Impacts of triglyceride-glucose index on prognosis of patients with type 2 diabetes mellitus and non-ST-segment elevation acute coronary syndrome: results from an observational cohort study in China

The present study is a single-center, observational, retrospective cohort study among patients with diabetes who were diagnosed with NSTE-ACS and treated with elective PCI at Beijing Anzhen Hospital between January and December 2015. The exclusion criteria were listed as follows: (1) explicit or suspected type 1 diabetes mellitus (T1DM); (2) missing clinical data; (3) history of coronary artery bypass grafting (CABG), cardiogenic shock, chronic infectious disease and malignant tumor; (4) taking TGs-lowering medications before admission (such as fibrates); (5) extreme body mass index (BMI) (BMI > 45 kg/m²) and suspected familial hypertriglyceridemia [plasma TGs ≥ 500 mg/dL (5.65 mmol/L)]; (6) renal dysfunction with estimated glomerular filtration rate (eGFR) < 30 mL/(min * 1.73 m²) or treated with renal replacement therapy, severe hepatic insufficiency with alanine transaminase (ALT) or aspartate transaminase (AST) ≥ 5 upper limit of normal; (7) PCI failure, PCI-related complications, and in-hospital death. Ultimately, a cohort of 798 patients who met the enrollment principles were included for the present analyses (Fig. 1).

Data of demographic and clinical characteristics, including age, sex, weight, height, heart rate, blood pressure [systolic blood pressure (SBP) and diastolic blood pressure (DBP)], medical history, family history, and medical treatment were extracted from the medical information recording system of Beijing Anzhen Hospital. BMI was calculated as follows: BMI = weight (kg)/[height (m)]². Criteria for diabetes include: (1) previously diagnosed diabetes under treatment of antidiabetic medication (diet, oral agents, and/or insulin); (2) the typical symptoms of diabetes with a random blood glucose ≥ 11.1 mmol/L, and/or FBG ≥ 7.0 mmol/L, and/or 2-h blood glucose after oral glucose tolerance test (OGTT) ≥ 11.1 mmol/L [23]; (3) glycosylated hemoglobin A1c (HbA1c) level ≥ 6.5% on admission [24]. NSTE-ACS was composed of non-ST-segment elevation myocardial infarction (NSTEMI) and unstable angina (UA), definitions of which were determined by appropriate guidelines [25]. NSTEMI was defined as having symptoms of ischemia and elevated cardiac troponin I (cTnI), and without an elevation of ST-segment. UA was diagnosed as ischemic symptoms at rest, or exacerbated or new-onset symptoms with transient ischemic ST-segment shifts, and without release of myocardial enzymes related to myocardial necrosis. Patients with SBP ≥ 140 mmHg and/or DBP ≥ 90 mmHg, or those receiving anti-hypertensive treatments were considered having hypertension. Peripheral vascular disease (PVD) was defined as aorta and other arteries than coronary arteries, with exercise related claudication, or reduced or absent pulsation, or angiographic stenosis of more than 50%.

Venous blood samples were collected after an overnight fasting on the day of the baseline coronary procedure. The routine hematology and biochemical parameters, including lipid profiles [TGs, total cholesterol (TC), low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C)], creatinine, uric acid, FBG, HbA1c, high-sensitivity C-reactive protein (hs-CRP), and other biomarkers, were determined by standard laboratory methods in central lab of Beijing Anzhen Hospital. Patients with fasting TC > 200 mg/dL, and/or LDL-C > 130 mg/dL, and/or TGs > 150 mg/dL, and/or HDL-C < 40 mg/dL at admission, and/or previously long-term use of lipid-lowering drugs were considered having dyslipidemia. The eGFR was calculated as previously described: eGFR [mL/(min * 1.73 m²)] = 186 * serum creatinine (mg/dL)^−1.154 * age^−0.203 (* 0.742 if female) [26]. Baseline TyG index was calculated based on fasting TGs and FBG values obtained at admission as previously reported: ln [fasting TGs (mg/dL) * FBG (mg/dL)/2] [11]. Left ventricular ejection fraction (LVEF) was evaluated by two-dimensional modified Simpson’s method using an ultrasonic cardiogram (Philips Company, Eindhoven, The Netherlands).

Coronary angiogram data were analyzed and recorded by at least two experienced cardiologists, and measurements of coronary artery lesion characteristics were obtained. The lesion characteristics were defined as follows: (1) multi-vessel lesion: more than two main coronary branches (vessel diameter ≥ 2 mm) with extent of stenosis ≥ 50%. (2) chronic total occlusion lesion: lesion with complete obstruction [thrombolysis in myocardial infarction (TIMI) flow grade 0] lasting longer than 3 months, which was judged from the previous medical history or coronary angiogram results. (3) diffuse lesion: a single stenotic lesion with a length of ≥ 20 mm. (4) bifurcation lesion: stenosis occurred adjacent to and/or involving the origin of a significant side branch that has too much functional value and so cannot be lost during the interventional procedure. (5) in-stent restenosis: stenosis of ≥ 50% occurring in the segment inside the stent, 5 mm proximal or distal to the stent [27]. The severity of coronary artery lesions was quantified by the synergy between PCI with taxus and cardiac surgery (SYNTAX) score. The SYNTAX score was calculated for each participant using the online calculator (http://​www.​syntaxscore.​com). PCI was performed in accordance with current practice guidelines in China [28], and detailed strategies were determined by experienced interventional cardiologists.

After baseline PCI, all patients were routinely followed up by trained professionals who were blinded to the baseline information at 3, 6, and 12 months and then annually for up to 36 months. The information about adverse prognostic events was obtained from patients or their family members by telephone questionnaire. The information was further confirmed by careful verification of corresponding medical records if necessary. The primary observational endpoint was defined as a composite of events including all-cause death, non-fatal myocardial infarction (MI) and ischemia-driven revascularization. The secondary observational endpoints were each component of the composite primary endpoint. MI was defined as elevated cardiac troponin higher than the upper reference limit with ischemia indicated from symptoms and/or electrocardiographic changes, with or without an elevation of ST-segment. Ischemia-driven revascularization was defined as the revascularization procedure associated with symptoms and/or electrocardiographic changes implicating ischemia. The first primary endpoint event that occurred during the follow-up was used for analysis in current study. For patients with multiple adverse outcomes occurring almost simultaneously during the follow-up, only the most severe event (all-cause death > non-fatal MI > ischemia-driven revascularization) was selected to perform our analyses. If the same event occurs multiple times, only the first occurrence was used for analysis.

Continuous variables were presented as mean ± standard deviation (SD) or median (25th and 75th percentiles: P25, P75) in the case of normal or non-normal distribution, and differences between the two groups were examined by independent-sample t-test or Mann–Whitney U test correspondingly. Categorical variables were described as counts (percentages) and compared by Pearson chi-square test (Pearson χ² test) or Fisher’s exact test appropriately. The Spearman’s rank correlation test or Pearson correlation test was used for evaluating the correlations between the TyG index and cardiovascular risk factors when appropriate. The Pearson correlation test was used to evaluate the correlation between two continuous variables with normal distribution, while the Spearman’s rank correlation test was applied in case that one or more of the variables being analyzed was non-normally distributed continuous variable or categorical variable. Receiver-operating characteristic (ROC) curve analysis was performed to determine the optimal cutoff point value of TyG index for predicting primary endpoint. The Kaplan–Meier survival analyses were performed to evaluate the incidence rate of adverse events between groups according to the optimal cutoff point of TyG index, and discrepancies between groups were evaluated by log-rank test. The predictive value of the variables for primary endpoint was evaluated by univariate and multivariate Cox proportional hazards analyses. The TyG index was analyzed in two ways: (1) as a categorical variable; and (2) as a continuous variable. In multivariate Cox proportional hazards analyses, four models were established to evaluate the predictive value of TyG index for primary endpoint, among which confounders were selected according to statistical significance (P < 0.2) in univariate analysis and clinical importance: (1) Model 1: adjusted for age, sex (female), BMI, SBP, DBP, smoking, drinking, duration of diabetes, dyslipidemia, prior MI, PCI, stroke and PVD; (2) Model 2: adjusted for variables included in Model 1 and diagnosis (NSTEMI), TC, HDL-C, eGFR, HbA1c, LVEF; (3) Model 3: adjusted for variables included in Model 2 and SYNTAX score, left main artery (LM) treatment, drug-coated balloon (DCB) use, complete revascularization and number of stents; (4) Model 4: adjusted for variables included in Model 3 and dual antiplatelet therapy (DAPT) at discharge, DAPT interruption in 12 months, statins at discharge, statins interruption in 12 months, oral hypoglycemic agents (metformin, alpha-glucosidase inhibitor, sulfonylurea, dipeptidyl peptidase 4 inhibitor) at discharge and insulin at discharge. The prognostic impact of TyG index for each component of primary endpoint was also assessed by using model 4. FBG and TGs were not introduced into multivariate analysis since the TyG index was calculated from them. Results of Cox proportional hazards analyses were presented as hazard ratio (HR) and 95% confidence intervals (CI). Further stratified analyses according to age (≤ 65 and > 65 years), sex, BMI (≤ 28 and > 28 kg/m²), hypertension, initial diagnosis (UA and NSTEMI), HbA1c (≤ 7 and > 7%), LDL-C (≤ 70 and > 70 mg/dL), and pre-admission medication including statins, oral hypoglycemic agents and insulin were employed to examine the consistence of the prognostic impact of TyG index for primary endpoint. The model used in the stratified analyses consisted of all covariates used in Model 4 except for the variables that were used for stratification. The interaction of TyG index and variables used for stratification was examined by likelihood ratio tests.

C-statistics including ROC curve analysis were performed to examine the incremental effects of TyG index on the predictive potential of the baseline risk model that including traditional risk factors. DeLong’s test was used to compare the area under the curve (AUC) from each of the models. We also calculated category-free net reclassification improvement (NRI) and integrated discrimination improvement (IDI) to determine the extent to which the addition of TyG index improves the predictive power of existing baseline risk model.

Statistical tests were performed with SPSS 23.0 (SPSS Inc., Chicago, Illinois, USA), the R Programming Language (version 3.5.1) and MedCalc version 19.1 (MedCalc Software, Belgium). A two-tailed P value < 0.05 was regarded as statistically significant.

Baseline characteristics of the total population and groups stratified by the occurrence of primary endpoint event were presented in Table 1. TyG index was significantly higher in patients with primary endpoint event compared with those without. Patients with a primary endpoint event showed higher age and SBP, longer duration of diabetes, and higher prevalence of dyslipidemia, previous MI and PCI history. In terms of laboratory indicators, participants with endpoint event had higher levels of TGs, TC, hs-CRP, FBG and HbA1c, but lower levels of HDL-C, eGFR and LVEF. As for the angiographic findings, those with an endpoint event showed higher proportions of LM disease, multi-vessel disease and other characteristics of complex coronary artery lesion. The SYNTAX score was significantly higher in subset with adverse prognosis. Correspondingly, more LM lesions were disposed and more coronary artery stents were implanted in patients with endpoint event. Moreover, the rate of complete revascularization was significantly lower in participants with adverse prognosis.

ROC curve analysis showed that the AUC of TyG index for predicting primary endpoint was 0.745 (95% CI 0.702–0.788, P < 0.001). The TyG index of 9.18 was determined as the optimal cutoff point for predicting primary endpoint with a sensitivity of 77.2% and a specificity of 62.8%. Baseline characteristics of groups according to the optimal cutoff point of TyG index were summarized in Table 2. Compared with patients in lower TyG index group, those with higher TyG index seemed to be younger, manifest higher levels of BMI and heart rate, and higher proportion of dyslipidemia. Laboratory indexes including TGs, TC, LDL-C, hs-CRP, uric acid, FBG and HbA1c were significantly higher in patients with higher TyG index, while HDL-C levels were relatively lower. In higher TyG index group, more patients were diagnosed as NSTEMI and prescribed insulin for treatment. Participants with higher TyG index also showed higher SYNTAX score compared to those with lower TyG index.

The Spearman’s rank or Pearson correlation analysis was performed to determine the correlation between the TyG index and traditional or commonly-used risk factors for cardiovascular disease. The TyG index was positively correlated with BMI, FBG, HbA1c, TGs, TC, LDL-C, uric acid, and hs-CRP, while negatively correlated with age and HDL-C (Table 3).

During the 36-month follow-up period, 180 (22.6%) endpoint events including 14 (1.8%) all-cause death, 37 (4.6%) non-fatal MI, and 129 (16.2%) ischemia-driven revascularization were documented to perform the present analyses. The incidence of adverse prognosis was compared between groups stratified by the optimal cutoff point of TyG index determined by ROC curve analysis. The incidence of primary endpoint, non-fatal MI and ischemia-driven revascularization increased significantly in patients with higher TyG index compared with those with lower TyG index (all chi-square P < 0.001). However, the all-cause death rate was similar between the two groups (chi-square P = 0.172) (Table 4).

Kaplan–Meier curves for incidence of primary endpoint and each component of it according to the optimal cutoff point of TyG index were shown in Fig. 2. Kaplan–Meier curves for primary endpoint showed a significant difference between the lower and higher TyG index group (Fig. 2a, Log-rank P < 0.001). The difference was mainly driven by the increased incidence of non-fatal MI and ischemia-driven revascularization (Fig. 2c, d, both Log-rank P < 0.001). Kaplan–Meier curves for all-cause death between the lower and higher TyG index group failed to reach statistical significance (Fig. 2b, Log-rank P = 0.167).

In multivariate Cox proportional hazard analysis, four models (Model 1–4 as described above) including variables that had statistical significance (P < 0.2) and/or clinical importance were constructed to evaluate the predictive potential of TyG index for primary endpoint. After adjusting for confounding variables, higher TyG index remained to be an independent risk predictor of primary endpoint, despite of regarding TyG index as a nominal or continuous variable (all P < 0.001 in Model 1–4) (Table 5). The detailed information of Model 4 was shown in Additional file 1: Table S1.

The predictive value of TyG index for each component of primary endpoint was also evaluated by using model 4. The results showed that a 1-unit increase of TyG index was independently associated with higher risk of non-fatal MI and ischemia-driven revascularization [HR (95% CI) for non-fatal MI: 3.332 (1.730–6.415), P < 0.001; HR (95% CI) for ischemia-driven revascularization: 3.021 (2.167–4.211), P < 0.001]. However, higher TyG index levels failed to be a predictor of all-cause death, which was consistent with the results of Kaplan–Meier curves (Table 6).

Further evaluation of the risk stratification value of TyG index for primary endpoint was performed in various subclasses of the study population. Increased TyG index (per 1-unit) was consistently related to primary endpoint in various subgroups, including age ≤ 65 or > 65 years, female or male, BMI ≤ 28 or > 28 kg/m², with or without hypertension, UA or NSTEMI, HbA1c ≤ 7 or > 7%, LDL-C ≤ 70 or > 70 mg/dL, with or without pre-admission medication including statins, oral hypoglycemic agents and insulin (Fig. 3). Interestingly, the predictive value of TyG index seemed to be more prominent in patients with BMI > 28 kg/m² [HR (95% CI) BMI > 28 kg/m² 5.513 (3.631–8.370) vs. BMI ≤ 28 kg/m² 2.178 (1.524–3.111), P for interaction < 0.001] and without pre-admission insulin therapy [HR (95% CI) without insulin 4.011 (2.827–5.691) vs. with insulin 2.255 (1.461–3.479), P for interaction = 0.024] (Fig. 3).

The addition of TyG index had a significant incremental effect on the AUC obtained from baseline risk model that consisted of risk factors including age, sex (female), smoking, SBP, prior MI, prior PCI, TC, HDL-C, eGFR, LVEF, SYNTAX score, LM treatment, complete revascularization, number of stents and statins at discharge (AUC: baseline risk model, 0.800 vs. baseline risk model + TyG index, 0.856, P for comparison < 0.001) (Table 7, Fig. 4d). Moreover, the addition of TyG index significantly improved the reclassification and discrimination ability beyond the baseline risk model with a category-free NRI of 0.346 and an IDI of 0.087 (both P < 0.001) (Table 8). Adding TGs to the baseline risk model also had a significant incremental effect on prognostic prediction (AUC: baseline risk model, 0.800 vs. baseline risk model + TGs, 0.842, P for comparison < 0.001; category-free NRI: 0.318, P < 0.001; IDI: 0.067, P < 0.001) (Tables 7 and 8, Fig. 4c). However, the addition of glycemic index including FBG or HbA1c did not have a significant incremental effect on the AUC of the baseline risk model (Table 7, Fig. 4a, b). A significant but relatively minor incremental effect on the reclassification and discrimination ability was found after adding HbA1c to the baseline risk model (Table 8).