Total dietary flavonoid intake and risk of cardiometabolic diseases: A dose-response meta-analysis of prospective cohort studies

Abstract Several epidemiological studies have suggested that flavonoid intake is associated with a decreased risk of cardiometabolic disease. However, the results remained inconsistent and there is no dose-response meta-analysis for specific outcomes. We conducted a meta-analysis to synthesize the knowledge about their associations and to explore their dose-response relationships. We comprehensively searched the PubMed, Embase, and Web of Science databases for prospective cohort studies published up to December 1, 2021. Summary relative risks (RR) and 95% confidence intervals (CI) were pooled for the association between flavonoid intake and cardiometabolic disease. Evaluations of linear or nonlinear dose-response were presented by restricted cubic splines. We identified 47 articles, including 1,346 676 participants and 127,507 cases in this meta-analysis. The summary of RR per 500 mg/d increase in flavonoid intake was 0.93 (95% CI 0.88–0.98) for cardiovascular disease, 0.89 (95% CI 0.84–0.94) for diabetes, and 0.97 (95% CI 0.94–0.99) for hypertension, respectively. We also found a linearity dose-response association between total flavonoid intake and cardiovascular disease (p nonlinearity = 0.541), and diabetes (p nonlinearity = 0.077). Our finding based on quantitative data suggested that a higher level of flavonoid intake is beneficial for the prevention of cardiometabolic diseases.


Introduction
A large body of epidemiological studies (Bruckdorfer 2008;Kent et al. 2020) has shown a clear relationship between a diet rich in vegetables and fruits and the decreased incidence of cardiometabolic diseases, including cardiovascular disease (CVD), diabetes, hypertension, and obesity.For example, the Mediterranean diet rich in plant foods is well known as healthy diet patterns (Sofi et al. 2008), which is closely related to reduced mortality, especially cardiovascular mortality.A common characteristic of plant-based diets is their richness in flavonoids.Flavonoid is a class of various compounds that are abundant in plants and have multiple biological effects, including antioxidant, anti-inflammatory, and free radical scavenging properties (Greenrod and Fenech 2003;Nijveldt et al. 2001;Middleton 1998).Especially, flavonoids have also played a significant role in reducing the risk of chronic noncommunicable diseases (Grosso, Godos, et al. 2017;Hooper et al. 2008).
Accumulative studies (Liu et al. 2017;Peterson et al. 2012;van Dam, Naidoo, and Landberg 2013;Guo et al. 2019;Godos et al. 2019;Marranzano et al. 2018) have evaluated the association between total flavonoids intake and the risk of cardiometabolic diseases.Some studies showed that higher daily intake of flavonoids did not reduce the risk of CVD (Hirvonen et al. 2000;Sesso et al. 2003), while other studies indicated that higher flavonoid intake had a protective effect on CVD (Arts et al. 2001;Keli et al. 1996).Therefore, the effect of flavonoid intake on the risk of cardiometabolic diseases has not yet been determined.Although many meta-analyses have been conducted on the association between flavonoid intake with CVD (Wang et al. 2014), diabetes (Liu et al. 2014), more recently published prospective cohort studies provided an opportunity to synthesize robust evidence.Furthermore, determining the does-response between flavonoid intake and outcomes was crucial for precise decision making of disease prevention.However, dose-response meta-analyses that demonstrate the association between flavonoid intake and CVD, diabetes, hypertension, and obesity are still lacking.
Therefore, we performed a comprehensive meta-analysis, including prospective cohort studies in adults (age ≥18 years) to synthesize the knowledge of the association between total flavonoid intake and cardiometabolic diseases, and to explore the dose-response relationship.

Literature search
This meta-analysis was performed according to the preferred reporting items for systematic reviews and meta-analysis (PRISMA) (Shamseer et al. 2015).We systematically searched the PubMed, Embase, and Web of Science databases until 1 December, 2021, for articles evaluating total dietary flavonoid intake and CVD, diabetes, hypertension, or obesity in humans.Additionally, the combination of MESH terms and free text terms were used to search.Details of the search terms are shown in Supplemental Table S1.To avoid missing any related studies, we also searched the reference lists of all included articles and previous systematic reviews.The International Prospective Register of Systematic Reviews (PROSPERO) registration number is CRD42021285414.

Study selection criteria
Studies were included if they (1) were prospective cohort studies; (2) participants over 18 years of age at baseline; (3) the exposure of interest was total flavonoid intake, and the outcomes of interest were CVD, diabetes, hypertension, or obesity (defined by body mass index (BMI) and/or waist circumference (WC); (4) reported relative risks (RR) with 95% confidence intervals (CI) or relevant data to calculate them; (5) to perform the dose-response meta-analysis, at least three levels of total flavonoid intake, the number of cases in each category/level, and the number of person-year/ participant or sufficient data for deriving these data were provided.Studies were also included if they only reported the corresponding increased RR values per increased amount of flavonoid intake; (6) without language restriction.Exclusion criteria included: (1) only reporting one subclass of flavonoids; (2) reviews, case reports, letters, and conference papers.If multiple studies were reported in the same cohort, the most recent was included with informative reporting of total flavonoid intake and/or the largest sample.Two authors (TL and HH) independently conducted the search by screening the title and abstract.A full text review was also evaluated by the two authors, and a third reviewer discussed any potential disagreement (JL).

Data extraction and quality assessment
Two authors (TL and LY) independently extracted the following information: the first author, the name of cohort studies, publication year, country, follow-up years, sample size, interested outcomes, participant characteristics at baseline (sex, mean or median age), measurement methods of total flavonoid intake, unit of total flavonoid intake, number of incident cases and participants for each flavonoid category, and confounding variables and RRs with 95% CI adjusted by the most confounders for each flavonoid category.The quality of inclusion publications was evaluated using the Newcastle Ottawa Scale (NOS) (Stang 2010), which consists of four aspects for study selection, three aspects to evaluate the outcome, and one aspect for comparability.We defined the NOS scores of 0-3, 4-6, and 7-9 as poor, general, and high quality, respectively.Any disagreements regarding the data extraction was resolved by consensus with the third author (JL).

Data synthesis
For cohort studies reporting hazard ratios (HR), we assumed that HR were approximately equal to RR (Orsini et al. 2012).If the case number in each category was missing, the total cases and RR were used to infer these data.If the number of person-year or participants was missing for each category, the categories were regarded as equal size (Bekkering et al. 2008).Furthermore, we assign the median or mean value in each category to the corresponding RR (Bekkering et al. 2008).If the median or mean total flavonoid per category was not available, the midpoint between the upper limit and the lower limit was used to replace it (Bekkering et al. 2008).Whenever the categories were open ended, we assumed that the width was the same as the preceding range (Tamakoshi et al. 2011).
The fixed effects model was used to calculate special article RR if one article reported different subgroups or different types of flavonoid intake data, respectively (Begg and Mazumdar 1994).In addition, we examined possible linear associations by modeling flavonoids intake levels with a restricted cubic spline, which chose three knots located at the 25th, 50th, and 75th percentiles of the distribution (Greenland 1995).Meanwhile, the least-squares estimate was used to evaluate the dose-response relationship.The p value for non-linearity (p non-linearity ) was calculated by testing the null hypothesis that the coefficient of the second spline is equal to 0 (Orsini et al. 2012).
Heterogeneity was tested by Cochran Q (the corresponding p heterogeneity value was calculated and presented) and I 2 statistics (Higgins et al. 2003).For the Q statistic, p heterogeneity < 0.1 was considered statistically significant, and I 2 values of approximately 25%, 50% and 75% were considered low, moderate, and high heterogeneity, respectively.
We used the fixed-effects model to pool study-specific RR (95% CI) and RR for I 2 values < 50% (Begg and Mazumdar 1994).On the contrary, if the I 2 values ≥ 50%, we chose the model affected by random to pooled RRs (DerSimonian and Laird 2015).
Subgroup analyses were carried out by sex, age (mean or median age > = 60 vs <60 years), region, follow-up years (mean or median > =10 years vs <10 years), sample size (> =10,000 vs <10,000), study quality (high vs non-high), and adjusted factors (e.g.BMI [kg/m 2 ], energy intake, smoking, and drinking).For some studies that do not report the median or mean age of the total participants, we chose the midpoint between upper limit and lower limit of the range as median age to conduct subgroup analyses according to the conventional method in the meta-analyses (Bekkering et al. 2008).We performed a meta-regression analysis to calculate the p-value for heterogeneity between groups (Santiago de Araújo Pio et al. 2017).Sensitivity analyses were further performed by omitting one study at a time to assess the stability of the results and explore potential sources of heterogeneity.The potential publication bias was evaluated using the Egger linear regression test (Begg and Mazumdar 1994).The two-sided p value < 0.05 indicated possible publication bias, and trim and fill methods were used to correct publication bias.All analyses involved the use of Stata 16 (Stata Corp, College Station, TX).

The characteristics of included studies
A total of 2,839 relevant records were identified in the literature search (2,835 from the computer database and four from the reference list).Because of duplication, 221 citations were excluded, 1,884 were excluded based on title and abstract, and 650 were excluded by record selection.Finally, eighty-four articles were left for full-text search.Among them, we also excluded 32 articles including reviews and meta-analysis (n = 4), exposure and outcome not interested (n = 21), not provided indispensable data (n = 5) and not available full text (n = 2).Finally, we included 52 articles (57 prospective cohort studies) in this meta-analysis (Figure 1).
When RR was merged for the highest versus lowest, all studies were included in the analysis.The summary RR was 0.88 (95% CI 0.82-0.94,I 2 = 61.7%,p heterogeneity = 0.003, Figure 3).We found no evidence of statistical publication bias using the Egger test (p = 0.239) or funnel plot (Figure S1B).With sensitivity analysis that removed one study at a time, none of the individual studies significantly changed the pooled risk.
The main summary RR did not change significantly by the sensitivity analysis.The Egger test (p = 0.111) and funnel plot (Figure S1E) did not show statistical publication bias.A negative linearity dose-response association was found between flavonoid intake and diabetes based on the pooled estimation of 11 studies (p non-linearity < 0.077, Figure 4B).

The risk of total dietary flavonoids intake and hypertension
For hypertension, eight (Lajous et al. 2016;do Rosario et al. 2021;Cassidy et al. 2011;Grosso et al. 2018) prospective cohort studies were included in the meta-analysis, including 212,985 participants and 47,720 incident cases with a mean follow-up period of 12.0 years.The countries where the studies were presented were USA (n = 3) (Cassidy et al. 2011), Europe (n = 3) (Lajous et al. 2016;Grosso et al. 2018) and Australia (n = 2) (do Rosario et al. 2021).The mean NOS score for quality assessment was 7.7.
All the eight studies were included for the dose-response analysis.We found a linear association between flavonoid intake and hypertension (p non-linearity = 0.300, Figure 4C).Compared to non-flavonoid consumption, the risk of incident hypertension was reduced 3% (RR 0.97, 95% CI 0.94-0.99)at 500 mg/d intake, with a high heterogeneity (I 2 = 60.6%, p heterogeneity = 0.013).We did not find that an increase in 100 mg/d and 300 mg/d of flavonoid consumption was beneficial in reducing the risk of hypertension (Table 3).
The main summary RR did not change significantly by sensitivity analysis (RR 0.97 95% CI 0.94-0.99).No significant statistical publication bias was detected by Egger's test (p = 0.136) and funnel plot (Figure S1F).

The risk of total dietary flavonoids intake and obesity
Two studies including 8,353 participants and 1,600 incident cases were included for the association between total dietary flavonoids intake and obesity in this meta-analysis.We only pooled a summary RR for the highest versus lowest categories of flavonoid consumption due to the limited number of studies.When comparing the highest versus lowest flavonoid consumption, the pooled RR was 0.47 (95% CI 0.32-0.69),with a low heterogeneity (I 2 = 0.0%, p heterogeneity = 0.329, Figure 3C).Sensitivity analysis gave a similar result.

Subgroup and meta-regression analyses
To explore the source of heterogeneity, subgroup analyses were performed by outcome, sex, age (mean or median age), region, follow-up years, NOS score, sample size and possible confounders (energy intake, BMI, physical activity, and family history).For the CVD subgroup (Table 2), the association was consistent in most subgroups, except in the sex subgroup.We found significant changes in the heterogeneity of the sex subgroup, which indicated that the different sexes may be the source of the heterogeneity.For the outcome of diabetes (Table 4), the negative association is not significant in the age subgroups (mean or median) ≥60 years, the subgroup of energy intake does not adjust, and the subgroup of physical activity does not adjust due to the limitation of the small sample.We did not conduct subgroup analyses for the outcomes of hypertension and obesity due to insufficient studies.

Discussion
In the present meta-analysis, we found a negative association between total flavonoid intake and cardiometabolic diseases.Compared to the lowest flavonoid consumption category, the risk of the highest category had a 15%, 12%, 4%, and 53% reduction in CVD, diabetes, hypertension, and obesity, respectively.With per 500 mg/d increment, the risk was reduced by 7%, 11%, and 3% for CVD, diabetes, and hypertension, respectively.For CVD and diabetes, a linear association was found with the consumption of total flavonoids.
The results were consistent with another meta-analysis (Wang et al. 2014), which also found an inverse association between flavonoid intake and CVD.Compared to the previous meta-analysis that included fourteen cohort studies, our dose-response diagram was drawn using restricted cubic spline methods.Furthermore, our meta-analysis included more original studies and conducted a more comprehensive subgroup analysis, increasing the precision and reliability of our results.Compared to a previous meta-analysis (Guo et al. 2019), we also included two prospective cohort studies to explore the association between total flavonoid intake and diabetes in our meta-analysis.Our results found an inverse linearity of total flavonoids intake and diabetes, which is consistent with previous meta-analysis.To our knowledge, our meta-analysis was the first dose-response meta-analysis to assess the association between total flavonoids intake and hypertension or obesity based on cohort studies.Although no dose-response association was found limited by the number of prospective cohort studies, the results of the highest versus lowest analysis showed that a higher consumption of food rich in flavonoids was beneficial to reduce the risk of hypertension and obesity.
Flavonoids reduce the risk of CVD and hypertension through various pathways (Perez-Vizcaino and Duarte   2010).First, flavonoids have a non-endothelium-dependent vasodilation effect in vitro.Under oxidative stress, it can improve nitric oxide bioavailability and bioactivity by inducing down-regulation of NO synthase and improve vascular parameters, including antiplatelet aggregation and inhibition of LDL oxidation (Kim et al. 1999).In addition, there is inconclusive evidence (Simonyi et al. 2005) that quercetin, a subgroup of flavonoids, has antihypertensive and antiatheroscierotic effects and can protect against myocardial ischemic injury.Furthermore, many acute cardiovascular events are caused by plaque damage and rupture, which is mediated by acute phase proteins (C-reactive protein, CRP) and cytokines (IL-4, IL-6).Flavonoids protect the endothelium by preventing excessive inflammatory stimulation.The mechanism includes inhibition of pro-inflammatory enzymes (cycde-2), lipoxygenase, inducible NO synthase, NF-KB, and activator protein-1 (AP-1) (Yoon and Baek 2005;Middleton, Kandaswami, and Theoharides 2000).Secondly, for diabetes, flavonoids can improve glucose homeostasis by reducing blood glucose concentration and improving liver insulin sensitivity (Kwon et al. 2011;Sasaki et al. 2007).Furthermore, lignin, phenolic, and stilbene compounds can reduce the activity of key enzymes associated with glucose and HbA1c levels, suppress oxidative stress, and finally reduce the risk of diabetic events (Ly et al. 2015).Evidence from randomized controlled trials and cohort studies suggested that ingestion of foods rich in flavonoids, including green tea, chocolate, and coco, can improve body fat distribution (Nagao, Hase, and Tokimitsu 2007;Kord-Varkaneh et al. 2019).Thus, the protect effects of flavonoid intake on diabetes may be achieved in part by achieving far-response metabolism in the body (Vinayagam and Xu 2015).Total flavonoid intake is affected by many intrinsic factors, such as geographical area, gender, population age characteristics, and different eating habits due to socio-cultural factors.The results of the EPIC study also suggested that the variability in flavonoid consumption is dependent on dietary differences in non-Mediterranean countries (Knaze et al. 2012).For example, Poland (Ilow et al. 2012) and Australia (Hanna, O'Neill, and Lyons-Wall 2010) have the highest flavonoids consumption (about 600 mg/d), followed by the USA (Sebastian et al. 2015) and South America (Nascimento-Souza et al. 2018; about mean 200 and 400 mg/d, respective).Asia, including China (Zhang et al. 2010) and South Korea (Kim et al. 2015), have the lowest intake of flavonoids (approximately 60 mg/d).Foods rich in flavonoids included mainly tea, coffee, red wine, fruits, and vegetables.In North and Central Europe, tea and coffee are the main flavonoid contributors, while in Southern Europe, the main sources of flavonoid are fruits and alcohol beverages, such as wine.In the USA, tea, citrus, and legumes are the main contributors to flavonoids.In Asia, apples and vegetables seem to be the main source of flavonoid in China and Korea, while green tea in Japan.Our meta-analysis has several strengths.First, we include the original studies reporting total flavonoids intake as exposure rather than a single subclass of flavonoids.The main sources of dietary flavonoid intake are tea, onions, apples, and red wine (Lin et al. 2016).These foods contain comprehensive flavonoids, not a single subclass, suggesting that the study on the relationship between total flavonoids intake and adverse health outcomes is more meaningful in guiding dietary habits.When the effects of the subgroups are combined, the results may be affected by the limitation of the small number of original studies.In our studies, the inclusion of several large sample studies reduced the risk.Second, compared to previous meta-analysis, our studies included more recent publications, up to 35 studies were included in the CVD analysis.This is also the first meta-analysis to assess the association between total flavonoids intake and hypertension and obesity based on the cohort studies.Third, several participants in this meta-analysis provided sufficient statistical power to assess the association of flavonoids intake and the risk of CVD, diabetes, hypertension, and obesity.Some limitations of our meta-analysis should be considered.First, the consumption of flavonoids depends on the type of food contained in the questionnaire, its frequency of intake, as well as the corresponding flavonoids content of each food in the food ingredient database.In most studies the food frequency questionnaire (FFQ) measured total flavonoids intake exposures, so the misclassification of exposures and recalled bias were inevitable.Some sources of flavonoids, rich in flavonoids but not often consumed, were not included in the questionnaire, which may lead to an underestimation of the intake of flavonoids.Furthermore, people with high consumption of flavonoids tend to eat more plants foods.Sufficient antioxidant vitamins, unsaturated fatty acids (Knekt et al. 2004;Rimm et al. 1993;Pietinen et al. 1996), and dietary fiber also contribute to the decrease in the risk of unhealthy outcomes.Therefore, we could not exclude the potential confounding bias due to some original studies without adjusting for dietary factors.Moreover, a part of our original study data comes from different food ingredient databases.It is worth noting that the database does not consider nonextractable polyphenols, 101 which may underestimate the intake of flavonoids.Finally, due to the limited number of studies, we are unable to conduct further research on the relationship between total flavonoids and obesity.

Conclusions
Our meta-analysis based on quantitative data suggested there is an inverse association between total flavonoids intake and unhealthy outcomes, including CVD, diabetes, hypertension, and obesity.Overall, the inverse linear trends show that a higher level of total flavonoids ingestion is beneficial for preventing these unhealthy outcomes compared to the absence of flavonoids.

Figure 1 .
Figure 1.Flow chart of study selection.

Figure 2 .
Figure 2. Forest plot of the highest versus lowest total flavonoids intake for Cvd risk.

Figure 4 .
Figure 4. dose-response associations between total flavonoids intake and the risk of Cvd (a), diabetes (B) and hypertension (C).

Figure 3 .
Figure 3. Forest plot of the highest versus lowest total flavonoids intake for risk of diabetes (a), hypertension (B), and obesity (C).

Table 1 .
summary of the characteristics of included studies in the meta-analysis.
t. Li et aL.

Table 2 .
subgroup analyses of the association between the highest versus lowest category of total flavonoid intake and Cvd.

Table 3 .
does-response analyses on association between total flavonoids intake and the risks of cardiometabolic diseases.

Table 4 .
subgroup analyses of the association between the highest versus lowest category of total flavonoid intake and diabetes.
BMI: body mass index, p a : p value for heterogeneity within each subgroup, RR: relative risk.