Differences in United States Adult Dietary Patterns by Cardiometabolic Health and Socioeconomic Vulnerability

Abstract

Background

Naturally occurring dietary patterns are not well described among individuals with cardiovascular disease (CVD) or cardiometabolic risk factors (i.e., diabetes, hypertension, obesity, and dyslipidemia), particularly considering socioeconomic vulnerability.

Objectives

We investigated major dietary patterns in the United States and their distribution by prevalent CVD, cardiometabolic risk factors, and socioeconomic vulnerability.

Methods

This cross-sectional study analyzed data from 32,498 noninstitutionalized adults who participated in the National Health and Nutrition Examination Survey (2009–2020). We used principal component analysis to identify dietary patterns. Using multiple linear regression, we tested the association of prevalent CVD, cardiometabolic risk factors, and socioeconomic vulnerability [number of social risk factors and Supplemental Nutrition Assistance Program (SNAP) participation status] with each pattern.

Results

Four dietary patterns were identified: processed/animal foods (high-refined grains, added sugars, meats, and dairy), prudent (high vegetables, nuts/seeds, oils, seafood, and poultry), legume, and fruit/whole grain/dairy, which together explained 29.2% of the dietary variance. After adjustment for age, gender, race and ethnicity, cohort year, and total energy intake, the processed/animals foods pattern associated (β-coefficient for difference in principal component score) positively with diabetes [0.08 (0.01, 0.14)], hypertension [0.11 (0.06, 0.16)], obesity [0.15 (0.11, 0.19)], higher social risk score (P-trend < 0.001), income-eligible SNAP nonparticipation [0.16 (0.09, 0.23)], and SNAP participation [0.23 (0.17, 0.29)]. The prudent pattern associated negatively with hypertension [−0.09 (−0.14, −0.04)], obesity [−0.11 (−0.16, −0.06)], higher social risk score (P-trend < 0.001), income-eligible SNAP nonparticipation [−0.14 (−0.21, −0.06)], and SNAP participation [−0.30 (−0.35, −0.24)]. The legume pattern was associated negatively with CVD [−0.09 (−0.15, −0.02)] and obesity [−0.08 (−0.12, −0.04)], and positively with income-eligible SNAP nonparticipation [0.11 (0.04, 0.18)]. The fruit/whole grain/dairy pattern was associated positively with diabetes [0.08 (0.01, 0.15)] and negatively with hypertension [−0.21 (−0.26, −0.15)], obesity [−0.23 (−0.28, −0.18)], higher social risk score (P-trend < 0.001), and SNAP participation [−0.19 (−0.25, −0.12)].

Conclusions

Empirical dietary patterns in the United States vary by CVD, cardiometabolic risk factors, and socioeconomic vulnerability. Initiatives to improve nutrition should consider these naturally occurring dietary patterns and their variation in key subgroups.

Introduction

Poor diet contributes greatly to premature death and morbidity in the United States, and poor diet is the top contributor to cardiovascular disease (CVD) mortality [1,2]. Modern health frameworks identify that clinical outcomes occur as downstream events related to interactions between individual factors (e.g., dietary behaviors), social contexts (e.g., social determinants of health), and socioeconomic policies [3]. This convergence is illustrated by the recognized connections between eating habits and CVD, cardiometabolic risk factors (i.e., diabetes, hypertension, obesity, and dyslipidemia), and socioeconomic vulnerability. For example, populations with food insecurity or other social risk factors have worse nutrition and higher burdens of diet-related chronic diseases, including CVD and cardiometabolic risk factors [[4], [5], [6], [7]]. These sociodemographic disparities in nutrition quality and CVD have persisted in the United States despite increased scientific understanding of nutrition and health and investments in government nutrition programs such as the Supplemental Nutrition Assistance Program (SNAP)—the largest food assistance program administered by the USDA with an annual budget of $120 billion [[8], [9], [10], [11], [12]]. A better understanding of how dietary patterns vary by cardiometabolic health and socioeconomic vulnerability may inform effective clinical and public health interventions.

The nutritional characteristics of overall diets can be evaluated by prespecified dietary pattern scores, that is, expert-derived metrics of adherence to dietary recommendations such as the Healthy Eating Index (HEI), Mediterranean diet score, or Dietary Approach to Stop Hypertension diet score [11,[13], [14], [15], [16], [17], [18]]. These scores were developed for general populations, and focus on externally recommended food consumption, rather than naturally observed patterns of food consumption. The assessment of empirically derived dietary patterns (combinations of foods that comprise an individual’s diet) provides a complementary paradigm to inform clinical and public health efforts [[19], [20], [21], [22]]. Because of structural barriers and cultural preferences, this assessment may be especially relevant to improve health equity because individuals with CVD, cardiometabolic risk factors, or socioeconomic vulnerability may have dietary habits that are overlooked by applying expert-derived dietary patterns. However, few studies have assessed how prevalent disease or socioeconomic vulnerability is associated with observed empirical dietary patterns.

To address this important knowledge gap, we characterized empirical dietary patterns among United States adults using updated nationally representative data from the NHANES, and evaluated whether adherence to these dietary patterns varied by CVD status, cardiometabolic risk factors, or socioeconomic vulnerability.

Methods

Design, setting, and population

NHANES is a nationally representative survey that assesses the health and nutrition of the United States population in 2-y cycles. Individuals are sampled using a 4-stage cluster design to represent the noninstitutionalized civilian population. We pooled data from the 2009–March 2020 NHANES cycles to provide a large, contemporary sample (overall response rate = 61.3%). All participants provided informed written consent, completed personal and household questionnaires, and received a health examination in a Mobile Examination Center (MEC). We included all participants who provided ≤2 d of interviewer-administered 24-h dietary recall using the multiple-pass method. The first recall was obtained during the mobile examination, and the second by telephone 3–10 d later. Full details of the sample design have been published [[23], [24], [25], [26]]. This study, using deidentified data, was deemed exempt by the University of Michigan Institutional Review Board. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology-Nutritional Epidemiology reporting guideline [27]. For this investigation of CVD and its risk factors, we included all participants aged >20 y (n = 32,498), given very low prevalence of CVD in younger individuals.

Dietary patterns

Major food categories (n = 29) were identified from USDA Food Pattern categories and converted to standardized servings using the Food Patterns Equivalents Database, including cup equivalents (fruit, vegetables, and dairy), ounce equivalents (grains and protein foods), teaspoon equivalents (added sugars), gram equivalents (solid fats and oils), and number of drinks (alcohol) [28]. To derive observed dietary patterns, we analyzed these food categories using principal components analysis (PCA), a dimensionality reduction method that identifies patterns of variation across multiple variables [29]. Eigenvalues were calculated and plotted (Supplemental Figure 1), which flattened after the fourth dietary pattern. We focused on these 4 patterns. For each participant, we also calculated the HEI-2020 score (range: 0–100), with higher scores reflecting greater compliance with the Dietary Guidelines for Americans [[30], [31], [32]].

CVD and cardiometabolic risk factors

Prevalent CVD was defined as a self-reported diagnosis by a health professional of coronary artery disease, stroke, or heart failure. Diabetes was defined by self-report, use of anti-diabetes medication, or hemoglobin A_1c ≥ 6.5%. Hypertension was defined by self-report, use of antihypertensive medication, or mean blood pressure (3 readings) ≥ 130/80 mmHg. Obesity was defined by BMI (in kg/m²) of ≥30 (or ≥27.5 among Asian adults), [33] calculated from measured weight in kilograms divided by measured height in meters squared. Hyperlipidemia was defined by use of cholesterol-lowering medication, low-density lipoprotein cholesterol ≥160 mg/dL (calculated by the Sampson method), or nonhigh-density lipoprotein cholesterol ≥190 mg/dL.

Socioeconomic vulnerability

Multiple social and economic factors can be used to define socioeconomic vulnerability. To assess a composite of multiple factors, we calculated a social risk factor score [34] based on having education less than high school graduation; being unmarried or not living with a partner; having Medicaid, other non-Medicare government health insurance, or being uninsured; experiencing food insecurity (low/very low food security); being unemployed (excluding retired individuals or students); or not having a routine location for healthcare. Housing status could not be included because of known systematic missingness in the 2017–2020 cycles [26]. One point was given for each social risk factor and summed to calculate an overall score, with a higher score representing greater socioeconomic vulnerability (range: 0–6). Given our focus on nutrition, SNAP participation status was evaluated separately. SNAP participation was defined as self-reported household participation within the last 12 mo. Income-eligible nonparticipants were defined by eligible annual family income [<130% of the federal poverty level (FPL), income-to-poverty ratio < 1.3] but without household SNAP participation within the last 12 mo. SNAP noneligible adults were those with annual family income >130% of FPL and no household SNAP participation within 12 mo.

Additional covariates

Additional demographic covariates included age (years), self-reported gender evaluated as a biologic variable, race and ethnicity [Hispanic, non-Hispanic Asian (hereafter Asian, available from 2011 forward), non-Hispanic Black (hereafter Black), non-Hispanic White (hereafter White), and others], total energy intake (kcal/day), and NHANES survey cycle (first survey year of cycle). Race and ethnicity were evaluated as sociocultural constructs that were self-reported based on NHANES-defined categories [[23], [24], [25], [26]].

Statistical design and analysis

Population characteristics were evaluated overall and by CVD, social risk score, and SNAP status, as weighted percent [95% confidence interval (CI)] for categorical variables and mean (SD) for continuous variables. Multiple linear regression models were constructed with each identified dietary pattern principal component score as the outcome. All models were adjusted for age, gender, race and ethnicity, total energy intake, and NHANES survey cycle; and included the predictors of CVD, cardiometabolic risk factors, social risk score, and SNAP status simultaneously to account for their interrelated effects. We set the maximum score as >4 risk factors because only 1.3% of the cohort had 5 or 6 social risk factors. Missing variables in NHANES (Supplemental Table 1) were multiply imputed, with results and analyses pooled from twenty imputed data sets.

We evaluated results overall and stratified by race and ethnicity. Statistical significance of interactions by race and ethnicity were tested by adding multiplicative interaction terms by each exposure (CVD status, cardiometabolic risk factors, social risk score, and SNAP status). A sensitivity analysis was conducted excluding food categories with Kaiser–Meyer–Olkin measure of sampling adequacy <0.5. A secondary analysis was done to evaluate proportions of the population aligned with each pattern; each participant was assigned to the dietary pattern for which they had the highest absolute principal components score. A single multinomial logit model examined the associations of 4 dietary patterns as the outcome with CVD status, social risk scores, and SNAP status each evaluated as categorical variables. Marginal outputs from this model estimated the proportion of individuals aligned with each dietary pattern in each subgroup. All analyses utilized NHANES MEC weights to account for the complex, multistage, and probability sampling design to produce nationally representative estimates. Adjustments for multiple comparisons were not performed because we were testing multiple separate hypotheses. Statistical significance was set at 2-sided alpha = 0.05. Analyses were performed using Stata v16 (StataCorp, LLC).

Results

Population characteristics

The analytic sample comprised 32,498 adults, representing 231 million United States adults aged >20 y (Table 1). Mean (SD) age was 47.7 (17.1). Self-reported race and ethnicity included Asian (4.7%), Black (11.4%), Hispanic (14.9%), and White (65.0%). Prevalent CVD was present in 8.3%. Cardiometabolic risk factors were common, including 45.7% with hypertension, 14.3% with diabetes, 38.6% with obesity, and 34.9% with dyslipidemia diagnoses. Social risk factors were also common: 14.8% had less than high school education, 37.5% were not married or living with a partner, 34.6% were publicly insured or uninsured, 15.7% had low or very low food security, 11.6% were unemployed, and 15.6% lacking a routine location for healthcare. The mean (SD) social risk score was 1.3 (1.3). With respect to SNAP participation status, 17.0% were SNAP participants, 10.2% were income-eligible SNAP nonparticipants, and 72.8% were noneligible for SNAP. The mean (SD) HEI score was 53.4 (13.6) out of 100 total points.

TABLE 1.

Weighted population characteristics (n = 32,498, representing 230,612,714 United States adults).

Characteristic	% (95% CI) or mean (SD)
Age (y)	47.7 (17.1)
Female	51.9% (51.3%, 52.6%)
Race
Asian1	4.7% (4.0%, 5.6%)
Black	11.4% (10.0%, 13.1%)
Hispanic	14.9% (13.0%, 17.1%)
White	65.0% (62.0%, 67.8%)
Other1	4.0% (3.5%, 4.5%)
Cardiovascular disease diagnosis	8.3% (7.8%, 8.9%)
Diabetes	14.3% (13.7%, 14.9%)
Hypertension	45.7% (44.5%, 47.0%)
Obesity	38.6% (37.5%, 39.7%)
Dyslipidemia	34.9% (34.0%, 35.8%)
Social risk factors
Not high school graduate or GED recipient	14.8% (13.7%, 16.0%)
Not married or living with partner	37.5% (36.2%, 38.9%)
Public insurance or uninsured	34.6% (33.2%, 36.1%)
Low, or very low food security2	15.7% (14.8%, 16.6%)
Unemployed	11.6% (10.9%, 12.4%)
No regular healthcare access	15.6% (14.8%, 16.5%)
Social risk score
0	34.5% (32.9%, 36.1%)
1	29.1% (28.3%, 30.0%)
2	18.7% (18.0%, 19.5%)
3	11.3% (10.6%, 12.1%)
4	5.1% (4.7%, 5.5%)
5	1.2% (1.1%, 1.4%)
6	0.11% (0.08%, 0.14%)
SNAP participation status
Noneligible	72.8% (71.1%, 74.4%)
Income-eligible nonparticipant	10.2% (9.6%, 10.9%)
Participant	17.0% (15.7%, 18.4%)
Total energy intake (kcal/d)	3936 (1679)
Healthy Eating Index score	53.4 (13.6)

Abbreviation: GED. General Education Development; SNAP, Supplemental Nutrition Assistance Program.

¹Other included non-Hispanic Asian individuals in the 2009–2010 cohort.

²Measured using the United States Household Food Security Survey Module.

In unadjusted (bivariate) analyses, those with a CVD diagnosis were more likely to be older, male, Black or White, have diabetes, hypertension, obesity, or dyslipidemia, less educated, unmarried or not living with a partner, food insecure, unemployed, have regular healthcare access, have a higher social risk score, and be SNAP participants or income-eligible SNAP nonparticipants (Supplemental Table 2), highlighting the intersections of CVD with socioeconomic vulnerability. Similarly, those with a higher social risk, compared with lower social risk score, were younger and more likely to be male, Black, or Hispanic, and to have diabetes, hypertension, obesity, dyslipidemia, and lower HEI scores (Supplemental Table 3). SNAP participants were younger, higher proportion of female and Black individuals, and were more likely to have diabetes, obesity, dyslipidemia, higher social risk scores, and lower HEI scores (Supplemental Table 4) compared with noneligible adults.

Identified dietary patterns

The 4 dietary patterns identified are shown in Table 2. The processed/animal foods pattern included higher intake of refined grains, meats, cured meats, milk, cheese, oils, solid fats, and added sugars (total population variance explained: 10.2%). The prudent pattern included higher vegetables, meat, poultry, seafood, eggs, and oils, and lower milk (variance explained: 8.0%). The legume pattern was heavily weighted by higher legumes (variance explained = 6.5%). The fruit/whole grain/dairy pattern included higher fruits, whole grains, soy protein, nuts/seeds, milk, and yogurt, and lower potatoes and eggs (variance explained = 4.6%).

TABLE 2.

Principal components analysis result showing component loading (>0.15) of the 4 dietary patterns¹.

Dietary component	Pattern 1 processed/ animal foods pattern	Pattern 2 prudent pattern	Pattern 3 legume pattern	Pattern 4 fruit/whole grain/dairy pattern
Citrus/melons/berries				0.28
Other fruits				0.43
Fruit juices
Dark green vegetables		0.26
Tomatoes	0.17	0.16
Other red orange vegetables		0.24
Starchy potatoes		0.21		−0.19
Starchy nonpotatoes
Other vegetables		0.36
Vegetable legumes			0.68
Whole grains				0.44
Refined grains	0.41
Meats	0.16
Cured meats	0.26
Organ meats
Poultry		0.22
Seafood (high omega-3)		0.20
Seafood (low omega-3)		0.26		−0.15
Eggs		0.19
Soy protein
Nuts/seeds		0.34
Protein legumes			0.68
Milk	0.25			0.41
Yogurt				0.34
Cheese	0.41
Oils	0.19	0.53
Solid fats	0.53
Added sugars	0.36
Alcohol		0.16		−0.26

¹Components shown after promax rotation. The 4 dietary patterns were named based on their strongest loading dietary patterns. In post hoc testing of the PCA, the Kaiser–Meyer-Olkin (KMO) measure of sampling adequacy revealed poor sampling of potatoes, organ meats, poultry, seafoods low in n-3 fatty acids, and nuts/seeds (KMO < 0.5). A sensitivity test was performed for the main analysis with these factors excluded. Note that positive values indicate positive correlation of these items with the pattern and negative values indicate negative correlations.

The mean (SD) HEI score for the top quartile of each pattern was 45.1 (10.3) for fats/processed/dairy, 60.7 (12.7) for prudent, 58.7 (12.6) for legume, and 65.2 (11.6) for fruit/whole grain/dairy patterns (Table 3).

TABLE 3.

Mean Healthy Eating Index scores by quartile of dietary pattern principal components score.

Quartile	Processed/animal foods pattern Mean (SD)	Prudent pattern Mean (SD)	Legume pattern Mean (SD)	Fruit/whole grain/dairy pattern Mean (SD)
Quartile 1	64.1 (13.2)	45.8 (12.2)	53.2 (13.4)	43.7 (10.0)
Quartile 2	55.9 (12.1)	50.8 (12.4)	51.5 (13.4)	47.7 (10.5)
Quartile 3	50.9 (11.4)	55.0 (12.4)	50.9 (13.5)	55.5 (10.7)
Quartile 4	45.1 (10.3)	60.7 (12.7)	58.7 (12.6)	65.2 (11.6)

CVD and cardiometabolic risk factors in relation to the dietary patterns

The processed/animal foods pattern associated positively with diabetes [β-coefficient for difference in principal component score = 0.08 (0.01, 0.14), P = 0.03], hypertension [0.11 (0.06, 0.16), P < 0.001], and obesity [0.15 (0.11, 0.19), P < 0.001; Figure 1]. The prudent pattern associated negatively with hypertension [−0.09 (−0.14, −0.04), P = 0.001] and obesity [−0.11 (−0.16, −0.06), P < 0.001]. The legume pattern associated negatively with CVD [−0.09 (−0.15, −0.02), P = 0.008] and obesity [−0.08 (−0.12, −0.04), P < 0.001]. The fruit/whole grain/dairy pattern associated positively with diabetes [0.08 (0.01, 0.15), P = 0.02] and negatively with hypertension [−0.21 (−0.26, −0.15)), P < 0.001] and obesity [−0.23 (−0.28, −0.18), P < 0.001]. Dyslipidemia did not associate with any patterns.

FIGURE 1.

Association of dietary patterns (principal components score) with cardiovascular disease and cardiometabolic risk factors for the entire cohort and by race and ethnicity. (A) Entire cohort; (B) non-Hispanic White individuals; (C) non-Hispanic Black individuals; (D) Hispanic individuals; (E) non-Hispanic Asian individuals. P values are indicated as follows: ∗P < 0.05; ∗∗P < 0.01; ∗∗∗P < 0.001. Values are β-coefficient for difference in principal component score. Positive values indicate positive association with the principal components score. Models were individual linear regression models with each principal components score as the outcome variable. All models were adjusted for age, gender, race and ethnicity, total energy intake, and NHANES survey cycle; and included the predictors of cardiovascular disease, cardiometabolic risk factors, social risk score, and Supplemental Nutrition Assistance Program (SNAP) status simultaneously to account for their interrelated effects.

Socioeconomic vulnerability in relation to the dietary patterns

The processed/animal foods pattern associated positively with higher social risk score (P-trend < 0.001), income-eligible SNAP nonparticipation [0.16 (0.09, 0.23), P < 0.001], and SNAP participation [0.23 (0.17, 0.29), P < 0.001; FIGURE 2, FIGURE 3]. The prudent pattern associated negatively with higher social risk score (P-trend < 0.001), income-eligible SNAP nonparticipation [−0.14 (−0.21, −0.06) P < 0.001], and SNAP participation [−0.30 (−0.35, −0.24) P < 0.001]. The legume pattern had no association with the social risk score, positive association with income-eligible SNAP nonparticipation [0.11 (0.04, 0.18), P = 0.002], and no association with SNAP participation [0.02 (−0.05, 0.09), P = 0.55]. The fruit/whole grain/dairy pattern and negative association with higher social risk score (P-trend < 0.001), no association with income-eligible SNAP nonparticipation [−0.06 (−0.13, 0.00), P = 0.07], and positive association with SNAP participation [−0.19 (−0.25, −0.12) P < 0.001]. For coefficients of all covariates, see Supplemental Table 5.

FIGURE 2.

Association of dietary patterns (principal components score) with number of social risk factors for the entire cohort and by race and ethnicity. (A) Entire cohort; (B) non-Hispanic White individuals; (C) non-Hispanic Black individuals; (D) Hispanic individuals; (E) non-Hispanic Asian individuals. P values are indicated as follows: ∗P < 0.05; ∗∗P < 0.01; ∗∗∗P < 0.001. Note: Values are β-coefficient for difference in principal component score. Positive values indicate positive association with the principal components score. Models were individual linear regression models with each principal components score as the outcome variable. All models were adjusted for age, gender, race and ethnicity, total energy intake, and NHANES survey cycle; and included the predictors of cardiovascular disease, cardiometabolic risk factors, social risk score, and Supplemental Nutrition Assistance Program (SNAP) status simultaneously to account for their interrelated effects.

FIGURE 3.

Association of dietary patterns (principal components score) with Supplemental Nutrition Assistance Program participation status for the entire cohort and by race and ethnicity. (A) Entire cohort; (B) non-Hispanic White individuals; (C) non-Hispanic Black individuals; (D) Hispanic individuals; (E) non-Hispanic Asian individuals. P values are indicated as follows: ∗P < 0.05; ∗∗P < 0.01; ∗∗∗P < 0.001. Note: Values are β-coefficient for difference in principal component score. Positive values indicate positive association with the principal components score. Models were individual linear regression models with each principal components score as the outcome variable. All models were adjusted for age, gender, race and ethnicity, total energy intake, and NHANES survey cycle; and included the predictors of cardiovascular disease, cardiometabolic risk factors, social risk score, and Supplemental Nutrition Assistance Program (SNAP) status simultaneously to account for their interrelated effects.

Findings by race and ethnicity

There were significant interactions by race and ethnicity in the relationships of diabetes, hypertension, obesity, dyslipidemia, social risk score, and SNAP status, but not with CVD, with several dietary patterns (Supplemental Table 6). In analyses stratified by race and ethnicity, there were notable findings that differed from the nonstratified analysis (Supplemental Tables 7–10; FIGURE 1, FIGURE 2, FIGURE 3). Among Black and Hispanic individuals but no other race and ethnicities, there was a negative association between the prudent pattern and CVD. Among Asian individuals, there was a unique positive association between the legume pattern and diabetes. Among only Black and Hispanic individuals, there was a notable lack of association between the processed/animal foods pattern and hypertension. Among only Hispanic individuals, the legume pattern associated positively with social risk score and SNAP participation.

Sensitivity analysis

In sensitivity analyses, we removed starchy potatoes, organ meats, poultry, seafood low in omega-3, and nuts/seeds because they did not meet the sampling adequacy of measures. All findings were generally similar to the primary analysis. As one exception, prevalent CVD was now more strongly associated with the prudent pattern [−0.10 (95% CI: −0.17, −0.02), P = 0.01; Supplemental Tables 11 and 12].

Secondary analysis: proportion aligning with each dietary pattern by prevalent CVD, social risk score, and SNAP status

In secondary analyses, about one-third (33.7%) of United States adults’ diets aligned most closely with processed/animal foods, followed by the fruit/whole grain/dairy (27.9%), prudent (22.6%), and legume (15.8%) patterns. Notable differences from the primary analysis were that a higher proportion of United States adults was consuming diets aligned with the processed/animal foods pattern with CVD [36.4% (95% CI: 33.7%, 39.0%)] compared with those without CVD [33.5% (32.6%, 34.4%), P = 0.04; Supplemental Figure 2 and Supplemental Table 13]. Also, the legume pattern was negatively associated with higher social risk score (P-trend < 0.001, Supplemental Figure 3). Finally, the legume pattern overall had a similar proportion of income-eligible SNAP nonparticipants [17.4% (15.8%, 19.1%)] and SNAP participants [17.1% (15.4%, 18.7%), P = 0.71] and both were higher than noneligible adults [15.2% (14.4%, 16.0%), P = 0.04 and P = 0.01, respectively; Supplemental Figure 4]. For coefficients of all covariates, refer to Supplemental Table 14.

Discussion

In this nationally representative study of United States adults, we identified 4 predominant dietary patterns and identified divergent associations with prevalent CVD, cardiometabolic risk factors (i.e., diabetes, hypertension, obesity, and dyslipidemia), and socioeconomic vulnerability. The processed/animal foods pattern was associated positively with diabetes, hypertension, higher social risk score, income-eligible SNAP nonparticipation, and SNAP participation. The prudent pattern was associated negatively with hypertension, obesity, higher social risk score, income-eligible SNAP nonparticipation, and SNAP participation. The legume pattern was associated negatively with CVD and obesity, and positively with income-eligible SNAP nonparticipation. The fruit/whole grain/dairy pattern associated positively with diabetes and negatively with hypertension, obesity, higher social risk score, and SNAP participation. The effect sizes were generally smaller for CVD and cardiometabolic risk factors than socioeconomic vulnerability markers. The largest effect sizes were with social risk scores of 2 or more. Analyses stratified by race and ethnicity revealed some associations that differ from nonstratified analyses.

These new findings provide further support for the role of unhealthy diet as a contributor to the convergence of CVD, cardiometabolic risk factors, and socioeconomic vulnerability. From a health equity lens, this emphasizes that the most vulnerable individuals, from either a health or socioeconomic standpoint, may require more concentrated nutritional supports to prevent diet-related adverse outcomes. In contrast, socioeconomically vulnerable adults were not less likely to align to a healthier dietary pattern marked by higher legume intake. Legumes have lower cost per serving than other protein foods, are rich in fiber and phytonutrients, and are associated with lower risk of CVD [[35], [36], [37]]. This suggests that, in some cases, there may be healthier dietary patterns that are accessible among socioeconomically vulnerable individuals. These findings support the need for further investigation of the role of the legume pattern, especially for socioeconomically vulnerable individuals. On the basis of our observed difference in analyses stratified by race and ethnicity, this may be further shaped by social and cultural influences. Our findings provide new findings relevant to prior observations that Hispanic individuals have a lower incidence of CVD despite a higher prevalence of CVD risk factors such as obesity, diabetes, and dyslipidemias [[38], [39], [40]]. There is increasing recognition that this “paradox” may be a result of unmeasured differences in social support structures and social risk factors. Our findings are hypothesis generating and require further exploration of how only among Hispanic individuals the legume pattern associates with more social risk factors and how this may contribute to eventual health outcomes.

Our investigation considered race and ethnicity as a sociocultural construct, [41] whereby individuals may be more likely to follow dietary patterns because of cultural and regional influences and family heritages. Except for the aforementioned findings, a majority of our observations were similar across race and ethnicity. Our findings are consistent with prior studies suggesting that a meaningful proportion of the relationships between socioeconomic factors, race and ethnicity, and life expectancy are mediated by behavioral and metabolic risk factors, including diet [42]. Additional research into how dietary patterns are achieved and impacted in the context of socioeconomic vulnerability is needed to help us understand our observed differences across race and ethnicity.

Our findings are also relevant for the burgeoning implementation of “Food is Medicine” approaches to integrate food-based nutritional therapies into healthcare [[43], [44], [45]]. Understanding naturally occurring patterns of dietary intake by CVD, cardiometabolic risk factors, and socioeconomic vulnerability can inform clinical interventions and practices to improve nutrition quality, health outcomes, and health equity for all Americans. Randomized trials have shown that healthier dietary patterns can reduce CVD events by 30%–40% [[46], [47], [48]]. Our primarily neutral findings of associations between CVD and dietary patterns are concerning in the context that diet is a major determinant of health outcomes among those with CVD. It is promising that the prudent, legume, and fruit/whole grain/dairy patterns each tended to associate negatively with cardiometabolic risk factors. Understanding how to introduce healthier food items within existent dietary patterns may facilitate sociocultural relevance of Food is Medicine programs, such as medically tailored groceries or meals; and better ways to incentivize healthier food selection and nutrition education within SNAP. This is especially important for groups at higher risk related to food insecurity, such as SNAP participants [13,[49], [50], [51], [52]]. Furthermore, people with diet-sensitive chronic conditions may benefit from dedicated dietary counseling. This approach is proposed in the Federal Medical Nutrition Therapy Act of 2023, which would expand Medicare coverage for dietary counseling to those with CVD and cardiovascular disease risk factors [[53], [54], [55], [56], [57]]. This coverage is sorely needed because dietary counseling is underutilized [58].

Our findings have similarities and differences with prior findings in the context of CVD and socioeconomic vulnerability. In the prospectively collected Multi-Ethnic Study of Atherosclerosis cohort, 4 empirical patterns were identified [19]. In this study, a fats and processed meat pattern (analogous to the processed/animal foods pattern in our study) was associated with higher CVD risk [hazard ratio (HR) of quintile 5 compared with 1: 1.82 (95% CI: 0.99, 3.35)]. Although we did not observe this association in the main analysis, we did see this in secondary analysis. Also, unlike the null finding in our cross-sectional study, a whole grain/fruit pattern was associated with lower CVD risk [HR: 0.54 (0.33, 0.91)] [19]. In the Nurses’ Health Study, 2 dietary patterns were found, with lower risk from a prudent [HR: 0.72 (0.60, 0.87)] compared with a western/processed food pattern [20]. Studies considering participants in SNAP have found that participants consume fewer healthy foods and nutrients and have lower scores on institutionally derived diet patterns, at least partly related to structural barriers faced in accessing and affording a healthful diet [13,49,51,52,59,60].

Few prior studies have evaluated naturally occurring dietary patterns in nationally representative United States populations. Among United States cancer survivors, 2 dietary patterns were identified, including a prudent pattern that was associated with lower food insecurity [61]. Two studies identified patterns in nutrient intake in relation to hypertension and bone health, but did not evaluate food groups [62,63]. One study from the early 1990s identified 4 dietary patterns among Mexican Americans in relation to gallbladder disease, including patterns marked by highly processed food, high vegetables, and high beans [64]. Our study builds upon and greatly extends this prior work by identifying contemporary food patterns and their relations to CVD and socioeconomic vulnerability.

Strengths

The study’s strengths include the large nationally representative cohort and racial and ethnic diversity, which support the reliability and generalizability of our findings. Sensitivity analyses to address shortcomings of the data did not change the results. Finally, we used rigorous and established methods for the PCA that help to account for correlation among food categories [65].

Limitations

Potential limitations include the cross-sectional nature of the data source, which limits causal inference. Also, NHANES relies on individual self-reports of CVD, SNAP, and social risk factors, which may increase the misclassification of these exposures. Dietary habits were assessed using the average of 2 24-h recalls, and longer dietary intake monitoring may increase the accuracy of observations. The Asian subgroup in the cohort was small, which limited the statistical power of analyses in this group.

In conclusion, empirical dietary patterns vary by both cardiometabolic health and especially socioeconomic vulnerability. Initiatives to improve nutrition in at-risk individuals should consider these naturally occurring dietary patterns and their variation in key subgroups. Interventions in health policy and clinical practice should be evaluated to promote healthier dietary patterns among people with or at risk for CVD or socioeconomic vulnerability.

Author contributions

The authors’ responsibilities were as follows – EJB, MK: assisted in procuring the data and writing the first draft, generated the coding for the analyses; and all authors: participated in conceptualization, reviewed and commented on subsequent drafts of the manuscript and read and approved the manuscript.

Data availability

Data are publicly available from the CDC.

Funding

This study was supported by the NIH NIMHD K23MD017253 (to EJB). The sponsor had no role in the study design, collection, analysis, interpretation of the data, writing, or any restrictions on the manuscript.

Conflict of interest

EJB reports financial support from the National Institute on Minority Health and Health Disparities. The other authors report no conflicts of interest.

Appendix A. Supplementary data

The following is the Supplementary data to this article: