The NHANES data were obtained and used for the present study. We conducted a cross-sectional study of participants within the time period of 2011–2020. NHANES is an investigation program of the National Center for Health Statistics (NCHS) in the USA, which collects health and nutritional information from the American population through interviews and laboratory or medical examinations on participants. The survey is conducted biennially and encompasses the collection of demographic, dietary, examination, and laboratory data. NHANES employs a multi-stage sampling design to represent the U.S. civilian population. It first selects primary units (counties), then divides them, lists the households, and randomly selects from these lists. Subsequently, individuals are chosen based on age, sex, and race/ethnicity. NHANES utilizes stratified sampling and oversampling techniques to enhance the accuracy of subgroup estimates. The weight assigned to each individual reflects their representation in the population, thereby ensuring a nationally representative sample for research purposes. NHANES is a publicly available database, the above-mentioned information can be accessed publicly via the following URL: https://www.cdc.gov/nchs/nhanes/.

In our study, data from NHANES 2011–2020 were aggregated to augment the sample sizes. We recruited individuals with complete blood metal information and cardiovascular disease (CVD) information. CVDs are include heart failure, coronary artery diseases such as angina, myocardial infarction or stroke. The verification of heart failure, coronary artery disease, angina, stroke, or heart attack is based on the questionnaire of MCQ160B, MCQ160C, MCQ160D, MCQ160F, and MCQ160E, respectively. Hypertension was verified by the questionnaire of BPQ020, which asked “Ever told you had high blood pressure”. Diabetes was verified by the questionnaire of DIQ010, which asking “Doctor mentioned you have diabetes”. Demographic information and characteristics, including age, gender, BMI, race, education, and current smoking status, were obtained through interviews. The indicators for blood heavy metal levels are sourced from the NHANES database, and their respective identifiers are LBDBPBSI: Blood lead (umol/L), LBDBCDSI: Blood cadmium (nmol/L), LBDTHGSI: Blood mercury, total (nmol/L), LBDBSESI: Blood selenium (umol/L) and LBDBMNSI: Blood manganese (umol/L). Meanwhile, we can retrieve their specific indicator description information (including the years of recording) from the search interface at https://wwwn.cdc.gov/nchs/nhanes/search/default.aspx. Data for these indicators are available for the years 2011–2020. Data from before 2011 have not been collected into our group due to the absence of test results for these indicators.

Detailed instructions regarding specimen collection and processing can be found in the NHANES Laboratory/Medical Technologists Procedures Manual (LPM). Vials are stored under appropriate frozen conditions (–30°C) until they are shipped to the National Center for Environmental Health for testing. Whole blood concentrations of lead (Pb), cadmium (Cd), total mercury (THg), manganese (Mn), and selenium (Se) are determined using inductively coupled plasma mass spectrometry (ICP-MS). This multi-element analytical technique is based on quadrupole ICP-MS technology. Then, we also analyzed the relationship between cardiovascular mortality and blood heavy metals from 2011 to 2018, using the accessible mortality data from that period. In the current analysis, 68,488 participants from NHANES were enrolled. This present study is a secondary data analysis which lacks personal identifiers and does not require institutional reviewing. However, all procedures conducted in the NHANES were approved by the National Center for Health Statistics Research Ethics Review Board, and all participants were provided with a written informed consent.

The data were processed using R or SPSS 20.0. Student's t-test analysis or the Mann–Whitney U test was used for comparing continuous variables, while the chi-square test was used to compare the composition ratios of each group. The Levene test was performed to assess the homogeneity of variances. Logistic regression analysis was conducted to assess the relationship between each blood metal and CVD after adjusting for covariates. In the logistic model, the included covariates are: Age, Gender, BMI, Race, Education, Diabetes, and Hypertension. In which, the variable number of Age (age) in the database is RIDAGEYR. The variable number of Gender, which clearly defined as male or female is RIAGENDR. The variable number of Body mass index (BMI, kg/m²) is BMXBMI. The variable number of race/ethnicity which includes Mexican American, other Hispanic, non-Hispanic white, non-Hispanic black, and other races or mixed races is RIDRETH1. The variable number of education level which divided into education level below senior high school and education level above senior high school is DMDEDUC2. The variable number of Hypertension is BPQ020. The variable number of Diabetes is DID040. Random forest analysis was performed using the blood metal with or without examination variables (BMI, age and cholesterol) using R and 10-fold cross-validation. We employed a rigorous method to organize and analyze death data and clinical outcomes. The variable were UCOD LEADING, MORTSTAT, and PERMTH INT, which facilitated the accurate identification and categorization of deaths and their underlying causes. The website that can be visited is: https://www.cdc.gov/nchs/data-linkage/mortality.htm. Kaplan-Meier curve, Receiver Operating Characteristic (ROC) curves and the area under curves (AUCs) were constructed using the pROC package in R. Random forest model utilized by Machine learning was used to calculate the AUC score (area under the curve) and explore the potentially important blood metals associated with CVDs. In all analyses, statistical significance was considered at a P-value of < 0.05.

There were 68,488 participants enrolled from NHANES 2011–2020 in the present analysis (Figure 1). In this study, CVD patients were defined as individuals with coronary heart disease, angina, congestive heart failure, stroke, or heart attack. First, we excluded participants without the information on cardiovascular disease (n = 36,645). Then, we excluded participants without examination of blood metals (n = 13,197). Finally, a total of 23,448 individuals were enrolled, including 20,772 non-CVD participants and 2,676 CVD patients.

The demographic characteristics of the present study are shown in Table 1. A total of 23,448 participants were included in the study, with a mean age of 50.32 ± 17.59 years old and 48.5% males. We found that the prevalence of coronary heart disease, congestive heart failure, heart attack, angina, and stroke was 4.1% (n = 963), 3.5% (n = 807), 4.3% (n = 1,016), 2.5% (n = 584), and 4.5% (n = 1,048), respectively. In addition, the overall prevalence of total cardiovascular disease was 11.4% (n = 2,676), consistent with previous findings in other countries (9). Most of the subjects were Non-Hispanic White and had a high level of education. Compared to participants in the non-CVD group, those in CVD group had higher levels of blood lead (0.045 μmol/L vs. 0.610 μmol/L, P < 0.001), and blood cadmium (2.687 nmol/L vs. 3.559 nmol/L, P < 0.001). Moreover, CVD group was likely to have lower levels of blood mercury (3.700 nmol/L vs. 3.240 nmol/L, P < 0.001), blood selenium (2.400 μmol/L vs. 2.340 μmol/L, P < 0.001) and blood manganese (169.260 nmol/L vs. 157.630 nmol/L, P < 0.001).

Table 2 represents the crude and adjusted odds ratios (ORs) for the association between blood metals and CVDs. Amazingly, the crude model indicates a statistically significant difference between increased odds of CVDs and higher blood lead levels [OR (95% CI) = 13.545 (8.470–21.662) P < 0.001]. Futrhermore, participants with higher blood cadmium levels exhibited a higher OR for CVDs [OR (95% CI) = 1.029 (1.022–1.035) P < 0.001]. However, blood mercury, blood selenium, and blood manganese demonstrated an opposite trend. Notably, higher blood selenium appears to be a protective factor against CVDs, associated with a lower OR for CVDs [OR (95% CI) = 0.650 (0.573–0.737) P < 0.001]. Surprisingly, there was also a significant difference between increased odds of CVDs and lower blood mercury levels [OR (95% CI) = 0.991 (0.987–0.995) P < 0.001], traditionally considered an important environmental pollutants. Additionally, individuals with higher levels of blood manganese had lower odds of CVDs [OR (95% CI) = 0.997 (0.996–0.997) P < 0.001].

After adjusting for gender, age, BMI, race, and educational background, the significance of blood lead and blood manganese disappeared. However, the relationship persisted in the adjusted model for blood cadmium, with higher levels showing a higher adjusted OR for CVDs [OR (95% CI) = 1.041 (1.032–1.049) P < 0.001]. Conversely, higher blood mercury levels were associated with a lower adjusted OR for CVDs [OR (95% CI) = 0.995 (0.900–0.999) P < 0.05]. Additionlly, the association between CVDs and blood selenium and CVDs remained significant in the adjusted model, with a lower OR for CVDs [OR (95% CI) = 0.786 (0.688–0.898) P < 0.001].

To further explore the relationship between CVDs and blood metals, we subsequently constructed prediction models for CVDs. Random forest analysis distinguished CVD patients from non-CVD individuals with an AUC of 0.87 using the clinical examination (BMI, age, and blood cholesterol) and 0.9 using blood metals, respectively (Figure 2a). Importantly, the AUC score increased to 0.91 when clinical indices and blood metals were combined. These results suggested that the blood metals have a significant predivtive effect (Figure 2a). Additionally, blood lead and blood cadmium levels were elevated in the CVDs group, whereas blood manganese, mercury, and selenium levels were decreased (Figure 2b). Consequently, the key blood metal indices included in constructing the random forest model were blood lead, manganese, cadmium, selenium, and mercury (Figure 2c). Blood lead emerged as the most important indicator in the random forest model, aligning with the findings from the logistic model.

Figure 3 displays the associations among the various blood metals. Blood lead and blood cadmium demonstrate the strongest correlation, with an R value of 0.215. Additionally, the correlations between blood lead and manganese, blood cadmium and manganese, as well as blood selenium and manganese, are all negative, according to Spearman analysis.

We divided the levels of heavy metals in blood samples into three groups based on tertiles, labeled as Level 1, Level 2, and Level 3 according to their low, medium, and high concentrations, respectively. The concentrations of blood heavy metals in each groups is showed in Table 3. While the differences between Level 1, Level 2, and Level 3 are showed in Figure 4. These results suggest that higher levels of blood lead or cadmium are associated with a higher risk of CVD mortality.