Diabetes–TyG interaction and mortality in critical cerebrovascular disease: a retrospective cohort study Authors: Hui Wang, Xiaojun Ding, Xuelei Cao, Jinglei Lv Published: 2026 ABSTRACT AbstractCritically ill cerebrovascular disease (CVD) patients face high mortality, yet modifiable prognostic factors remain underexplored. The triglyceride-glucose (TyG) index has emerged as a surrogate marker of insulin resistance, but its association with mortality in neurocritical care and its interaction with diabetes status remain unclear. This retrospective cohort study analyzed 2,584 CVD adults from the MIMIC-IV database (2008–2019). The TyG index was calculated from admission glucose and triglyceride values. Multivariable Cox models assessed TyG-mortality associations, with interaction and stratified analyses by diabetes status. Outcomes included 30-, 90-, and 365-day all-cause mortality. In the overall cohort, TyG showed no significant association with 30-day (HR 1.15, 95%CI 1.00–1.32) or 365-day mortality (HR 1.11, 95%CI 0.98–1.24), with a borderline association at 90 days (HR 1.14, 95%CI 1.00–1.30). However, diabetes significantly modified this effect (P for interaction = 0.033, 0.004, and 0.005 at 30, 90, and 365 days, respectively). In non-diabetic patients, each TyG unit increment was associated with significantly higher mortality at all time points (30-day: HR 1.25, 95%CI 1.04–1.50; 90-day: HR 1.26, 95%CI 1.07–1.49; 365-day: HR 1.21, 95%CI 1.03–1.40), whereas no association was observed in diabetic patients. Diabetes critically modifies the TyG-mortality relationship in CVD. The TyG index may serve as a prognostic biomarker primarily in non-diabetic patients, suggesting that diabetes-stratified approaches warrant further investigation in neurocritical care.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-64021-w.Keywords: TyG index, Mortality, Cerebrovascular disease, MIMIC-IV databaseSubject terms: Biomarkers, Diseases, Endocrinology, Medical research, Neurology Introduction Cerebrovascular disease (CVD), encompassing stroke and related pathologies, remains a leading global cause of death and long-term disability, imposing a substantial socioeconomic burden1–5. Annually, millions are admitted to intensive care units (ICUs) worldwide due to critical cerebrovascular events, such as severe ischemic or hemorrhagic stroke, subarachnoid hemorrhage, or complex post-operative states6–8. Despite advances in acute management, mortality rates among critically ill cerebrovascular patients remain alarmingly high, ranging from 5.7% to 50.2% during hospitalization or within the first months post-event6,9,10. Survivors often face significant functional impairment and reduced quality of life11–13. Identifying robust, modifiable prognostic factors in this vulnerable population is therefore paramount for risk stratification and guiding targeted interventions to improve outcomes. The triglyceride-glucose (TyG) index has emerged as a simple, cost-effective surrogate marker of insulin resistance (IR) and systemic metabolic dysfunction14,15. Substantial evidence links elevated TyG index to increased risks of incident type 2 diabetes (T2D), cardiovascular events, and all-cause mortality in general and specific patient populations16. Critically ill patients frequently exhibit profound metabolic derangements, including hyperglycemia, dyslipidemia, and heightened IR17,18. These disturbances are strongly implicated in exacerbating secondary brain injury, immune dysfunction, and poor outcomes in neurological critical care19. However, recent evidence on the prognostic value of the TyG index in critical illness has been conflicting. Several studies have reported significant associations between elevated TyG and mortality in heterogeneous ICU populations, including patients with sepsis, heart failure, and coronary artery disease20–22. In stroke-specific cohorts, Wu et al. found that TyG independently predicted short-term mortality in critically ill stroke patients23. Conversely, other studies have observed attenuated or absent associations among patients with pre-existing diabetes. Xia et al. reported no association between TyG-BMI and cardiovascular risk in patients with T2D24, and Cai et al. demonstrated that the TyG-mortality association was largely confined to non-diabetic individuals with ischemic stroke25. These discrepant findings suggest that the metabolic context—particularly diabetes status—may fundamentally alter the prognostic meaning of the TyG index, yet this interaction has not been systematically evaluated in critically ill cerebrovascular disease patients. Despite the recognized importance of metabolic dysregulation in critical cerebrovascular illness, significant knowledge gaps persist. Crucially, the prognostic utility of the TyG index specifically within the heterogeneous and high-risk population of critically ill cerebrovascular disease patients remains inadequately validated. Furthermore, pre-existing diabetes is highly prevalent in CVD patients and profoundly alters baseline metabolic homeostasis and the response to acute stress17,18. People with diabetes often exhibit exaggerated metabolic disturbances during critical illness, potentially modifying the relationship between TyG index and all-cause mortality risk17,18. Current evidence lacks rigorous investigation into whether and how the presence of diabetes modifies the association between TyG index and mortality in this specific ICU cohort. Understanding this interaction is essential for refining risk prediction models and developing personalized management strategies. Therefore, this study primarily aims to evaluate whether diabetes status modifies the association between the TyG index and mortality — an interaction that has not been formally tested as the primary hypothesis in critically ill cerebrovascular patients. Materials and methods Data source This retrospective cohort investigation employed version 3.1 of the Medical Information Mart for Intensive Care IV database (MIMIC-IV), an open-access repository aggregating ICU admissions at Beth Israel Deaconess Medical Center (Boston, MA) during 2008–2019. All protected health information underwent de-identification. Ethical clearance with waiver of informed consent was granted by institutional review boards at Massachusetts Institute of Technology and Beth Israel Deaconess Medical Center. Researcher certification in human subjects protection was verified through NIH course completion (certificate ID: 12432638). Study conduct observed STROBE reporting standards for observational research and aligned with Declaration of Helsinki ethical principles26. Clinical trial number: not applicable. Study population This investigation enrolled patients experiencing first-time intensive care unit admission for cerebrovascular disorders. Disease diagnoses were verified employing International Classification of Diseases codes (ICD-9/10; Supplementary Table S1). Exclusion criteria included: (1) previous hospitalizations, (2) missing admission glucose or triglyceride measurements, and (3) discharge or death ≤ 24 h after ICU admission. Demographic comparisons between excluded and analyzed cohorts appear in Supplementary Table S2. Demographic and laboratory variables Structured query language facilitated data extraction encompassing: demographics: age, sex, weight, racial background; clinical history: diabetes mellitus, myocardial infarction, congestive heart failure, peripheral vascular disease, chronic pulmonary disease, hypertension, renal impairment, malignancy, severe hepatic disease, Charlson Comorbidity Index, and sepsis status; laboratory parameters: hemoglobin, platelet count, white blood cell (WBC), anion gap, bicarbonate, blood urea nitrogen (BUN), calcium, chloride, creatinine, sodium, potassium, international normalized ratio (INR), alanine aminotransferase (ALT); interventions: invasive mechanical ventilation (IMV), renal replacement therapy (RRT), vasoactive medication administration; severity metrics: Acute Physiology Score III (APS III), Simplified Acute Physiology Score II (SAPS II); physiological variables: heart rate, mean arterial pressure (MBP), respiratory frequency, body temperature, peripheral oxygen saturation (SpO2); outcome measure: survival duration. Diabetes diagnoses were validated using ICD-9/10 codes (Supplementary Table S3). All laboratory parameters represented initial ICU admission values. For variables measured multiple times within the first 24 h, we selected values reflecting peak illness severity: maximum readings for heart rate, respiratory rate, temperature, leukocyte count, anion gap, bicarbonate, blood urea nitrogen, calcium, chloride, creatinine, sodium, potassium, international normalized ratio, and alanine aminotransferase; minimum values for mean arterial pressure, hemoglobin, platelet count, and oxygen saturation. Supplementary Table S4 details missing data patterns. Multiple imputation (m = 3 datasets) was performed using predictive mean matching for continuous variables and logistic regression for binary variables27. The imputation model included all covariates, the survival time, and the event indicator. A complete-case sensitivity analysis (excluding imputed observations) was conducted to assess robustness. This investigation’s principal exposure variable constituted the triglyceride-glucose (TyG) index, an integrated biomarker derived from admission serum triglycerides and blood glucose concentrations. Because strict fasting status (≥ 8 h without caloric intake) cannot be definitively confirmed for all patients in a retrospective ICU database, we used the first glucose and triglyceride measurements recorded within 24 h of ICU admission. The index was computed according to the established equation: TyG = ln[TG(mg/dL) × Glucose(mg/dL)] / 228. The primary outcome measure comprised 30-day all-cause mortality. Secondary outcomes included 90-day and 365-day all-cause mortality. All-cause mortality events were defined as deaths occurring following ICU admission within each respective time interval. Statistical analysis Critically ill cerebrovascular disease patients were stratified by diabetes status. Baseline characteristics are summarized for each cohort. Normality distribution was assessed histographically. Continuous variables are expressed as mean ± standard deviation (SD) or median (interquartile range), while categorical variables appear as frequencies with percentages. Statistical analyses were performed using chi-square tests for categorical variables, Student’s t-tests for normally distributed continuous variables, and Mann–Whitney U tests for non-normally distributed continuous variables. Covariate selection incorporated clinical rationale, published evidence, univariate analysis findings (P < 0.05), or variables altering effect estimates by > 10%29. Multicollinearity was evaluated via variance inflation factors (VIF), with VIF ≥ 4 indicating collinearity. No significant multicollinearity was detected (Supplementary Table S5). Cox proportional hazards (PH) regression modeled relationships between TyG index and mortality endpoints, generating hazard ratios (HRs) with 95% confidence intervals (CIs). Analyses included an unadjusted model and a series of multivariable models that were sequentially adjusted. Model 1 included sex, age, race, and weight. Model 2 built upon Model 1 by adding myocardial infarction, congestive heart failure, peripheral vascular disease, chronic pulmonary disease, hypertension, renal disease, malignant cancer, and severe liver disease. Model 3 further expanded on Model 2 by incorporating heart rate, MBP, respiratory rate, temperature, SpO₂, hemoglobin, platelets, WBC, anion gap, bicarbonate, BUN, calcium, chloride, creatinine, sodium, potassium, INR, and ALT. Model 4 added Charlson comorbidity index, sepsis, IMV, RRT, and vasoactive drug use to the variables in Model 3. Because APS III and SAPS II incorporate glucose and other metabolic parameters that overlap with TyG components, these scores were excluded from the primary Model 4 to avoid overadjustment; a sensitivity analysis additionally adjusting for these scores is provided. Finally, Model 5 supplemented Model 4 by further adjusting for diabetes status. This hierarchical approach ensured comprehensive confounder adjustment while evaluating TyG index-mortality associations. The PH assumption was tested using scaled Schoenfeld residuals. At 365 days, a time-dependent coefficient Cox model was fitted to assess robustness. Detailed results are provided in Supplementary Tables S7 and S8. Prespecified sensitivity analyses included: (1) complete-case analysis excluding imputed data (Supplementary Table S9); (2) stratification by prediabetic status among non-diabetic patients (Supplementary Table S10); (3) stratification by diabetes subtype — type 1 vs. type 2 diabetes (Supplementary Table S11); and (4) additional adjustment for APS III and SAPS II to assess overadjustment (Supplementary Table S12). Diabetes-stratified subgroup analyses were performed, with multiplicative interaction assessed via likelihood ratio testing. To comprehensively evaluate effect modification, additive interaction was additionally assessed using the Relative Excess Risk due to Interaction (RERI), Attributable Proportion (AP), and Synergy Index (SI) with 95% confidence intervals, using TyG index dichotomized at the lower tertile (8.497). Results are presented in Supplementary Table S13. Forest plot visualizations are provided. Statistical analyses employed R software (version 4.2.2; R Foundation) and Free Statistics package (v2.0). Statistical significance threshold was established at P < 0.05. Results Baseline characteristics of study subjects A total of 2,584 patients as depicted in the flow chart (Fig. 1) were included in this analysis. Among these patients, 30-day, 90-day, and 365-day mortality rates were 18.7%, 23.2%, and 28.9%, respectively. Baseline characteristics categorized by diabetes are presented in Table 1. The mean age of the cohort was 70.2 ± 15.1 years, and 50.5% were male. Flow diagram of the screening and enrollment of study participants. Table 1Baseline characteristics and outcomes of participants.VariablesTotal (n = 2584)Without diabetes (n = 1818)With diabetes (n = 766)PTyG index8.8 ± 0.78.7 ± 0.69.2 ± 0.8< 0.001 Sex0.087 Male1306 (50.5)899 (49.4)407 (53.1) Female1278 (49.5)919 (50.6)359 (46.9)Age, years70.2 ± 15.169.9 ± 16.070.7 ± 12.90.260Race White1467 (56.8)1098 (60.4)369 (48.2)< 0.001 Black233 (9.0)120 (6.6)113 (14.8) Asian72 (2.8)41 (2.3)31 (4.0) Others812 (31.4)559 (30.7)253 (33)Weight, kg79.7 ± 22.277.8 ± 21.484.1 ± 23.6< 0.001Vital signs Heart rate, bpm98.9 ± 19.298.1 ± 19.3100.8 ± 18.90.001 MBP, mmHg68.2 ± 15.168.3 ± 14.867.9 ± 15.70.579 Respiratory rate, breaths/min27.4 ± 6.027.3 ± 6.127.7 ± 5.90.229 Temperature, °C37.4 ± 0.637.4 ± 0.637.5 ± 0.70.040 SpO2, %92.3 ± 5.392.3 ± 5.192.2 ± 5.80.649Laboratory parameters Hemoglobin, g/dL11.9 ± 2.112.0 ± 2.111.7 ± 2.20.001 Platelets, ×10⁹/L209.7 ± 77.3208.2 ± 78.2213.1 ± 75.20.140 WBC, ×10⁹/L11.7 ± 7.011.6 ± 7.312.1 ± 6.10.090 Anion gap, mmol/L15.4 ± 4.015.1 ± 3.816.1 ± 4.4< 0.001 Bicarbonate, mmol/L24.0 ± 3.224.0 ± 3.124.1 ± 3.40.605 BUN, mg/dL18.0 (13.0, 25.0)17.0 (13.0, 24.0)20.0 (15.0, 29.0)< 0.001 Calcium, mg/dL8.9 ± 0.68.9 ± 0.69.0 ± 0.70.002 Chloride, mmol/L105.3 ± 5.2105.6 ± 5.1104.6 ± 5.3< 0.001 Creatinine, mg/dL1.0 (0.8, 1.2)0.9 (0.8, 1.2)1.0 (0.8, 1.4)< 0.001 Sodium, mmol/L140.8 ± 4.5140.9 ± 4.5140.6 ± 4.40.101 Potassium, mmol/L4.3 ± 0.84.3 ± 0.84.4 ± 0.8< 0.001 INR1.3 ± 0.41.3 ± 0.51.3 ± 0.40.343 ALT, U/L19.0 (14.0, 29.0)19.0 (14.0, 29.0)20.0 (14.0, 30.0)0.325Comorbidities Myocardial infarction352 (13.6)198 (10.9)154 (20.1)< 0.001 Congestive heart failure487 (18.8)287 (15.8)200 (26.1)< 0.001 Peripheral vascular disease217 (8.4)152 (8.4)65 (8.5)0.917 Chronic pulmonary disease371 (14.4)242 (13.3)129 (16.8)0.019 Hypertension2059 (79.7)1368 (75.2)691 (90.2)< 0.001 Renal disease383 (14.8)186 (10.2)197 (25.7)< 0.001 Malignant cancer163 (6.3)120 (6.6)43 (5.6)0.346 Severe liver disease26 (1.0)18 (1.0)8 (1.0)0.900 Sepsis839 (32.5)566 (31.1)273 (35.6)0.025 Charlson comorbidity index6.9 ± 2.66.3 ± 2.48.2 ± 2.5< 0.001Severity of illness APS III score42.1 ± 21.040.3 ± 20.146.5 ± 22.5< 0.001 SAPS II score32.5 ± 11.831.9 ± 11.633.8 ± 12.0< 0.001Interventions IMV794 (30.7)531 (29.2)263 (34.3)0.010 RRT75 (2.9)35 (1.9)40 (5.2)< 0.001 Vasoactive drug346 (13.4)236 (13.0)110 (14.4)0.347Outcomes 30-day mortality484 (18.7)320 (17.6)164 (21.4)0.023 90-day mortality600 (23.2)396 (21.8)204 (26.6)0.008 365-day mortality748 (28.9)493 (27.1)255 (33.3)0.002Data presented are mean ± SD, or N (%). Baseline characteristics and outcomes of participants. Data presented are mean ± SD, or N (%). People with diabetes exhibited significantly higher values for weight, heart rate, temperature, anion gap, BUN, calcium, creatinine, potassium, Charlson Comorbidity Index, APS III, and SAPS II scores, along with a higher prevalence of myocardial infarct, congestive heart failure, chronic pulmonary disease, hypertension, renal disease, sepsis, IMV, RRT, and higher 30-day, 90-day, and 365-day mortality rates compared to the group without diabetes. Conversely, they were less likely to be White and had lower hemoglobin and chloride levels. Notably, no significant differences were observed between the two groups regarding sex, age, MBP, respiratory rate, SpO₂, platelets, WBC, bicarbonate, sodium, INR, ALT, peripheral vascular disease, malignant cancer, severe liver disease, and the use of vasoactive drugs. Univariate analysis The outcomes of univariate analyses are shown in Supplementary Table S6. Elevated TyG index was significantly associated with increased mortality risk at 30 days (HR 1.42, 95% CI 1.27–1.59), 90 days (HR 1.38, 95% CI 1.25–1.53), and 365 days (HR 1.30, 95% CI 1.19–1.43; all P < 0.001). Other factors significantly associated with mortality are detailed in Supplementary Table S6. Unadjusted and adjusted Cox proportional hazards model As shown in Table 2, unadjusted analyses demonstrated significant associations between elevated TyG index and increased mortality risk at 30 days, 90 days, and 365 days in critically ill cerebrovascular patients (all P < 0.001). These relationships intensified following demographic adjustment (Model 1: HR 1.69, 95%CI 1.51–1.90 for 30-day mortality; HR 1.65, 95%CI 1.49–1.83 for 90-day mortality; HR 1.59, 95%CI 1.44–1.75 for 365-day mortality). Significance persisted after comorbidity adjustment (Model 2: HR 1.65, 95%CI 1.47–1.85 for 30-day mortality; HR 1.60, 95%CI 1.44–1.78 for 90-day mortality; HR 1.52, 95%CI 1.38–1.68 for 365-day mortality) and attenuated though remained significant following laboratory/vital parameter adjustment (Model 3: HR 1.25, 95%CI 1.10–1.43, P = 0.001 for 30-day mortality; HR 1.26, 95%CI 1.12–1.42, P < 0.001 for 90-day mortality; HR 1.21, 95%CI 1.09–1.35, P < 0.001 for 365-day mortality). After further adjustment for Charlson comorbidity index, sepsis, and clinical interventions (Model 4), the associations were substantially attenuated and no longer significant at 30 days (HR 1.14, 95%CI 0.99–1.30, P = 0.064) or 365 days (HR 1.09, 95%CI 0.98–1.22, P = 0.120), while a borderline association persisted at 90 days (HR 1.13, 95%CI 1.00–1.28, P = 0.049). Additional adjustment for diabetes status (Model 5) yielded similar results (30-day: HR 1.15, 95%CI 1.00–1.32, P = 0.055; 90-day: HR 1.14, 95%CI 1.00–1.30, P = 0.043; 365-day: HR 1.11, 95%CI 0.98–1.24, P = 0.095), indicating that the TyG index has limited independent prognostic value in the overall cohort. Table 2Association between TyG index and 30-day, 90-day, and 365-day mortality in critically ill patients with cerebrovascular disease.OutcomeUnadjustedP-valueModel 1P-valueModel 2P-valueModel 3P-valueModel 4P-valueModel 5P-value30-day mortality1.42 (1.27, 1.59)< 0.0011.69 (1.51, 1.90)< 0.0011.65 (1.47, 1.85)< 0.0011.25 (1.10, 1.43)0.0011.14 (0.99, 1.30)0.0641.15 (1.00, 1.32)0.05590-day mortality1.38 (1.25, 1.53)< 0.0011.65 (1.49, 1.83)< 0.0011.60 (1.44, 1.78)< 0.0011.26 (1.12, 1.42)< 0.0011.13 (1.00, 1.28)0.0491.14 (1.00, 1.30)0.043365-day mortality1.30 (1.19, 1.43)< 0.0011.59 (1.44, 1.75)< 0.0011.52 (1.38, 1.68)< 0.0011.21 (1.09, 1.35)< 0.0011.09 (0.98, 1.22)0.1201.11 (0.98, 1.24)0.095Data presented are HRs and 95%CIs.Model 1 was adjusted for sex, age, race, and weight.Model 2 was further adjusted for myocardial infarction, congestive heart failure, peripheral vascular disease, chronic pulmonary disease, hypertension, renal disease, malignant cancer, and severe liver disease.Model 3 was additionally adjusted for heart rate, MBP, respiratory rate, temperature, SpO2, hemoglobin, platelets, WBC, anion gap, bicarbonate, BUN, calcium, chloride, creatinine, sodium, potassium, INR, and ALT.Model 4 was adjusted for Model 3 + Charlson comorbidity index, sepsis, IMV, RRT, and vasoactive drug use.Model 5 was adjusted for Model 4 + diabetes.Abbreviations: ALT, alanine aminotransferase; BUN, blood urea nitrogen; IMV, invasive mechanical ventilation; INR, international normalized ratio; MBP, mean blood pressure; RRT, renal replacement therapy; SpO2, peripheral oxygen saturation; TyG, triglyceride-glucose; WBC, white blood cell. Association between TyG index and 30-day, 90-day, and 365-day mortality in critically ill patients with cerebrovascular disease. Data presented are HRs and 95%CIs. Model 1 was adjusted for sex, age, race, and weight. Model 2 was further adjusted for myocardial infarction, congestive heart failure, peripheral vascular disease, chronic pulmonary disease, hypertension, renal disease, malignant cancer, and severe liver disease. Model 3 was additionally adjusted for heart rate, MBP, respiratory rate, temperature, SpO2, hemoglobin, platelets, WBC, anion gap, bicarbonate, BUN, calcium, chloride, creatinine, sodium, potassium, INR, and ALT. Model 4 was adjusted for Model 3 + Charlson comorbidity index, sepsis, IMV, RRT, and vasoactive drug use. Model 5 was adjusted for Model 4 + diabetes. Abbreviations: ALT, alanine aminotransferase; BUN, blood urea nitrogen; IMV, invasive mechanical ventilation; INR, international normalized ratio; MBP, mean blood pressure; RRT, renal replacement therapy; SpO2, peripheral oxygen saturation; TyG, triglyceride-glucose; WBC, white blood cell. The PH assumption was tested using scaled Schoenfeld residuals. The primary exposure variable, TyG index, satisfied the PH assumption (χ² = 2.611, P = 0.106). Although the global test was significant (P < 0.001), this is attributable to several covariates (e.g., sepsis, electrolyte parameters) and does not affect the validity of the TyG-mortality estimates. Results were similar at 90 days (χ² = 3.084, P = 0.079). Detailed results are provided in Supplementary Table S7. At 365 days, the Schoenfeld test was statistically significant (P = 0.006). However, in a time-dependent coefficient model, the TyG × time interaction term was not clinically meaningful (HR = 0.999, 95%CI 0.998–1.000, P = 0.043), and the TyG effect estimate changed by less than 10% compared with the standard Cox model (Supplementary Table S8), indicating that the PH violation is negligible. Subgroup and sensitivity analyses Diabetes-stratified analyses revealed differential TyG index-mortality associations (Fig. 2). The P values for interaction were 0.033, 0.004, and 0.005 at 30, 90, and 365 days, respectively. Among people without diabetes, significant positive associations emerged for 30-day (HR 1.25, 95%CI 1.04–1.50), 90-day (HR 1.26, 95%CI 1.07–1.49), and 365-day mortality (HR 1.21, 95%CI 1.03–1.40). People with diabetes demonstrated no independent associations. When stratifying TyG index into tertiles (T1 < 8.497; T2 8.497–9.025; T3 ≥ 9.026), an overall positive trend was observed. Subgroup analyses of the association between TyG index and all-cause mortality. A 30-day mortality; B 90-day mortality; C 365-day mortality. Adjusted for sex, age, race, weight, myocardial infarct, congestive heart failure, peripheral vascular disease, chronic pulmonary disease, hypertension, renal disease, malignant cancer, severe liver disease, heart rate, mean blood pressure, respiratory rate, temperature, peripheral oxygen saturation, hemoglobin, platelets, white blood cell, anion gap, bicarbonate, blood urea nitrogen, calcium, chloride, creatinine, sodium, potassium, international normalized ratio, alanine aminotransferase, Charlson comorbidity index, sepsis, invasive mechanical ventilation, renal replacement therapy, and vasoactive drug use. TyG, triglyceride-glucose. TyG index tertiles: T1 (< 8.497), T2 (8.497–9.025), T3 (≥ 9.026). Complete-case sensitivity analyses excluding participants with missing covariates (n = 1,950) yielded no significant TyG-mortality association at any time point (Supplementary Table S9), consistent with the primary findings at 30 and 365 days, though the borderline association at 90 days was attenuated. In this cohort, mortality rates were 19.5%, 24.3%, and 29.8% at 30, 90, and 365 days. Furthermore, in sensitivity analyses of cerebrovascular patients without diabetes (n = 1,818), stratification by prediabetic status revealed differential TyG index-mortality relationships (Supplementary Table S10). Among normoglycemic individuals (n = 1,761), each unit increase in TyG index conferred significantly elevated mortality risk after comprehensive multivariable adjustment: 30 days (HR 1.28, 95% CI 1.04–1.58; P = 0.020), 90 days (HR 1.23, 1.02–1.49; P = 0.034), and 365 days (HR 1.17, 0.98–1.39; P = 0.089). Conversely, in the prediabetic subgroup (n = 57), the limited number of outcome events (≤ 8 deaths per time point) precluded reliable adjusted modeling. Formal interaction testing indicated no statistically significant effect modification by prediabetic status on the association between the TyG index and all-cause mortality at 30, 90, or 365 days. Sensitivity analyses confined to people with diabetes (n = 766) revealed no TyG index-mortality associations across diabetes subtypes at 30, 90, or 365 days (Supplementary Table S11). Among type 1 diabetes (T1D) patients (n = 10), sparse outcome events (≤ 4 deaths per timepoint) precluded adjusted modeling due to limited statistical power. In T2D patients (n = 433), unadjusted analyses demonstrated significantly elevated all-cause mortality risk with high TyG index at 30-day (HR 1.37, 95%CI 1.06–1.77; P = 0.017) and 365-day (HR 1.27, 1.02–1.57; P = 0.030) endpoints, with borderline significance at 90 days (HR 1.25, 0.99–1.59; P = 0.059). However, full covariate adjustment attenuated these associations to non-significance (HR 1.16, 95%CI 0.80–1.70, P = 0.434 for 30-day mortality; HR 1.10, 95%CI 0.79–1.52, P = 0.586 for 90-day mortality; HR 1.23, 95%CI 0.92–1.65, P = 0.164 for 365-day mortality). No significant interaction by diabetes subtype emerged at any timepoint. In sensitivity analyses further adjusting for APS III and SAPS II (Supplementary Table S12), the TyG-mortality association remained consistent with the primary analysis, confirming that overadjustment by these severity scores did not drive the primary findings. Additive interaction analyses using RERI, AP, and SI were performed with TyG dichotomized at the lower tertile (8.497). A significant antagonistic additive interaction was observed across all time points (AP range: -0.66 to -0.53; S range: 0.19–0.21), consistent with the multiplicative interaction from the primary continuous analysis (Supplementary Table S13). These findings support that the TyG-mortality association is significantly modified by diabetes status. Discussion This investigation of 2,584 critically ill cerebrovascular patients demonstrates that diabetes status critically modifies the TyG index–mortality relationship. In the overall cohort, the TyG index showed no significant association with 30-day (HR 1.15, 95%CI 1.00–1.32) or 365-day mortality (HR 1.11, 95%CI 0.98–1.24), with only a borderline association at 90 days (HR 1.14, 95%CI 1.00–1.30). However, significant positive associations were confined to individuals without diabetes (30-day: HR 1.25, 95%CI 1.04–1.50; 90-day: HR 1.26, 95%CI 1.07–1.49; 365-day: HR 1.21, 95%CI 1.03–1.40), whereas no association was observed in diabetic patients (all P > 0.05). These findings establish diabetes status as a key effect modifier in TyG index–based mortality prediction in critical cerebrovascular disease. The TyG index, a surrogate marker of insulin resistance, has emerged as a potential predictor of adverse outcomes in cardiovascular and cerebrovascular diseases. However, recent studies report conflicting associations between the TyG index and all-cause mortality risk, particularly in critically ill patients with cerebrovascular disease. While some studies suggest a strong positive correlation between elevated TyG levels and increased all-cause mortality risk, others indicate no significant association or even protective effects in specific subpopulations, especially people with diabetes. For instance, a study by Xia et al. (2024) found no association between the TyG-BMI index and the risk of heart disease in patients with T2D24. Similarly, research by Sun et al. (2025) indicated that the C-reactive protein-triglyceride glucose index (CTI) was not significantly associated with CVD mortality among people with diabetes30. Cai et al. also found that the association between the TyG index and all-cause mortality was particularly pronounced in patients without diabetes among those critically ill with ischemic stroke25. These findings are consistent with our conclusions. However, studies by Xiao et al. (2025) and Huang et al. (2024) reported that TyG levels served as an independent predictor for aneurysmal subarachnoid hemorrhage and hemorrhagic stroke31,32. This discrepancy may reflect differences in study populations, sample size, and the breadth of confounder adjustment. Meanwhile, Wu et al. (2025) found that an elevated TyG index was independently associated with higher mortality among critically ill stroke patients, supporting its utility for risk stratification23. It is worth noting that their study focused on short-term mortality during hospitalization and ICU stay, which differs from our outcome measures at 30, 90, and 365 days. This suggests that the TyG index may be more reflective of acute metabolic stress impacting short-term outcomes, whereas our long-term findings highlight the dominant role of diabetes status itself and other chronic factors. Our findings suggest that diabetes status serves as a critical effect modifier in this relationship, consistent with prior observations in cardiovascular cohorts. Specifically, significant TyG index-mortality associations persist in individuals without diabetes but become substantially attenuated in patients with diabetes. To our knowledge, few studies have formally evaluated this interaction using both multiplicative and additive scales in critically ill cerebrovascular patients, and our findings add to the growing evidence that diabetes status critically modifies the TyG–mortality association. Its clinical significance lies in establishing individuals without diabetes as a high-risk subgroup warranting metabolic-targeted interventions, while methodological rigor was ensured through comprehensive sensitivity analyses confirming robustness. Pathophysiological divergence underlies the diabetes-dependent TyG-mortality relationship. Among people without diabetes, acute insulin resistance may provoke cerebral energy crises through impaired glucose utilization23,33, while unmasked dysmetabolism and mitochondrial vulnerability34 amplify metabolic risk. This pattern is consistent with observations in ischemic stroke, where stress hyperglycemia has been associated with worse outcomes predominantly in non-diabetic patients35. Conversely, people with diabetes exhibit metabolic preconditioning36 that attenuates TyG effects through chronic adaptive changes, with competing mortality drivers and therapeutic interventions37 potentially mitigating TyG-related risk. Our findings suggest three practical implications. First, in non‑diabetic patients, an admission TyG ≥ 9.026 (upper tertile) may identify those at higher metabolic risk who could benefit from enhanced glycemic monitoring and early management of stress hyperglycemia. In diabetic patients, TyG should not be used as a primary prognostic indicator. Second, TyG is easily calculated from routine admission glucose and triglycerides, without requiring additional laboratory testing, and could be integrated into electronic health records to generate real‑time risk alerts. Third, our results support diabetes‑stratified management: non‑diabetic patients with elevated TyG need closer attention to acute insulin resistance, while diabetic patients require a holistic approach addressing multifactorial risks. These recommendations are hypothesis‑generating and require prospective validation. This study has several limitations. First, residual confounding from unmeasured variables (such as prestroke glycemic control) may constrain result validity. Second, the single-center design necessitates external validation across diverse populations to ensure generalizability. Third, our analysis of type 1 diabetes patients was severely underpowered due to the very small sample size (n = 10), precluding any reliable conclusions for this subgroup. Fourth, reliance on a single TyG index measurement may fail to capture dynamic metabolic changes during critical illness evolution. Additionally, because strict fasting status (≥ 8 h) cannot be definitively confirmed, the glucose values used for TyG calculation represent admission rather than true fasting glucose. However, admission glucose is the clinically relevant value in acute ICU settings and reflects the metabolic state at the time of critical illness. Conclusion Diabetes status modulates the TyG index–mortality relationship in cerebrovascular critical care. The TyG index may be useful as a biomarker of metabolic vulnerability primarily among people without diabetes, whereas its prognostic value is attenuated in patients with established diabetes. This pathophysiology-specific interaction warrants further investigation and suggests that diabetes-stratified approaches to metabolic monitoring may be appropriate in neurocritical settings. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Acknowledgements Gratitude to Jie Liu of the PLA General Hospital for his invaluable guidance on our manuscript. Abbreviations Alanine aminotransferase Acute physiology score III Blood urea nitrogen Invasive mechanical ventilation International normalized ratio Mean blood pressure Renal replacement therapy Simplified Acute Physiology Score II Standard deviation Peripheral oxygen saturation Triglyceride-glucose White blood cell Author contributions HW designed the study, conducted data collection and analysis, and drafted the manuscript. XD, XC, and JL performed data analysis and supervision. All authors contributed to the article, reviewed the final manuscript, and approved the submitted version. Data availability The data supporting this study’s findings originate exclusively from the MIMIC-IV database, a publicly accessible repository. No original datasets were created or analyzed during this investigation. Declarations Competing interests The authors declare no competing interests. Ethics approval and consent to participate This investigation adhered strictly to ethical principles outlined in the Declaration of Helsinki. Authorization for MIMIC-IV database utilization was secured through institutional review boards at Massachusetts Institute of Technology and Beth Israel Deaconess Medical Center. The pre-existing ethical clearance governing MIMIC data application encompasses this analysis, thereby precluding additional ethics review or patient consent requirements. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally to this work: Hui Wang and Xiaojun Ding. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement The data supporting this study’s findings originate exclusively from the MIMIC-IV database, a publicly accessible repository. No original datasets were created or analyzed during this investigation.