Abstract
Objective
The relationship between serum selenium levels and heart failure risk remains unclear. This study aimed to investigate the potential nonlinear association between serum selenium level and heart failure risk and explore whether hepatic steatosis and dyslipidemia mediate this relationship.
Methods
Data from 6969 adults in the National Health and Nutrition Examination Survey 2017–2020 cohort were analyzed. Logistic regression, restricted cubic spline, and random forest models were used to assess associations between serum selenium level and heart failure risk. Mediation analysis was used to evaluate the indirect effects of hepatic steatosis and lipid parameters. Mendelian randomization was used to infer causality.
Results
A U-shaped association was identified between serum selenium level and heart failure risk (P for nonlinearity = 0.003), with the lowest risk observed at 150–160 µg/L. Both low and high selenium levels were associated with increased heart failure risk. Hepatic steatosis and lipid markers partially mediated this association. Mendelian randomization analysis suggested a potential causal effect of genetically predicted serum selenium level on heart failure risk.
Conclusion
These findings highlight a nonlinear association between selenium exposure and heart failure and suggest possible metabolic pathways underlying this association. However, further research is needed before any clinical recommendations can be made.
Keywords
Introduction
Selenium (Se), a trace element with antioxidant, immunomodulatory, and anti-inflammatory properties, plays a crucial role in maintaining human health through the activity of various selenoproteins.1–3 Both Se deficiency and excess have been implicated in a wide range of diseases, including cardiovascular conditions.4–6 Although previous studies have linked low Se levels to cardiomyopathy and impaired myocardial function,7–9 the relationship between Se and heart failure (HF) risk in the general population remains controversial. Some studies suggest a protective effect of Se, 10 while others report neutral or even adverse associations at higher levels.11,12
HF, a growing global health burden, is strongly associated with metabolic dysfunction, including insulin resistance, hepatic steatosis, and dyslipidemia.13–15 Emerging evidence indicates that Se status may influence hepatic lipid accumulation and liver function, potentially impacting cardiovascular health through metabolic pathways.16–18 However, the potential mediating roles of hepatic steatosis and lipid abnormalities in the Se–HF association have not been clearly elucidated.
Moreover, most existing studies have relied on linear models or basic subgroup analysis, which may not adequately capture complex nonlinear relationships. Machine learning approaches and nonlinear regression techniques, such as restricted cubic spline (RCS), offer advanced methods to explore these associations more accurately.19,20 In addition, Mendelian randomization (MR) can strengthen causal inference by minimizing confounding and reverse causation. 21
The study aims were as follows: (a) to investigate the potential nonlinear relationship between serum Se levels and HF risk in a nationally representative sample of the US population; (b) to evaluate whether hepatic steatosis and dyslipidemia mediate this relationship; and (c) to validate potential causality using MR analysis. Our findings may offer new insights into the cardiometabolic implications of Se and its optimal range for HF prevention.
Materials and methods
Study population
The National Health and Nutrition Examination Survey (NHANES) is designed to assess the health and nutritional status of the US civilian population through a nationally representative sample. The NHANES protocols were approved by the Institutional Review Board of the National Center for Health Statistics (NCHS), part of the US Centers for Disease Control and Prevention (CDC). Informed consent was obtained from all participants. All potentially identifiable information was removed to ensure the confidentiality of participants and their households. This study was conducted in accordance with the Helsinki Declaration of 1975, as revised in 2024. The reporting of this study conforms to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines. 22
We initially identified 15,560 participants from the 2017–2020 NHANES dataset. After excluding 6328 individuals aged <20 years and 87 pregnant participants, a total of 9145 adults remained. We further excluded 1223 individuals with missing hepatic steatosis data and 953 with incomplete serum or dietary Se data. The final analysis included 6969 participants (Supplementary Figure 1).
Measurement of dependent and independent variables
HF was identified through a question regarding participants’ medical history: “Has a doctor or healthcare professional ever diagnosed you with congestive heart failure?” Positive responses were considered indicative of congestive HF. Hepatic steatosis was identified using a controlled attenuation parameter threshold of ≥248 dB/m. Chronic kidney disease (CKD) stages were classified according to the Kidney Disease: Improving Global Outcomes (KDIGO) guidelines. Detailed information on Se measurement as well as the definitions of hepatic steatosis and CKD can be found in the Supplementary materials.
Covariates
This study considered a range of covariates, including age, sex, ethnicity, family income (economic status measured by the poverty income ratio (PIR), with low income defined as PIR <1.3 and high income as PIR ≧ 1.3), marital status, educational level, smoking status, alcohol consumption, physical activity, body mass index (BMI), waist circumference, diabetes, hypertension, triglycerides, and glycosylated hemoglobin (HbA1c) levels. Definitions of these covariates can be found in the Supplementary materials. All standard blood and biochemical tests were conducted in accordance with the procedures outlined in the NHANES Laboratory/Medical Technologist Procedure Manual. 23
Statistical analysis
Survey-weighted regression and mediation analyses
Following NHANES analytical protocols, sample weights, stratification, and primary sampling units were applied to account for the complex survey design. For continuous variables, study population characteristics were presented as means with standard deviation, while categorical variables were reported as percentages. Participants were divided into four groups based on serum Se quartiles. To examine the relationships between serum Se level, hepatic steatosis, and HF risk, multivariate logistic regression was employed, using three distinct statistical models. In Model 1, adjustments were made for age (continuous), sex (male or female), and race/ethnicity (non-Hispanic White, others). Model 2 further adjusted for BMI (categorized into the following ranges: <18.0, 18.0–24.9, 25.0–29.9, 30.0–34.9, or ≥35.0 kg/m2); alcohol consumption (nondrinker, light drinker, moderate drinker, and heavy drinker); smoking status (never smoked, former/current smoker); diabetes medication use (yes or no); total cholesterol (continuous); and blood pressure (continuous). Model 3 incorporated additional adjustments for HbA1c levels (<6.5% or ≥6.5%) and waist circumference (continuous). Linear trends were assessed by treating the median of each category as a continuous variable. No multicollinearity was detected among the covariates and the serum Se framework (Supplementary Table 1). To address potential sample size reductions due to missing covariate data, multiple imputation was performed using the R JOMO package. 24 This method estimates missing values based on observed data, applying hierarchical imputation for sampling units.
To evaluate the dose–response relationship between Se and both hepatic steatosis and HF, an RCS regression model with four knots was applied after logarithmically transforming the serum Se values. Nonlinearity was assessed by comparing models with and without the cubic spline term using the likelihood ratio test. Mediation analyses were conducted using the R MEDIATION package, 25 which explored the mediating roles of serum Se, hepatic steatosis, liver fibrosis, CKD, blood lipids, and HF. The direct effect (DE) represents the influence of serum Se level on HF risk without considering mediators, while the indirect effect (IE) reflects the impact of serum Se level on HF risk mediated by hepatic steatosis. The total effect (TE) is the sum of DE and IE, and the proportion of the mediation effect is calculated by dividing IE by TE.
Stratified analyses were conducted to assess various subgroups based on race (non-Hispanic White, others), age (≤60 years, >60 years), BMI (<30 kg/m2, ≥30 kg/m2), HbA1c level (<6.5%, ≥6.5%), smoking status (never smoked, former/current smoker), and alcohol consumption (nondrinker, former/current drinker). Interaction significance was evaluated by testing P values for the product terms between the serum Se level and stratified categories.
We performed several sensitivity analyses to ensure the reliability of our results. First, extreme values (n = 26) were excluded from the dataset. Then, dietary Se was added to the model to account for potential biases related to dietary intake. Additionally, as serum Se binding is influenced by serum albumin levels, albumin was included in the model adjustments. To address potential cognitive biases related to educational attainment, economic status, and physical activity, further adjustments were made as follows: educational level (above or below junior college), economic status (<1.3 or ≥1.3), and physical activity (inactive or otherwise).
Nonlinear modeling and machine learning
The data were subsequently divided into training and testing sets in an 8:2 ratio. A random forest model was trained to effectively capture the nonlinear relationship between serum Se level and HF risk. To address class imbalance, optional sample weights were incorporated, and downsampling strategies were applied to enhance model stability and generalizability. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), and a comparative analysis was conducted with a traditional logistic regression model. To further investigate the nonlinear effects, partial dependence plots (PDPs) were used to visually demonstrate the marginal effect of Se on HF predictions, thereby revealing its potential nonlinear influence.
Mediation MR analysis
To investigate the potential causal effects of Se on HF, we conducted a two-sample MR analysis. Genetic instruments (single nucleotide polymorphisms (SNPs)) associated with serum Se levels were extracted from a large-scale genome-wide association study conducted in European populations. SNPs were selected using a genome-wide significance threshold of P < 5 × 10−8 and clumped to ensure independence by applying a linkage disequilibrium threshold of r2 < 0.001 within a 10,000-kb window. Palindromic SNPs with ambiguous strand alignment (i.e. same alleles on forward and reverse strands) were excluded to prevent strand misalignment. We calculated the F-statistic for each SNP to assess instrument strength and excluded variants with F <10 to avoid weak instrument bias. The two-sample MR analysis was used to assess the linear causal effect of genetically predicted Se levels on HR risk. This method is not capable of evaluating nonlinear dose–response relationships. Summary-level genetic association data for HF were obtained from publicly available datasets (https://www.ebi.ac.uk/gwas/). The inverse variance-weighted method was used as the primary MR estimator. To evaluate the robustness of the causal estimates and assess horizontal pleiotropy, we performed sensitivity analyses using the MR-Egger, weighted median, and MR-PRESSO methods.
All analyses were conducted using R version 4.2.0 (R Foundation). A two-sided P value <0.05 was considered to indicate statistical significance.
Results
Baseline data
Table 1 presents the baseline characteristics of the 6969 participants, stratified by serum Se quartiles. The mean age of the participants was 48.16 years, and 49.0% of them were males. The median serum Se level was 183.70 (interquartile range: 168.70–200.46) μg/L. Higher Se levels were associated with male sex, unmarried, higher educational level, lower income, greater dietary Se intake, and increased prevalence of hepatic steatosis (all P ≤ 0.05). Additionally, high-density lipoprotein cholesterol level tended to decrease with rising Se levels, whereas low-density lipoprotein cholesterol (LDL-C), triglyceride, and total cholesterol levels increased.
Baseline characteristics of patients stratified by serum selenium level in NHANES 2017–2020 (N = 6969).
Normally distributed continuous variables are described as means ± SD, and continuous variables without a normal distribution are described as medians (interquartile ranges). Categorical variables are presented as numbers (percentages). All estimates accounted for complex survey designs.
ALB: albumin; ALT: alanine aminotransferase; AST: aspartate transaminase; CAP: controlled attenuation parameter; CCR: creatinine clearance rate; FBG: fasting blood glucose; GGT: γ-glutamyl transferase; HDL-C: high-density lipoprotein cholesterol; LDL-C: low-density lipoprotein cholesterol; LSM: liver stiffness measurement; PLT: platelets; TC: total cholesterol; TG: triglycerides; WC: waist circumference.
Relationship between serum Se level, hepatic steatosis, and HF risk
Weighted logistic regression models (Table 2) showed an inverse association between Se level and HF. After adjusting for confounders, the odds ratios (ORs) and 95% confidence interval (CIs) across Se quartiles were as follows: quartile 1 (reference); quartile 2, OR = 0.58 (95% CI: 0.29–1.14); quartile 3, OR = 0.45 (95% CI: 0.21–0.97), and quartile 4, OR = 0.42 (95% CI: 0.15–1.20), with P for trend = 0.041. The association between log-transformed Se and HF was not statistically significant (OR = 0.13, P = 0.23), but the trend indicated a protective direction.
Odds ratios for heart failure across quartiles of serum selenium levels.
No. = cases/total.
Logistic regression analyses were performed to estimate the odds ratio (OR) and 95% confidence interval (CI) for heart failure across different quartiles of serum selenium levels. In Model 1, adjustments were made for age (continuous), sex (male or female), and race/ethnicity (non-Hispanic White or other). Model 2 included additional adjustments for body mass index (<18.0, 18.0–24.9, 25.0–29.9, 30.0–34.9, or ≥35.0 kg/m2), alcohol consumption (nondrinker, light drinker, moderate drinker, and heavy drinker), smoking status (never smoked, former/current smoker), diabetes medication use (yes or no), total cholesterol (continuous), and blood pressure (continuous). Model 3 was further adjusted for glycosylated hemoglobin (<6.5% or ≥6.5%) and waist circumference (continuous).
For hepatic steatosis, higher Se level was associated with a greater risk. Fully adjusted ORs (95% CIs) across quartiles were as follows: quartile 1 (reference); quartile 2, OR = 1.10 (95% CI: 0.77–1.59); quartile 3, OR = 1.46 (95% CI: 1.00–2.12), and quartile 4, OR = 1.37 (95% CI: 0.90–2.09), with P for trend = 0.033. Log-transformed Se was associated with a 115% reduction in risk (P = 0.09), suggesting a nonlinear relationship (Supplementary Tables 2 to 5).
Stratified analysis of serum Se level and HF risk
Across stratified analyses, no significant interaction was observed between serum Se level and HF risk across subgroups, including age, race/ethnicity, BMI, HbA1c, smoking status, and alcohol consumption subgroups (all P for interaction >0.05; Table 3). The inverse association between Se level and HF risk remained broadly consistent across these subgroups.
Stratified analyses of the association between quartiles of serum selenium levels and heart failure risk.
Logistic regression analyses were used to calculate the odds ratio (OR) and 95% confidence interval (CI) for heart failure across different quartiles of serum selenium levels. In Model 1, adjustments were made for age (continuous), sex (male or female), race/ethnicity (non-Hispanic White or other), and body mass index (<30.0 or ≥30.0 kg/m2). Additional adjustments included alcohol consumption (nondrinker or former/current drinker), smoking status (never smoked or former/current smoker), diabetes medication use (yes or no), total cholesterol (continuous), blood pressure (continuous), glycosylated hemoglobin (<6.5% or ≥6.5%), and waist circumference (continuous). Stratification was performed by excluding the stratification variable from the model. The Wald test was used to assess the interaction between continuous selenium levels and stratification variables.
Nonlinearity detection
To explore the shape of the relationship between serum Se level and HF risk, we employed RCS modeling and machine learning–based prediction analysis.
RCS regression revealed a significant nonlinear U-shaped association between Se levels and HF risk (P for nonlinearity =0.003). As shown in Figure 1(a), the risk of HF was lowest at moderate Se levels (approximately 150–160 µg/L), while both low (<120 µg/L) and high (>170 µg/L) Se levels were associated with an increased risk of HF. A similar nonlinear pattern was observed for the association between Se level and hepatic steatosis (Figure 1(b)), supporting its role as a potential mediator in the Se–HF pathway.
Restricted cubic spline regression. Associations between serum selenium levels and hepatic steatosis (a) and heart failure risk (b). A restricted cubic spline regression model with four knots was used to estimate the dose–response relationship. Odds ratios (ORs) were adjusted for age (continuous), sex (male or female), race/ethnicity (non-Hispanic White or other), body mass index (<18.0, 18.0–24.9, 25.0–29.9, 30.0–34.9, or ≥35.0 kg/m2), alcohol consumption (nondrinker, light drinker, moderate drinker, and heavy drinker), smoking status (never smoked, former/current smoker), diabetes medication use (yes or no), total cholesterol (continuous), blood pressure (continuous), glycosylated hemoglobin (<6.5% or ≥6.5%), and waist circumference (continuous). CI: confidence interval.
To further characterize the complex relationship between Se and clinical outcomes, we implemented a random forest model incorporating all candidate covariates. The variable importance ranking confirmed that serum Se level was among the top predictors of HF risk. PDPs generated from the random forest model revealed the same U-shaped pattern as observed in the RCS model, reinforcing the nonlinear nature of the relationship (Supplementary Figure 2).
Moreover, the random forest model (AUC = 0.970) significantly outperformed the traditional logistic regression model in terms of discriminative power (AUC =0.827) (Supplementary Figure 3), indicating the superiority of machine learning methods in capturing nonlinear interactions and complex feature relationships in this context.
Mediation analysis
Parallel mediation models showed that hepatic steatosis explained 8.5% (3.5%–17.5%) of the Se–HF relationship (P < 0.05) (Figure 2 and Supplementary Figures 4 and 5). When fibrosis and CKD were included (Supplementary Figures 6 to 8), hepatic steatosis explained 14.2% (7.3%–25.9%) of the mediation (P < 0.05). Se level showed protective effects against both CKD and HF (Supplementary Figure 9).
Mediation analysis. Mediation analysis models examining the role of hepatic steatosis in the relationship between serum selenium levels and heart failure risk. (a) Basic mediation model. (b–d) Blood lipids as additional mediators: (b) triglycerides, (c) low-density lipoprotein cholesterol (LDL-C), and (d) high-density lipoprotein cholesterol (HDL-C). The results suggest that hepatic steatosis significantly mediates the association between serum selenium levels and heart failure risk.
Mediation MR
Two-sample MR using genetic instruments for Se confirmed a protective effect on HF (OR = 1.054; 95% CI: 1.028–1.081; P = 3.75 × 10−5) (Supplementary Table 6). MR-Egger showed no pleiotropy (P > 0.05), and Cochran’s Q test found no heterogeneity (P = 0.34) (Supplementary Tables 7 and 8). Leave-one-out analysis confirmed the result’s robustness (Supplementary Figure 10), supporting a causal role of Se in HF prevention.
Discussion
This study is among the first to comprehensively examine the relationship between serum Se level and HF risk using a multimethod approach. RCS regression revealed a significant U-shaped association, suggesting that both deficient and excessive Se levels may increase HF risk, with an optimal protective range of 150–160 µg/L. Although the machine learning–based random forest model did not directly demonstrate the dose–response shape, it confirmed serum Se level as a top predictor of HF risk, and its PDP demonstrated a nonlinear pattern broadly consistent with the spline analysis. Mediation analysis further indicated that hepatic steatosis and dyslipidemia partially mediated this relationship, supporting the existence of a cardio–hepatic–metabolic axis in HF pathogenesis. Additionally, two-sample MR analysis provided evidence for a causal association between genetically predicted serum Se levels and HF risk. Together, these findings offer a mechanistic and epidemiological basis for Se’s dual role in cardiovascular health and suggest that maintaining the serum Se level within an optimal range may be clinically beneficial.
Our findings are consistent with those of prior studies demonstrating both the protective and detrimental effects of Se, depending on dosage and population characteristics. Se deficiency has long been associated with cardiomyopathy and impaired cardiac contractility, particularly in the context of Keshan disease.8,9 Conversely, excessive Se intake may induce oxidative stress and impair metabolic function, leading to adverse cardiovascular outcomes.11,12,26–29 This duality may explain the U-shaped pattern observed in our study and highlights the narrow therapeutic window for Se in cardiovascular health.
The role of hepatic steatosis in mediating the Se–HF association is noteworthy. Se has been implicated in hepatic lipid metabolism through selenoprotein regulation, influencing oxidative balance, inflammation, and insulin sensitivity.16–18 Studies have shown that Se can modulate the expression of key transcription factors such as SREBP-1c and PPARγ, thereby promoting or inhibiting lipogenesis depending on concentration. Our mediation analysis showed that hepatic steatosis accounted for a significant proportion of the IE, suggesting that liver fat accumulation may serve as a metabolic conduit between Se exposure and cardiac dysfunction. 30 Moreover, the involvement of triglycerides and LDL-C further underscores the contribution of systemic lipid dysregulation. 31
The application of machine learning models, particularly random forest, added robustness to our findings. These models confirmed Se level as one of the top predictors of HF and reproduced the nonlinear association observed in traditional regression models. The superior performance of random forest (AUC = 0.970) compared with logistic regression (AUC = 0.827) suggests potential utility in risk stratification and variable prioritization in clinical settings. 32
Our study also leveraged MR analysis to strengthen causal inference. Using genetic variants as instrumental variables, we observed a significant association between genetically predicted Se levels and HF risk, independent of confounding and reverse causation. These results provide complementary genetic evidence supporting Se’s role in HF pathogenesis. Although the MR analysis supports a potential causal link between Se level and HF risk by reducing confounding and reverse causality, its linear modeling framework limits its ability to confirm the U-shaped association observed in the cross-sectional analysis. Future nonlinear MR approaches or stratified genetic analyses may provide a more precise understanding of this complex relationship. MR is increasingly used in nutritional epidemiology to infer causality in cases where randomized controlled trials are infeasible.33–35
Subgroup analyses confirmed the robustness of the inverse Se–HF association across sex, age, race/ethnicity, BMI, HbA1c, smoking status, and alcohol consumption. No significant interactions were observed, suggesting that the nonlinear Se–HF relationship is consistent across diverse demographic and metabolic profiles. These findings support the generalizability of our results and Se’s potential cardiometabolic relevance.
Nevertheless, several limitations merit consideration. First, the cross-sectional design precludes temporal inference, although MR analysis partially addresses causality. Second, HF was self-reported, potentially leading to misclassification. Third, Se measurement was based on total serum levels without assessing selenoprotein subtypes. Finally, residual confounding cannot be fully excluded.
This study reveals a U-shaped relationship between serum Se levels and HF risk in the general US adult population, mediated in part by hepatic steatosis and lipid abnormalities. These findings emphasize the importance of maintaining optimal Se status and suggest a potential metabolic link between Se exposure and cardiovascular health. 36 Further prospective and interventional studies are warranted to validate these findings and refine Se intake recommendations.
Conclusion
This study identifies a U-shaped relationship between serum Se levels and HF risk, with both low and high Se levels associated with an increased risk. Hepatic steatosis and dyslipidemia partly mediate this association, highlighting the potential role of the cardio–hepatic–metabolic axis in HF pathogenesis. These findings underscore the importance of precise Se management in individuals with metabolic comorbidities. However, the observed nonlinear pattern warrants further validation in longitudinal and interventional studies before clinical recommendations can be made.
Supplemental Material
sj-pdf-1-imr-10.1177_03000605251370320 - Supplemental material for Exploring optimal serum selenium range for heart failure prevention: A U-shaped association and mediation by hepatic steatosis in US adults
Supplemental material, sj-pdf-1-imr-10.1177_03000605251370320 for Exploring optimal serum selenium range for heart failure prevention: A U-shaped association and mediation by hepatic steatosis in US adults by Han Li, Wenhu Liu, Jingjie Xiong, Xuehua Wang, Qian Xu, Ni Xiong, Yanling Huang, Yan Wang, Jing Hu and Zhaohui Wang in Journal of International Medical Research
Supplemental Material
sj-pdf-2-imr-10.1177_03000605251370320 - Supplemental material for Exploring optimal serum selenium range for heart failure prevention: A U-shaped association and mediation by hepatic steatosis in US adults
Supplemental material, sj-pdf-2-imr-10.1177_03000605251370320 for Exploring optimal serum selenium range for heart failure prevention: A U-shaped association and mediation by hepatic steatosis in US adults by Han Li, Wenhu Liu, Jingjie Xiong, Xuehua Wang, Qian Xu, Ni Xiong, Yanling Huang, Yan Wang, Jing Hu and Zhaohui Wang in Journal of International Medical Research
Footnotes
Acknowledgments
We are grateful to the participants and staff of the National Health and Nutrition Examination Survey study.
Author contributions
Conceptualization, Han Li, Wenhu Liu, Jingjie Xiong, Yanling Huang, and Jing Hu; Data curation, Han Li; Investigation, Ni Xiong; Methodology, Han Li and Wenhu Liu; Project administration, Zhaohui Wang; Resources, Xuehua Wang; Supervision, Jing Hu and Zhaohui Wang; Writing–original draft, Han Li and Wenhu Liu; Writing–review & editing, Jingjie Xiong, Xuehua Wang, Qian Xu, Yan Wang, Jing Hu, and Zhaohui Wang.
Data availability
Data will be made available on request.
Consent to participate
Not applicable.
Consent for publication
Not applicable
Declaration of conflicting interests
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Ethical considerations
The NHANES study protocols were approved by the National Center for Health Statistics (NCHS) Research Ethics Review Board. As this study involved secondary analysis of publicly available deidentified data, additional ethical approval was not required.
Funding
This research received financial support from the National Natural Science Foundation of China under grant NO. 82070400 and NO. 82270367.
Supplemental material
Supplemental material for this article is available online.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
