The National Health and Nutrition Examination Survey (NHANES) is a series of health-related research aimed at determining non-institutionalized Americans' health and nutritional status. As a representative sample, a multistage, stratified probability strategy was used to select survey participants (36). This cross-sectional study used the data from 2011–2012, 2013–2014, and 2015–2016 cycles from the NHANES, as the interesting trace metal was only examined in these three survey waves. Demographic, socioeconomic and health-related information were collected through questionnaires, physical examinations, and laboratory tests. Health interviews were conducted at participants' homes, while thorough physical examinations, including blood sample collection, were carried out at the Mobile Examination Center (MEC). The collected serum specimens were then tested at the National Center for Environmental Health's Division of Laboratory Sciences of the Centers for Disease Control and Prevention (37).

The NHANES was authorized by the National Center for Health Statistics Ethics Review Board (https://www.cdc.gov/nchs/nhanes/irba98.htm). Before participating, all participants completed written informed consent forms. The secondary analysis did not require additional Institutional Review Board approval (38). The NHANES data are accessible through the NHANES website (http://www.cdc.gov/nchs/nhanes.htm; accessed on 19 Oct 2023).

Our study's participants were above the age of 20 and had completed an interview and evaluation at a MEC.

We excluded pregnant women or individuals with missing data on serum zinc, fasting plasma glucose (FPG), triglyceride (TG) or covariates. And we excluded participants with extreme energy intake, consuming <500 or >5,000 kcal per day.

Serum zinc was detected at the Environmental Health Sciences Laboratory of the CDC National Center for Environmental Health using the inductively coupled plasma dynamic reaction cell mass spectrometry following extensive quality control procedures. The lower limit of detection (LLOD) for serum zinc was 2.9 μg/dL, and all the data was above the LLOD for all tests. In the multivariable linear models, log-transformed serum zinc was categorized into quartiles: Q1 (1.69–1.89 μg/dL; n = 397), Q2 (1.90–1.93 μg/dL; n = 408), Q3 (1.94–1.97 μg/dL; n = 401), Q4 (1.98–2.37 μg/dL; n = 404).

Triglyceride-glucose index (TyG) was calculated using the formula Ln [fasting TG (mg/dL) × fasting plasma glucose (FPG; mg/dL)/2] (18). Blood samples were taken in the morning after fasting overnight to measure the levels of TG and glucose in the blood. The concentration of TG and FPG was measured using an automatic biochemistry analyzer. The serum TG levels were determined using a Roche Cobas 6000 chemistry analyzer and a Roche Modular P chemistry analyzer. A Roche/Hitachi Cobas C 501 chemistry analyzer was used to measure FPG using the hexokinase-mediated reaction.

The covariates considered in this study consisted of sociodemographic, behavioral, health characteristics and laboratory data deemed a priori as potential confounders.

Sociodemographic variables consisted of age groups (20–59 years and ≥60 years) (39–41), gender (female and male), race/ethnicity (non-Hispanic White, non-Hispanic Black, Mexican American, or other races), education level (<9, 9–12, or >12 years), marital status (married, living with a partner, or living alone). According to a US government report (42), family income was categorized into three groups by the poverty income ratio (PIR): low (PIR ≤ 1.3), medium (PIR > 1.3–3.5), and high (PIR > 3.5).

Behavioral characteristics comprised smoking status, drinking status, and physical activity. According to previous literature definitions (43), smoking status was classified into three categories: never smokers (participants who had smoked fewer than 100 cigarettes), current smokers, and former smokers (those who had quit smoking after smoking more than 100 cigarettes). Furthermore, individuals who consumed at least 12 alcoholic drinks per year throughout their lifetime were classified as drinkers (37). Physical activity was categorized as sedentary, moderate (involving at least 10 min of movement within the past 30 days, resulting in light sweating or a mild to moderate increase in breathing or heart rate), and vigorous (involving at least 10 min of activity within the past 30 days, resulting in profuse sweating or a significant increase in breathing or heart rate) (43).

Health factors included body mass index (BMI), trouble sleeping, hypertension (no or yes) (44), diabetes (no or yes) (45) and failing kidneys (no or yes) (46). BMI was computed using a standardized technique which is weight (kg) divided by height (m) and divided into four categories with cut-off values of 18.5, 25, and 30 kg/m² (underweight, normal, overweight, and obese) (47). Hypertension was diagnosed based on a self-reported physician diagnosis (a positive response to “Have you been diagnosed with hypertension?”), and/or recent use of an antihypertensive agent (a positive response to “Are you currently taking any antihypertensive drugs to treat or control your blood pressure?”), and/or a systolic blood pressure/diastolic blood pressure ≥140/90 mmHg (44). Diabetes cases were defined as participants who fulfilled the inclusion criteria: (1) FPG ≥126 mg/dL, (2) 2-h plasma glucose ≥200 mg/dL on an oral glucose tolerance test (OGTT), (3) HbA1c ≥6.5%, and (4) current use of insulin or diabetes pills to lower blood glucose levels, or a self-report questionnaire that indicates a previously diagnosed of T2DM by a physician (45). Failing kidneys was determined for participants who positively responded to the question has he/she ever been told by a doctor or other health professional that had weak or failing kidneys (excluding kidney stones, bladder infections, or incontinence) (46). A dietary recall interview preceded and interview including total energy intake.

Laboratory data including HbA1c, high density lipoprotein cholesterol (HDL-C), triglyceride, creatinine, total cholesterol, FPG and uric acid.

Statistical analyses were performed using the statistical software programs R (The R Foundation) and Free Statistics software version 1.9.2 (Beijing Free Clinical Medical Technology Co., Ltd.) (48). All statistical tests were two-sided, and significance was considered at P < 0.05. Analyses were conducted according to the Centers for Disease Control and Prevention (CDC) guidelines for the analysis of NHANES data. As the sample size was determined based solely on the available data, no a priori statistical power estimates were conducted. We used fasting subsample MEC weights for the weighted analysis. For the combined analyses of NHANES 2011–2016 data, a 6-year fasting subsample MEC weights (WTSAF2YR) set was used, stratum (SDMVSTRA), and primary sampling units (SDMVPSU) were taken into account for the complex survey design (48).

Categorical data were expressed as unweighted numbers (weighted percentages), whereas continuous data were expressed as means (standard deviation, SD). One-way analyses of variance (continuous variables) and chi-square tests (categorical variables) were used to compare differences between the groups. To analyze the association between serum zinc and TyG, we used univariate and multivariable linear regression models. The models integrated regression coefficients (β) and 95% confidence intervals (CI) while controlling for significant covariates. Log-transformed serum zinc was considered a continuous variable after undergoing a logarithm 10 transformation. The selection of confounding variables was guided by clinical relevance, existing scientific literature, the significance of covariates in univariate analysis, their correlation with the outcomes of interest, or a change in effect estimate exceeding 10%. In multivariable linear regression, we showed (1) unadjusted models, (2) model 1 adjusted covariates with a change in effect estimate exceeding 10%, including sex, BMI, HDL-C, TC, uric acid, diabetes and trouble sleeping, (3) model 2 adjusted for variables from model 1 plus covariates that P values were < 0.05 in the univariate analysis, including age, race and ethnicity, educational level, physical activity, smoking status, HbA1c, failing kidneys, hypertension, and (4) model 3 adjusted for variables from model 2 plus covariates that on the basis of previous findings and clinical constraints, including marital status, PIR, drinking status, creatinine, total energy intake (17, 49).

In addition, we examined possible dose-response relationships between log-transformed serum zinc and TyG after adjusting variables in model 3, restricted cubic spline (RCS) regression was performed with 4 knots at the 5th, 35th, 65th, and 95th percentiles of the distribution (43).

We used a two-piece-wise linear regression model with smoothing to analyze the association threshold between log-transformed serum zinc and TyG after adjusting the variables in model 3, the likelihood-ratio test and the bootstrap resampling method were used in determining inflection points, in addition to conducting separate analyses for the age groups <60 years and age ≥60 years.

Furthermore, we compared potential modifications of the relationship between log-transformed serum zinc and TyG in the groups with age <60 years and age ≥60 years. The heterogeneity in the subgroup were assessed using multivariable linear regression and interactions between the subgroup and log-transformed serum zinc were examined through likelihood ratio testing.

In total, 29 902 participants completed the interview, of whom 12,854 participants were < 20 years old. We excluded pregnant women (n = 192), those missing data on TG (n = 9,680), those missing data on FPG (n = 9), those missing data on serum zinc (n = 4,841), or those with covariates (n = 636). And we excluded participants with extreme energy intake, consuming < 500 or >5,000 kcal per day (n = 80). Ultimately, this cross-sectional study included 1 610 participants from the NHANES between 2011 and 2016 in the analysis. The detailed inclusion and exclusion process is shown in Figure 1. The figure delineates the study's design, sampling, and exclusion procedures. This study included American adults (aged ≥20 years) who participated in the 2011–2012, 2013–2014, and 2015–2016 cycles of NHANES, as the serum zinc is only assessed during these survey waves.

The Supplementary Table S1 describes the baseline characteristics of the excluded and included participants. Table 1 illustrates the baseline characteristics of all subjects based on their age, categorized into the age < 60 group and age ≥60 group. The average age of the study participants was 49.6 (17.6) years, and 839 (52.1%) individuals were male. In comparison to the age < 60 group, the age ≥60 group exhibited elevated levels of HbA1c, HDL-C, creatinine, FPG, uric acid, and TyG. Furthermore, they demonstrated a higher prevalence of hypertension, diabetes, renal impairment, and sedentary physical activity, along with a lower prevalence of current smoking, current alcohol use, and lower educational attainment. The log-transformed serum zinc levels in the age ≥60 group were comparable to those in the age < 60 group (P = 0.29).

The univariate analysis demonstrated that age, sex, race/ethnicity, education level, BMI, physical activity, smoking status, trouble sleeping, HbA1c, HDL-C, TC, uric acid, failing kidneys, hypertension, diabetes, and log-transformed serum zinc were associated with TyG (Table 2).

The findings of the multivariable linear regression analysis are shown in Table 3. In the unadjusted model, there was a positive association of log-transformed serum zinc with TyG (β = 1.16, 95% CI = 0.55–1.78). Results were similar after adjusting for sex, BMI, HDL-C, TC, uric acid, diabetes and trouble sleeping (β = 0.50, 95% CI = 0.04–0.96). After adjusting for other possible confounders, including age, race and ethnicity, educational level, physical activity, smoking status, HbA1c, failing kidneys, hypertension, marital status, PIR, alcohol, creatinine, and total daily energy intake, the positive association remained significant (β = 0.50, 95% CI = 0.08–0.93; P < 0.05). When log-transformed serum zinc was analyzed using quartiles, the association between TyG was consistent across all models, indicating their robustness (Table 3). The individuals with quartile 3 (Q3) group of log-transformed serum zinc (1.94–1.97 μg/dL) were used as the baseline reference, those with Q4 group of log-transformed serum zinc (1.98–2.37 μg/dL) had an adjusted β for TyG of 0.092 (95% CI 0.013–0.17, P < 0.05; Table 3) after adjusting for the variables in Model 3.

Accordingly, in the fully adjusted analyses, the restricted cubic spline (RCS) regression (Figure 2) indicated a non-linear relationship between log-transformed serum zinc and TyG levels in a J-shaped manner (P for nonlinearity = 0.019) (A). A segmented regression model was employed to delineate the intervals and calculate threshold effects, with an inflection point at ~1.94 μg/dL. The results are presented in Table 4. When the log-transformed serum zinc was < 1.94 μg/dL, the estimated dose-response curve exhibited a consistent horizontal trend, and the relationships between the log-transformed serum zinc and TyG was not significant (P > 0.05). Likewise, the TyG exhibited an increase with rising log-transformed serum zinc after the inflection point, with a correlation coefficient (β) of 0.96 (95% CI: 0.25–1.66) after adjusting for the variables in Model 3 (Table 4). Furthermore, within the age <60 years group, the TyG demonstrated an increase with escalating log-transformed serum zinc after the inflection point, with a correlation coefficient (β) of 1.21 (95% CI: 0.49–1.94) after adjusting for the variables in Model 3. Conversely, within the age ≥60 years group, the TyG exhibited an increase with increasing log-transformed serum zinc prior to the inflection point, with a correlation coefficient (β) of 1.98 (95% CI: 0.62–3.35) after adjusting for the variables in Model 3 (Table 5).

And the RCS analysis was also applied to investigated the dose-response association between log-transformed serum zinc level and TyG with age ≥60 years and age <60 years. Figure 2 shows that in the fully adjusted analyses log-transformed serum zinc was related to level of TyG in a J shaped nonlinear manner (P for nonlinearity = 0.014) in age <60 years group (B), but in a inverted-L shaped nonlinear manner (P for nonlinearity < 0.001) in age ≥60 years group (C).

A stratified analysis was conducted according to age, gender, BMI and diabetes to determine whether there were differential effects in the association between log-transformed serum zinc and TyG in American adults (aged ≥20 years). The results demonstrated that no statistically significant interactions were identified in any of the subgroups after stratification by gender, BMI and diabetes in model 3 (Supplementary Table S2). However, when log-transformed serum zinc was transformed into a categorical variable, an interaction between log-transformed serum zinc and TyG was observed in individuals aged <60 years and those aged ≥60 years (P value for the likelihood ratio test for the interaction was P = 0.017; Table 6). The individuals with the log-transformed serum zinc quartile 3 (Q3) group (1.94–1.97 μg/dL) were used as the baseline reference. In the group of individuals aged <60 years, those with the Q4 group of log-transformed serum zinc (1.98–2.37 μg/dL) exhibited an adjusted β for TyG of 0.096 (95% CI 0.009–0.18; P < 0.05) in comparison to the Q3 group, after adjusting for the variables in Model 3. In the group of individuals aged ≥60 years, those with the Q1 group of log-transformed serum zinc levels (1.69–1.89 μg/dL) exhibited an adjusted β for TyG of −0.169 (95% CI −0.280 to −0.058; P < 0.05) in comparison to the Q3 group.