CVD remains the leading cause of mortality worldwide, accounting for approximately 18 million deaths annually and representing nearly one-third of all global deaths(Reference Roth, Mensah and Johnson1,Reference Vaduganathan, Mensah and Turco2) . Despite substantial advances in pharmacological interventions and procedural techniques, the persistent burden of CVD underscores the critical need to identify and address modifiable risk factors, particularly those amenable to population-level interventions(Reference Lloyd-Jones, Allen and Anderson3). While traditional risk factors such as hypertension, diabetes and dyslipidaemia have been well-characterised and aggressively targeted in clinical practice(Reference Arnett, Blumenthal and Albert4), emerging evidence suggests that nutritional deficiencies may represent an underappreciated yet substantial contributor to cardiovascular pathogenesis(Reference Mozaffarian5,Reference Afshin, Sur and Fay6) .
Micronutrients – including vitamins, minerals and trace elements – play fundamental roles in cardiovascular physiology, participating in diverse processes ranging from endothelial function and oxidative stress modulation to lipid metabolism and myocardial energy production(Reference Tangvoraphonkchai and Davenport7). Vitamin D regulates vascular smooth muscle proliferation and inflammation through genomic and non-genomic pathways(Reference Zittermann, Pilz and Hoffmann8,Reference Autier, Mullie and Macacu9) , with observational studies reporting associations between deficiency and increased CVD risk(Reference Pilz, Verheyen and Grübler10), though randomised trials have yielded inconsistent results(Reference Barbarawi, Kheiri and Zayed11). Fe serves as an essential cofactor for Hb synthesis and cellular respiration, with both deficiency and excess implicated in adverse cardiovascular outcomes. B vitamins, particularly folate and vitamin B12, mediate homocysteine metabolism, while Ca maintains vascular tone and myocardial contractility. Despite their biological plausibility as cardiovascular determinants, the epidemiological evidence linking specific micronutrient deficiencies to CVD risk remains inconsistent and fragmented.
However, existing research has important limitations: most studies examine individual micronutrients in isolation, ignoring cumulative effects of multiple deficiencies (‘hidden hunger’); large-scale biomarker-based studies are scarce and cross-population validation is insufficient. Although folic acid fortification has eliminated deficiency, supplementation trials show only modest stroke protection without coronary benefit. Despite affecting 1·2 billion people globally, Fe deficiency research remains limited to heart failure populations, and nonlinear dose–response relationships have not been adequately characterised using flexible modeling approaches. Therefore, this study aimed to characterise the independent and cumulative associations between micronutrient deficiencies and CVD using binational data from NHANES and CHARLS, examining both biomarker-based and dietary pattern approaches to inform evidence-based recommendations for nutritional screening and risk stratification.
Experimental methods
Study design and study populations
This cross-national comparative study integrated cross-sectional biomarker data from the US National Health and Nutrition Examination Survey (NHANES, 2007–2018) with prospective dietary data from the China Health and Retirement Longitudinal Study (CHARLS, 2011–2018) to examine associations between nutritional status and CVD risk.
NHANES uses complex multistage probability sampling to recruit non-institutionalised US civilians. We included adults aged 40 years or older with complete CVD status, serum micronutrient measurements and covariate data. From 60 142 examined participants, we excluded those younger than 40 years (n 27 348), with missing CVD status (n 1523), lacking micronutrient data (n 24 986) or incomplete covariates (n 437), yielding 3848 participants (304 with CVD, 3544 without).
CHARLS uses multistage probability-proportional-to-size sampling across twenty-eight Chinese provinces to recruit adults aged 45 years or older. Baseline enrollment occurred in 2011 with biennial follow-ups through 2018 (retention rate exceeding 80 %). From 17 708 baseline participants, we excluded those younger than 45 years (n 1245), with missing dietary data (n 2834), or lost to follow-up (n 661), yielding 11 391 participants. For incident CVD analysis, we further excluded 1577 with baseline CVD, leaving 9814 at risk.
Exposure assessment
Serum micronutrient biomarkers
Fasting venous blood samples in NHANES were analysed using standard laboratory methods. Serum 25-hydroxyvitamin D was measured by LC-tandem mass spectrometry, folate by microbiologic assay, vitamin B12 by electrochemiluminescence immunoassay, Ca by potentiometry and Fe by colorimetric assay. Deficiencies were defined as vitamin D below 50 nmol/l, folate below 10 nmol/l, vitamin B12 below 148 pmol/l, Ca below 2·2 mmol/l and Fe below 10·7 μmol/l based on clinical guidelines. A cumulative micronutrient deficiency score (range 0–5) was created by summing the number of deficiencies present for each participant. All participants were included in this score; for nutrients where deficiency was not present (e.g. folate in the vast majority due to fortification), a score of ‘0’ was contributed for that nutrient.
Dietary patterns
Dietary intake in CHARLS was assessed using a validated FFQ covering seven food groups (milk, soya products, fruits and vegetables, meat, fish, eggs and nuts). Participants reported consumption frequency over the past year. Regular consumption was defined as once per week or more v. less frequent or no consumption.
Outcome assessment
Prevalent CVD in NHANES was defined as self-reported physician diagnosis of coronary heart disease, myocardial infarction, angina or stroke. Incident CVD in CHARLS was ascertained at follow-up surveys (2013, 2015 and 2018) as new physician diagnosis of cardiovascular conditions. Time to event was calculated from baseline to the midpoint between last CVD-free visit and first event, with censoring at last follow-up or 31 December 2018.
Covariate assessment
Demographic variables included age, sex, race or ethnicity (NHANES only), education and income. Anthropometric measures included BMI and waist circumference (NHANES only). Lifestyle factors included smoking, alcohol consumption and physical activity. Comorbidities included hypertension, diabetes mellitus and hyperlipidaemia (NHANES only), defined using standard clinical criteria combining self-report, measurements and medication use.
Statistical analysis
All analyses used R version 4.2.0 with two-sided P less than 0·05 considered significant. NHANES analyses incorporated survey weights, strata and primary sampling units using the survey package to generate nationally representative estimates. Baseline characteristics were compared using design-adjusted Rao-Scott χ 2 tests for categorical variables and weighted linear regression for continuous variables.
We examined associations between micronutrient deficiencies and prevalent CVD using logistic regression with three adjustment levels: model 1 (unadjusted), model 2 (demographics) and model 3 (demographics plus anthropometrics, lifestyle and comorbidities). A comprehensive model including all risk factors simultaneously was constructed. Results are presented as OR with 95 % CI.
Nonlinear relationships between continuous biomarker levels and CVD risk were assessed using restricted cubic splines with four knots at the 5th, 35th, 65th and 95th percentiles. Reference values were set at clinically relevant levels (vitamin D: 75 nmol/l, Fe: 75 μg/dl). Nonlinearity was tested using likelihood ratio tests.
Kaplan–Meier curves illustrated CVD-free survival by dietary consumption patterns, with log-rank tests comparing groups. Cox proportional hazards models estimated hazard ratios for associations between dietary factors and incident CVD, with univariate and multivariable adjustment. The proportional hazards assumption was verified using Schoenfeld residuals.
Random-effects meta-analysis combined estimates from NHANES and CHARLS using the DerSimonian–Laird method. Effect sizes were log-transformed to a common scale. Heterogeneity was assessed using I-squared statistics. Sensitivity analyses varied baseline event rate assumptions (15 %, 25 % and 35 %) to test the robustness of the meta-analytic findings. These assumed rates were selected to cover a plausible range observed in general and at-risk populations: 15 % represents a lower-bound conservative estimate, 25 % approximates the age-standardised pooled prevalence across our cohorts and 35 % represents an upper-bound high-risk scenario.
Result
Study population and baseline characteristics in National Health and Nutrition Examination Survey
Table 1 presents the baseline characteristics of 3848 participants (304 with CVD, 3544 without CVD) from NHANES 2007–2018. Compared with non-CVD participants, those with CVD were significantly older (mean age difference: 18·7 years, P < 0·001) and exhibited lower socio-economic status, as indicated by reduced educational attainment and income-to-poverty ratios (Table 1). Anthropometric measures revealed higher adiposity indices in the CVD group, with elevated BMI (29·73 v. 28·82 kg/m2, P = 0·023) and waist circumference (103·80 v. 98·32 cm, P < 0·001).
Baseline characteristics of study participants according to cardiovascular disease status in NHANES 2007–2018
Table 1. Long description
The table presents baseline characteristics of 3848 participants, divided into two groups: 304 with cardiovascular disease (CVD) and 3544 without CVD. The table includes data on age, BMI, waist circumference, Vitamin D levels, folate, B12, calcium, iron, income ratio, sex, race, education, hypertension, diabetes, current smoking, current alcohol use, and physical activity. Each characteristic is presented with mean and standard deviation for continuous variables and counts and percentages for categorical variables. Notable trends include higher mean age, BMI, waist circumference, and lower socio-economic status in the CVD group compared to the non-CVD group.
NHANES, National Health and Nutrition Examination Survey; VitD, serum 25-hydroxyvitamin D; B12, vitamin B12.
Notably, CVD participants demonstrated paradoxically higher levels of serum 25(OH)D (71·36 v. 63·96 nmol/l, P < 0·001) and folate (604·42 v. 481·58 nmol/l, P < 0·001), while serum Fe was significantly lower (79·88 v. 85·56 μg/dl, P = 0·007). No significant between-group differences were observed for vitamin B12 or Ca. Cardiometabolic comorbidities were markedly more prevalent in the CVD group, with 78·3 % having hypertension and 33·4 % having diabetes, compared with 31·9 % and 10·6 % in the non-CVD group, respectively (both P < 0·001). The CVD group also had higher smoking prevalence (60·2 % v. 42·1 %, P < 0·001) and lower physical activity participation (11·6 % v. 17·9 %, P = 0·007) (Table 1).
Association between individual nutritional deficiencies and CVD
Table 2 presents the associations between individual nutritional deficiencies and prevalent CVD across three progressively adjusted models. Among the five nutritional biomarkers examined, Fe deficiency emerged as the only consistent predictor of CVD. In crude analysis (model 1), Fe deficiency was significantly associated with increased CVD odds (OR 1·45, 95 % CI: 1·11, 1·87, P = 0·005). This association persisted and strengthened after adjustment for demographic factors (model 2: OR 1·53, 95 % CI: 1·14, 2·03, P = 0·004) and remained significant in the fully adjusted model accounting for anthropometric measures and cardiometabolic risk factors (model 3: OR 1·49, 95 % CI: 1·09, 2·01, P = 0·012).
Associations between individual nutritional deficiencies and cardiovascular disease in NHANES 2007–2018
Table 2. Long description
The table presents data on the associations between various nutritional deficiencies and cardiovascular disease (CVD) across three models: Model 1 (Crude), Model 2 (Demographics), and Model 3 (Fully adjusted). The table includes five nutritional biomarkers: Vitamin D, Folate, Vitamin B12, Calcium, and Iron. Each biomarker's odds ratio (OR) and 95% confidence interval (CI) are provided for each model, along with the P-value indicating statistical significance. Notably, Iron deficiency is consistently associated with increased CVD odds across all models. In Model 1, the OR for Iron deficiency is 1.45 with a 95% CI of 1.11 to 1.87 and a P-value of 0.005. In Model 2, the OR is 1.53 with a 95% CI of 1.14 to 2.03 and a P-value of 0.004. In Model 3, the OR is 1.49 with a 95% CI of 1.09 to 2.01 and a P-value of 0.012. Other nutritional deficiencies do not show consistent significant associations across all models.
NHANES, National Health and Nutrition Examination Survey.
Model 1: Unadjusted. Model 2: Adjusted for age, sex, race/ethnicity, education and income-to-poverty ratio. Model 3: Additionally adjusted for BMI, waist circumference, hypertension, diabetes mellitus, current smoking, alcohol use and physical activity. Micronutrient deficiencies were defined as: serum 25-hydroxyvitamin D < 50 nmol/l, folate < 10 nmol/l, vitamin B12 < 148 pmol/l, Ca < 2·2 mmol/l and Fe < 10·7 μmol/l (or < 60 μg/dl as analysed in this table).
Vitamin D deficiency showed an initial protective association in the unadjusted model (OR 0·76, 95 % CI: 0·58, 0·98, P = 0·036), which was completely attenuated after controlling for demographic characteristics (model 2: OR 1·16, P = 0·342), indicating confounding by age and socio-economic status. Vitamin B12 deficiency demonstrated a non-significant trend toward increased CVD risk across all models, with wide confidence intervals suggesting limited statistical power. Similarly, Ca deficiency showed elevated but non-significant odds ratios with notably wide confidence intervals (model 3: OR 2·58, 95 % CI: 0·36, 12·10), likely reflecting the low prevalence of Ca deficiency in the study population. Folate deficiency could not be analysed due to insufficient cases (prevalence < 1 %).
To evaluate the relative importance of nutritional deficiencies among established cardiovascular risk factors, we constructed a comprehensive multivariable model simultaneously including all demographics, lifestyle factors, comorbidities and nutritional biomarkers (Figure 1). Hypertension (OR 3·26, 95 % CI: 2·39, 4·51) and diabetes (OR 1·79, 95 % CI: 1·32, 2·42) were the strongest independent predictors of CVD (both P < 0·001). Age demonstrated a dose-dependent association (OR 1·06 per year, 95 % CI: 1·05, 1·07, P < 0·001), while higher socio-economic status was protective (income-to-poverty ratio OR 0·85, 95 % CI: 0·76, 0·94, P < 0·01). Among modifiable risk factors, current smoking (OR 1·47, 95 % CI: 1·10, 1·97, P < 0·05) and Fe deficiency (OR 1·49, 95 % CI: 1·09, 2·02, P < 0·05) remained independently associated with CVD after full adjustment. Other nutritional deficiencies, lifestyle factors (BMI, alcohol and physical activity) and demographic characteristics showed no significant independent associations.
Forest plot showing multivariable-adjusted associations between risk factors and prevalent cardiovascular disease. OR and 95 % CI were derived from a fully adjusted logistic regression model including all variables shown. The vertical dashed line represents an OR of 1·0 (no association). Statistical significance: *P < 0·05, **P < 0·01, ***P < 0·001.
Figure 1. Long description
A horizontal dot plot displays the odds ratios (OR) and 95 percentage confidence intervals (CI) for various risk factors associated with prevalent cardiovascular disease. The x-axis represents the odds ratio ranging from 0.3 to 10.0, while the y-axis lists different risk factors categorized under Clinical, Demographics, Lifestyle, and Nutrients. Each dot represents the OR for a specific risk factor, with horizontal lines indicating the 95 percentage CI. Dots are color-coded, with blue indicating significant associations and grey indicating non-significant associations. The vertical dashed line at OR 1.0 signifies no association. Notable significant associations include hypertension with an OR of 3.26, diabetes with an OR of 1.79, and age with an OR of 1.06. Lifestyle factors such as current smoking show an OR of 1.47, while nutritional deficiencies like iron deficiency have an OR of 1.49. The plot highlights the varying degrees of association between different risk factors and cardiovascular disease, emphasizing the importance of addressing both clinical and lifestyle factors in cardiovascular health.
Individual micronutrient deficiencies and CVD risk
Figure 2 illustrates the weighted prevalence of five micronutrient deficiencies among 5432 US adults aged ≥ 40 years. Vitamin D deficiency was the most prevalent, affecting 33·1 % (95 % CI, 30·8 %, 35·4 %) of the population, equivalent to approximately one in three adults. Fe deficiency was the second most common, with a prevalence of 23·4 % (95 % CI, 21·2 %, 25·6 %), affecting nearly one in four individuals. The remaining micronutrient deficiencies were substantially less common. Vitamin B12 deficiency was observed in 1·7 % (95 % CI, 1·1 %, 2·3 %) of participants. Ca deficiency was rare, affecting only 0·3 % (95 % CI, 0·1 %, 0·5 %) of the population. Folate deficiency was virtually absent (0·0 %), with insufficient cases to generate a reliable prevalence estimate. This near-elimination of folate deficiency reflects the success of the mandatory folic acid fortification program implemented in the USA in 1998, which requires enrichment of cereal grain products.
Prevalence of micronutrient deficiencies among USA adults aged ≥ 40 years in NHANES (2011–2012). NHANES, National Health and Nutrition Examination Survey.
Figure 2. Long description
The bar graph compares the prevalence of deficiencies in various micronutrients among USA adults aged 40 years and older, based on data from the National Health and Nutrition Examination Survey (NHANES) conducted in 2011-2012. The x-axis lists the nutrients: Vitamin D, Iron, Vitamin B12, Calcium, and Folate. The y-axis represents the deficiency prevalence in percentage, ranging from 0 to 40 percent. There are five vertical bars, each representing a different nutrient. Vitamin D has the highest deficiency prevalence at 33.1 percent, followed by Iron at 23.4 percent. Vitamin B12 shows a deficiency prevalence of 1.7 percent, while Calcium has a very low prevalence of 0.3 percent. Folate shows no deficiency prevalence at 0.0 percent. The bars are colored in blue. All values are approximated.
Individual micronutrient deficiencies showed variable associations with prevalent CVD, though none reached statistical significance (Figure 3). Point estimates suggested elevated odds for vitamin B12 deficiency (OR 1·44, 95 % CI: 0·61, 3·05), Ca deficiency (OR 2·58, 0·36, 12·10) and Fe deficiency (OR 1·49, 1·09, 2·01). Vitamin D deficiency showed a non-significant protective trend (OR 1·14, 0·82, 1·56). Wide confidence intervals reflected limited precision, particularly for Ca deficiency.
Association between individual micronutrient deficiencies and prevalent cardiovascular disease in NHANES. NHANES, National Health and Nutrition Examination Survey.
Figure 3. Long description
A line graph illustrates the association between individual micronutrient deficiencies and prevalent cardiovascular disease in the National Health and Nutrition Examination Survey (NHANES). The x-axis represents the odds ratio on a logarithmic scale ranging from 0.3 to 10.0, while the y-axis lists five micronutrient deficiencies: Vitamin D, Folate, Vitamin B 12, Calcium, and Iron. Each deficiency is represented by a data point with an associated confidence interval. Vitamin D deficiency has an odds ratio of 1.14 with a confidence interval of 0.82 to 1.56. Folate deficiency does not show a significant data point. Vitamin B 12 deficiency has an odds ratio of 1.44 with a confidence interval of 0.61 to 3.06. Calcium deficiency has an odds ratio of 2.58 with a confidence interval of 0.36 to 12.10. Iron deficiency has an odds ratio of 1.49 with a confidence interval of 1.09 to 2.01. All values are approximated.
Non-linear relationships between continuous micronutrient levels and CVD risk
Restricted cubic spline models demonstrated nonlinear associations between micronutrient levels and CVD risk (Figure 4). Serum vitamin D showed an inverse relationship (P-nonlinearity < 0·001), with risk decreasing sharply below 50 nmol/l and plateauing thereafter (Figure 4(a)). Serum Fe exhibited a U-shaped curve (P-nonlinearity = 0·003), with nadir risk at 50–100 μg/dl; both deficiency (< 50 μg/dl) and excess (> 150 μg/dl) were associated with elevated OR (Figure 4(b)). All models were adjusted for age, sex, race/ethnicity, BMI, smoking, alcohol use, physical activity, socio-economic status and comorbidities.
Dose–response relationships between serum micronutrient levels and cardiovascular disease risk.
Figure 4. Long description
The image contains two line graphs labeled A and B. Graph A shows the relationship between Vitamin D levels and cardiovascular disease risk, while Graph B shows the relationship between Iron levels and cardiovascular disease risk. Both graphs have the odds ratio on the y-axis and the micronutrient levels on the x-axis. The shaded areas represent confidence intervals. The graphs indicate that as Vitamin D levels increase, the odds ratio for cardiovascular disease risk decreases slightly. Similarly, as Iron levels increase, the odds ratio for cardiovascular disease risk also decreases slightly. The data is adjusted for various factors including age, sex, race, education, income ratio, body mass index, hypertension, diabetes, smoking, alcohol consumption, and physical activity.
To examine the cumulative effect of multiple micronutrient deficiencies, we constructed a micronutrient deficiency score by summing the number of deficiencies present for each participant (online Supplementary Figure 1). Among the study population, 45·2 % had no deficiencies, 38·1 % had 1 deficiency and 16·7 % had ≥ 2 deficiencies. Compared with participants with no micronutrient deficiencies, those with 1 deficiency had a 17 % non-significant increase in CVD odds (OR, 1·17; 95 % CI, 0·86, 1·58; P = 0·31). Participants with ≥ 2 deficiencies demonstrated a 91 % increased odds of CVD (OR, 1·91; 95 % CI, 1·20, 2·99; P = 0·006), and this association was statistically significant. The test for linear trend across micronutrient deficiency score categories was significant (P for trend = 0·008), indicating a dose–response relationship between the number of micronutrient deficiencies and CVD risk.
Baseline characteristics in China Health and Retirement Longitudinal Study cohort
Among 11 391 CHARLS participants, 1577 (13·8 %) had prevalent CVD at baseline. Compared with participants without CVD, those with CVD were older (mean age 60·0 (sd 9·2) v. 57·4 (sd 9·1) years, P < 0·001) and more likely to be female (60·2 % v. 52·9 %, P < 0·001). Participants with CVD had significantly lower consumption of all assessed food groups: milk (13·7 % v. 21·9 %), soya products (35·6 % v. 41·4 %), fruits and vegetables (62·3 % v. 77·5 %), fish (18·3 % v. 32·4 %) and nuts (18·3 % v. 26·5 %) (all P < 0·001). CVD prevalence was also higher among non-smokers (64·6 % v. 61·1 %, P = 0·009) and non-drinkers (66·7 % v. 60·5 %, P < 0·001). No significant differences were observed in educational attainment (P = 0·151) or egg consumption (P = 0·091) (Table 3).
Sociodemographic and dietary characteristics of CHARLS participants stratified by CVD status at baseline
Table 3. Long description
The table presents data on sociodemographic and dietary characteristics of CHARLS participants, stratified by CVD status at baseline. It includes 1577 participants with CVD and 9814 without CVD. The table has 12 rows and 7 columns, with headers for Level, CVD, NoCVD, and P-value. Key variables include age, sex, education, smoking, alcohol consumption, and dietary habits such as milk, soya, fruit and vegetable, meat, fish, egg, and nut consumption. Notable trends include older age, higher percentage of females, lower consumption of milk, soya products, fruits and vegetables, fish, and nuts among participants with CVD compared to those without CVD. Significant differences are observed in smoking and alcohol consumption, with higher prevalence of non-smokers and non-drinkers among those with CVD. No significant differences are noted in educational attainment or egg consumption.
CHARLS, China Health and Retirement Longitudinal Study.
Associations between dietary factors and incident CVD
In univariate analysis, several dietary factors showed significant protective associations: fruits and vegetables (HR 0·78, 95 % CI: 0·69, 0·88, P < 0·001), nuts (HR 0·80, 0·70, 0·92, P = 0·004), fish (HR 0·83, 0·72, 0·95, P = 0·009), milk (HR 0·85, 0·75, 0·96, P = 0·01) and soya products (HR 0·90, 0·81, 0·99, P = 0·035). Conversely, meat consumption was associated with increased risk (HR 1·12, 1·01, 1·25, P = 0·041), while egg consumption showed no association (HR 0·97, 0·86, 1·10, P = 0·62). After adjustment for demographic, socio-economic and lifestyle factors, most dietary associations persisted but were attenuated. Fruits and vegetables remained strongly protective (adjusted HR (aHR) 0·81, 0·70, 0·93, P = 0·006), as did nuts (aHR 0·82, 0·71, 0·95, P = 0·008) and fish (aHR 0·85, 0·74, 0·98, P = 0·027). Milk consumption showed borderline significance (aHR 0·88, 0·77, 1·00, P = 0·048). The associations for soya products (aHR 0·93, 0·84, 1·04, P = 0·12) and meat (aHR 1·10, 0·98, 1·23, P = 0·09) were no longer statistically significant after adjustment. Among non-dietary factors, age (aHR 1·03 per year, 1·02, 1·04, P < 0·001), male sex (aHR 1·20, 1·05, 1·37, P = 0·004), low education (aHR 1·15, 1·02, 1·30, P = 0·012), low income (aHR 1·18, 1·05, 1·34, P = 0·008) and smoking (aHR 1·25, 1·10, 1·43, P < 0·001) were independent risk factors. Physical activity was protective (aHR 0·85, 0·74, 0·98, P = 0·021), while alcohol consumption showed no association (aHR 1·03, 0·90, 1·18, P = 0·52).
Kaplan–Meier survival analysis of dietary intake and CVD-free survival
Kaplan–Meier survival analyses were performed to compare unadjusted cardiovascular disease-free survival by consumption frequency for seven dietary factors (Figure 5). The unadjusted survival curves showed significant divergence for soya products (log-rank P < 0·0001; Figure 5(b)) and meat consumption (log-rank P < 0·0001; Figure 5(d)). In contrast, differences for milk (P = 0·27; Figure 5(a)), fruits and vegetables (P = 0·068; Figure 5(c)), fish (P = 0·16; Figure 5(e)), eggs (P = 0·25; Figure 5(f)) and nuts (P = 0·14; Figure 5(g)) did not reach statistical significance. Notably, the absence of statistically significant unadjusted survival differences for factors such as nuts and fish contrasts with the protective associations observed in multivariable-adjusted Cox models (aHR 0·82 and 0·85, respectively), highlighting the influence of confounding factors (e.g. age and socio-economic status) that were controlled for in the latter analyses. The widening CI at later time points across all curves reflect decreasing numbers at risk due to censoring. Overall, these unadjusted survival analyses are consistent with the direction of associations found in adjusted models, with soya products showing a clear protective pattern and meat consumption an adverse pattern.
Kaplan–Meier survival curves for cardiovascular disease-free survival by dietary intake patterns in CHARLS (2011–2018). CHARLS, China Health and Retirement Longitudinal Study.
Figure 5. Long description
Seven Kaplan-Meier survival curves compare cardiovascular disease-free survival by dietary intake patterns over eight years. Each graph shows survival probability on the y-axis and time in years on the x-axis. The graphs are labeled as Milk, Soya, Fruit and Vegetable, Meat, Fish, Egg, and Nut. Each graph includes two lines representing 'No' and 'Yes' for the respective dietary intake. The p-values indicate the statistical significance of the differences between the two groups. The Milk graph shows a p-value of 0.27, the Soya graph shows a p-value of less than 0.0001, the Fruit and Vegetable graph shows a p-value of 0.068, the Meat graph shows a p-value of less than 0.0001, the Fish graph shows a p-value of 0.16, the Egg graph shows a p-value of 0.25, and the Nut graph shows a p-value of 0.14. The survival probabilities decrease over time in all graphs, with varying degrees of separation between the 'No' and 'Yes' lines.
Cross-cohort synthesis of nutritional biomarkers and dietary patterns
In NHANES, several serum biomarkers showed significant associations with prevalent CVD (Figure 6(a)). Protective associations were observed for vitamin D (OR 0·82, 95 % CI: 0·77, 0·87) and HDL-cholesterol (OR 0·78, 0·71, 0·86). Adverse associations were identified for C-reactive protein (OR 1·32, 1·18, 1·48), glucose (OR 1·28, 1·15, 1·43), TAG (OR 1·18, 1·09, 1·28), Fe (OR 1·15, 1·06, 1·25) and vitamin B12 (OR 1·12, 1·01, 1·24). Folate, Ca and albumin showed non-significant protective trends.
Cross-cohort synthesis: biomarkers, dietary proxies, meta-analysis, and sensitivity analysis. (a) Associations between serum biomarkers and CVD in NHANES, (b) meta-analysis of vitamin D (NHANES) and fruits/vegetables (CHARLS), (c) associations between dietary proxies and incident CVD in CHARLS, (d) sensitivity analysis varying assumed event rates. CHARLS, China Health and Retirement Longitudinal Study; NHANES, National Health and Nutrition Examination Survey.
Figure 6. Long description
The image contains four distinct graphs analyzing the relationship between biomarkers, dietary factors, and cardiovascular disease (CVD). The first graph (A) is a scatter plot showing the odds ratio of various biomarkers such as Vitamin D, triglycerides, iron, HDL, glucose, folate, CRP, calcium, B12, and albumin against CVD in the NHANES study. The second graph (B) is a forest plot presenting the meta-analysis of Vitamin D from NHANES and fruits/vegetables from CHARLS, indicating the effect on CVD on a log scale. The third graph (C) is another scatter plot illustrating the hazard ratio of dietary proxies like soy, nuts, milk, meat, fruits/vegetables, fish, and eggs against incident CVD in the CHARLS study. The fourth graph (D) is a sensitivity analysis line graph showing the hazard ratio of fruits/vegetables with varying assumed event rates. Each graph uses different axes and color schemes to represent the data accurately. All values are approximated.
Meta-analysis combining vitamin D effects from NHANES with fruits/vegetables from CHARLS yielded a null pooled estimate (−0·01, 95 % CI: −0·06, 0·05) with moderate heterogeneity (Figure 6(b)), reflecting challenges in harmonising biomarker and dietary proxy measures across cohorts.
In CHARLS, multivariable-adjusted analysis revealed protective associations for soya products (HR 0·82, 95 % CI: 0·74, 0·91), nuts (HR 0·82, 0·72, 0·93), fish (HR 0·83, 0·74, 0·93) and milk (HR 0·88, 0·78, 0·99) (Figure 6(c)). Meat consumption showed an association with increased risk that was attenuated after multivariable adjustment (aHR 1·10, 0·98, 1·23; see Table 4). Fruits/vegetables and eggs demonstrated null associations after adjustment.
Cox proportional hazards regression analysis of factors associated with incident cardiovascular disease in CHARLS
Table 4. Long description
The table presents a Cox proportional hazards regression analysis of factors associated with incident cardiovascular disease in CHARLS. It includes demographic, socio-economic, and lifestyle factors, along with dietary factors. The table has 12 rows and 7 columns, with column headers for Univariate analysis HR, Univariate analysis 95 percentage CI, Univariate analysis P, Multivariable analysis HR, Multivariable analysis 95 percentage CI, and P. Row labels include Age per year, Male sex, Low education, Low income, Current smoking, Alcohol consumption, Regular physical activity, Milk consumption, Soya products, Fruits and vegetables, Meat, and Fish. Notable trends include significant associations for age, male sex, low education, low income, current smoking, regular physical activity, milk consumption, soya products, fruits and vegetables, and fish with incident cardiovascular disease.
CHARLS, China Health and Retirement Longitudinal Study.
Sensitivity analysis using a range of plausible baseline CVD event rates (15 %, 25 %, 35 %) demonstrated moderate robustness of the pooled findings (Figure 6(d)). Pooled hazard ratios ranged from 0·80 (95 % CI: 0·72, 0·88) at 15 % event rate to 1·05 (0·96, 1·14) at 35 % event rate, with overlapping confidence intervals indicating stability of core findings despite assumption perturbations.
Discussion
In this binational study integrating cross-sectional biomarker data from NHANES with prospective dietary data from CHARLS, we found that Fe deficiency was the only micronutrient deficiency independently associated with prevalent CVD after full adjustment (OR 1·49, 95 % CI: 1·09, 2·01). Cumulative micronutrient deficiencies demonstrated dose–response relationships, with participants having two or more deficiencies showing 91 % increased odds of CVD. Restricted cubic spline analysis revealed a U-shaped relationship for Fe and a threshold effect for vitamin D below 50 nmol/l. In the Chinese cohort, plant-based dietary patterns – particularly soya products, nuts and fish – were protective, while meat consumption was associated with a non-significant trend towards increased risk in the fully adjusted model (aHR = 1·10). Cross-cohort meta-analysis yielded null pooled estimates, highlighting challenges in harmonising biomarker v. dietary proxy measures.
Our finding that Fe deficiency independently predicts CVD risk is consistent with recent evidence from heart failure populations. Randomised controlled trials have demonstrated that intravenous Fe supplementation improves functional capacity in heart failure patients with Fe deficiency independent of anaemia status(Reference Anker, Comin Colet and Filippatos12,Reference Ponikowski, van Veldhuisen and Comin-Colet13) , supporting direct cardiac effects beyond oxygen-carrying capacity. However, the IRONOUT-HF trial showed that oral Fe supplementation did not improve outcomes(Reference Lewis, Malhotra and Hernandez14), indicating that bioavailability and delivery route are critical. The U-shaped dose–response relationship we observed reconciles previously inconsistent findings, with both deficiency and excess associated with elevated CVD risk through distinct mechanisms: hypoxia-induced cardiac remodelling at low levels(Reference Anker, Comin Colet and Filippatos12) and Fenton-mediated oxidative damage at high levels(Reference Sullivan15).
In contrast, vitamin D deficiency showed no independent association after multivariable adjustment despite a crude protective association. This aligns with recent null findings from large randomised controlled trials. The VITAL trial found no reduction in major cardiovascular events with vitamin D supplementation(Reference Manson, Cook and Lee16), and similar null results were reported by the D-Health(Reference Neale, Baxter and Romero17) and DO-HEALTH trials(Reference Bischoff-Ferrari, Vellas and Rizzoli18). Importantly, CVD participants in our study paradoxically had higher vitamin D levels than controls (71·36 v. 63·96 nmol/l), indicating post-diagnosis supplementation or confounding by indication(Reference Kyriacou and Lewis19). The complete attenuation of the crude association after demographic adjustment strongly suggests that observational vitamin D-CVD associations reflect confounding by lifestyle and socio-economic factors rather than causality.
For dietary patterns, our findings from CHARLS corroborate evidence from Western populations. The protective effects of plant-based foods, particularly soya products (18 % risk reduction), align with the Shanghai Women’s Health Study showing substantially lower coronary disease risk with high soya intake(Reference Zhang, Shu and Gao20). Mechanisms include isoflavone-mediated endothelial function improvement and LDL-cholesterol reduction through soya protein(Reference Anderson and Bush21). Although the association observed in this study did not reach statistical significance (adjusted hazard ratio 1·10, P = 0·09), the direction of the association is consistent with large meta-analyses reporting approximately 40 % higher coronary disease risk per daily serving of processed meat(Reference Micha, Wallace and Mozaffarian22). Notably, our finding extends to total meat consumption, predominantly fresh pork in the Chinese diet, suggesting mechanisms beyond Na and nitrite preservatives – potentially heme Fe oxidative stress(Reference Kell and Pretorius23), heterocyclic amines from cooking(Reference Cross and Sinha24) and gut microbiome-mediated TMAO production(Reference Koeth, Wang and Levison25,Reference Tang, Wang and Levison26) . The null finding for fruits and vegetables after adjustment contrasts with the INTERHEART study showing 30 % risk reduction(Reference Yusuf, Hawken and Ounpuu27) and meta-analyses reporting 28 % lower risk at high intake levels(Reference Aune, Giovannucci and Boffetta28). This discrepancy likely reflects measurement limitations in our dietary assessment, which used crude frequency thresholds without portion quantification and high baseline consumption (77·5 %) limiting contrast.
Several plausible mechanisms link the observed associations to cardiovascular pathophysiology. Fe deficiency impairs oxygen-carrying capacity and mitochondrial function, leading to compensatory cardiac remodelling, increased cardiac output and eventual heart failure in severe cases(Reference Anker, Comin Colet and Filippatos12). Conversely, Fe excess catalyses Fenton reactions generating reactive oxygen species that promote lipid peroxidation, endothelial dysfunction and atherosclerotic plaque instability(Reference Sullivan15), explaining the U-shaped risk curve.
Plant-based dietary patterns exert cardiovascular protection through multiple pathways. Soy isoflavones act as selective estrogen receptor modulators, improving endothelial nitric oxide bioavailability and arterial compliance. n-3 fatty acids from fish reduce TAG, improve endothelial function, decrease platelet aggregation, and exert anti-inflammatory effects, as validated by the REDUCE-IT trial showing 25 % reduction in cardiovascular events with high-dose EPA(Reference Bhatt, Steg and Miller29). Nuts provide cardioprotective MUFA and PUFA, plant sterols, L-arginine and magnesium(Reference Ros30).
The adverse effects of meat consumption involve multiple mechanisms. Heme Fe undergoes Fenton chemistry more readily than non-heme plant Fe(Reference Kell and Pretorius23). High-temperature cooking produces heterocyclic amines and polycyclic aromatic hydrocarbons with pro-inflammatory properties(Reference Cross and Sinha24). Gut microbial metabolism of L-carnitine and choline abundant in meat generates TMAO, which enhances platelet reactivity and accelerates atherosclerosis(Reference Koeth, Wang and Levison25,Reference Tang, Wang and Levison26) . A randomised crossover trial confirmed that red meat consumption increases plasma TMAO levels compared with white meat or plant protein(Reference Wang, Bergeron and Levison31).
The synergistic effects of cumulative micronutrient deficiencies likely reflect interconnected metabolic pathways. Vitamin B12 and folate deficiencies both elevate homocysteine(32), while concurrent Fe deficiency impairs erythropoiesis despite adequate B vitamin status. Vitamin D and Ca interact in vascular smooth muscle regulation and calcification(Reference Reid and Bolland33). Multiple marginal deficiencies may exceed adaptive metabolic capacity, leading to cumulative dysfunction.
Our findings have several actionable implications. First, screening for Fe deficiency in cardiovascular risk assessment warrants consideration, particularly given the independent association after extensive adjustment. Current guidelines recommend screening only for symptomatic anaemia, but subclinical Fe deficiency may independently contribute to CVD risk. However, the U-shaped relationship mandates caution against indiscriminate supplementation. Individualised assessment using ferritin and transferrin saturation to maintain Fe stores in the optimal range (corresponding to serum Fe 50–100 micrograms per decilitre) is essential.
Second, dietary counseling for CVD prevention should emphasise plant-based patterns while reducing meat consumption. These recommendations align with the 2021 American Heart Association dietary guidance emphasising fruits, vegetables, whole grains, legumes, nuts and fish and the EAT-Lancet Commission recommendations for sustainable healthy diets. Practical implementation strategies include institutional food service modifications, economic policies such as fruit and vegetable subsidies and addressing barriers to healthy food access.
Third, the dose–response relationship for cumulative deficiencies highlights the importance of overall dietary quality rather than single-nutrient supplementation. Dietary pattern approaches such as the Mediterranean diet or DASH diet that naturally provide adequate micronutrients may be more effective. The Lyon Diet Heart Study demonstrated 70 % reduction in CVD recurrence with Mediterranean diet, while the DASH-Sodium trial showed substantial blood pressure reductions.
This study has several strengths. The integration of NHANES biomarker data with CHARLS prospective dietary data provided complementary evidence addressing limitations inherent to each design. Large nationally representative samples enhanced generalisability and statistical power. Comprehensive covariate adjustment using progressively adjusted models minimised residual confounding. Advanced statistical methods including restricted cubic splines for nonlinear relationships, survey-weighted analysis accounting for complex sampling and random-effects meta-analysis increased methodologic rigor. The cumulative deficiency analysis provided novel insights into synergistic nutritional effects.
Several limitations warrant consideration. First, the cross-sectional NHANES design precludes causal inference and is susceptible to reverse causation. The higher vitamin D levels in CVD participants exemplify this limitation, likely reflecting post-diagnosis supplementation. Prospective biomarker studies are needed to establish temporal relationships. Second, CHARLS lacked serum micronutrient measurements, requiring dietary proxies that inadequately capture bioavailable nutrient status. The FFQ assessed only consumption frequency without portion size, which introduces non-differential misclassification. Methodological studies suggest that such instruments may attenuate true diet–disease associations towards the null, with correlation coefficients for nutrient intake often ranging between 0·3 and 0·6 compared with detailed dietary records. Although this limitation likely resulted in conservative bias (i.e. underestimation of the protective associations), statistically significant findings for plant-based foods persisted after multivariable adjustment, strengthening the credibility of our conclusions. Additionally, the FFQ did not distinguish between processed (e.g. sausages and cured meats) and unprocessed meat consumption. Given the potentially distinct cardiovascular risk profiles mediated by components such as Na and nitrites in processed meats, our reported association for total meat intake may not capture these differential effects and should be interpreted with this caveat. Third, residual confounding by unmeasured factors such as genetic susceptibility, medication use, dietary supplement intake beyond single nutrients and overall dietary quality cannot be excluded. Finally, our composite micronutrient deficiency score assigned equal weight to each deficiency, which may not reflect their potentially disparate biological impacts or synergistic interactions (e.g. between vitamin B12 and folate). While weighted scores or interaction models could offer more granularity, the low prevalence of certain deficiencies in our sample limited the statistical power for such analyses. Future studies with larger samples or more balanced deficiency profiles should explore more sophisticated modelling of cumulative nutritional risk.
Conclusions
In conclusion, this binational study provides evidence that Fe deficiency independently predicts CVD risk with a U-shaped dose–response relationship, while cumulative micronutrient deficiencies exert synergistic effects. Prospective dietary analysis identified protective associations for plant-based foods and adverse associations with meat consumption. The null association for vitamin D after adjustment and null cross-cohort meta-analytic findings underscore the importance of rigorous confounder control and highlight methodological challenges in harmonising heterogeneous exposure measures across populations. These findings support screening for Fe deficiency in cardiovascular risk assessment, dietary guidance emphasising plant-based patterns while reducing meat intake and population health strategies to improve overall dietary quality. However, randomised controlled trials are needed to establish causality, and implementation research is required to translate evidence into effective interventions that reduce the global burden of diet-related CVD.
Supplementary material
For supplementary material/s referred to in this article, please visit https://doi.org/10.1017/S0007114526107661
Acknowledgements
The authors would like to express their sincere gratitude to all participants and staff involved in the National Health and Nutrition Examination Survey (NHANES) and the China Health and Retirement Longitudinal Study (CHARLS) for their invaluable contributions to data collection and public data sharing. We also acknowledge the support of the Department of Cardiology at The Affiliated Huaian No.1 People’s Hospital of Nanjing Medical University for providing the institutional infrastructure and collaborative environment that made this research possible. Special thanks are extended to our colleagues and mentors for their insightful discussions and constructive feedback throughout the study. No external funding was received for this work.
This research received no specific grant from any funding agency, commercial, or not-for-profit sectors.
Z. L.: Conceptualisation, Data curation, Formal analysis, Writing – original draft. C. Y.: Conceptualisation, Data curation, Formal analysis. Z. W.: Conceptualisation, Data curation, Formal analysis. J. G.: Conceptualisation, Data curation, Formal analysis. P. G.: Conceptualisation, Writing – original draft, Writing – review & editing. H. L.: Conceptualisation, Writing – original draft, Writing – review & editing, Supervision. All authors read and approved the final manuscript.
The authors declare no conflicts of interest.
