Associations between patterns of blood heavy metal exposure and health outcomes: insights from NHANES 2011–2016

The National Health and Nutrition Examination Survey (NHANES) is an ongoing research program designed to assess the health and nutrition status of a representative population of noninstitutionalized individuals in the United States. The NHANES survey follows a multistage, stratified sampling approach and has been conducted biennially since 1999. For our study, we utilized publicly available NHANES data from the years 2011–2012, 2013–2014, and 2015–2016. The study focused on individuals aged 20 years and above. To ensure data completeness, participants with missing information on heavy metals in blood (Cd, Mn, Pb, Hg, Se, Cu, and Zn), malignancies, and severe cardiovascular diseases (e.g., previous myocardial infarction, heart failure, or stroke) were excluded from the analysis. Ultimately, 7,236 participants (3,728 males and 3,508 females) were included in the final analysis. The NHANES study and its protocols were approved by the Institutional Review Board of the National Center for Health Statistics, Centers for Disease Control and Prevention. All participants provided informed consent before participating in the survey. For detailed information regarding the study methodology, it can be accessed on the NHANES website at www.​cdc.​gov/​nchs/​nhanes/​irba98.​html.

Biospecimens for Pb, Cd, Mn, Hg, Se, Cu, and Zn in blood were collected at the mobile examination center (MEC). Trained phlebotomists at the MEC conducted venipuncture to collect non-fasting blood samples, ensuring a minimum volume of 0.25 ml per vial. These samples were subsequently processed, stored, and shipped to various laboratories across the United States for analysis. The heavy metal concentrations in whole blood specimens (Cd, Mn, Hg, Pb, Se, Cu, and Zn) were directly determined via mass spectrometry following a simple dilution preparation step. Rigorous scrutiny was applied to the measured values of heavy metals, with any incomplete or improbable data sent back to the laboratories for confirmation. The lowest detection limits were as follows: Cd (0.10 µg/L), Mn (0.99 µg/L), Hg (0.28 µg/L), Pb (0.07 µg/dL), Cu (2.5 µg/dL), Zn (2.9 µg/dL), and Se (4.5 µg/dL). Values falling below the detection limit were calculated using the square root of 2 divided by the lowest detection limit. Units of Cd, Mn, and Hg (µg/L) were converted to nmol/L using conversion factors of 8.897, 18.202, and 4.99, respectively. Pb concentrations in µg/dL were converted to µmol/L by multiplying by 0.0483, while Se concentrations in µg/L were converted to µmol/L using a factor of 0.0127. Cu concentrations in µg/dL were converted to µmol/L by multiplying by 0.1570, and Zn concentrations in µg/dL were converted to µmol/L by multiplying by 0.1530. Sample weights were not factored into the analysis due to the limited number of replicates in which heavy metals were measured. For detailed instructions on specimen collection and processing, publicly available information can be accessed on the NHANES website, ensuring transparency and reproducibility.

Demographic factors were collected through household interviews, encompassing sex (male and female), age (recorded as a continuous variable), ethnic origin (Mexican-American, Hispanic, non-Hispanic White, African American, and other races), and education level (categorized as high school and below [≤ high school] or some college and above [≥ college]). Smoking status was self-reported and categorized as never, current, or former smoker. Additionally, participants’ body mass index (BMI) was determined using height and weight measurements obtained during the medical examination. BMI values were then categorized as follows: normal [< 25.0 kg/m²], overweight [25.0–29.9 kg/m²], or obese [≥ 30.0 kg/m²].

Systolic blood pressure (SBP) and diastolic blood pressure (DBP) were determined as the average of four measurements. Hypertension was diagnosed based on the use of anti-hypertensive treatment, or when the mean SBP was ≥ 140 mmHg or the mean DBP was ≥ 90 mmHg at baseline. Diabetes mellitus was confirmed either by the use of anti-hyperglycemic therapy or when the HbA1c level was ≥ 6.5%. A history of heart failure, coronary artery disease, stroke, angina pectoris, or myocardial infarction was self-reported, and the presence of any one of these conditions was considered indicative of heart diseases. Malignant neoplasms were identified when physicians were informed of a cancer diagnosis or any malignancy.

The data were analyzed and presented based on NHANES year (NHANES 2011–2012, 2013–2014, and 2015–2016). Continuous variables were expressed as mean ± standard deviation (SD), while categorical variables were presented as percentages. Due to the skewed nature of metal concentrations in blood, their distribution was described using the median (interquartile range, IQR). To achieve a more normal distribution for analysis purposes, a logarithmic transformation of the metal concentrations was conducted.

Clustering analysis was employed to identify patterns of exposure to different heavy metals. This unsupervised learning technique is used to discern natural groupings of observations based on the inherent structure of the dataset, thereby reducing highly multivariate datasets. Cluster analysis categorizes individuals into relatively uniform groups, enabling direct comparison among these groups. For the identification of blood heavy metal exposure patterns, the FASTCLUS program was utilized. Participants were classified based on similarities in their blood heavy metal exposure using the K-means method, which is one of the most commonly used clustering algorithms. This method computes cluster centers based on least squares estimation. Prior to the analysis, the number of clusters needed to be specified. Initially, the FASTCLUS program generated 20 clusters, then temporarily removed participants from clusters containing fewer than 5 individuals. From this subset, the number of clusters varied from 2 to 6 to ascertain the optimal number that would provide a reasonably sized and interpretable solution. Ultimately, three clustering solutions were selected to represent the blood heavy metal exposure patterns observed within this population.

The sample matrix eigenvectors were calculated, and the top three eigenvariables extracted 91.10%, 91.48%, and 91.47% of the sample information in 2011–2012, 2013–2014, and 2015–2016, respectively. These eigenvariables served to identify three patterns of metal concentrations in blood using a stepwise clustering method. Differences among the three patterns across NHANES years were compared using a nonparametric test for multiple group comparisons (Kruskal-Wallis test). Additionally, logistic regression analysis was employed to investigate the relationship between these patterns and outcome variables (such as malignant neoplasms, hypertension, heart diseases, and diabetes), while accounting for covariates. The evolutionary characteristics of the three patterns concerning the outcome variables were visualized using a Sankey plot. Statistical analyses were conducted using SAS (version 9.4) and R software (version 3.3.2) (http://​www.​r-project.​org/​), with statistical significance set at P < 0.05.

Table 1 illustrates the baseline characteristics of NHANES samples collected during 2011–2012, 2013–2014, 2015–2016, and the combined years. It presents median values (IQR) and percentages for demographic factors such as age, sex, BMI, race, education, smoking status, and the prevalence of malignant neoplasms, hypertension, heart diseases, and diabetes. No statistically significant differences were observed when comparing these characteristics across the examined years. Table 2 showcases median values (IQR) of metal concentrations (Cd, Mn, Pb, Hg, Se, Cu, Zn). None of the inter-year comparisons yielded statistical significance. These findings suggest consistently obtained samples, ensuring a balanced representation across different periods.

The cluster analysis results revealed a three-cluster solution effectively distinguishing heavy metal concentrations in blood. Pattern 1, labeled ‘low Cd-Pb-Hg-Se’, consistently exhibited low exposure levels to Cd, Pb, Hg, and Se. Pattern 2, termed ‘low Cu & high Hg-Se’, showed low Cu exposure but high Hg and Se exposure. Pattern 3 indicated high exposure to Cu, Cd, and Pb, referred to as ‘high Cu-Cd-Pb’. These patterns remained consistent throughout NHANES 2011–2015, demonstrating stability. Figure 1 visually demonstrates the variations in heavy metal concentrations among these patterns. As depicted in Fig. 1 and detailed in Tables S1-S3, all heavy metal concentrations, except Zn, exhibited statistically significant differences among the patterns. Notably, in NHANES 2011–2012 and 2013–2014, Mn levels were elevated in pattern 2 and reduced in pattern 3. However, in NHANES 2015–2016, Mn exposure increased in pattern 1 while remaining lower in pattern 3.

Figure 2 depict the relationship between the identified patterns and adverse outcome events. In the case of malignant neoplasms, participants in pattern 1 exhibited odds ratios (ORs) of 1.29 (95% CI 1.01–1.65) for 2015–2016 and 1.33 (95% CI 1.02–1.72) for 2011–2012 when compared to pattern 2. Pattern 3 showed a significant association in 2013–2014, with an OR of 0.62 (95% CI 0.45–0.84) versus pattern 2. Regarding hypertension, participants in pattern 1 displayed ORs of 1.41 (95% CI 1.22–1.63), 1.31 (95% CI 1.14–1.51), and 1.26 (95% CI 1.08–1.47) for the years 2015–2016, 2013–2014, and 2011–2012, respectively, compared to pattern 2. Conversely, participants in pattern 3 exhibited ORs of 0.67 (95% CI 0.57–0.78) and 0.72 (95% CI 0.62–0.84) for the years 2015–2016, and 2013–2014. No statistically significant trends were observed for the ORs concerning pattern 1 and pattern 3 compared to pattern 2 for diabetes. Regarding heart diseases, participants in pattern 1, contrasted with those in pattern 2, had ORs of 1.68 (95% CI 1.33–2.13), 1.76 (95% CI 1.38–2.24), and 1.34 (95% CI 1.05–1.71) for 2015–2016, 2013–2014, and 2011–2012, respectively. Pattern 3 exhibited significance solely in the years 2015–2016 and 2013–2014, with ORs of 0.52 (95% CI 0.38–0.71) and 0.59 (95% CI 0.42–0.82), respectively.

Figure 3 illustrates that Pattern 1 comprises 37.33% of the participants. Within this group, 3.82%, 4.62%, 8.34%, and 20.55% were associated with heart diseases, malignant neoplasms, diabetes, and hypertension, respectively. Pattern 2 includes 42.81% of the subjects, with 5.16%, 5.40%, 9.02%, and 23.23% linked to heart diseases, malignant neoplasms, diabetes, and hypertension, respectively. Pattern 3 accounted for 64.96% of the subjects, among whom 11.12%, 9.48%, 11.25%, and 33.11% were associated with heart diseases, malignancy, diabetes mellitus, and hypertension, respectively. These results suggest an association with hypertension among the major participants in the three patterns.