Insulin resistance has been identified as an important risk factor for the development of cardiovascular disease (CVD) [1], which a leading cause of morbidity and mortality in China and worldwide [2, 3]. A meta-analysis of cohort studies or nested case-control studies have showed that a positive prospective relationship between insulin resistance and risk of CVD in adults without diabetes [4]. Furthermore, a Mendelian randomization analysis established a causal relationship between the two conditions [5]. Therefore, early identifying insulin resistance is essential to reduce the disease burden of CVD.
In the clinical setting, measurement of insulin resistance can be challenging as there are limitations to the homeostasis model assessment for insulin resistance (HOMA-IR) and the gold standard of the euglycemic clamp is time-consuming and burdensome, hence a simpler measure of insulin resistance is needed [6]. The triglyceride-glucose (TyG) index, which is the logarithmized product of fasting triglyceride and glucose, has been shown to be a simple measure of insulin resistance [7]. Previous studies have shown that TyG index is significantly related to an increased risk of cardiovascular events [8-16]. Additionally, a recent meta-analysis of cohort studies included 5,731,294 participants without CVD at baseline showed that the highest TyG index category was associated with 1.61-fold increased risk of CVD [17]. However, most of prior studies were based on a single baseline TyG index measurement [8-12], which may not reflect long-term exposure. Few studies have examined repeated TyG measurements to evaluate the impact of longitudinal TyG index on the risk of CVD. To our knowledge, only one study examined the association between the change TyG index at two time points and incident CVD [14]. Additionally, three studies have reported on the association between time updated average TyG index, the number of visits with a high TyG index and CVD events [8, 9]. Furthermore, although the TyG index can change over time, no study has examined the association between long-term TyG index visit-to-visit variability and CVD development.
Therefore, we conducted a large population-based study involving more than 52,000 Chinese adults who had repeated measurements in TyG index to investigate the longitudinal association between baseline TyG level, visit-to-visit variability in TyG index, and CVD incidence during a median 9.0-year follow-up in a general population. We hypothesized that higher baseline TyG as well as greater variability are both associated with increased risk of CVD.
Study population
Details of the Kailuan study cohort design, methods, and data collection have been published previously [8, 9, 14, 18]. In brief, the Kailuan study recruited 101,510 community-dwelling adults aged 18 years and over between June 2006 and October 2007 in the Kailuan community, Tangshan City, China. A standardized interview and health examinations were conducted at baseline and follow-up. This study was approved by the Ethics Committee of Kailuan General Hospital and Beijing Tiantan Hospital, Capital Medical University, and it was conducted according to the principles of the Declaration of Helsinki. All participants provided written informed consent.
In this analysis, we included participants who underwent 3 health examinations between June 2006 and December 2010 (baseline and index year). Of 56,833 participants, we excluded those who had missing data on fasting blood glucose or triglycerides (n = 1,034) and those who had a previous diagnosis of stroke or myocardial infarction during the 4-year washout period (n = 2,874). Therefore, 52,925 eligible participants were included in the current study (Additional file 1: Figure 1).
Data collection and definitions
At baseline and follow-up surveys, demographics, lifestyles, medical histories, anthropometric measurements, and laboratory tests were collected. Information on age, sex, education (< high school or ≥ high school), income level (≤ 1000 or > 1000 RMB/month), current smoking status (yes or no), and current drinking status (yes or no), physical activity (active: > 4 times/week and > 20 min at a time or inactive: < 80 min/week or none) and history of diseases (e.g., hypertension), and medications use (e.g., antihypertensive agents) was collected via a standardized questionnaire. Trained physicians or nurses measured participants’ height, weight, systolic blood pressure (SBP), and diastolic blood pressure (DBP). Body mass index (BMI) was calculated as weight (kg)/height (m)2. Fasting blood samples were collected and were measured using the Hitachi 747 auto-analyzer (Hitachi, Tokyo, Japan). Serum total cholesterol (TC), triglycerides (TG), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), fasting blood glucose (FBG), creatinine, and high-sensitivity C-reactive protein (hs-CRP) were measured via a standardized protocol. Hypertension was defined as a self-reported history of hypertension, use of antihypertensive medications, SBP ≥140 mm Hg, or DBP ≥ 90 mm Hg. Diabetes was defined as self-reported history of diabetes, use of glucose-lowering drugs, or FBG ≥ 7.0 mmol/l. Hypercholesterolemia was defined as self-reported history of dyslipidemia, use of lipid-lowering medications or TC ≥ 5.17 mmol/l. Estimated glomerular filtration rate (eGFR) was calculated by the Chronic Kidney Disease Epidemiology Collaboration creatinine equation [19]. Chronic kidney disease was defined as an eGFR < 60 mL/min/1.73 m². High hs-CRP was defined as a hs-CRP > 3 mg/dL [20].
Definition of Baseline, Mean and Variability of TyG index
According to previous studies [8, 14], the TyG index was calculated as Ln[TG (mg/dL) × FBG (mg/dL)/2]. The baseline TyG index was calculated by using serum TG and FBG measured in 2010. Mean TyG index was defined as the average of TyG index measured in 2006, 2008, and 2010. The TyG index variability was defined as intraindividual variability in TyG levels measured on these three health examinations. As previously described [21, 22], we used residual SD, defined as visit-to-visit TyG index variability calculated as the root-mean-square error (RMSE) of the residuals (i.e., differences between observed TyG and predicted TyG) obtained from a simple linear regression analysis of the three TyG measurements of each participant.
Study outcome and follow-up
The study outcome was newly diagnosed CVD, which was defined as a composite of myocardial infarction and stroke. As previously described [14, 23], all participants were linked to the Municipal Social Insurance Institution and the Hospital Discharge Register for incidence of CVD, which cover all the Kailuan study participants. To further identify potential CVD cases, we reviewed the discharge lists from the 11 hospitals during 2006 to 2019 and asked for a history of CVD via a questionnaire during the biennial interview. For all suspected CVD events, three experienced physician adjudicators who were blinded to the study design reviewed the medical records. Myocardial infarction was defined as the recording of ICD-10 codes I21. Stroke was defined as the recording of ICD-10 codes I63, or I60 to I61. The vital status was obtained from Hebei Provincial Vital Statistics Offices or directly contacting the participants’ family members.
Statistical Analysis
Baseline characteristics of Kailuan study participants were described across TyG variability tertiles. Continuous variables were summarized as mean ± SD or median (interquartile range) depending on variable distribution, and categorical variables as count (proportion). Continuous variables were compared using one-way ANOVA or the Kruskal-Wallis test, and categorical variables were compared using the χ2 test across tertiles of TyG variability.
We analyzed risk of CVD according to the tertiles of baseline, mean, and variability of TyG index, and the combination of baseline TyG tertiles and TyG variability tertiles. Person-years of follow-up for each participant were calculated as the amount of time from the index date to the first of the following events: incident CVD, death, or December 31st, 2019. Incidence rate of CVD per 1000 person-years was calculated. Time to first CVD event was first examined using Kaplan-Meier survival curves and compared using log-rank test. Then, multivariable-adjusted Cox proportional hazards regression models were used to estimate the hazard ratios (HR) and 95% confidence intervals (CI). The proportional hazards assumption was verified by inspecting negative log-log survival plots and no violation was observed. Model 1 was adjusted for age and sex. Model 2 was adjusted for age, sex, education, income, current smoking status, current drinking status, physical activity, BMI, diabetes, hypertension, chronic kidney disease, and hs-CRP. Model 3 was additionally adjusted for use of lipid-lowering agents, TC, LDL-C, and LDL-C. Model 4 was additionally adjusted for baseline TyG index. In the trend test, the categorical variable (i.e., TyG variability tertile) was statistically examined as an ordinal variable (continuous variable) in Cox regression model.
Net reclassification improvement (NRI) for survival data were used to estimate the improvement in discrimination and reclassification after adding baseline and variability of TyG levels to the conventional clinical risk model [24]. A risk threshold of 5% for 9-year CVD risk (median follow-up time) was used to calculate categorical NRI. 95% CI for continuous and categorical NRI were estimated with 500 bootstrap replications.
To estimate the population impact of TyG measures on CVD risk, we estimated the absolute risk difference for between baseline, mean, and variability of TyG index and incident CVD. Predicted cumulative incidence and absolute risk differences were presented as per 1000 population over 10 years and were estimated by use of flexible parametric survival models on the cumulative hazard scale [25], similar to what has been done previously [26]. We also plotted the adjusted cumulative incidence curves for CVD by extrapolating to 10 years using stpm2 and standsurv command in Stata, which were standardized to the baseline covariates.
Several sensitivity analyses were conducted as follows: (1) excluding lipid-lowering agent users; (2) excluding antidiabetic agent users; (3) excluding FBG ≥ 7.0 mmol/L or TG ≥ 1.7 mmol/L at baseline; (4) using Fine-Gray competing risk regression treating deaths as competing risk events [27] ; (5) excluding CVD events that occurred within 2 years of follow-up (2-y lag analysis); (6) other variability, including SD, coefficient of variation (CV), and independent of the mean (VIM) were calculated [18]; (7) to assess the influence of unmeasured confounding, E-value, which is defined as the minimum strength of association, was calculated based on the estimated HR and 95% CI for CVD [28].
Statistical analyses were conducted using STATA MP, version 16.0 (StataCorp) and R software, version 4.1.3. All P-values were 2-sided and a P < 0.05 was considered statistically significant, unless otherwise stated.