The associations of plant-based food and metabolic syndrome using NHANES 2015–16 data

Abstract

Background

Consuming a diet rich in plant-based foods (PBFs) may be protective for risk of metabolic syndrome (MetS) and chronic disease. However, the impact of consuming healthy versus all types of PBF on MetS is unknown.

Methods

The relationship between consumption of PBF (both healthy and all) was examined using data from the 2015 to 2016 National Health and Nutrition Examination Survey (NHANES). The amount of PBFs consumed was calculated as average daily servings, whereas dichotomous MetS variables were based on the National Cholesterol Education Adult Program Treatment Panel III (2005). After weighting and multiple imputation, adjusted associations were examined using logistic regression and marginal probabilities.

Results

Consumption of healthy PBF was significantly associated with reduced risk for elevated waist circumference (P = 0.017; odds ratio, OR 0.96, 95% confidence interval, CI 0.94–0.98) and MetS (P = 0.006; OR 0.96, 95% CI 0.93–0.99). Interactions revealed significant protective effects for females who were |$\ge$| 60 years.

Conclusions

In the adjusted model, a 1-unit increase in daily servings of healthy PBF was associated with a 4% lower risk for prevalence of elevated waist circumference and a 4% decrease in risk for prevalence of MetS. Increasing consumption of PBF may offer a viable strategy for reducing risk of MetS.

Metabolic Syndrome, plant-based food, NHANES

Introduction

A wealth of studies suggests that, over the long term, diet has a profound effect on health, and is strongly correlated with development of chronic disease.^1–5 In particular, there is evidence that consuming a plant-based diet is protective for diet-related chronic disease prevalence. Research has found associations between consumption of a plant-based diet and lower risk for obesity,^6,7 heart disease,^8,9 cancer,^10,11 and Type 2 diabetes.^4,12

Thus, consumption of a plant-based diet holds promise for providing a simple and feasible approach for reducing the risk of chronic disease. However, no studies examining the association of plant-based diets and chronic disease have been conducted in nationally representative samples in the US, limiting the generalizability of any conclusions. In addition, there is controversy over what, exactly, constitutes a plant-based diet. Definitions range from Lea et al.’s ¹³ characterization of ‘an eating pattern that is dominated by fresh or minimally processed plant foods and decreased consumption of meat, eggs and dairy products’, to more restrictive definitions that exclude all processed foods, meat, dairy and eggs.¹⁴ We propose to circumvent this obstacle by avoiding labels altogether, and to simply assess the association of plant-based foods (PBF) with chronic disease. PBFs are defined as foods that come from plants, including fruits and vegetables (including juice), grains, legumes, nuts and seeds, soy, oils from plants (i.e. olive, corn, etc.) and sugar.

The potential pro-health benefits of a diet high in PBFs may be confounded by the inclusion of highly processed plant foods. For example, Satija et al.⁵ created healthy and unhealthy plant-based diet indices, giving positive scores to healthy plant foods and reverse scores to less healthy plant foods such as juices, sweetened beverages, refined grains, potatoes, fries and sweets. Their findings indicated that an unhealthy plant-based diet was positively associated with incidence of T2D, whereas a healthy plant-based diet was negatively associated. Therefore, it is important to assess the association of healthy versus general PBFs to understand these relationships. NOVA guidelines provide a framework for categorizing PBFs as healthy, or not, based on the level of processing.¹⁵ We propose to assess the associations of consumption of servings of plant-based food, using NOVA guidelines to identify healthy plant-based foods (HealthyPBFs) as those which exclude added sugars, refined grains and oils, versus consumption of servings of all plant-based food (AllPBF), in order to disaggregate their importance to chronic disease risk.

Finally, we propose to utilize a general indicator for chronic disease risk, metabolic syndrome (MetS), in order to capture greater heterogeneity in a representative sample and provide a general indicator associated with incidence of chronic disease. MetS describes a set of co-occurring, interrelated risk factors that predict incidence of chronic disease, as demonstrated by studies in Canada¹⁶ and the US.^17,18 Globally, the prevalence of MetS is 25%, whereas in the US that number is 34.3%.¹⁹

Based on these factors, our primary research question follows: Is higher consumption of PBFs associated with lower risk for MetS and MetS components (hyperglycemia, abdominal obesity, hypertriglyceridemia, hypertension and low HDL), adjusted for relevant confounders (age, sex, income and race/ethnicity)? Secondarily, are these adjusted associations different for higher consumption of healthy versus all PBFs? We hypothesize that higher levels of consumption of PBFs will confer protection for risk of MetS components and MetS, and higher consumption of HealthyPBFs will confer greater protection than consumption of PBFs generally. We will utilize data from the National Health and Nutrition Examination Survey (NHANES) to examine the association of healthy and unhealthy plant-based diet with MetS and MetS components in a nationally representative sample of US adults.

Methods

Study design

This study is a secondary analysis of NHANES 2015–16 cross-sectional data. NHANES includes surveys (demographic, socioeconomic, dietary and health-related questions), two 24-h dietary recalls (24hDRs) and a physical examination (medical, dental and physiological measurements), as well as laboratory tests on plasma and urine samples collected via NHANES Mobile Examination Centers (MECs). Oversampling of select subgroups (Hispanics, African Americans, Asians, people ≥ 185% of Health and Human Services poverty guidelines, and people over 60) and sampling weights are used to ensure representative data.²⁰ In 2015–16, the total sample was 9544 adults.²¹

Data preparation

Components for inclusion in this secondary analysis are completion of three NHANES evaluation methods: (i) two 24hDR’s, (ii) Family and Sample Person Questionnaires and (iii) complete physical examination. An additional criterion was being ≥20 years old. Information from NHANES Demographic Variables and Sample Weights were used to create variables for sex (male = 0, female = 1), age (⁠|$\ge$|20 and |$<$|40 years, |$\ge$|40 and |$<$|60 years, |$\ge$|60 years), income (<$25K, $25–|$<$|⁠$65K, |$\ge$|⁠$65K) and race/ethnicity (1 = Mexican American, 2 = Other Hispanic, 3 = Non-Hispanic White, 4 = Non-Hispanic Black and 5 = Other).²¹

Plant-based food variables

The amount of PBFs consumed was calculated as average servings per day, using the Food Patterns Equivalents Database (FPED) 2015–16, based on the USDA Food Pattern definitions presented in the 2015–20 Dietary Guidelines for Americans (DGA).²² The FPED disaggregates the 8700 foods and beverages in the Food and Nutrient Database for Dietary Studies (FNDDS)²³ into the amounts of Fruits, Vegetables, Grains, Protein Foods, Dairy, Oils, Added Sugars, Solid Fats and Alcoholic Drinks present in each food.²² Combined with the data from the NHANES 2015–16 24hDRs, daily servings of each food group were calculated and averaged over the 2 days of dietary recalls, then combined to create an AllPBF variable and a HealthyPBF variable. The categories included in the total plant-based foods variable (AllPBFs) were fruits (including fruit juice), vegetables (including legumes), nuts and seeds, soy foods, grains (both whole and refined), oils and added sugars. The HealthyPBF variable, in alignment with NOVA guidelines,¹⁵ excluded fruit juice (due to high added sugars), refined grains, added sugars and oils.

MetS and component variables

The National Cholesterol Education Adult Program Treatment Panel III (NCEP ATP III) (2005) has defined MetS as the presence of three or more of the following conditions: hyperglycemia (glucose |$\ge$|100 mg/dl), abdominal obesity (waist circumference |$\ge$|88 cm in women and |$\ge$|102 cm in men), hypertriglyceridemia (⁠|$\ge$|150 mg/dl), hypertension (⁠|$\ge$|130/85 mm Hg or on treatment for hypertension) and/or low-HDL cholesterol (<40 mg/dl in women and <50 mg/dl in men).²⁴ For this analysis, MetS components were coded as dichotomous variables where 0 = within guidelines and 1 = at or above guidelines. A variable for MetS was also created, where the occurrence of ≥3 MetS components was coded 0 = no MetS and 1 = MetS.

Analyses

All NHANES 2015–16 data were weighted as directed in NHANES analytic guidelines, using weights provided in the Laboratory dataset, as this was the smallest of the merged datasets (wtsaf2yr), as well as appropriate sampling unit, stratum and VCE variables²¹ as provided in the NHANES Demographic file. We incorporate the sampling weights in all statistical analyses presented here.

Before analysis, the data were assessed for missingness. The independent variables (IVs) (AllPBF and HealthyPBF) were found to have 30.1% missing data, well above the recommended 10% level. Further, these data were not Missing Completely at Random (MCAR); therefore, multiple imputation was performed. Because the dataset included both continuous and categorical variables, the Multiple Imputation for Chained Equations (MICE) method²⁵ was used to generate the imputed values. In alignment with both White²⁵ and Bodner’s²⁶ recommendations to set the number of imputations to be higher than the percentage of incomplete cases, forty iterations were run. All variables in the final model were included in the imputation: AllPBF, HealthyPBF, all five MetS components (hyperglycemia, abdominal obesity, hypertriglyceridemia, hypertension and low HDL) and confounders (age, gender, race/ethnicity and income). Diagnostics indicate that error terms were stable, and both relative variance increase and fraction of missing information (FMI) were small and below suggested limits.²⁵ Trace plots for variables with the highest FMI (HealthyPBF and AllPBF) were examined (Appendix 1) and found to demonstrate good convergence with no discernable trend and relatively constant predicted values.

Descriptive statistics (mean and SD for continuous variables, and proportions for categorical and dichotomous variables) were generated. Four logistic regressions were run to assess the associations of MetS components and MetS (dependent variables, DVs) with AllPBFs and healthy PBFs (IVs), both crude and adjusted. Contrasts were generated to assess significant interactions, which were added to the final, adjusted model, and results were stratified by age. Finally, marginal probabilities were generated for the final adjusted and stratified models. For logistic regression analyses, referent categories for confounders were ‘Male’, ‘|$\ge$|20 and <40 years’ ‘<$25 000’, and ‘Non-Hispanic White’. All tests were conducted with a significance level of |$\alpha$| = 0.05, using Stata/SE 16.1 (Stata Corp, 4905 Lakeway Dr., College Station, TX)

Results

Descriptive statistics for MetS components are reported in Table 3. Per NCEP ATP III guidelines, hypertension is defined as a ratio of diastolic/systolic equal or above 130/80²⁴; both elements are reported in Table 1.

Table 1

Descriptive statistics for dependent variables (metabolic syndrome components) from NHANES 2015–16 data

Variable	n	Min	Max	Weighted mean	Weighted SD
Waist Cir. (cm)	5116	58.7	171.6	100.2	16.61
Glucose (mg/dl)	2452	21	479	113.6	41.3
Triglycerides (mg/dl)	2254	15	2141	114.6	96.4
HDL (mg/hl)	5157	6	226	54.2	17.7
Systolic blood pressure (mm Hg)	5315	84	231.3	125.5	18.3
Diastolic blood pressure (mm Hg)	5315	0	138.7	69.8	12.6

Table 2 captures the coded and weighted variables, both IVs and DVs, presented by demographic variables. The majority of the weighted sample was Non-Hispanic White (63.7%), and half made $60K or more annually. MetS components varied significantly by demographics. Men had a higher prevalence of all MetS components except for waist circumference, where women had a higher prevalence. The prevalence of MetS and MetS components was higher in older adults, increasing across higher age categories, except for low HDL, which decreased slightly in prevalence for the |$\ge$|60 years category. Non-Hispanic Whites had higher prevalence of high waist circumference and hypertriglyceridemia compared with other ethnic groups, whereas Non-Hispanic Blacks had a higher prevalence of hypertension. Mexican-Americans had higher prevalence of hyperglycemia compared with other ethnic groups, which was different than Other Hispanics, which had a higher prevalence of low HDL.

Table 2

Weighted demographic composition of sample, all plant-based food (PBF) servings, healthy PBF servings, and percentage of participants at or above cut points for metabolic syndrome components, and metabolic syndrome for NHANES 2015–2016 participants

	% of n (SE)	All PBF servings mean (SD)	Healthy PBF servings mean (SD)	Hyperglycemia % = 1 (Yes)	Waist circum. % = 1 (Yes)	Hypertriglycridemia % = 1 (Yes)	Hypertension % = 1 (Yes)	Low HDL % = 1 (Yes)	METS % = 1 (Yes)
Sex
Male	48.1 (0.01)	45.5 (25.0)	6.9 (4.3)	67.8	50.1	49.6	47.1	58.3	52.6
Female	51.9 (0.01)	36.1 (17.5)	6.3 (4.2)	50.3	71.5	29.2	43.3	8.4	36.8
Age
|$\ge$|20 and |$<$|40 years	36.0 (0.01)	43.3 (22.4)	6.1 (3.8)	40.2	49.5	16.7	19.8	33.9	23.8
|$\ge$|40 and |$<$|60 years	35.9 (0.01)	41.1 (22.4)	6.5 (3.9)	62.3	64.7	34.3	49.3	33.5	49.4
|$\ge$|60 years	28.1 (0.02)	36.6 (18.2)	7.3 (5.1)	77.8	71.6	54.7	72.2	29.0	65.4
Income
|$<$|⁠$25K	15.5 (0.01)	40.3 (37.2)	5.2 (5.2)	64.9	61.7	36.3	51.9	30.1	48.8
$25K–|$<$|⁠$65K	34.4 (0.01)	41.5 (21.6)	6.1 (4.5)	59.2	62.6	36.6	45.9	34.1	46.7
|$\ge$|⁠$65K	50.0 (0.02)	40.3 (17.5)	7.3 (3.7)	56.6	60.5	33.2	42.8	31.4	41.9
Race/ethnicity
Mexican-American	8.3 (0.02)	43.4 (29.9)	6.3 (5.3)	64.7	63.5	32.5	37.7	38.9	43.4
Other Hispanic	6.8 (0.01)	39.9 (29.7)	6.1 (6.1)	54.9	52.1	28.7	37.7	38.8	37.9
Non-Hispanic White	63.7 (0.04)	40.9 (16.5)	6.8 (3.3)	60.3	64.8	37.7	44.9	32.9	47.2
Non-Hispanic Black	11.4 (0.03)	39.4 (29.8)	5.3 (5.4)	47.3	60.2	23.8	57.2	20.8	38.0
Other	9.8 (0.01)	38.3 (23.4)	7.3 (5.2)	59.2	41.7	34.2	43.6	32.6	38.9
Total	100.0	40.6 (0.6)	6.6 (4.3)	58.7	61.1	34.7	45.1	32.4	44.4

After adjusting for confounding by sex, age, income and ethnicity, HealthyPBF was significantly associated with elevated waist circumference (P = 0.017; odds ratio, OR 0.96, 95% confidence interval, CI 0.94–0.98) and MetS (P = 0.006; OR 0.96, 95% CI 0.93–0.99). Holding other covariates in the adjusted model constant, a 1-unit increase in daily servings of HealthyPBD was associated with a 4% lower risk for prevalence of elevated waist circumference and a 4% decrease in risk of MetS. For the significant association between HealthyPBF and elevated waist circumference, increased risk was found for being female and being in the older age categories, whereas lower risk was significantly associated with being ‘Other’ race/ethnicity. No interactions were significant for this relationship. For the significant association between HealthyPBF and risk for MetS, being Non-Hispanic Black (compared with Non-Hispanic White) and making |$\ge$|⁠$65K (compared with making |$>$|⁠$25K) were significantly protective for risk of MetS in the adjusted model. For this model, interactions between age and sex were found to be significant, and therefore, results were stratified by age.

Marginal probabilities for each decile of HealthyPBF revealed significant changes at every decile increase (i.e. from 0 to 10) for the association of MetS, suggesting a dose–response inverse relationship for probability of MetS as daily servings HealthPBF increased (Fig. 1). This pattern was also present for elevated waist circumference, but was not significant at every decile. Instead, HealthyPBF was significantly associated with increased probability of elevated Waist Circumference at lower deciles (1 and 2), then significant again at higher deciles (8, 9 and 10), but for decreased probability. For probability of both waist circumference and MetS, significant protective associations of HealthyPBF were present at the higher deciles of daily servings of HealthyPBF, and either neutral or increased at lower deciles.

Fig. 1

Marginal Risk across deciles of healthy plant-based food for elevated waist circumference and metabolic syndrome following fully adjusted logistic regressions.

Results of stratified and adjusted logistic regressions are presented in Table 3. In the younger age categories (⁠|$\ge$|20 and |$<$|40 years, |$\ge$|40 and |$<$|60 years), servings of Healthy PBF conferred no significant reduction in risk for MetS, while being Female was significantly protective in each category. However, the protective effect of being Female for risk of MetS diminished in the |$\ge$| 60 years category. For participants |$\ge$| 60 years old, higher daily servings of HealthyPBF significantly lowered risk for MetS.

Table 3

Multivariate logistic regression stratified by age (includes adjustment for demographic covariates) for the significant association of healthy plant-based food (HealthyPBF) with metabolic syndrome

	MetS |$\boldsymbol{\ge}$|20 and |$<$|40 years			MetS |$\boldsymbol{\ge}$|40 and |$<$|60 years			MetS |$\boldsymbol{\ge}$|60 years
	OR	P-value	95% CI	OR	P-value	95% CI	OR	P-value	95% CI
HealthyPBF	0.97	0.40	0.90–1.05	0.96	0.07	0.91–1.01	0.96	0.04	091–1.00
Sex
Male	1	–	–	1	–	–	1	–	–
Female	0.21	<0.001	0.12–0.36	0.49	0.003	0.32–0.75	0.75	0.26	0.44–1.28
Race/ethnicity
Mexican-American	1.27	0.55	0.54–2.99	0.94	0.84	0.70–1.71	1.08	0.85	0.49–2.38
Other Hispanic	0.88	0.77	0.35–2.22	0.61	0.12	0.53–1.09	0.96	0.87	0.55–1.67
Non-Hispanic White	1	–	–	1	–	–	1	–	–
Non-Hispanic Black	0.58	0.12	0.29–1.17	0.74	0.10	0.53–0.94	0.84	0.5	0.49–1.44
Other	1.06	0.89	044–2.52	0.60	0.08	0.55–1.27	0.94	0.87	0.42–2.11
Income
|$<$|$25K	1	–	–	1	–	–	1	–	–
$25K–|$<$|$65K	0.90	0.72	0.48–1.68	0.88	0.54	0.58–1.35	0.80	0.49	0.41–1.58
|$\boldsymbol{\ge}$|$65K	0.98	0.94	0.51–1.88	0.50	0.007	0.32–0.80	0.63	0.08	0.37–1.07

Bolded odds ratios (ORs) are significant at p<0.05.

Discussion

Main findings of this study

In our study, increased daily servings of HealthyPBFs were significantly associated with reductions in risk of elevated waist circumference and MetS in adjusted models, whereas AllPBF was not. This outcome refutes our primary hypothesis—that consuming higher amounts of PBFs generally would be protective for risk of MetS—but confirms our secondary hypothesis that consuming health plant-based food would be more protective than plant-based food generally. It appears that the quality of a plant-based diet is critical for reducing risk of MetS. This result aligns with other research into diet and chronic disease risk, such as Rauber et al.’s findings that a diet higher in ultra-processed foods contained markedly higher amounts of added sugars and saturated fat, and leading to increased risk of chronic disease prevalence.²⁷ Although increasing the consumption of HealthyPBFs appears to be a feasible strategy for reducing risk of MetS and associated chronic disease in US adults, the distinction of minimally processed plant foods is critical, as illustrated by our findings that only HealthyPBF conferred protection from MetS and elevated waist circumference.

Although the effect size of the association of HealthyPBF with MetS and elevated waist circumference were small (both ORs = 0.96), this 4% reduction in risk for each 1-unit change in daily servings of HealthyPBF translates into a substantial effect for small changes in diet. Age-stratified results suggest that the protective influence of sex on risk of MetS diminished in those females |$\ge$|60 years old, whereas the benefits of consuming HealthyPBF became significant. This is in agreement with other research, which found that lower risk for chronic disease among females attenuated or even reversed in later years, generally tracking with the end of reproductive years.^28–31

What is already known on this topic

Our findings align with other evidence that a plant-based diet is an important and modifiable risk factor for MetS. For example, the Adventist Health Study-2 found that prevalence of MetS increased as the proportion of plants in the diet decreased (trend P-value < 0.001).³² Similar significant and inverse associations between a plant-based diet and MetS have been found in Chinese adults,³ Iranian adults with impaired glucose tolerance,³³ and a non-representative sample of the US population.³⁴ Research has also found associations for plant-based diets and reductions in BMI, including among Seventh Day Adventists in the US,³² adults in the UK,³⁵ and in a large cohort of Taiwanese adults.³⁶ Although this analysis did not detect improvements in hyperglycemia in association with increased servings of PBFs, other studies have found associations between a plant-based dietary pattern (i.e. vegan or vegetarian eating patterns) and improved fasting glucose.^32,37,38 Fewer studies have found associations between plant-based diets and improvements in hypertriglyceridemia, and none have found associations with improved HDL levels.^39,40

What this study adds

Our study indicates that increasing consumption of HealthyPBFs may offer a viable strategy for reducing risk of MetS in a manner that is not absolutist, as a recommendation to avoid animal products altogether might be, and therefore more feasible and acceptable at the population level. To anchor these results in the Dietary Guidelines for Americans 2015–20 definition of a serving, a 1 unit (or 1 serving) change in HealthyPBF would mean consuming an extra 1/2 cup of fresh fruits or vegetables, or an additional 1-ounce serving of nuts and seeds daily in order to reduce the risk of elevated waist circumference and MetS by 4%. It is important to note that no previous studies utilized a nationally representative US sample or 24hDR to assess diet. Instead, other research has used non-representative populations and/or food frequency questionnaires (FFQ) to assess diet (Nurse’s Health Studies, Adventist Health Studies, European Prospective Investigation into Cancer and Nutrition etc.), which are considered less accurate than the repeated 24hDRs available in NHANES data,^41–44 or have been conducted outside the US.²⁷ Although some of the associations we detected may have been due to a substitution mechanism, whereby consuming more servings of HealthyPBFs displaces the consumption of less healthy foods, this is a beneficial by-product of this strategy, rather than a disadvantage. Finally, our approach has the advantage of utilizing a continuous measure of the daily servings of PBFs, rather than categorizing consumption patterns (i.e. vegan versus Western diet), thus capturing a more realistic and accurate measure of consumption.

Limitations of this study

Limitations of this study include a lack of precision in categorizing foods consumed as either plant-based or not, which was dictated by the granularity offered by the FPED database. For example, in the FPED, soy milk was aggregated with dairy, and so was not included in either category of PBFs (AllPBF or HealthyPBF), even though it is technically a plant food. These challenges to accurate classification of foods may have obscured associations between PBFs and MetS components and/or MetS, making these analyses more conservative. In addition, these data are cross-sectional, limiting the characterization of a causal relationship. Further, these data did not allow a nuanced analysis of other important confounders of dietary quality and chronic disease risk, such as stress, acculturation and environmental factors.

Future research into associations between plant-based food, chronic disease and chronic disease risk factors may benefit from more detailed and flexible food codes to allow for greater precision in classifying foods as plant-based, or not. Large randomized control trails (RCTs) to prospectively assess the impact of plant-based food consumption on MetS components and MetS are needed to better understand the mechanisms by which this relationship may operate, explore potential causal relationships, and provide better guidance for the support and encouragement of optimal diets for diverse populations.

Christine E.S. Jovanovic, PhD, MPH

Deanna M. Hoelscher, Professor

Baojiang Chen, Professor

Nalini Ranjit, Professor

Alexandra E. van den Berg, Professor