National Health and Nutrition Examination Survey, coordinated through the National Center for Health Statistics (NCHS) (18), aims to evaluate the health and nutritional status of the United States population. Its stratified multistage probability sampling methodology ensures that each survey cycle produces a sample representative of the whole country (19). The survey is divided into two parts: a household interview and a medical examination at a Mobile Examination Center (MEC). Ethical oversight for NHANES was administered by the NCHS and all individuals providing informed consent.

This investigation utilized data spanning ten NHANES survey cycles (1999–2018), with rigorous exclusion protocols applied to ensure an accurate sample size. Exclusion criteria for the analysis included participants who did not meet the inclusion criteria for postmenopausal women (n = 45,419) and missing data for the TyG index and its related adiposity metrics (n = 5,360). The analysis framework included 4,302 eligible participants, as depicted in Figure 1.

The TyG index and its associated indices in the study were defined and calculated as follows: BMI was determined by weight (kg)/height squared (m²), while WC was the waist circumference (cm), and WHtR was determined by WC (cm)/height (cm). TyG index = ln [fasting triglyceride (mg/dL) × fasting glucose (mg/dL)/2]; TyG-WC = TyG × WC; TyG-WHtR = TyG × WHtR; TyG-BMI = TyG × BMI.

To determine menopausal status in this study, we used a two-step questionnaire from the NHANES Reproductive Health Questionnaire. First, participants were asked whether they had experienced at least one menstrual period in the past 12 months. They were subsequently inquired about the reason for not having menstruated in the previous 12 months. Women were classified as postmenopausal if they responded “no” to the first question and indicated either “menopause” or “hysterectomy” in response to the second question.

Hypertension was defined as meeting any of the subsequent conditions: (1) an average systolic blood pressure of ≥ 140 mmHg and/or an average diastolic blood pressure of ≥ 90 mmHg according to the guidelines established by the International Society of Hypertension; (2) self-reported physician-diagnosed hypertension; or (3) current use of prescription antihypertensive medications.

In our analysis, we controlled for several potential confounders, guided by existing literature and clinical expertise, including sociodemographic factors such as age, racial/ethnic categorization, marital status, educational attainment, and family income; behavioral modifiers including smoking, drinking habits, and physical activity; Assayed variables included high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), total cholesterol (TC), glycosylated hemoglobin (HbA1c), and uric acid; and finally, characteristics of menopausal metabolic consisted of menopausal age and hormone replacement therapy. Marital status was divided into two groups: married or living with a partner and living alone. Race categories included Mexican American, Non-Hispanic Black, Non-Hispanic White, other Hispanic, and other races (20). Educational attainment was categorized into three levels: less than high school, high school or equivalent, and more than high school (21). Family income was stratified into three groups by the poverty income ratio (PIR): below 1.3, between 1.3 and 3.5, and above 3.5. Physical activity was determined by the amount of time (in minutes) spent in different activities per week. Smoking status was categorized as never smoker (smoked < 100 cigarettes), current smoker, and former smoker (quit smoking after smoking ≥ 100 cigarettes). Drinking status was divided into never drinker (lifetime alcohol consumption < 12 drinks), former drinker (≥ 12 drinks in a year but abstained in the past year, or no drinks last year but ≥ 12 lifetime drinks), and current drinker. Menopausal age was considered as a continuous variable in the Cox regression analysis. The determination of hormone replacement therapy (HRT) was derived from participants’ questionnaire responses indicating past use of female hormones, including estrogen and progesterone.

This study was a secondary analysis of publicly available datasets. Baseline characteristics of the study participants were stratified by hypertension status and compared using appropriate statistical tests. Due to the percentages of missing values being found to be less than 20%, multiple imputations were used to impute missing data for the covariates, resulting in the generation of five independent datasets that were subsequently evaluated collectively; detailed information regarding multiple imputation is provided in the supplementary methods. Descriptive statistics were used to describe the baseline characteristics; categorical data were expressed as frequencies and percentages, while continuous data were expressed as means ± SD and medians (interquartile range), taking into account any skewed data distribution. Normality was assessed using the Shapiro-Wilk test to validate the selected statistical methods. Chi-squared analyses were used for categorical variables, and analysis of variance or Mann-Whitney U tests were used for continuous variables.

To investigate the independent and combined effects of the TyG index and its integration with adiposity metrics (TyG-BMI, TyG-WC, and TyG-WHtR) on the prevalence of hypertension among postmenopausal women, we implemented multivariable logistic regression modeling. We standardized (Z-score) the TyG index and its obesity indicators, then included them in the multivariable logistic analyses. Additionally, we categorized the continuous variables into tertiles, with the first tertile of the TyG index and its obesity indicators serving as the reference. Odds ratios (ORs) with 95% confidence intervals (CIs) were computed per standard deviation (SD) increment and across tertiles (T1–T3) of the TyG index and its associated indices. Four hierarchical models were constructed to adjust for confounders: Model 1 was the crude model, not accounting for any covariates. Model 2 was adjusted for age, race/ethnicity, marital status, education level, PIR, and NHANES cycles. Model 3 was additionally adjusted for smoking, drinking status, and physical activity. Model 4 included all preceding variables augmented along with HDL-C, LDL-C, HbA1c, uric acid, menopausal age, and hormone replacement therapy. When analyzing dose-response gradients across tertile divisions, median values within each stratified subgroup were operationalized as continuous parameters.

To investigate the potential curvilinear associations, restricted cubic spline (RCS) modeling with three knots was implemented, complemented by likelihood ratio tests examining the goodness-of-fit.

The classification accuracy of TyG-related indices was assessed through the implementation of a receiver operating characteristic (ROC) framework. The discriminative ability of these indices was quantified by calculating the area under the curve (AUC) metrics, with 1,000 bootstrap-resampled iterations employed to enhance the robustness of the estimates. Intergroup comparisons were conducted via two-sample independent t-test procedures.

To assess the validity of TyG-related indices, we conducted sensitivity analyses using multivariate regression and ROC curves to compare their associations with hypertension relative to HOMA-IR, a well-established indicator of insulin resistance.

Analyses were performed with R (version 4.3.1) and Free Statistics software (version 2.1, Beijing, China¹). P < 0.05 (two-sided) was defined as statistically significant.

The current study included 4,302 postmenopausal women (mean age 64.1 ± 11.6 years), of whom 2,826 (65.7%) were diagnosed with hypertension. As delineated in Table 1, marked epidemiological and phenotypic divergences were identified through rigorous parametric comparisons between the hypertensive and non-hypertensive groups. Compared with non-hypertensive participants, those with hypertension were older and disproportionately identified as Non-Hispanic Black, less educated, lower family income, and more prone to residing alone. Those who developed hypertension had lower rates of current smoking and alcohol consumption, along with significantly less weekly physical activity; had higher HbA1c, uric acid levels, TyG index, and TyG-related indices; had a slightly later median age at menopause; and had no difference in hormone replacement therapy utilization.

Covariate-adjusted logistic regression results are presented in Tables 2, 3. After full adjustment (Model 4), each SD increase in the TyG index (OR = 1.33, 95% CI: 1.18–1.49), TyG-BMI (OR = 1.42, 95% CI: 1.28–1.57), TyG-WC (OR = 1.42, 95% CI: 1.28–1.58), and TyG-WHtR (OR = 1.49, 95% CI: 1.34–1.66) remained significantly associated with elevated hypertension risk (all P < 0.001). Quartile analyses revealed graded relationships: compared to T1, T3 of TyG-WHtR exhibited the highest risk (OR = 2.13, 95% CI: 1.69–2.68), followed by TyG-WC (OR = 2.11, 95% CI: 1.68–2.65), TyG-BMI (OR = 1.95, 95% CI: 1.56–2.44), and TyG index (OR = 1.52, 95% CI: 1.20–1.92). All trend P-values were < 0.001, indicating robust linear associations. The RCS curve delineated a J-shaped curve association between TyG-WHtR (SD) and hypertension risk (P for non-linearity = 0.026) (Figure 2D), whereas TyG index (SD), TyG-BMI (SD), and TyG-WC (SD) displayed near-linear trends (P for non-linearity > 0.05) (Figures 2A–C).

Receiver operating characteristic analysis evaluating the TyG index and its obesity-integrated derivatives against hypertension prevalence revealed distinct discriminatory abilities. TyG-WHtR showed superior predictive validity (AUC = 0.643, 95% CI 0.625–0.660), outperforming TyG-WC (AUC = 0.632, 0.614–0.650) and TyG-BMI (AUC = 0.615, 0.597–0.632). The stand-alone TyG index showed a comparatively lower diagnostic precision (AUC = 0.588, 0.570–0.606) (Figure 3).

In order to assess the robustness of the findings, sensitivity analyses were performed using the complete-case dataset (pre-imputation). The associations between TyG-associated indices (TyG index, TyG-BMI, TyG-WC, and TyG-WHtR) and hypertension remained statistically significant across all multivariable-adjusted models, both for continuous and categorical (tertile-based) analyses. These results closely aligned with the primary analyses derived from multiply imputed data (Supplementary Tables 1, 2).

Furthermore, sensitivity analyses using multivariate regression and ROC curves demonstrated that the TyG index, its obesity indicators (TyG-BMI, TyG-WC, and TyG-WHtR), and HOMA-IR were significantly associated with hypertension risk. Notably, the TyG-WHtR index exhibited superior discriminative capacity for hypertension, with an AUC of 0.643 (95% CI: 0.626–0.661), compared to HOMA-IR (AUC = 0.630, 95% CI: 0.612–0.647). Detailed results of these analyses are provided in Supplementary Tables 3, 4, while the ROC analysis findings are summarized in Supplementary Figure 1 and Supplementary Table 5.