Data were obtained from the Kailuan study, which is prospective cohort study conducted in the Kailuan community in Tangshan, China. Detailed study design and procedures have been described previously.[22-24] Briefly, since June 2006, a total of 101,510 participants (81,110 men and 20,400 women, aged 18-98 years) were enrolled in the baseline survey and completed questionnaires, health assessments and laboratory tests every 2 years. In the present study, we examined data for 49,749 participants with normal weight (BMI 18.5-24.9 kg/m2) according to the Guidelines and Prevention and Control of Overweight and Obesity in Chinese Adults. The TyG trajectories were developed from 2006 to 2012 to predict incident CVD risk after 2012 (Figure 1). Participants were excluded if they had MI or stroke prior to or in 2012, had missing data on FBG or TG at baseline or at all three measurements at 2008, 2010, and 2012, or with extreme outliers on the TyG index. Following these exclusions, we included 40,473 participants in the current analysis (Figure S1). The study was performed according to the guidelines of the Helsinki Declaration and was approved by the Ethics Committee of Kailuan General Hospital (approval number: 2006-05) and Beijing Tiantan Hospital (approval number: 2010-014-01). All the participants agreed to take part in the study and provided written informed consent.
Demographic data, including age, sex, educational levels, and family income, lifestyle behaviors, including smoking, alcohol drinking and physical activity, as well as medical and medication history were collected via standardized questionnaires. Active physical activity was defined as physical activity ≥4 times per week and ≥20 minutes at a time. BMI was calculated as weight (kg)/height (m)2. Blood pressure was measured in the seated position using a mercury sphygmomanometer, and the average of three measurements of the systolic blood pressure (SBP) and diastolic blood pressure (DBP) were recorded. All the blood samples were analyzed using an auto-analyzer (Hitachi 747, Hitachi, Tokyo, Japan) on the day of the blood draw. The biochemical indicators tested included FBG, serum lipids, serum creatinine, and high-sensitivity C-reactive protein (hs-CRP).
Hypertension was defined as SBP ≥140 mm Hg or DBP ≥90 mm Hg, any use of the antihypertensive drug, or a self-reported history of hypertension. Diabetes was defined as FBG ≥7.0mmol/L, any use of glucose-lowering drugs, or a self-reported history of diabetes. Dyslipidemia was defined as any self-reported history or use of lipid-lowering drugs, or total cholesterol (TC) ≥5.17 mmol/L.
Assessment of the TyG index
The TyG index was calculated as ln (fasting TG [mg/dl] × FBG [mg/dl]/2), as descried previously.[26-28] The TyG index trajectories during 2006 to 2012 were the primary exposure of the current study. As secondary exposure, the annual increase of the TyG index (TyG index slope) was calculated using linear regression model in which the TyG index in 2006, 2008, 2010, and 2012 was the response variable and follow-up duration (years) was the independent variable.
Assessment of incident CVD
The outcome in our study were the first occurrence of CVD. The types of CVD included total stroke, ischemic stroke (IS), hemorrhagic stroke (HS), and myocardial infarction (MI). We used ICD-10th revision codes to identify CVD cases (I63 for IS, I60-I61 for HS, and I21 for MI). The database of CVD diagnosed was obtained from the Municipal Social Insurance Institution and Hospital Discharge Register and was updated annually during the follow-up period. An expert panel collected and reviewed annual discharge records from 11 local hospitals to identify patients who were suspected of CVD. Incident stroke was diagnosed based on neurological signs, clinical symptoms, and neuroimaging tests, including computed tomography or magnetic resonance, according to the World Health Organization criteria. MI was diagnosed according to the criteria of the World Health Organization on the based on the clinical symptoms, changes in the serum concentrations of cardiac enzymes and biomarkers, and electrocardiographic results.[16, 30]
The TyG index trajectories during 2006 to 2012 were identified using latent mixture modeling within the PROC TRAJ procedure in SAS. We used a censored normal model appropriate for continuous outcomes. We initiated a model with 1 trajectory and then added 2, 3, 4 and up to 5 trajectory patterns. Model fit was assessed using the Bayesian information criterion (BIC), with the smallest negative number indicating the best fit model. We then compared the model with different functional forms. Cubic, quadratic, and linear terms were considered and evaluated based on their significance level, starting with the highest polynomial. In our final model, we had two patterns with linear order terms and three patterns with up to quadratic order terms.
Baseline characteristics were described as mean with standard deviation for continuous variables and frequency with percentage for categorical variables. Group differences were compared with Kruskal-Wallis test and chi-square test. The person-years were determined from the date when the message was collected at baseline to either the date of CVD diagnosis, death, or the end of the follow-up (December 31, 2019), whichever came first. The CVD and its subtypes probabilities were estimated by Kaplan-Meier method and the differences among groups were evaluated by log-rank test.
Cox proportional hazards model was used to investigate the association between exposures (the TyG index trajectories and annual increase rate of the TyG index from 2006 to 2012) and risk of developing CVD. According to Schoenfeld residuals and log-log inspection, the models met the proportional assumption criteria. Three models were built: model 1 was adjusted for age and sex; model 2 was further adjusted for education, income, smoking status, drinking status, physical activity, BMI, SBP, DBP, high density lipoprotein cholesterol (HDL-C), low density lipoprotein cholesterol (LDL-C) and hs-CRP; and model 3 was further adjusted for history of hypertension, diabetes, dyslipidemia, use of antihypertensive agents, antidiabetic agents, and lipid-lowering agents.
For more detailed of the association between annual increase rate of the TyG index from 2006 to 2012 and subsequent CVD risk, we also used restricted cubic spline models with 4 knots defined at the 5th, 35th, 65th, and 95th percentiles of the TyG index. Several sensitivity analyses were performed to validate the robustness of the results. First, we further adjusted the TyG index at baseline to control the regression-to-the mean influence. Second, we performed competing risk model considering non-CVD deaths as competing risk events. Third, we excluded participants who developed CVD cases within the first two years of follow up to minimize potential reverse causation. Likelihood ratio tests were conducted to examine statistical interactions between TyG index trajectories and age (<60 vs. ≥60 years), sex, hypertension (yes vs. no), diabetes (yes vs. no), and dyslipidemia (yes vs. no) in relation to CVD risk, by comparing -2 log likelihood c2 between nested models with and without the cross-product terms.
All analyses were conducted using SAS version 9.4 (SAS Institute Inc., Cary, NC, USA). A two-sided P<0.05 was considered statistically significant.