This study used data from the China Health and Retirement Longitudinal Study (CHARLS), a nationwide longitudinal survey of people aged 45 years or above (21). A stratified four-stage probability sampling strategy was adopted to recruit participants from 150 counties and urban districts across 28 provinces in China. The baseline survey was conducted in 2011, with the respondent rate of 80.5% for questionnaires, 78.9% for anthropometric measurements, and 67% for blood sample collection (22). Respondents were then followed up with a face-to-face computer-assisted personal interview every 2 years (2013, 2015, and 2018, respectively). Detailed study procedures can be found elsewhere (21). CHARLS was approved by the Behavioral and Social Research division of the National Institute on Aging of the National Institute of Health, the Natural Science Foundation of China, the World Bank, and Peking University. All participants provided their written informed consent to participate in this study.

A total of 17,708 participants were recruited at baseline in CHARLS 2011. After excluding those under the age of 45 (n = 289) or with incomplete data on sex, education, economic status, tobacco use, systolic blood pressure (SBP), diastolic blood pressure (DBP), weight, height, and waist circumstance (WC) (n = 6,018), who had incomplete data on fasting triglycerides (TG), fasting plasma glucose (FPG), high-density lipoprotein cholesterol (HDL-C), and low-density lipoprotein cholesterol (LDL-C) (n = 3,192), finally 8,209 participants were included in analyses (Figure 1).

Blood sample collection was done once every two follow-up periods (2011 and 2015). Biomarkers were measured using overnight fasting blood samples, which were promptly transported to the local laboratory and stored at 4°C. The blood samples were centrifugated and stored at −20°C before being transported to the China Center for Disease Prevention and Control (CDC) in Beijing and frozen at −70°C before analysis (21). Routine blood tests were performed at township hospitals or the local CDC. FPG, TG, HDL-C, and LDL-C were measured using the enzymatic colorimetric method. High sensitivity C-reactive protein (hs-CRP) was measured using the immunoturbidimetric assay.

The TyG index was calculated as ln ⁡ ⁡ T G   m g / d l × F P G   m g / d l / 2 (13).

Blood pressure (BP) was measured three times in a sitting position at a 45-s interval using an electronic monitor (Omron model HEM-7200). The average of the three measurements was calculated to the nearest 0.1 mmHg for analysis. Hypertension statuses were divided into hypertension and normotension. Hypertension was defined as mean SBP ≥ 140 mmHg and/or mean DBP ≥ 90 mmHg and/or having hypertension treatments (12, 21). Normotension was defined as SBP < 140 mmHg and DBP < 90 mmHg. After excluding those with antihypertensive medications, hypertension stages were further classified as stage 1 hypertension (SBP ≥ 140 mmHg and < 159 mmHg and/or DBP ≥ 90 mmHg and < 99 mmHg), stage 2 hypertension (SBP ≥ 160 mmHg and < 179 mmHg and/or DBP ≥ 100 mmHg and < 109 mmHg), and stage 3 hypertension (SBP ≥ 180 mmHg and/or DBP ≥ 110 mmHg). Given the sample size constraints, we combined stage 2 and stage 3 hypertension into stage 2 hypertension. Hypertension phenotypes were also classified as ISH (SBP ≥ 140 mmHg and DBP < 90 mmHg), IDH (SBP < 140 mmHg and DBP ≥ 90 mmHg), and SDH (SBP ≥ 140 mmHg and DBP ≥ 90 mmHg). More details are shown in Supplementary Table S1. The progressions of hypertension statuses were classified as maintained normotension, normotension to hypertension, and maintained hypertension. The progressions of hypertension stages were classified as maintained normotension, normotension to stage 1, normotension to stage 2, maintained stage 1, stage 1 to stage 2, and maintained stage 2. The progressions of hypertension phenotypes were classified as maintained normotension, normotension to ISH, normotension to IDH, normotension to SDH, maintained ISH, ISH to SDH, and maintained SDH.

Anthropometric measurements, including height, weight, and WC, were made at every 2-year follow-up. Information on age, sex, residence, education, economic status, tobacco use, and alcohol consumption were collected through questionnaires. The residence was classified as urban and rural according to the location of the participants. Education was divided into illiterate, primary school and below, and junior high school or above. Economic status was evaluated by the natural logarithm of per capita expenditures (ln [PCE]) (23). The bottom, middle, and top tertile of ln (PCE) indicated poor, middle, and rich economic status, respectively. Tobacco use was categorized as never smoking and ever smoking (including former smokers and current smokers). Likewise, alcohol consumption was classified as never drinking and ever drinking (including former alcohol consumers and current alcohol consumers). Body mass index (BMI) was calculated as w e i g h t   k g   / h e i g h t   [ m 2 . The categories of general obesity included normal (BMI < 24 kg/m²), overweight (BMI ≥ 24 kg/m² and BMI < 28 kg/m²), and obesity (BMI ≥ 28 kg/m²) (24). Diabetes was defined as FPG ≥ 7.0 mmol/L, and/or random plasma glucose ≥ 11.1 mmol/L, and/or HbA1c ≥ 6.5%, and/or self-reported treatment (25).

The normality test showed that all continuous variables were skewed. Characteristics of participants at baseline (CHARLS 2011) were described in the form of the median with interquartile range (IQR) for continuous variables with skewed distributions, and number with percent (%) for categorical variables. The basic characteristics of participants between the included participants and excluded participants, between normotension and hypertension, among different hypertension stages, hypertension phenotypes, and hypertension progressions were compared using Wilcoxon rank-sum test for continuous variables and the Chi-square test for categorical variables.

The included participants were divided into four groups based on their TyG index levels (quartile 1 [Q1], quartile 2 [Q2], quartile 3 [Q3] and quartile 4 [Q4]) (18, 26). In the data analysis, the TyG index was considered as a continuous (per-unit) variable and converted to a categorical (quartiles) variable.

For the cross-sectional analysis in CHARLS 2011, we used binomial logistic regression to assess the associations between TyG index and hypertension statuses. Meanwhile, we used multinomial logistic regressions to investigate the associations of the TyG index with hypertension stages and phenotypes, with normotension as reference. The model was adjusted for age, sex, residence, education, economic status, tobacco use, alcohol consumption, general obesity, WC, LDL-C, HDL-C, and hs-CRP. Meanwhile, considering the sex-TyG index interaction, we conducted sex-stratified multinomial logistic regression as well (27). The goodness of fit of the multinomial logistic regression model has been verified using the Hosmer-Lemeshow Goodness-of-Fit Test. Consider the impacts of diabetes, hypoglycemic agents, insulin, or dyslipidemia drugs on glucose, lipid metabolism, and BP. Sensitivity analyses were also conducted as follows to explore the stability of our findings: 1) excluding people with diabetes; 2) further adjusted diabetes and dyslipidemia treatment. We further investigated the optimal cutoff point of TyG index for prediction of hypertension according to best Youden index (YI, sensitivity+specificity−1). The receiver operating characteristics (ROC) curve and the area under the curve (AUC) of TyG index were also conducted.

For the longitudinal analysis using CHARLS 2011 and 2015, we excluded those who had hypertension treatments to ensure proper classification of hypertension stages and phenotypes. Participants interviewed in both waves with complete data on SBP and DBP were included to further investigate the associations between the TyG index and hypertension progressions. Those who had non-attenuated hypertension progression patterns including maintained normotension, normotension to hypertension, and maintained hypertension based on their hypertension statuses from 2011 to 2015 were included. Subsequently, we selected participants who had non-attenuated hypertension stages (i.e., maintained normotension, normotension to stage 1, normotension to stage 2, maintained stage 1, stage 1 to stage 2, and maintained stage 2) and phenotypes (i.e., maintained normotension, normotension to ISH, normotension to IDH, normotension to SDH, maintained ISH, ISH to SDH, and maintained SDH) to explore whether the TyG index was associated with hypertension stages and phenotypes progressions, respectively. Those who had unrelated progressions, which referred to attenuated hypertension progression patterns (i.e., hypertension to normotension; stage 2 to stage 1, stage 2 to normotension, stage 1 to normotension; ISH to normotension, ISH to IDH, IDH to normotension, IDH to ISH, SDH to normotension, SDH to ISH, SDH to IDH), were excluded, accordingly. The model and adjustments were consistent with those above. All of these analyses were conducted using sex-stratified multinomial logistic regression with maintained normotension as reference.

This study was done following the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines (28). SAS statistical software (version 9.4; SAS Institute Inc., Cary, NC, United States) was used. All analyses were two-sided, and statistical significance was defined as a p-value <0.05.

Characteristic comparisons between the included (n = 8,209) and excluded (n = 9,499) participants were shown in Supplementary Table S2. The average age was 59.00 (52.00–65.00) of the included participants, with a greater proportion of female (53.0%). More than half of the participants lived in rural area (65.7%), never smoked (60.4%), never had alcohol consumption (74.7%). Table 1 showed that of the 8,209 individuals included in CHARLS 2011, 5,040 (61.4%) had normotension, and 3,169 (38.6%) had hypertension. Among them, 406 participants received diabetes treatment (took medicine or insulin injection) and dyslipidemia treatment (took medicine or other treatment). The median TyG index was 8.52 (IQR: 8.17–8.94) for normotension, and 8.74 (IQR: 8.34–9.20) for hypertension, with p values <0.05. Significant differences in age, residence, education, economic status, alcohol consumption, general obesity, WC, LDL-C, HDL-C, and hs-CRP between different hypertension statuses were also observed (all p values <0.05).

For the cross-sectional analysis in CHARLS 2011, we found that when compared with the bottom quartile of baseline TyG index, a higher quartile of TyG index was significantly associated with hypertension (adjusted odds ratio [aOR] 1.17, 95% confidence interval [CI] 1.01–1.35 for Q2; aOR 1.48, 95% CI 1.28–1.72 for Q3; aOR 1.90, 95% CI 1.63–2.23 for Q4). Similar results were also found when the TyG index was used as a continuous variable and in the sex-stratified analyses (Table 2). In Supplementary Tables S3, S4, the sensitivity analyses of the TyG index with hypertension in 2011 among population without diabetes and further adjusted for diabetes and dyslipidemia treatment showed similar results in line with our primary analysis. Supplementary Figure S1 presented the ROC curve of the TyG index for prediction of hypertension, with AUC of 0.595. According to YI, the optimal cutoff point of TyG index in predicting hypertension was 8.637.

After excluding participants who had hypertension treatment in 2011 (n = 1,625), a total of 6,584 participants were finally included in the classification of hypertension stages or phenotypes, according to SBP and DBP levels. Baseline characteristics of the remaining 6,584 participants by hypertension stages and phenotypes were shown in Supplementary Tables S5, S6, respectively. We also found that compared with people with the lowest quartile of baseline exposure, those with a higher TyG index had an increased risk of stage 1 hypertension (aOR 1.26, 95% CI 1.02–1.55 for Q3; aOR 1.71, 95% CI 1.38–2.13 for Q4) and stage 2 hypertension (aOR 1.52, 95% CI 1.13–2.04 for Q3; aOR 1.74, 95% CI 1.27–2.38 for Q4). Furthermore, the TyG index was also positively associated with ISH (aOR 1.35, 95% CI 1.08–1.68 for Q3; aOR 1.66, 95% CI 1.31–2.11 for Q4), IDH (aOR 2.04, 95% CI 1.03–4.04 for Q3; aOR 2.52, 95% CI 1.26–5.05 for Q4), and SDH (aOR 1.65, 95% CI 1.23–2.23 for Q4) (Table 3). Similarly, a higher TyG index (per-unit) was significantly associated with the development of stage 1 hypertension (aOR 1.43, 95% CI 1.27–1.61), stage 2 hypertension (aOR 1.42, 95% CI 1.20–1.67), ISH (aOR 1.41, 95% CI 1.24–1.61), IDH (aOR 1.78, 95% CI 1.30–2.45), SDH (aOR 1.35, 95% CI 1.15–1.58). However, when stratified by sex, neither as a continuous variable nor as a categorical variable of the TyG index had a significant association with stage 2 hypertension in males.

In the longitudinal study using CHARLS 2011 and 2015, participants with incomplete data on SBP and DBP (n = 1,887) or unrelated hypertension progressions (n = 893) were further excluded, leaving 5,429 participants for longitudinal analyses. During the 4-year follow-up, 796 individuals (14.7%) developed hypertension (Supplementary Table S7). Participants who were in the highest quartile TyG index were more likely to have the progression of normotension to hypertension (normotension to hypertension vs. maintained normotension: aOR 1.48, 95% CI 1.14–1.92) and maintained hypertension (maintained hypertension vs. maintained normotension: aOR 2.16, 95% CI 1.74–2.69), compared with those in the lowest quartile TyG index, as shown in Supplementary Table S8. Associations of the per-unit TyG index with the progression of normotension to hypertension (aOR 1.34, 95% CI 1.16–1.54) and maintained hypertension (aOR 1.59, 95% CI 1.41–1.79) were also found when compared with maintained normotension. However, when stratified by sex, there was no evidence of significant associations between the TyG index (per-unit) and normotension to hypertension (aOR 1.18, 95% CI 0.96–1.45), compared with maintained normotension in females (Supplementary Table S8).

Furthermore, after excluding participants who had hypertension treatment in 2011 and 2015 (n = 1,188), with unrelated progressions of hypertension stages (n = 97) and unrelated progressions of hypertension phenotypes (n = 115), 4,144 and 4,126 participants were enrolled separately to explore whether the TyG index was associated with the progressions of hypertension stages and phenotypes. Characteristics of these participants were described in Supplementary Tables S9, S10, respectively. As shown in Supplementary Table S11 and Figure 2, individuals with a higher baseline TyG index were more likely to have progressions of normotension to stage 1 (for highest quartile vs. lowest quartile: aOR 1.45, 95%CI 1.05–2.00; for per-unit: aOR 1.39, 95% CI 1.16–1.65), maintained stage 1 (for highest quartile vs. lowest quartile: aOR 1.68, 95%CI 1.10–2.56; for per-unit: aOR 1.32, 95% CI 1.05–1.67), and maintained stage 2 (for highest quartile vs. lowest quartile: aOR 2.62, 95%CI 1.23–5.62; for per-unit: aOR 2.00, 95% CI 1.37–2.91).

In addition, we also found significant associations of the TyG index with normotension to ISH (per-unit: aOR 1.28, 95% CI 1.04–1.56), normotension to IDH (for highest quartile vs. lowest quartile: aOR 3.46, 95%CI 1.42–8.44; for per-unit: aOR 1.94, 95% CI 1.27–2.97), maintained ISH (for highest quartile vs. lowest quartile: aOR 1.63, 95%CI 1.03–2.57; for per-unit: aOR 1.34, 95% CI = 1.05–1.70), and maintained SDH (for highest quartile vs. lowest quartile: aOR 2.66, 95%CI 1.36–5.20; for per-unit: aOR 1.71, 95% CI 1.22–2.40) as shown in Supplementary Table S12 and Figure 3.

The principal strength of this study is that we are the first to explore the associations of the TyG index with various hypertension stages and phenotypes, as well as their progressions, in a large Chinese general population. Since the TyG index is a recently proposed and high-profile indicator, our findings assist to accelerate the clinical application of the TyG index to facilitate early hypertension screening. Additionally, hypertension is known to be common in elderly with higher risks of other diseases, such as metabolic syndrome, type 2 diabetes, and cardiovascular diseases (26, 53). Our results indicate that the TyG index is helpful in predicting the risk of hypertension in elderly, suggesting that the TyG index may have a valuable application value in the clinic risk assessment of other diseases. Moreover, we used data from the CHARLS, which has broad geographic coverage, rigorous implementation procedures, and nationwide samples to ensure the national representation of our conclusion.

Despite these strengths, several limitations need to be recognized. First, the TyG index was only assessed in 2011, and its changes during follow-up were not taken into account. Second, there were significant characteristic differences between included and excluded participants (Supplementary Table S2), which may lead to selection bias and affect generalizability. Third, since all participants were Chinese adults aged 45 and over, our findings cannot be directly extrapolated to younger groups. Fourth, we merged stage 2 and stage 3 hypertension into stage 2 hypertension due to the limited sample size, which may result in misjudging the association between the TyG index and stage 2 hypertension, or the progressions. Moreover, the participants in our study are the general population but not the patients with hypertension, which may pull down the average TyG level and lead to the non-significant association of the TyG index with normotension to stage 2 and normotension to SDH. More high-quality epidemiologic investigations and primary prevention studies are needed to explore the role of TyG index in the progression of hypertension. Besides, the BP variability and the different environments, times of measurement (seasonal variations or diurnal variations) may lead to inaccuracy of BP values (54–57). Finally, even though the model we used was adjusted for multiple covariates, certain possible confounding factors, such as exercise habits and family history of hypertension, were not included.

In conclusion, our study revealed that a higher TyG index was significantly associated with hypertension stages, phenotypes, and their progressions. Given the rising prevalence of hypertension and diabetes in China, the TyG index may serve as a useful indicator for the early prevention of hypertension and help to identify persons at higher risk of hypertension.