Can the triglyceride-glucose index predict the risk of stroke? A meta-analysis of high-quality studies with 12.8 million participants

Introduction

Stroke is now the second major cause of mortality worldwide leading to about 1/3rd of all disabilities (Lozano et al., 2012). Statistics indicate that about 13 million new incident cases of stroke were diagnosed in 2016 with about 87% being ischemic stroke (Saini, Guada & Yavagal, 2021). Chinese data from 2020 shows that approximately 3.4 million patients were diagnosed with first-ever stroke causing about 2.3 million deaths (Tu et al., 2023). About 76% of strokes develop in individuals without a prior history of the disease (Saver et al., 2015). The illness is a life-altering event for those who experience the disease as well as for families and caregivers. About 26% of elderly experiencing stroke become dependent on activities of daily living and about 46% have cognitive impairment (Go et al., 2014). These findings support the fact that active and effective prevention can help in reducing the disease burden and there is an urgent need for scaling up the primary prevention programs (Lozano et al., 2012). In this context, the development of accurate risk prediction markers and models can help identify high-risk individuals who can be monitored and targeted by effective interventions to reduce the risk of stroke (Xu et al., 2021). Despite the establishment of several stroke prediction models and biomarkers in the past few decades, researchers have been unable to identify a single model or marker that is highly effective in predicting the risk of stroke (Xu et al., 2021; Lu et al., 2021; Lip et al., 2022; Ihle-Hansen et al., 2023).

Research indicates that insulin resistance could have a major role in the pathogenesis of stroke (Ding et al., 2022). Insulin acts as a protective agent for the brain by preventing ischemia, oxidative stress, and apoptosis-induced brain tissue damage. It also modulates cholesterol metabolism in neural tissues and astrocytes and is known to improve cognitive dysfunction in Alzheimer’s disease (Agrawal et al., 2021; Ding et al., 2022). Increased insulin resistance has been linked with a higher risk of stroke in the general population (Zheng et al., 2024). However, the current gold standard for assessing insulin resistance, i.e., the hyperinsulinemic-euglycemic clamp is too complex and expensive to be applied routinely in clinical practice (Cersosimo et al., 2014). A more accessible marker can be the homeostasis model assessment of insulin resistance (HOMA-IR) index (Matthews et al., 1985). Nevertheless, its routine clinical application is also not economical and convenient. The triglyceride-glucose (TyG) index, is considered to be a biomarker for insulin resistance which is calculated using the formula: fasting triglycerides (mg/dl)×fasting blood glucose (mg/dl)/2 (Liao et al., 2022). Studies have shown that high TyG levels are significantly associated with increased risk of coronary artery disease (CAD), contrast-induced nephropathy, hypertension (HT), diabetes mellitus (DM), atrial fibrillation, metabolic dysfunction associated fatty liver disease, metabolic syndrome, and stroke. Moreover, a positive association has also been demonstrated between high TyG and the prognosis of CAD and stroke (Yin et al., 2024; Nayak et al., 2024).

On the question of its predictive ability for stroke, three prior meta-analysis studies (Liao et al., 2022; Feng et al., 2022; Yang et al., 2023) with eight to eleven studies each have shown that high TyG may be a potential marker for stroke. However, these reviews have several limitations including inclusion of studies on specific populations (like HT, DM, CAD), use of cross-sectional data, and studies with overlapping data. Moreover, the low number of studies with a small number of participants is an additional hindrance that limits the acceptability of the evidence. To overcome the limitations of prior reviews and to present the best possible evidence in the literature, we conducted an updated meta-analysis including only high-quality cohort studies to assess the ability of TyG to predict stroke in the general population.

Materials and Methods

Results

Search results

We have presented the search results in Fig. 1. Of the 3,400 studies found from all databases, we removed 2,488 duplicates. A total of 1,122 studies underwent meticulous screening by the reviewers. Jointly, they selected 40 studies for further analysis. After full-text reading, 13 were selected for the review (Hong, Han & Park, 2020; Zhao et al., 2021; Cho et al., 2022; Liu et al., 2022b; Liu et al., 2022a; Che et al., 2023; Wang et al., 2023; Muhammad et al., 2023; Wan et al., 2023; Yao et al., 2024; Li et al., 2024a; Li et al., 2024b; Rafiee et al., 2024). Agreement between reviewers was high (kappa = 0.95). The search of additional sources did not reveal any missed study. A list of excluded studies can be found in Table S2.

TyG as a categorical variable

Meta-analysis of all 13 studies indicated a statistically significant increased risk of stroke between higher vs lower values of TyG (RR: 1.27 95% CI [1.19–1.35]) (Fig. 2). I² was found to be 66% indicating high heterogeneity. No change in the significance of RR was found by the reviewers during sensitivity analysis. Publication bias was noted on the funnel plot (Fig. 3). Egger’s test was not significant (p = 0.69). Subgroup analysis showed that the association between TyG and stroke persisted after dividing studies based on location, excluded population, stroke diagnosis, TyG data, and follow-up (Table 2). Heterogeneity was found to be reduced to zero in some of the subgroup analyses like studies on western cohorts, excluding only baseline stroke, using TyG as tertiles, and with follow-up ≥10 years. Meta-regression analysis found that age, male gender, DM, HT, TyG cut-off, stroke incidence, and follow-up did not have a significant effect on the meta-analysis results (Table 3).

Table 2:

Subgroup analysis details.

Covariates	Groups	Number of studies	Risk ratio [95% confidence intervals]	I2
TyG categorical variable
Location	Chinese	7	1.40 [1.24, 1.57]	32
	All Asian	11	1.27 [1.18, 1.37]	71
	Western	2	1.29 [1.18, 1.41]	0
Excluded population	All CVD	10	1.25 [1.16, 1.35]	72
	All Stroke	3	1.33 [1.22, 1.45]	0
Stroke diagnosis	ICD codes	6	1.22 [1.12, 1.33]	81
	Physician diagnosed	2	1.70 [1.28, 2.27]	8
TyG data	Quartile	11	1.25 [1.17, 1.34]	70
	Tertile	2	1.48 [1.16, 1.88]	0
Follow-up	≥10 years	6	1.27 [1.20, 1.35]	0
	<10 years	7	1.27 [1.15, 1.41]	81
TyG continuous variable
Location	Chinese	6	1.17 [1.10, 1.24]	49
	All Asian	6	1.17 [1.10, 1.24]	49
	Western	2	1.13 [0.84, 1.51]	95
Excluded population	All CVD	5	1.16 [1.04, 1.30]	92
	All Stroke	3	1.18 [1.06, 1.30]	60
Stroke diagnosis	ICD codes	3	1.13 [0.92, 1.38]	90
	Medical records	2	1.32 [1.18, 1.47]	0
Follow-up	≥10 years	4	1.22 [1.09, 1.36]	63
	<10 years	4	1.12 [0.98, 1.29]	90

DOI: 10.7717/peerj.19994/table-2

Notes:

CVD

cardiovascular disease

ICD

international classification of diseases

TyG

Triglyceride glucose index

Table 3:

Details of meta-regression analysis.

Variable	Beta	SE	+95% CI	−95% CI	p-value
TyG categorical variable
Sample size	0.000000005	0.000000005	−0.000000006	0.00000002	0.34
Age	−0.0008	0.002	−0.005	0.003	0.64
Male (%)	−0.00002	0.001	−0.003	0.002	0.99
DM (%)	0.002	0.005	−0.008	0.014	0.59
HT (%)	0.002	0.001	−0.001	0.004	0.12
TyG cut-off	−0.004	0.003	−0.011	0.002	0.11
Stroke incidence (%)	0.007	0.004	−0.001	0.015	0.66
Follow-up (years)	0.001	0.004	−0.008	0.010	0.79
TyG continuous variable
Sample size	−0.0000006	0.0000002	−0.0000011	−0.0000001	0.004
Age	0.0002217	0.0182591	−0.0429542	0.0433976	0.99
Male (%)	−0.0024111	0.0048529	−0.0138865	0.0090642	0.61
DM (%)	0.0193193	0.0088024	−0.0014951	0.0401336	0.20
HT (%)	0.0051778	0.0024720	−0.0006676	0.0110231	0.40
Stroke incidence (%)	0.0205690	0.0132036	−0.0106526	0.0517907	0.12
Follow-up (years)	0.0071913	0.0069041	−0.0091343	0.0235170	0.29

DOI: 10.7717/peerj.19994/table-3

Notes:

SE

standard error

CI

confidence intervals

DM

diabetes mellitus

HT

hypertension

TyG

Triglyceride glucose index

TyG as a continuous variable

Meta-analysis of eight studies showed that a per unit increase in TyG was associated with a statistically significant increase in the risk of stroke (RR: 1.16 95% CI [1.07–1.27]) (Fig. 4). I² was 89%, indicating high heterogeneity again. No major asymmetry of the funnel plot was noted (Fig. 5). Egger’s test was not significant (p = 0.76). The RR remained statistically significant on sensitivity analysis. Subgroup analyses revealed non-significant results for studies on the Western population and those using ICD codes for the identification of stroke. For the other subgroups, the results remained statistically significant (Table 2). On meta-regression analysis (Table 3), only sample size was found to inversely influence the effect size. A larger sample size was associated with a weaker association between TyG and stroke. None of the other moderators were found to be statistically significant.

Discussion

A wide range of stroke prediction models and biomarkers have been proposed in the literature to predict the risk of stroke in the general population (Xu et al., 2021; Lu et al., 2021; Lip et al., 2022; Ihle-Hansen et al., 2023). There are several routine laboratory markers like albumin, brain natriuretic peptide, serum creatinine, red cell distribution width, total cholesterol, high-density lipoprotein, low-density lipoprotein, and non-high-density lipoprotein cholesterol which have been associated with a higher risk of stroke (Sughrue et al., 2016). Other non-routine markers like fibrinogen, E-selectin, interferon-γ-inducible-protein-10, resistin, and total adiponectin have also been linked with increased risk of stroke (Prugger et al., 2013). On the other hand, there are also complex prediction models and machine-learning-based algorithms that claim to accurately predict the risk of stroke (Lu et al., 2021; Lip et al., 2022). Lastly, specific investigation-based markers like carotid plaque scores are also linked with the risk of stroke (Ihle-Hansen et al., 2023). Despite a plethora of research, no single marker has been identified as the gold standard as there are issues of accuracy with simpler laboratory markers and difficulty of application with more complex models.

During the search for an ideal marker, the TyG has generated considerable interest in the past few years. Its association with a large number of diseases as well as disease outcomes has prompted significant research on its ability to predict stroke in the general population (Yin et al., 2024; Nayak et al., 2024). However, despite three prior reviews (Liao et al., 2022; Feng et al., 2022; Yang et al., 2023), there remains uncertainty about its clinical application. One major limitation is the quality of these meta-analysis studies. Yang et al. (2023) in their review have not only included overlapping studies using the same dataset (Kailuan community cohort) but also have combined data of cross-sectional and cohort studies in the same meta-analysis. Repeated inclusion of the same data can overestimate the effect size generating skewed results. Also, cross-sectional studies cannot establish causality like cohort studies. Secondly, all three reviews (Liao et al., 2022; Feng et al., 2022; Yang et al., 2023) have also pooled studies assessing the risk of stroke in the general population with studies on specific disease cohorts like DM, HT, CAD, etc. The risk of stroke in the general population is considerably different compared to those with known risk factors like DM, HT, and CAD (Saini, Guada & Yavagal, 2021). Lastly, these reviews have not uniformly excluded studies including participants with prior stroke, which is another risk factor for recurrent stroke.

Overcoming these issues, the current systematic review and meta-analysis aimed to generate the best possible evidence on the clinical utility of TyG in predicting the risk of stroke in the general population. After the exclusion of a large number of studies with overlapping data, studies on disease populations, studies with short follow-ups, and studies including prior stroke participants, we were still left with 13 cohorts with about 12.8 million participants. Separate data analysis was conducted for TyG as a categorical as well as continuous variable. Our results revealed that high levels of TyG resulted in a 27% increase in the risk of stroke in the general population whereas a per unit increase in TyG was associated with a 16% increase of the same. We found the results to be robust on sensitivity analysis highlighting the credibility of the outcomes. Lack of publication bias also supplements the results. Our results are in agreement with the large study of Lopez-Jaramillo et al. (2023) which could not be included in the review as it did not exclude prior stroke patients. However, the study is worth mentioning as it was a prospective analysis of 141,243 individuals aged 35–70 years from 22 countries. After a median follow-up of 13.2 years, the authors reported an increased risk of stroke (hazard ratio: 1.16 95% CI [1.05–1.28]) with the highest TyG tertile. The association was the strongest in low-income countries followed by middle-income countries but non-significant in high-income countries.

Despite the robust results of our study, the small increase in the risk of stroke associated with high TyG may raise questions about its routine use in clinical practice. In the realm of stroke prevention, especially within the general population, establishing a precise minimal clinically important difference (MCID) is challenging due to varying baseline risks and individual patient factors. MCID represents the smallest change in the risk of disease or treatment outcome that a patient or clinician would identify as meaningful (Salas Apaza et al., 2021). Nevertheless, the perception of what constitutes a meaningful risk can be gauged from current CVD and stroke guidelines. For example, the American College of Cardiology and American Heart Association recommend considering statin therapy for individuals with a 10-year atherosclerotic cardiovascular disease risk of 7.5% or higher (Virani, 2022). The risk of stroke noted with high TyG in our review was much higher and therefore should be clinically relevant and prompt monitoring and risk reduction measures in high-risk individuals.

An important caveat in understanding the utility of a marker for stroke prediction is the role of confounding factors. The risk of stroke depends on numerous variables like age, gender, ethnicity, family history, physical inactivity, alcohol consumption, smoking, obesity, DM, HT, CVD, etc (Tu et al., 2023). Much variation was observed in the included populations in terms of such baseline characteristics which could have led to the high heterogeneity in the meta-analysis. However, we were able to somewhat circumvent this limitation by including only adjusted data and conducting several subgroup and meta-regression analyses. Most of the included studies adjusted their data for age, gender, smoking, physical inactivity, body mass index, DM, HT, and alcohol consumption thereby eliminating the impact of major confounders. On subgroup analysis, there was no change in the significance of the results especially for TyG as a categorical variable. Segregation of data based on the region of the study, exclusion of CVD or stroke, diagnosis of stroke, TyG data as tertiles or quartiles, and follow-up had no impact albeit without complete elimination of inter-study heterogeneity. This indicates that there may be other unmeasured factors at play requiring further research. Likewise, meta-regression using relevant moderators like age, male gender, DM, HT, TyG cut-off, stroke incidence, and follow-up also did not have a significant association with the pooled analysis. The only significant association noted was between sample size and TyG as a continuous variable which could have been a statistical artifact given the small number of studies in the analysis.

The underlying mechanism supporting the link between high TyG and stroke remains unclear. Substantial evidence reinforces the relationship between the TyG index and insulin resistance (Guerrero-Romero et al., 2010; Sánchez-García et al., 2020). In fact, the TyG index has been identified as one of the best markers of insulin resistance performing better than visceral adiposity indicators and other lipid parameters (Du et al., 2014). Insulin resistance has been linked with stroke via several mechanisms like interference with insulin signaling and sensitivity, amplification of chronic systemic inflammation, and accelerating foam cell generation causing atherosclerosis and advanced plaques (Bornfeldt & Tabas, 2011; Kosmas et al., 2023). Insulin resistance interferes with the function of insulin-like growth factors, cyclic guanosine monophosphate, and nitric oxide thereby causing adhesion, activation, and aggregation of platelet function which in turn causes vascular occlusion and stroke (Randriamboavonjy & Fleming, 2009; Guo et al., 2021). Moreover, insulin resistance may impact the cerebrovascular reserve via several chemical, neuronal, and metabolic mechanisms leading to reduced cerebral perfusion during stroke (Banks & Rhea, 2021; Fan et al., 2022). There is also evidence to show that admission hyperglycemia and DM can negatively impact stroke outcomes itself (Belge Bilgin et al., 2025).

Despite the robust results, there are certain major limitations of the review. Firstly, we were unable to identify the ideal cut-off of TyG to predict the risk of stroke due to a lack of sensitivity and specificity data and varied cut-offs of the included studies. Guerrero-Romero et al. (2010) have shown that the best TyG cut-off for the diagnosis of insulin resistance was 4.68. However, there remains limited data on the ideal cut-off for predicting stroke. Secondly, the studies considered only baseline glucose and triglyceride measurements and did not account for changes over time. It remains unclear how TyG changes affected the risk of stroke. Thirdly, despite the studies adjusting several potential cardiovascular and metabolic risk factors, we cannot eliminate the possibility of residual confounding due to unmeasured factors. Fourthly, data was derived from observational studies which have inherent bias. Fifthly, the large heterogeneity in the meta-analysis cannot be ignored and hence the results must be interested with caution. Lastly, the predominance of Chinese studies limits the generalization of data to other regions. More robust studies from Western countries are needed to add to current evidence.

Conclusions

High TyG is associated with increased risk of stroke in the general population. Since the index is easy to measure and calculated from routinely available laboratory values, it may be incorporated into daily clinical practice to screen individuals at high risk of stroke.

Supplemental Information