United Kingdom Department of Health, and the Scottish Government. The UK Biobank has also received funding from the Welsh Assembly Government, the British Heart Foundation, DiABETES UK, Northwest Regional Development Agency, and the Scottish Government. In addition, WB and AST hold a PhD studentship of the Department for the Economy, Northern Ireland. The project was in part funded by the co-center for Sustainable Food Systems (funded by Science Foundation Ireland, United Kingdom Research and Innovation, and the Department of Agriculture, Environment and Rural Affairs).
Data availability
UK Biobank data requests can be made by all researchers for approved projects, including replication, through https://www.ukbiobank.ac.uk/. This research was conducted using UK Biobank-funded and sourced data (application 64426).
Conflict of interest
AC acts as an advisor to the United States Highbush Blueberry Council (with oversight from the USDA) grant committee and has received funding from them for a randomized control trial and population-based work. All other authors report no conflicts of interest.
Footnotes
Supplementary data to this article can be found online at https://doi.org/10.1016/j.ajcnut.2024.09.022.
Contributor Information
Tilman Kühn, Email: t.kuhn@qub.ac.uk.
Aedín Cassidy, Email: a.cassidy@qub.ac.uk.
Appendix A. Supplementary data
The following is the Supplementary data to this article:
References
- 1.Quek J., Chan K.E., Wong Z.Y., Tan C., Tan B., Lim W.H., et al. Global prevalence of non-alcoholic fatty liver disease and non-alcoholic steatohepatitis in the overweight and obese population: a systematic review and meta-analysis. Lancet Gastroenterol. Hepatol. 2023;8(1):20–30. doi: 10.1016/S2468-1253(22)00317-X. [DOI] [PubMed] [Google Scholar]
- 2.Riazi K., Azhari H., Charette J.H., Underwood F.E., King J.A., Afshar E.E., et al. The prevalence and incidence of NAFLD worldwide: a systematic review and meta-analysis. Lancet Gastroenterol. Hepatol. 2022;7(9):851–861. doi: 10.1016/S2468-1253(22)00165-0. [DOI] [PubMed] [Google Scholar]
- 3.Loomba R., Friedman S.L., Shulman G.I. Mechanisms and disease consequences of nonalcoholic fatty liver disease. Cell. 2021;184(10):2537–2564. doi: 10.1016/j.cell.2021.04.015. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Mundi M.S., Velapati S., Patel J., Kellogg T.A., Abu Dayyeh B.K., Hurt R.T. Evolution of NAFLD and its management. Nutr. Clin. Pract. 2020;35(1):72–84. doi: 10.1002/ncp.10449. [DOI] [PubMed] [Google Scholar]
- 5.Maurice J., Manousou P. Non-alcoholic fatty liver disease. Clin. Med. (Lond.). 2018;18(3):245–250. doi: 10.7861/clinmedicine.18-3-245. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Buzzetti E., Pinzani M., Tsochatzis E.A. The multiple-hit pathogenesis of non-alcoholic fatty liver disease (NAFLD) Metabolism. 2016;65(8):1038–1048. doi: 10.1016/j.metabol.2015.12.012. [DOI] [PubMed] [Google Scholar]
- 7.Quek J., Lim G., Lim W.H., Ng C.H., So W.Z., Toh J., et al. The association of plant-based diet with cardiovascular disease and mortality: A meta-analysis and systematic review of prospect cohort studies. Front Cardiovasc. Med. 2021;8 doi: 10.3389/fcvm.2021.756810. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.McMacken M., Shah S. A plant-based diet for the prevention and treatment of type 2 diabetes. J Geriatr. Cardiol. 2017;14(5):342–354. doi: 10.11909/j.issn.1671-5411.2017.05.009. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Wang Y., Liu B., Han H., Hu Y., Zhu L., Rimm E.B., et al. Associations between plant-based dietary patterns and risks of type 2 diabetes, cardiovascular disease, cancer, and mortality – a systematic review and meta-analysis. Nutr. J. 2023;22(1):46. doi: 10.1186/s12937-023-00877-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Cassidy A., Mukamal K.J., Liu L., Franz M., Eliassen A.H., Rimm E.B. High anthocyanin intake is associated with a reduced risk of myocardial infarction in young and middle-aged women. Circulation. 2013;127(2):188–196. doi: 10.1161/CIRCULATIONAHA.112.122408. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Hooper L., Kroon P.A., Rimm E.B., Cohn J.S., Harvey I., Le Cornu K.A., et al. Flavonoids, flavonoid-rich foods, and cardiovascular risk: a meta-analysis of randomized controlled trials. Am. J Clin. Nutr. 2008;88(1):38–50. doi: 10.1093/ajcn/88.1.38. [DOI] [PubMed] [Google Scholar]
- 12.Cassidy A., Minihane A.M. The role of metabolism (and the microbiome) in defining the clinical efficacy of dietary flavonoids. Am. J Clin. Nutr. 2017;105(1):10–22. doi: 10.3945/ajcn.116.136051. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Holden J.M., Bhagwat S.A., Haytowitz D.B., Gebhardt S.E., Dwyer J.T., Peterson J., et al. Development of a database of critically evaluated flavonoids data: application of USDA’s data quality evaluation system. J Food Compos. Anal. 2005;18(8):829–844. doi: 10.1016/j.jfca.2004.07.002. [DOI] [Google Scholar]
- 14.Neveu V., Perez-Jiménez J., Vos F., Crespy V., du Chaffaut L., Mennen L., et al. Phenol-Explorer: an online comprehensive database on polyphenol contents in foods. Database (Oxford) 2010:bap024. doi: 10.1093/database/bap024. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Bondonno N.P., Dalgaard F., Murray K., Davey R.J., Bondonno C.P., Cassidy A., et al. Higher habitual flavonoid intakes are associated with a lower incidence of diabetes. J Nutr. 2021;151(11):3533–3542. doi: 10.1093/jn/nxab269. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Dalgaard F., Bondonno N.P., Murray K., Bondonno C.P., Lewis J.R., Croft K.D., et al. Associations between habitual flavonoid intake and hospital admissions for atherosclerotic cardiovascular disease: a prospective cohort study. Lancet Planet Health. 2019;3(11):e450–e459. doi: 10.1016/S2542-5196(19)30212-8. [DOI] [PubMed] [Google Scholar]
- 17.Curtis P.J., Berends L., van der Velpen V., Jennings A., Haag L., Chandra P., et al. Blueberry anthocyanin intake attenuates the postprandial cardiometabolic effect of an energy-dense food challenge: results from a double blind, randomized controlled trial in metabolic syndrome participants. Clin. Nutr. 2022;41(1):165–176. doi: 10.1016/j.clnu.2021.11.030. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Curtis P.J., van der Velpen V., Berends L., Jennings A., Feelisch M., Umpleby A.M., et al. Blueberries improve biomarkers of cardiometabolic function in participants with metabolic syndrome-results from a 6-month, double-blind, randomized controlled trial. Am. J Clin. Nutr. 2019;109(6):1535–1545. doi: 10.1093/ajcn/nqy380. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Mastroiacovo D., Kwik-Uribe C., Grassi D., Necozione S., Raffaele A., Pistacchio L., et al. Cocoa flavanol consumption improves cognitive function, blood pressure control, and metabolic profile in elderly subjects: the Cocoa, Cognition, and Aging (CoCoA) Study--a randomized controlled trial. Am. J Clin. Nutr. 2015;101(3):538–548. doi: 10.3945/ajcn.114.092189. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Romero-Gómez M., Zelber-Sagi S., Trenell M. Treatment of NAFLD with diet, physical activity and exercise. J Hepatol. 2017;67(4):829–846. doi: 10.1016/j.jhep.2017.05.016. [DOI] [PubMed] [Google Scholar]
- 21.Petermann-Rocha F., Carrasco-Marín F., Boonpor J., Parra-Soto S., Shannon O., Malcomson F., et al. Association of five diet scores with severe NAFLD incidence: A prospective study from UK Biobank. Diabetes Obes. Metab. 2024;26(3):860–870. doi: 10.1111/dom.15378. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Yaskolka Meir A., Rinott E., Tsaban G., Zelicha H., Kaplan A., Rosen P., et al. Effect of green-Mediterranean diet on intrahepatic fat: the DIRECT PLUS randomised controlled trial. Gut. 2021;70(11):2085–2095. doi: 10.1136/gutjnl-2020-323106. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Hadefi A., Arvanitakis M., Trépo E., Zelber-Sagi S. Dietary strategies in non-alcoholic fatty liver disease patients: from evidence to daily clinical practice, a systematic review. United European Gastroenterol. J. 2023;11(7):663–689. doi: 10.1002/ueg2.12443. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Zhong Q.W., Wu Y.Y., Xiong F., Liu M., Liu Y.P., Wang C., et al. Higher flavonoid intake is associated with a lower progression risk of non-alcoholic fatty liver disease in adults: a prospective study. Br. J Nutr. 2021;125(4):460–470. doi: 10.1017/S0007114520002846. [DOI] [PubMed] [Google Scholar]
- 25.Sudlow C., Gallacher J., Allen N., Beral V., Burton P., Danesh J., et al. UK Biobank: an open access resource for identifying the causes of a wide range of complex diseases of middle and old age. PLOS Med. 2015;12(3) doi: 10.1371/journal.pmed.1001779. PMID 25826379. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Bondonno N.P., Liu Y.L., Zheng Y., Ivey K., Willett W.C., Stampfer M.J., et al. Change in habitual intakes of flavonoid-rich foods and mortality in US males and females. BMC Med. 2023;21(1):181. doi: 10.1186/s12916-023-02873-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Thompson A.S., Jennings A., Bondonno N.P., Tressera-Rimbau A., Parmenter B.H., Hill C., et al. Higher habitual intakes of flavonoids and flavonoid-rich foods are associated with a lower incidence of type 2 diabetes in the UK Biobank cohort. Nutr. Diabetes. 2024;14(1):32. doi: 10.1038/s41387-024-00288-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Wilman H.R., Kelly M., Garratt S., Matthews P.M., Milanesi M., Herlihy A., et al. Characterisation of liver fat in the UK Biobank cohort. PLOS ONE. 2017;12(2) doi: 10.1371/journal.pone.0172921. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Mojtahed A., Kelly C.J., Herlihy A.H., Kin S., Wilman H.R., McKay A., et al. Reference range of liver corrected T1 values in a population at low risk for fatty liver disease-a UK Biobank sub-study, with an appendix of interesting cases. Abdom. Radiol. (NY) 2019;44(1):72–84. doi: 10.1007/s00261-018-1701-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Carr S., Bryazka D., McLaughlin S.A., Zheng P., Bahadursingh S., Aravkin A.Y., et al. A burden of proof study on alcohol consumption and ischemic heart disease. Nat. Commun. 2024;15(1):4082. doi: 10.1038/s41467-024-47632-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.NHS. Alcohol-related liver disease, www.nhs.uk2022 [cited 19, November, 2023]. Available from: https://www.nhs.uk/conditions/alcohol-related-liver-disease-arld/.
- 32.Lipsitch M., Tchetgen Tchetgen E., Cohen T. Negative controls: a tool for detecting confounding and bias in observational studies. Epidemiology. 2010;21(3):383–388. doi: 10.1097/EDE.0b013e3181d61eeb. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.R Core Team . R Foundation for Statistical Computing; 2023. R: A language and environment for statistical computing. [Google Scholar]
- 34.Li L., Ji K., Du F., Jin N., Boesch C., Farag M.A., et al. Does flavonoid supplementation alleviate non-alcoholic fatty liver disease? A systematic review and meta-analysis of randomized controlled trials. Mol. Nutr. Food Res. 2023;67(23) doi: 10.1002/mnfr.202300480. [DOI] [PubMed] [Google Scholar]
- 35.do Rosario V.A., Fitzgerald Z., Broyd S., Paterson A., Roodenrys S., Thomas S., et al. Food anthocyanins decrease concentrations of TNF-α in older adults with mild cognitive impairment: A randomized, controlled, double blind clinical trial. Nutr. Metab. Cardiovasc. Dis. 2021;31(3):950–960. doi: 10.1016/j.numecd.2020.11.024. [DOI] [PubMed] [Google Scholar]
- 36.Muscarà C., Molonia M.S., Speciale A., Bashllari R., Cimino F., Occhiuto C., et al. Anthocyanins ameliorate palmitate-induced inflammation and insulin resistance in 3T3-L1 adipocytes. Phytother. Res. 2019;33(7):1888–1897. doi: 10.1002/ptr.6379. [DOI] [PubMed] [Google Scholar]
- 37.Curtis P.J., Sampson M., Potter J., Dhatariya K., Kroon P.A., Cassidy A. Chronic ingestion of flavan-3-ols and isoflavones improves insulin sensitivity and lipoprotein status and attenuates estimated 10-year CVD risk in medicated postmenopausal women with type 2 diabetes: a 1-year, double-blind, randomized, controlled trial. Diabetes Care. 2012;35(2):226–232. doi: 10.2337/dc11-1443. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Wijarnpreecha K., Thongprayoon C., Ungprasert P. Coffee consumption and risk of nonalcoholic fatty liver disease: a systematic review and meta-analysis. Eur. J Gastroenterol. Hepatol. 2017;29(2):e8–e12. doi: 10.1097/MEG.0000000000000776. [DOI] [PubMed] [Google Scholar]
- 39.Dennis A., Kelly M.D., Fernandes C., Mouchti S., Fallowfield J.A., Hirschfield G., et al. Correlations between MRI biomarkers PDFF and cT1 with histopathological features of non-alcoholic steatohepatitis. Front Endocrinol. (Lausanne.) 2020;11 doi: 10.3389/fendo.2020.575843. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Jeon S., Carr R. Alcohol effects on hepatic lipid metabolism, J Lipid. Res. 2020;61(4):470–479. doi: 10.1194/jlr.R119000547. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
UK Biobank data requests can be made by all researchers for approved projects, including replication, through https://www.ukbiobank.ac.uk/. This research was conducted using UK Biobank-funded and sourced data (application 64426).