Data sources and study population

The dataset used in this investigation was exclusively sourced from the National Health and Nutrition Examination Survey (NHANES) website [16]. The NHANES orchestrated by The National Center for Health Statistics (NCHS) is an all-encompassing national survey specifically crafted to obtain data emblematic of the population. This study utilized a sophisticated stratified multistage probabilistic cluster sampling technique to conduct an exhaustive evaluation of the health and nutritional status of the general population throughout the U.S. [17]. In this cross-sectional investigation, we utilized an aggregated 6-year dataset for our analyses to ensure robust and accurate estimates. Data acquisition was conducted in early May 2024. Data analysis was conducted in May 2024. Before participating in the study, all individuals provided written informed consent, and the NHANES investigation obtained approval from the Ethics Review Board of NCHS [18]. During or after data collection, the authors did not have access to information that could identify individual participants. The NHANES study was authorized by the NCHS Ethics Review Committee, and all participants provided written informed consent prior to their participation. This study was exempted from ethical review and approval for secondary analysis, as it was determined that no supplementary institutional review board clearance was required. This study was conducted in accordance with the guidelines of the Declaration of Helsinki. We employed a weighted analysis to ensure that the conclusions drawn from our sample-based analysis were generalizable to the overall U.S. population.

Our analysis used data collected from participants spanning three biennial NHANES cycles (2007–2008, 2009–2010, and 2011–2012). Within the NHANES 2007–2012 dataset, 11,638 participants met the inclusion criteria and possessed complete data on TH measurements and LE8 metrics. Participants were excluded if they fell under the following categories: under the age of 20 years, pregnant, had missing data for the TH or LE8 tests, or had incomplete weight information. Ultimately, the analysis encompassed a total of 3,019 participants within the age of 20 years old and above. A visual representation outlining the selection methodology employed for the study participants is presented in Fig 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Flow chart of patient selection.

https://doi.org/10.1371/journal.pone.0329194.g001

Clarification of the definitions of thyroid hormones and Life’s Essential 8

The AHA initially defined the cardiovascular health paradigm in 2010 based on seven health behaviors and factors (collectively termed LS7) associated with greater CVD-free survival, total lifespan, and quality of life [14]. In 2022, the AHA’s LE8 initiative was a critical advancement in the enhancement of cardiovascular health with components including diet, physical activity, nicotine exposure, sleep health, body mass index (BMI), blood lipids, blood glucose, and blood pressure [15]. Compared to the LS7, LE8 incorporates an additional metric of sleep health, building upon the original cardiovascular health framework [15]. LE8, an enhanced and comprehensive set of health metrics and assessment frameworks, has been used extensively in population-based studies [19–21]. This study employed meticulous data collection techniques as previously documented in the literature. These techniques included the use of dietary questionnaires to determine the Dietary Approaches to Stop Hypertension (DASH) diet score; self-reported minutes per week of moderate to vigorous physical activity; exposure to tobacco and nicotine; length of sleep; and measurements of blood pressure, weight, height, blood glucose, and non-high-density lipoprotein cholesterol (non-HDL) levels [15]. Each metric was evaluated using a scoring algorithm and the aggregate cardiovascular health score was derived from the unweighted mean of these metrics, with scores ranging from 0 to 100 points. Based on literature findings, the LE8 score was categorized into low (< 50), moderate (50–80), and high groups (≥ 80) [15,22]. The LE8 represents a comprehensive assessment of cardiovascular health, integrating both self-reported and objective measures to provide a critical evaluation of health factors essential for cardiovascular risk [15,23].

Serum TH levels were used to evaluate thyroid function. The TH profile encompasses a series of tests aimed at assessing thyroid function, including levels of total thyroxine (TT4), FT4, total triiodothyronine (TT3), FT3, thyroglobulin (TG), thyroglobulin antibodies (TgAbs), thyroid peroxidase antibodies (TPOAb), and TSH. To establish population-based reference ranges for these hormone levels, comprehensive instructions regarding specimen collection and processing were derived from the Laboratory Methods Manual and Laboratory/Medical Technologists Procedure Manual available on the NHANES website [24]. Specifically, the reference ranges for TH levels are as follows: TSH levels (measured in uIU/mL) between 0.24–5.4 define the normal level group, below 0.24 define the low-level group, and above 5.4 define the high-level group. FT3 levels (measured in pg/mL) between 2.5–3.9 define the normal level group, below 2.5 the low-level group, and above 3.9 the high-level group. FT4 levels (measured in ng/dL) are considered normal between 0.6–1.6, below 0.6 indicate a lower level, and above 1.6 indicate a higher level. TT3 levels (measured in ng/dL) fall within the normal range at 87–178, below 87 ng/dL suggest a lower level, and above 178 ng/dL suggest a higher level. TT4 levels (measured in ug/dL) are within the normal range at 6.1–12.2, below 6.1 indicate a lower level, and above 12.2 indicate a higher level. TG levels (measured in ng/mL) are considered normal below 35.0 and categorized as high at or above 35.0. TgAb levels (measured in IU/mL) are within the normal range at < 4.0 and are classified as high at or above 4.0. TPOAb levels (measured in IU/mL) are considered normal below 9.0 and categorized as high at or above 9.0.

Covariates

Covariates were selected based on three criteria: (1) inclusion of established confounders informed by prior evidence of thyroid-cardiovascular associations, clinical relevance, and sufficient outcome events; (2) variables altering effect estimates by > 10% when added to or removed from core models; and (3) variables with p < 0.1 in univariate screening. Multivariate linear regression models were adjusted for demographic characteristics, clinical biomarkers, self-reported comorbidities, and medication use. A multitude of potential confounding variables was scrutinized, drawing from the extant literature on TH or LE8 [12,25–31]. A comprehensive range of covariates was considered, encompassing demographic factors such as age (in years), sex (male and female), race/ethnicity (categorized as non-Hispanic White, non-Hispanic Black, Mexican American, other Hispanic, other race-including Multi-Racial), educational attainment (three segments: less than 9th grade, 9th grade-high school grade, or more than high school grade), marital status (four categories: married, never married, living with partner, or other including widowed, divorced, or separated individuals), poverty income ratio (PIR) divided into low (≤ 1.30), medium (1.31–3.50), and high (> 3.50), and alcohol user (never: had < 12 drinks in lifetime; former: had ≥ 12 drinks in 1 year and did not drink last year, or did not drink last year but drank ≥ 12 drinks in lifetime; current heavy alcohol user: ≥ 3 drinks per day for females, ≥ 4 drinks per day for males, or binge drinking ≥ 4 drinks on same occasion for females or ≥ 5 drinks on same occasion for males on 5 or more days per month; current moderate alcohol use: ≥ 2 drinks per day for females, ≥ 3 drinks per day for males, or binge drinking ≥ 2 days per month; current mild alcohol use: ≤ 1 drinks per day for females, ≤ 2 drinks per day for males) [30,32]. Biomedical indicators were also noted including uric acid (mg/dL), creatinine (mg/dL), alanine aminotransferase (Alt, [U/L]), aspartate aminotransferase (Ast, [U/L]), and urine Iodine (ug/L) [30]. Thyroid disease status was determined using two validated questions: “Do you currently have any thyroid problems?” or “Have you ever been diagnosed with thyroid cancer?” [33]. If the answer to either of the two questions was “Yes,” this variable was classified as “having thyroid disease”; otherwise, it was defined as “No.” CVD was defined as a self-reported condition diagnosed by a physician, including congestive heart failure, coronary heart disease, angina, heart attack, and stroke [34]. Medications associated with thyroid dysfunction, TH modulation, and thyroid disease were identified using these two approaches (see S1 Table). First, we referenced prescription medication records corresponding to thyroid diseases from the International Classification of Diseases-Tenth Revision-Clinical Modification (ICD-10-CM) information in other years of the NHANES database [35,36]. Second, we consulted previously published literature that mentioned the names of medications related to THs in the NHANES dataset [31,37–41].

Statistical analysis

This study re-examined datasets that were openly accessible for analysis purposes by employing a weighted analysis. Due to the absence of weights for thyroid testing included in the 2007–2008 cycle (WTSA2YR) and the inclusion of fasting blood glucose data in LE8, the analysis was conducted using the Fasting Subsample 2 Year Mobile Examination Center (MEC) Weight as the analytical weight (WTSAF2YR). The weights were calculated by dividing the WTSAF2YR values by the number of 2-year cycles included in the analysis (WTSAF2YR/3). Normality assessment for continuous variables was conducted using two methods: visualizing the variables through histograms and examining whether the mean exceeded twice the standard deviation (see S1 Fig). Participant characteristics are presented as means with standard deviations (for normally distributed continuous variables) or medians with interquartile ranges (for non-normally distributed continuous variables). Categorical variables are represented as counts and percentages, as appropriate. LE8 was examined in both continuous and categorical forms. THs were processed for analysis as follows: they were transformed using logarithmic conversion, treated as continuous variables, and categorized into subgroups based on hormone reference ranges, specifically the normal range, low-level group, and high-level group.

To evaluate the differences among groups, we employed various statistical techniques: weighted one-way analysis of variance (suitable for normally distributed data), weighted Kruskal-Wallis tests (applied to non-normally distributed data), and weighted chi-square tests (used for categorical variables). Post hoc analysis was conducted using Tukey’s honest significant difference (HSD) method for pairwise comparisons, and the results were visualized using box plots to illustrate intergroup differences (see S2–S7 Figs). Additionally, we employed weighted linear regression models to determine effect sizes (β) and 95 percent confidence intervals (95% CI) while examining the association between TH (after log transformation) and LE8, as well as the association between LE8 and covariates. Before constructing the multivariate model, we assessed the collinearity between LE8 and other variables by calculating the generalized variance inflation factor (GVIF) (see S2 Table). A variable was considered significantly collinear if the GVIF^(1/2Df) was equal to or greater than two degrees of freedom (Df). We used three models to adjust for various covariates, thereby elucidating the interrelationship between variables and LE8. In Model 1, statistical adjustments were applied specifically to sociodemographic factors, including age, sex, and race/ethnicity. Model 2 expanded these adjustments to include age, sex, race/ethnicity, and urinary iodine. Model 3 (fully adjusted model) was adjusted for age, sex, race/ethnicity, urine iodine, education, marital status, PIR, CVD, creatinine, uric acid, Alt, AST, alcohol use, prescription drugs affecting thyroid function, and thyroid disease. A residual diagnosis was made for the fully adjusted linear regression model to validate its robustness by assessing whether the model residuals followed a normal distribution (see S8–S15 Figs). For all primary analyses, we used the E-value to assess the validity against unmeasured confounders [42,43]. The E-value can test the strength of the association between exposure and outcome while evaluating the evidence of causality. Unmeasured confounders must be associated with both THs and LE8, with a risk ratio equal to the E-value, to fully explain the observed association between THs and cardiovascular health. Additionally, the association between serum TH levels (after log transformation) and LE8 score was examined for potential nonlinearity. Restrictive cubic spline (RCS) analysis was used to delineate the nonlinear relationships between these two variables. The capping method was utilized by applying a 2.5% tail trimming to address extreme values in the RCS analysis.

In the sensitivity analysis, we examined the relationship between a specific TH (after log transformation) and LE8 after grouping according to the reference range of TH. If the sample size within a subgroup was insufficient, the analysis was omitted. To validate the robustness of the results, the sensitivity analysis also included a fully adjusted linear regression analysis, excluding participants with a history of CVD and thyroid disease. Single imputation methods were employed to handle missing data. Data processed using various missing value handling methods and different weighting analyses were employed for the sensitivity analyses in the fully adjusted model. All reported probabilities (p values) were two-sided, and statistical significance was established at a threshold of p < 0.05. We performed analyses using R software (version available at http://www.R-project.org, provided by The R Foundation) and Free Statistics software version 1.9.2. During the preparation for this study, the authors used Kimi (Moonshot AI) and DeepSeek (AI) to improve language and readability.

Study population

Of the 11,638 participants who underwent interviews, 2,083 were less than 20-years-old. Exclusions were made for pregnant women (n = 99), individuals with missing data for TH (n = 929) and incomplete LE8 data (n = 1,961), as well as those with incomplete weight information (n = 3,547). Consequently, the analysis encompassed 3,019 participants from the NHANES dataset from the 2007–2012. study. Fig 1 shows a detailed and fully comprehensive representation of the inclusion and exclusion criteria.

Baseline characteristics

Table 1 presents an exhaustive overview with the results of the weighted analysis elucidating the foundational characteristics of the entire subject pool segregated by their LE8 scores (the demarcation points situated at 50 and 80). The results indicate a total participant pool of 3,019 individuals, with 386 in the low category (< 50), 2,023 in the moderate group (50–80), and 610 in the high category (≥ 80). The mean age of the participants was 47.12 (± 16.44) years and 1,530 individuals (51.4%) identified as women. Individuals with a high LE8 score were more likely to be younger, female, of Non-Hispanic White background, have an educational attainment exceeding 9th grade, be married, have a high income, be mild alcohol users, and not have CVD. Additionally, they exhibited lower levels of Alt, Ast, uric acid, creatinine, and urine iodine; did not use prescription medications; and had no thyroid diseases. The results of the post-hoc test provided a clear illustration of the pairwise comparisons between the groups (see S2–S7 Figs). Visualization of the eight TH graphs indicated that TSH, Tg, TGAb, and TPOAb followed a non-normal distribution (see S1 Fig). Therefore, to facilitate the analysis, a logarithmic transformation using the natural base (e) was applied to all TH variables.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Demographic profiles and initial participant attributes in the current research inquiry.

https://doi.org/10.1371/journal.pone.0329194.t001

Relationship between thyroid hormone and Life’s Essential 8

Weighted univariate analysis revealed associations between LE8 and several variables including age, sex, race/ethnicity, marital status, education level, PIR, alcohol use, Alt level, Ast level, uric acid level, creatinine level, urine iodine level, CVD, prescription drugs affecting thyroid function, and thyroid disease (see S3 Table).

The relationship between TH (after log transformation) and LE8 is presented in Table 2. In the non-adjusted model, ln(TT4) and ln(Tg) negatively correlated with LE8. However, ln(FT3) (β (95%CI): −6.31(−12.13, −0.49), p = 0.035), ln(TT3) (β (95%CI): −5.18(−8.22, −2.15), p = 0.002), and ln(Tg) (β (95%CI): −0.98(−1.74, −0.22), p = 0.014) were negatively correlated with LE8, after full adjustment for all covariates. The substantially elevated E-values suggest the robustness of these associations. The residuals of the linear regression model approximately followed a normal distribution (see S8–S15 Figs).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Association between thyroid hormone and life’s essential 8 in different models.

https://doi.org/10.1371/journal.pone.0329194.t002

To visualize the relationship between the TH levels (after log transformation) and LE8 scores, a nonlinearity test was conducted. For variables exhibiting a nonlinear relationship, the RCS was fitted as shown in Fig 2. Notably, nonlinear relationships were observed between ln(Tg), ln(TgAb), ln(TPOAb), and LE8. The relationships among ln(Tg), ln(TgAb), ln(TPOAb), and LE8 were approximately inverse L-shaped.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Restricted cubic spline plots between ln(Tg), ln(TgAb), ln(TPOAb), and LE8, with vertical lines indicating the cut points.

Abbreviation: Tg, thyroglobulin; TgAb, thyroglobulin antibody; TPOAb, thyroid peroxidase antibody. The model was fully adjusted (corrected for age, sex, race/ethnicity, urine iodine, education, marital status, PIR, CVD, creatinine, uric acid, Alt, AST, alcohol use, prescription drugs affecting thyroid function, and thyroid disease). The extreme values (2.5% of each tail) were trimmed to preserve 95% of the participants.

https://doi.org/10.1371/journal.pone.0329194.g002

Sensitivity analysis

To assess the robustness of the relationship between TH levels (after log transformation) and LE8 scores, subgroup analyses were performed to examine the association within the subgroups stratified by normal and abnormal hormone levels (Table 3). In the subgroup with normal TT3 levels, we found a significant negative correlation between ln(TT3) and LE8 (β (95% CI): −7.73 (−15.09, −0.37), p = 0.04). In the subgroup with normal TT4 levels, ln(TT4) showed a significant positive correlation with LE8 (β (95% CI): −7.59 (−14.32, −0.86), p = 0.03). Notably, in the subgroup with normal Tg levels, ln(Tg) exhibited a negative correlation with LE8 (β (95% CI): −1.53 (−2.85, −0.20), p = 0.03). The substantially elevated E-values suggest the strength of these associations. Linear regression analyses conducted after excluding populations with cardiovascular or thyroid diseases revealed a similar negative correlation, indicating the validity of the results (see S4 Table and S16–S17 Figs). Similarly, linear regression analyses using different missing data handling methods also demonstrated consistent negative correlations, further supporting the robustness of our findings (see S5 Table and S18–S20 Figs). Analyses incorporating MEC weights showed a comparable negative correlation between ln(FT3), ln(TT3), ln(Tg), and LE8 as well as an additional negative correlation between ln(TT4) and LE8. Moreover, ln(FT4) and ln(TPOAb) were positively correlated with LE8 levels. The explanation for this result may be that MEC weight analyses retained a larger sample size, with nearly double the number of participants in the SAF weight analyses.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Association between different level groups of thyroid hormone and life’s essential 8.

https://doi.org/10.1371/journal.pone.0329194.t003