Triglyceride-glucose index and triglyceride to high-density lipoprotein cholesterol ratio as potential cardiovascular disease risk factors: an analysis of UK biobank data

UK Biobank is a community-based prospective cohort of over half a million individuals aged 40–69 years at recruitment between 2006 and 2010. Potential participants attended one of 22 assessment centers across England, Wales, and Scotland where they underwent physical examinations, provided biological samples, and completed baseline questionnaires, as described in detail elsewhere [24]. After excluding those who had missing data on the TyG index or TG/HDL-C ratio (n = 73,128), withdrew at the time of the study (n = 1298), or had prevalent CVD at baseline (n = 24,744), a total of 403,335 individuals remained for the final analysis.

UK Biobank was constructed under ethical approval obtained by the North West Multi-Centre Research Ethics Committee (REC reference: 11/NW/03820) and all participants provided written informed consent prior to participation. The current analyses were carried out under Application Number 52632.

We used the baseline touchscreen questionnaire to derive information on several potential confounders: age, sex, ethnicity, Townsend Deprivation Index, current smoking status, physical activity (< 150 or  ≥ 150 min/week based on the total time spent in moderate physical activity in minutes each week [25]), and medication use at baseline (aspirin, insulin, and antihypertensive and cholesterol-lowering medications). The Townsend deprivation index is a composite measure of deprivation based on unemployment, non-car ownership, non-home ownership, and household overcrowding. It is derived from the residential postcode, with a negative value representing high socioeconomic status [26]. Two measurements of systolic and diastolic blood pressure were taken using the Omron HEM-7015IT digital blood pressure monitor or a manual sphygmomanometer, and the mean of the two measurements was used for analysis. Body mass index was calculated as the weight in kilograms (kg) divided by the square of the height in meters (m²).

Peripheral venous blood samples were collected at baseline from all participants, and collection procedures for the UK Biobank study were validated [27]. Blood samples were taken at random due to their future applicability to a wide range of diseases and the difficulty in collecting and processing fasting blood samples in a very large population with a distributed assessment center setting [27]. Non-fasting serum biochemical measurements including glucose, triglycerides (TG), total cholesterol, high density lipoprotein cholesterol (HDL-C), low density lipoprotein cholesterol (LDL-C), high sensitivity C-reactive protein, creatinine, and uric acid were performed on a Beckman Coulter AU5800 clinical chemistry analyzer at the central laboratory. Coefficients of variation for concentrations of TG, high sensitivity C-reactive protein, and creatinine were less than 3% and for glucose, total cholesterol, HDL-C, LDL-C, and uric acid were less than 2%. Glycated hemoglobin (HbA_1c) was measured by high-performance liquid chromatography analysis on a Bio-Rad VARIANT II Turbo. The TyG index was calculated using the formula: ln [triglycerides (mg/dL) × glucose (mg/dL)/2] [28]. The TG/HDL-C ratio was calculated as TG (mg/dL) divided by HDL-C (mg/dL). Estimated glomerular filtration rate (eGFR) was calculated using the Chronic Kidney Disease Epidemiology Collaboration creatinine equation [29], based on the measurement of creatinine at baseline visit. Chronic kidney disease was defined as an eGFR < 60 mL/min/1.73 m² [30]. Details of these assessments can be found in the UK Biobank’s online protocol (www.​ukbiobank.​ac.​uk).

The primary outcome of this study was incident CVD, defined as fatal or non-fatal coronary heart disease (CHD) or stroke. The secondary outcomes were individual diagnoses of CHD and stroke. Participant follow-up started at inclusion in the UK Biobank and was censored on March 31, 2017 or on the date of the first CHD or stroke. The date of the first incident CHD or stroke after baseline was ascertained through record linkage with hospital episode statistics in England, Scotland, and Wales and national death registers. Incident CHD was defined by the 10th revision of the International Classification of Diseases (ICD-10) codes I20–I25. Incident stroke was defined by ICD codes I60–I64. Prevalent comorbidities including dyslipidemia (ICD-10 codes E78), type 1 diabetes (ICD-10 code E10), type 2 diabetes (ICD-10 code E11), retinopathy (ICD-10 codes E10.3, E11.3, E12.3, E13.3, E14.3, H28.0, H33, H35.3, H36.0, H40–H42) and hypertension (ICD-10 codes I10–I15) were captured from the health records (Additional file 1: Appendix S1).

The data were expressed as mean (SD) for continuous variables and frequency (percentage) for categorical variables. Participants were stratified into four groups according to the quartiles of the TyG index and TG/HDL-C ratio at baseline. Trends across quartiles were tested by the generalized linear regression analysis for continuous variables and the Cochran-Armitage trend chi-square test for categorical variables, respectively. Pearson’s (for continuous variables) and Point-biserial (for dichotomized variables) correlation tests were used to assess the correlations of the TyG index and TG/HDL-C ratio with participant characteristics.

Kaplan-Meier cumulative incidence plots were generated to assess the relationship between the quartiles of the TyG index and TG/HDL-C ratio and incident CVD (including total CVD, CHD, and stroke) during follow-up, and the log-rank test was used for statistical assessment. Cox proportional hazard models were applied to calculate hazard ratios and 95% confidence intervals of higher quartiles in relation to the lowest quartiles and of 1-SD increments in log-transformed TyG index and TG/HDL-C ratio for all endpoints. P values for linear trends in hazard ratios across the quartiles of the TyG index and TG/HDL-C ratio were tested using the median value within each quartile as the predictor. Three models were established with incremental degrees of adjustment for potential confounders of CVD: model 1 was adjusted for none, model 2 was adjusted for baseline age (years), sex, ethnicity (White and others), region (England, Scotland, and Wales), and Townsend Deprivation Index, and model 3 was adjusted for the same variables as model 2 and also for current smoking status (yes or no), physical activity (< 150 or  ≥ 150 min/week), body mass index (kg/m²), systolic blood pressure (mm Hg), total cholesterol (mg/dL), LDL-C (mg/dL), uric acid (mg/dL), HbA_1c (mmol/mol), eGFR (mL/min/1.73 m²), high sensitivity C-reactive protein (mg/L), aspirin use (yes or no), undergoing insulin treatment (yes or no), use of antihypertensive medication (yes or no), use of cholesterol-lowering medication (yes or no), prevalent retinopathy (yes or no), and chronic kidney disease (yes or no). To determine whether there was a nonlinear dose-response relationship of the TyG index and TG/HDL-C ratio with the risk of CVD after multivariable-adjustment, restricted cubic splines were fitted, with four knots placed at the 5th, 35th, 65th, and 95th percentiles and the 1% highest and lowest TyG index and TG/HDL-C ratio observations were trimmed.

Mediation analysis was performed using the publicly available SAS macro %mediate (http://​www.​hsph.​harvard.​edu/​donna-spiegelman/​software/​mediate/​) [31] to evaluate the proportional contribution of prevalent dyslipidemia, diabetes, and hypertension on the associations of the TyG index and TG/HDL-C ratio with CVD risk. In brief, the mediation proportion is a statistical measure used to estimate how much of the total exposure-outcome association is explained by a particular mediator [32]. Mediation analysis models were adjusted for the same set of confounders as in model 3 of the primary analyses. In sensitivity analyses, we first adjusted for prevalent dyslipidemia, diabetes, and hypertension in addition to the variables in model 3. Moreover, to check for reverse causation, we excluded individuals who developed CVD within the first 3 years of follow-up.

Participant characteristics both overall and stratified by the quartiles of the TyG index were presented in Table 1, and for the TG/HDL-C ratio in Additional file 1: Table S1. Of 403,335 included participants, 222,586 (55.2%) were women, 381,880 (94.7%) were White, and the mean (SD) age at baseline was 56.2 (8.1) years. The prevalence of dyslipidemia, type 1 diabetes, type 2 diabetes, and hypertension were 6.7%, 0.3%, 3.5%, and 13.9%, respectively. The median TyG index and TG/HDL-C ratio were 8.67 (IQR 8.31–9.07) and 2.39 (IQR 1.49–3.90), respectively. Higher TyG index and TG/HDL-C ratio were both associated with male sex, older age, White ethnicity, lower socioeconomic status, smoking, less physical activity, higher body mass index and blood pressure, increased serum concentrations of uric acid, glycated hemoglobin, total cholesterol, LDL-C, C-reactive protein, and creatinine="null">, lower eGFR, and more frequent use of aspirin, insulin, and antihypertensive and cholesterol-lowering medication at baseline. Moreover, individuals with higher TyG index or TG/HDL-C ratio levels had greater prevalence of retinopathy="null">, chronic kidney disease, dyslipidemia, type 1 diabetes, type 2 diabetes, and hypertension than those with lower levels (all P < 0.001 for trend). The results of Pearson’s or Point-biserial correlations of the TyG index and TG/HDL-C ratio with participant characteristics were displayed in Additional file 1: Table S2. Significant differences in participant characteristics were observed in England, Scotland, and Wales (P < 0.001 for difference), except for triglycerides, the TG/HDL-C ratio, and insulin use (Additional file 1: Table S3).

During a median follow-up of 8.1 years (IQR 7.4–8.8, totaling 3,222,292 person-years), 19,754 (4.9%) participants developed CVD, including 16,404 (4.1%) cases of CHD and 3976 (1.0%) cases of stroke. There were differences in rates of incident CVD in England, Scotland, and Wales (Additional file 1: Table S3). Kaplan-Meier curves indicated a graded increased risk of CVD with higher quartiles of the TyG index and TG/HDL-C ratio (all log-rank P < 0.001, Fig. 1). In unadjusted Cox regression analyses, both higher TyG index and TG/HDL-C ratio were associated with increased risks of total CVD, CHD, and stroke, and these associations persisted after adjusting for sociodemographic factors. After full covariate adjustment, the hazard ratios (95% confidence intervals) of total CVD in higher quartiles versus the lowest quartiles were 1.05 (1.00–1.10), 1.05 (1.00–1.10), and 1.19 (1.14–1.25), respectively, for the TyG index, and 1.07 (1.02–1.13), 1.13 (1.07–1.18), and 1.29 (1.23–1.36), respectively, for the TG/HDL-C ratio. There were significant trends toward an increasing risk of CVD across the quartiles of the TyG index and TG/HDL-C ratio (both P trend  < 0.001). Each 1-SD increment in log-transformed values of the TyG index and TG/HDL-C ratio was associated with 8% and 12% increase in the risk of CVD, respectively (Tables 2 and 3). Similarly, in the fully adjusted model, hazard ratios (95% confidence intervals) of CHD in higher quartiles versus the lowest quartiles were 1.07 (1.01–1.13), 1.09 (1.03–1.15), and 1.25 (1.18–1.32), respectively, for the TyG index, and 1.11 (1.05–1.17), 1.18 (1.11–1.24), and 1.37 (1.30–1.45), respectively, for the TG/HDL-C ratio. The trends across quartiles were significant for both the TyG index and TG/HDL-C ratio (both P trend < 0.001). Per 1-SD increment in log-transformed values of the TyG index and TG/HDL-C ratio was associated with 10% and 15% increase in the risk of CHD, respectively. After full adjustment, the associations of the TyG index and TG/HDL-C ratio with stroke were attenuated to non-significance (Tables 2 and 3).

Multivariable-adjusted restricted cubic spline analyses suggested a nonlinear association of the TyG index with total CVD, CHD, and stroke. They also indicated a nonlinear association between the TG/HDL-C ratio and CHD. There was no evidence of a nonlinear association of TG/HDL-C ratio with total CVD or stroke. There was evidence of a significant linear relationship of TG/HDL-C ratio with total CVD (P for linearity < 0.001), but not with stroke (Fig. 2). In sensitivity analyses additionally adjusted for prevalent dyslipidemia, diabetes, and hypertension (Additional file 1: Table S4) or with exclusion of 6080 incident CVD cases with less than 3 years of follow-up (Additional file 1: Table S5), we still observed significant associations of the TyG index and TG/HDL-C ratio with the risk of total CVD and CHD (all P < 0.001).

In mediation analyses (Table 4), dyslipidemia accounted for 45.8% and 36.1% of the associations of the TyG index with incident CVD and CHD, respectively, and 40.0% and 33.5% of the TG/HDL-C ratio’s associations with CVD and CHD, respectively. The proportions mediated through type 2 diabetes were 27.0% and 20.6% for the associations of the TyG index with incident CVD and CHD, respectively, and 11.8% and 9.6% for the TG/HDL-C ratio’s associations with CVD and CHD, respectively. Hypertension also played a significant mediating role in the associations of the TyG index with incident CVD and CHD (proportions mediated: 15.0% and 11.1%, respectively), and in the associations of the TG/HDL-C ratio with incident CVD and CHD (proportions mediated: 13.3% and 10.6%, respectively). The combined mediation effect of dyslipidemia, type 2 diabetes, and hypertension accounted for 56.5% and 43.6% of the total effect of TyG index on CVD and CHD, respectively, and 46.6% and 38.3% of the total effect of TG/HDL-C ratio on CVD and CHD, respectively (All P < 0.001, Table 4). There were no significant mediation effects observed for prevalent type 1 diabetes.