Association of triglyceride-glucose index with clinical outcomes in patients with acute ischemic stroke receiving intravenous thrombolysis

Intravenous tissue plasminogen activator (tPA) remains the cornerstone of recanalization therapy for acute ischemic stroke (AIS), albeit with varying degrees of response. The triglyceride-glucose (TyG) index is a novel marker of insulin resistance, but association with outcomes among AIS patients who have received tPA has not been well elucidated. We studied 698 patients with AIS who received tPA from 2006 to 2018 in a comprehensive stroke centre. TyG index was calculated using the formula: ln[fasting triglycerides (mg/dL) × fasting glucose (mg/dL)/2]. TyG index was significantly lower in patients that survived at 90-days than those who died (8.61 [Interquartile Range: 8.27–8.99] vs 8.76 [interquartile range: 8.39–9.40], p = 0.007). In multivariate analysis, TyG index was significantly associated with 90-day mortality (OR: 2.12, 95% CI: 1.39–3.23, p = 0.001), poor functional outcome (OR: 1.41 95% CI: 1.05–1.90, p = 0.022), and negatively associated with early neurological improvement (ENI) (OR: 0.68, 95% CI: 0.52–0.89, p = 0.004). There was no association between TyG index and symptomatic intracranial hemorrhage. ‘High TyG’ (defined by TyG index ≥ 9.15) was associated with mortality, poor functional outcomes and no ENI. In conclusion, the TyG index, a measure of insulin resistance, was significantly associated with poorer clinical outcomes in AIS patients who received tPA.

www.nature.com/scientificreports/ While Insulin Resistance (IR), part of the underlying pathophysiology of disorders related to metabolic syndrome 10 , has shown correlation with poor post-stroke outcomes 11 , current measures of IR, such as the hyperglycemic clamp, the hyperinsulinemic-euglycemic clamp test (HEC) and homeostatic model assessment of insulin resistance (HOMA-IR), are not practical in a clinical setting. Alternative measures of IR that reflect various components of metabolic syndrome have been proposed, but among these, the Triglyceride-Glucose (TyG) Index was shown to be superior to other similar markers for identifying IR risk and risk of developing diabetes 12,13 . The TyG Index has recently been established as a reliable, cost-effective and easily accessible surrogate marker for IR 14,15 , and performed better than HOMA-IR in estimating IR when compared to the hyperglycemic clamp 16 . Furthermore, the TyG index performed superiorly to fasting glucose and triglycerides alone in predicting development of Type 2 diabetes 17 , and notably increased the risk of cardio-cerebrovascular disease 18 , suggesting the TyG index may potentially be a good marker of metabolic disease to predict post-stroke outcomes.
While the TyG index has demonstrated predictive effects in major adverse cardiovascular event outcomes in patients with acute coronary disease undergoing percutaneous coronary intervention 19 , and was superior to fasting glucose and hemoglobin A1c (HbA1C) in this setting 20 , it has not been well investigated in stroke populations. While previous authors have found an association between incidence of AIS and TyG index 21,22 , limited studies have reported the effect of the TyG index on post-stroke clinical outcomes or recurrence [23][24][25] , and the relationship of the TyG index in a thrombolyzed patient cohort has been studied scarcely. The only two studies that have investigated the effect of IR on AIS patients' response to thrombolysis both utilized HOMA-IR and were limited by sample size or analyzed only non-diabetic patients 26,27 . Hence, further studies in a current, larger cohort of patients are required to establish the significance of a relationship between the TyG index and post-thrombolysis stroke outcomes. Therefore, we sought to understand the association of the TyG index, as a reflection of IR, in AIS patients who underwent thrombolysis.

Methods
Study design. This is a retrospective cohort study of consecutive patients who received IV tPA from September 2006 to June 2018 and was approved by the local institutional review board. Patients were assessed by a neurologist to be eligible to receive intravenous thrombolysis according to institutional protocol and American Heart Association/American Stroke Association guidelines at a standard dose of 0.9 mg/kg body weight 6 . All stroke patients who were thrombolyzed underwent standard non-contrast head Computed Tomography and Computed Tomography brain and neck angiography. Patients with valid laboratory readings of their fasting triglycerides and venous glucose levels, taken within 24 h of their AIS admission were included. Patients who were deemed unsuitable for IV tPA were excluded from the study. Other baseline demographics, clinical parameters and ischemic stroke characteristics were collected and tabulated within 24 h of AIS admission. Diabetes Mellitus (DM) was defined as pre-existing diagnosis of diabetes mellitus or an admitting fasting blood glucose level greater than or equal to 7.0 mmol/L (126 mg/dl) or an HbA1c greater than or equal to 6.5%. The presence of Large Vessel Occlusion (LVO) was defined as occlusions of the first and second segment of the middle cerebral artery (MCA), the Internal Carotid Artery (ICA) and as well as its terminus, tandem occlusions involving ICA-MCA, or occlusion of the basilar artery. The severity of stroke at presentation was assessed using the National Institute of Health Stroke Scale (NIHSS) 28 , recorded as admitting NIHSS. This assessment was made by credentialed nurses as part of the acute stroke response team. The Trial of Org 10172 in Acute Stroke Treatment (TOAST) criteria was used by the treating stroke neurologist to classify stroke subtypes 29 . A cut-off for TyG index of 9.15 was calculated using the internationally recognised triglyceride cut-off for hypertriglyceridemia of 150 mg/dL (1.7 mmol/L) 30 and the cut-off of fasting glucose of 126 mg/dL (7.0 mmol/L), which was selected based on the fasting glucose cut-off for the diagnosis of diabetes 31 . All procedures were performed in accordance with the relevant guidelines.
The TyG Index was calculated using the formula: ln[fasting triglycerides (mg/dL) × fasting glucose (mg/ dL)/2] 32 . The primary outcome studied was all-cause 90-day mortality. The secondary outcomes included poor functional outcome, for which the dichotomous definition of poor functional outcome was defined as having a 90-day modified Rankin Scale (mRS) of 3-6. The 90-day mRS was evaluated during the follow-up visit to the stroke clinic, and if patients were not able to attend the follow-up visit physically, mRS was evaluated via telephone call instead. Other secondary outcomes included early neurological improvement (ENI), defined as a decrease in NIHSS score by 4 points or more within 24 h or complete resolution of neurological deficits at 24 h, and SICH, based on the European Cooperative Acute Stroke Study (ECASS) II definition: any type of intracerebral hemorrhage on any post-treatment imaging after the start of thrombolysis and increase of ≥ 4 NIHSS points from baseline, or from the lowest value within 7 days, or leading to death 33 . Data analysis. Statistical tests were performed using R version 4.0.2 (R Core Team, Vienna, Austria), and a p-value of < 0.05 was considered statistically significant. We performed statistical analysis using Mann-Whitney U test for non-normally distributed continuous variables, and Chi-square analysis for categorical variables. All continuous variables were non-normally distributed as determined by visual histogram and Shapiro-Wilk statistical tests, and median and interquartile range (IQR) were reported for each variable. A logistic regression model was then performed to identify independent predictors of mortality, 90-day functional outcome, ENI and SICH. Covariates that were selected a priori for variable adjustment were age, hypertension, NIHSS recorded at presentation (admitting NIHSS), and LVO status. The adjusted multivariate logistic regression model was then employed to evaluate the associations of TyG index with the respective outcomes and presented as adjusted odds ratio (OR) with 95% confidence intervals (CI).
Characteristics of patients with 'Low TyG' and 'High TyG' indices were then compared using ≥ 9.15 as a cut-off value, including the distributions of 90-day functional outcome (measured by mRS) between the two groups of patients. Receiver-operating characteristic (ROC) curve analysis was performed in R to determine the performance of the TyG index and the cut-off value in predicting the primary outcome of mortality. Logistic regression was again performed to determine the relationships between 'High TyG' and 'Low TyG' and dichotomous outcomes including mortality, ENI and SICH. For functional outcome, ordinal shift analysis was carried out to compare associations between patient cohorts with 'Low TyG' and 'High TyG' indices and ordinal functional outcome (mRS 0-6), where multivariate ordinal logistic regression was presented as OR with 95% CI. After analyzing the main cohort, subgroup analysis was conducted to investigate the relationship of TyG index in patients with and without diabetes respectively. Similarly, baseline characteristics and multivariate logistic regression were carried out for both cohorts of patients with and without diabetes.
Multivariate Imputation by Chained Equations (MICE) in R was employed for missing baseline data to conduct multivariate regression analyses. Variables with originally more than 25% missing data were excluded from multivariate analysis but were used as predictors in imputation. Missing baseline data was imputed with MICE. The multiple imputation method used was Predictive Mean Matching (PMM) for scale variables and logistic regression for categorical variables. Predictors used in the imputation included age, race, gender, smoking status, hypertension, hyperlipidemia, DM, atrial fibrillation (AF), fasting glucose, random plasma glucose, HbA1c, serum total cholesterol, low-density lipoprotein cholesterol (LDL), high-density lipoprotein cholesterol (HDL), triglycerides, the cholesterol:HDL (cholHDL) ratio, serum aspartate transaminase (AST), alanine transaminase (ALT), admitting hematocrit, total white blood cells, lymphocytes, neutrophils, platelets, admitting NIHSS, NIHSS at 2 h and 24 h, systolic blood pressure on admission, 2 h and 24 h, diastolic blood pressure on admission, 2 h and 24 h, baseline mRS, mRS at discharge, 90-day mRS, TOAST classification, SICH, onset-to-treatment time for tPA administration, and LVO status. 30 imputed datasets were created and estimates from the multivariate analysis were pooled.
Ethics approval and consent to participate. Ethics approval was obtained from National Healthcare Group-Domain Specific Review Board (NHG DSRB Reference: 2010/00509) where approval for informed consent was waived as this was patient level data.

Comparing 'High TyG' and 'Low TyG' patients using a TyG index cut-off point. Patients with a
TyG index of < 9.15 were considered to have 'Low TyG' (n = 563) and patients with TyG index of ≥ 9.15 were considered to have 'High TyG' (n = 135), utilizing literature cut-offs for triglyceride (150 mg/dL or 1.7 mmol/L) and fasting glucose (7.0 mmol/L or 126 mg/dL) 30,31,34 . ROC curve analysis was employed to evaluate the association between the TyG index and the primary outcome of mortality, demonstrating an Area Under the Curve (AUC) of 0.604 (95% CI: 0.530-0.678, p = 0.007) (Fig. 1). The value of 9.15 derived as the cut-off for TyG index served as a possible optimal cut-off point on the ROC curve, predicting mortality with 33.3% sensitivity and 81.7% specificity. There were a higher proportion of deaths among 'High TyG' (n = 21, 15.6%) patients than 'Low TyG' patients (n = 42, 7.6%, p = 0.006). ( (Fig. 2).  (Table 5).

Discussion
In this study, we observed an association between increased TyG index and mortality, poor functional outcome and poor ENI among post-thrombolysis AIS patients. Furthermore, on subgroup analyses, the TyG index was associated with an increased odds of mortality in patients without diabetes.
The TyG index was first introduced as a marker of insulin resistance in 2008 by Simental-Mendia to find a reliable alternative to HOMA-IR, with high sensitivity in detecting IR amidst non-diabetic subjects 32 . Compared  www.nature.com/scientificreports/ to the HOMA-IR which utilizes glucose and insulin levels, the TyG index uses glucose and triglycerides values. Insulin is not routinely measured in clinical practice and the insulin assays are variable across platforms 35,36 . In comparison, serum triglyceride and plasma glucose levels are routinely performed in clinical practice and are well-calibrated across platforms 14 . The gold standard for evaluating IR is the hyperinsulinemic-euglycemic clamp (HEC), however, this labour-intensive and time-consuming procedure is should be performed in specialized metabolic research units requiring trained staff 37,38 . In 2010, it was then found that TyG index was comparable to and as reliable as the HEC test in assessing IR in a generic patient population including both diabetics and nondiabetics 14 . The TyG index has been shown to be a good predictor of metabolic health and early identification of IR, by identifying individuals with metabolic derangements 39 , and indexes involving the TyG index performed better than anthropometric measurements and traditional lipid markers to identify metabolic disease 12,40 . The role of the TyG index in various cardiovascular and cerebrovascular diseases has also been increasingly studied. It is associated with arterial stiffness 41 , increased risk of cardiovascular disease and worse post-percutaneous coronary intervention outcomes 19 . In stroke, the TyG index has been associated with increased risk of stroke 22 , in-hospital mortality in critically-ill stroke 24 , and early recurrent ischemic lesions 25 . Baseline TyG index has been associated with increased mortality and neurologic deterioration in ischemic stroke 23 . However, studies have not been conducted to investigate the role of TyG index in patients post-thrombolysis, a key component of stroke management.
Calculation of the TyG index involves fasting glucose and triglycerides, both of which have been shown to have some association with stroke independently. Hyperglycemia in stroke is well known to be associated with  Table 4. Association of 'High TyG' (≥ 9.15) with outcomes including ordinal functional outcome mRS 0-6. Adjusted odds ratio, 95% confidence interval and p-value calculated from logistic regression after adjusting for hypertension, age, admitting NIHSS and LVO status. 'High TyG' triglyceride-glucose index ≥ 9.15, OR odds ratio, 95% CI 95% confidence interval, p p-value, poor functional outcome dichotomous outcome of modified Rankin scale of 3 to 6, ENI early neurological improvement, SICH symptomatic intracranial hemorrhage, ordinal function outcome ordinal outcomes using modified Rankin scale 0 to 6. www.nature.com/scientificreports/ poor survival and outcomes, increased infarct volume, increased risk of hemorrhagic conversion and reduced recanalization after IV-tPA 42,43 . While studies have not been conducted on the associations of post-thrombolysis outcomes with fasting triglycerides, the Reduction of Cardiovascular Events with Icosapent Ethyl-Intervention Trial (REDUCE-IT) study suggested that triglyceride lowering via the use of icosapent ethyl, a purified eicosapentaenoic acid (EPA) ethyl ester, significantly reduces the risk of composite cardiovascular outcomes and stroke 44 . However, there is evidence to suggest the TyG index is more strongly associated with cardiovascular and cerebrovascular outcomes than fasting glucose and triglycerides alone 18,45 . A large-scale prospective cohort study comprising 273,368 individuals found a significant association of the TyG index in cerebrovascular and ischemic stroke events, compared to the relationship of joint exposure of glucose and triglycerides scores with cerebrovascular disease which did not reach significance, recommending the TyG index to be a sensitive pre-diagnostic indicator for disease 18 . Furthermore, the TyG index is shown to be superior to fasting glucose or  www.nature.com/scientificreports/ triglycerides in diagnosing metabolic syndrome and for diabetes prediction 17,46 . These argue for the TyG index's utility as an effective way to assess outcomes in cerebrovascular disease as an appropriate measure of IR. We observed worse mortality, neurological and functional outcomes in patients with high TyG index. A possible mechanism is that IR causes an increase in chronic proinflammatory cytokines and prothrombotic responses and endothelial dysfunction, exacerbating brain damage post stroke [47][48][49] . It has been suggested that the highly related concept of metabolic syndrome could increase resistance to thrombolysis due to an impaired fibrinolytic system or increased clot density 50 , and those with more metabolic syndrome components could increase the density of the clot structure and make it more resistant to lysis 51 . This suggests that identifying and targeting IR for patients to prevent future poor response to thrombolysis may be beneficial, but further studies will be required.
As there is a high prevalence of stroke patients who have concomitant diabetes in our local population (43.2%) 52 , and patients with diabetes are known to have poorer post-thrombolysis stroke outcomes compared to those without diabetes 53 , we proceeded to perform a subgroup analysis comparing the outcomes of patients with and without diabetes. Our study observed no association between TyG index and stroke outcomes in patients with diabetes, although the association with mortality and poorer functional outcomes was approaching statistical significance. There are several possible mechanisms for poor response to thrombolysis in diabetic patientsincluding vascular changes such as cerebral vascular endothelial dysfunction, arterial stiffness and thickening of capillary basal membrane 54 , hyperglycemia-induced overproduction of reactive oxygen species 55 , and increased and deregulated neuroinflammation, which could lead to worsened cerebral ischemia and incomplete recanalization after thrombolysis in a patient with diabetes 56,57 . These affect neurological outcomes through processes that may not be related to IR. Furthermore, in our study, subjects with diabetes were older, and had more cardiovascular risk factors including hypertension and hyperlipidemia. (Supplementary Table 2 [see Additional File 1]) Hence, after adjustment for these comorbidities in multivariate analysis, and accounting for other mechanisms that influence neurological outcomes in diabetes, the TyG index could exert less predictive effect on mortality and functional outcomes in a patient with established diabetes. Nonetheless, the associations of TyG index with mortality and functional outcomes were approaching statistical significance, which also may be related to the relatively small sample size of patients with diabetes in this study (n = 214).
Meanwhile, we found that higher TyG index was significantly associated with mortality but not functional outcomes, ENI or SICH after thrombolysis in patients without diabetes. IR has been associated with poor outcomes after ischemic stroke in patients without diabetes 58,59 , including in those who received thrombolysis 27 . The TyG index was also shown to be a good predictor of neurological worsening and stroke recurrence in AIS patients without diabetes 23 . Although our study did not replicate previous results in demonstrating an association of the TyG index with functional outcomes or ENI, this could be due to differing study populations and sample size. Nonetheless, our results imply that identifying and treating elevated IR even among AIS patients without diabetes could be important to improving post-thrombolysis stroke outcomes including mortality.
Emerging studies have suggested targeting IR as a therapeutic strategy to improve post-stroke outcomes using anti-diabetic drugs in both patients with and without diabetes. The Insulin Resistance Intervention after Stroke (IRIS) trial which involved 3876 patients, demonstrated that pioglitazone, an insulin-sensitizer, reduced the risk of recurrent stroke and myocardial infarction in patients without diabetes who had recent ischemic stroke or TIA 11 . Others observed associations between prior metformin use and better neurological outcomes in patients with AIS and those who received thrombolytic therapy 60 . A meta-analysis on glucagon-like peptide-1 (GLP-1) receptor agonists like liraglutide, which work by glucose-dependent insulin release, have shown benefits for primary stroke prevention 61 . Experimental studies also showed how GLP-1 agonists may exert neuroprotective effects including through mechanisms that decrease insulin resistance in the brain 62 . These results suggest that trials targeting IR using anti-hyperglycemic insulin-sensitising agents could be explored to further target IR in a post-thrombolysis population, possibly even in patients without diabetes with high levels of IR.
With respect to limitations, this was a retrospective cohort study where causality could not be established. We were not able to compare the TyG index with other markers of insulin resistance, such as the HOMA-IR or hyperinsulinemic-euglycemic clamp, as only data which had been collected on admission as part of routine clinical practice was used. Furthermore, we did not have sufficient data on height, weight, or waist circumference (WC) to compare other related TyG indices such as TyG-BMI (Body Mass Index) or TyG-WC Index. The retrospective nature of our cohort study performed in a single centre warrants future prospective studies to validate the role of the TyG index in predicting thrombolysis outcomes, and could include comparison with HOMA-IR or between patients with and without diabetes. Further studies could also investigate the effect of anti-diabetic agents which affect IR on post-thrombolysis stroke outcomes.

Conclusion
In AIS patients who received tPA, the TyG index, a measure of insulin resistance, was significantly associated with increased 90-day mortality and poorer neurological and functional outcomes.

Data availability
The data that support the findings of this study are available from National University Hospital but restrictions apply to the availability of these data, and so are not publicly available. Data are however available from the authors upon reasonable request and with permission of National University Hospital.