Dairy consumption and cardiovascular disease risk: a multi-level analysis with inflammatory biomarker mediation

Abstract

Objective

Including dairy products in the diet may help reduce the risk of developing cardiovascular disease (CVD). Nevertheless, how various categories of dairy items influence CVD risk remains insufficiently explored. This study aimed to evaluate the associations between different types of dairy products, their intake frequency, fat content, and the risk of CVD, addressing existing research gaps.

Methods

Dairy intake data—including yogurt, cheese, and milk—were obtained from the Global Dietary Database, covering 185 countries between 1990 and 2018. Corresponding CVD burden metrics—including incidence, prevalence, and disability-adjusted life years (DALYs)—were extracted from the Global Burden of Disease study. These data were used to investigate how dairy intake relates to CVD burden globally. Additionally, individual-level data from 30,341 participants in the National Health and Nutrition Examination Survey (NHANES) were analyzed to evaluate the association between dairy product consumption and CVD risk. Finally, directed acyclic graph and causal mediation models were used to confirm whether inflammatory markers partially mediated the association between dairy consumption and CVD risk.

Results

Globally, individuals who consumed greater amounts of milk tended to have a lower risk of CVD. Individually, milk intake was found to be negatively associated with the risk of CVD, coronary heart disease (CHD), and heart attack (HA). Exploratory mediation analysis indicated that systemic inflammatory markers (SIRI), the monocyte-to-lymphocyte ratio (MLR), and the neutrophil-to-lymphocyte ratio (NLR) partially mediated the relationship between milk intake and CVD risk, contributing − 6.09%, −5.02%, and − 4.13% of the total association, respectively. Further analysis showed that individuals who reported drinking milk daily or at least weekly tended to have a lower risk of developing CVD than those who consumed it less than once per week.

Conclusion

Milk intake may offer protective effects against CVD, CHD, and HA, potentially through the mitigation of inflammation. The frequency of milk intake appears to be a key factor influencing CVD risk. These findings highlight the potential public health benefits of promoting regular milk intake as part of a balanced dietary pattern to help reduce the global burden of CVD.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12986-025-01060-6.

Introduction

Cardiovascular disease (CVD), a leading non-communicable disease, poses a major global public health challenge due to its high incidence, long duration, high fatality, and substantial healthcare burden [1]. Its development is closely linked to multiple lifestyle-related risk factors, including hypertension, obesity, smoking, air pollution, and unhealthy diets [2].

Among various interventions, dietary regulation is considered a key strategy for preventing and managing CVD. According to the 2021 Global Burden of Disease (GBD) study [3], dietary risks were responsible for 5.83 million CVD deaths globally in 2021, as well as approximately 134 million disability-adjusted life years (DALYs). Although moderate dairy intake is recommended in many guidelines, its role in heart-healthy diets remains controversial [4]. Dairy products provide proteins, phospholipids, vitamins, and minerals that may protect against CVD, yet their saturated and trans fats raise concerns [5]. Epidemiological evidence on dairy intake and CVD risk is inconsistent. Several cohort studies suggest protective effects, particularly for fermented and low-fat dairy products, which may improve lipid metabolism and reduce inflammation [6–8]. In contrast, others link high consumption of full-fat milk or cheese to elevated CVD risk, possibly due to their saturated fat content and energy density [9, 10]. These discrepancies likely reflect differences in study design, populations, and types of dairy assessed. Hence, large-scale, standardized studies are needed to clarify the role of specific dairy subtypes in CVD prevention.

Inflammation plays a central role in the onset and progression of CVD [11, 12]. Biomarkers such as the neutrophil-to-lymphocyte ratio (NLR) [13], high-sensitivity C-reactive protein (hs-CRP) [14], and the C-reactive protein-albumin-lymphocyte (CALLY) index [15] are established predictors of CVD risk. Fermented dairy products may have anti-inflammatory effects and potentially confer protection against chronic inflammation-related diseases [16]. For example, postbiotics from Lactiplantibacillus plantarum isolated from fermented dairy products have been shown to alleviate intestinal inflammation [17]. However, most studies on dairy intake and CVD risk have been geographically limited and have varied by demographic factors such as gender, age, education, residence, country, and survey year [18]. Therefore, comprehensive analyses using large-scale, multi-level datasets are needed to better characterize the relationship.

This study examined the association between dairy consumption and CVD risk from both global and individual perspectives. We hypothesized that higher dairy consumption particularly cheese, yogurt, and milk, is associated with a lower CVD burden, partly through anti-inflammatory mechanisms. To test this hypothesis, we combined data from 185 countries using the Global Dietary Database (GDD) and the Global Burden of Disease (GBD) datasets to evaluate the association between dairy consumption and CVD burden. We then validated the association using data from the National Health and Nutrition Examination Survey (NHANES), and further investigated whether it was mediated by inflammatory biomarkers. Finally, we examined the associations between milk intake frequency, fat content, and the CVD risk.

Methods

Data source

This research utilized data from the GDD, which compiles dietary survey records from 185 countries and regions across seven waves (1990, 1995, 2000, 2005, 2010, 2015, and 2018). Data were obtained through collaboration with international researchers and publicly available sources [19, 20]. Priority was given to nationally or regionally representative surveys using standardized dietary assessment methods, such as 24 h recalls, food frequency questionnaires, and food records, to ensure comparability and validity. Dietary data were extracted according to unified standards and stratified by age, gender, education level (low: < 6 years, medium: 6–12 years, high: >12 years), and residence (rural/urban). The GDD pre-processed these data, applying a Bayesian hierarchical modeling framework to estimate energy-adjusted mean dairy intake and its uncertainty for each subgroup. To enhance comparability across population subgroups and address potential heterogeneity in dietary assessment methods and definitions, the GDD pre-processed the data by standardizing food, beverage, and nutrient classifications, implementing quality control measures, and, where necessary, applying a Bayesian hierarchical modeling framework to estimate energy-adjusted mean dairy intake and its uncertainty for each subgroup [21].

CVD burden data were obtained from the GBD Study 2021 (GBD 2021), covering the period from 1990 to 2021 and coordinated by the Institute for Health Metrics and Evaluation at the University of Washington. We extracted matched estimates of CVD incidence, prevalence, and DALYs. To ensure temporal consistency, the GDD dietary data were matched with GBD estimates by survey year, country (or region) and 5-year age group.

Individual-level data were derived from NHANES, a nationally representative, large-scale cross-sectional survey implemented by the U.S. Centers for Disease Control and Prevention and the National Center for Health Statistics [22]. Among 86,618 individuals surveyed in eight cycles of NHANES (2003–2020), 30,341 eligible participants were ultimately identified after excluding individuals aged < 20 years (N = 38,165), those with missing data on dairy intake, frequency, or fat content (N = 18,108), and those lacking CVD outcome data (N = 4). Notably, exclusion categories were not mutually exclusive. The sample selection process is detailed in Fig. 1.

Fig. 1.

Study design and sample selection process. CVD, cardiovascular disease; GDD, Global Dietary Database; GBD, Global Burden of Disease; NHANES, National Health and Nutrition Examination Survey

Definition of dairy products

For the global-level analysis, dairy consumption data were obtained from the GDD. We extracted daily intake data (g/day) for dairy products, including yogurt, cheese, and milk, across the survey years 1990, 1995, 2000, 2005, 2010, 2015, and 2018. These data were stratified by age, gender, education level, place of residence, and country/region.

For the individual-level analysis, dairy intake data were obtained from the NHANES database, based on two 24-hour dietary recall interviews conducted with each participant. The assessment included daily intake (g/day) of various types of dairy products—including cheese, yogurt, and milk—as well as total dairy intake. Frequency of milk product consumption was assessed using the NHANES question: “During the past 30 days, how often did you consume milk products (including chocolate or other flavored milk and hot cocoa made with milk, but excluding small amounts added to coffee or tea)?” Participants were categorized as rarely (< 1/week), sometimes (1–6/week), often (≥ 1/day), or varied. The type of milk consumed was determined from the NHANES item, “What type of milk do you usually drink?” with response options including whole/regular milk, 2% fat milk, 1% fat milk, free/skim milk, another type of milk, or varied (i.e., mixed varieties). As information on mixed-type consumption is not available, milk fat content was classified based on the type participants reported as their usual choice.

Collection and definition of CVDs

For the global-level analysis, data on the incidence, prevalence, and DALYs associated with nine specific subtypes of CVD were obtained from the GBD 2021 study. The detailed classification of these nine CVD subcategories is presented in Fig. 1. For more information on the definitions of CVD outcomes and detailed descriptions of data collection methodologies, please refer to the official GBD website: https://www.healthdata.org/gbd/methods-appendices-2021.

For the individual-level analysis, CVD status was determined based on a standardized questionnaire. Participants were asked: “Has a doctor or other health professional ever told you that you had congestive heart failure (CHF), coronary heart disease (CHD), angina, heart attack (HA), or stroke?” Participants who answered “yes” to any of these conditions were classified as having CVD.

Inflammatory biomarkers

Complete blood cell (CBC) counts, including neutrophil counts, lymphocyte counts, white blood cell counts, and monocyte counts were measured using Beckman Coulter technology. Using these four hematological parameters, the five inflammation-derived indices were calculated: systemic immune-inflammation index (SII) [23], systemic inflammation response index (SIRI) [24], aggregate index of systemic inflammation (AISI) [25], neutrophil-to-lymphocyte ratio (NLR) [25], and monocyte-to-lymphocyte ratio (MLR) [25]. Detailed calculation formulas are shown in Supplementary Table 1.

Covariate assessment

In the global-level, several socioeconomic, lifestyle, and disease-related factors may be associated with CVD; therefore, relevant internationally available data were included as covariates. Gross domestic product (GDP, US dollars per capita) and urbanization rate (percentage of urban population) were obtained from the World Bank. Total energy intake was obtained from the Food and Agriculture Organization of the United Nations (FAO), while male prevalence (percentage of male population), aging rate (percentage of the population aged ≥ 60 years), obesity rate (body mass index (BMI) > 30%), high low density lipoprotein (HLDL) rate, smoking rate, alcohol consumption rate, socio-demographic Index (SDI), hypertension, diabetes, chronic kidney disease, cancer, and survey year were obtained from the GBD database. As a result, further adjustment for additional covariates was not feasible.

In the individual-level NHANES analysis, a total of 20 covariates were selected based on published clinical studies [19, 26] and the construction of directed acyclic graphs (Supplementary Fig. 1). These variables covered demographic characteristics, lifestyle behaviors, laboratory parameters, and health-related conditions. Demographic characteristics included age, gender (male/female), survey year, race (Mexican American, Other Hispanic, Non-Hispanic White, Non-Hispanic Black, and Other Race), educational level (less than high school, high school or equivalent, college or above), marital status (married/living with a partner, divorced/separated/widowed, never married), and poverty income ratio (PIR). Lifestyle factors included smoking status (never smoker, former smoker, current smoker), drinking status (never drinker, former drinker, current drinker), and BMI. Laboratory measurements included serum creatinine (SCr, mg/dL), serum uric acid (SUA, mg/dL), serum total cholesterol (STC, mg/dL), urinary albumin-to-creatinine ratio (ACR, mg/g), estimated glomerular filtration rate (eGFR, mL/(min*1.73 m²)), and total energy intake (Kcal/d). Health conditions included the presence of chronic kidney disease (CKD, yes/no), diabetes (yes/no), hypertension (yes/no), hyperlipidemia (yes/no), and cancer (yes/no).

Statistical analysis

In the global-level analysis, descriptive statistics were first calculated for each country by survey year. Analysis of variance and general linear models were used to assess differences and temporal trends, respectively. Negative binomial regression models with a log(population) offset were then applied to evaluate the associations of yogurt, cheese, and milk intake with CVD incidence, prevalence, and DALYs. Measurement uncertainty was incorporated through inverse-variance weighting based on reported uncertainty intervals. To account for repeated measure within countries, mixed-effects models with random intercepts was employed. Non-linear associations between dietary exposure variables and CVD outcomes were explored using generalized linear mixed models incorporating restricted cubic splines (RCS).

For the individual-level NHANES analysis, all statistical procedures incorporated appropriate survey weights to reflect the complex, multistage sampling design. The combined survey weight was derived using the standardized algorithm: Final weight = WTDR2D × 1/ (7 + 1.625) + WTDR2DPP × 1.625/ (7 + 1.625), where WTDR2D denotes the 2-day dietary weight for the 2003–2016 NHANES cycles and WTDR2DPP for the 2017–2020 cycle.

Baseline characteristics of participants with and without CVD were summarized using survey-weighted models. Continuous variables were expressed as mean ± standard deviation (SD) or median and interquartile range (IQR), compared using ANOVA or Kruskal-Wallis H test as appropriate, while categorical variables were expressed as frequencies (percentages) and compared using the chi-square test. Weighted logistic regression models were used to examine associations between dairy intake (amount, frequency, and fat content) and CVD risk. RCS and subgroup analyses were conducted to explore potential relationships. The mediating role of inflammation-related biomarkers in the association between dairy intake and CVD was evaluated, with sensitivity analyses for unmeasured mediator–outcome confounding [27]. The total effect was partitioned into the average direct effect (ADE) and average causal mediation effect (ACME), and the mediation proportion was estimated as ACME / (ACME + ADE) using 1,000 bootstrap replications. Sensitivity analyses assessed robustness by excluding participants with missing covariate data and applying multiple imputation by chained equations (MICE) for missing values, accounting for survey design variables (final weights, strata, and primary sampling units). Missing data patterns are summarized in Supplementary Table 2.

All analyses were performed using R software (version 4.3.1), with key packages including glmmTMB (for negative binomial regression modeling), survey (for complex survey design), rms (for logistic regression modeling), ggplot2 (for data visualization), mediation (for mediation analysis), and mice (for multiple imputation). A two-tailed P value < 0.05 was considered statistically significant.

Results

Characteristics of variables by year at the global level

Supplementary Table 3 presents temporal trends in dairy product consumption (including yogurt, cheese, and milk) and other demographic, socioeconomic, lifestyle, and disease-related factors. Median intakes of yogurt, cheese, and milk, as well as the incidence and prevalence of CVD, showed no significant changes over time, whereas CVD-related DALYs significantly declined from 1990 to 2018 (p < 0.001). In contrast, the aging rate, urbanization rate, total energy intake, SDI, and the risks of hypertension, diabetes, and cancer increased significantly during the same period (p < 0.05).

Association between global dairy consumption and CVD/its subtypes

Figure 2 illustrates the associations between dairy product consumption and CVD and its subtypes. In Model 1, which was unadjusted for covariates, milk intake was inversely associated with the incidence, prevalence, and DALYs of CVD. After adjustment for multiple covariates in Model 3, milk intake remained significantly associated only with a lower prevalence of CVD. However, the associations differed across CVD subtypes. For instance, milk intake was negatively associated with the incidence, prevalence, and DALYs of non-rheumatic degenerative mitral valve disease. RCS analyses further revealed non-linear associations between the consumption of cheese, yogurt, and milk and the incidence, prevalence, and DALYs of CVD (P for nonlinearity < 0.05), as illustrated in Supplementary Fig. 2.

Fig. 2.

Association between total dairy consumption (yogurt, cheers, and milk) and the global burden of CVDs. A incidence, B prevalence, C DALYs. CVD, cardiovascular disease; DALYs, disability-adjusted life years; GDP, gross domestic product; HLDL, high low density lipoprotein; SDI, socio-demographic index; CKD, chronic kidney diseases. Model 1: crude model. Model 2: adjusted for male rate, aging rate, urban rate, GDP, obesity rate, HLDL rate, smoke rate, alcohol rate, energy intake, SDI. Model 3: adjusted for male rate, aging rate, urban rate, GDP, obesity rate, HLDL rate, smoke rate, alcohol rate, energy intake, SDI, Hypertension, Diabetes, CKD, cancer and survey year

Baseline characteristics of participants in the NHANES

As shown in Table 1, a total of 30,341 participants in the NHANES dataset met the inclusion criteria. The median age was 49.0 years (IQR: 34.0–65.0). Among them, 3,505 participants (11.6%) were identified as having CVD. Relative to non-CVD individuals, participants diagnosed with CVD tended to be older and exhibited higher levels of SCr, SUA, and ACR, but lower levels of STC, energy intake, and eGFR. In addition, the weighted proportion of CVD patients was significantly higher among the following subgroups, including non-Hispanic Whites, males, individuals with an education level of high school or equivalent or less than high school, those who were divorced/separated/widowed, participants with a PIR ≤ 3.5, former smokers, former drinkers, individuals with a BMI ≥ 30, and those with comorbid conditions such as diabetes, hypertension, CKD, or cancer.

Table 1.

Clinical characteristics of participants with and without CVD in NHANES

Characteristics	Total (n = 30,341)	CVD (n = 3,505)	Non-CVD (n = 26,836)	p-value
Age, median [IQR], (years)	49.00 [34.00, 65.00]	69.00 [60.00, 78.00]	46.00 [32.00, 62.00]	< 0.001
Gender				< 0.001
Male	14,498 (47.62)	1985 (53.50)	12,513 (47.05)
Female	15,843 (52.38)	1520 (46.50)	14,323 (52.95)
Race, n (%)				< 0.001
Mexican American	4819 (8.12)	333 (3.99)	4486 (8.52)
Other Hispanic	2600 (4.68)	210 (2.57)	2390 (4.88)
Non-Hispanic White	14,058 (71.36)	2064 (78.30)	11,994 (70.69)
Non-Hispanic Black	6277 (10.05)	720 (10.27)	5557 (10.03)
Other Race	2587 (5.79)	178 (4.88)	2409 (5.88)
Education level, n (%)				< 0.001
Less than high school	7021 (15.89)	1076 (25.13)	5945 (14.99)
High school or equivalent	7173 (23.40)	932 (28.04)	6241 (22.95)
College or above	16,117 (60.72)	1489 (46.83)	14,628 (62.06)
Marital status, n (%)				< 0.001
Married/living with partner	18,463 (65.65)	1940 (60.19)	16,523 (66.18)
Divorced/separated/ widowed	6582 (17.72)	1322 (33.83)	5260 (16.16)
Never married	5280 (16.63)	241 ( 5.98)	5039 (17.66)
PIR, n (%)				< 0.001
< 1.3	8434 (20.42)	1141 (26.83)	7293 (19.80)
1.3–3.5	10,815 (36.69)	1377 (45.11)	9438 (35.88)
> 3.5	8755 (42.89)	712 (28.06)	8043 (44.32)
Smoking status, n (%)				< 0.001
Never	16,858 (54.84)	1354 (37.41)	15,504 (56.53)
Former	7597 (25.11)	1468 (41.54)	6129 (23.52)
Current	5871 (20.05)	680 (21.05)	5191 (19.96)
Drinking status, n (%)				< 0.001
Never	3936 (11.65)	444 (12.73)	3492 (11.54)
Former	5784 (12.85)	858 (18.58)	4926 (12.29)
Current	19,143 (75.50)	2059 (68.69)	17,084 (76.17)
BMI, n (%), (kg/cm2)				< 0.001
< 25	8268 (30.11)	705 (20.62)	7563 (31.02)
25–30	9942 (33.40)	1114 (32.43)	8828 (33.49)
≥ 30	11,787 (36.48)	1582 (46.95)	10,205 (35.49)
SCr, mean ± SD, (mg/dL)	0.91 ± 0.43	1.13 ± 0.72	0.88 ± 0.36	< 0.001
SUA, mean ± SD, (mg/dL)	5.41 ± 1.44	5.97 ± 1.61	5.33 ± 1.41	< 0.001
STC, mean ± SD, (mg/dL)	193.49 ± 42.30	180.69 ± 45.17	195.15 ± 41.63	< 0.001
Energy intake, mean ± SD, (Kcal/d)	2122.37 ± 992.73	1846.21 ± 877.20	2158.44 ± 1001.24	< 0.001
ACR, median [IQR], (mg/g)	7.16 [4.57, 13.88]	12.12[6.27, 37.40]	6.83 [4.44, 12.50]	< 0.001
eGFR, median [IQR], (ml/(min*1.73 m2))	91.67[74.65, 107.62]	69.48[53.89, 86.13]	94.13[78.10, 109.48]	< 0.001
Diabetes, n (%)	5797 (14.69)	1475 (38.41)	4322 (12.39)	< 0.001
Hypertension, n (%)	9301 (81.40)	2324 (87.72)	6977 (79.82)	< 0.001
CKD, n (%)	5826 (14.92)	1657 (42.85)	4169 (12.22)	< 0.001
Cancer, n (%)	3086 (10.22)	758 (22.60)	2328 (9.01)	< 0.001

PIR, poverty income ratio; BMI, body mass index; SCr, serum creatinine; SUA, serum uric acid; STC, serum total cholesterol; ACR, albumin-to-creatinine ratio; eGFR, estimated glomerular filtration rate; SD, standard deviation; IQR, interquartile range

Association between individual dairy intake and CVD/its subtypes

Figure 3 and Supplementary Tables 4–9 illustrate the associations between varying dairy consumption and the risk of CVD and its subtypes within the NHANES dataset. After controlling for various confounding variables, higher milk intake was associated with reduced risks of both CVD and HA; and a comparable inverse relationship was likewise observed with total dairy consumption. Participants consuming milk more than once a week or daily had significantly lower risks of developing CVD compared to those consuming it less than once a week (Fig. 4). Furthermore, the analysis did not reveal any meaningful association linking milk fat levels with CVD risk (Fig. 5). However, compared to regular whole milk consumers, individuals who frequently consumed skim milk were more likely to develop CHD but had the lowest risk of stroke.

Fig. 3.

Association between dairy intake and CVD incidence. CVD, cardiovascular disease; CHF, congestive heart failure; CHD, coronary heart disease; HA, heart attack. Model 1 Adjusted for age, gender, race, education level, marital status, PIR, and survey year. Model 2 Adjusted for age, gender, race, education level, marital status, PIR, survey year, smoking status, drinking status, and energy intake. Model 3 Adjusted for age, gender, race, education level, marital status, PIR, survey year, smoking status, drinking status, energy intake, BMI, SCr, SUA, STC, ACR, eGFR, diabetes, hypertension, hyperlipidemia, CKD, and cancer

Fig. 4.

Association between milk intake frequency (per month) and CVD incidence. CVD, cardiovascular disease; CHF, congestive heart failure; CHD, coronary heart disease; HA, heart attack. Model 1 Adjusted for age, gender, race, education level, marital status, PIR, and survey year. Model 2 Adjusted for age, gender, race, education level, marital status, PIR, survey year, smoking status, drinking status, and energy intake. Model 3 Adjusted for age, gender, race, education level, marital status, PIR, survey year, smoking status, drinking status, energy intake, BMI, SCr, SUA, STC, ACR, eGFR, diabetes, hypertension, hyperlipidemia, CKD, and cancer

Fig. 5.

Association between milk fat content and CVD incidence. Abbreviations: CVD, cardiovascular disease; CHF, congestive heart failure; CHD, coronary heart disease; HA, heart attack. Model 1 Adjusted for age, gender, race, education level, marital status, PIR, and survey year. Model 2 Adjusted for age, gender, race, education level, marital status, PIR, survey year, smoking status, drinking status, and energy intake. Model 3 Adjusted for age, gender, race, education level, marital status, PIR, survey year, smoking status, drinking status, energy intake, BMI, SCr, SUA, STC, ACR, eGFR, diabetes, hypertension, hyperlipidemia, CKD, and cancer

Moreover, RCS analysis revealed a linear dose–response relationship between both milk intake and total dairy intake and CVD risk (P for non-linearity > 0.05), as shown in Supplementary Fig. 3.

Mediation analysis

We further evaluated whether inflammation-related biomarkers (SII, SIRI, AISI, MLR, and NLR) mediated the association between milk intake and CVD risk (Fig. 6A–E and supplementary Table 10). It was found that SIRI, MLR, and NLR may partially explain the association, with mediational contributions of -6.09%, -5.02%, and − 4.13%, respectively. Furthermore, mediation sensitivity analysis indicated that these results were less affected by unmeasured mediator-outcome confounders (Supplementary Fig. 4).

Fig. 6.

Mediation analysis of the association between milk intake and CVD through inflammation-related biomarkers. (A) SII; (B) SIRI; (C) AISI; (D) MLR; (E) NLR. CVD, cardiovascular disease; SII, systemic immune-inflammation index; SIRI, systemic inflammation response index; AISI, systemic inflammation response index; MLR, monocyte-to-lymphocyte ratio; NLR, neutrophil-to-lymphocyte ratio

All models were adjusted for age, gender, race, education level, marital status, PIR, smoking status, drinking status, BMI, SCr, SUA, STC, ACR, eGFR, energy intake, survey year, diabetes, hypertension, hyperlipidemia, CKD, and cancer.

Subgroup analysis

Subgroup analyses were conducted according to major covariates (Table 2). Fully adjusted models (excluding the stratification variables) revealed an inverse correlation was observed linking milk intake with the risk of CVD in specific subgroups, including females, married or living with partner, PIR between 1.3 and 3.5, current drinkers, diabetes, and non-CKD. In addition, no significant interaction effects were identified between milk intake frequency and CVD across different subgroups (P for interaction > 0.05).

Table 2.

Subgroup analysis and interaction tests of the association between milk intake and CVD incidence

Characteristics	N (%)	Milk intake
		OR (95%CI)	p	P for interaction
Age				0.267
20–49	12,113 (49.6)	0.96 (0.91, 1.02)	0.158
≥ 50	12,321 (50.4)	0.96 (0.83, 1.10)	0.524
Gender				0.076
Male	11,821 (48.4)	0.96 (0.90, 1.03)	0.278
Female	12,613 (51.6)	0.90 (0.82, 0.99)	0.035
Marital status				0.315
Married/living with partner	15,098 (61.8)	0.90 (0.84, 0.97)	0.006
Divorced/separated/ widowed	5211 (21.3)	1.00 (0.92, 1.09)	0.955
Never married	4125 (16.9)	1.02 (0.84, 1.23)	0.857
PIR				0.657
< 1.3	7200 (29.5)	0.97 (0.89, 1.06)	0.554
1.3–3.5	9434 (38.6)	0.92 (0.84, 1.00)	0.048
> 3.5	7800 (31.9)	0.95 (0.85, 1.07)	0.393
BMI				0.653
< 25	6658 (27.2)	0.98 (0.87, 1.09)	0.704
25–30	8157 (33.4)	0.93 (0.85, 1.02)	0.143
≥ 30	9619 (39.4)	0.93 (0.86, 1.01)	0.099
Smoking status				0.569
Never	13,460 (55.1)	0.95 (0.87, 1.04)	0.254
Former	6266 (25.6)	0.96 (0.89, 1.05)	0.365
Current	4708 (19.3)	0.91 (0.81, 1.02)	0.090
Drinking status				0.811
Never	3178 (13)	0.99 (0.84, 1.16)	0.906
Former	4803 (19.7)	0.95 (0.85, 1.07)	0.400
Current	16,453 (67.3)	0.94 (0.88, 1.00)	0.048
Diabetes				0.009
Yes	4709 (19.3)	0.86 (0.79, 0.95)	0.002
No	19,725 (80.7)	0.99 (0.93, 1.06)	0.789
Hypertension				0.790
Yes	10,689 (43.7)	0.95 (0.90, 1.01)	0.132
No	13,745 (56.3)	0.93 (0.84, 1.04)	0.187
CKD				0.202
Yes	4694 (19.2)	0.98 (0.90, 1.07)	0.661
No	19,740 (80.8)	0.92 (0.86, 0.98)	0.016
Cancer				0.411
Yes	2471 (10.1)	0.92 (0.81, 1.03)	0.157
No	21,963 (89.9)	0.96 (0.90, 1.01)	0.139

The model adjust for age, gender, race, education level, marital status, PIR, survey year, smoking status, drinking status, energy intake, BMI, SCR, SUA, STC, ACR, eGFR, diabetes, hypertension, hyperlipidemia, CKD, and cancer, excluding the stratification variables

Sensitivity analysis

First, we removed 5,907 individuals with incomplete covariate information from the dataset, and reanalyzed the data. The findings were largely in agreement with the primary analysis outcomes (Supplementary Table 11). Next, to handle missing covariate information, we applied multiple imputation techniques before reanalyzing the data. The results were consistent with primary findings, reinforcing the study’s conclusions (Supplementary Table 12).

Discussion

This study was the first to systematically analyze how dairy intake, including yogurt, cheese, and milk, affects CVD risk at both global and individual levels. This study further elucidated the potential mediating role of inflammation in this association. Our results demonstrated an inverse association between regular milk intake and CVD risk, with partial mediation by inflammatory markers (SIRI, MLR, NLR). Notably, the observed association remained significant after adjustment for key confounders. Further analysis demonstrated that the frequency of milk intake-rather than its fat content-was the key determinant of CVD risk. Individuals with frequent milk intake were found to have a significantly lower probability of CVD risk than those consuming it less often, whereas milk fat content showed no meaningful correlation with CVD risk. This study suggest that milk intake may have a preventive effect on CVD, particularly among individuals at elevated risk. However, further research is needed to verify these findings.

Dairy consumption represents a culturally integral component of the diet in many Western societies. Numerous clinical studies have provided evidence suggesting that milk intake is linked to a lower likelihood of developing conditions such as hyperglycemia, dyslipidemia, and hypertension [28]. Thorning et al. [29] conducted a review and found that consumption of full-fat milk products was inversely associated with CVD risk. However, another comprehensive meta-analysis indicated that total milk intake did not show a significant link with CVD risk, although it was related to a lower likelihood of hypertension [30]. Our research consistently identified a negative correlation between milk intake and CVD risk, observable across both global and individual levels. Milk contains a variety of bioactive components with potential cardioprotective effects, including specific saturated fatty acids (SFAs), conjugated linoleic acid, sphingolipids, calcium, and tripeptides. Certain SFAs, such as C15 and C17, have been suggested to protect against CVD [31], while conjugated linoleic acid and sphingolipids may improve endothelial function and reduce inflammation [32]. Adequate calcium intake has been linked to improved lipid profiles through reductions in STC and low-density lipoprotein cholesterol [33]. Moreover, milk-derived tripeptides exert antihypertensive effects [34], and milk polar lipids from the fat globule membrane can modulate cholesterol absorption and decrease lipid biomarkers associated with CVD risk [35]. Taken together, these findings support the view that milk intake contributes to cardiovascular protection rather than harm. Promoting moderate milk intake as part of a balanced diet may therefore represent an effective dietary strategy for CVD prevention.

Yogurt and cheese contain unique fermentation-derived metabolites such as short-chain fatty acids, bioactive peptides, and probiotics [36, 37], which may modulate gut microbiota, improve lipid metabolism, and reduce inflammation, thereby potentially lowering CVD risk [38]. However, in the present study, no significant associations were observed between yogurt or cheese intake and CVD incidence. Several explanations are plausible. Many commercially available yogurts, particularly flavored varieties, contain added sugars to enhance palatability. Excessive sugar intake may promote obesity, insulin resistance, and dyslipidemia, thereby counteracting the potential cardioprotective effects of fermentation metabolites. Cheese, while rich in protein, calcium, and other beneficial nutrients, typically contains a relatively high amount of sodium. High sodium intake is known to elevate blood pressure and cardiovascular risk, which may offset the beneficial effects of its other components. Moreover, the anti-inflammatory and lipid-lowering effects of fermented dairy products may be relatively modest and thus insufficient to yield measurable reductions in CVD incidence at the population level. In summary, although fermented dairy products contain bioactive components with potential cardiovascular benefits, their net effect on CVD risk is likely influenced by the sodium, sugar, and fat content of these products.

Inflammation, whether systemic or localized, is deeply implicated in both the onset and advancement of CVD. This process spans multiple phases, from early endothelial dysfunction to atherosclerotic plaque development and eventual rupture leading to adverse clinical outcomes [39]. Previous research has indicated that many inflammatory biomarkers such as the NLR [13], CALLY [15], and hs-CRP [14], can effectively predict CVD risk independently of traditional risk factors. Milk contains bioactive molecules like conjugated linoleic acid, immunoglobulins, antioxidant enzymes, and anti-inflammatory peptides, which provide various health benefits to humans [40]. These active ingredients act through various mechanisms, including regulating immune responses, downregulation of inflammatory mediators, and promoting tissue repair, to produce anti-inflammatory effects [40, 41]. Moreover, lactoferrin, as an important bioactive molecule, not only plays a crucial role in milk but also demonstrates significant benefits in anti-inflammatory and immune-regulatory aspects [41, 42]. Based on NHANES data, our results indicate that SIRI, MLR, and NLR may partly mediate how milk intake influences CVD risk, with mediation effects of -6.09%, -5.02%, and − 4.13%, respectively. The relatively small mediating effect indicates that inflammation explains only a minor part of the association between milk intake and CVD. This aligns with the multifactorial nature of CVD, which arises from complex interactions among metabolic, inflammatory, hormonal, and lifestyle pathways. Although inflammation is a plausible biological mechanism linking dairy consumption to cardiovascular health, other processes, such as lipid metabolism, endothelial function, insulin sensitivity, and gut microbiota regulation, are likely to play important roles. Future longitudinal and mechanistic studies are warranted to clarify the contributions of these interrelated pathways.

This study further investigated the impact of milk intake frequency and fat content on CVD risk. Our findings showed that individuals with milk intake more than once per week or on a daily basis exhibited a markedly reduced risk of developing CVD, relative to those with less frequent consumption. This finding further supports current dietary guidelines that recommend including milk as part of a healthy diet. This conclusion is consistent with evidence from related studies. For example, researchers in Japan carried out a prospective cohort investigation and found that women with a weekly milk intake between 7 and 12 cups exhibited a significantly reduced risk of ischemic stroke relative to those consuming less than 2 cups per week [43]. This protective effect may be partly explained by the fact that frequent milk consumers tend to follow healthier diets. These diets typically include more nutrient-rich foods such as fish, dark-colored berries, whole grains, and soy-based items [44, 45]. In addition, no significant association was observed between milk fat content and CVD in this study. This finding is consistent with a prospective study by Patterson et al. [46], which reported that substituting regular-fat milk (≥ 3.0%) with low-fat milk (≤ 1.5%) did not significantly affect the risk of myocardial infarction. However, the association between whole-fat and low-fat milk intake and CVD outcomes remains controversial. Some studies have suggested potential cardiometabolic benefits of whole-fat milk. For instance, Talaei et al. [47] reported in the Isfahan Cohort Study that participants consuming 1–3 cups of whole milk per week had a significantly lower risk of CVD compared with non-consumers. Conversely, other studies have favored low-fat milk; for example, the Norway cohort study found that whole milk intake was associated with higher CVD and all-cause mortality, whereas low-fat milk was not [48]. In the present analysis, milk fat categories were derived from self-reported usual intake, and information on fortified, flavored, or mixed dairy products was unavailable, which may have led to misclassification bias. Therefore, the lack of a significant association observed in this study does not preclude a possible role of milk fat composition in cardiovascular health. Further studies using detailed dietary classification and longitudinal designs are warranted to clarify these associations.

This study did not fully elucidate the underlying mechanisms linking milk intake, inflammation reduction, and decreased CVD risk. However, existing literature offers several plausible biological explanations. First, milk contains several biofunctional constituents—including lactoferrin as well as vitamins D and K—which may act synergistically to alleviate systemic inflammation through multiple pathways, including downregulation of inflammatory mediators, regulating immune responses, and promoting tissue repair [41, 49, 50]. Second, specific milk components, including SFAs (especially C15 and C17), bioactive tripeptides, and polar lipids, are believed to protect the cardiovascular system by lowering blood pressure and reducing cholesterol absorption [34, 35]. In addition, inflammation and immune dysregulation are central mechanisms in the development of many CVD [51]. Therefore, milk intake may indirectly reduce the risk of CVD through this pathway. This hypothesis merits further investigation through experimental and interventional studies in the future.

The unique strength of this study lies in its dual-perspective approach, examining the associations between dairy product intake (yogurt, cheese, and milk) and the risk of CVD and its subtypes at both the global and individual levels. This multi-level research design helps reduce potential biases arising from cross-country variability, including demographic composition and timing of data collection, thereby enhancing the significant universal value and credibility of research conclusions. Nevertheless, several potential limitations should be acknowledged. First, dairy intake in both the GDD and NHANES was assessed through 24-hour dietary recalls, which are inherently prone to misreporting and recall bias. While previous validation studies have indicated acceptable population-level accuracy in both datasets, some degree of under- or over-reporting of intake is likely unavoidable. Furthermore, the GDD incorporates food frequency questionnaires, dietary records, household consumption and food balance data. However, because it does not specify which assessment method was used for each data point, we were unable to perform subgroup analyses to compare the different methods. Second, accurately quantifying complex dairy intake in real-world diets (e.g., milk tea, lattes, milk tablets) is challenging due to difficulties in standardization and categorization. In addition, differences in sugar content across dairy products may independently influence CVD risk, potentially introducing bias. The NHANES database also does not distinguish between fortified and non-fortified milk or capture detailed co-consumption patterns. Furthermore, milk fat content was classified solely based on self-reported usual intake, which may affect the observed associations. Third, inherent methodological limitations exist in both the GDD and GBD studies, including the varying quality of data across countries and regions, which may introduce systematic errors. Fourth, although adjustments were made for several common covariates at both the global and individual levels, residual confounding cannot be completely ruled out. Fifth, CVD status in NHANES was determined from self-reported physician diagnoses, which may introduce misclassification due to recall bias or undiagnosed conditions. Although this non-differential misclassification would likely bias the results toward the null, the possibility of differential misclassification cannot be excluded. Sixth, due to its cross-sectional nature, the NHANES design is inherently limited in inferring causality between dairy intake and CVD risk. Moreover, CBC measurements were obtained only once, making them susceptible to intraindividual variability and measurement error. Future large-scale prospective cohort studies with repeated measurement and long-term follow-up are warranted to validate these associations and further elucidate potential causal links. Seventh, multiple dietary exposures and cardiovascular outcomes were examined, which may increase the likelihood of false-positive findings due to multiple comparisons. However, the present study was structured as a two-stage analysis (global discovery followed by independent validation using NHANES), and the consistency of associations across datasets and analytical approaches strengthens the credibility of the observed associations. Future research with prespecified hypotheses and multiplicity adjustment is warranted to further confirm these findings. Eighth, as some findings of this study were derived solely from NHANES data, these results are specific to the U.S. population and may not be generalizable to populations in other countries or regions. Lastly, temporal relationships among exposure, mediator, and outcome cannot be inferred for the cross-sectional of NHANES. As mediation analysis depends on strong, untestable assumptions, our findings should be viewed as exploratory and hypothesis-generating, not causal. Therefore, despite the consistency of associations observed across multiple levels of analysis, the results warrant prudent interpretation due to inherent study limitations.

Conclusion

This study reveals a negative association between milk intake and the risk of CVD, with this association partially mediated by reductions in inflammatory biomarkers. Furthermore, the frequency of milk intake emerges as a significant factor influencing CVD risk. The study found that milk intake may have a preventive effect on CVD and its subtypes, providing favorable epidemiological evidence and having important public health implications.

Abbreviations

AISI

Aggregate index of systemic inflammation

ACME

Average causal mediation effect

ADE

Average direct effect

BMI

Body mass index

CALLY

C-reactive protein-albumin-lymphocyte

CVD

Cardiovascular disease

CBC

Complete blood cell

CKD

Chronic kidney disease

CHF

Congestive heart failure

CHD

Coronary heart disease

DALYs

Disability-adjusted life years

eGFR

Estimated glomerular filtration rate

GBD

Global burden of disease

GDD

Global dietary database

GDP

Gross domestic product

HA

Heart attack

HLDL

High low density lipoprotein

hs-CRP

High-sensitivity C-reactive protein

IQR

Interquartile range

MLR

Monocyte-to-lymphocyte ratio

IQR

Median and interquartile range

NCHS

National center for health statistics

NHANES

National health and nutrition examination survey

NLR

Neutrophil-to-lymphocyte ratio

PIR

Poverty income ratio

RCS

Restricted cubic spline

SFAs

Saturated fatty acids

SDI

Socio-demographic Index

SCr

Serum creatinine

STC

Serum total cholesterol

SUA

Serum uric acid

SD

Standard deviation

SII

Systemic immune-inflammation index

SIRI

Systemic inflammation response index

ACR

Urinary albumin-to-creatinine ratio

Author contributions

D.M.contributed to conceptualization, investigation, formal analysis, and writing – original draft.X.C. contributed to conceptualization, investigation, formal analysis, and writing – original draft.Q.F. contributed to investigation and writing – original draft.S.N. contributed to formal analysis.F.W. contributed to data curation (secondary data analysis).P.H. contributed to conceptualization, supervision, and writing – review & editing.All authors read and approved the final manuscript.

Funding

No funding was received for this study.

Data availability

The data utilized in this study are publicly available from the following sources: the Global Dietary Database (GDD) at [https://globaldietarydatabase.org/](https:/globaldietarydatabase.org) ; the Global Burden of Disease (GBD) data used in the analyses can be accessed at [https://vizhub.healthdata.org/gbd-results/](https:/vizhub.healthdata.org/gbd-results) ; the National Health and Nutrition Examination Survey (NHANES) at https://www.cdc.gov/nchs/nhanes/index.htm.

Declarations

Ethics approval

Ethical approval was not applicable, as this study used publicly available and deidentified data from the GBD 2021 and GDD database. The NHANES protocol was reviewed and approved by the by the Institutional Review Board of the National Center for Health Statistics, with all participants providing written informed consent. As the current study utilized publicly available deidentified data, it was exempted from the requirement to obtain informed consent in accordance with the relevant regulations of the Zhejiang Chinese medical University Institutional Review Board.

Human ethics and consent to participate

Not applicable. This study analyzed only publicly available, de-identified data and involved no direct interaction with human participants.

Competing interests

The authors declare no competing interests.