Cobalt exposure in relation to cardiovascular disease in the United States general population

Cobalt exposure has adverse health effects on the cardiovascular system in occupational and laboratory studies, but these effects have not been assessed in the general population. We aimed to determine whether serum cobalt levels had relationship with the prevalence of cardiovascular disease (CVD) in the general population. Using data from the National Health and Nutrition Examination Survey (NHANES) (2015–2016), we performed the cross-sectional study. We analyzed the baseline characteristics of 3389 participants (1623 men and 1766 women). Generalized linear models and restricted cubic spline plots curve were undertaken to elucidate the relationship. Stratified subgroup analysis was tested to exclude interaction between different variates and cobalt. Our results showed that the adjusted odds ratios (ORs) with 95% confidence intervals (CIs) for CVD prevalence across the quartiles of cobalt were 0.94 (0.67, 1.30), 1.55 (1.15, 2.10), and 1.74 (1.28, 2.35) compared with lowest quartile. The restricted cubic spline curve also suggested nonlinear and positive association between cobalt and CVD (P for nonlinearity = 0.007). In summary, our cross-sectional results verify that higher cobalt levels are associated with a higher prevalence of cardiovascular disease.


Introduction
Cardiovascular disease (CVD) has caused substantial health and economic burdens around the world (Mozaffarian et al. 2016). The prevalence and control of CVD involve some important risk factors, such as genetic factors, environmental factors, adverse lifestyles, and metabolic changes (Bhatnagar 2017;Yu et al. 2016). However, the risks following exposure to heavy metals and the cardiotoxicity of some drugs and devices are often underestimated (Hantson 2019).
Cobalt is a relatively rare metal element (Barceloux 1999). Studies have shown that heavy metals in environmental pollution can elevate blood pressure and even may cause myocardial toxicity (Wu et al. 2018;Zhang et al. 2016). Besides, anemia has been reported association with elevated cobalt in blood and urine (Fort et al. 2015;Junqué et al. 2020;Wai et al. 2020). Additionally, environmental cobalt or cobalt added to beer was noted to cause CVD, mainly cardiomyopathy (Milon et al. 1968). Humans can be exposed to cobalt in many ways in daily life, including the use of glass, paint and diet being the main source for the general population, and occupational contact while working in the metal industry (Barceloux 1999;Lauwerys and Lison 1994).
Case studies of severe cardiomyopathy have revealed the association between poisoning cobalt doses and CVD due to the increased use of metal-to-metal (MoM) hip implants in patients undergoing total hip arthroplasty (THA) (Choi et al. 2019;Sanz Pérez et al. 2019). The blood cobalt levels of these patients could reach a fatal dose, as high as 641.6 μg/L (Dahms et al. 2014;Fox et al. 2016). Therefore, serum cobalt levels > 20 μg/L are not allowed in patients given a MoM hip implant (Sanz Pérez et al. 2019), and at > 7 μg/L, clinicians should begin to observe the heart function (Tower 2010).
In addition, cobalt can induce hypoxia in cardiomyocytes (Wang et al. 2019). The mechanism may be that cobalt interferes with the production of ATP by interfering with the synthesis of mitochondrial respiratory chain enzymes (Clyne et al. 2001), and at the same time increases oxidative stress (Akinrinde et al. 2016;Lewis et al. 1991), which is also involved in the pathophysiology of CVD (Alfatni et al. 2020). However, it is currently uncertain whether the "normal" range of cobalt exposure in the general population has an effect on CVD.
Therefore, based on the National Health and Nutrition Examination Survey (NHANES) providing serum cobalt levels in samples from the USA in 2015-2016, this study was performed to investigate the relationship between cobalt exposure and the prevalence of CVD in the general population.

Study population
We used data from the NHANES accessible online for design and procedures through the Centers for Disease Control and Prevention (CDC) National Center for Health Statistics (NCHS) (https://www.cdc.gov/nchs/nhanes/index.htm) to estimate the cobalt-related factors of CVD. The data was obtained from the NHANES surveys conducted in the US population in 2015-2016. Totally 9971 patients were admitted, and for whom (n = 3442) the serum concentrations were available. We also excluded participants with missing data on CVD (n = 49) and those pregnant (n = 4). Ultimately, 3389 participants involved in our study as shown in Fig. 1.

Cobalt measurements
Blood samples from the participants were collected after confirming that there was no background contamination. Concentrations of cobalt (micrograms per liter (μg/L)) were measured in blood by inductively coupled plasma mass spectrometry (ICP-MS, PerkinElmer/SCIEX model 500; PerkinElmer, Shelton, CT, USA, http://www.cdc.gov/nchs/ data/nhanes/nhanes_03_04/l06_c_met_hm.pdf) at the Environmental Health Sciences Laboratory of the CDC and the National Center for Environmental Health (NCEH). The normal blood cobalt concentration is defined as < 1 μg/L (Samar et al. 2015). The lower limit of detection (LLOD) for cobalt in this study is 0.06 μg/L. Exposure was divided into weighted quartiles based on the measured distribution of blood cobalt concentration. This study modeled the quartiles of the exposure distribution to assess the potential nonlinear association between cobalt and the target endpoint.

CVD ascertainment
Statistics on CVD consisted of congestive heart failure (CHF), coronary heart disease (CHD), angina pectoris, heart attack, and stroke. The prevalence of the disease was recorded by questionnaire whether the patient had new-onset or preexisting disease at this time. Participants were asked the following question: "Has a doctor or other health professional ever told you that you have congestive heart failure (CHF)/ coronary heart disease (CHD)/angina pectoris/heart attack/ stroke?" with a standardized medical condition questionnaire that was administered by trained interviewers.

Covariates
Participants self-reported the following covariates: age, gender, ethnicity (Mexican American, other Hispanic American, non-Hispanic black, non-Hispanic white, and other races), education level, total energy intake, poverty-income ratio, and health behaviors including smoking, alcohol consumption, and physical activity, which were collected through online questionnaires (http://www.cdc.gov/nchs/nhanes/ nhanes_questionnaires.htm). Hypertension and diabetes mellitus were defined as self-reported diagnoses. Body mass index (BMI in kg/m 2 ) ≧ 30 was defined as obesity. The concentrations of plasma lipids were publicly available on the NCHS and USDA websites (https://www.cdc.gov/nchs/ nhanes/index.htm).

Statistical analysis
Continuous variables expressed as the mean (standard deviation, SD), and skewed data expressed as median (quartile range) were analyzed by t-test. Categorical variables presented as numbers (percentages, %) were analyzed by the chi-square test to demonstrate the demographic and health characteristics with CVD and non-CVD. Data on serum cobalt concentration was in log2-transformation because of its skewed distribution. Adjusted odds ratios (ORs) with 95% confidence intervals (CIs) and probability value (P) were used to evaluate the strength of the associations between variables.
We used a multivariate logistic generalized linear model to determine the relationship between cobalt quartiles and CVD. In model 1, we calculated the OR adjusted for age and gender. Model 2 were adjusted for other potential confounders which included age, gender, education level, ethnicity, BMI, alcohol user, diabetes mellitus, and hypertension (model 2). Based on model 2, the full adjustment was made for poverty-income ratio, energy intake, physical activity, high density, total cholesterol, and hemoglobin (model 3).
We also modeled each cobalt quartile as an ordinal variable to test for linear trends. Subgroup analysis was performed to examine whether the effect of cobalt on CVD could be modified by age, gender, physical exercise, diabetes, hypertension, and obesity. We used restricted cubic spline regression with 3 knots at the 10th, 50th, and 90th percentages to graph the relationship between cobalt and the prevalence of CVD. The identified nonlinear associations were tested by Wald χ2 tests (Desquilbet and Mariotti 2010). The interaction between cobalt and stratified covariates in the multiple regression model was tested. We performed all statistical analyses with R software version 3.6.0 and defined P < 0.05 as significant.

Baseline characteristics
The demographic characteristics of the exposure levels in the study population are presented in Table 1. Significant differences were found in age, gender, ethnicity, hypertension, alcohol intake, energy intake, physical activity, high cholesterol, total cholesterol, and hemoglobin (all P ≤ 0.001) among the different levels of cobalt exposure groups (see Table 1).

Association between cobalt and total CVD
In the fully adjusted model, the ORs (95% CIs) for the CVD prevalence across the cobalt quartiles were 0.94 (0.67, 1.30), 1.58 (1.15, 2.10), and 1.74 (1.28, 2.35) (P for trend < 0.001), compared with the first quartile of cobalt exposure (see Table 2). This characteristic of increased ORs for the prevalence with increased cobalt concentration persisted in the other two adjusted models. Figure 2 shows that the restricted cubic spline plot curve graphs the nonlinear association between cobalt and total CVD (p for nonlinearity = 0.007).

Stratified associations between cobalt and total CVD
Subgroup analysis stratified by different variates including age, gender, physical exercise, diabetes, hypertension, and obesity confirmed a positive association between the total CVD prevalence and cobalt. There was no significant trend toward an association between total CVD and cobalt among participants who were female, younger than 60 years old, or without a history of hypertension and participated in vigorous physical activity. The stratified model of obesity and diabetes also showed that the trend was statistically significant as the cobalt concentration increased and the prevalence of CVD increased (all P < 0.05) ( Table 4).  Data are presented as mean (SD) or n (%) Model 1 was adjusted for age and gender. Model 2 was adjusted for age, gender, education level, ethnicity, diabetes mellitus, smoker, alcohol user, hypertension, body mass index. Model 3 was adjusted for age, gender, education level, ethnicity, diabetes mellitus, smoker, alcohol user, hypertension, body mass index, poverty-income ratio, physical activity, and energy intake, high density cholesterol total cholesterol and hemoglobin. * P < 0.05, ** P < 0.01, *** P < 0.001; CVD, cardiovascular disease; OR, odd ratio; CI, confidence interval

Discussion
Our cross-sectional study found a nonlinear positive correlation between cobalt exposure and the prevalence of CVD, and the correlation was associated with the concentration of blood cobalt in the general population. Previous studies have reported many cases of CVD caused by a cobalt poisoning dose and suggested that THA patients with higher doses of cobalt presented more severe CVD (Dahms et al. 2014;Fox et al. 2016;Samar et al. 2015;Sanz Pérez et al. 2019). Due to a lack of data about a possible association between cobalt and CVD in the general population, our research aimed to fill this knowledge gap.
Our results are similar to prior findings in THA patients. A positive correlation between cobalt exposure and the prevalence of CVD existed in both THA patients and the general population. However, a prospective study found no significant differences in cardiac magnetic resonance (CMR) performance in a high blood cobalt group compared to a lower cobalt group among THA patients (Berber et al. 2017). We speculate the reason for this is that the average blood cobalt concentration (30.0 ± 29.1 μg/L) in the high concentration group in that study was much lower than that among the THA patients with CVD (≧120 μg/L) (Samar et al. 2015).
For occupational cobalt exposed people, a cross-sectional survey in Belgium found no dose-effect relationship between cobalt exposure and dilated cardiomyopathy by measurement of echocardiography (echo) and ECG (Lantin et al. 2013). Simultaneously, another study, the first follow-up cohort study investigating occupational cobalt exposure, also confirmed no differences in the echo and ECG parameters between exposed and unexposed workers and no differences in the prevalence of myocardial infarction, CHD, or heart failure (Linna et al. 2019).
These results seem to be contrary to our results. However, the occupational population in the Belgium study was Fig. 2 The restricted cubic spline plot of the association between cobalt and the total CVD prevalence. Cobalt is log2 transformed. The association was adjusted for age, gender, education level, ethnicity, diabetes mellitus, smoker, alcohol user, hypertension, body mass index, poverty-income ratio, physical activity, energy intake, high density cholesterol, total cholesterol and hemoglobin. CVD: cardiovascular disease Analysis were adjusted for age, gender, education level, ethnicity, diabetes mellitus, smoker, alcohol user, hypertension, body mass index, povertyincome ratio, physical activity, and energy intake, high density cholesterol, total cholesterol and hemoglobin. CVD, cardiovascular disease; *P < 0.05, **P < 0.01; OR, odd ratio; CI, confidence interval significantly younger than the population in our study. Another possible explanation is that it is hard to detect early structural and ECG changes without clinical symptoms in occupational populations. Additionally, regional environmental differences such as seasons, sunlight exposure, geographic characteristics of the natural environment, and industrial production exist between our population and the occupational populations in the two other studies, which can exert an important influence on CVD (Bhatnagar 2017). In addition, the sample sizes of the two other studies are significantly smaller than our study, which may increase the false-negative rate. Additionally, our study defined CVD using a questionnaire asking about clinical symptoms rather than using early clinical testing methods. Furthermore, two specific types of CVD (CHF and heart attack) analyzed in our study show the same trend with cobalt, which is consistent with previous results. CHF results from any functional or structural heart disorder, and it clinically manifests with impaired ventricular filling or ejection of the blood into the systemic circulation (Verbrugge et al. 2013). A previous study explored myocardial specimens of cobalt poisoning patients and verified a higher cobalt concentration, more cardiac hypertrophy, more interstitial fibrosis in the myocardium, and a lower ejection fraction compared with the non-exposed group (Wyles et al. 2017). In addition, as the concentration of cobalt increases, the structure of the heart changes as the left ventricular end-diastolic, end-systolic, and the left atrium volume increases (Lodge et al. 2018).
Heart attack, mostly a sudden and fatal occurrence of coronary thrombosis, was estimated to be associated with the serum concentration of cobalt in our study (Murray and Lopez 1997). However, normal coronary angiography has been reported for patients with cobalt poisoning (Samar et al. 2015), and no significant difference was found in serum troponin levels (Berber et al. 2017). Additionally, with the common use of cobalt in drug-eluting stents, no cases of cobalt poisoning have been reported (Kozuma et al. 2019;Romaguera et al. 2014;Sotomi et al. 2018). However, another study has shown significantly increased serum enzymes and low voltages on ECG in patients with cobalt poisoning (Packer 2016). Therefore, there are still controversies about this correlation. This may be due to the rarity of cases reported so far, so the results are quite different. Therefore, further research and verification in prospective follow-up studies and in animal studies are warranted.
The underlying mechanism of the association of cobalt and CVD is mainly related to mitochondrial respiratory and oxidative stress. Studies use cobalt chloride (CoCl 2 ), a chemical mimetic of hypoxia, to inhibit mitochondrial function through decreasing ATP production rate in the respiratory chain Analyses was adjusted for age, sex, education level, race, diabetes mellitus, smoker, alcohol user, hypertension, body mass index, poverty-income ratio, physical activity, and energy intake, high density cholesterol, total cholesterol and hemoglobin when they were not the strata variables. * P < 0.05, ** P < 0.01, *** P < 0.001; OR, odd ratio; CI, confidence interval (Cheng et al. 2017;Chimeh et al. 2018;Clyne et al. 2001;Wang et al. 2019). Ultimately, CoCl 2 constructs hypoxic condition models and decreased cell viability (Sundaram et al. 2018). As the cobalt concentration increasing, mitochondrial dysfunction may contribute to cobalt-induced oxidative stress in macrophages through an increase in reactive oxygen species (ROS) production (Salloum et al. 2018). ROS in most cells, particularly in cardiac cells, are mainly generated from the mitochondria (Assies et al. 2014). Many cell metabolic cascades are linked and modulated by ROS in a complicated pattern, which form the pathogenesis basis of CVD (Cervantes Gracia et al. 2017). In addition, CoCl 2 causes a significant increase in oxidative stress indicators (hydrogen peroxide (H 2 O 2 ) and malondialdehyde (MDA)), an increase in the expression of caspase 8 (apoptosis initiating factor) in the heart, and a significant decrease in glutathione (GSH) (Akinrinde et al. 2016). Therefore, cobalt may eventually lead to CVD by changing the hypoxia state of the respiratory chain and promoting oxidative stress. The major clinical value of the study is that CVD increases are observed at low concentrations of cobalt and that the associations are statistically significant. To make primary prevention to reduce it, we need to increase the blood metal detection rate for possibly affected patients.
The current study still has several limitations. Firstly, we used the cross-sectional design of NHANES with a lack of temporality, so causality cannot be determined. Secondly, though the regression model was adjusted for a wide range of covariates, unmeasured confounders correlated to CVD may also play a role. In addition, retrospective bias may exist in this study, as this is a retrospective study and subjects may have a "recall bias" toward some of the relevant results. Thus, the impact of cobalt on CVD in the general population merits further prospective research.

Conclusions
In summary, our findings revel that elevated serum cobalt levels are nonlinearly and positively associated with a higher prevalence of cardiovascular disease, suggesting that cobalt exposure may be a risk factor for cardiovascular disease. Although the mechanism of cobalt action on CVD is unclear, the current results may contribute to an epidemiological evidence of day-to-day environmental risk factors for cardiovascular disease.