Association of Healthy Eating Index-2020 scores and its components with epilepsy in US adults: a cross-sectional study of NHANES 2005–2020

Abstract

Background

Epilepsy is a prevalent neurological disorder associated with a substantial global health burden. Emerging evidence suggests that dietary intake may influence its risk and progression. This cross-sectional study aimed to examine the association between overall dietary quality, as measured by the Healthy Eating Index-2020 (HEI-2020), and the risk of epilepsy.

Methods

Data were obtained from the National Health and Nutrition Examination Survey (NHANES, 2005–2020). Dietary intake was assessed using two 24-hour recalls. Multivariable logistic regression, restricted cubic splines (RCS), subgroup analyses, quantile g-computation (Qgcomp), and weighted quantile sum (WQS) regression were employed to evaluate associations between HEI-2020 and epilepsy, and the joint effects of dietary components.

Results

The epilepsy group exhibited substantially lower HEI-2020 scores than the non-epilepsy group (p = 0.006). The highest quartile group (Q4) had a 28% lower risk of epilepsy compared to the lowest group (Q1) (OR = 0.720, 95% CI: 0.549–0.945, p = 0.018). RCS analysis further suggested a linear association of HEI-2020 with epilepsy risk (p > 0.050 for nonlinearity). Both WQS regression and Qgcomp models identified a consistent negative correlation of the combined effect of dietary components with epilepsy risk (WQS: β=-0.453, 95% CI: -0.591 to -0.315, Qgcomp: β=-0.497, 95% CI: -0.340 to -0.654).

Conclusions

The current investigation demonstrated a negative correlation between HEI-2020 scores and epilepsy risk. A healthier diet (as reflected by higher HEI-2020 scores) is associated with a lower risk of epilepsy. This observational association suggests that future research should prioritize randomized controlled trials or prospective cohort studies examining dietary patterns involving an increased intake of high-quality protein, vegetables, low-sugar fruits, and healthy fatty acids, and a decreased intake of refined grains and high-sodium foods. However, longitudinal studies and randomized intervention trials are still needed to confirm the causal relationship and quantify the actual benefits of dietary optimization for the primary prevention of epilepsy.

Trial registration

Not applicable.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12883-026-04763-2.

Introduction

Epilepsy, a persistent neurological condition characterized by pathological hypersynchronous neuronal activity, manifests clinically as recurrent, transient disturbances in the central nervous system (CNS). As a common neurological condition worldwide, epidemiological data indicate that epilepsy affects around 70 million individuals globally [1]. Pharmacological intervention represents the primary treatment modality for epilepsy. Over the last 30 years, more than 20 anti-seizure medications (ASMs) have been developed that exhibit reduced adverse effects and improved ease of use, while potentially mitigating the severity of epileptic symptoms. Nevertheless, approximately one-third of individuals exhibit drug resistance [2]. Recurrent epileptic seizures increase the risk of accidental injury, cognitive impairment, and psychiatric comorbidities (e.g., depression, anxiety), severely impacting quality of life and social functioning while imposing a considerable socioeconomic burden [3]. This clinical reality presents a significant challenge to epilepsy prevention and management. In recent years, increasing interest has arisen regarding alternative and nutritional therapies for treatment-refractory epilepsy, including neuromodulation, ketogenic dietary therapies, and surgery. These approaches offer new hope by improving treatment outcomes and quality of life for individuals with epilepsy, and they provide important implications for future research and therapy. These advancements motivate ongoing efforts to identify safer and more effective therapeutic strategies.

In this context, exploring modifiable dietary factors holds significant importance for the primary prevention and disease management of epilepsy. Recent research has demonstrated associations between dietary vitamin K1 intake, comprehensive dietary antioxidant index, vitamin B1 intake, dietary carbohydrate-to-energy ratio, and epilepsy [4–7]. However, the majority of these investigations have examined the effects of individual nutrients or specific dietary components. A systematic evaluation from the perspective of overall dietary patterns is lacking. This disparity underscores the necessity of meticulously researching the correlation of overall dietary quality with epilepsy. The Healthy Eating Index (HEI), developed in alignment with the Dietary Guidelines for Americans (DGA), is regarded as the gold standard for quantifying dietary quality. It has been widely implemented as a reliable tool for evaluating overall dietary quality. Substantial evidence has revealed that an elevated HEI score is linked to improved health status and reduced risk of various conditions, including cystic fibrosis (CF) [8], periodontitis [9], and stroke [10]. This study innovatively utilizes the large-sample data from the National Health and Nutrition Examination Survey (NHANES) database to systematically examine the correlation of Healthy Eating Index-2020 (HEI-2020) scores with epilepsy risk for the first time. The study also quantifies the contribution of different dietary factors through component analysis. The aim is to provide evidence-based support for dietary interventions in epilepsy management.

Materials and methods

Information collection

NHANES, a nationwide cross-sectional survey program implemented by the National Center for Health Statistics (NCHS), a division of the Centers for Disease Control and Prevention (CDC), served as the data source for the current study. Official data are accessible at https://www.cdc.gov/nchs/nhanes/index.htm [11]. A biennial survey design is employed by this program to comprehensively evaluate dietary patterns and health metrics of the non-hospitalized U.S. population. The goal is to contribute scientific evidence for informed health policy decisions. According to institutional guidelines and national regulations, written informed consent was not required for the present study. The analysis incorporated eight consecutive NHANES survey waves collected between 2005 and 2020, yielding an initial sample size of 76,496 people. To ensure the quality of the research, rigorous eligibility criteria were established for the study population. We excluded participants who were < 20 years of age (n = 33,075), had missing dietary or epilepsy medication data (n = 10,040), or had incomplete data on key covariates (marital status, race, sex, education level, age, body mass index [BMI], diabetes mellitus [DM], poverty income ratio [PIR], hypertension, smoking status, and alcohol consumption) (n = 5,052). A flowchart illustrating the selection process is illustrated in Fig. 1.

Fig. 1.

Selection process

HEI-2020

Created by the U.S. Department of Agriculture’s Center for Nutrition Policy and Promotion, HEI-2020 serves as a dietary evaluation tool aligned with the 2020–2025 DGA [12]. The index incorporates 13 distinct dietary elements: dark green vegetables and legumes, total vegetables, sodium, total fruits, whole grains, refined grains, whole fruits, total protein foods, dairy, seafood and plant proteins, fatty acids, added sugars, and saturated fats. The possible range for HEI-2020 scores is 0-100. Higher scores indicate a healthier diet [13]. In the 2005–2020 NHANES surveys, a relatively consistent dietary assessment methodology was adopted. The NHANES collected 24-hour dietary recall data for each participant on two non-consecutive days. This data included all meals (breakfast, lunch, dinner, and snacks).

Definition of epilepsy

During data collection, NHANES assessed whether participants had taken any prescription medications in the past 30 days and recorded the generic names, primary reasons for use, and any corresponding International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) codes. Therefore, the case selection for epilepsy patients in this study was based on the report by Wen et al. [14] and identified through the following method. Patients who had taken ASMs within the past 30 days and whose main reason for taking the medication was coded as G40 (epilepsy and recurrent seizures) according to ICD-10-CM were included [15, 16]. Since the NHANES lacks detailed clinical diagnoses, seizure frequency, and lifetime epilepsy history, this drug-based definition is a common and practical method for population-based studies.

Covariate

Covariate selection in this study was based on prior research and clinical protocols [17, 18]. Covariates encompassed the following demographic characteristics and clinical indicators: sex (male, female), age, race/ethnicity (Mexican American, non-Hispanic Black, non-Hispanic White, other Hispanic, other race), marital status (married/cohabitation, unmarried, widowed/divorced/separated), education level (above high school, high school or below), and PIR. Smoking status was categorized in accordance with the NHANES questionnaire. Adults ≥ 20 years of age self-reported smoking habits. They were divided into three groups: never smokers (smoked ≤ 100 cigarettes in their lifetime), former smokers (smoked > 100 cigarettes in their lifetime but have quit), and current smokers (smoked > 100 cigarettes in their lifetime and currently smoke daily or occasionally). This categorization is consistent with previous literature [19]. Alcohol consumption was simplified into a dichotomous variable (alcohol consumption or no alcohol consumption) based on previous studies [20]. BMI was classified as obese (≥ 30.0 kg/m²), overweight (25.0 to 29.9 kg/m²), and normal/underweight (< 25.0 kg/m²) based on World Health Organization (WHO) standards. Hypertension was defined [21, 22] as a self-reported physician diagnosis or a systolic blood pressure (SBP) ≥ 130 mmHg or a diastolic blood pressure (DBP) ≥ 80 mmHg. DM diagnosis was determined using the following criteria [23]: (i) self-reported physician-confirmed history, (ii) current use of insulin, (iii) use of hypoglycemic medications to lower blood glucose (BG), and (iv) laboratory-measured glycated hemoglobin (HbA1c) ≥ 6.5% or fasting blood glucose (FBG) ≥ 6.0 mmol/L. Fulfillment of any of these criteria resulted in a classification of DM.

Statistical analysis method

Consistent with NHANES’s complex survey design, all analyses in this study applied sampling weights and accounted for clustering/stratification to preserve national representativeness. Statistical descriptions of continuous variables were performed using different methods based on their distribution characteristics. Non-normally distributed variables were expressed as interquartile ranges and medians. Intergroup comparisons were conducted utilizing the Mann-Whitney U test. Normally distributed variables were displayed as means ± standard deviations (SDs). Intergroup comparisons were implemented via t-tests. Categorical variables were presented as percentages and counts. Intergroup comparisons were carried out utilizing the chi-square (χ²) test. Variance inflation factors (VIFs) were computed to assess multicollinearity among covariates. All included variables had VIF values < 5, indicating the absence of significant multicollinearity in the model. Weighted multivariable logistic regression (MLR) models were utilized to analyze the association of HEI-2020 with epilepsy. Three progressively adjusted models were constructed. Model 1 was a crude model with no adjustments. Model 2 made adjustments for demographic variables (race, age, sex). Beyond the adjustments in Model 2, Model 3 included adjustments for socioeconomic factors (marital status, education level, PIR) and clinical factors (smoking, alcohol consumption, BMI, hypertension, DM). To explore the possible nonlinear correlation of HEI-2020 with epilepsy, an RCS model was constructed with knot locations defined at the 5th, 35th, 65th, and 95th percentiles. To assess effect heterogeneity and interactions across different populations, subgroup analysis was carried out. The joint effect of HEI-2020 components on epilepsy was evaluated using the following two methods. First, the weighted quantile sum (WQS) regression model [24] was used to assess the cumulative effects of each component and the contribution of individual components. Given that the WQS model requires all component effects to be in the same direction, we further used the quantile g-computation (Qgcomp) model, which allows for both positive and negative associations of components, for a supplementary analysis to overcome the limitations of WQS. To verify the robustness of the results, two sensitivity analyses were implemented. First, the consistency of the analysis results was compared between weighted and unweighted data. Second, the main analysis process was repeated following 1:1 nearest neighbor matching to balance baseline characteristics between the case and control groups. All data analyses were conducted in R version 4.3.3 (R Foundation), with statistical significance defined as two-tailed P-values < 0.05.

Results

Baseline characteristics of subjects

This NHANES analysis incorporated 28,329 subjects, with 1,724 (6.26%) identified as having epilepsy. The results (Table 1) indicated significant intergroup differences with respect to marital status, sex, race, education level, age, PIR, hypertension, DM, smoking status, BMI, and alcohol consumption (p < 0.05 for all). Furthermore, compared to the non-epilepsy group, the epilepsy group exhibited a substantially lower HEI-2020 score (p = 0.006).

Table 1.

Baseline characteristics of subjects by epilepsy

Characteristic	Overall	Non-epilepsy group	Epilepsy group	p-value
Sex				< 0.001
Female	14,622 (51.69%)	13,616 (51.09%)	1,006 (60.66%)
Male	13,707 (48.31%)	12,989 (48.91%)	718 (39.34%)
Age	47.00 (33.00, 60.00)	46.00 (33.00, 60.00)	55.00 (41.00, 65.00)	< 0.001
Race				< 0.001
Mexican American	4,036 (7.95%)	3,900 (8.20%)	136 (4.28%)
Non-Hispanic Black	6,197 (10.68%)	5,894 (10.91%)	303 (7.28%)
Non-Hispanic White	12,918 (69.04%)	11,897 (68.36%)	1,021 (79.21%)
Other Hispanic	2,528 (5.25%)	2,377 (5.30%)	151 (4.49%)
Other Race	2,650 (7.08%)	2,537 (7.23%)	113 (4.74%)
Education				0.028
< High school diploma	6,071 (13.70%)	5,671 (13.62%)	400 (14.81%)
> High school diploma	15,713 (62.98%)	14,828 (63.26%)	885 (58.76%)
High school diploma/equivalent	6,545 (23.32%)	6,106 (23.12%)	439 (26.43%)
Marital status				< 0.001
Married/cohabitation	17,139 (63.59%)	16,225 (63.93%)	914 (58.47%)
Unmarried	6,218 (18.28%)	5,634 (17.62%)	584 (28.22%)
Widowed/divorced/separated	4,972 (18.13%)	4,746 (18.45%)	226 (13.31%)
PIR	3.10 (1.55, 5.00)	3.14 (1.58, 5.00)	2.43 (1.18, 4.38)	< 0.001
HEI-2020	50.39 (42.51, 59.32)	50.48 (42.57, 59.47)	49.01 (41.36, 57.74)	0.006
Hypertension				< 0.001
No	12,974 (50.72%)	12,449 (51.64%)	525 (36.98%)
Yes	15,355 (49.28%)	14,156 (48.36%)	1,199 (63.02%)
DM				< 0.001
No	23,485 (87.31%)	22,276 (87.96%)	1,209 (77.59%)
Yes	4,844 (12.69%)	4,329 (12.04%)	515 (22.41%)
Smoking status				< 0.001
Former	7,257 (25.60%)	6,691 (25.25%)	566 (30.87%)
Never	15,471 (55.06%)	14,777 (55.88%)	694 (42.71%)
Current	5,601 (19.35%)	5,137 (18.87%)	464 (26.42%)
BMI (kg/m²)				< 0.001
<25	7,774 (29.01%)	7,405 (29.29%)	369 (24.73%)
25–30	9,245 (32.79%)	8,780 (33.28%)	465 (25.38%)
>30	11,310 (38.21%)	10,420 (37.43%)	890 (49.89%)
Alcohol consumption				< 0.001
No	9,080 (27.21%)	8,427 (26.66%)	653 (35.49%)
Yes	19,249 (72.79%)	18,178 (73.34%)	1,071 (64.51%)

Abbreviations: PIR Poverty income ratio, HEI-2020 Healthy Eating Index-2020, BMI Body mass index, DM Diabetes mellitus

Logistic regression-RCS curve

Three weighted MLR models were constructed. Model 1 made no adjustments. Model 2 incorporated adjustments for demographic factors, including marital status, sex, race, PIR, age, and education level. Model 3 encompassed further adjustments for lifestyle and health condition variables, including smoking status, hypertension, alcohol consumption, BMI, and DM, besides the adjustments in Model 2. All three models revealed (Table 2) that HEI-2020 was negatively linked to the risk of epilepsy [(Model 1: OR = 0.989, 95% CI: 0.982–0.996), (Model 2: OR = 0.980, 95% CI: 0.973–0.987), (Model 3: OR = 0.990, 95% CI: 0.982–0.997)]. The continuous variable analysis demonstrated that a 1% decrease in epilepsy risk was observed for each 1-unit elevation in HEI-2020 (Model 3: OR = 0.990, 95% CI: 0.982–0.997, p = 0.010). Quartile analysis revealed a dose-response trend. In Model 3, Q4 exhibited a 28% decrease in epilepsy risk compared to Q1 (OR = 0.720, 95% CI: 0.549–0.945, p = 0.018). The RCS curve results displayed a linear association of HEI-2020 with epilepsy risk (nonlinear p-value > 0.05) (Fig. 2). This result indicates an inverse association between superior dietary quality and epilepsy risk.

Table 2.

Relationship between HEI-2020 and epilepsy (weighted)

Participants	Model 1		Model 2		Model 3
	OR (95% CI)	p-value	OR (95% CI)	p-value	OR (95% CI)	p-value
HEI-2020
Continuous	0.989 (0.982, 0.996)	0.002	0.980 (0.973, 0.987)	< 0.001	0.990 (0.982, 0.997)	0.01
Q1	ref.		ref.		ref.
Q2	0.904 (0.730, 1.118)	0.348	0.815 (0.651, 1.020)	0.074	0.893 (0.713, 1.119)	0.323
Q3	0.912 (0.737, 1.129)	0.396	0.761 (0.613, 0.947)	0.015	0.946 (0.758, 1.179)	0.616
Q4	0.710 (0.559, 0.901)	0.005	0.535 (0.413, 0.693)	< 0.001	0.720 (0.549, 0.945)	0.018

Model 1: crude model; Model 2: made adjustments for marital status, sex, race, PIR, age, and education level; Model 3: included adjustments for marital status, sex, race, education, PIR, age, smoking status, alcohol consumption, BMI, hypertension, and DM

Fig. 2.

Relationship between the variable and the predicted probability

Subgroup analyses

Subgroup analyses (with Q1 as the reference group) revealed (Table 3) that the correlation of HEI-2020 with epilepsy risk exhibited population-specific differences. A significant inverse association was observed among females in Q4 (OR = 0.632, 95% CI: 0.474–0.843), but no marked differences were noted in males (p = 0.637). Marital status analysis illustrated a significant 38.5% reduction in epilepsy risk in the Q4 group among married/cohabitating individuals (OR = 0.615, 95% CI: 0.425–0.891, p = 0.011). Race-specific analyses demonstrated a significant risk reduction in the higher HEI-2020 groups among Mexican Americans (OR = 0.423, p = 0.027) and non-Hispanic Blacks (OR = 0.572–0.665, p < 0.05). Notably, a notable inverse association was also observed in the Q4 group for both high school-educated participants (OR = 0.558–0.561, p < 0.050) and obese individuals (BMI ≥ 30, OR = 0.685, p = 0.026). Furthermore, subjects with DM (OR = 0.532, p = 0.006) or hypertension (OR = 0.670, p = 0.010) appeared to benefit more significantly from high-quality diets. Despite these subgroup differences, all interaction terms were statistically non-significant (p for interaction > 0.05).

Table 3.

Correlation of HEI-2020 with epilepsy according to different variables

Subgroup	Q1		Q2		Q3		Q4		p for interaction
	OR (95% CI)	p-value	OR (95% CI)	p-value	OR (95% CI)	p-value	OR (95% CI)	p-value
Sex									0.68
Female	ref.		0.879 0.664, 1.165	0.367	0.890 0.667, 1.188	0.427	0.632 0.474, 0.843	0.002
Male	ref.		0.909 0.639, 1.293	0.592	1.037 0.727, 1.477	0.841	0.907 0.602, 1.366	0.637
Race									0.053
Mexican American	ref.		0.943 0.416, 2.137	0.887	0.814 0.349, 1.898	0.629	0.423 0.198, 0.906	0.027
Non-Hispanic Black	ref.		0.665 0.445, 0.993	0.046	0.572 0.351, 0.931	0.025	0.612 0.370, 1.011	0.055
Non-Hispanic White	ref.		0.922 0.702, 1.210	0.553	1.069 0.819, 1.395	0.622	0.784 0.572, 1.075	0.129
Other Hispanic	ref.		0.428 0.215, 0.853	0.016	0.655 0.294, 1.458	0.296	0.440 0.187, 1.037	0.06
Other Race	ref.		1.769 0.844, 3.708	0.129	0.457 0.192, 1.086	0.076	0.742 0.355, 1.549	0.423
Education									0.078
< High school diploma	ref.		0.931 0.611, 1.418	0.736	1.070 0.694, 1.651	0.756	1.274 0.708, 2.295	0.416
> High school diploma	ref.		1.062 0.796, 1.419	0.679	1.154 0.853, 1.560	0.349	0.738 0.511, 1.064	0.103
High school diploma/equivalent	ref.		0.637 0.398, 1.018	0.059	0.561 0.377, 0.834	0.005	0.558 0.344, 0.906	0.019
Marital status	ref.								0.161
Married/cohabitation	ref.		0.760 0.579, 0.996	0.047	0.979 0.728, 1.316	0.886	0.615 0.425, 0.891	0.011
Unmarried	ref.		1.195 0.836, 1.707	0.325	0.927 0.666, 1.290	0.651	0.844 0.566, 1.260	0.403
Widowed/divorced/separated	ref.		0.920 0.458, 1.849	0.813	0.732 0.431, 1.242	0.245	1.100 0.576, 2.098	0.771
Hypertension									0.77
No	ref.		0.936 0.641, 1.366	0.729	1.014 0.740, 1.391	0.928	0.796 0.475, 1.335	0.384
Yes	ref.		0.856 0.665, 1.100	0.221	0.911 0.689, 1.205	0.51	0.670 0.495, 0.907	0.010
DM									0.449
No	ref.		0.894 0.685, 1.167	0.405	0.962 0.749, 1.236	0.761	0.765 0.565, 1.035	0.082
Yes	ref.		0.870 0.576, 1.315	0.505	0.832 0.579, 1.195	0.315	0.532 0.341, 0.828	0.006
Smoking status									0.095
Former	ref.		0.893 0.619, 1.287	0.540	0.699 0.502, 0.974	0.035	0.701 0.471, 1.045	0.080
Never	ref.		1.079 0.722, 1.613	0.707	1.226 0.826, 1.821	0.309	0.746 0.512, 1.087	0.125
Current	ref.		0.717 0.508, 1.012	0.059	0.875 0.568, 1.348	0.541	0.954 0.553, 1.644	0.863
BMI (kg/m²)									0.728
<25	ref.		0.811 0.514, 1.278	0.362	0.868 0.567, 1.327	0.508	0.792 0.425, 1.474	0.458
25–30	ref.		1.127 0.794, 1.598	0.500	0.902 0.584, 1.393	0.639	0.686 0.439, 1.071	0.097
>30	ref.		0.839 0.618, 1.138	0.256	1.025 0.770, 1.364	0.865	0.685 0.491, 0.955	0.026
Alcohol consumption
No	ref.		0.755 0.552, 1.032	0.078	0.892 0.654, 1.217	0.469	0.619 0.429, 0.891	0.011	0.635
Yes	ref.		0.985 0.741, 1.309	0.916	0.985 0.741, 1.308	0.914	0.785 0.579, 1.064	0.117

WQS-Qgcomp

The current study evaluated the correlation of HEI-2020 dietary components with epilepsy risk utilizing two analysis methods: WQS regression and Qgcomp. As illustrated in Table 4, the WQS analysis results indicated a significant inverse correlation of HEI-2020 dietary components with epilepsy risk (β=-0.453, 95% CI: -0.315 to -0.591, p < 0.001). Following adjustment for confounders including race, sex, age, and education level, the WQS model (Fig. 3) identified the top five dietary components showing the strongest inverse associations with epilepsy risk, which were total protein foods (37.7%), total vegetables (14.37%), total fruits (11.34%), vegetables and legumes (10.90%), and whole fruits (7.57%). These findings suggest that the HEI-2020 dietary pattern, particularly a dietary structure rich in high-quality protein and plant-based foods, may be linked to a lower risk of epilepsy. Supplementary analysis using the Qgcomp model (Fig. 4) revealed that refined grains, sodium, green leafy vegetables, and legumes were positively related to epilepsy risk. All other components exhibited a negative correlation. These 13 components jointly affect epilepsy prevalence. In this population, appropriately increasing total protein foods, total vegetables, total fruits, and fatty acid intake is related to a lower risk of epilepsy. Excessive intake of refined grains and sodium may be linked to a higher risk of epilepsy.

Table 4.

Association of HEI-2020 mixture components with epilepsy by weighted quantile regression (WQS) and quantile g-computation (Qgcomp) analyses

	β (95% CI)	p-value
WQS	-0.453 (-0.315, -0.591)	< 0.001
Qgcomp	-0.497 (-0.340, -0.654)	< 0.001

Included adjustments for sex, race, education, PIR, marital status, smoking status, age, alcohol consumption, BMI, DM, and hypertension

Fig. 3.

WQS model regression index weights for HEI-2020 components and epilepsy

Fig. 4.

Qgcomp model regression index weights for HEI-2020 contribution weights of dietary components in the Qgcomp model and epilepsy

Sensitivity analysis

To ensure the robustness of our findings, two distinct approaches to sensitivity analysis were implemented. Initially, the analysis was repeated using unweighted data. The results revealed a negative association of HEI-2020 with epilepsy risk across all models: Model 1 (OR = 0.984, 0.980–0.989), Model 2 (OR = 0.975, 0.971–0.979), and Model 3 (OR = 0.985, 0.980–0.989) (Table S1). In addition, to eliminate potential confounding bias, we employed 1:1 nearest neighbor matching to balance the baseline characteristics of the case and control groups and then reconstructed the logistic regression model.

Following PSM, intergroup differences in marital status, sex, age, race, education level, PIR, DM, hypertension, smoking, BMI, and alcohol consumption were eliminated (Table S2). The regression results consistently demonstrated a negative association of HEI-2020 with epilepsy risk: Model 1 (OR = 0.986, 0.977–0.995), Model 2 (OR = 0.986, 0.977–0.996), and Model 3 (OR = 0.984, 0.974–0.994) (Table S3).

Discussion

This cross-sectional study included 28,329 adult participants in the United States to explore the correlation between overall dietary quality and the risk of epilepsy. Consistent results across multiple models indicated a significant inverse correlation of higher HEI-2020 scores with lower epilepsy risk (p < 0.05 for all models). A high-quality diet pattern was associated with a 28% lower risk of epilepsy. RCS curves displayed a negative dose-response association of HEI-2020 scores with epilepsy risk. Qgcomp and WQS models confirmed that the 13 HEI-2020 dietary components collectively influence epilepsy risk. Total protein foods, vegetables, fruits, and healthy fatty acids contribute the most significantly to a better outcome. Notably, excessive consumption of refined grains and sodium may elevate epilepsy risk. These findings remained consistent across different analytical methods.

By analyzing NHANES data, this study discovered that higher HEI-2020 scores (reflecting healthier dietary patterns) are significantly associated with a lower risk of epilepsy. This result supports the hypothesis proposed by previous studies that dietary patterns may influence neuronal excitability. More importantly, the heterogeneous effects of different HEI-2020 components on epilepsy risk were identified. Higher consumption of vegetables, total protein, fruits, and healthy fatty acids is associated with a lower risk of epilepsy. Higher intake of refined grains and sodium is associated with a higher risk of epilepsy. The potential biological mechanisms and clinical implications are discussed below.

Dietary components associated with lower risk

Total protein and epilepsy risk

Protein is a major component of human nutrition. There have been reports indicating a link between malnutrition and epileptic seizures [25]. Protein intake may affect epilepsy risk through a variety of routes. First, certain amino acids (e.g., taurine) act as neurotransmitter precursors, regulating GABAergic neuron activity and inhibiting abnormal discharges to produce antiepileptic effects [26]. Second, oxidative stress damage represents a key pathophysiological mechanism in epilepsy. Studies have revealed the effects of oxidative damage on neuronal excitability and susceptibility to seizures [27]. Animal experiments have demonstrated that protein restriction may exacerbate oxidative stress. Appropriate supplementation with high-quality protein (e.g., whey protein) can enhance the body’s antioxidant defenses. Additionally, research has illustrated that the Composite Dietary Antioxidant Index (CDAI) is negatively related to epilepsy risk [28]. Thus, a high-antioxidant dietary pattern may help prevent epilepsy. This discovery has considerable potential significance for the development of future strategies in epilepsy treatment and prevention. The current study aligns with previous studies in that total protein was negatively correlated with the risk of epilepsy. This suggests that the association between high-quality protein and reduced epilepsy risk may be related to its antioxidant defense mechanism.

Vegetables, fruits, and potential neuroprotective pathways

Many animal models of epilepsy and individuals with drug-resistant epilepsy exhibit significant inflammatory responses. Studies have revealed that to date, anti-neuroinflammatory drugs are mostly used to treat children with epilepsy who are resistant to common ASMs and difficult to control [29]. Drugs that act on various inflammatory pathways can alleviate epileptic phenotypes. Inhibiting neuroinflammatory processes may be particularly effective in stopping the chain reaction of seizure recurrence [30]. Vegetables and fruits are rich in polyphenols, vitamins C and E, and dietary fiber. These nutrients are associated with anti-inflammatory and antioxidant properties. For optimal nutrition, health, and well-being, consumers should obtain bioactive compounds, antioxidants, nutrients, and phytochemicals from a balanced diet containing a variety of vegetables, fruits, whole grains, and other plant-based foods.

Modulatory role of healthy fatty acids

HEI-2020 emphasizes the intake of unsaturated fatty acids (e.g., omega-3 polyunsaturated fatty acids [ω-3 PUFA]). The current study supports their antiepileptic potential. Research has revealed that dietary supplements containing ω-3 PUFA can reduce seizure frequency in individuals with CNS diseases who are taking anticonvulsant medication [31]. A randomized, placebo-controlled, crossover study found that supplementation with low-dose fish oil (n-3 fatty acids) represents a cost-effective and well-tolerated approach that may decrease seizure frequency [32]. The present study is consistent with these results. One possible reason is that the diet contains low doses of n-3 fatty acids, which help reduce seizures in individuals with epilepsy.

Potential harm of risk dietary components

Refined grains and BG fluctuations

The high glycemic index of refined grains can cause drastic postprandial BG fluctuations, which induce increased neuronal excitability. Studies have indicated that cortical excitability fluctuates with changes in BG levels. This variation is more pronounced in individuals with epilepsy [33]. Animal models illustrate that a high-sugar diet can result in oxidative injury to the hippocampus and abnormal synaptic plasticity. Furthermore, clinical case reports display that refined grains are deficient in B vitamins (e.g., B6, folate), which are key cofactors for GABA synthesis and homocysteine metabolism.

High-sodium diet and neuronal hyperexcitability

Excessive sodium intake may directly interfere with blood-brain barrier integrity and increase brain tissue osmotic pressure. This promotes neuronal depolarization. High-potassium and low-sodium fruits and vegetables may affect neuronal excitability by regulating ion channel stability. Epidemiological studies have also suggested that hypertension (associated with a high-sodium diet) is an independent risk factor for epilepsy. Research has indicated [34] that oxidative stress and dysregulation of synaptic proteins and neurotrophic factors are linked to memory impairment caused by high-salt diets. Oxidative stress and dysregulation of synaptic proteins and neurotrophic factors are also important pathogenic mechanisms of epilepsy. Additionally, studies have revealed that excessive salt intake in the diet can alter the excitability of central sympathetic neural networks [35]. Abnormal neuronal excitability is an important mechanism for the occurrence of epilepsy. Therefore, the current study is consistent with previous studies indicating that excessive salt intake elevates epilepsy risk.

Reverse causality

The cross-sectional design inherently limits causal inference. Even after statistical adjustment, a reverse causal pathway cannot be excluded where epilepsy and its pathological processes precede a deterioration in dietary quality. The observed lower HEI scores in this investigation more likely reflect the cumulative burden of the disease and its consequences. First, epilepsy frequently co-occurs with depression and anxiety [36]. The ensuing emotional eating can reduce dietary diversity. Second, first-line ASMs like valproate and topiramate often induce gastrointestinal discomfort and altered taste perception [37], which may prompt individuals to reduce fruit and vegetable intake while increasing reliance on high-calorie foods. Third, recurrent seizures impair cognitive function [38], diminishing an individual’s capacity for informed dietary decisions. Concurrently, the financial strain of poverty secondary to epilepsy—driven by factors such as unemployment and driving license revocation—can force a shift towards inexpensive, highly processed foods. Therefore, while a low HEI score could be a contributing factor, it is more plausibly a consequence of epilepsy and its downstream sequelae, such as mood disorders, medication side effects, cognitive impairment, and economic hardship.

Robustness and population heterogeneity

Subgroup analyses indicated some heterogeneity in the inverse association between the HEI-2020 and epilepsy risk across different populations. Although females, specific racial groups, and individuals with certain comorbidities exhibited significant protective effects in the highest quartile group, none of the interaction terms reached statistical significance (P > 0.05). This finding suggests that these apparent differences likely stem from random error rather than genuine biological heterogeneity. The observed sex difference merits particular attention. Females demonstrated a marked protective effect (OR = 0.632), while the association for males remained non-significant (OR = 0.907). Nevertheless, the broad confidence interval overlapping with that of females, combined with the non-significant interaction, implies that this disparity could primarily result from limited sample size and consequent reduced statistical power in the male subgroup. Additionally, patients with DM, hypertension, or obesity showed a stronger protective signal, hinting at a potentially critical role for metabolic health in the link between diet and epilepsy. These subgroup findings are exploratory, and the interactions lack statistical significance. Therefore, the results should be interpreted with caution. While definitive inter-subgroup differences cannot be confirmed, these observations generate important hypotheses for future large-scale studies, particularly regarding potential effect modification by metabolic status and sex.

Limitations

This study observed a positive association between dietary patterns and epilepsy risk. However, the inherent constraints of a cross-sectional design limit causal interpretation, and conclusions must be drawn with caution. Several key limitations exist. First, causality cannot be established. The NHANES provides cross-sectional data and lacks key clinical details such as epilepsy duration, seizure type, and time of diagnosis. The temporal sequence between dietary habits and epilepsy onset remains unclear, and reverse causation bias is plausible. Dietary patterns assessed after an epilepsy diagnosis may reflect post-diagnosis adaptations rather than pre-morbid exposures, potentially distorting the true association. Second, while the analysis adjusted for multiple demographic factors and BMI, residual confounding factors from unmeasured variables like occupational exposures and genetic susceptibility are still possible. The impact of these confounders is difficult to quantify and could introduce bias into the effect estimates. Third, dietary exposure assessment relied on two non-consecutive 24-hour dietary recalls to calculate the HEI-2020. This method may inadequately capture long-term dietary habits, and the recall error caused by self-reporting may dilute the true association. Defining epilepsy cases based on the use of ASM also presents limitations. Certain medications (e.g., gabapentin, valproate, topiramate) have non-epilepsy indications. Thus, some non-epilepsy individuals might be misclassified as cases. Conversely, individuals with well-controlled epilepsy or those not on regular medication might be missed. Such non-differential misclassification would also bias results towards the null. Finally, the study sample derives from the non-institutionalized civilian US population. Genetic background, lifestyle, and dietary structures differ substantially in other countries and regions. Therefore, generalizing these findings to non-US populations is inadvisable. Future research should employ prospective longitudinal cohorts or intervention studies to clarify the causal relationship between dietary patterns and epilepsy onset. Such work will provide higher-level evidence to inform preventive strategies and personalized nutritional interventions for epilepsy.

Conclusions

In summary, this cross-sectional analysis utilized NHANES data to investigate the association of HEI-2020 scores with epilepsy prevalence, demonstrating an inverse association. Higher dietary quality (as measured by HEI-2020) is associated with a lower risk of epilepsy. This observational association suggests that future research should prioritize randomized controlled trials or prospective cohort studies examining dietary patterns involving an increased intake of high-quality protein, vegetables, low-sugar fruits, and healthy fatty acids, and a decreased intake of refined grains and high-sodium foods. However, longitudinal studies and randomized intervention trials are still needed to confirm any causal relationship and to quantify the potential impact of dietary optimization for the primary prevention of epilepsy.

Abbreviations

CNS

Central Nervous System

ASMs

Anti-seizure Medications

HEI

Healthy Eating Index

DGA

Dietary Guidelines for Americans

CF

Cystic Fibrosis

NHANES

National Health and Nutrition Examination Survey

HEI-2020

Healthy Eating Index-2020

NCHS

National Center for Health Statistics

CDC

Centers for Disease Control

WHO

World Health Organization

SBP

Systolic Blood Pressure

DBP

Diastolic Blood Pressure

BG

Blood Glucose

FBG

Fasting Blood Glucose

SDs

Standard Deviations

VIFs

Variance Inflation Factors

CDAI

Composite Dietary Antioxidant Index

Authors’ contributions

All authors contributed to the study conception and design. **Writing - original draft preparation: ** [Jifen Wang] **; Writing - review and editing: ** [Xinyong Zhang, Jifen Wang] **; Conceptualization: ** [Kui Duan, Jifen Wang] **; Methodology: ** [Jifen Wang, Changling Chen] **; Formal analysis and investigation: ** [Jifen Wang, Xianni Wang, Lan Ye] **; Funding acquisition: ** [Jifen Wang, Zhanhui Feng] **; Resources: ** [Chunlin Zhang] **; Supervision: ** [Zhanhui Feng], and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.

Funding

This work was supported by the National Natural Science Foundation of China (Grant No. 82360266) and the Qianxinan Medical Research Joint Project (Grant No. Zhoukehe Medical 2024-57).

Data availability

All data generated or analyzed during the course of this study came from the NHANES database.

Declarations

Ethics approval and consent to participate

Data for this study were obtained from the NHANES database, which ensures ethical collection and anonymization to protect individual privacy. As our research uses only pre-existing, publicly available data without any direct interaction with human subjects or new data collection, no additional ethical approval was necessary. Thus, no ethical statement is provided.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.