This study presents a secondary analysis of data sourced from CHARLS, a comprehensive nationwide cohort study designed to represent the national population (16). The design and methodology of CHARLS have been elaborated in previous publications (16). Succinctly, the study sampled 17,708 participants from 10,257 households, utilizing a multistage stratified probability-proportional-to-size sampling method. Participants were enlisted from 150 counties or districts and 450 villages across 28 provinces in China between June 2011 and March 2012. Using a standardized questionnaire, data on sociodemographic, lifestyle factors, and health-related aspects were collected. The initial survey, Wave 1, achieved an 80.5% response rate. Follow-up assessments were conducted biennially: Wave 2 in 2013, Wave 3 in 2015, and Wave 4 in 2018. For the current analysis, only participants aged 45 or above with comprehensive data on fasting blood glucose (FBG), triglycerides (TG), hypertension, and blood pressure were considered. Those with a prior history of stroke were excluded.

The CHARLS study secured approval from Peking University's institutional review board, and all participants provided written informed consent. The current study is compliant with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines (17).

The Beijing-based Chinese Center for Disease Control and Prevention swiftly received venous blood samples within two weeks of their dispatch from the CDC station. These samples were promptly preserved at −20°C until transport. Post assay completion at the Chinese Medical University laboratory, they were relocated to a deep freezer, with a maintained temperature of −80°C. TG and FPG concentrations were analyzed using the enzyme colorimetric assay at the Youanmen Clinical Laboratory of Capital Medical University. The TyG index was computed using the formula ln [TG (mg/dl) × FPG (mg/dl)/2] (18).

Hypertension was characterized by a systolic blood pressure of ≥140 mmHg, diastolic blood pressure of ≥90 mmHg, ongoing antihypertensive medication use, or a self-reported history of the condition. Blood pressure measurements were undertaken by trained nursing personnel.

The primary outcome of this study was the incidence of stroke. In accordance with previous studies (19, 20), stroke incidents were identified through self-reports wherein participants confirmed a physician's diagnosis of stroke. The recorded date of the stroke diagnosis was demarcated as the interval between the date of the latest interview and the date when the incident stroke was reported (19, 20).

During the initial assessment at Wave 1, trained interviewers gathered sociodemographic and health-related data via a structured questionnaire, including age, gender, residence, marital status, and educational level (categorized as no formal education, primary school, middle or high school, or college and above). Marital status was classified as either married or another marital status (including individuals who were never married, separated, divorced, or widowed). Health-related variables included self-reported smoking and drinking status (categorized as never, former, or current), self-reported physician-diagnosed medical conditions (diabetes, heart disease, kidney disease, and dyslipidemia), and the use of medications for diabetes, hypertension, and dyslipidemia. Laboratory tests included measurements of total cholesterol (TC), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), estimated glomerular filtration rate (eGFR), and glycosylated hemoglobin, type A1c (HbA1c) (21). Diabetes was defined as fasting plasma glucose ≥126 mg/dl, current use of antidiabetic medication, or self-reported history of diabetes. Dyslipidemia was defined as total cholesterol levels ≥240 mg/dl, triglyceride levels ≥150 mg/dl, low-density lipoprotein cholesterol levels ≥160 mg/dl, high-density lipoprotein cholesterol levels <40 mg/dl, current use of lipid-lowering medication, or self-reported history of dyslipidemia. Height and weight were measured by a trained nurse. Body mass index was calculated as weight in kilograms divided by height in meters squared. The eGFR was calculated using the Chronic Kidney Disease Epidemiology Collaboration's 2009 creatinine equation (22).

Statistical analysis was conducted from October 1, 2023, to October 23, 2023. Data were described as medians and interquartile ranges for continuous variables. The frequency with percentage was used to describe categorical variables. The χ² test and Kruskal-Wallis rank sum test were utilized to compare differences in the baseline characteristics between the different groups, as appropriate. Five percent (452 of 9,145) of total data items were missing, were assumed to be missing at random, and, thus, were imputed with the multiple imputation of chained equations method using the baseline characteristics.

We used Fine and Gray competing risk models (23) to estimate the hazard ratios (HRs) and 95% confidence intervals (CIs) of stroke associated with hypertension, systolic blood pressure (categorized as quartiles), diastolic blood pressure (categorized as quartiles), and TyG index (categorized as quartiles); we also used logistic regression models to estimate the odds ratios (ORs) and 95% CIs of hypertension associated with TyG index. Three models were estimated: in model 1, age and sex were adjusted; in model 2, age, sex, residence, marital status, educational level, body mass index, smoking status, and drinking status were adjusted; and in model 3, the variables in model 2 plus history of diabetes, heart disease, dyslipidemia, kidney disease, history of medication use for diabetes, history of medication use for dyslipidemia, TC, HDL-C, LDL-C, HbA1c, hsCRP, and eGFR were adjusted. In addition, we explored the potential nonlinear associations using 4-knotted restricted cubic spline (RCS) regression, and the models were conducted with 4 knots at the 5th, 35th, 65th, and 95th percentiles of TyG index, systolic blood pressure, or diastolic blood pressure (reference is the 5th percentile).

To quantify the additive and multiplicative interactions, we additionally included a product term of blood pressure (hypertension, systolic blood pressure, or diastolic blood pressure) and TyG index (quartiles) in the model. The HR with its 95% CIs of the product term was the measure of interaction on the multiplicative scale, and the multiplicative interaction was statistically significant when its CIs did not include 1. We assessed additive interactive effects using the synergy index (SI). This metric captures different aspects of interaction, including the part of the effect attributable to interaction, the proportion of the combined effect arising from interaction, and the ratio between the combined effect and individual effects. Specifically, SI = 1 indicates the absence of interactive effects between blood pressure and TyG index concerning stroke incidence. Conversely, when SI > 1, this signifies that the combined effects of blood pressure and TyG index on stroke incidence exceed the sum of their individual effects, suggesting synergistic effects. Conversely, if SI < 1, it indicates that the combined effects are smaller than the sum of the individual effects of blood pressure and TyG index. We computed the corresponding 95% CIs for SI using the delta method, and the additive interaction was statistically significant when its CIs did not include 1.

We proceeded to assess indirect associations of TyG index with stroke mediated by hypertension and direct associations unmediated by blood pressure (hypertension, systolic blood pressure, or diastolic blood pressure). Using the “mediation” package in R, we applied a regression-based approach to compute the total effect, natural indirect effects (NIE), and natural direct effects (NDE) of the TyG index on stroke incidence (Supplementary Figure S1). Two models were constructed: one involved a multivariable logistic regression model for blood pressure (mediator), conditioned on TyG index (exposure) and confounders, while the other encompassed a multivariable Cox proportional hazard regression model for stroke (outcome), conditioned on TyG index, blood pressure, and confounders. NDE conveyed the impact the of TyG index on stroke independent of blood pressure, whereas NIE represented the proportion of TyG index influenced by its connection with blood pressure over time. To gauge the extent of mediation, we computed the proportion of the association mediated by TyG as NIE divided by the sum of NDE and NIE.

To test the robustness and potential variations in different subgroups, we repeated all analyses stratified by gender (male and female), and age groups (<60 years, and ≥60 years, defined as elders by the World Health Organization (24). And, sensitivity analyses were conducted as follows: (1) repeating all analyses using the complete data set (8,693 participants) without multiple imputations; (2) using blood pressure (hypertension, systolic blood pressure, or diastolic blood pressure) at Wave 2 and excluded participants who have hypertension at baseline when analyzed their mediating effects to minimize the possibility of reverse causality on the observed associations. All analyses were performed using R statistical software version 4.3.0 (R Foundation). We considered two-sided P values <0.05 to be significant.

Of the 17,708 CHARLS participants at study baseline, we excluded 777 individuals younger than 45 years, 634 with stroke at baseline, 5,471 missing TyG index, 643 missing systole blood pressure, 2 missing diastolic blood pressure, and 36 missing hypertension. Finally, 9,145 participants were included for analysis. A detailed description of the selection process for the study analytic sample is included in Supplementary Figure S2. A comparison of baseline characteristics between participants included and those who were not included in the analysis is shown in Supplementary Table S1.

The mean (SD) age at baseline was 59.20 (9.33) years; 4,251 (46.5%) of the participants were men. Table 1 shows the characteristics of the participants. At baseline, the mean (SD) TyG index was 8.67 (0.68), the mean (SD) systole blood pressure was 129.59 (21.50) mmHg, while the mean (SD) diastolic blood pressure was 75.32 (12.23) mmHg (Supplementary Figure S3–S4), and 41.1% (3,757/9,145) participants experienced hypertension. When compared with Q1 of TyG index, participants in the other quartiles were more likely to be female, body mass index ≥28.0 kg/m² and married; to reside in an urban setting; to have no formal education, fewer current smokers and drinkers, and higher systolic and diastolic blood pressure; to have a higher prevalence of hypertension, diabetes, dyslipidaemia, and heart disease; to have a history of medication use for diabetes and dyslipidemia; to have higher TC, LDL-C HbA1, hsCRP, and eGFR; and to have lower HDL-C levels (Table 1).

During a median follow-up of 7.1 years, 637 (7.0%) participants experienced an incident stroke. Table 2 shows the associations between TyG index and incident stroke. After adjusting for confounders (in model 3), when compared with Quartile 1, the adjusted HRs (95% CIs) for incident stroke were 1.27 (0.98–1.65) for Quartile 2, 1.55 (1.19–2.02) for Quartile 3, and 1.21 (0.88–1.66) for Quartile 4. A one SD increase in TyG index was associated with an elevated risk of stroke (HR, 1.16; 95% CI, 1.02–1.33). A nonlinear association between the TyG index and risk of incident stroke using RCS regression was also found (for association, P = 0.004; for nonlinear, P = 0.009; Supplementary Figure S5A). Hypertension, systole blood pressure ≥140 mmHg, and diastolic blood pressure ≥90 mmHg were independently associated with increased risk of incident stroke, the HRs were 2.10 (95% CI, 1.76–2.50), 1.62 (95% CI, 1.37–1.92), and 1.70 (95% CI, 1.39–2.08), respectively. Similar results were found when modeling the systole blood pressure and diastolic blood pressure as quartiles, and a linear and positive association of systole blood pressure and diastolic blood pressure with the risk of incident stroke using RCS regression were also found (Supplementary Figures S5B,C). Similar results were found when complete data analyses were conducted (Supplementary Table S2).

Table 3 shows the associations between TyG index and incident hypertension. After adjusting for confounders (in model 3), by comparing Quartile 4 with Quartile 1, the adjusted ORs for incident hypertension were 1.52 (95% CI, 1.27–1.81). A one SD increase in TyG index was associated with an elevated risk of hypertension (OR, 1.29; 95% CI, 1.19–1.41). A linear and positive association between the TyG index and risk of incident hypertension using RCS regression was also found (for association, P < 0.001; for nonlinear, P = 0.438; Supplementary Figure S5D). Similar results were found when complete data analyses were conducted (Supplementary Table S3).

Figure 1 shows the interaction and joint association of TyG index and hypertension on the stroke. Both multiplicative and additive interactions were observed between TyG index and hypertension on stroke (HR for multiplicative: 2.34, 95% CI, 1.57–3.48; SI: 4.13, 95% CI, 2.73–6.25; Figure 1A). Similar patterns were found for TyG index and systole blood pressure on stroke (Figure 1B), as well as, TyG index and diastolic blood pressure on stroke (Figure 1C). The results remained similar in all sensitivity analyses (Supplementary Table S4) and subgroup analyses (Supplementary Tables S5–S6).

As shown in Figure 2, significant mediated effects by hypertension, systole blood pressure, or diastolic blood pressure were observed on the associations between TyG index and stroke, the mediated proportion were 20.0%, 13.2%, and 9.8%, respectively. Similar results were found when complete data analyses were conducted (Supplementary Table S7), used blood pressure data at Wave 2 (Supplementary Table S8), or stratified by gender (Supplementary Table S9). However, the mediated proportion was more pronounced among participants with age <60 years when compared with those with age ≥60 years (mediated by hypertension: 27.6% vs. 14.3%; by systole blood pressure: 24.6% vs. 6.9%; by diastolic blood pressure: 20.4% vs. 4.1%; Supplementary Table S10).