Relationship between the triglyceride-glucose index and risk of cardiovascular diseases and mortality in the general population: a systematic review and meta-analysis

The PubMed, Embase, and the Cochrane Library databases were searched for studies from inception to April 16, 2022, without language restrictions. The search terms were as follows: (1) Exposure: For the TyG index: “TyG index” or “triglyceride-glucose index”. (2) Outcomes: For CVD: “cardiovascular disease”, “cardiovascular diseases”, “coronary artery disease”, “coronary disease”, “coronary arteriosclerosis”, “myocardial infarction”, “heart attack”, “heart failure”, “heart decompensation”, “atrial fibrillation”, “sudden cardiac death”, “arrhythmia”, “cardiomyopathy”, “hypertrophic cardiomyopathy”, and “dilated cardiomyopathy”. For mortality: “mortality”, “death”, “cardiovascular death”, and “all-cause mortality”. We did not apply keywords for the population and study design in the database search to capture as many studies as possible that met the inclusion criteria. The detailed search strategy is described in Additional file 1: Table S2.

We used Endnote X9 software (Thomson Reuters, New York, NY, USA) to organize all studies. After removing duplicates, titles and abstracts were checked, and an initial screening of the relevant literature was performed, and the full text of studies was retrieved for full-text assessment of eligibility for inclusion.

According to the PICOS (population, intervention, comparison, outcome, and study design), the criteria for considering studies for this review were as follows: (1) types of participants: adult (age > 18 years) general population. (2) Exposure and comparator: high versus low TyG index level. (3) Outcomes: all kinds of cardiovascular diseases, composite of cardiovascular disease, cardiovascular mortality, and all-cause mortality. (4) Types of studies: random controlled trails (RCTs), post hoc analyses of RCTs, or observational cohort studies.

Our exclusion criteria included: Cross-sectional studies due to the high risk for bias; Studies reporting the association of the TyG index with prognosis in established conditions (e.g., cardiovascular diseases, diabetes); Studies focusing on children and adolescents; Studies not reporting multivariable adjusted outcomes. In addition, articles without sufficient data (reviews, editorials, preclinical studies) and studies not relevant for the purpose of the current meta-analysis were excluded. If the same population was used in multiple studies, we selected the article with the most information or the largest sample size.

Literature searches, data extraction and quality assessment of the included studies were conducted independently by two authors (Z. Q-T and X-L) based on predetermined criteria. The following data were extracted: (1) author names, year of publication, and country; (2) study design and duration of follow-up; (3) study population characteristics, including data source, sample size, mean age, and sex; (4) outcomes and determination of outcomes; and (5) risk estimates and adjustments.

The Newcastle–Ottawa Scale (NOS) was used to assess the quality of included cohort studies. Scores ranged from 0 to 9 to assess article selection, comparability, and outcomes. Studies with an NOS score of > 6 were considered to be of high quality [22].

We used RevMan software version 5.4 (The Cochrane Collaboration 2014, Nordic Cochrane Center Copenhagen, Denmark) for statistics and analysis. We used risk estimates of hazard ratios (HRs) and their corresponding 95% confidence intervals (CIs) as general indicators of the association between the TyG index and cardiovascular disease in the general population. The TyG index was calculated using the following equation: TyG = Ln (TG [mg/dL] × fasting glucose [mg/dL]/2) [23, 24]. The TyG index was analyzed as a categorical variable with the highest TyG index levels compared to the group with the lowest TyG index levels. We also conducted a supplemental analysis of the moderate (if there were four groups, we chose the third highest TyG index level group) versus lowest TyG index levels. If the article reported sex subgroups, we first used a random effects model to combine to obtain effect sizes. The TyG index was analyzed as a continuous variable, and the units of the TyG index (per 1 unit) were standardized. We estimated linear trends and 95% CIs from the natural logarithm of the effect size and CI for the TyG index category, as described by Greenland and Longnecker [25]. A random-effects model was used to improve reliability [26]. Robust error meta-regression methods were used to fit nonlinear dose–response analysis described by Xu and Doi et al. [27, 28]. This method is based on a “one-stage approach”, which treats each study as a cluster of the whole sample and considers the within-study correlations by clustered robust error. It requires known levels of the TyG index and risk estimates with variance estimates for at least two quantitative exposure categories [12]. As described in previous articles [29, 30], if quantitative TyG values were not reported, we used the median or mean TyG of the categories. If the median or mean TyG was not provided and reported in ranges, we estimated the midpoint of each category by averaging the lower and upper boundaries of that category. If the highest or lowest category was open-ended, we assumed that the open-ended interval length was the same as the adjacent interval [14].

We assessed the degree of heterogeneity using the I² test, whereby 25%, 50%, and 75% represented low, moderate, and high heterogeneity, respectively [29]. Publication bias was evaluated using funnel plots. Sensitivity analyses were performed by omitting each study in turn. Subgroup analyses were performed stratified by sex, age, and diabetes if the number of included studies exceeded ten for each outcome [31]. P < 0.05 was considered statistically significant.

We assessed the quality and strength of the evidence for each outcome according to the Grading of Recommendations Assessment, Development and Evaluation (GRADE) method [32, 33]. Two authors rated the quality of the evidence for each outcome separately. We used GRADEprofiler software and provide evidence profile tables. We present the results of the outcomes as described in the outcome metric type section; footnotes are used to justify all decisions to lower or raise the quality of evidence.

The flow chart shows the process of the database search (Fig. 1). An initial search of 594 articles (PubMed = 274; Embase = 307; Cochrane Library = 13) was conducted according to the search protocol. Of these, 174 duplicate articles were excluded; in addition, 336 articles were excluded as irrelevant after screening titles and abstracts. Subsequently, the full texts of 84 articles were reviewed. Of these, 72 were further excluded for the following reasons: (1) insufficient data (n = 9); (2) certain publication type for which no data were available (n = 5); (3) without an appropriate study design (n = 5); (4) without an appropriate population (n = 20); (5) without an appropriate exposure (n = 6); and (6) did not report the target outcome (n = 27). Ultimately, no RCT studies met the inclusion criteria, 12 cohort studies were used for this meta-analysis [18‐20, 34‐42]. All excluded studies and the reasons (n = 72) are shown in Additional file 1: Table S3.

The basic characteristics of all included observational articles are summarized in Table 1. The twelve cohort studies [18‐20, 34‐42], including 6,354,990 participants with a mean age between 44.9 and 70.5 years, were published from 2014 to 2021; sample sizes ranged from 5014 to 5,593,134. Of these, three studies reported CAD, two reported MI, six reported CVD, and four reported mortality outcomes (cardiovascular [CV] mortality had three and all-cause mortality had four). The definitions of CVD in each study are listed in Additional file 1: Table S4. All articles reported the follow-up time, with eleven exceeding 5 years [18‐20, 34‐39, 41, 42]. The span of follow-up ranged from 4.1 years to 16.1 years. Of these studies, ten studies were performed in Asia (four in China [19, 20, 37, 38], four in Korea [34‐36, 40] and two in Iran [18, 39]), one was from the Americas (The United States of America [USA] [42]) and one was from Europe (Spain [41]). In addition, six were prospective cohort studies [18‐20, 39‐41]; the others were retrospective cohort studies [34‐38, 42].

Only one study had an NOS score of 6 [42], which raised concerns about selection and outcome bias. The remaining studies were all of high quality, with a score higher than 7 (Additional file 1: Table S5). All studies adjusted age and sex, with adjustment for other confounders varying considerably.

Sensitivity analysis by deleting one study showed consistent results (Additional file 1: Fig. S2). Due to the limited number of included studies (N < 10), subgroup and publication bias analyses were not performed according to the guidelines and predefined criteria.

We determined the quality of the evidence using GRADE. CVD and CAD outcomes were upgraded due to a dose–response relationship and the presence of a large effect for CAD. As there was greater heterogeneity in CVD (I2 = 82%) and mortality (CV mortality I² = 76% and all-cause mortality I² = 87%), and all outcomes not process analysis of publication bias, downgrades were given. Eventually, of the five outcomes included, GRADE assessment indicated very low certainty for CVD, MI, cardiovascular mortality and all-cause mortality, and moderate certainty for CAD (Additional file 1: Tables S7, S8).

Wang et al. found that the TyG index a positive association between TyG index and hypertension risk in the general adult population [51], and Li et al. demonstrated that the index can predict adverse events in patients with acute coronary syndrome [66]. Previous meta-analyses have also examined the relationship between the TyG index and the prognosis of patients with acute coronary syndromes [12, 66]. Compared to previous meta-analysis, the population in our study was the general population. In addition, our study included more types of cardiovascular diseases and mortality, and to produce more accurate results, the association between various cardiovascular diseases and the TyG index was analyzed separately.

TyG may be a reliable indicator of insulin resistance [23], and a prior study demonstrated its validity as an alternative marker of insulin resistance [67], with high sensitivity (96.5%) and specificity (85.0%) [46]. Therefore, insulin resistance can be determined by measuring the TyG index and inferring the incidence of CVD. Insulin resistance is characterized by a low degree of systemic inflammation, which can lead to endothelial dysfunction [51]. At the same time, insulin resistance contributes to atherosclerosis and plaque progression through a variety of mechanisms, including changes in classical cardiovascular disease risk factors and downregulation of insulin signaling pathways [68].

Several risk factors that epitomize the development of cardiovascular disease are also closely related to TyG index results, such as coronary artery calcification (CAC), which was not included in our meta-analysis. Park's study (1175 participants followed up for 4.2 years) showed the TyG index to be an independent predictor of CAC progression (OR = 1.82, 95% CI 1.20–2.77) [69]. Furthermore, a Korean-based study found the TyG index to be better at predicting CAC (OR = 1.95, 95% CI 1.23–3.11) than HOMA-IR (OR = 1.64, 95% CI 1.12–2.40) [70]. Additionally, the development of CAC provides evidence for the development of CAD.

In estimates performed with the Archimedes Model, insulin resistance is an important cause of coronary artery disease [71], which may be assessed with the TyG index. A genetic study including 63,746 CAD patients and 130,681 healthy individuals reported that lipid metabolism and inflammation are key biological processes involved in the pathogenesis of coronary atherosclerosis, supporting the role of insulin resistance in the pathophysiology of CAD [72]. Insulin resistance in liver and adipose tissues drives the development of atherosclerotic dyslipidemia, generates a low-grade inflammatory state, and increases release of inflammatory markers [73]. It also affects blood pressure, endothelial cells, and macrophages [73].

Our meta-analysis has several limitations. First, this was a meta-analysis based on cohort studies. Despite the inclusion of studies with multivariate analysis, residual confounding factors may have influenced our results; thus, causation cannot be proven. Second, a total of 12 studies were included in our analysis, which is a relatively small sample size. Outcomes were inconsistent across the 12 studies we included, resulting in a small number of studies included in each outcome (MI = 2, CV death = 3). This may be another cause of bias and inaccurate results. Continued attention to relevant research and refinement of each analysis are needed in the future. In addition, of the 12 studies included, most were conducted in Asia, with only two conducted in Western countries (Europe and America). However, the limitation of the number of articles precluded us from examining regional variation, and future research should elucidate whether there are regional differences to demonstrate the association of the TyG index with CVD and mortality in the general population. Third, substantial heterogeneity was found, which might have derived from differences in TyG index measurement methods, study design, follow-up, diabetes status, and other clinical characteristics across studies. Fourth, the limited number of included studies prevented us from conducting subgroup analysis; hence, the potential influence of confounding and potential intermediate factors needs further investigation. Finally, limited studies excluded individuals with preexisting diabetes, and subgroup analysis stratified by diabetes was not available. Diabetes has been considered a main risk factor for CVD and mortality; therefore, application of the TyG index for CVD incidence can be easily biased by diabetes, and additional studies in nondiabetes populations are needed to reduce reverse causality before applying the TyG index to examine CVD incidence and mortality.

Although the TyG index is not directly applied in clinical guidelines, the role of glycemic and triglyceride control and diabetes management in the prevention of cardiovascular disease is reflected in some guidelines. Recently, the American Diabetes Association (ADA) published its 2022 diabetes criteria, which detail that patients with elevated triglyceride levels (≥ 150 mg/dL [1.7 mmol/L]) should implement enhanced lifestyle interventions and optimal glycemic control [74]. Control of blood glucose and triglycerides is effective in reducing the TyG index, an indicator of insulin resistance, which is one of the main triggers of T2DM. There is a strong association between these factors and clinical guidelines.

Our results show that an elevated TyG index is strongly associated with a high risk of CAD, MI and CVD, suggesting that the incidence of CVD can be significantly reduced by controlling TyG index-related factors or triggers that contribute to elevated TyG, such as blood glucose. As a simple-measurement indicator, the TyG index may be used more often in clinical practice. With respect to public health implications, early detection of CVD onset can be achieved by assessing the TyG index of the population, and prompt intervention after disease risk is identified may be effective in reducing the incidence of CVD in the general population. The TyG index might be added to routine health check-up programs to screen for the development of diabetes and cardiovascular disease. However, more research is needed before application of the TyG index to evaluate CVD incidence in the clinic.

Based on our study, the TyG index may be considered an independent predictor for CVD incidence in the general population. Further studies are required to investigate dose–response relationships and determine cutoff values. On the other hand, addition of the TyG index to the Framingham Risk Score (FRS) did not lead to improvement in its predictive power; thus, adding the TyG index to the FRS does not improve CVD risk prediction [18]. Further studies and more articles are needed to confirm the relevance of the TyG index to CVD, determine the predictive effect of the TyG index, and ascertain whether the TyG index can improve the predictive power of existing cardiovascular risk scores.