Correlation between surrogate indicators of insulin resistance and all-cause mortality in patients with severe hemorrhagic stroke: a multicenter retrospective cohort study in the United States

Abstract

Background

Hemorrhagic stroke (HS), including non-traumatic intracerebral hemorrhage (ICH) and subarachnoid hemorrhage (SAH), accounts for approximately 30% of all stroke cases and over 40% of stroke-related deaths. With high mortality and disability rates, HS imposes a heavy global health burden. As a core metabolic disorder, insulin resistance (IR) has been proven to be associated with all-cause mortality (ACM) in patients with ischemic stroke in previous studies. Nevertheless, its prognostic value remains unclear in critically ill HS patients admitted to the intensive care unit (ICU). Given the markedly higher mortality and morbidity of HS compared with ischemic stroke, it is essential to explore this association. This study aimed to investigate the correlations between multiple surrogate markers of insulin resistance and all-cause mortality among critically ill HS patients in the ICU setting.

Methods

Data were extracted from the public eICU-CRD database. Patients with severe HS were identified based on the International Classification of Diseases (ICD)-9/10 diagnostic codes. A total of 1538 ICU-admitted patients with severe HS were enrolled and stratified according to quartiles of various IR surrogate markers. The primary endpoint was in-hospital mortality. Cox regression analysis, Kaplan-Meier survival curves, restricted cubic splines (RCS) and receiver operating characteristic (ROC) curves were adopted for statistical analyses.

Results

Among the 1538 enrolled patients, males accounted for 54.6%, and the overall in-hospital all-cause mortality was 26.59%. Multivariate Cox regression analyses revealed that all IR surrogate markers were significantly correlated with all-cause mortality in severe HS patients. SPISE was negatively correlated with all-cause mortality, while other indicators showed positive correlations. Restricted cubic spline analyses demonstrated non-linear relationships between TyG, SPISE, TG_HDL, METS_IR, TyG_BMI, TyG_RC and mortality. No significant effect modification was observed in interaction analyses. ROC curve analysis indicated that TyG exhibited the highest predictive accuracy.

Conclusion

In conclusion, insulin resistance surrogate markers were significantly associated with all-cause mortality in critically ill HS patients. Despite their weak-to-moderate discriminative performance, these indices may serve as auxiliary prognostic references for risk stratification. Clinical application of these indicators is expected to optimize therapeutic strategies and disease progression management. Furthermore, this study enriches current evidence regarding the association between insulin resistance surrogate markers and hemorrhagic stroke, and clarifies their roles in predicting mortality across different stroke subtypes.

Graphical abstract

Introduction

According to statistics from the World Health Organization (WHO), stroke has become the second leading cause of death worldwide and the primary cause of long-term disability [1, 2]. Hemorrhagic stroke (HS) mainly consists of two subtypes: intracerebral hemorrhage (ICH) occurring within brain parenchyma and spontaneous subarachnoid hemorrhage (SAH). It presents more critical clinical conditions, accounting for approximately 30% of all stroke cases and over 40% of stroke-related deaths [3, 4]. Notably, hemorrhagic stroke has distinctly higher mortality and disability rates than ischemic stroke, especially in low- and middle-income countries [5]. Current clinical management strategies for hemorrhagic stroke mainly involve supportive care, complication prevention, hemostatic therapy, blood pressure control and surgical intervention. However, these approaches show limited efficacy in alleviating primary brain injury [6, 7], particularly in critically ill HS patients with highly variable clinical prognoses. Recent advances have indicated that multi-dimensional comprehensive management, including aggressive blood pressure control, glycemic regulation, standardized temperature management and anticoagulation adjustment, may improve functional outcomes in patients with acute intracerebral hemorrhage, further highlighting the significance of integrated therapeutic strategies [8]. With the accelerating global population aging, the number of stroke patients admitted to intensive care units (ICU) keeps rising, leading to a growing medical burden associated with stroke. Accordingly, it is urgent to identify effective indicators for predicting adverse prognoses in stroke patients. Ideal prognostic biomarkers should be simple, convenient, cost-effective and easy to popularize in routine clinical practice.

Insulin resistance (IR), a hallmark of metabolic syndrome, is characterized by impaired insulin-mediated glucose metabolism [9] and serves as a high-risk factor for macrovascular and microvascular diseases [10]. The hyperinsulinemic-euglycemic clamp is regarded as the gold standard for evaluating insulin resistance, yet its complicated and invasive procedures restrict its application in clinical research. Homeostasis model assessment of insulin resistance (HOMA-IR) is a widely accepted surrogate method calculated based on fasting blood glucose and fasting insulin levels [11]. Given that serum insulin is rarely routinely tested in primary care settings, various convenient surrogate indicators for insulin resistance have been gradually developed and applied. Accumulating evidence has validated that multiple IR surrogate markers can effectively predict cardiovascular diseases and other clinical endpoints [12–17]. Nevertheless, most existing studies only analyzed a small number of IR-related indicators, and systematic researches comprehensively exploring the associations between diverse IR biomarkers and cardiovascular diseases remain scarce. To date, no study has systematically investigated the intrinsic relationships between multiple IR-related biomarkers and all-cause mortality (ACM) in critically ill HS patients.

This study focused on the potential of IR surrogate markers as predictive indicators for mortality among critically ill HS patients, who carry substantially higher risks than those with ischemic stroke. Using data from the eICU-CRD database, we aimed to clarify the correlations between IR surrogate marker levels and all-cause mortality in this vulnerable population. This research aims to supplement existing evidence on the prognostic associations of IR surrogate markers among critically ill HS patients, provide auxiliary reference information for clinical risk stratification, and support the formulation of targeted monitoring and intervention plans for high-risk patients.

Materials and methods

Study design and data source

This was a retrospective observational cohort study. All clinical data were extracted from the eICU Collaborative Research Database (eICU-CRD), a public critical care database. This database integrates complete clinical data of critically ill patients admitted to intensive care units in more than forty tertiary hospitals across the United States, covering demographic characteristics, admission diagnoses, vital signs, laboratory test results, illness severity scores, clinical treatment regimens and long-term prognostic outcomes. All data have been standardized and quality-checked with satisfactory authenticity and population representativeness.

All data used in this study were fully de-identified without any identifiable personal privacy information. Ethical approval was waived by the institutional ethics committee, and all research procedures strictly complied with the ethical principles of the Declaration of Helsinki.

Study participants

Inclusion criteria

Eligible patients were screened strictly according to the following criteria: (1) Adult inpatients aged ≥ 18 years old; (2) Definite diagnosis of hemorrhagic stroke including non-traumatic intracerebral hemorrhage and subarachnoid hemorrhage confirmed by ICD-9 and ICD-10 diagnostic codes; (3) Admitted to ICU for monitoring and treatment with an ICU stay duration of no less than 24 h; (4) Complete data of insulin resistance-related laboratory indicators, baseline clinical information and in-hospital prognostic outcomes.

Exclusion criteria

Patients meeting the following conditions were excluded: Missing core laboratory indicators such as fasting blood glucose and triglycerides, or abnormal values including zero, negative numbers and data beyond normal clinical reference ranges; Incomplete survival outcome data that failed to determine in-hospital death events; Severe lack of key baseline information including age, gender and previous underlying diseases; Complicated with severe systemic diseases such as end-stage hepatorenal failure, malignant tumors and septic shock, which could independently affect short-term prognosis; Patients with ICU stay less than 24 h, voluntary discharge or inter-hospital transfer leading to incomplete follow-up of in-hospital outcomes. The detailed participant screening flowchart is shown in Fig. 1.

Fig. 1.

The flow chart of the enrolled patients throughout the trial. TyG, Triglyceride-glucose index; SPISE, Single-point insulin sensitivity estimator; TG_HDL, Triglyceride to high-density lipoprotein cholesterol ratio; METS_IR, Metabolic score for insulin resistance; AIP, Atherogenic index of plasma; TyG_BMI, Triglyceride-glucose body mass index; TyG_RC, Triglyceride-glucose remnant cholesterol index; CHG, Cholesterol, HDL, and Glucose index

Study indicators and grouping methods

Definition and calculation of exposure indicators

Multiple surrogate indices of insulin resistance were set as core exposure variables, and eight commonly used clinical indicators were enrolled. All indicators were calculated based on fasting venous blood samples collected within 24 h after admission [18–25].

Grouping strategy

To explore the dose-response relationship between each insulin resistance surrogate marker and in-hospital mortality risk in patients with severe hemorrhagic stroke, all enrolled participants were divided into four quartile groups (Q1, Q2, Q3, Q4) according to the values of each indicator. The lowest quartile Q1 was defined as the reference group, and the other three groups were regarded as high-level groups. The cut-off values were determined based on actual data of the enrolled population. Baseline characteristic balance tests were performed to ensure inter-group comparability.

Collection of baseline clinical data

Baseline information was systematically collected as follows: Demographic data: age, gender, height, body weight and body mass index (BMI); Disease severity scores: Acute Physiology Score (APS), Acute Physiology and Chronic Health Evaluation Ⅱ (APACHE Ⅱ); Past medical history: hypertension, diabetes mellitus, coronary heart disease, atrial fibrillation, heart failure and prior stroke; Laboratory parameters: fasting blood glucose, fasting triglyceride, routine blood tests, liver and renal function, platelet count and other admission laboratory results. All laboratory indicators mentioned above were extracted from the first blood test results obtained immediately upon ICU admission. To minimize potential bias, variables with high missing rates were excluded, including activated partial thromboplastin time ratio (missing rate = 95.3%), C-reactive protein (missing rate = 96.7%), high-sensitivity C-reactive protein (missing rate = 98.9%), erythrocyte sedimentation rate (missing rate = 87.9%), and uric acid (missing rate = 82.4%). The remaining variables had missing rates below 20%, and missing data were imputed using predictive mean matching (PMM) with 50 iterations.

Definition of outcome variables

The primary outcome was all-cause in-hospital mortality, referring to all-cause deaths occurring from ICU admission to the end of hospitalization.

Statistical analysis

All statistical analyses were performed using R 4.3.0 software. A two-tailed test was adopted with α = 0.05, and P < 0.05 was considered statistically significant.

Normally distributed continuous variables were presented as mean ± standard deviation (SD), with standard deviation used for dispersion reporting in Table 1. Non-normally distributed continuous variables and ordinal data were described as median (interquartile range, IQR), and interquartile range was reported alongside medians in Table 1 instead of 95% confidence intervals for descriptive statistics. Between-group comparisons for normally distributed continuous variables were performed using independent t-test or one-way ANOVA, while Kruskal-Wallis H test was adopted for non-normal continuous and ranked data. Categorical variables were expressed as number of cases and percentages [n (%)], and compared using the chi-square test or Fisher’s exact test as appropriate.

Table 1.

Baseline characteristics of the survivors and Non-survivors groups

Categories	Overall (N=1538)	Non-survivor (N=409)	Survivor (N=1129)	P Value
Baseline characteristics of the Survivors and Non-survivors groups
IR index
TyG	8.737 (8.712–8.762)	8.945 (8.891–8.999)	8.662 (8.635–8.689)	< 0.001
SPISE	6.345 (6.231–6.458)	6.122 (5.907–6.336)	6.425 (6.292–6.559)	0.019
TG-HDL	2.556 (2.427–2.685)	2.972 (2.717–3.227)	2.406 (2.257–2.554)	< 0.001
METS-IR	42.896 (42.287–43.504)	43.968 (42.869–45.066)	42.507 (41.781–43.233)	0.030
AIP	0.763 (0.736–0.789)	0.885 (0.828–0.941)	0.718 (0.689–0.747)	< 0.001
TyG-BMI	241.532 (238.277–244.787)	247.728 (241.753–253.702)	239.287 (235.424–243.15)	0.020
TyG-RC	223.819 (215.46–232.178)	258.856 (240.208–277.504)	211.126 (202.067–220.185)	< 0.001
CHG	13.012 (12.988–13.037)	13.124 (13.074–13.174)	12.972 (12.945–12.999)	< 0.001
Demographic
Male	840 (54.6%)	235 (57.5%)	605 (53.6%)	0.197
Female	698 (45.4%)	174 (42.5%)	524 (46.4%)	0.197
Age	69.718 (69.016–70.42)	69.33 (68.011–70.649)	69.858 (69.029–70.687)	0.506
BMI	27.604 (27.249–27.959)	27.664 (27.03–28.298)	27.582 (27.156–28.008)	0.834
Hospital time	16814.474 (16401.548–17227.4)	16906.543 (16149.469–17663.617)	16781.12 (16289.814–17272.427)	0.785
Weight	79.257 (78.115–80.4)	80.824 (78.585–83.063)	78.69 (77.363–80.017)	0.109
APS	41.263 (40.138–42.388)	52.778 (50.159–55.396)	37.091 (35.983–38.199)	<0.001
APACHE	55.577 (54.391–56.762)	66.68 (63.962–69.397)	51.554 (50.357–52.752)	<0.001
Laboratory indicators
HDL	46.982 (46.382–47.582)	46.235 (45.065–47.405)	47.252 (46.554–47.951)	0.144
LDL	84.613 (83.541–85.686)	87.02 (84.451–89.588)	83.741 (82.618–84.864)	0.022
PT	13.98 (13.74–14.221)	14.357 (13.757–14.956)	13.844 (13.6–14.089)	0.121
INR	1.184 (1.16–1.208)	1.213 (1.154–1.273)	1.173 (1.149–1.197)	0.222
APTT	30.04 (29.663–30.417)	29.82 (29.109–30.531)	30.12 (29.676–30.565)	0.482
Glucose	134.721 (132.167–137.275)	151.443 (145.38–157.505)	128.663 (126.051–131.275)	< 0.001
PLT	214.667 (210.462–218.873)	225.02 (216.358–233.681)	210.917 (206.14–215.694)	0.005
TC	156.729 (155.255–158.203)	161.599 (158.238–164.96)	154.965 (153.38–156.55)	< 0.001
TG	106.453 (103.094–109.811)	121.797 (114.139–129.455)	100.894 (97.308–104.479)	< 0.001
LYM	15.928 (15.472–16.383)	16.154 (15.228–17.081)	15.845 (15.324–16.367)	0.569
MON	8.322 (8.16–8.485)	8.242 (7.976–8.507)	8.352 (8.152–8.551)	0.516
NEU	72.957 (72.424–73.491)	72.589 (71.482–73.696)	73.091 (72.484–73.697)	0.436
Past medical history
Coronary heart disease	225 (14.6%)	67 (16.4%)	158 (14%)	0.276
Hypertension	1163 (75.6%)	313 (76.5%)	850 (75.3%)	0.665
Arrhythmia	95 (6.2%)	22 (5.4%)	73 (6.5%)	0.508
Atrial fibrillation	273 (17.8%)	83 (20.3%)	190 (16.8%)	0.135
Heart failure	156 (10.1%)	50 (12.2%)	106 (9.4%)	0.125
Heart valve disease	74 (4.8%)	27 (6.6%)	47 (4.2%)	0.066
Peripheral vascular disease	41 (2.7%)	13 (3.2%)	28 (2.5%)	0.567
Deep Vein Thrombosis	65 (4.2%)	23 (5.6%)	42 (3.7%)	0.135
Pulmonary embolism	19 (1.2%)	3 (0.7%)	16 (1.4%)	0.433
Hypothyroidism	177 (11.5%)	56 (13.7%)	121 (10.7%)	0.127
Diabetes	141 (9.2%)	34 (8.3%)	107 (9.5%)	0.549
Stroke	313 (20.4%)	84 (20.5%)	229 (20.3%)	0.970
Dementia	102 (6.6%)	24 (5.9%)	78 (6.9%)	0.543
TIA	81 (5.3%)	20 (4.9%)	61 (5.4%)	0.788
Epilepsy	89 (5.8%)	22 (5.4%)	67 (5.9%)	0.773
Malignant neoplasm	189 (12.3%)	42 (10.3%)	147 (13%)	0.172
Hematological malignancy	22 (1.4%)	7 (1.7%)	15 (1.3%)	0.752
COPD	117 (7.6%)	31 (7.6%)	86 (7.6%)	1.000
Asthma	66 (4.3%)	17 (4.2%)	49 (4.3%)	0.988
Respiratory failure	12 (0.8%)	5 (1.2%)	7 (0.6%)	0.391
Renal insufficiency	145 (9.4%)	34 (8.3%)	111 (9.8%)	0.423

In Table, the bolded P-values indicate statistically significant differences

Cox proportional hazards regression model was used to analyze the association between quartiles of insulin resistance surrogate markers and in-hospital mortality risk. Three progressively adjusted models were established: Model 1 was the crude model without any confounders; Model 2 was adjusted for age, gender and BMI on the basis of Model 1; Model 3 further adjusted for APS score, APACHE Ⅱ score, hypertension, diabetes mellitus, coronary heart disease, atrial fibrillation, heart failure, prior stroke and platelet count. For the time-to-event structure of Cox regression, the time origin was defined as hospital admission, and the follow-up period covered the entire hospital stay. The primary event was in-hospital all-cause mortality. Patients discharged alive and those transferred to other hospitals were treated as censored observations at the time of discharge or transfer. The proportional hazards assumption of each Cox model was formally verified by Schoenfeld residual test. Results were presented as hazard ratios (HR) with 95% confidence intervals (95% CI).

Kaplan-Meier survival curves were plotted, and Log-rank test was used to compare survival differences among quartile groups. Restricted Cubic Splines (RCS) with four knots were applied to explore the non-linear dose-response relationship between continuous insulin resistance indicators and in-hospital mortality risk. All insulin resistance indices were analyzed in their original raw units without standardization or logarithmic transformation; no winsorization or truncation was performed on extreme values. Knots were automatically placed at the default quantiles by the rms package. The reference value was set to zero for each exposure variable in the Predict function, and the vertical dashed red reference line at HR = 1 was marked in all figures. The y-axis of each plot represented adjusted hazard ratios (HR) derived by exponentiating the linear predictors from Cox models, with shaded bands indicating corresponding 95% confidence intervals. The P value for non-linearity was extracted from the second row of the ANOVA table of each Cox proportional hazards model constructed with RCS terms, and all non-linear P values were uniformly formatted (reported to three decimal places or as P < 0.001 where appropriate). All RCS Cox models were fully adjusted for the same set of confounders as Model 3 described above.

Static binary ROC analysis was selected rather than time-dependent ROC because the primary endpoint was in-hospital mortality, with complete follow-up for every enrolled patient until hospital discharge. No participants were lost to follow-up or censored before the end of hospitalization, so standard binary ROC yielded unbiased discriminative estimates without the need for time-dependent AUC frameworks. Delong–Delong nonparametric 95% CIs for AUC were computed for each raw biomarker.

Results

Baseline characteristics

A total of 1538 patients with severe hemorrhagic stroke (HS) were enrolled in this study. The median age of participants was 69.718 years (95%CI: 69.016–70.42), including 840 males (54.6%) and 698 females (45.4%). The overall in-hospital mortality was 26.59%. Differences in baseline characteristics between survivors and non-survivors are presented in Table 1.

In terms of insulin resistance (IR) surrogate markers, non-survivors had higher levels of TyG, TG_HDL, METS_IR, AIP, TyG_BMI, TyG_RC and CHG, as well as lower SPISE values. Regarding disease severity, non-survivors exhibited higher APS and APACHE Ⅱ scores. For laboratory parameters, non-survivors had elevated levels of LDL, blood glucose, PLT, TC and TG. All differences were statistically significant (all P < 0.05).

Clinical outcomes

Kaplan-Meier survival curves were plotted to compare survival disparities among different groups (Fig. 2). Higher levels of TyG, TG_HDL, AIP, TyG_RC and CHG were significantly correlated with increased mortality risk in patients with severe HS.

Fig. 2.

Kaplan-Meier survival curves of in-hospital all-cause mortality across groups of different IR surrogate indices in severe hemorrhagic stroke patients. A: TyG; B: SPISE; C: TG_HDL; D: METS_IR; E: AIP; F: TyG_BMI; G: TyG_RC; H: CHG

Cox regression models were adopted to quantify the associations between each IR surrogate marker and mortality, with indicators analyzed both as continuous variables and quartile categorical variables (Table 2). Across all regression models, elevated TyG, TG_HDL, METS_IR, AIP, TyG_BMI, TyG_RC and CHG were consistently linked to higher in-hospital mortality risk. In contrast, SPISE showed an inverse correlation with mortality risk, namely higher SPISE values indicated lower mortality risk. These associations remained statistically stable after multivariate stepwise adjustment (all P for trend < 0.05).

Table 2.

Cox proportional hazard ratios (HR) for all-cause mortality associated with IR surrogate indices

IR (Quartile)	Model 1			Model 2			Model 3
	HR (95% CI)	P-value	P for trend	HR (95% CI)	P-value	P for trend	HR (95% CI)	P-value	P for trend
Cox proportional hazard ratios (HR) for all-cause mortality with IR index
TyG
Continues variable per unit	1.808 (1.549–2.111)	< 0.001		2.089 (1.751–2.493)	< 0.001		2.047 (1.712–2.448)	< 0.001
Quartile			< 0.001			< 0.001			< 0.001
Q1 (n = 385)	Reference			Reference			Reference
Q2 (n = 385)	1.11 (0.798–1.544)	0.535		1.17 (0.834–1.643)	0.363		1.181 (0.836–1.669)	0.344
Q3 (n = 384)	1.301 (0.949–1.782)	0.102		1.346 (0.973–1.862)	0.073		1.325 (0.954–1.84)	0.093
Q4 (n = 384)	2.351 (1.773–3.116)	< 0.001		2.58 (1.928–3.452)	< 0.001		2.649 (1.971–3.561)	< 0.001
SPISE
Continues variable per unit	0.989 (0.944–1.037)	0.652		0.855 (0.769–0.952)	0.004		0.837 (0.748–0.937)	0.002
Quartile			0.387			0.015			0.003
Q1 (n = 385)	Reference			Reference			Reference
Q2 (n = 385)	1.061 (0.817–1.379)	0.656		0.718 (0.512–1.006)	0.054		0.706 (0.498–1.001)	0.051
Q3 (n = 384)	0.978 (0.746–1.282)	0.87		0.582 (0.384–0.884)	0.011		0.58 (0.378–0.889)	0.012
Q4 (n = 384)	0.897 (0.677–1.187)	0.446		0.406 (0.239–0.69)	0.001		0.386 (0.223–0.667)	0.001
TG-HDL
Continues variable per unit	1.024 (1.004–1.045)	0.019		1.032 (1.009–1.056)	0.007		1.022 (0.999–1.046)	0.06
Quartile			< 0.001			< 0.001			< 0.001
Q1 (n = 385)	Reference			Reference			Reference
Q2 (n = 385)	0.685 (0.491–0.956)	0.026		0.734 (0.519–1.038)	0.08		0.762 (0.537–1.082)	0.128
Q3 (n = 384)	1.293 (0.982–1.701)	0.067		1.396 (1.044–1.867)	0.024		1.45 (1.082–1.944)	0.013
Q4 (n = 384)	1.572 (1.206–2.051)	0.001		1.754 (1.33–2.313)	< 0.001		1.791 (1.355–2.369)	< 0.001
METS-IR
Continues variable per unit	1.002 (0.994–1.009)	0.684		1.032 (1.015–1.049)	< 0.001		1.029 (1.012–1.047)	0.001
Quartile			0.113			0.001			< 0.001
Q1 (n = 385)	Reference			Reference			Reference
Q2 (n = 385)	1.154 (0.859–1.55)	0.341		1.607 (1.147–2.253)	0.006		1.75 (1.223–2.504)	0.002
Q3 (n = 384)	1.31 (0.982–1.749)	0.066		2.154 (1.466–3.164)	< 0.001		2.216 (1.478–3.32)	< 0.001
Q4 (n = 384)	1.234 (0.93–1.638)	0.146		3.162 (1.873–5.338)	< 0.001		3.357 (1.967–5.73)	< 0.001
AIP
Continues variable per unit	1.42 (1.212–1.665)	< 0.001		1.498 (1.265–1.774)	< 0.001		1.473 (1.246–1.743)	< 0.001
Quartile			< 0.001			< 0.001			< 0.001
Q1 (n = 385)	Reference			Reference			Reference
Q2 (n = 385)	0.685 (0.491–0.956)	0.026		0.734 (0.519–1.038)	0.08		0.762 (0.537–1.082)	0.128
Q3 (n = 384)	1.293 (0.982–1.701)	0.067		1.396 (1.044–1.867)	0.024		1.45 (1.082–1.944)	0.013
Q4 (n = 384)	1.572 (1.206–2.051)	0.001		1.754 (1.33–2.313)	< 0.001		1.791 (1.355–2.369)	< 0.001
TyG-BMI
Continues variable per unit	1 (0.999–1.002)	0.65		1.021 (1.015–1.027)	< 0.001		1.02 (1.014–1.025)	< 0.001
Quartile			0.095			< 0.001			< 0.001
Q1 (n = 385)	Reference			Reference			Reference
Q2 (n = 385)	1.331 (0.99–1.79)	0.058		2.075 (1.458–2.951)	< 0.001		2.172 (1.491–3.165)	< 0.001
Q3 (n = 384)	1.374 (1.027–1.84)	0.033		2.805 (1.831–4.296)	< 0.001		2.947 (1.87–4.644)	< 0.001
Q4 (n = 384)	1.308 (0.98–1.747)	0.068		5.009 (2.693–9.319)	< 0.001		5.071 (2.662–9.662)	< 0.001
TyG-RC
Continues variable per unit	1.001 (1–1.001)	0.001		1.001 (1–1.001)	< 0.001		1.001 (1–1.001)	0.001
Quartile			< 0.001			< 0.001			< 0.001
Q1 (n = 385)	Reference			Reference			Reference
Q2 (n = 385)	1.199 (0.878–1.637)	0.254		1.277 (0.921–1.771)	0.143		1.308 (0.942–1.818)	0.109
Q3 (n = 384)	1.019 (0.744–1.395)	0.908		1.119 (0.808–1.55)	0.5		1.105 (0.795–1.536)	0.55
Q4 (n = 384)	2.106 (1.609–2.757)	< 0.001		2.438 (1.835–3.239)	< 0.001		2.53 (1.894–3.378)	< 0.001
CHG
Continues variable per unit	1.484 (1.221–1.805)	< 0.001		1.605 (1.305–1.974)	< 0.001		1.579 (1.28–1.946)	< 0.001
Quartile			< 0.001			< 0.001			< 0.001
Q1 (n = 385)	Reference			Reference			Reference
Q2 (n = 385)	1.154 (0.848–1.569)	0.362		1.207 (0.878–1.659)	0.246		1.202 (0.87–1.661)	0.265
Q3 (n = 384)	1.304 (0.972–1.751)	0.077		1.338 (0.985–1.817)	0.063		1.306 (0.957–1.782)	0.093
Q4 (n = 384)	1.711 (1.298–2.255)	< 0.001		1.825 (1.372–2.427)	< 0.001		1.81 (1.354–2.42)	< 0.001

Model 1: unadjusted. Model 2: adjusted for gender + age + BMI. Model 3: adjusted for gender + age + BMI + APS score + APACHE II score + previous coronary heart disease + previous hypertension + previous atrial fibrillation + previous heart failure + previous diabetes mellitus + previous stroke, and platelet count. In Table, the bolded P-values indicate statistically significant differences

Restricted Cubic Spline (RCS) analysis with adjustment for potential confounders further verified the above correlations (Fig. 3). All IR surrogate markers exerted strong overall associations with mortality in severe HS patients. Obvious non-linear relationships were observed for TyG, SPISE, TG_HDL, METS_IR, TyG_BMI and TyG_RC (non-linear P < 0.05), while other indicators presented approximately monotonic risk trends within the observed range.

Fig. 3.

Dose-response associations between various IR surrogate markers and mortality among severe hemorrhagic stroke patients, examined by RCS analysis. The solid line represents the adjusted mortality probability, and the shaded area represents the 95% confidence interval. The models were adjusted for gender, age, BMI, APS, APACHE II score, history of coronary heart disease, hypertension, atrial fibrillation, heart failure, diabetes mellitus, previous stroke, and platelet count. A: TyG; B: SPISE; C: TG_HDL; D: METS_IR; E: AIP; F: TyG_BMI; G: TyG_RC; H: CHG

Subgroup analysis

Subgroup analysis was conducted to verify the consistency of predictive efficacy of IR indices for short-term mortality across different subgroups (Fig. 4). Specifically, the protective effect of SPISE against mortality was more prominent in patients without prior stroke history (0.94 vs. 1.17, P for interaction < 0.001). TG_HDL showed stronger correlation with mortality among patients with BMI < 30 (1.07 vs. 1.01, P for interaction = 0.029). METS_IR had a more remarkable prognostic effect in patients with previous stroke history (1.01 vs. 0.97, P for interaction = 0.001). TyG_BMI was more closely associated with mortality in patients without prior stroke (1 vs. 0.99, P for interaction = 0.001). No significant heterogeneity was found in the associations between other IR surrogate markers and mortality risk across other stratified subgroups.

Fig. 4.

Forest plots of hazard ratios for the hospital mortality in different subgroup. HR: hazard ratio; CI: confidence interval. A: TyG; B: SPISE; C: TG_HDL; D: METS_IR; E: AIP; F: TyG_BMI; G: TyG_RC; H: CHG

Comparison of predictive efficacy

Receiver operating characteristic (ROC) curves were used to evaluate the predictive performance of eight insulin resistance-related indicators for in-hospital mortality in patients with severe hemorrhagic stroke (Fig. 5). The area under the curve (AUC) and corresponding 95% confidence intervals were as follows: TyG (AUC = 0.66, 95%CI: 0.628–0.692), TyG_RC (AUC = 0.626, 95%CI: 0.593–0.659), CHG (AUC = 0.596, 95%CI: 0.563–0.629), TG_HDL and AIP (both AUC = 0.58, 95%CI: 0.549–0.611), TyG_BMI (AUC = 0.556, 95%CI: 0.525–0.588), METS_IR (AUC = 0.555, 95%CI: 0.523–0.587), SPISE (AUC = 0.547, 95%CI: 0.515–0.578). The results demonstrated that TyG had the highest AUC value, indicating its superior predictive ability for in-hospital mortality compared with other insulin resistance surrogate markers.

Fig. 5.

ROC curves of various IR surrogate indices for predicting mortality in severe hemorrhagic stroke patients

Discussion

In this study, we explored the associations between insulin resistance surrogate markers and in-hospital all-cause mortality among critically ill HS patients from a US ICU cohort, and found that these markers remained significantly associated with mortality after adjustment for multiple confounding clinical covariates. Interaction analyses did not reveal meaningful effect modification across predefined subgroups. Collectively, IR surrogate markers may provide auxiliary reference for clinical risk stratification in critically ill HS patients and show independent statistical associations with mortality. Among all tested indices, the TyG index exhibited the strongest correlative discrimination for mortality outcome.

Insulin resistance (IR) is one of the core characteristics of metabolic syndrome. It is not only closely associated with diabetes mellitus, obesity and dyslipidemia, but also an independent risk factor for the occurrence and development of cardio-cerebrovascular diseases, and is remarkably correlated with poor prognosis in patients suffering from these diseases [26]. Multiple studies have confirmed that IR is positively associated with the risk of cardiovascular diseases in prediabetic individuals [27]. A Korean cohort study with a median follow-up of 9.83 years revealed that IR could increase the risks of all-cause mortality, cardiovascular mortality and adverse cardiovascular events by 87%, 133% and 267% respectively among patients with cardiovascular diseases [28]. The hyperinsulinemic-euglycemic clamp test is the gold standard for diagnosing IR, yet its complicated operation and high cost restrict its widespread application in large-scale clinical studies. Accordingly, a variety of simple surrogate indicators for IR evaluation have been widely applied in clinical research. Among them, homeostasis model assessment of insulin resistance (HOMA-IR) is the most commonly used surrogate marker currently, while its calculation depends on the detection of fasting insulin levels [29]. The triglyceride-glucose index (TyG index) has gained wide attention due to convenient detection and simple calculation. Previous studies have shown that it has a sensitivity of 96.5% and a specificity of 85.0% in diagnosing IR compared with the hyperinsulinemic-euglycemic clamp test [30]. In addition, multiple derived indicators of TyG have also been extensively used in clinical research [31–33].

At present, there are abundant studies focusing on the relationship between IR surrogate indicators and cardiovascular disease prognosis, whereas relevant evidence targeting critically ill HS patients remains insufficient. Most existing studies concentrate on short-term in-hospital outcomes. For example, two cohort studies conducted in China found that the TyG index was positively correlated with 1-year follow-up adverse events including all-cause mortality among hospitalized patients with coronary heart disease combined with hypertension [34, 35]. Other studies demonstrated that the TyG-BMI index was positively related to adverse cardiovascular outcomes in elderly and female patients with coronary heart disease who underwent percutaneous coronary intervention [36]. As for HOMA-IR, relevant researches indicated that it was a favorable predictor of all-cause mortality in patients with coronary heart disease complicated with hypertension [37].

Nevertheless, considering the potential confounding effect of hyperglycemia, the clinical application of IR surrogate indicators has aroused prudent discussions among numerous scholars. As a pathological state closely linked to obesity, hyperlipidemia, hypertension and other manifestations of metabolic syndrome, insulin resistance has always been a core research focus in relevant fields. To verify the clinical application value of TyG, this indicator has been widely adopted to assess IR status in various high-risk populations defined as disease-vulnerable groups in numerous clinical studies. Relevant results have confirmed that TyG is an effective tool for IR evaluation, and presents prominent accuracy in predicting disease risks especially among young and middle-aged populations [30]. Essentially, compared with separate detection of blood glucose or triglyceride, TyG can more comprehensively reflect the characteristics of disease progression in specific clinical scenarios [38]. Hence, TyG has been proven to be a reliable biomarker for evaluating a series of cardio-cerebrovascular diseases and other disorders related to metabolic dysfunction.

The specific pathophysiological mechanisms underlying the associations between IR surrogate indicators and the onset, progression and mortality risks of cerebrovascular diseases remain to be further elucidated. Studies have indicated that blood glucose levels can reflect liver metabolism-related IR status to a certain extent, while triglyceride levels mainly reflect IR conditions in adipose tissue, so TyG can comprehensively evaluate IR status derived from the above two aspects [39]. Furthermore, IR surrogate indicators are closely involved in endothelial dysfunction, inflammatory response, accelerated foam cell formation and vascular smooth muscle proliferation, all of which are key pathological links in the early stage of atherosclerosis [40–42]. A study conducted by Miao et al. confirmed that IR surrogate indicators were significantly correlated with the severity of carotid atherosclerosis, suggesting their potential value as biomarkers for atherosclerosis [43]. In addition, Che et al. found that elevated levels of IR surrogate indicators were significantly associated with increased risk of cerebrovascular diseases after adjusting for known confounding factors [44]. The influence of IR is not limited to the initiation of atherosclerosis; it can also promote the progression of advanced atherosclerotic plaques by inducing apoptosis of vascular smooth muscle cells. IR surrogate indicators can comprehensively reflect glucose metabolism, inflammatory response, oxidative stress status [45], as well as advanced glycation end-product metabolism and platelet activity, all of which may affect endothelium-dependent vasodilation function. Moreover, elevated TyG levels usually indicate increased free fatty acids, which are commonly accompanied by IR status [9, 46, 47]. Therefore, reducing the levels of IR surrogate indicators may become an auxiliary intervention target for populations susceptible to cerebrovascular diseases. The interaction of the above various pathophysiological changes jointly promotes the occurrence and progression of cerebrovascular diseases, and ultimately leads to adverse clinical outcomes.

The core innovation of this study lies in confirming for the first time based on an American critical care cohort that IR surrogate indicators possess predictive value for mortality risk among critically ill HS patients, among which the TyG index shows the optimal predictive efficacy. Meanwhile, several limitations of this study need to be acknowledged. Firstly, the observational research design fails to establish causal relationships between IR surrogate indicators and mortality in critically ill HS patients. Secondly, all laboratory parameters were only collected at baseline, making it impossible to assess the dynamic changes of these indicators and their potential impacts on disease progression. Thirdly, despite comprehensive adjustment for sociodemographic factors and clinical covariates, residual confounding caused by unmeasured variables and inherent measurement variability cannot be completely excluded. In addition, due to the lack of key biomarkers including fasting insulin level and visceral fat area in the eICU-CRD database, not all available IR surrogate indicators were included for analysis. Fourthly, all blood samples used to calculate insulin resistance surrogate markers were obtained within 24 h after ICU admission. Critically ill patients with hemorrhagic stroke experience severe acute neuroendocrine stress at early hospitalization. Uncertain fasting status, stress hyperglycemia, fluid resuscitation, nutritional support, insulin or lipid-lowering agents, concomitant infection and organ dysfunction can substantially alter circulating glucose and lipid concentrations. Therefore, single early-admission laboratory measurements may not accurately reflect long-term baseline insulin resistance, introducing potential measurement bias to our analyses. Fifthly, the insulin resistance surrogate markers analyzed in this study (TyG, TG/HDL-C, METS-IR, SPISE, AIP and other related indices) were originally derived and validated in general or chronic metabolic populations. Their reliability and clinical validity for reflecting chronic insulin resistance remain unconfirmed in critically ill hemorrhagic stroke patients under acute stress conditions. Sixthly, we performed numerous parallel statistical tests across eight IR surrogate markers, multiple Cox models and subgroup stratifications without multiple comparison correction, which may increase the risk of type I error and false positive associations. The significant correlations observed in this exploratory cohort need to be prospectively validated in independent external hemorrhagic stroke populations before clinical extrapolation. Seventhly, this study exclusively enrolled patients with established hemorrhagic stroke, which renders our analysis vulnerable to index-event bias. Multiple biological and clinical factors could simultaneously contribute to the initial onset of hemorrhagic stroke and drive post-admission in-hospital mortality; such shared confounders may generate spurious statistical correlations or distort the magnitude of the observed associations between IR surrogate markers and mortality outcomes. While this methodological limitation is inherent to all retrospective analyses focusing on single-disease inpatient cohorts, we explicitly acknowledge this potential bias to temper the interpretation of our effect estimates. Finally, given that this study was conducted based on an American ICU cohort, caution should be exercised when extending these conclusions to other populations and ethnic groups. Further verification studies among diverse populations are required to confirm the general applicability of our findings.

Conclusion

In conclusion, surrogate markers of insulin resistance are significantly correlated with all-cause mortality in critically ill HS patients. Despite only weak-to-moderate discriminative performance, these markers still show prognostic relevance for stratifying mortality risk among HS patients and may provide auxiliary reference information for clinical risk assessment. Furthermore, this study enriches current evidence regarding the association between insulin resistance surrogate markers and hemorrhagic stroke, and clarifies their roles in predicting mortality across different stroke subtypes.

Abbreviations

ACM

all-cause mortality

AF

Atrial fibrillation

AIP

Atherogenic Index of Plasma

APACHE Ⅱ

Acute Physiology and Chronic Health Evaluation Ⅱ

APS

Acute Physiology Score

APTT

Activated partial thromboplastin time

AUC

area under the curve

BMI

body mass index

CHD

Coronary heart disease

CHG

Cholesterol, HDL, and Glucose index

CI

confidence intervals

COPD

Chronic obstructive pulmonary disease

eICU-CRD

eICU Collaborative Research Database

FBG

Fasting blood glucose

HDL

High-density lipoprotein

HOMA-IR

Homeostasis model assessment of insulin resistance

HR

hazard ratios

HS

Hemorrhagic stroke

ICD

International Classification of Diseases

ICH

intracerebral hemorrhage

ICU

intensive care unit

INR

International normalized ratio

IR

insulin resistance

LDL

Low-density lipoprotein

LYM

Lymphocyte count

METS-IR

Metabolic Score for Insulin Resistance

MON

Monocyte count

NEU

Neutrophil count

PLT

Platelet count

PT

Prothrombin time

RCS

restricted cubic splines

ROC

receiver operating characteristic

SAH

subarachnoid hemorrhage

SPISE

Single Point Insulin Sensitivity Estimator

TC

Total cholesterol

TG

Triglyceride

TG/HDL-C

Triglyceride-to-high-density lipoprotein cholesterol ratio

TyG

Triglyceride-glucose index

WHO

World Health Organization

Author contributions

Dewei Zou, Bo Lin and Bo Wu conceived and designed the study; Wanli Yu, Gang Zhang and Haotian Jiang extracted and curated the data; Dewei Zou, Bo Lin and Bo Wu performed the statistical analyses and drafted the manuscript; Chuan Shao critically reviewed the data and revised the manuscript for important intellectual content. Nan Wu and Chuan Shao supervised the study and checked the manuscript. All authors made intellectual contributions to the manuscript and approved the submission.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not–for—profit sectors.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

This study was conducted in accordance with the ethical principles outlined in the Declaration of Helsinki. Since the data were obtained from the publicly available eICU-CRD database, ethical approval and informed consent were waived for this study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.