We collected data from the NHANES, a national population survey conducted by the National Center for Health Statistics (NCHS) in the US. It uses complex, multi-stage, and probability sampling techniques and is released on a 2-year cycle. It aims to investigate the nutritional and health status of the entire population in the United States (31). Information can be retrieved on the NHANES website.¹ The NCHS Ethics Review Board approved the NHANES study protocol, and each participant signed an informed consent form.

We collected 29,940 participants from the NHANES database in consecutive NHANES cycles 2007–2010 and 2017–2018. We excluded 5,786 participants with missing data on a hyperlipidemia diagnosis, 4,786 participants with missing data on flavonoid intake, 5,511 participants younger than 20 years of age, 463 participants with caloric intake greater than 4,200 calories, or Participants with caloric intake less than 700 calories, 153 participants. Pregnant participants and 1,462 cancer participants, for a total of 8,940 participants (Figure 1).

We collected data on dietary flavonoid intake from the USDA Survey of Food and Beverage Flavonoid Values Database (“flavonoid database”). This database provides intakes of compounds from foods and beverages from the USDA Dietary Study Food and Nutrient Database (32) and corresponding dietary data from WWEIA (33) and NHANES. The USDA Nutrient Data Laboratory measured the content (mg/100 g) of 29 flavonoids in each food/beverage. Dietary flavonoids include the following seven flavonoid classes and the total daily intake of all flavonoids (the sum of 29 flavonoids) calculated from all foods and beverages. This study collected dietary flavonoid intake data from the flavonoid database from 2007–2010 to 2017–2018. We defined dietary flavonoid intake as the average of 2 days for each flavonoid.

Hyperlipidemia was identified when any of the following criteria were met: triglycerides ≥150 mg/dL; total cholesterol ≥200 mg/dL; low-density lipoprotein ≥130 mg/dL; high-density lipoprotein ≤40 mg/dL (male); high-density lipoprotein ≤50 mg/dL (female); or utilization of antihyperlipidemic agents.

We included the following covariates: age, race, education, Family income-poverty ratio (PIR), body mass index (BMI), smoking status, alcohol drinking, caloric intake, Protein, Carbohydrate, total fat, total saturated fatty acids (total sfat), total polyunsaturated fatty acids (total mfat), total monounsaturated fatty acids (total pfat), total cholesterol, vitamin D, vitamin E, AST/ALT, ACR, eGFR, lipid-lowering drugs, hypertension, diabetes, heart attack, stroke, and coronary heart disease.

Participants were divided into the following three groups according to age: <30 years, 30–59 years, and ≥ 60 years. Race was divided into non-Hispanic White, non-Hispanic Black, Mexican-American, and others. The family income-poverty ratio was classified as <1.5, 1.5–3.5, and > 3.5. Education level was categorized as less than high school, high school or equivalent, some college or AA degree, and college graduate or above. Smoking status was classified as never (smoked fewer than 100 cigarettes in life), former (smoked more than 100 cigarettes in life and smoke not at all now), now (smoked more than 100 cigarettes in life and smoked some days or every day). Alcohol drinking was classified as never (had <12 drinks in a lifetime); former (had ≥12 drinks in 1 year and did not drink last year, or did not drink last year but drank ≥12 drinks in a lifetime); Mild (defined as two drinks per day for men and one drink per day for women); moderate (defined as three drinks per day for men and two drinks per day for women, or binge drinking 2–4 days per day); heavy (defined as ≥ four drinks per day for men and ≥ three drinks per day for women, or binge drinking ≥5 days per day) (34).

Disease covariates in this study include hypertension, diabetes, heart attack, stroke, and coronary heart disease. Based on the questionnaire and physical examination results, participants were diagnosed with hypertension if they met one of the following three conditions: (1) the average systolic blood pressure ≥ 130 mmHg or the average diastolic blood pressure ≥ 80 mmHg; (2) the answer to the question “have you ever been told to take a prescription for hypertension” was “yes”; (3) the answer to the question “have you ever been told that you had high blood pressure” was “yes.” All blood pressure determinations (systolic and diastolic) were taken at a mobile examination center. The following protocol calculated average blood pressure: The diastolic reading with zero was not used to calculate the diastolic average. If all diastolic readings were zero, then the average would be zero. If only one blood pressure reading was obtained, that reading is the average. If there was more than one blood pressure reading, the first reading was excluded from the average. The diagnostic criteria for diabetes were as follows: the doctor told the patient they have diabetes, HbA1c ≥ 6.5%; fasting glucose ≥7.0 mmol/L; random blood glucose ≥11.1 mmol/L; two-hour OGTT blood glucose ≥11.1 mmol/L; utilization of diabetes medication or insulin. DM: diabetes mellitus; IFG: impaired fasting glycemia (fasting glucose 6.1–7.0 mmol/L); IGT: Impaired Glucose Tolerance (two-hour OGTT blood glucose 7.8–11.1 mmol/L). Cardiovascular disease was diagnosed based on whether questionnaires and physical examination results were used, and cardiovascular disease was defined as heart attack, stroke, and coronary heart disease. The diagnostic criteria for heart attack: the answer to the question “Have you ever been told that you had a heart attack?” was “yes.” The diagnostic criteria for stroke: the answer to the question “Have you ever been told that (the patient) had a stroke?” was “yes.” The diagnostic criteria for coronary heart disease: the answer to the question “Have you ever been told that you had coronary heart disease?” was “yes.”

All statistical analyses were performed using R software (version 4.1.3). Data preparation and statistical analysis were performed using the R packages “NHANESR” and “survey.” In the analysis of baseline information, continuous variables were expressed as weighted means ± standard deviations using one-way analysis of variance to compare differences between groups; categorical variables were expressed as frequencies and percentages and compared using the chi-square test. We used four weighted logistic regression models to examine the relationship between flavonoid consumption and hypertension prevalence. The crude model was unadjusted. Model 1 was adjusted for age, race, and sex. Model 2 was adjusted for age, race, sex, caloric intake, smoking status, alcohol drinking, and PIR. Model 3 was adjusted for age, race, sex, caloric intake, smoking status, alcohol drinking, education, PIR, protein, total fat, total sfat, total mfat, total pfat, total cholesterol, vitamin D, vitamin E, ACR, eGFR, AST/ALT, lipid-lowering drugs, hypertension, heart attack, stroke, coronary heart disease, and diabetes.

We used weighted restricted cubic splines from the “rms” package to evaluate potential nonlinear associations. Subgroup weighted logistic regression was used to analyze the effect of flavonoid intake on the prevalence of hyperlipidemia, stratified by age, race, sex, caloric intake, smoking status, alcohol drinking, education, PIR, protein, total fat, total sfat, total mfat, total pfat, total cholesterol, vitamin D, vitamin E, ACR, eGFR, AST/ALT, lipid-lowering drugs, hypertension, heart attack, stroke, coronary heart disease, and diabetes. Weighted logistic regression was used to calculate odds ratios and corresponding 95% confidence intervals. A significance level of p < 0.05 was used as the threshold for statistical significance. The P for interaction was based on the log-likelihood ratio test to assess the heterogeneity of the relationship between subgroups.

We included 8,940 participants. The baseline characteristics were grouped according to hyperlipidemia (Table 1). There were 6,520 patients with hyperlipidemia and 2,420 without. The average age of healthy participants was 38.89 (0.51) years, and that of participants with hyperlipidemia was 48.90 (0.37) years. Non-Hispanic white, Mexican-American participants and participants in less than high school and high school or equivalent were more likely to have hyperlipidemia than healthy participants. In terms of smoking status, more participants with hyperlipidemia were current or former smokers than healthy participants. Among participants with hyperlipidemia, the proportion of former or mild drinkers was higher than that of healthy participants. Participants with hyperlipidemia had a higher BMI, lower caloric intake, lower total pfat intake, and lower vitamin E intake than healthy participants. Hyperlipidemic participants had lower ACR, higher eGFR, and lower AST/ALT compared to healthy participants. Among 8,940 participants, 1753 participants took lipid-lowering medications and 7,187 participants did not take lipid-lowering medications (Supplementary Table 1). Among the 6,520 participants with hyperlipidemia, 1753 participants took lipid-lowering drugs and 4,767 participants did not take lipid-lowering drugs. In terms of comorbidities, participants with hyperlipidemia had higher rates of hypertension, diabetes, stroke, heart attack, and coronary heart disease than healthy participants. There were no significant differences in sex, PIR, protein, total fat, total sfat, total mfat, total cholesterol, and vitamin D between healthy participants and those with hyperlipidemia. However, the participants with hyperlipidemia had a higher intake of subtotal catechins, total flavan-3-ols, and total flavonoids. The specific content of participants’ flavonoid dietary assessment is provided in Supplementary Table 2.

We analyzed weighted logistic regression to evaluate the potential association between flavonoid intake, and hyperlipidemia. Age, race, sex, caloric intake, smoking status, alcohol drinking, education, PIR, protein, total fat, total sfat, total mfat, total pfat, total cholesterol, vitamin D, vitamin E, ACR, eGFR, AST/ALT, lipid-lowering drugs, hypertension, heart attack, stroke, coronary heart disease, and diabetes were fully adjusted. We divided isoflavone intake into four groups based on flavonoid subclass intake; because more than 50% of participants did not report isoflavone intake, we divided isoflavone intake into two groups based on the median intake (Table 2).

In model 3, we observed that compared with the first quartile, there was an inverse relationship between subtotal catechins intake and the prevalence of hyperlipidemia in the third quartile [0.74 (0.56, 0.98), p = 0.04]. However, the p-value for the trend was insignificant (p = 0.6). Similarly, there was an inverse relationship between total flavan-3-ols intake and the prevalence of hyperlipidemia in the third quartile [0.76 (0.59, 0.98), p = 0.03], with a non-significant p-value for trend (p = 0.83). Compared with the first quartile, there was an inverse relationship between total anthocyanins intake and the prevalence of hyperlipidemia in the third [0.77 (0.62, 0.95), p = 0.02] and fourth quartiles [0.77 (0.60, 0.98), p = 0.04], the p-value for trend was significant (p = 0.04). Total flavonols intake in the fourth quartile [0.75 (0.60, 0.94), p = 0.02] was inversely related to the prevalence of hyperlipidemia, and the p-value for trend was significant (p = 0.02).

Because the p-value for a trend of the prevalence of hyperlipidemia and subtotal catechins intake and total flavan-3-ols intake were not significant, we considered a possible nonlinear relationship. We performed analyses using restricted cubic splines to explore whether there might be a nonlinear relationship between the prevalence of hyperlipidemia and subtotal catechins intake and total flavan-3-ols intake. There was a significant nonlinear relationship between the prevalence of hyperlipidemia and subtotal catechins intake (Figure 2A, p = 0.0002), total flavan-3-ols intake (Figure 2B, p = 0.0006), and total anthocyanidins (Figure 2C, p = 0.0153). In the above results, the nonlinear relationship between the prevalence of hyperlipidemia and subtotal catechins intake and total flavan-3-ols intake showed a U-shaped correlation. Through analysis, we observed that when the subtotal catechin intake was less than 25.36 mg/day, there was a significant negative linear relationship between the prevalence of hyperlipidemia and subtotal catechin intake. When the total flavan-3-ols intake is less than 41.09 mg/day, the prevalence of hyperlipidemia had a significant negative linear relationship with the total flavan-3-ols intake. However, the nonlinear relationship between the prevalence of hyperlipidemia and total flavones did not reach significance (Figure 2D).

To assess the robustness of the association between flavonoid intake and hyperlipidemia， we performed subgroup analysis using weighted logistic regression to determine the subgroup interaction effect between flavonoid intake and the prevalence of hyperlipidemia. Stratified analyses were adjusted for age, sex, race, education, smoking status, alcohol drinking, PIR, caloric intake, hypertension, heart attack, stroke, coronary heart disease, diabetes, and other variables.

After analysis, we found that the interaction between the prevalence of hyperlipidemia and subtotal catechins intake (Table 3, p for interaction = 0.005) was significant when stratified by sex. However, when stratified by other variables, the relationship between the prevalence of hyperlipidemia and subtotal catechins intake was not statistically significant. This finding showed that age, race, education, smoking status, alcohol drinking, PIR, caloric intake, hypertension, heart attack, stroke, coronary heart disease, diabetes, and other variables did not significantly affect the relationship between the prevalence of hyperlipidemia and subtotal catechins intake (p for interaction >0.05). The interaction between the prevalence of hyperlipidemia and total flavan-3-ols intake was significant when stratified by sex (Supplementary Table 3, p for interaction = 0.01) and heart attack (Supplementary Table 3, p for interaction = 0.02). However, age, sex, race, education, smoking status, alcohol drinking, PIR, caloric intake, hypertension, heart attack, stroke, coronary heart disease, diabetes, and other variables did not significantly affect the relationship between the prevalence of hyperlipidemia and total anthocyanidin intake (Supplementary Table 4, p for interaction > 0.05). There was no significant interaction between hyperlipidemia prevalence and total flavonols intake when stratified by age, sex, race, education, smoking status, alcohol drinking, PIR, caloric intake, hypertension, heart attack, stroke, coronary heart disease, diabetes, and other variables (Supplementary Table 5, p for interaction >0.05).

There was an interaction between the prevalence of hyperlipidemia and flavonoid intake in sex stratification. We used restricted cubic splines analysis to evaluate the association between flavonoid intake and the prevalence of hyperlipidemia in sex stratification. The nonlinear associations between hyperlipidemia prevalence and subtotal catechins intake (Figure 3A, p = 0.0001) and total flavan-3-ols intake (Figure 3B, p = 0.0002) were significant among female participants.