NHANES was an ongoing cross-sectional study done by the Centers for Disease Control and Prevention (CDC) to measure the health and nutritional status of the civilian, non-institutionalized US population. The NHANES study protocols were approved by the National Center for Health Statistics (NCHS) Ethics Review Board. All study participants signed written informed consent forms. The NHANES datasets are available on DataDryad (https://doi.org/10.5061/dryad.d5h62).

Data from the 1999–2000 and 2001–2002 cycles of NHANES were pooled because of available data of LTL in these two data collection cycles. In total, 7827 participants with LTL data were included. Participants with missing triglyceride data (n=7), glucose data (n=2), or using lipid-lowering drugs (n=551) were excluded. Considering the possibility of IR bias among the population with diabetes, we further excluded participants with diabetes (n=778). Finally, a total of 6489 participants were included in this study. The selection process of the study analytic sample is detailed in Figure 1 .

All information was collected by trained researchers. Data included demographics characteristics (sex, age, race/ethnicity, poverty to income ratio), health-related behavior (smoking, alcohol consumption, physical activity), education, history of diseases (hypertension, diabetes), using medications, anthropometric measurements (height, weight, waist circumference, blood pressure) as well as laboratory data [fasting blood glucose (FBG), total cholesterol (TC), TG, high-density lipoprotein-cholesterol (HDL-C]), serum uric acid (SUA), serum creatinine, and C-reactive protein (CRP)]. Fasting venous blood samples were collected. Laboratory data were determined using automatic clinical analyzers. The details of the covariate measurement process are available at http://cdc.gov/nchs/nhanes.

The following formula was used to compute the body mass index (BMI): BMI=weight (kg)/height (m²). Hypertension was defined as a mean systolic blood pressure (SBP) ≥140 mmHg or a mean diastolic blood pressure (DBP) ≥90 mmHg or a positive response to the questions: “Have you been told on two or more different visits that you had hypertension?” or “Are you taking prescribed medicine to lower blood pressure” (27). Diabetes was defined as a fasting glucose concentration >126 mg/dL or a self-reported medical diagnosis of diabetes. Estimated glomerular filtration rate (eGFR) was calculated via using the newly developed Chronic Kidney Disease Epidemiology Collaboration equation (28): eGFR = 141 × min(Scr/κ, 1)α × max(Scr/κ, 1)-1.209 × 0.993Age × 1.018 [if woman] × 1.159 [if black], where Scr is serum creatinine, κ is 0.7 for women and 0.9 for men, α is -0.329 for women and -0.411 for men, min indicates the minimum of Scr/κ or 1, and max indicates the maximum of Scr/κ or 1.

In our study, the exposure variable was the TyG index and HOMA-IR. The TyG index was calculated by the formula ln [fasting TG (mg/dL) ×fasting glucose (mg/dL)/2] (29). TyG index was divided into 3 tertiles: <8.27 (tertile 1), 8.27-8.77 (tertile 2), and ≥8.77 (tertile 3). HOMA-index was calculated as fasting plasma glucose (mmol/L) × fasting serum insulin (mU/mL)/22.5. The outcome variable was LTL. Aliquots of purified DNA were provided by the laboratory of the CDC. DNA was isolated from whole blood using the Puregene (D-50K) kit protocol (Gentra Systems, Inc., Minneapolis, MN) and stored at -80°C. One milliliter of extraction buffer was added to 10⁵ -10⁸ cells in whole blood, then centrifuged for 10 min at 13,000 g at 4-25°C. There was 500 μL binding buffer added to the spin column, centrifuged at 12,000 g for 30 s. There was 10 to 200 μL TE buffer added and incubated at room temperature for 1 min. The buffer in the microcentrifuge tube contains the DNA. DNA concentrations were measured by running aliquots on 0.8% agarose gel and by reading absorbance at 260 nm with a spectrophotometer. LTL was measured relative to standard reference DNA (T/S ratio), which was evaluated with samples from the human diploid fibroblast cell line IMR90. Using real-time quantitative polymerase chain reaction (qPCR), the relative ratio of telomere repeat copy number to single-copy gene copy number (T/S ratio) was determined in the blood leukocytes. Purified DNA samples were diluted in 96-well microtiter source plates to ∼1.75 ng/µl in 10 mM Tris–HCl, 0.1 mM EDTA, pH 7.5 (TE–4, final volume 300 µl/well), heated to 95°C for 5 min in a thermal cycler, quick chilled by transfer to an ice/water bath for 5 min, centrifuged briefly at 730 g, sealed with adhesive aluminum foil, and stored at 4°C until the time of assay. For each individual in whom the T/S ratio was assayed, three identical 20-µl aliquots of the DNA sample (35 ng/aliquot) were added to plate 1 and another three aliquots were added to the same well positions in plate 2. For each standard curve, one reference DNA sample was diluted serially in TE–4 by ∼1.68-fold per dilution to produce five concentrations of DNA ranging from 0.63 to 5 ng/µl, which were then distributed in 20-µl aliquots to the standard curve wells on each plate. The final concentrations of reagents in the PCR were 150 nM 6-ROX and 0.2× Sybr Green I (Molecular Probes), 15 mM Tris–HCl pH 8.0, 50 mM KCl, 2 mM MgCl2, 0.2 mM each dNTP, 5 mM DTT, 1% DMSO, and 1.25 U AmpliTaq Gold DNA polymerase (Applied Biosystems). The final telomere primer concentrations were: tel 1, 270 nM; tel 2, 900 nM. To standardize the variability between the two tests, each panel included eight control DNA samples. The sample characteristics were unknown to the LTL assay lab. The coefficient of variance between assays was 6.5%. The equation for conversion from the T/S ratio to bp pairs is 3274 + 2413 × (T/S) (30). The full details of this standard protocol have been described elsewhere (30–32). The CDC reviewed the quality of TL measurement before releasing the data to the public.

Sample weights were used for analyses to account for the complex survey design and non-response of NHANES (30). Continuous and categorical variables were expressed as means ± standard deviation (SD) or proportions and were compared using weighted linear regression model or weighted χ2 test according to different TyG index groups (tertiles), respectively. The distribution of LTL levels according to sex, age, smoking status, and BMI were also described. The association between the TyG index and LTL was examined as a continuous variable per one unite increase and also as a categorical variable using tertiles with tertile 1 (T1) as the reference group. Multivariate linear regression analysis was applied to obtain beta coefficient (β) and 95% confidence interval (CI) for the association of the TyG index with LTL. Three models were constructed: model 1, adjusted for age; model 2, adjusted for age, sex, education, smoking status, alcohol consumption, physical activity, poverty to income ratio, and BMI; model 3, adjusted for age, sex, education, smoking status, alcohol consumption, physical activity, BMI, waist circumference, SBP, DBP, history of hypertension, TC, SUA, eGFR, and CRP. Variables considered as confounders based on existing literature, clinical judgment, and statistical significance in the univariate analysis were included. We further explored a dose-response relation between the TyG index by a decile approach [≤ 7.81 (reference), 7.81-8.04, 8.04-8.21, 8.21-8.37, 8.37-8.52, 8.52-8.68, 8.68-8.83, 8.83-9.03, 9.03-9.33, and >9.33] and LTL.

We conducted the sensitivity analysis to confirm the accuracy of our data analysis. If covariates with missing data were omitted from data analyses, we imputed missing values using the reference category for qualitative variables or the median value for quantitative variables. The imputed datasets in Supplementary Table S1 were also used to get the main results. Second, the effects of the following variables on the relationship between the TyG index and LTL were investigated: sex (women vs. men), age (<45 vs. ≥45 years), BMI (<25 vs. ≥25 kg/m²), current smoking (yes vs. no), education (<high school vs. ≥high school), physical activity (sedentary, low, moderate vs. high), hypertension (yes vs. no), total cholesterol (<202 [median] vs. ≥202 mg/dL), and eGFR (<60 vs. ≥60 mL/min per 1.73 m²). Potential interactions were examined by including the interaction terms into those linear regression models with the greatest number of confounding variables.

All P values were two-sided with a significance level of < 0.05. Statistical analyses were performed using the statistical package R (http://www.R-project.org, The R Foundation) and Empower (R) (www.empowerstats.com; X&Y Solutions, Inc., Boston, MA). The figures were drawn using origin.

A total of 6489 adults without diabetes were included in our study, with a mean age of 46.67 ± 18.55 years old. There were 3060 (47.16%) participants that were men. The mean (SD) TyG index was 8.6 (0.6) (median 8.5). The mean (SD) LTL was 5796.1 (691.4) bp (median 5722.3 bp). The weighted characteristics of the participants by TyG index tertiles are presented in Table 1 . There were significant differences between TyG index tertiles. Participants with the highest TyG index in T3 (≥ 8.77) were more likely to be older, to be men, to be current smokers, to have lower educational levels and physical activity levels, to have a high rate of hypertension, to have higher values in BMI, waist circumference, SBP, DBP, FBG, TC, TG, SUA, and CRP, and to have lower values in poverty to income ratio, alcohol consumption, eGFR, and LTL than those of the other groups (all P < 0.01).

Figure 2 shows the distribution of LTL levels according to sex, age, smoking status, and BMI. As shown in Figure 2A , women had higher LTL levels than that in men (5836.1 vs 5751.3 bp, P < 0.01). Participants with increasing level of age had lower LTL values (P < 0.001) ( Figure 2B ). The mean LTL values in never smoker, former smoker, and current smoker were 5834.9 ± 736.7, 5852.3 ± 640.0, and 5660.6 ± 617.2 bp, respectively ( Figure 2C ). Current smoker had lower LTL levels than that in other groups. There was no statistical difference in LTL levels between never smoker and former smoker (P = 0.431). The mean LTL values for BMI of < 25, 25-30, and ≥ 30 kg/m² were 5868.9 ± 801.2, 5773.8 ± 630.6, and 5758.1 ± 616.3 bp, respectively ( Figure 2D ). Participants with normal weight had higher LTL levels than that in overweight/obese subjects (P < 0.001). There was no statistical difference in LTL levels between overweight and obese people (P = 0.421).

The associations between various baseline characteristics and LTL levels using univariate analysis are shown in Supplementary Table S1 . The LTL levels were significantly associated with sex, BMI, waist circumference, SBP, DBP, history of hypertension, education levels, smoking status, alcohol consumption, physical activity, TC, CRP, SUA, eGFR, and the TyG index, but not poverty to income ratio.

The association between the TyG index and LTL in the absence of diabetes using multivariate linear regression analyses is further illustrated in Table 2 . We constructed 3 models for analyzing the role of the TyG index in LTL levels. After adjustment for the important confounders, the positive association between the TyG index and LTL was still found in all models. In model 3, for each 1 unit increase in the TyG index, the change in LTL was -72.62 bp (95%CI: -107.36~-37.87). The TyG index was converted from a continuous variable to a categorical variable (tertiles). Compared with those with TyG index < 8.27, subjects with TyG index ≥ 8.77 had statistically significantly shorter LTL (β = -95.98, 95%CI: -145.08~-46.89, P<0.001), while no association was found between middle TyG index tertile (8.27-8.77) and LTL (β = -4.31, 95%CI: -48.12~39.49, P=0.847). Similarly, a significantly shorter LTL (β = -93.33, 95%CI: -134.33~-52.32) was found in participants in tertile 3 (≥ 8.77) compared with participants in tertiles 1-2 (< 8.77). The results did not change qualitatively when we reanalyzed the association between TyG index and LTL using imputation data ( Supplementary Table S2 ). It seems that the association between the TyG index and LTL is likely to be nonlinear. TyG index ≥ 8.77 was inversely associated with LTL, supporting a threshold effect of the TyG index on LTL among U.S. adults without diabetes.

We further explored a dose-response relation between the TyG index by a decile approach [≤ 7.81 (reference), 7.81-8.04, 8.04-8.21, 8.21-8.37, 8.37-8.52, 8.52-8.68, 8.68-8.83, 8.83-9.03, 9.03-9.33, and >9.33] and LTL ( Figure 3 ). We constructed 2 models: model 1, only adjusted for age; model 2, adjusted for age, sex, education, smoking status, alcohol consumption, physical activity, BMI, waist circumference, SBP, DBP, history of hypertension, TC, SUA, eGFR, and CRP. Compared to TyG index ≤ 7.81, 5 subgroups (TyG index 7.81-8.04, 8.04-8.21, 8.21-8.37, 8.37-8.52, and 8.52-8.68) did not show significant effect on LTL, while there was a significantly shorter LTL for participants with TyG index > 8.68. The results also suggest that there was a threshold effect of the TyG index on LTL.

We further analyzed the association of HOMA-IR with LTL. As shown in Supplementary Table 3S , after adjustment for important confounders, HOMA-IR was not associated with LTL, suggesting that the TyG index was better associated with LTL than HOMA-IR.