Causal effect of kimchi intake on HDL-cholesterol levels in middle aged Korean men: a two-sample Mendelian randomization analysis

Abstract

Background

High-density lipoprotein cholesterol (HDL-C) helps remove excess cholesterol, and low levels are a key metabolic risk factor for cardiovascular disease. Although several studies have investigated the association between kimchi consumption and HDL-C levels, these observational findings are often limited by confounding factors and reverse causality.

Objective

We aimed to assess the causal association between kimchi intake and the reduced HDL-C using Mendelian randomization (MR), with analyses stratified by sex.

Methods

We conducted this study using participants from two cohorts of the Korean Genome and Epidemiology Study (KoGES): the Health Examinees (HEXA) cohort (n = 53,060) and the Ansan/Ansung cohort (n = 4,907). Kimchi intake was assessed via a semi-quantitative food frequency questionnaire, estimating total daily consumption (g/day) of various kimchi types including baechu (cabbage) kimchi, kkakdugi/mu-kimchi, nabak-kimchi/dongchimi, and other varieties. Single nucleotide polymorphisms (SNPs) associated with kimchi consumption were identified via genome-wide association studies (GWAS), adjusting for covariates and applying a significance threshold of p < 5 × 10⁻⁵ and linkage disequilibrium cutoff (r² < 0.001). A total of 86 SNPs in men and 82 in women were selected as instrumental variables (IVs) for MR analysis. The MR analysis was conducted using inverse-variance weighted (IVW), MR-Egger, weighted median, and MR-PRESSO methods. Sensitivity analyses included heterogeneity testing, leave-one-out analysis, and funnel plot inspection.

Results

In men, a significant causal relationship between higher kimchi consumption and lower odds of reduced HDL-C was identified by both the IVW (OR = 0.997, 95% CI = 0.996–0.999, p < 0.05) and MR-PRESSO (OR = 0.998, 95% CI = 0.996–0.999, p < 0.05) analyses, with no evidence of heterogeneity or horizontal pleiotropy. No significant associations were observed in women.

Conclusion

This study provides evidence for a causal effect of kimchi consumption in preventing decreases in HDL-C levels among middle-aged Korean men. These findings support kimchi as a sex-specific dietary intervention for cardiovascular health.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12263-025-00785-6.

Introduction

Reduced high-density lipoprotein cholesterol (reduced HDL-C), defined as blood HDL-C levels below 40 mg/dL for men and 50 mg/dL for women [1, 2], represents a major global health concern.

The prevalence varies substantially across populations yet consistently affects a considerable portion of the global population. In developed countries, prevalence ranges from 13.8% in the United States [3] to 24.5% in China [4]. Rural populations tend to exhibit even higher burdens, as shown in Poland where prevalence reached 20.0% in men and 30.9% in women [5]. The situation is particularly concerning in Korea, where 24.3% of men and 34.6% of women suffer from reduced HDL-C [2] among the highest rates globally highlighting the urgent need for population-specific interventions.

The clinical significance of reduced HDL-C extends far beyond a simple laboratory abnormality. As an independent risk factor for coronary artery disease [6], reduced HDL-C contributes to atherosclerosis development [7] and is directly associated with increased risk of myocardial infarction, stroke, and sudden cardiac death [6]. The high prevalence observed, particularly in East Asian populations such as Korea, suggests that a substantial proportion of individuals are at elevated risk for cardiovascular disease, rendering reduced HDL-C both a significant public health burden and a critical target for prevention.

Kimchi, a traditional Korean fermented food, contains lactic acid bacteria, dietary fiber, and various bioactive compounds such as 3-(4′-hydroxy-3′,5′-dimethoxyphenyl)propionic acid (HDMPPA), capsaicin and allicin, all of which have been reported to exert anti-inflammatory and lipid-lowering effects [8]. Among these, HDMPPA, isolated from the dichloromethane fraction of kimchi, has been shown to penetrate macrophages and modulate cholesterol metabolism by upregulating the expression of liver X receptor α (LXRα), peroxisome proliferator-activated receptor α (PPARα), and ATP-binding cassette transporter A1 (ABCA1) [9], with ABCA1 playing a key role in nascent HDL formation [10]. Other compounds in kimchi, including capsaicin from red pepper and allicin from garlic, have also been associated with increased HDL-C levels and improved lipid profiles. Capsaicin supplementation was shown to elevate HDL-C levels in a randomized, double-blind, placebo-controlled clinical trial involving adults with reduced HDL-C [11]. Similarly, a 12-week randomized, single-blind, placebo-controlled clinical trial involving patients with type 2 diabetes and newly diagnosed dyslipidemia (n = 70) reported a significant increase in HDL-C in the garlic-treated group [12], and the lipid-improving impact of allicin had been confirmed in animal studies [13]. These findings suggest a potential role of kimchi in modulating HDL-C levels and underscore the need to further investigate its effects on lipid metabolism.

Study findings regarding the influence of kimchi consumption on HDL-C levels have been inconsistent across studies. While some experimental and cohort studies have demonstrated beneficial effects of kimchi intake on increasing HDL-C [14–16], others have shown no significant association [17, 18]. These conflicting findings, with some studies demonstrating significant HDL-C improvements while others show no effect, may be attributed to the inherent methodological challenges in nutritional epidemiology, including difficulties in accurately measuring dietary intake and the complex nature of diet as a multifaceted behavioral exposure that involves interactions among various components [19]. Experimental studies often suffer from short intervention periods and small sample sizes, which may limit the generalizability of their findings [20]. Observational studies, on the other hand, are inherently limited in causal inference due to residual confounding, measurement error, and reverse causation [21].

Mendelian randomization (MR) is an epidemiological method that utilizes genetic variants as instrumental variables (IVs) to estimate causal associations between exposures and outcomes [21]​. By exploiting the randomization of genetic material at conception and the independence of genotypes from confounding factors, MR minimizes bias from reverse causality and residual confounding [22]. This approach is especially useful for investigating the causal effects of dietary exposures on metabolic outcomes, as it addresses key limitations of observational studies [23]. To ensure sufficient statistical power and external validity, this research used data from the Korean Genome and Epidemiology Study (KoGES), a large-scale, population-based cohort [24]. The use of the KoGES helps overcome the limited sample sizes commonly observed in experimental investigations, thereby contributing to greater credibility in causal inference.

Despite the growing attention to the potential health benefits of fermented foods, few studies have evaluated the causal relationship between kimchi intake and HDL-C levels. Based on current evidence, this study is one of the first to apply MR to assess the causal impact of traditional fermented foods on HDL-C risk. Therefore, we hypothesized that kimchi consumption may be associated with lower odds of reduced HDL-C and tested this hypothesis using sex-stratified MR analysis based on the KoGES data.

Materials and methods

Study design and population

We utilized data from the Health Examinees (HEXA) cohort and the Ansan-Ansung cohort of the KoGES. Participants in the KoGES were aged between 40 and 70 years. Baseline assessments of the HEXA cohort were conducted between 2004 and 2013. Baseline recruitment for the Ansan-Ansung cohort occurred between 2001 and 2002. An in-depth description of this method has been reported in past investigations [24].

From the HEXA cohort, a total of 173,195 subjects were recruited at baseline. Individuals without genotype data (n = 114,502), without information on kimchi intake (n = 549), without covariate data (n = 3,537), and without information on HDL-C and other metabolic biomarkers (blood pressure, triglyceride, fasting glucose, waist circumference, and total cholesterol) (n = 1,547) were excluded. The final analytic sample comprised 53,060 individuals (18,477 men and 34,583 women) (Fig. 1).

Fig. 1.

Flowchart of study participant selection from HEXA cohort. Flowchart showing the sequential exclusion criteria for study participants from the KoGES HEXA cohort, including missing data for KCHIP, kimchi intake, covariates, and metabolic variables. The final analytic sample consisted of 18,477 men and 34,583 women. KoGES, Korean Genome Epidemiology Study; HEXA, Health Examinees Study; KCHIP, Korean Chip

In the Ansan-Ansung cohort, a total of 10,030 individuals were initially enrolled. Participants without covariate data (n = 932), without metabolic biomarkers including HDL-C, blood pressure, triglycerides, fasting glucose, waist circumference and total cholesterol; (n = 273), and without genotype data (n = 3,918) were excluded. The final analytic sample included 4,907 individuals—comprising 2,379 men and 2,528 women (Fig. 2).

Fig. 2.

Flowchart of study participant selection from Ansan-Ansung cohort. Flowchart showing the sequential exclusion criteria for study participants from the KoGES Ansan-Ansung cohort, including missing data for KCHIP, covariates, and metabolic variables. The final analytic sample consisted of 2,379 men and 2,528 women. KoGES, Korean Genome Epidemiology Study; KCHIP, Korean Chip

The study adhered to the principles outlined in the Declaration of Helsinki and received approval from the Institutional Review Board of Inha University (IRB No. 240307–1 A). Written informed consent was obtained from all participants.

Demographic, anthropometric, biochemical, and clinical parameter assessment

Participants provided information on smoking status, alcohol consumption, education, income, marital status, physical activity, age and body mass index (BMI). Smoking status was determined by asking whether participants had smoked more than 400 cigarettes in their lifetime. Those who responded “no” were classified as never-smokers. Participants who had smoked more than 400 cigarettes but were not currently smoking were categorized as past smokers while those who reported current smoking were categorized as current smokers. Alcohol consumption status was assessed through structured questionnaires. Participants who reported avoiding alcohol for personal or religious reasons were defined as never-drinkers. Among the remaining participants, those who denied current alcohol consumption were classified as past drinkers, and those who reported current drinking were categorized as current drinkers. Educational attainments were classified into four categories: elementary school or less, middle school, high school, and college or above. Income levels were categorized as follows: lower-income (< 1 M KRW/month), moderate-income (1–3 M KRW/month), and higher-income (≥ 3 M KRW/month). Marital status was categorized as single, married, or other. Physical activity was assessed by asking whether participants exercised regularly at an intensity sufficient to cause sweating. A “yes” response indicated active status; a “no” response indicated inactive status.

Height, weight, and waist circumference were measured with participants barefoot and wearing light clothing, following standardized procedures [24]. BMI was calculated by dividing weight (kg) by the height squared (m²). Blood samples were collected after at least 12 h of fasting, both with and without anticoagulants. Plasma and serum were separated by centrifugation. The lipid profiles—including HDL-C, total cholesterol (TC), low-density lipoprotein cholesterol (LDL-C), and triglycerides (TG)—as well as glucose levels, were analyzed using an automatic analyzer (Hitachi 7600, Hitachi, Tokyo, Japan), respectively [25]. Systolic blood pressure (SBP) and diastolic blood pressure (DBP) were measured three times on the right arm at heart level in a seated position, and the average of the three measurements was used for analysis.

Definition of reduced HDL–cholesterol

Serum HDL-C levels were measured directly using the Cholest N HDL reagent (Sekisui, Japan) with a homogeneous enzymatic method on the Hitachi Automatic Analyzer 7600 − 210 (Hitachi, Japan) [26]. Individuals were classified as being at reduced HDL-C if serum HDL-C levels were < 40 mg/dL for men and < 50 mg/dL for women; all others were considered to have normal HDL-C levels [1, 2].

Food and nutrient intake

Dietary intake was evaluated utilizing a semi-quantitative food frequency questionnaire (SQFFQ) specifically tailored to typical Korean dietary habits and previously validated within the framework of the KoGES cohort [27]. Participants were requested to provide information about their usual intake of 106 food items over the preceding six-month period [28]. Kimchi consumption was categorized into four types: baechu kimchi (cabbage kimchi), kkakdugi/mu kimchi, nabak kimchi/dongchimi, and other types of kimchi (e.g., green onion kimchi, mustard leaf kimchi, godulppaegi kimchi). Consumption frequency was assessed using a 9-point scale (rarely, once monthly, 2–3 times a month, weekly, 2–4 times weekly, 5–6 times per week, once daily, twice daily, and up to three times daily), which is the format adopted in the KoGES SQFFQ [29].

Portion sizes were standardized as follows: 25 g (0.5 serving), 50 g (1 serving), and 75 g (1.5 servings) for baechu kimchi, kkakdugi/mu kimchi, and other types; and 47.5 g (0.5 serving), 95 g (1 serving), and 142.5 g (1.5 servings) for nabak kimchi/dongchimi. Daily kimchi intake was calculated by multiplying the reported frequency of consumption by the corresponding portion size for each kimchi. Total kimchi intake was derived by summing the daily amounts across all four categories. To estimate individual energy and nutrient intake, the consumption frequency of each food item was multiplied by its corresponding caloric and nutrient content, and the totals were summed across all reported foods [30].

Genotyping and quality control

Genomic DNA was extracted from peripheral blood specimens. Genotype data suitable for imputation were performed using the Korea Biobank Array (Korean Chip, KCHIP, Seoul, Korea), a microarray platform specifically developed for Korean populations and widely used in genome-wide association studies [31]. Genotyping accuracy was evaluated using the Bayesian Robust Linear Model for Microarray (BRLMM) algorithm, which incorporates Mahalanobis distance to improve the robustness of genotype calling [32]. To ensure data quality, several filters were applied during quality control, including: absence of sex mismatch, low heterozygosity (< 30%), high genotyping concordance (≥ 98%), low missingness per SNP (< 4%), minor allele frequency (MAF) above 1%, and conformity with Hardy–Weinberg Equilibrium (HWE; p >0.05) [33].

Instruments variables selection

To identify genetic variants related to total kimchi consumption, a genome-wide association analysis (GWAS) was conducted. Daily kimchi intake (g/day) was treated as a continuous variable, and analyses were stratified by sex (for men and women). We conducted the analysis using PLINK 1.9 software (https://www.cog-genomics.org/plink/1.9, accessed on 8 April 2025) within a city hospital-based cohort. The Manhattan and quantile–quantile (Q–Q) plots were generated in R using the ggplot2, ggrepel, and data.table packages (Figs. 3 and 4). The genomic inflation factor (λGC) from the Q–Q plot was 1.012(men) and 1.037(women), indicating minimal inflation and a low risk of false positives in the GWAS [34]. SNP annotations were verified using the dbSNP database (https://www.ncbi.nlm.nih.gov/snp, accessed on 7 May 2025).

Fig. 3.

Manhattan plots of GWAS for kimchi intake. (A) Men, (B) Women. The x-axis represents chromosome position, and the y-axis represents the –log₁₀(p-value) for each SNP. The red dashed line indicates the genome-wide significance threshold (p = 5 × 10⁻⁸). GWAS, Genome-Wide Association Study; SNP, Single Nucleotide Polymorphism

Fig. 4.

QQ plots of GWAS for kimchi intake. (A) Men, (B) Women. λGC indicates the genomic inflation factor. The plots show no substantial genomic inflation. QQ plot, Quantile-Quantile plot; λGC, Genomic Inflation Factor; GWAS, Genome-Wide Association Study

Linkage disequilibrium (LD) was assessed using a 10,000 kb window, and variants were retained only if their pairwise r² was below 0.001, ensuring independence among selected SNPs [35]. Previous studies have shown that lifestyle-related traits typically exhibit weaker associations with genetic variants; thus a liberal p-value threshold of 5 × 10⁻⁵ has often been used for selecting IVs in MR analyses [35, 36]. In our analysis, despite the large sample size, only one SNP in men and none in women reached the conventional genome-wide significance threshold of p < 5 × 10⁻⁸. Therefore, we adopted a significance threshold of 5 × 10⁻⁵ to select SNPs related to kimchi intake. For both men and women, the F-statistic showed no weak instrument bias (all F-statistic >10). See Supplementary Table 1 and Supplementary Table 2.

A two-sample MR analysis design

A two-sample MR analysis was conducted to assess the causal relationship between kimchi intake (exposure) and reduced HDL-C. Genetic variants associated with kimchi consumption were used as IVs. For the MR design to be valid, 3 key assumptions must be satisfied: (1) the selected genetic variants must be strongly associated with the exposure (kimchi intake); (2) they must not be associated with confounders; accordingly, SNPs with evidence of association with any prespecified covariate were excluded; and (3) they must affect the outcome (reduced HDL-C) only through the exposure [37]. Potential confounders were adjusted for, including categorical factors (smoking, alcohol intake, education, marital status, income and physical activity) and continuous variables (age, BMI, and total energy intake). To ensure the robustness in instrument selection, we applied the p < 5 × 10⁻⁵ threshold as described above. Based on these criteria, our two-sample MR analysis offers insights into the potential causal role of kimchi intake in lowering the odds of reduced HDL-C.

To satisfy the assumptions of a two-sample MR analysis, two independent cohorts from the KoGES were used: (1) the HEXA cohort (18,477 men and 34,583 women), and (2) the Ansan-Ansung cohort (n = 4,907). In the HEXA cohort, IVs were identified through linear regression-based GWAS using daily kimchi intake as a continuous exposure. The association between the selected IVs and the reduced HDL-C was then evaluated in the Ansan-Ansung cohort using logistic regression-based GWAS.

The primary MR method was the inverse-variance weighted (IVW) approach, which estimates the overall causal effect by calculating a weighted average of the SNP-specific Wald ratios, with weights based on the inverse of the squared standard errors [38]. To enhance robustness, we also applied MR-Egger regression and the weighted median method. MR-Egger regression allows for directional pleiotropy by incorporating an intercept term, assuming the Instrument Strength Independent of Direct Effect (InSIDE) assumption [39]. The weighted median method provides consistent estimates if at least 50% of the weight comes from valid IVs and is less sensitive to outliers [40]. In addition, the MR Pleiotropy RESidual Sum and Outlier (MR-PRESSO) method was used to detect and correct for horizontal pleiotropy in the IVW model. MR-PRESSO performs a global test to identify outlier SNPs, removes them, and re-estimates the causal effect, thereby improving the reliability of estimate [41]. As a sensitivity analysis, we conducted leave-one-out test to assess symmetry of the effect estimates, helping detect potential directional horizontal pleiotropy. Forest plots were also generated to visualize SNP-specific causal estimates along with 95% confidence intervals allowing for evaluation of both SNP-level and overall consistency in causal estimates [42]. All analyses were performed independently for men and women using the same procedures. All MR effect estimates are expressed as odds ratios per 1 standard deviation (SD) increase in genetically predicted kimchi intake, as clarified in the Methods and figure legends.

Secondary analyses

In addition to sex-stratified analyses, we conducted a prespecified sex-combined two-sample MR using the same instrument selection (p < 5 × 10⁻⁵; LD r² < 0.001), harmonization, and sensitivity procedures. Because kimchi is a composite food that intrinsically contains sodium, our primary MR (without sodium adjustment) targets the total effect of kimchi intake. To approximate a sodium-independent effect, we re-estimated the exposure GWAS additionally adjusting for dietary sodium (mg/day) and repeated the MR pipeline. These estimates are interpreted as sensitivity analyses, not as an alternative primary analysis. To assess potential heterogeneity by kimchi type, we performed MR treating baechu kimchi, kkakdugi/mu-kimchi, nabak-kimchi/dongchimi, and other types as separate exposures, applying identical instrument criteria.

Statistical analysis

All statistical computations were conducted using SAS 9.4 (SAS Institute Inc., Cary, NC, USA) and R 4.4.3 (R Foundation for Statistical Computing, Vienna, Austria) [43]. Categorical variables were examined using frequency analysis across tertiles of kimchi intake. Chi-squared tests were applied to assess differences in the distribution of categorical variables across the tertile groups. Continuous variables were expressed as means ± standard deviations (SD) according to tertile groups of kimchi intake. Differences among the groups were evaluated using one-way analysis of variance (ANOVA). Statistical significance was determined at p-value < 0.05 [44].

Results

Demographic, Lifestyle, anthropometric and clinical characteristics

Table 1 shows the baseline characteristics of the study subjects categorized by sex and tertiles of kimchi intake. A total of 53,060 individuals (18,477 men and 34,583 women) were retained in the analysis. Participants were classified into low, middle, and high intake groups based on sex-specific tertile cut-offs for kimchi consumption.

Table 1.

Demographic and lifestyle characteristics according to sex and kimchi consumption tertiles

	Total (n = 53,060)
	Men (n = 18,477)				Women (n = 34,583)
	Kimchi intake				Kimchi intake
	Tertile 1 (0–87.7 g/day) (n = 6,160)	Tertile 2 (87.8–171.3 g/day) (n = 6,157)	Tertile 3 (171.4–1,102.5 g/day) (n = 6,160)	p	Tertile 1 (0–75.0 g/day) (n = 11,712)	Tertile 2 (75.1–155.4 g/day) (n = 11,389)	Tertile 3 (155.6–1,102.5 g/day) (n = 11,482)	p
Age (year)	55.38 ± 8.28	55.25 ± 8.37	54.52 ± 8.58	< 0.0001	52.97 ± 7.70	52.75 ± 7.64	53.05 ± 7.69	0.0103
Education (%)				0.0394				0.0015
≤Elementary school	488 (7.92)	536 (8.71)	552 (8.96)		2,018 (17.23)	1,933 (16.97)	2,059 (17.93)
Middle school	730 (11.85)	802 (13.03)	729 (11.83)		2,042 (17.44)	1,975 (17.34)	2,083 (18.14)
High school	2,378 (38.60)	2,335 (37.92)	2,426 (39.38)		4,954 (42.30)	4,884 (42.88)	4,945 (43.07)
≥College	2,564 (41.62)	2,484 (40.34)	2,453 (39.82)		2,698 (23.04)	2,597 (22.80)	2,395 (20.96)
Marital status (%)				0.0002				< 0.0001
Married	5,759 (93.49)	5,853 (95.06)	5,855 (95.05)		10,135 (86.54)	10,099 (88.67)	10,299 (89.70)
Single	142 (2.31)	127 (2.06)	112 (1.82)		318 (2.72)	241 (2.12)	182 (1.59)
Other	259 (4.20)	177 (2.87)	193 (3.13)		1,259 (10.75)	1,049 (31.70)	1,001 (8.72)
Income (%)				0.9509				0.0138
< 1,000,000	506 (8.21)	500 (8.12)	514 (8.34)		1,328 (11.34)	1,256 (11.03)	1,350 (11.76)
1,000,000 ~  3,000,000	2,625 (42.61)	2,596 (42.16)	2,625 (42.61)		5,207 (44.46)	4,919 (43.19)	5,127 (44.65)
≥ 3,000,000	3,029 (49.17)	3,061 (49.72)	3,021 (49.04)		5,177 (44.20)	5,127 (44.65)	5,005 (43.59)
Smoking status (%)				< 0.0001				0.0103
Never	1,977 (32.09)	1,882 (30.57)	1,755 (28.49)		11,355 (96.95)	11,094 (97.41)	11,218 (97.70)
Past	2,567 (41.67)	2,591 (42.08)	2,586 (33.39)		131 (1.12)	102 (0.90)	93 (0.81)
Current	1,616 (26.23)	1,684 (27.35)	1,819 (29.53)		226 (1.93)	193 (1.69)	171 (1.49)
Drinking status (%)				< 0.0001				0.9778
Never	1,349 (21.90)	1,311 (21.29)	1,154 (18.73)		7,937 (67.77)	7,739 (67.95)	7,804 (67.97)
Past	523 (8.49)	416 (6.76)	445 (7.22)		203 (1.73)	191 (1.68)	187 (1.63)
Current	4,288 (6961)	4,430 (71.95)	4,561 (74.04)		3,572 (30.50)	3,459 (30.37)	3,491 (30.40)
Physical activity (%)				0.0968				0.2162
Yes	2,543 (41.28)	2,531 (41.11)	2,436 (39.55)		5,521 (47.14)	5,499 (48.28)	5,465 (47.60)
No	3,617 (58.7)	3,626 (58.89)	3,724 (60.45)		6,191 (52.86)	5,890 (51.72)	6,017 (52.40)
Energy(kcal)	1,737.98 ± 474.99	1,876.65 ± 491.83	2,097.07 ± 606.49	< 0.0001	1,581.16 ± 514.16	1,716.88 ± 507.30	1,916.98 ± 585.89	< 0.0001
BMI (kg/m2)	24.33 ± 2.72	24.40 ± 2.70	24.52 ± 2.64	0.0007	23.42 ± 2.89	23.56 ± 2.90	23.80 ± 2.98	< 0.0001
HDL-cholesterol (mg/dL)	49.31 ± 11.92	49.50 ± 12.16	49.09 ± 11.75	0.1686	56.91 ± 13.34	56.54 ± 13.27	55.35 ± 12.89	< 0.0001
Diastolic blood pressure (mmHg)	77.99 ± 9.47	78.06 ± 9.57	78.50 ± 9.54	0.005	74.05 ± 9.55	74.24 ± 9.45	74.57 ± 9.54	0.0002
Systolic blood pressure (mmHg)	124.84 ± 13.90	125.76 ± 14.24	125.74 ± 14.04	0.0002	120.42 ± 14.81	120.68 ± 14.74	121.05 ± 14.97	0.0051
Waist circumference (cm)	85.39 ± 7.46	85.74 ± 7.46	85.92 ± 7.39	0.0003	77.50 ± 7.98	78.01 ± 8.00	78.83 ± 8.13	< 0.0001
Blood glucose (mg/dL)	99.43 ± 22.56	99.55 ± 22.41	99.04 ± 22.51	0.4095	92.94 ± 17.35	92.74 ± 17.90	93.06 ± 17.89	0.3871
Triglycerides (mg/dL)	145.35 ± 101.23	147.32 ± 101.46	150.13 ± 103.70	0.0333	111.14 ± 72.91	112.34 ± 72.66	115.87 ± 73.80	< 0.0001
Total-cholesterol (mg/dL)	191.76 ± 35.18	192.89 ± 35.22	192.25 ± 34.14	0.194	200.12 ± 36.09	200.21 ± 35.63	199.83 ± 35.70	0.6969

Values are presented as mean ± standard deviation for continuous variables and number (%) for categorical variables. Kimchi intake was categorized into tertiles separately for men and women. p-values were obtained using ANOVA for continuous variables and the chi-square test for categorical variables. p < 0.05 was considered statistically significant

Participants in the high kimchi consumption group exhibited significantly different characteristics compared to those in the low intake group in terms of age, education, marital status, income, smoking, drinking, energy intake, BMI, HDL-C, systolic blood pressure, diastolic blood pressure, waist circumference, and fasting blood glucose (p < 0.05). Specifically, age showed opposite trends by sex: higher kimchi intake was related with younger age in men, but with older age in women (p < 0.05). Educational attainment significantly differed across intake groups in both sexes, with the proportion of participants holding a college degree or higher decreasing as kimchi intake increased, while the proportions of lower education levels increased (p < 0.05). Regarding marital status, the proportion of married individuals increased with higher kimchi consumption in both men and women, whereas other marital categories decreased (p < 0.05). Income level showed significant differences only among women, with the high intake group having a lower proportion of participants in the highest income category (p < 0.05). A significant difference in smoking status was observed across kimchi intake levels in both men and women, with opposite trends by sex (p < 0.05). In men, current smoking was more prevalent in the high intake group compared to the low intake group, while in women, the opposite trend was observed. A significant difference in alcohol consumption was observed only in men, where current drinking was more prevalent in the high intake group (p < 0.05). Daily energy intake and BMI were significantly higher in the high kimchi intake group for both sexes (p < 0.05). In women, HDL-C levels were significantly lower in the high intake group compared to low intake group (p < 0.05). Systolic blood pressure, diastolic blood pressure, waist circumference, and fasting blood glucose were all significantly higher in the high intake group compared to low intake group for both men and women (p < 0.05).

Nutrient intake

Nutrient intake increased significantly across tertiles of kimchi consumption in both men and women (p < 0.05). Intake levels of all nutrients progressively increased from the low to high kimchi intake groups. This pattern was observed for all assessed nutrients, including macronutrients (protein, fat, carbohydrates, fiber), minerals (calcium, potassium, sodium, zinc, iron), and vitamins (vitamin A, vitamin C, vitamin E, vitamin B₁, vitamin B₂, vitamin B₆, folate, β-carotene, retinol) as well as ash and cholesterol. For instance, protein intake increased from 52.67 ± 19.13 g in the lowest tertile to 70.39 ± 28.80 g in the highest tertile among men, and from 48.42 ± 20.67 g to 64.14 ± 27.50 g among women. Similar upward trends were observed for all other nutrients such as fat, sugar, fiber, ash, calcium, phosphorus, iron, potassium, zinc, sodium, vitamin A, retinol, β-carotene, vitamin E, vitamin B₁, vitamin B₂, vitamin B₆, niacin, vitamin C, folate, and cholesterol (Table 2).

Table 2.

Nutrient intake of study participants by sex-specific tertiles of kimchi intake

	Total (n = 53,060)
	Men (n = 18,477)				Women (n = 34583)
	Low intake (n = 6,160)	Middle intake (n = 6,157)	High intake (n = 6,160)	p	Low intake (n = 11,712)	Middle intake (n = 11,389)	High intake (n = 11,482)	p
Protein (g)	52.67 ± 19.13	59.49 ± 21.06	70.39 ± 28.80	< 0.0001	48.42 ± 20.67	55.05 ± 20.73	64.14 ± 27.50	< 0.0001
Fat (g)	24.98 ± 14.97	28.62 ± 16.71	33.54 ± 21.16	< 0.0001	22.27 ± 15.54	25.14 ± 15.13	28.33 ± 19.58	< 0.0001
Carbohydrate (g)	318.25 ± 81.73	338.02 ± 83.36	371.76 ± 95.67	< 0.0001	291.90 ± 90.12	313.20 ± 88.97	347.42 ± 95.85	< 0.0001
Ash(g)	7.22 ± 3.05	9.45 ± 3.34	13.80 ± 6.07	< 0.0001	7.22 ± 3.38	9.31 ± 3.67	12.91 ± 6.00	< 0.0001
Fiber(g)	19.11 ± 6.11	23.37 ± 6.34	30.67 ± 9.71	< 0.0001	19.07 ± 7.15	23.30 ± 7.65	30.06 ± 10.60	< 0.0001
Ca(mg)	342.38 ± 188.07	439.17 ± 191.61	585.04 ± 270.57	< 0.0001	363.49 ± 209.71	460.91 ± 228.82	586.96 ± 292.97	< 0.0001
P(mg)	688.75 ± 267.46	809.07 ± 287.64	976.97 ± 395.65	< 0.0001	674.87 ± 294.06	790.44 ± 308.17	936.32 ± 406.60	< 0.0001
Fe(mg)	7.41 ± 2.50	8.87 ± 2.75	11.19 ± 4.04	< 0.0001	7.06 ± 2.78	8.42 ± 2.86	10.55 ± 4.08	< 0.0001
K(mg)	1,481.45 ± 642.63	1,872.65 ± 684.10	2,508.45 ± 1,048.72	< 0.0001	1,548.49 ± 738.18	1,923.23 ± 790.50	2,476.63 ± 1,124.02	< 0.0001
Zinc(mg)	7.56 ± 2.86	8.60 ± 3.27	10.25 ± 4.41	< 0.0001	7.00 ± 3.13	7.89 ± 2.93	9.26 ± 4.02	< 0.0001
Na(mg)	1,217.68 ± 609.50	1,857.58 ± 605.12	3,009.35 ± 1,249.51	< 0.0001	1,081.67 ± 581.50	1,664.26 ± 26	2,640.91 ± 1,131.87	< 0.0001
Vit.A(R.E)	338.15 ± 220.88	447.36 ± 243.25	679.63 ± 443.61	< 0.0001	361.63 ± 246.69	467.37 ± 276.48	659.30 ± 438.90	< 0.0001
Retinol(µg)	81.42 ± 76.48	92.28 ± 78.64	106.33 ± 98.68	< 0.0001	88.55 ± 84.26	98.35 ± 84.00	105.48 ± 97.30	< 0.0001
β-Carotene(µg)	1,541.23 ± 1,126.62	2,131.37 ± 1,268.00	3,440.98 ± 2,413.17	< 0.0001	1,639.37 ± 1,251.01	2,215.07 ± 1,439.95	3,324.02 ± 2,379.87	< 0.0001
Vit.E(mg)	6.33 ± 3.69	7.61 ± 3.74	9.94 ± 5.14	< 0.0001	5.96 ± 3.84	7.25 ± 3.97	9.13 ± 5.16	< 0.0001
Vit.B1(mg)	0.70 ± 0.30	0.85 ± 0.32	1.08 ± 0.46	< 0.0001	0.67 ± 0.33	0.81 ± 0.32	1.01 ± 0.45	< 0.0001
Vit.B2(mg)	0.90 ± 0.46	1.06 ± 0.47	1.33 ± 0.62	< 0.0001	0.86 ± 0.46	1.02 ± 0.48	1.23 ± 0.61	< 0.0001
Niacin(mg)	8.52 ± 3.89	9.99 ± 4.18	12.23 ± 5.96	< 0.0001	8.15 ± 4.16	9.50 ± 4.28	11.33 ± 5.82	< 0.0001
Vit.C(mg)	48.27 ± 32.12	63.01 ± 34.18	84.42 ± 51.04	< 0.0001	58.57 ± 40.23	73.51 ± 42.91	93.64 ± 56.73	< 0.0001
Vit.B6(mg)	0.50 ± 0.19	0.69 ± 21	0.94 ± 0.31	< 0.0001	0.49 ± 0.22	0.66 ± 0.22	0.90 ± 0.31	< 0.0001
Folate(µg)	172.46 ± 76.74	223.67 ± 80.57	299.50 ± 122.83	< 0.0001	182.09 ± 87.78	232.48 ± 95.66	302.28 ± 133.87	< 0.0001
Cholesterol(mg)	126.09 ± 85.32	150.13 ± 96.09	176.21 ± 121.23	< 0.0001	120.66 ± 89.82	141.89 ± 92.74	162.01 ± 118.46	< 0.0001

Values are presented as mean ± standard deviation. Intake values were derived from validated semi-quantitative food frequency questionnaires. Kimchi intake was categorized into tertiles separately for men and women. p-values were calculated using ANOVA to assess differences across intake groups. p < 0.05 was considered statistically significant

Two-Sample MR analysis

In men, MR analysis suggested a plausible causal connection between increased kimchi intake and lower odds of reduced HDL-C. The IVW method yielded a statistically significant inverse association (OR = 0.997, 95% CI: 0.996–0.999, p = 0.007). This finding was directionally consistent with the weighted median estimate (OR = 0.998, 95% CI: 0.996–1.001), although it was not statistically significant (p = 0.217). Similarly, MR-Egger regression indicated no significant causal effect (OR = 0.998, 95% CI: 0.994–1.001, p = 0.224), and the intercept test provided no evidence of directional pleiotropy (intercept = − 0.001, SE = 0.017, p = 0.958).

The MR-PRESSO method identified a significant association (OR = 0.998, 95% CI: 0.996–0.999, p = 0.005), further supporting the IVW results, with no evidence of horizontal pleiotropy (p > 0.05). Heterogeneity tests for both the IVW and MR-Egger models were statistically non-significant (p > 0.05), indicating the robustness of the causal estimates (Table 3; Fig. 5A). In men, the causal association was statistically significant, but the effect sizes were very close to unity (IVW OR = 0.997, p = 0.007; MR-PRESSO OR = 0.998, p = 0.006), and the consistency across these two primary methods reinforces the robustness of the finding despite its minimal magnitude.

Table 3.

Mendelian randomization estimates for the association between kimchi intake and risk of reduced HDL-cholesterol in men

Men
	MR			Heterogeneity		Pleiotropy
Method	Number of SNPs	OR (95% CI)	p	Q	p	Intercept	SE	p
MR Egger	86	0.998 (0.994, 1.002)	0.224	68.873	0.884	-0.001	0.017	0.958
Weighted median	86	0.998 (0.996, 1.001)	0.217
Inverse variance weighted	86	0.997(0.996, 0.999)	0.007	68.876	0.898
Weighted mode	86	0.999 (0.993, 1.006)	0.910
MR-PRESSO	86	0.998 (0.998, 0.999)	0.005

Odds ratios (ORs) and 95% confidence intervals (CIs) for the effect of genetically predicted kimchi intake on the risk of reduced HDL-cholesterol in men, estimated using different MR methods. MR, Mendelian randomization; SNP, single nucleotide polymorphism; HDL, high-density lipoprotein; OR, odds ratio; CI, confidence interval; SNP, single nucleotide polymorphism; Q, Cochran’s Q statistic; SE, standard error

Fig. 5.

Scatter plots of SNP effects on kimchi intake and low HDL cholesterol risk. Each black dot represents the estimated effect of a single SNP on the exposure (x-axis, kimchi intake) and the outcome (y-axis, reduced HDL cholesterol risk), with horizontal and vertical lines indicating standard errors. The regression lines indicate causal estimates from three MR methods: IVW, MR-Egger, and weighted median. The slope of each line reflects the direction and strength of the estimated causal association

In women, all MR methods (IVW, MR-Egger, and weighted median) produced non-significant estimates, indicating no evidence of a causal relationship with reduced HDL-C.

MR-PRESSO detected no outlier SNPs. Using all selected instruments, we found no evidence of a causal association (OR = 0.999, 95% CI: 0.996–1.002, p = 0.597), and no horizontal pleiotropy was detected. Heterogeneity tests from both IVW and MR-Egger models were statistically non-significant, supporting the consistency of the IVs (Table 4; Fig. 5B).

Table 4.

Mendelian randomization estimates for the association between kimchi intake and risk of reduced HDL-cholesterol in women

Women
	MR			Heterogeneity		Pleiotropy
Method	Number of SNPs	OR (95% CI)	p	Q	p	Intercept	SE	p
MR Egger	82	0.997 (0.990, 1.005)	0.467	96.455	0.102	0.011	0.021	0.589
Weighted median	82	1.002 (0.997, 1.006)	0.463
Inverse variance weighted	82	0.999 (0.996, 1.002)	0.574	96.811	0.111
Weighted mode	82	1.006 (0.994, 1.017)	0.360
MR-PRESSO	82	0.999 (0.996, 1.002)	0.547

Odds ratios (ORs) and 95% confidence intervals (CIs) for the effect of genetically predicted kimchi intake on the risk of reduced HDL-cholesterol in women, estimated using different MR methods. MR, Mendelian randomization; SNP, single nucleotide polymorphism; HDL, high-density lipoprotein; OR, odds ratio; CI, confidence interval; SNP, single nucleotide polymorphism; Q, Cochran’s Q statistic; SE, standard error

Sex-combined MR

When men and women were analyzed jointly, estimates were non-significant across methods. There was no evidence of directional pleiotropy or excess heterogeneity. See Supplementary Table 3.

Sodium-adjusted MR

After adjusting the exposure GWAS for dietary sodium, associations were not statistically significant in both sexes (men: IVW OR = 0.999, 95% CI 0.995–1.003, p = 0.650; women: IVW OR = 1.001, 95% CI 0.995–1.006, p = 0.819). Full results are provided in Supplementary Table 4.

Subtype-specific MR

Estimates across baechu, kkakdugi/mu-kimchi, nabak/dongchimi, and other types were not statistically significant; details are in Supplementary Table 5.

Sensitivity analyses

Leave-one-out sensitivity analysis

The leave-one-out sensitivity analysis demonstrated consistent causal estimates when each SNP was sequentially excluded from the analysis, in both men and women. This indicates that no single SNP had a disproportionate influence on the overall MR estimate, thus supporting the robustness of the findings (Fig. 6).

Fig. 6.

Leave-one-out sensitivity analysis. Forest plots showing the effect estimates obtained by excluding each SNP one at a time from the MR analysis for kimchi intake on reduced HDL-C in (A) men and (B) women. Dots represent MR estimates; horizontal lines indicate 95% confidence intervals

Forest plot

In the forest plot, the MR-Egger estimate for men was not statistically significant. However, the combined causal estimate from the IVW method was statistically significant, suggesting that higher kimchi intake was associated with lower odds of reduced HDL-C (Fig. 7A). For women, the inferred causal associations derived from both the IVW and MR-Egger methods were not statistically significant, indicating no clear supporting evidence of a causal link between kimchi intake and the reduced HDL-C in women (Fig. 7B).

Fig. 7.

Forest plots for Mendelian randomization estimates. Forest plots of individual SNP effects on reduced HDL-C per standard deviation increase in kimchi intake for (A) men and (B) women. Black dots represent log ORs; horizontal lines indicate 95% confidence intervals. Red markers indicate IVW and MR-Egger estimates, respectively

Funnel plot

Funnel plot analysis revealed symmetrical distributions around the causal estimate in men and women under the IVW method, and no noticeable asymmetry was observed in MR-Egger analysis. This implies a low probability of horizontal pleiotropy and reporting bias (Fig. 8).

Fig. 8.

Funnel plots for Mendelian randomization estimates. Funnel plots assessing potential pleiotropy and small-study bias in MR analysis of kimchi intake and reduced HDL-C for (A) men and (B) women. Symmetry suggests absence of directional pleiotropy

Discussion

This study used MR to investigate the causal link between kimchi intake and the risk of reduced HDL-C using two independent population-based cohorts from the KoGES. The results indicated a significant inverse association between genetically estimated kimchi intake and lower odds of reduced HDL-C among men, suggesting a potential protective effect, while no significant association was found in women.

Prior investigations assessing the association between kimchi consumption and HDL-C have primarily been based on observational studies or short-term intervention trials, and their findings have been inconsistent. In a clinical trial investigating the lipid-modulating effects of kimchi, a six-week supplementation with freeze-dried kimchi resulted in reduced serum triglyceride levels and increased HDL-C concentrations [45]. In an experimental study involving 102 healthy Korean adult men aged 40–64 years, daily kimchi intake ranged from 68 g (1st quartile) to 383 g (4th quartile), and higher kimchi intake was positively associated with HDL-C levels (p < 0.05) [14]. Similarly, a positive association between baechu kimchi intake and HDL-C concentrations was observed in men in an observational study of 61,761 adults aged 40–69 years, based on the HEXA cohort. Dietary intake was assessed by FFQ and categorized into 1 serving, 1–2 servings, 2–3 servings, and at least 3 servings, with a mean follow-up duration of approximately 5 years [15].

In contrast, some clinical studies [17, 18] have reported that HDL-C levels either decreased or showed no significant changes following kimchi consumption. For example, a 7-day intervention study with 100 young healthy adults compared low (15 g/day) and high (210 g/day) kimchi intake groups, and found no statistically meaningful variation in HDL-C between the groups [17]. Another crossover trial with 22 overweight and obese participants (BMI >25 kg/m²) compared the impact of 4-week consumption of fresh vs. fermented kimchi. Although total cholesterol and triglycerides decreased more in the fermented group, HDL-C levels remained statistically unchanged [18]. Such inconsistent findings may be attributed to differences in participant characteristics, such as age (middle-aged adults aged 40–64 years vs. younger adults), sex (men vs. women), and BMI (normal-weight individuals vs. those who are overweight or obese), kimchi intake levels (ranging from 15 g to over 380 g/day), and intervention durations (from 7 days to approximately 5 years).

Given this background, the present study applied MR analysis using genetic instruments to strengthen the causal inference beyond the associations identified in previous studies. Notably, a statistically significant inverse association between kimchi intake and lower odds of reduced HDL-C was detected only in men (IVW: OR = 0.997, 95% Cl = 0.996–0.999, p < 0.05; MR-PRESSO: OR = 0.998, 95% Cl = 0.996–0.999, p < 0.05), which aligns with earlier findings reporting more pronounced effects of kimchi intake on HDL-C levels in men populations, thereby enhancing the consistency of the current results [14, 15]. Similarly, no statistically significant association was found in women, consistent with prior studies that also reported non-significant associations in women participants [15]. This consistency between our MR-based causal estimates and previous observational findings reinforces the credibility of the evidence, suggesting that the observed lower odds of reduced HDL-C in men with higher kimchi intake is unlikely to be explained by confounding or reverse causation, but instead reflects a likely causal effect. Although the male-specific associations were statistically significant, the effect sizes were extremely close to 1. To aid interpretation, we translated these values into absolute risk differences: the IVW OR = 0.997 corresponds to a reduction in low HDL-C prevalence from 24.3% to approximately 24.24% (–0.06%p), and the MR-PRESSO OR = 0.998 corresponds to about 24.26% (–0.04%p). These changes represent very limited clinical impact at the individual level. Nonetheless, because MR estimates reflect lifelong exposure, even such modest effects may accumulate at the population level, potentially preventing several hundred cases per million men if intake distribution shifts by 1 SD. At the same time, given the high sodium content of kimchi, our findings should not be interpreted as a recommendation for indiscriminate increases in intake but rather as etiological evidence of a modest, male-specific protective association within a balanced dietary pattern.

Kimchi is a fermented food composed of cabbage, various vegetables, red pepper, garlic, and other ingredients, and contains an array of components including bioactive compounds such as vitamins, minerals, probiotics, and phytochemicals beneficial to health. It is rich in vitamins, minerals, dietary fiber, probiotics, capsaicin, gingerol, and allicin all of which possess antioxidant and bioactive properties [8]. Probiotics that proliferate during the fermentation process can improve gut microbiota composition and alter bile acid metabolism, thereby reducing cholesterol reabsorption and ultimately lowering blood cholesterol levels [46, 47]. In particular, Lactobacillus plantarum produces bile salt hydrolase (BSH) enzymes in the small intestine, which deconjugate bile acids and reduce their reabsorption, thereby contributing to the reduction of total cholesterol levels [48]. Several studies have reported that Lactobacillus plantarum significantly increases HDL-C levels, thereby improving the lipid profile [49, 50]. Furthermore, Leuconostoc mesenteroides has been reported to improve HDL-C levels and reduce body weight, potentially through fermentation-associated metabolites and improvements in the gut microbiome environment [51].

Kimchi contains flavonoids [52], a diverse class of polyphenolic compounds including anthocyanins, flavanones, and isoflavones [53]. Among them, isoflavones have been shown to improve HDL-C levels under oxidative stress conditions [54]. Among the components of kimchi, clinical trials have suggested that capsaicin supplementation can raise HDL-C by approximately 3 mg/dL, which may contribute to a 9% reduction in cardiovascular disease risk [11, 55]. Allicin, a bioactive compound found in garlic, is known to decrease triglycerides, total cholesterol, and LDL-C levels, while increasing HDL-C levels, thereby exerting lipid-modulating effects [12, 56]. Taken together, the various components of kimchi may synergistically influence lipid metabolism, particularly by contributing to the improvement of HDL-C levels.

In this study, an inverse association between kimchi intake and the risk of reduced HDL-C was observed only in men, while no statistically significant association was found in women. This sex-specific discrepancy may also reflect age-related hormonal changes within the study population, which consisted of individuals aged 40 to 70 years. In men, sex hormone levels—especially total testosterone—gradually decline at an approximate rate of 0.4% per year [57]. In contrast, women experience a more abrupt hormonal shift during the menopausal transition, typically occurring between the ages of 45 and 55 [58], with circulating estradiol (E2) levels dropping sharply from 100 to 250 pg/mL pre-menopause to around 10 pg/mL post-menopause [59]. The age range of women in this study, all of whom were 40 years of age and older, encompasses the typical window for menopausal transition. E2 is primarily synthesized in the ovaries using LDL-C as a substrate [60]. Thus, the reduction in E2 synthesis following menopause implies that LDL-C is no longer utilized for hormone production, remaining in systemic circulation. Consequently, postmenopausal women tend to exhibit elevated LDL-C levels, which are associated with increased risks of abdominal obesity, insulin resistance, dyslipidemia, hypertension, and cardiovascular disease [61]. Such an elevation in circulating LDL-C may also have downstream effects on HDL-C metabolism. Prior research has shown that LDL-rich environments can promote HDL catabolism, leading to reduced HDL-C concentrations. In particular, a kinetic study in postmenopausal women by Matthan et al. (2004) demonstrated that an increase in LDL-C was accompanied by a higher fractional catabolic rate of HDL apoA-I, suggesting that higher LDL-C levels may indirectly contribute to lower HDL-C through accelerated turnover [62].

In this study, women aged 40–70 years represented a heterogeneous group in terms of menopausal status, including those in pre-menopause, post-menopause, or those who may have undergone surgical menopause (4,091 out of 34,583 women, 11.83%, had undergone hysterectomy). Circulating E2 concentrations—greatly affected by menopausal status—may act as a confounding factor in the relationship between kimchi intake and HDL-C metabolism in women. To further contextualize these sex differences, we conducted an exploratory menopausal-stratified MR in women (pre-menopausal n = 13,064, post-menopausal n = 21,475). In post-menopausal women, the IVW estimate was not statistically significant (OR = 1.000, 95% CI 0.998–1.002, p = 0.956; Cochran’s Q p = 0.801; MR-Egger intercept p = 0.257). In pre-menopausal women, the IVW estimate was also not significant (OR = 1.001, 95% CI 0.999–1.002, p = 0.494; Q p = 0.246); the MR-Egger slope reached nominal significance (OR = 1.004, 95% CI 1.000–1.008, p = 0.046) with a borderline intercept (p = 0.058). Given smaller strata, these analyses are exploratory and do not alter our primary conclusions. Full results are provided in Supplementary Table 6.

In addition to hormonal changes, bioactive compounds in kimchi—particularly isoflavones—have been suggested to exert phytoestrogenic effects [63]. These compounds can bind to estrogen receptors and elicit weak estrogenic responses, particularly under conditions of low endogenous estrogen, such as in postmenopausal women [64]. However, a quantitative bioassay study reported that the phytoestrogens genistein (0.11), daidzein (0.08), and genistin (0.06) exhibited only 6–11% of the estrogenic activity of 17β-E2, which was used as the reference compound (potency = 1) [65]. These findings indicate that the estrogenic potency of phytoestrogens is substantially lower than that of endogenous E2 and may be insufficient to compensate for the abrupt decline in estrogen levels after menopause. Therefore, although kimchi’s bioactive compounds may theoretically modulate lipid metabolism via estrogen receptor-mediated pathways, their actual impact in postmenopausal women is likely too modest to yield a statistically significant association in this study.

In this research, the inferred causal connection between kimchi intake and the reduced HDL was analyzed using IVW, MR-Egger, weighted median, and MR-PRESSO analyses. Among men, a significant causal association was observed with IVW and MR-PRESSO, whereas the results from MR-Egger and the weighted median were not statistically significant. This inconsistent pattern—where IVW shows statistical significance but MR-Egger and the weighted median do not—has been frequently reported in previous studies [66, 67], as MR-Egger tends to yield wider standard errors and consequently lower statistical power, especially when the instrument count is limited [68].

IVW is widely accepted as the primary method for estimating the genetically predicted causal effect between exposure and outcome. In contrast, MR-Egger and the weighted median are primarily employed as sensitivity analyses to assess the validity of MR assumptions [37, 39]. In particular, the MR-Egger intercept test evaluates the presence of horizontal pleiotropy—that is, whether the instrumental variables collectively influence the outcome through pathways unrelated to the exposure [39]. In this study, the MR-Egger intercept was not statistically significant, suggesting no evidence of horizontal pleiotropy. This finding supports the validity of the IVW estimate as an unbiased causal effect.

According to recent MR guidelines, IVW is recommended as the primary approach, while MR-Egger and the weighted median serve as sensitivity analyses [69]. Although the inconsistency across MR methods necessitates cautious interpretation, the agreement between IVW and MR-PRESSO—which additionally corrects for potential outlier bias—reinforces the robustness of our findings. Together, these results provide strong evidence supporting a protective causal effect of kimchi intake on HDL-C metabolism in men. This research has the strengths of using MR analysis to assess the causal association between kimchi consumption and lower odds of reduced HDL-C through genetic instruments, thereby effectively reducing the issues of confounding and reverse causation. Because genetic variants are randomly assigned at conception, they are generally unrelated to lifestyle or environmental influences like dietary habits and are thus less likely to be influenced by traditional confounders. In addition, this study utilized data from the KoGES, incorporating GWAS results on kimchi consumption and HDL-C related traits derived from a Korean population. Multiple MR methods were applied to examine the consistency and robustness of the findings. In particular, the MR-PRESSO approach was used to account for potential bias introduced by horizontal pleiotropy [41], thereby enhancing the reliability of the causal estimates. Furthermore, this research is the first MR analysis to evaluate the causal effect of a specific fermented food, kimchi, on blood lipid profiles. It provides important scientific evidence that complements and extends previous findings from observational studies. While direct, drug-like antagonistic interactions between kimchi and other foods have not been elucidated, the overall dietary pattern—particularly high refined-carbohydrate intake—may modify kimchi’s metabolic effects. At the population level, the high prevalence of low HDL-C in Korea is likely multifactorial; although kimchi may contribute beneficially to lipid metabolism, its physiological impact is modest compared with the negative influence of refined-carbohydrate diets and other lifestyle factors [70] . Accordingly, the observed population-level burden of low HDL-C should be interpreted in the context of broader dietary patterns and health behaviors, rather than as evidence of a lack of efficacy of kimchi itself.

This research has several limitations. First, dietary consumption was estimated using an FFQ, which is prone to measurement error and may compromise the accuracy of exposure estimation. Second, the study population consisted of Korean adults in midlife, which could restrict how broadly the results apply to other populations or ethnic groups. Third, although increased kimchi intake was associated with lower odds of reduced HDL-C in men (OR = 0.997 by IVW and 0.994 by MR-PRESSO), these effect sizes are very close to 1, suggesting that the clinical impact at the individual level may be limited. However, the statistical significance of these findings remains meaningful, as they are derived from large-scale epidemiological data with strong population-level representativeness. Finally, kimchi is traditionally prepared with substantial amounts of salt, with baechu kimchi containing approximately 534–783 mg of sodium per 100 g [71]. Given that the WHO recommends a daily sodium intake of less than 2,000 mg (equivalent to 5 g of salt), consuming just 100 g of kimchi may account for over a quarter of this limit. Although some epidemiological studies have reported no significant adverse effects of kimchi consumption on blood pressure [72, 73], its high sodium content remains a potential concern. Therefore, excessive intake may pose health risks, particularly among individuals with salt-sensitive hypertension. Future studies should determine optimal intake thresholds that maximize benefits while minimizing potential harms. A further limitation is that the observed effect sizes (ORs of 0.997 and 0.998) are extremely close to unity, indicating minimal clinical impact at the individual level, even though they were statistically significant.

Conclusions

In conclusion, this study using MR suggests a potential causal relationship between kimchi consumption and lower odds of low HDL-C among middle-aged Korean men. While the MR approach reduces concerns about confounding and reverse causation compared with observational studies, the findings should be interpreted cautiously given the study’s sample size, population specificity, and other limitations. Further research in larger and more diverse populations is warranted to confirm these observations and to inform the development of dietary recommendations.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

HDL-C

High-density lipoprotein cholesterol

MR

Mendelian randomization

KoGES

Korea Genome and Epidemiology Study

HEXA

Health Examinees Study

BMI

Body mass index

TC

Total cholesterol

LDL-C

Low-density lipoprotein cholesterol

TG

Triglycerides

SBP

Systolic blood pressure

DBP

Diastolic blood pressure

GWAS

Genome-wide association study

IVs

Instrumental variables

IVW

Inverse-variance weighted

Author contributions

DS conceptualized and designed the study, secured funding, and supervised the research. JC and DS designed the research methodology. JC analyzed the data and performed statistical analysis. DS provided essential materials. JC, SHK, and DS wrote and revised the manuscript. DS had primary responsibility for the final content. All authors read and approved the final manuscript.

Funding

This work was supported by a National Research Foundation of Korea grant funded by the Korean Government (MSIT) (RS-2024-00340086).

Data availability

The data described in the manuscript, code book, and analytic code will not be made available. These data are available with permission from the National Biobank of Korea of the Korea Disease Control and Prevention Agency.

Declarations

Ethical approval

The study was approved by the Institutional Review Board of Inha University (IRB No. 240307–1 A). All participants provided written informed consent prior to participation, and the study complied with the Declaration of Helsinki.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.