CVD is the leading cause of death worldwide( Reference Mozaffarian, Benjamin and Go 1 ). Abnormal levels of lipids in the blood are risk factors for CVD. High levels of LDL-cholesterol and TAG and low levels of HDL-cholesterol lead to atherosclerosis and stroke, increasing the risk of CVD( 2 , Reference Hokanson and Austin 3 ). Metabolic risk factors such as large waist circumference, high blood pressure, high TAG, high fasting blood glucose and low HDL-cholesterol have been used to assess CVD risk( Reference Grundy, Brewer and Cleeman 4 ). Other risk factors include overweight and obesity, elevated LDL-cholesterol and insulin resistance( Reference Resnick, Jones and Ruotolo 5 , Reference Wilson, D’Agostino and Sullivan 6 ).
Diet is a modifiable risk factor for reducing the risk of CVD. Greater consumption of fruits and vegetables is associated with reduced risk of CVD( Reference Bhupathiraju, Wedick and Pan 7 ), and flavonoids found in these foods may contribute to this risk reduction( Reference Hjartaker, Knudsen and Tretli 8 ). Flavonoids are polyphenolic phytochemicals commonly found in fruits, vegetables, herbs and teas( Reference Spencer 9 ). There is evidence that flavonoids may reduce the risk of CVD by inhibiting LDL oxidation( Reference O’Reilly, Sanders and Wiseman 10 ), reducing endothelial wall damage and preventing atherosclerosis( Reference Bolduc, Baraghis and Duquette 11 , Reference Siow and Mann 12 ).
Although dietary flavonoids have been reported to reduce the risk of CVD( Reference Ivey, Lewis and Prince 13 , Reference Wang, Ouyang and Liu 14 ), findings from studies on the effects of specific flavonoid compounds or flavonoid-rich foods are still inconsistent( Reference Song, Manson and Buring 15 , Reference Sacks 16 ). Hooper et al.( Reference Hooper, Kroon and Rimm 17 ) provided a comprehensive review of 133 randomised-controlled trials, which confirmed significant heterogeneity by different effects among flavonoid subclasses or foods. Furthermore, there have been no validated investigative tools or flavonoid food composition tables available for estimating flavonoid intake in observational studies. As the US Department of Agriculture (USDA) has released an updated flavonoid database( 18 – 20 ), several studies have been conducted to estimate flavonoid intake based on this database( Reference Cassidy, O’Reilly and Kay 21 , Reference Wedick, Pan and Cassidy 22 ). One study observed an inverse association between intake of anthocyanins and anthocyanin-rich foods and incidence of type 2 diabetes( Reference Wedick, Pan and Cassidy 22 , Reference Zamora-Ros, Forouhi and Sharp 23 ). Another study reported that the higher intake of anthocyanins was associated with reduction in risk of hypertension( Reference Cassidy, O’Reilly and Kay 21 ). Several epidemiological studies have focused on flavonoid intake and CVD events or mortality( Reference Mink, Scrafford and Barraj 24 – Reference Ivey, Hodgson and Croft 27 ). However, studies on the association of flavonoid intake with comprehensive CVD risk factors including blood lipid profile, blood pressure and anthropometric measures are lacking. Therefore, in this study, we aimed to investigate the association of dietary flavonoid intake with CVD risk factors among US adults by utilising the National Health and Nutrition Examination Survey (NHANES) 2007–2012.
Methods
Study population
This study utilised data from 4042 US adults aged 19 years and older from the NHANES 2007–2012( 28 – 30 ). Exclusion criteria included the following: subjects who reported fasting for <8 h (n 9859), pregnant or breast-feeding women (n 112), those with dietary recalls coded as unreliable or incomplete (n 1086), those whose dietary recalls were coded as ‘much more than usual’ or ‘much less than usual’ or those who answered yes to ‘Are you currently on any kind of diet, either to lose weight or for some other health-related reason?’, because these might affect biomarkers of interest (n 2256).
Estimation of dietary flavonoid intake
This study used databases on flavonoid, isoflavone and proanthocyanidin contents of US foods: the USDA Database for the Flavonoid Content of Selected Foods, version 3.1( 18 ), containing values for 506 food items of twenty-six dietary flavonoid compounds classified by flavonoid subclasses such as flavonols, flavones, flavanones, flavan-3-ols and anthocyanidins; the USDA Database for the Isoflavone Content of Selected Foods, version 2.0( 19 ), containing values for individual isoflavone compounds for 557 foods; and the USDA Database for the Proanthocyanidin Content of Selected Foods, containing values for proanthocyanidins for 205 food items( 20 ). These three databases were combined to a single database: flavonols (isorhamnetin, kaempferol, myricetin, quercetin), flavones (apigenin, luteolin), flavanones (eriodictyol, hesperetin, naringenin), flavan-3-ols (flavan-3-ol monomers ((+)-catechin, (+)-gallocatechin, (−)-epicatechin, (−)-epigallocatechin, (−)-epicatechin 3-gallate, (−)-epigallocatechin 3-gallate); flavan-3-ol derived compounds (theaflavin, theaflavin-3-gallate, theaflavin 3′-gallate, theaflavin 3,3′-digallate, thearubigins); proanthocyanidins (dimers, trimers, 4-6mers, 7-10mers and polymers)), anthocyanidins (cyanidin, delphinidin, malvidin, pelargonidin, peonidin, petunidin) and isoflavones (daidzein, genistein, glycitein). Flavan-3-ols intake was estimated by summing three subclasses of flavan-3-ol monomers, flavan-3-ol derived compounds and proanthocyanidins (dimer to polymers). As the analytical values were available for cooked or processed foods in the USDA flavonoid database, retention factors for processed or cooked foods based on the Phenol-Explorer database were applied only for expansion process of the flavonoid database( Reference Rothwell, Perez-Jimenez and Neveu 31 ). The flavonoid contents in USDA databases are expressed as aglycone. The flavonoid and isoflavone databases were expanded to include additional foods as described in other publications( Reference Chun, Chung and Song 32 ). This significantly improved the coverage of the flavonoid database by increasing the proportion of major food sources having flavonoid composition data from 36 to 67 %( Reference Kim, Vance and Chun 33 ). Average daily food intake was calculated from 2-d 24-h dietary recall data in the NHANES 2007–2012. Dietary flavonoid intake was estimated by combining the flavonoid, isoflavone and proanthocyanidin databases with the food consumption data of the NHANES 2007–2012. Total flavonoid intake was determined by summing the daily intake of individual flavonoid compounds.
CVD risk factors
Waist circumference, height, weight and blood pressure were measured in the mobile examination centre( 34 ). BMI values were calculated using measured height and weight values (kg/m2). Serum total cholesterol (TC), HDL-cholesterol, TAG, fasting plasma glucose and insulin were measured as described in the NHANES Laboratory Procedures Manual( 34 ). LDL-cholesterol was calculated by the following equation: LDL=TC−HDL−0·2×TAG( 34 ). Homoeostasis model assessment for insulin resistance (HOMA-IR) is a method used to quantify insulin resistance and was calculated as (fasting serum glucose (mg/dl)×insulin (μU/ml))/405( Reference Matthews, Hosker and Rudenski 35 ). High ratios of TAG:HDL-cholesterol and TC:HDL-cholesterol have been reported as better predictors of CVD risk than changes in their absolute levels( Reference Lemieux, Lamarche and Couillard 36 , Reference da Luz, Favarato and Faria-Neto 37 ).
Statistical analysis
All the statistical analyses were performed with SAS, version 9.4 (SAS Institute Inc.), using SAS survey procedures and the appropriate weight, strata, domain and cluster variables to account for the complex survey design. Dietary flavonoids were log-transformed and adjusted for average energy intake using the residual method( Reference Willett, Howe and Kushi 38 ).
Participants were grouped into quartiles based on flavonoid intake, and the means of CVD risk factors and proportion of subjects by socio-demographic and lifestyle variables were calculated across quartiles of flavonoid intake. For descriptive statistics, poverty income ratio (PIR) was classified as ≤1·3 and >1·3. Physical activity was expressed as metabolic equivalence of tasks (MET) based on weekly minutes of walking/bicycling and moderate/vigorous recreational activities by multiplying the number of days per week by the average minutes of activities on a typical day( Reference Ainsworth, Haskell and Herrmann 39 ). MET-min/week were determined by multiplying weekly minutes of activities by the assigned MET values. Subjects who reported no walking/bicycling or moderate/vigorous recreational activities were defined as inactive. Alcohol consumption was defined based on the number of drinks of any type of alcoholic beverage per day, with consumption of no drinks as none, no more than 2 drinks/d for men and no more than 1 drink/d for women as moderate and more than two drinks for men and more than one drink for women as high intake( Reference Krauss, Eckel and Howard 40 ). Positive smoking status was defined as smoking 100 cigarettes/year, with current smokers defined as those who had not quit and former smokers as those who had reported quitting by the time of the interview.
In regression models, after inspecting residual plots, all CVD risk factors were log-transformed. A simplified representation of the model for a given CVD risk factor and flavonoid can be described by the following regression equation:
$$\log _{{\rm e}} \,({\rm CVD}\,{\rm risk}\,{\rm factor})\,{\equals}\, \beta _{0} {\plus}\log _{{\rm e}} \,{\rm flavonoid}\,\beta _{1} .$$
As the model was fit with both the predictor and the outcome on the logarithmic scale, the slope from the regression model above was used to calculate the % change in CVD risk factor for a 100 % increase in flavonoid intake (the choice of percentage is arbitrary):
$$\,\%\,\,{\rm Change}\,{\rm in}\,{\rm CVD}\,{\rm risk}\,{\rm factor}\,{\equals}\,(2^{{\beta _{1} }} {\minus}1){\times}100.$$
To determine both statistical significance and precision, the 95 % CI of the % change determined above was calculated using the standard error of β
1 (95 % CI=% change±(
${\rm 2}^{{1 \cdot 96{\times}{\rm SE}_{{\beta _{{\rm 1}} }} }} {\minus}{\rm 1}$
). Multivariate models were adjusted for the following variables: age, sex, ethnicity, PIR, alcohol consumption, smoking status, physical activity, educational level, BMI, SFA, fibre and vitamin C intakes, and blood pressure medication and insulin use. A multivariate model of BMI was adjusted for all variables except BMI. All P-values reported are two sided (α=0·05).
Results
The socio-demographic and lifestyle characteristics of study participants by quartiles of total flavonoid intake from the NHANES 2007–2012 are shown in Table 1. Age, PIR and education level were positively associated and smoking status was inversely associated with total flavonoid intake. Women in the lowest quartile of flavonoid intake had higher waist circumference. Subjects in the lowest quartile of flavonoid intake had higher fasting glucose, insulin and HOMA-IR and lower HDL-cholesterol than those in the higher quartile of flavonoid intake (Table 2). Subjects with higher flavonoid intake had lower TAG:HDL-cholesterol ratio and TAG levels than those with lower flavonoid intake.
Table 1 Socio-demographic and lifestyle characteristics by quartiles of total dietary flavonoid intake among US adults in the NHANES 2007–2012 (Ranges and mean; numbers and percentages; n 4042)
PIR, poverty income ratio; MET, metabolic equivalence of task.
* Percentage is weighted percentage considering the complex sampling design in the National Health and Nutrition Examination Survey.
† Alcohol consumption: no consumption of any type of alcoholic beverage per day was defined as none, no more than two drinks for men and no more than one drink for women as moderate and more than two drinks for men and more than one drink for women as high intake.
‡ Current smoking: former meant to have smoked at least 100 cigarettes in their entire life but do not smoke cigarettes now. Current meant to have smoked at least 100 cigarettes in their entire life and still smoke.
§ Physical activity: inactive meant not walking/bicycling or performing moderate/vigorous recreational activities for at least 10 min continuously in a typical week.
|| Blood pressure medication: yes meant taking prescribed medicine for high blood pressure.
¶ Insulin use: yes meant taking insulin or medication to control blood glucose levels.
Table 2 CVD risk factors according to the quartile (Q) of dietary flavonoid intake among US adults in the National Health and Nutrition Examination Survey (NHANES) 2007–2012 (Ranges and mean; mean values with their standard errors; n 4042)
TC, total cholesterol; HOMA-IR, homoeostasis model assessment for insulin resistance.
* Test for linearity of the trend was carried out after adjusting for age, sex, ethnicity, physical activity, poverty income ratio, smoking status, alcohol consumption, education level, BMI, blood pressure medication and insulin use, and vitamin C, SFA and fibre intakes (model of BMI was adjusted for all variables except BMI).
Serum TAG and TAG:HDL-cholesterol ratio were inversely associated with total dietary flavonoids after adjusting for age, sex, ethnicity, physical activity, PIR, smoking status, alcohol consumption, education level, BMI, SFA, fibre and vitamin C intakes, and blood pressure medication and insulin use (Table 3). The % changes in TAG for a 100 % increase in anthocyanidins and total flavonoid intakes were −1·25 % (95 % CI −2·44, −0·04) and −1·31 % (95 % CI −2·34, −0·26), respectively. TAG:HDL-cholesterol ratio was inversely associated with anthocyanidins (−1·60 % change; 95 % CI −3·12, −0·04) and total flavonoid intake (−1·83 % change; 95 % CI −3·03, −0·62). Increased HDL-cholesterol was associated with higher total dietary flavonoid intake (0·54 % change; 95 % CI 0·14, 0·94). Serum insulin and HOMA-IR were inversely associated with flavone (for insulin, −3·18 % change; 95 % CI −5·85, −0·44; for HOMA-IR, −3·10 % change; 95 % CI −5·93, −0·19) and isoflavone intakes (for insulin, −3·11 % change; 95 % CI −5·46, −0·70; for HOMA-IR, −4·01 % change; 95 % CI −6·67, −1·27). We observed that the % changes in TAG and TAG:HDL-cholesterol ratio for a 100 % increase in flavone intake were −2·15 % (95 % CI −4·70, 0·47) and −2·62 % (95 % CI −5·80, 0·66), respectively. However, they were not statistically significant. BMI was found to be negatively associated with anthocyanidin intake (−0·60 % change; 95 % CI −1·03, −0·16) after adjusting for all variables except for BMI.
Table 3 Association between flavonoid intake and CVD risk factors among US adults in the National Health and Nutrition Examination Survey (NHANES) 2007–2012 (Percentages and 95 % confidence intervals; n 4042)Footnote †
TC, total cholesterol; HOMA-IR, homoeostasis model assessment for insulin resistance. *P<0·05.
† Multivariate linear regression analysis of cardiovascular risk factors. Values are changes in percentages of cardiovascular risk factors with a 100 % increase in flavonoid intake. Models were adjusted for age, sex, ethnicity, physical activity, poverty income ratio, smoking status, alcohol consumption, education level, BMI, blood pressure medication, insulin use, vitamin C, SFA and fibre intakes.
‡ Flavan-3-ol intake was estimated by the sum of intakes of flavan-3-ol monomers, flavan-3-ol derived compounds and proanthocyanidins (dimer to polymers).
§ Multivariate model of BMI was adjusted for all variables except BMI.
Discussion
In spite of accumulating evidence that dietary flavonoids have effects on improving CVD risk factors in experimental studies( Reference Panchal, Poudyal and Brown 41 – Reference Bornhoeft, Castaneda and Nemoseck 43 ), a few observational studies have reported the association between flavonoid intake and CVD risk factors( Reference Cassidy, O’Reilly and Kay 21 , Reference Wedick, Pan and Cassidy 22 ). Furthermore, even though studies have reported the protective effects of flavonoid-rich foods such as red wine, tea, chocolate, cocoa and soya products on CVD risk( Reference Goodman-Gruen and Kritz-Silverstein 44 , Reference Shrime, Bauer and McDonald 45 ), it is still not clear whether the observed health benefits are attributed to flavonoids themselves rather than other ingredients. Established flavonoid composition databases for foods are essential for the reliable estimation of flavonoid intake and studies on flavonoids and disease prevention. Recently, two cross-sectional studies have documented the associations of higher consumption of flavonoids with improved metabolic syndrome or lipid profile by utilising flavonoid data from the Phenol-Explorer database or published composition database for estimation of flavonoid intake in the Chinese population( Reference Li, Zhu and Zhang 46 , Reference Sohrab, Hosseinpour-Niazi and Hejazi 47 ). Our research group has developed an investigative research tool by which we could expand the number of foods covered by the USDA flavonoid database( Reference Chun, Chung and Song 32 ), and using these data from the NHANES 1999–2002 we documented an inverse association between dietary flavonoid intakes and serum C-reactive protein concentrations in US adults( Reference Chun, Chung and Claycombe 48 ).
In this cross-sectional investigation using NHANES 2007–2012 data, we found some associations between flavonoid intake and CVD risk factors. BMI was inversely associated with greater consumption of anthocyanidins. Although epidemiological studies on the effect of anthocyanidins on obesity are scarce, a few animal studies showed that anthocyanidins have a significant advantage for preventing obesity by improving adipocyte dysfunction( Reference Tsuda 49 ). For HDL-cholesterol, a positive association was observed with total flavonoid intake, which is supported by a study that reported flavonoids-rich cocoa powder and orange juice increased HDL-cholesterol in human intervention trials( Reference Kurowska, Spence and Jordan 50 , Reference Mursu, Voutilainen and Nurmi 51 ). We found that insulin and HOMA-IR were negatively associated with flavone intake, which may be explained by the fact that flavone reduced insulin resistance and ameliorated insulin resistance-related endothelial dysfunction by blocking inhibitor of nuclear factor κ-B kinase β/NF-κB activation, leading to the down-regulation of TNF-α and IL-6 gene expressions( Reference Deqiu, Kang and Jiali 52 , Reference Jennings, Welch and Spector 53 ). Insulin and HOMA-IR were also inversely associated with isoflavone intakes, which is in accordance with previous studies that reported isoflavones favourably altered insulin resistance( Reference Shi, Ryan and Jones 54 , Reference Jayagopal, Albertazzi and Kilpatrick 55 ). Some studies have demonstrated that isoflavones have anti-diabetic effects mediated by increased β cell proliferation, reduced apoptosis and glucose-stimulated insulin release( Reference Gilbert and Liu 56 ). However, as other human studies have reported that isoflavones showed no significantly beneficial effects on CVD risk factors( Reference Nikander, Tiitinen and Laitinen 57 , Reference Liu, Ho and Chen 58 ), further research is warranted.
TAG and TAG:HDL-cholesterol ratio were inversely associated with anthocyanidin and total dietary flavonoid intakes. These are consistent with the results from previous experimental studies that showed higher flavonoid consumption decreased TAG and TAG:HDL-cholesterol ratio( Reference Pfeuffer, Auinger and Bley 59 , Reference Cho, Kim and Andrade 60 ). These findings are also supported by the report that showed supplementation of anthocyanin or anthocyanin-rich foods reduced TAG in human intervention trials( Reference Asgary, Kelishadi and Rafieian-Kopaei 61 , Reference Li, Zhang and Liu 62 ). Blood TAG:HDL-cholesterol ratio has been identified to be a strong predictor for extensive CHD among subjects at high risk for the development of coronary disease( Reference da Luz, Favarato and Faria-Neto 37 ), cardiometabolic risk( Reference Weiler Miralles, Wollinger and Marin 63 ) and major adverse cardiovascular events among patients with acute coronary syndrome( Reference Wan, Zhao and Huang 64 ). Therefore, these results indicate that flavonoid intake may lower CVD risk by improving atherogenic blood lipid profile.
In Table 3, the changes in percentages of cardiovascular risk factors with a 100 % increase in flavonoid intake are presented. For example, a −3·18 % change in the association of flavones with insulin means that an insulin level of 80 pmol/l might be decreased by 2·5 pmol/l (3·18 %) if 1·2 mg/d average intake of flavones would be doubled to 2·4 mg/d.
This study has several strengths. First, we used a relatively large sample of the US population. Second, we used a modified flavonoid database in an effort to reduce the incompleteness of the database and provided better estimates of dietary flavonoid intakes( Reference Kim, Vance and Chun 33 ). However, this study also has several limitations. First, this study was based on cross-sectional data, which only allows showing statistical associations and cannot make causal inference. Second, we did not consider the bioavailability or metabolism of flavonoids. Third, flavonoid intake may have been underestimated because of the limited food composition data and the exclusion criteria for flavonoid intake used in this study. Fourth, the estimation of flavonoid intake was based on 2-d 24-h dietary recalls, which are limited by within-person variability and recall error. Fifth, the values of thearubigins in the USDA database are crude approximations using an indirect method. Sixth, there may still have been residual confounding, although we adjusted available confounding factors. As we tested a set of statistical inferences simultaneously, some significant results might be false positives due to chance.
In conclusion, higher dietary flavonoid intake was associated with improved blood lipid profile. Our findings may support the beneficial effects of dietary flavonoids on lowering CVD risk. However, further research is warranted to confirm the findings from this study as these associations were moderate in strength.
Acknowledgements
The authors thank Dr Sang Jin Chung, Professor of the Department of Food and Nutrition in Kookmin University, for her statistical consultation on this project. This research did not receive financial support from any funding agency, commercial or not-for-profit sectors.
K. K., T. M. V and O. K. C. designed the research; K. K. drafted the manuscript and analysed the data; K. K. and T. M. V. performed statistical analyses; K. K., T. M. V. and O. K. C. contributed to the interpretation of the results and critically reviewed the manuscript. All the authors read and approved the final version of the manuscript.
None of the authors has any conflicts of interest to declare.
