This study utilized data from the 2017 China National Nutrition and Health Surveillance of Children and Lactating Women (34). This surveillance employed a stratified multi-stage cluster random sampling design to select participants. A total of 125 survey sites were systematically selected across 31 provinces in China. Site selection took into account the distribution balance of regional and urban-rural stratification factors, existing work basis and conditions. These 125 monitoring sites were allocated proportionally to four types of regions based on population size: 5 large cities, 57 medium/small cities, 50 rural areas, and 13 impoverished rural areas. In each survey site (city/district/county), two townships or sub-districts were randomly selected, and two villages or neighborhood committees were randomly chosen from each township or street district. For children and adolescents, students from 10 grades were surveyed at each site, including grades 1–6 in two primary schools, grades 7–8 in two junior high schools, and grades 10–11 in one senior high school (the schools were randomly selected from each monitoring site). Considering the academic load and the poor compliance, grade 9 and grade 12 were not included in this surveillance. Subsequently, one class was randomly selected for each grade, and 28 students with equal numbers of males and females were chosen from each selected class. The collected surveillance data is representative at both the provincial and national levels.

Due to limited access to the entire national dataset, we used data from five provinces: Shandong, Jiangsu, Guangdong, Guizhou, and Inner Mongolia, which are geographically located in the East, South, Southwest, and North of China. The initial dataset of the five provinces included 15,673 children and adolescents. As our study focused on participants aged 7 to 17 years, we excluded 866 individuals who fell outside this age range. Additionally, 259 participants lacked laboratory data for calculating TyG and METS-IR, and 193 participants missing variables necessary to determine BP status were excluded. Subsequently, 2,268 individuals were excluded due to missing covariates. A flowchart of the subjects’ exclusion process is presented in Supplementary Figure 1 . A total of 12,087 participants entered the final analysis. Informed consent was obtained from all participants or their legal guardians.

The participants received a standardized questionnaire to obtain their demographic, health-related, and lifestyle information. All questionnaires were asked and completed in person by uniformly trained investigators.

Measurements were taken using uniform instruments at each monitoring site. Height was measured in the standing position with shoes removed using a metal TZG-type stadiometer with an accuracy of 0.1 cm. Body weight was measured in the fasted state, with the subjects in underwear and without shoes, using an electronic scale (TANITA, HD-390) that was accurate at 0.01 kg. Waist circumference (WC) was measured horizontally at the midpoint between the inferior edge of the rib cage and the iliac crest along the mid-axillary line in the fasting state using a tape. Systolic blood pressure (SBP) and diastolic blood pressure (DBP) were measured three times at one-minute intervals using an electronic sphygmomanometer (Omron HBP 1300, Tokyo, Japan) in the morning on the left arm (unless otherwise specified). The mean of the two closest readings among the three measurements was taken. The participants were instructed to avoid intense physical activity, eating, or drinking within one hour before the measurement.

Fasting blood samples (6 ml) were collected for biochemical parameters. Fasting plasma glucose (FPG) was measured using the glucokinase method (Roche P800 automatic biochemical analyzer). Triglyceride (TG), total cholesterol (TC), low-density lipoprotein cholesterol (LDL-C), and high-density lipoprotein cholesterol (HDL-C) were measured using a Roche Cobas C701 automatic analyzer.

TyG was calculated using the formula: ln[TG (mg/dL) × FPG (mg/dL)/2] (15). The METS-IR was calculated as ln[(2 × FPG (mg/dL) + TG (mg/dL)) × BMI (kg/m²)/ln[HDL-C (mg/dL)] (15).

The current definition of pediatric hypertension is based on the normative distribution of BP in healthy children (6, 35). According to sex, age, and height percentiles, normal blood pressure (NBP) was defined as SBP and DBP < 90th percentile, prehypertension as SBP and/or DBP ≥ 90th and < 95th percentile. Prehypertension is previously referred to as “high normal blood pressure”, a term now replaced by “elevated blood pressure” (EBP) (6), and they are equivalent. Hypertension was defined as SBP and/or DBP ≥ 95th percentile for sex, age, and height (35). To align our definition with the China National Health Industry Standard for children and adolescents, this study refers to SBP and/or DBP ≥ 95th percentile as high blood pressure (HBP) in later text (36). Isolated systolic HBP (ISH) was defined as SBP ≥ 95th and DBP < 95th percentile. Isolated diastolic HBP (IDH) was defined as DBP ≥ 95th and SBP < 95th percentile. Systolic-diastolic HBP (SDH) was defined as both SBP and DBP ≥ 95th percentile.

The covariates included in this study were demographic variables (age, sex, parental education level, residence), health-related and lifestyle variables [abdominal obesity, estimated glomerular filtration rate (eGFR), moderate-vigorous physical activity (MVPA), sleep sufficiency, passive smoking, alcohol intake, unhealthy dietary quality score, and family history of hypertension], and biochemical indicators (serum uric acid, total protein, TC, LDL-C).

Age was calculated by subtracting the date of birth from the survey date, and the full years were taken. Parental education level was categorized into three groups based on the highest level of education attained by both parents: Low—both parents had a primary school education/lower, or one had a primary school education/lower and the other had a secondary school education/diploma; Medium—both parents had a secondary school education/diploma, or one had a primary school education/lower and the other had a bachelor’s degree/higher; High—both parents had a bachelor’s degree/higher, or one had a secondary school education/diploma and the other had a bachelor’s degree/higher.

Abdominal obesity was defined as a WC at or above the 90th percentile for age and sex, determined according to the cutoff points specified in the National Health Industry Standard of China for children and adolescents (37). The eGFR was calculated using the formula recommended by the Chinese guideline for early screening of pediatric chronic kidney disease (38): K × height (cm) × 88.4/serum-creatinine (μmol/L). The constant K is defined as follows: for children aged 2–12 years, K = 0.55; for individuals aged >12 years, K = 0.77 for boys and 0.55 for girls.

Sleep sufficiency was defined based on recommendations by the Ministry of Education of the People’s Republic of China (39), with sleep duration greater than 8 hours for high school students, greater than 9 hours for middle school students, and greater than 10 hours for primary school students. Participants meeting these criteria were classified as “yes” for sufficient sleep, and “no” otherwise. Family history of hypertension was defined as having at least one of the following family members diagnosed with hypertension: father, mother, paternal grandparents, or maternal grandparents.

The unhealthy dietary quality score was calculated based on data from a food frequency questionnaire recording the various foods consumed by participants over the past month and the Chinese Food Guide Pagoda issued by the Chinese Nutrition Society (40). The detailed scoring process for unhealthy dietary quality score can be found in our previous article (41). The remaining covariates were simple self-reports, investigator reports, or laboratory tests.

All the statistical analyses were conducted using R Project for Statistical Computing version 4.2.3 (Vienna, Austria). A two-sided p-value <0.05 was considered statistically significant. Continuous variables were presented as medians and interquartile ranges, whereas categorical variables were reported as frequencies (percentages). Kruskal-Wallis tests, Welch’s ANOVA, and chi-square tests were used to compare variable differences across groups where appropriate. Post hoc pairwise comparisons with Bonferroni correction were performed between the NBP group and the EBP group, as well as between the NBP group and the HBP group, yielding a statistically significance threshold of p-value <0.025 (0.05/2). For pairwise comparisons, the Mann-Whitney U tests with a Bonferroni correction were applied to continuous variables, while the chi-square tests with a Bonferroni correction were used for categorical variables.

Multinomial logistic regression analyses were performed to evaluate the associations of TyG and METS-IR with different BP groups, with the NBP group as the reference group. The results were expressed as odds ratios (ORs) and 95% confidence intervals (CIs). Multicollinearity was evaluated with the Variance Inflation Factor (VIF), ensuring all variables had VIF values below 5. Restricted cubic spline (RCS) regression was used to explore potential non-linear relationships, with the number of knots set to 3 for smooth curve fitting.

Stratified analysis by age, sex, sleep sufficiency, MVPA, and family history of hypertension was performed to investigate potential heterogeneity across subgroups. Sensitivity analysis was conducted using data after propensity score matching (PSM) with a 1:1 nearest-neighbor matching algorithm and a caliper width of 0.20. The propensity score was estimated using a logistic regression model, and the variables used in calculating the score included age, abdominal obesity, MVPA, sleep sufficiency, alcohol intake, family history of hypertension, parental education level, serum creatinine, total protein, TC, and LDL-C, which were covariates with p<0.05 in baseline table, excluding those used to calculate TyG and METS-IR.

This study analyzed data for 12,087 subjects (50.24% girls and 49.76% boys) aged 7–17 years and their basic characteristics by BP status are presented in Table 1 . The median age of the participants was 11 years. Compared with the NBP group, both the EBP group and the HBP group had a significantly greater proportion of individuals with abdominal obesity (p<0.025). Furthermore, BMI, total protein, TG, FPG, TyG, and METS-IR levels were significantly higher in both the EBP and HBP groups than in the NBP group (p<0.025). TC and LDL-C in the HBP group were significantly higher than those in the NBP group (p<0.025) but showed no statistically significant differences between EBP and NBP (p>0.025). In contrast, the serum creatinine level was significantly higher in the NBP group than in the other two groups (p<0.025). Additionally, sleep sufficiency, MVPA, alcohol intake, family history of hypertension, and parental education level differed significantly across groups (p<0.05).

Table 2 presents the associations of TyG and METS-IR with EBP and HBP. TyG and METS-IR were divided into quartiles, with the lowest quartile (Q1) as the reference group. In the fully adjusted Model 2, the ORs for EBP and HBP was 1.2-fold and 1.4-fold higher in the highest quartile groups of TyG (OR: 1.24, 95%CI: 1.06–1.45, p<0.01; OR: 1.44, 95%CI: 1.26–1.65, p<0.001) and 2.3-fold and 3.1-fold higher in the highest quartile groups of METS-IR (OR: 2.32, 95%CI: 1.88–2.86, p<0.001; OR: 3.06, 95%CI: 2.56–3.66, p<0.001), compared with the lowest quartile groups of TyG and METS-IR. Higher quartiles of METS-IR were more strongly linked to the presence of EBP and HBP than the corresponding quartiles of TyG. Supplementary Figure 2 illustrated that TyG and METS-IR had linear relationships with EBP and HBP (p for non-linear >0.10).

Table 3 shows the associations of these two NI-IR indices with various HBP subtypes. Both Model 1 and Model 2 indicated that the third quartile (Q3) and the highest quartile (Q4) of TyG and METS-IR were positively associated with all HBP phenotypes, as were continuous TyG and METS-IR. After adjusting for potential confounders in Model 2, the ORs for ISH, IDH and SDH in the highest TyG quartile groups compared with the lowest groups were 1.39 (95%CI: 1.19–1.63, p<0.001), 1.56 (95%CI: 1.13–2.16, p<0.01), and 1.53 (95%CI: 1.16–2.01, p<0.01), respectively. For METS-IR, the ORs for ISH, IDH and SDH were 3.5-fold, 1.9-fold and 3.0-fold higher in the highest quartile groups than in the lowest quartile groups (OR: 3.46, 95%CI: 2.81–4.26, p<0.001; OR: 1.85, 95%CI: 1.23–2.80, p<0.01; OR: 3.02, 95%CI: 2.12–4.32, p<0.001). The RCS analysis indicated linear relationships between TyG, METS-IR, and all HBP phenotypes (p for non-linear >0.10), as illustrated in Figure 1 .

Figure 2 consistently demonstrates a significantly positive association of TyG and METS-IR with ISH across all strata. The results for SDH are similar. However, for IDH, in the strata where individuals were aged 12 years and older, had sufficient sleep, and engaged in daily MVPA, neither the TyG nor the METS-IR showed a statistically significant association with IDH. The subgroup sample sizes are detailed in Supplementary Table 1 .

Sensitivity analysis was conducted using data after PSM. The results of matching and the characteristics of the new sample are presented in Supplementary Table 2 . Utilizing the new dataset, we re-explored the associations between TyG, METS-IR, and different HBP subtypes, and it remained consistent with our previous findings, suggesting that TyG and METS-IR may link to increased risk of all kinds of HBP phenotypes in children and adolescents ( Supplementary Table 3 ).

Several strengths can be attributed to this study. Our surveillance data are representative at the provincial level and are collected with rigorous quality assurance and control. We have examined the multiplicative interaction between the province and the IR indices on hypertension prevalence and observed that the interaction was not statistically significant (p for interaction >0.05), which enhances the national generalization of our findings. In addition, this study extends the evidence on this research topic from adults to pediatric populations, thereby contributing to the validation of IR index in children. However, we acknowledge that the present study has several limitations. First, although BP was measured three times for each participant, all readings were obtained during a single visit, an unavoidable constraint in epidemiological studies, which may overestimate the prevalence of hypertension (79). Evaluating BP on three or more separate occasions is helpful, thereby reducing measurement variability. Second, despite extensive efforts to adjust for potential confounders, including demographic information, family history, health-related behaviors and status, diet, and kidney health, residual confounding from unmeasured or unknown factors, such as medication use, may still influence the results. Third, since the study population consisted solely of Chinese minors, the applicability of these findings to other ethnicities remains uncertain. Given that the relation between IR and BP differs among racial groups (80), we propose further research among diverse pediatric ethnic populations to improve the global generalizability of these observations.

Finally, the cross-sectional design of this study limits causal inference, as it collects data on exposure and outcome at a single point in time. Consequently, it precludes establishing the temporal sequence of events—we cannot determine whether IR preceded BP elevation, vice versa, or whether they developed concurrently. This ambiguity poses challenges: reverse causality remains a possible alternative explanation, and the dynamic interplay between IR and BP over time cannot be tracked. This limitation could not be overcome through statistical methods. Therefore, the associations reported in this study should be interpreted with caution and not as direct evidence of causality. To address this, prospective cohort studies with repeated measurements of exposure and BP over time are essential.