Triglyceride-glucose index linked to hospital mortality in critically ill stroke patients: an observational multicentre study on eICU database


 Background: The triglyceride‑glucose (TyG) index is a reliable surrogate of insulin resistance and a marker for ischemic stroke (IS) incident. Whether the TyG index predicts stroke outcome remains uncertain. This study investigated the prognostic value of the TyG index in critically ill stroke patients.Methods: This was a retrospective observational study that included stroke patients, and all data were extracted from the eICU Collaborative Research Database. The TyG index was calculated as the ln (fasting glucose level [mg/dL] × triglyceride level [mg/dL]/2). The outcomes included the hospital and intensive care unit (ICU) death. Multivariate logistic regression was used to determine independent risk factors. The smoothing curves and forest plots were illustrated.Results: A total of 4570 eligible subjects were enrolled. The mean level of TyG index was 9.1 ± 0.7. The hospital and ICU mortality rate were 10.3% and 5.0% respectively. TyG index as a continuous variable was associated hospital mortality in univariate analysis (OR 1.723, 95% CI 1.524-1.948, P < 0.001), adjusted model 1 (OR 1.861, 95% CI 1637-2.116, P < 0.001) and adjusted model 2 (OR 2.543, 95% CI 1.588-4.073, P < 0.001). TyG was also associated ICU mortality in univariate analysis (OR 2.146, 95% CI 1.826-2.523, P < 0.001), adjusted model 1 (OR 2.183, 95% CI 1.847-2.580, P < 0.001), and adjusted model 2 (OR 2.672, 95% CI 1.376-5.188, P < 0.001). The smoothing curves observed a continuous linear association after adjusting all covariates both in hospital and ICU mortality. Subgroup analysis demonstrated TyG index was associated with increased risk of hospital and ICU death in critically ill IS (P < 0.05), but not in hemorrhage stroke (P > 0.05).Conclusion: The TyG index is a potential predictor for hospital and ICU mortality in critically ill stroke patients, especially in IS patients.

Stroke is a leading cause of mortality and disability worldwide. According to global burden of disease (GBD) 2016 stroke surveys, it was estimated that there were 5.5 million deaths due to stroke, and the global lifetime risk of stroke from the age of 25 years onward was approximately 25%. [1,2] The economic costs of treatment and post-stroke care are substantial, while the global burden of stroke has been increasing. Furthermore, an increasing number of stroke patients, who have higher hospital mortality, are being admitted to an intensive care unit (ICU) for management of severe neurological impairment and post-stroke complications. [3,4] Therefore, it is still important to identify the controllable risk factors targeted by preventive strategies and health-care management that lead to a larger decrease in stroke mortality.
The triglyceride-glucose (TyG) index combines both levels of fasting plasma glucose and triglyceride, and it has been reported to be signi cantly correlated with insulin resistance (IR) and to be a reliable surrogate marker of IR. [5] Previous studies have indicated that the TyG index is associated with cardiovascular disease morbidity and mortality in the general population and different types of patient cohorts. [6][7][8][9] Lately, the TyG index has been proposed as a direct marker for the risk of incident ischemic stroke (IS) in general population. [10] To date, no relevant study has focused on the impact of the TyG index on mortality in patients with critically ill stroke including IS and hemorrhage stroke (HS). Therefore, this study aimed to speci cally investigate the association between TyG index and mortality of stroke in a large observational multicentre study on eICU database. The primary outcome was hospital mortality, with ICU mortality as a secondary outcome.

Methods
Variables with more than 10% missing values were excluded from the analysis. Multiple imputation was performed for variables with missing values of less than 10%.

Statistical analyses
Categorical variables were expressed in absolute number with percentage and analyzed by chi-square or Fisher exact test. Continuous variables were rst assessed for normality. Normal data were expressed in mean with standard deviation (SD) and compared using Student T test or one-way ANOVA. Non-normal data were expressed in median with interquartile range (IQR) and were compared using Wilcoxon ranksum test. Variables with two-tailed p value less than 0.05 were considered to be statistically signi cant and were included in the regression model. Multivariate-adjusted odds ratios (OR) and 95% con dence intervals (CI) for the study outcomes and TyG index (1 unit and quartile) were calculated by logistic regression analysis. The multivariate model 1 included gender, age, ethnicity, while model 2 included model 1 plus stroke types, coronary heart disease, diabetes mellitus, heart failure, plasma glucose, total cholesterol (TC), low-density lipoprotein-cholesterol (LDL-C), triglyceride, mechanical ventilation, APACHE IV score, LOS. All of the analyses were performed using the SPSS version 26 (Chicago, Illinois). All p values were two-tailed and a p value < 0.05 was considered statistically signi cant. The smoothing curves and forest plots were illustrated by EmpowerStats (X&Y Solutions, Inc., Boston, MA).

Results
The initial search identi ed 200,859 ICU admissions from the eICU database. A total of 6849 subjects with a primary diagnosis of brain stroke were identi ed. A total of 2279 subjects were excluded because they did not have mortality values. The nal cohort included 4570 patients, including 4,101 (89.7%) survivors before hospital discharge and 4341 (95.0%) survivors before ICU discharge. The study patients had an average age of 66.3 ± 14.2 years and 2358 (51.6%) patients were male. The mean level of TyG index was 9.1 ± 0.7. The hospital and ICU mortality rate were 10.3% and 5.0% respectively.

Baseline characteristics
The baseline characteristics between survivors and non-survivors groups are described in Table 1.
Hospital non-survivor group had higher age, plasma glucose, triglyceride, ICU LOS than hospital survivor group (P < 0.05), while lower TC and LDL-C (P < 0.05). There was no signi cant difference on sex, ethnicity, height, weight, HDL-C, hypertension, atrial brillation, hospital LOS (P > 0.05). Furthermore, hospital non-survivor group had more frequent diabetes mellitus, coronary heart disease, heart failure, mechanical ventilation than hospital survivor group (P < 0.05). Stroke type was signi cantly different between the two groups (P < 0.05). Patients in the hospital non-survivor group had higher APACHE IV score (70.8 ± 25.0 vs 47.7± 19.0, P < 0.001) and TyG index (9.3 ± 0.7 vs 9.0 ± 0.7, P < 0.001) than those survivor group.
In addition, there was no signi cant difference on age and TC between ICU survivors and ICU nonsurvivors groups (P > 0.05). ICU non-survivor group had lower hospital LOS than ICU survivor group (P < 0.001). Patients in the ICU non-survivor group had higher APACHE IV score (70.8 ± 25.0 vs 47.7± 19.0, P < 0.001) and TyG index (9.3 ± 0.7 vs 9.0 ± 0.7, P < 0.001) than those survivor group. Other results in ICU groups were consistent with those in-hospital groups.
TyG index and mortality of critically ill stroke Univariate and multivariate logistic regression revealed the association between TyG index and critically ill stroke mortality ( In addition, all participants were strati ed into four groups based on the quartile of TyG index. The mean levels of TyG index were 8.3 ± 0.3, 8.8 ± 0.1, 9.2 ± 0.1, and 10.0 ± 0.1, respectively. The hospital and ICU mortality signi cantly increased with increasing quartiles of the TyG index (P < 0.001) (Fig. 1). When dividing TyG index into quartiles, we observed a quartile increment in TyG index was associated with the increased hospital and ICU mortality of critically ill stroke in univariate, adjusted model 1 and adjusted model 2 logistic regression analysis, with a signi cant trend across the quartiles (P for trend<0.001, Table  2).
To further explore the relationship between TyG index and the mortality of critically ill stroke, we plotted the smoothing curves of TyG index against the hospital and ICU mortality of critically ill stroke (Fig. 2). In this plot, we observed a continuous linear association after adjusting all covariates both in hospital ( Fig.  2A) and ICU (Fig. 2B) mortality. This nding was consistent with the stepwise increased OR in the analysis of multivariate logistic regression.

Relationship of TyG index to mortality according to stroke types
The stroke patients were grouped into 4 types: IS group, HS group, IS with HS group, and unknown/others group. Multivariate logistic regression models were analyzed to identify the association between TyG index and mortality according to the stroke type and TyG index. We found that a higher TyG index was signi cantly associated with the increased hospital and ICU mortality in IS group and unknown/others group ( Fig. 3 and Fig. 4, adjusted P <0.05). However, the similar result did not occur in HS group and IS with HS group (Fig. 3 and Fig. 4, adjusted P > 0.05).

Discussion
To the best of our knowledge, this is the rst study to evaluate the association of the TyG index with hospital and ICU mortality in critically ill stroke patients. The main ndings are as follows: (1) the TyG index is an independent predictor for hospital and ICU mortality in patients with critically ill stroke; (2) the hospital and ICU mortality correlated proportionally with the increment of TyG index, implicating the linearity of TyG index as an indicator of critically ill stroke; (3) subgroup analysis demonstrated TyG index was associated with increased risk of hospital and ICU death in critically ill IS, but not in HS.
The TyG index, as the product of fasting plasma glucose and triglyceride, is a novel index that has been well recognized as a simple and reliable surrogate of IR. [12] The homeostatic model assessment of IR (HOMA-IR) has been traditionally used to estimate IR. [13] However, insulin levels must be required to calculate the HOMA-IR index. Compared with the inconvenient HOMA-IR, the TyG index does not require levels of insulin and may apply to all of the patients and healthy population. Recent studies indicated that the TyG index has been shown to be superior to HOMA-IR in predicting IR. [14] Furthermore, several studies conducted in Asia and Europe validated the strong association between TyG and incidence of diabetes mellitus. [12,15] Won et al. reported the TyG index was independently associated with arterial stiffness in a relatively healthy Korean population. [16] Other studies demonstrated the TyG index was an independent predictor of coronary artery calci cation progression and risk of cardiovascular diseases. [8,17] In addition, previous studies suggested the TyG index predicted severity and outcomes in patients with acute coronary syndrome. [6,18,19] Wang et al. reported the TyG index predicted future cardiovascular events in patients with diabetes and acute coronary syndrome independently of known cardiovascular risk factors. [7] However, in a previous study based on a Caucasian population, the TyG index displayed an insigni cant association with stroke. [20] Later, another study based on the same Caucasian cohort, identi ed metabolic health and obesity states groups based on TyG index was signi cantly associated with the risk of IS. [21] Additionally, a recently published epidemiological investigation expanded the use of TyG index as a direct marker for the risk of IS. [10] However, no relevant study focus on the association between TyG index and outcome of stroke. The present study investigated the relationship between the TyG index and the hospital mortality in critically ill stroke on eICU database.
We found that the TyG index was signi cantly associated with mortality in ICU stroke after adjusting for confounding factors. Furthermore, the TyG index had a continuous linear correlation with the mortality of ICU stroke.
Although the mechanism underlying the relationship between the TyG index and stroke is not fully elucidated, the TyG index has been deemed as a useful atherogenic indicator linked to IR and metabolic syndrome. IR may be the mechanism in increasing mortality in stroke. Firstly, IR may increase proin ammatory cytokines and enhance prothrombotic responses, thus exacerbating damage in the brain after stroke. [22,23] Secondly, IR may cause sympathetic activity and catabolism in muscles, thus enhancing muscle loss and leading to poor functional outcomes. [24] Thirdly, IR may increase platelet adhesion, enhance atherosclerosis progression, facilitate plaque instability, and therefore contribute to severity of stroke, via promoting apoptosis of vascular smooth muscle cells macrophages, and endothelial cells. [25,26] Fourthly, recent evidence has proved that triglyceride and glucose disorder are the risk factors of stroke, while the formula of TyG index is composed of triglyceride and glucose. [27,28] Lastly, IR may augment the role of modi able risk factors of stroke, such as hypertension, atrial brillation. [10] Although the present study had showed the association between TyG index and poststroke outcomes, the underlying molecular mechanisms involved in this association should be further investigated in the future study.
Stroke accounts for almost 5% of all disability-adjusted life-years and 10% of all deaths worldwide. [2] From different epidemiological surveys, stroke mortality in different periods varies from 5%-72 %. [29,30] Stroke mortality has been declining since the early 20th century. [31] The reasons for this are not completely understood, although the decline is welcome. [31] However, due to the increasing complexity of stroke treatment and severe conditions, an increasing proportion of acute stroke patients are being admitted to an ICU. [32,33] A study including 4958 consecutive stroke patients reported 347 (7.0%) patients required ICU admission at any time point during their index hospitalization. [4] In-hospital mortality of ICU stroke reported in the literature varies widely, highly depending on the patient characteristics. A previous small study found the mortality was 38.7% (43/111) in stroke patients requiring ICU admission. [32] An United States study including 448 ICU stroke patients provided hospital mortality was 30%. [34] Recently, a prospective observation trial reported the mortality was 7.5% in critically ill stroke patients. [35] These patients with high hospital mortality mainly associated with older age, poor neurological severity at admission, high APACHE score, impaired consciousness, intracranial hemorrhage and need for mechanical ventilation. [32,33] Moreover, functional outcomes in survivors appear to be poor. [36] The current study based on eICU database, indicated the hospital and ICU mortality were 10.3% and 5.0%, respectively. Furthermore, we found a new risk factor, the TyG index, for the mortality of critically ill stroke.
In subgroup analysis, we demonstrated TyG index was associated with increased risk of hospital and ICU death in critically ill IS, but not in HS. In agreement with previous GBD 2016 report, we did not estimate the mortality due to subarachnoid hemorrhage and intracerebral hemorrhage separately. [1] Of the total number of prevalent strokes, over 80% were IS. However, the number of global deaths due to IS was slightly lower than the number due to HS deaths. [1] In contrast, a previous neurology ICU study demonstrated HS was more frequent than IS (71.9% vs 28.1%). [37] Several studies, including our present study, described TyG index was a useful marker in IS. However, to our knowledge, no relevant study has evaluated the relationship between the TyG index and HS. Although previous studies had focusing on the relationship between IR and HS, the results had been discrepancy. The Rotterdam study and the Uppsala study had previously examined the association between IR and risk of HS, nding virtually no evidence of an association. [38,39] Later, a large United States stroke cohort showed IR may be a protective effect on HS.
[40] In the current study, we did not observe a signi cant association of the TyG index with all-cause death in critically ill HS in either unadjusted or adjusted analysis.
Although our study based on a large multicentre critical care database, it still has some limitations. First, this was a retrospective analysis derived from an observational study, which could not de nitively establish causality. Second, the eICU v2.0 did not contain the data on head imaging, neurological severity scores, and follow-up outcomes after discharge. Residual confounding could exist. Third, the baseline levels of plasma glucose and triglyceride could be affected by the use of antidiabetic and lipid-lowering drugs before ICU admission. The TyG index might have changed during hospital; therefore, it is unknown whether the change in the TyG index could have predicted the mortality. Fourth, we did not measure HOMA-IR because the examination of insulin levels is not included in the eICU v2.0. Last, the data were from the United States, and thus the results may not apply fully to ICUs elsewhere with different practices or resources.

Conclusions
In this multicentre critically ill stroke cohort, we observed the association of the TyG index with hospital and ICU mortality in stroke patients. For the rst time, this study demonstrated that the TyG index is a potential predictor for hospital and ICU mortality in critically ill stroke patients, especially in IS patients. Furthermore, the TyG index has a linear correlation with the mortality of ICU stroke. Most importantly, these ndings suggested that the TyG index may be a useful indicator for risk strati cation and prognosis in patients with critically ill stroke. Further prospective studies are required to con rm our ndings.   The hospital and ICU mortality according to TyG index quartiles. ICU intensive care unit, TyG triglycerideglucose Figure 2 The smoothing curves of the hospital (A) and ICU (B) mortality of critically ill stroke against TyG index.
ICU intensive care unit, TyG triglyceride-glucose