Association of Stress Hyperglycemia Ratio with Sepsis and Mortality in Critically Ill Stroke Patients: A Retrospective Cohort Study from MIMIC-IV

Abstract

Background

Stress-induced hyperglycemia (SIH) has been associated with poor outcomes in stroke patients. However, the relationship between SIH and sepsis in this population remains understudied. We aimed to evaluate the association of SIH, measured using the stress hyperglycemia ratio (SHR), with the development of sepsis and mortality among critically ill stroke patients.

Methods

We retrospectively analyzed stroke patients requiring ICU admission from the MIMIC-IV database. Primary outcome was sepsis, and secondary outcomes were 30-day and 90-day all-cause mortality. Multivariable Cox and logistic regression models were used to evaluate associations.

Results

A total of 3018 patients were included (66.8% ischemic stroke). After full adjustment for confounders, SHR was independently associated with an increased risk of sepsis (Q4 vs Q1: OR 1.46, 95% CI: 1.12-1.89, P = 0.005; continuous SHR: OR 1.31, P = 0.010). SHR also demonstrated a strong dose-response relationship with mortality; patients in Q4 had significantly higher risks of 30-day (OR 2.95, 95% CI: 2.25-3.88, P < 0.001) and 90-day mortality (OR 2.25, 95% CI: 1.80-2.82, P < 0.001). Subgroup analyses revealed significant interactions between SHR and stroke type for sepsis (P for interaction = 0.014), with a more pronounced effect observed in ischemic stroke patients. The associations between SHR and both sepsis and mortality were consistently maintained regardless of the presence of diabetes (all P < 0.050).

Conclusion

Elevated stress hyperglycemia ratio is independently associated with higher risks of sepsis and short-to long-term mortality among critically ill patients with stroke, with consistent associations observed irrespective of diabetes status. In contrast, no statistically significant association between SHR and sepsis was identified in the hemorrhagic stroke subgroup.

Introduction

Stroke accounts for over 12.5 million new cases and more than 6.5 million deaths annually, ¹ which also accounts for more than 140 million disability-adjusted life years lost each year. ² A substantial proportion of patients with severe stroke require intensive care; for instance, in a US Medicare cohort, 19.9% of acute ischemic stroke (AIS) admissions received ICU-level care. ³ However, the prognosis for these critically ill patients remains suboptimal, with some retrospective cohorts reporting 6-month mortality rates as high as 64.5% and a functional independence rate of only 19.5%.^3,4 This poor prognosis is primarily driven by frequent and severe post-stroke complications, particularly infection and sepsis. In contemporary neurological ICU settings, where two-thirds of admissions are due to AIS or hemorrhagic stroke (HS), up to 63% of patients develop infections and sepsis.^4,5 Notably, sepsis is present in the majority of death cases, independently associated with markedly worse 3-month functional outcomes. ⁶ Accordingly, early risk stratification and the optimization of modifiable physiological targets are central to guideline-based neurocritical care to improve patient outcomes.^7,8

Metabolic derangement, particularly dysglycemia, is a hallmark of the acute stress response in stroke, affecting approximately two-thirds of patients. ⁹ Dysglycemia in stroke patients may stem from diagnosed diabetes and the activation of the hypothalamic-pituitary-adrenal axis and subsequent insulin resistance under acute stress conditions, also known as stress-induced hyperglycemia (SIH).^10,11 To account for the influence of a patient’s baseline glycemia and chronic diabetes, the stress hyperglycemia ratio (SHR) has been proposed as a more refined metric to quantify stress-induced metabolic derangement. By normalizing admission glucose to estimated average glucose (derived from HbA1c), SHR provides a more accurate reflection of SIH that is independent of background glycemia.^12,13 Extensive evidence has demonstrated that higher SHR is significantly associated with increased short- and long-term mortality, as well as poor neurological recovery in stroke patients. Beyond these outcomes, emerging research has explored the impact of SHR on infectious complications following a stroke.^14-22 For instance, a large-scale 2025 study of 865,765 patients from the Chinese Stroke Center Alliance cohort found that SHR is significantly associated with in-hospital medical complications, particularly pneumonia and urinary tract infections. ²³

However, several critical gaps remain in the current understanding of the relationship between SIH and infectious complications. First, while extensive research has linked SIH to pneumonia, few studies have systematically investigated its association with sepsis—a more severe systemic manifestation of clinical deterioration. SIH may be linked to sepsis through hyperglycemia-mediated immune impairment, barrier dysfunction, and invasive devices. Second, most prior explorations into SIH and infection have relied on absolute glucose levels measured at a single time point, failing to account for the relative intensity of stress represented by SHR. Thirdly, the predictive value of SIH in patients with pre-existing diabetes remains a subject of ongoing debate, with inconsistent evidence regarding whether chronic glycemia modulates the impact of acute stress. To address these gaps, this study aims to evaluate the association of SHR with the development of sepsis and mortality among critically ill stroke patients.

Method

Data Source

The data utilized in this study were obtained from the MIMIC-IV (v3.1) database, a comprehensive, publicly accessible resource developed and maintained by the MIT Computational Physiology Laboratory (https://physionet.org/content/mimiciv/3.1/). This database includes detailed information on all patients admitted to the Beth Israel Deaconess Medical Center between 2008 and 2019. It records a wide range of patient data, including length of stay, laboratory results, medication treatments, vital signs, and other clinical information. To ensure patient privacy, all personal identifiers have been de-identified, and random codes are used to replace direct patient identifiers. Consequently, informed consent and ethical approval for patient participation are not required. Data for this study were extracted by author Shuai Yuan, who successfully completed the online training course offered by the National Institutes of Health (authorization code: 67811155). The data extraction process was carried out using PostgreSQL tools, version 1.12.3.

Population Selection Criteria

Hospital admission information for stroke patients was extracted based on the diagnostic codes for AIS, non-traumatic subarachnoid hemorrhage, and intracerebral hemorrhage according to the 9th and 10th revisions of the International Classification of Diseases (ICD-9 and ICD-10) codes (Supplemental Table 1). The exclusion criteria were as follows: (1) age <18 years; (2) not a first-time ICU visit; (3) ICU stay of less than 24 hours; (4) missing HbA1c or glucose records. Ultimately, a total of 3018 stroke patients admitted to the ICU were included in the study (Figure 1).

Figure 1.

Flow chart of the study

Data Extraction

The clinical data of the patients included demographic information (age, gender), medical history (myocardial infarction, congestive heart failure, chronic pulmonary disease, diabetes, renal disease, severe liver disease), vital parameters (temperature, mean blood pressure (MBP), respiratory rate, blood oxygen saturation (SpO2), clinical features (stroke type, Glasgow Coma Scale (GCS), Sequential Organ Failure Assessment (SOFA), Acute Physiology Score III (APACHE III)), and laboratory variables (white blood cell count (WBC), red blood cell distribution width (RDW), anion gap, bicarbonate, blood urea nitrogen (BUN), creatinine, sodium, potassium). Vital signs, laboratory values, and clinical scores were based on the average values within the first 24 hours of ICU admission. For missing laboratory data, forward imputation was performed using a 24-hour backward search window. Regarding GCS, since the database automatically assigns a GCS score of 15 for intubated patients, we extracted the first GCS score recorded after extubation for patients who were intubated upon admission. For HbA1c, the data correspond to the last measurement within 1 week prior to ICU admission, while glucose levels reflect the first measurement obtained after ICU admission. The SHR was calculated using the following formula: SHR = (admission blood glucose (mg/dl))/(28.7 × HbA1c (%) − 46.7). To minimize the underdiagnosis of pre-existing diabetes, we employed a comprehensive diagnostic strategy. Patients were identified as having diabetes if they met either of the following criteria: (1) a documented medical history of diabetes in the electronic health records (comorbidities), or (2) an admission HbA1c level ≥6.5%. The data extraction code is publicly available on GitHub (https://github.com/MIT-LCP/mimic-iv).

Outcomes and Follow-Up

The primary outcome of this study was sepsis, while the secondary outcomes included 30-day and 90-day all-cause mortality. In this study, sepsis was defined using 2 criteria: 1. A suspected infection, indicated by the ordering of microbiological cultures within a window of 72 hours before and 24 hours after the initiation of antibiotic therapy. 2. An increase in the SOFA score of ≥2 points, consistent with the Sepsis-3 criteria. Mortality information was obtained from the admissions and patients tables in the MIMIC-IV database. To enhance completeness of death ascertainment beyond the index hospitalization, MIMIC-IV links hospital electronic health record data with the Massachusetts Registry of Vital Records and Statistics, allowing capture of both in-hospital and out-of-hospital deaths for at least 365 days after discharge. Accordingly, 30-day mortality and 90-day mortality were operationally defined as any recorded death occurring within 30 days or 90 days, respectively, from the index ICU admission date. Patients without a recorded date of death (ie, null death date fields) were treated as alive throughout the available follow-up and were not classified as deceased at the specified time points. Patients transferred to another facility were accounted for using recorded discharge disposition; subsequent mortality was determined through the same linked mortality fields, and individuals without a death record were considered alive at 30 and 90 days.

Statistical Analysis

Variables with a normal distribution were expressed as mean ± standard deviation and compared using analysis of variance. For non-normally distributed variables, values were presented as median with interquartile range (IQR), and comparisons were made using the Mann–Whitney U test or Kruskal–Wallis test, as appropriate. Categorical variables were presented as counts and percentages, and comparisons were made using the chi-square (χ²) test or Fisher’s exact test when applicable. SHR was analyzed as both a continuous variable and a categorical variable (stratified into quartiles, Q1–Q4) to ensure the robustness of our findings. When treated as a continuous variable, we assessed the odds ratio (OR) or hazard ratio (HR) with corresponding 95% confidence intervals (CIs) per unit increase in SHR. When treated as a categorical variable, the first quartile (Q1) was designated as the reference group to evaluate the dose-response relationship across increasing SHR levels. To evaluate the independent association between the SHR and clinical outcomes, we constructed 3 multivariable regression models using a sequential adjustment strategy. Model 1 was a preliminary model adjusted for basic demographic factors, including age and gender. Model 2 further incorporated pre-existing comorbidities to account for the patients’ baseline health status, including myocardial infarction, congestive heart failure, chronic pulmonary disease, diabetes, renal disease, and severe liver disease. Model 3 was the fully adjusted model, which integrated Model 2 variables along with acute clinical factors and laboratory markers to minimize residual confounding. These included stroke type, neurological and organ dysfunction severity (GCS and SOFA scores), and admission laboratory values (hemoglobin, white blood cell count, blood urea nitrogen, sodium, and potassium).^15-18 Kaplan–Meier survival analysis was used to evaluate the association between SHR and overall mortality, with statistical significance assessed using the log-rank test. Survival curves were generated using Kaplan–Meier analysis and compared with the log-rank test. The proportional hazards assumption for the Cox regression models was assessed by Schoenfeld residuals test to ensure model validity. And a non-significant result (P > 0.050) in the Schoenfeld test was considered indicative of compliance with the proportional hazards assumption. Receiver operating characteristic (ROC) curve analysis was used to assess the predictive performance of SHR, with the area under the (AUC) indicating discriminative ability. All statistical tests were two-tailed, and a P-value <0.050 was considered statistically significant. Statistical analyses were conducted using R software (version 4.4.1). Missing data were addressed using Multiple Imputation by Chained Equations, implemented with m = 20. This method was applied to variables with <10% missingness, including laboratory and vital sign data. Variables with >20% missingness were excluded from multivariable models. Variance Inflation Factor (VIF) was conducted for each covariate in the models and a VIF value of less than 5 was predefined as the threshold for acceptable levels of collinearity.

Result

Patient Characteristics and Outcomes

The baseline characteristics and clinical outcomes of the 3018 stroke patients included in the study are summarized in Table 1. The proportion of missing data for baseline variables was generally low (<1%) and is summarized in Supplemental Table 2. Key variables such as SHR and mortality outcomes had complete data. The median age of the cohort was 71.77 years (IQR: 60.24-82.33), with 1543 males (50.9%). Ischemic stroke was the predominant stroke type, accounting for 2015 (66.8%) of the total population. Initially, 946 patients were identified based solely on documented comorbidities. After incorporating the HbA1c criterion (≥6.5%), an additional 59 patients were identified, bringing the total number of patients with diabetes to 1005 (33.3% of the total cohort). The median SHR was 1.05 (IQR: 0.89-1.27). Patients were stratified into quartiles based on the SHR at ICU admission, with the following cutoff values: Q1 = 0.18-0.89, Q2 = 0.89-1.05, Q3 = 1.05-1.27, and Q4 = 1.27-13.9. The corresponding median SHR values were 0.80 (Q1), 0.97 (Q2), 1.15 (Q3), and 1.47 (Q4). Higher SHR quartiles were associated with lower GCS scores, MBP, and bicarbonate levels. Conversely, patients in the higher SHR groups showed significantly higher SOFA and APS III scores, body temperature, SpO₂, WBC, anion gap, creatinine, and BUN levels. There were no significant differences across SHR quartiles in terms of gender, RDW, sodium, potassium, or medical history of myocardial infarction, congestive heart failure, chronic pulmonary disease, renal disease, or severe liver disease. Additionally, patients with higher SHR were more likely to experience adverse outcomes, including increased 30-day and 90-day mortality, as well as higher incidence rates of sepsis.

Table 1.

Baseline Characteristics According to SHR Quartiles

	Overall (n = 3018)	Q1	Q2	Q3	Q4	P
		0.18-0.89	0.89-1.05	1.05-1.27	1.27-13.9
		(n = 756)	(n = 754)	(n = 755)	(n = 753)
Age, years	71.77 [60.24, 82.33]	72.09 [61.09, 82.49]	72.71 [60.24, 82.88]	71.44 [59.99, 81.82]	70.97 [59.96, 81.49]	0.257
Gender, male	1543 (50.9)	386 (51.1)	397 (52.7)	389 (51.5)	363 (48.2)	0.357
Comorbidities
Myocardial infarct	378 (12.5)	96 (12.7)	83 (11.0)	89 (11.8)	110 (14.5)	0.180
Congestive heart failure	652 (21.5)	169 (22.3)	162 (21.4)	136 (18.0)	185 (24.4)	0.021
Chronic pulmonary disease	444 (14.7)	124 (16.4)	99 (13.1)	108 (14.3)	113 (14.9)	0.334
Diabetes	1005 (33.3)	290 (38.4)	200 (26.5)	226 (29.9)	289 (38.4)	<0.001
Renal disease	469 (15.5)	135 (17.8)	113 (14.9)	93 (12.3)	128 (16.9)	0.015
Severe liver disease	34 (1.1)	6 (0.8)	4 (0.5)	7 (0.9)	17 (2.2)	0.007
Vital signs
MBP, mmHg	87.20 [79.76, 95.71]	88.22 [80.54, 97.00]	88.21 [80.62, 97.61]	87.06 [79.04, 95.46]	85.25 [78.12, 92.93]	<0.001
Temperature, °C	36.90 [36.73, 37.14]	36.83 [36.70, 37.00]	36.92 [36.74, 37.13]	36.92 [36.74, 37.19]	36.95 [36.76, 37.24]	<0.001
SPO2, %	96.96 [95.68, 98.29]	96.75 [95.44, 97.95]	96.77 [95.54, 98.00]	97.00 [95.79, 98.36]	97.44 [95.92, 98.85]	<0.001
Clinical features
SOFA	2.00 [1.00, 3.38]	1.62 [0.88, 2.75]	1.79 [0.92, 2.88]	2.00 [1.00, 3.46]	2.79 [1.56, 4.25]	<0.001
APSIII	35.00 [27.00, 46.00]	32.00 [24.00, 41.00]	34.00 [25.00, 43.00]	35.00 [27.00, 46.00]	41.00 [31.00, 52.00]	<0.001
GCS	14.00 [11.00, 14.00]	14.00 [13.00, 14.00]	14.00 [12.00, 14.00]	14.00 [11.00, 14.00]	13.00 [10.00, 14.00]	<0.001
Hemorrhagic stroke	1003 (33.2)	180 (23.8)	238 (31.6)	280 (37.1)	305 (40.5)	<0.001
Laboratory tests
Hemoglobin, g/dL	12.20 [10.73, 13.60]	12.30 [10.80, 13.60]	12.40 [11.10, 13.65]	12.40 [10.90, 13.70]	11.90 [10.30, 13.27]	<0.001
WBC, 109/L	9.85 [7.70, 12.60]	8.52 [6.98, 10.70]	9.30 [7.50, 11.80]	10.37 [8.50, 12.90]	11.70 [9.00, 14.70]	<0.001
RDW, %	13.70 [13.05, 14.70]	13.80 [13.10, 14.80]	13.70 [13.06, 14.50]	13.70 [13.00, 14.60]	13.80 [13.10, 14.97]	0.039
Anion gap, mmol/L	14.00 [12.00, 15.67]	13.00 [11.50, 15.00]	14.00 [12.00, 15.33]	14.00 [12.00, 16.00]	14.00 [12.50, 16.00]	<0.001
Bicarbonate, mmol/L	23.00 [21.00, 25.00]	23.50 [22.00, 25.00]	23.50 [22.00, 25.00]	23.00 [21.00, 25.00]	22.00 [20.00, 24.00]	<0.001
Bun, mg/dL	16.00 [12.00, 22.33]	16.00 [12.00, 22.00]	15.50 [12.00, 21.00]	16.00 [12.00, 21.00]	17.50 [13.00, 25.50]	<0.001
Creatinine, mg/dL	0.90 [0.70, 1.13]	0.90 [0.70, 1.15]	0.90 [0.70, 1.10]	0.87 [0.70, 1.10]	0.95 [0.75, 1.25]	<0.001
Sodium, mmol/L	139.50 [137.00, 142.00]	140.00 [138.00, 142.00]	139.50 [137.50, 142.00]	139.50 [137.00, 142.00]	139.00 [137.00, 142.00]	0.237
Potassium, mmol/L	4.00 [3.70, 4.30]	4.00 [3.70, 4.35]	4.00 [3.70, 4.30]	4.00 [3.70, 4.30]	4.03 [3.75, 4.35]	0.125
Outcomes
30-day mortality	578 (19.1)	72 (9.5)	116 (15.3)	147 (19.4)	243 (32.1)	<0.001
90-day mortality	726 (24.0)	115 (15.2)	146 (19.3)	182 (24.0)	283 (37.4)	<0.001
Sepsis	1018 (33.6)	171 (22.6)	211 (27.9)	281 (37.1)	355 (46.9)	<0.001

SHR, stress hyperglycemia ratio; MBP, mean blood pressure; SP02, blood oxygen saturation; GCS, Glasgow Coma Scale; SOFA, Sequential Organ Failure Assessment; APS III, Acute Physiology Score III; WBC, white blood cells; BUN, blood urea nitrogen; RDW, Red Blood Cell Distribution Width.

Association Between SHR and Risk of Sepsis

The association between SHR and the incidence of sepsis is summarized in Table 2. In the unadjusted model (Model 1), higher SHR was significantly associated with an increased risk of sepsis, whether analyzed as a categorical or continuous variable. Compared to the lowest quartile (Q1), the ORs for Q2, Q3, and Q4 were 1.30 (95% CI: 1.05-1.61, P = 0.016), 1.93 (95% CI: 1.56-2.38, P < 0.001), and 2.75 (95% CI: 2.23-3.39, P < 0.001), respectively.

Table 2.

Logistic and Cox Regression Analyses of SHR and Clinical Outcomes

Variable	Model 1 OR or HR (95% CI)	P-value	Model 2 OR or HR (95% CI)	P-value	Model 3 OR or HR (95% CI)	P-value
Sepsis
Q1	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Q2	1.30 (1.05-1.61)	0.016	1.36 (1.09-1.69)	0.006	1.20 (0.93-1.56)	0.168
Q3	1.93 (1.56-2.38)	<0.001	2.05 (1.66-2.54)	<0.001	1.53 (1.19-1.98)	0.001
Q4	2.75 (2.23-3.39)	<0.001	2.74 (2.21-3.40)	<0.001	1.46 (1.12-1.89)	0.005
SHR (continuous)	2.90 (2.34-3.61)	<0.001	2.78 (2.23-3.46)	<0.001	1.31 (1.07-1.61)	0.010
30-day mortality
Q1	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Q2	1.69 (1.25-2.27)	<0.001	1.68 (1.25-2.25)	<0.001	1.64 (1.21-2.21)	0.001
Q3	2.29 (1.73-3.04)	<0.001	2.31 (1.74-3.07)	<0.001	2.11 (1.58-2.81)	<0.001
Q4	4.10 (3.15-5.34)	<0.001	4.03 (3.09-5.25)	<0.001	2.95 (2.25-3.88)	<0.001
SHR (continuous)	1.32 (1.25-1.39)	<0.001	1.33 (1.26-1.40)	<0.001	1.27 (1.19-1.36)	<0.001
90-day mortality
Q1	1.00 (Reference)		1.00 (Reference)		1.00 (Reference)
Q2	1.34 (1.05-1.72)	0.019	1.33 (1.04-1.71)	0.021	1.28 (1.00-1.65)	0.050
Q3	1.82 (1.44-2.30)	<0.001	1.85 (1.46-2.34)	<0.001	1.67 (1.31-2.11)	<0.001
Q4	3.11 (2.50-3.86)	<0.001	3.05 (2.45-3.79)	<0.001	2.25 (1.80-2.82)	<0.001
SHR (continuous)	1.31 (1.25-1.38)	<0.001	1.32 (1.26-1.39)	<0.001	1.25 (1.18-1.33)	<0.001

Note: SHR was analyzed both as a continuous variable and as a categorical variable (quartiles: Q1–Q4). Model 1: Adjusted for age and gender. Model 2: Adjusted for Model 1 variables plus comorbidities (myocardial infarction, congestive heart failure, chronic pulmonary disease, diabetes, renal disease, and severe liver disease). Model 3: Adjusted for Model 2 variables plus stroke type, clinical severity scores and laboratory parameters. Abbreviations: SHR, stress hyperglycemia ratio; CI, confidence interval; OR, odds ratio; HR, hazard ratio.

After adjusting for core confounders including age, gender, and stroke severity (Model 2), the association remained robust. In the fully adjusted model (Model 3), which further accounted for laboratory parameters and clinical scores, SHR remained an independent risk factor for sepsis. Specifically, patients in Q3 (OR: 1.53, 95% CI: 1.19-1.98, P = 0.001) and Q4 (OR: 1.46, 95% CI: 1.12-1.89, P = 0.005) exhibited significantly higher risks compared to those in Q1. When analyzed as a continuous variable, each unit increase in SHR was associated with a 31% increase in the risk of sepsis (adjusted OR: 1.31, 95% CI: 1.07-1.61, P = 0.010).

Association Between SHR and 30-Day and 90-Day Mortality

The relationship between SHR and all-cause mortality is presented in Table 2 and Figure 2. For 30-day mortality, the unadjusted analysis (Model 1) showed a strong dose-response relationship, with the highest quartile (Q4) carrying more than a four-fold risk compared to Q1 (OR: 4.10, 95% CI: 3.15-5.34, P < 0.001). This association remained highly significant in the fully adjusted model (Model 3). In Model 3, compared to Q1, the adjusted HRs for 30-day mortality were 1.64 (95% CI: 1.21-2.21, P = 0.001) for Q2, 2.11 (95% CI: 1.58-2.81, P < 0.001) for Q3, and 2.95 (95% CI: 2.25-3.88, P < 0.001) for Q4. Continuous SHR analysis further confirmed this independent association (adjusted HR: 1.27, 95% CI: 1.19-1.36, P < 0.001). Similarly, for 90-day mortality, elevated SHR was independently associated with worse outcomes. In the fully adjusted model (Model 3), patients in the higher quartiles demonstrated a significantly increased risk of death (Q3 vs Q1: HR: 1.67, 95% CI: 1.31-2.11, P < 0.001; Q4 vs Q1: HR: 2.25, 95% CI: 1.80-2.82, P < 0.001). While the association for Q2 was borderline significant (HR: 1.28, 95% CI: 1.00-1.65, P = 0.050), the overall trend across quartiles and the continuous SHR analysis (adjusted HR: 1.25, 95% CI: 1.18-1.33, P < 0.001) consistently supported SHR as a significant predictor of short- and long-term mortality in critically ill stroke patients. Kaplan–Meier survival curves further demonstrated that higher SHR quartiles were associated with significantly worse survival outcomes over both 30 and 90 days (Figure 2). Furthermore, the Schoenfeld residuals test confirmed the validity of the Cox models for both 30-day mortality (global P = 0.428) and 90-day mortality (global P = 0.356), indicating no significant violation of the proportional hazards assumption.

Figure 2.

Kaplan–Meier survival curves for 30-day and 90-day mortality stratified by stress hyperglycemia ratio quartiles

Predictive Performance of SHR for Sepsis and Mortality

ROC curve analyses were conducted to assess the predictive performance of the SHR for sepsis, 30-day mortality, and 90-day mortality (Figure 3). We compared the baseline multivariable model (Model 3, including age, gender, comorbidities, and clinical severity scores) with the enhanced model that integrated SHR. As illustrated in Figure 3, the addition of SHR to the multivariable baseline model resulted in a modest but consistent improvement in the Area Under the Curve (AUC) for predicting sepsis, 30-day mortality, and 90-day mortality. Specifically, the predictive performance of SHR as a biomarker was evaluated across all 3 outcomes. For Sepsis, the SHR demonstrated a discriminatory power with an AUC of 0.625. According to the Youden index, the optimal cutoff value for identifying patients at high risk of sepsis was 1.0635. For 30-day Mortality: the SHR showed the highest predictive accuracy for short-term mortality, with an AUC of 0.662. The optimal threshold was determined to be 1.1239. For 90-day Mortality: the AUC for predicting long-term mortality was 0.637, with an optimal cutoff value of 1.1226. These results indicate that SHR serves as a valuable supplementary biomarker to conventional clinical risk factors, providing a more refined risk stratification for critically ill stroke patients.

Figure 3.

Receiver operating characteristic curves illustrating the predictive performance of SHR for sepsis, 30-day mortality, and 90-day mortality. AUC, sensitivity, and specificity are presented for each outcome

Subgroup Analysis of the Association Between SHR and Sepsis

To evaluate the robustness of the association between the SHR and clinical outcomes, we performed stratified analyses across several predefined subgroups, including age, gender, stroke type, and history of diabetes. These analyses were conducted using the fully adjusted Model 3, with SHR treated as a continuous variable. The results are summarized in Figure 4. For the risk of sepsis, the predictive value of SHR was significantly modulated by several clinical factors. We observed significant interactions between SHR and age group, GCS score, and stroke type. Specifically, elevated SHR was strongly associated with an increased risk of sepsis in patients aged <65 years (OR: 2.60, 95% CI: 1.56-4.34, P < 0.001), those with ischemic stroke (OR: 1.71, 95% CI: 1.26-2.33, P < 0.001), and those with GCS scores >8 (OR: 1.46, 95% CI: 1.14-1.87, P = 0.003). However, these associations were not statistically significant in patients aged ≥65 years, those with hemorrhagic stroke, or those with GCS scores ≤8 (all P > 0.050). No significant interaction was found for gender, diabetes status, or SOFA score.

Figure 4.

Subgroup analyses of the association between SHR and clinical outcomes. (A) Sepsis, (B) 30-day mortality, and (C) 90-day mortality. Forest plots illustrating the odds ratios (OR) or hazard ratios (HR) with 95% confidence intervals (CIs) for SHR across predefined subgroups. SHR was analyzed as a continuous variable. All models were adjusted for potential confounders (model3). P for interaction was calculated to assess the consistency of the SHR effect across different clinical strata

Regarding mortality, SHR demonstrated a remarkably robust and consistent association across all examined subgroups (all P < 0.001). For 30-day mortality, although the association was universal, the effect size was significantly greater in younger patient, those with baseline SOFA scores >4, and those with pre-existing diabetes. Similarly, significant interactions were observed for 90-day mortality, where the detrimental effect of high SHR was more pronounced in younger and more severely ill patients. The association with 90-day mortality remained consistent regardless of stroke type or gender.

Discussion

This study demonstrates that in critically ill stroke patients, an SHR at ICU admission is a robust and independent risk factor for sepsis, as well as 30-day and 90-day all-cause mortality. Even after adjustment for baseline confounders, SHR maintained its predictive value for adverse clinical outcomes. Notably, the association with sepsis was particularly pronounced in specific clinical phenotypes, including younger patients, those with ischemic stroke, and those presenting with higher GCS scores. Furthermore, the integration of SHR into traditional prognostic models significantly enhanced their discriminative power, highlighting its potential as a valuable biomarker for early risk stratification in neurocritical care.

In our study, SHR demonstrated statistically significant associations with post-stroke sepsis when modeled both as a continuous exposure and as quartiles. This pattern is concordant with prior stroke literature in which SHR showed consistent signal across categorical and continuous parameterizations for infectious complications, particularly stroke-associated pneumonia.^16,23,24 Biologically, elevated SHR reflects a state of severe neuroendocrine stress and systemic inflammation, which triggers significant immune dysfunction.^25,26 Specifically, acute hyperglycemia can impair neutrophil function and suppress cellular immunity, while simultaneously compromising the microvascular endothelial barrier. When superimposed on stroke-induced immunodepression, this “metabolic-immune” synergism creates a permissive environment for pathogen invasion and dysregulated host responses. Consequently, SHR serves not merely as a marker of glucose levels, but as a surrogate for the systemic physiological frailty that predisposes critically ill stroke patients to sepsis and subsequent organ failure.^27,28

In contrast to ischemic stroke, the association between SHR and sepsis was not statistically significant in the hemorrhagic stroke subgroup, characterized by a non-monotonic dose-response relationship in restricted cubic spline modeling (Supplemental Figure 1). This discrepancy may be attributed to the overwhelming impact of primary brain injury in HS. Specifically, the acute stress response in HS is primarily driven by elevated intracranial pressure and the neurotoxic effects of intraparenchymal blood components, which may exert a “ceiling effect” that masks the predictive value of the metabolic stress ratio.^29,30 In such cases, the severity of the initial neurological insult becomes the dominant determinant of outcome, potentially decoupling SHR from systemic infectious risks. Furthermore, HS patients frequently undergo intensive clinical interventions, including invasive neuromonitoring, mechanical ventilation, and neurosurgical procedures. These iatrogenic factors represent potent, independent drivers of nosocomial infection that may overshadow the influence of baseline glycemic stress, thereby attenuating the statistical power of SHR as a specific predictor for sepsis in this subpopulation.^31,32

Beyond its association with infectious complications, SHR demonstrated a remarkably consistent and robust predictive value for both 30-day and 90-day mortality aligning with previous studies. Additionally, subgroup analyses confirmed the association between SHR and mortality across all strata, with no evidence of effect modification, aligning with previous reports on its universal prognostic value.^17,19,33 This consistency, however, contrasts with the sepsis endpoint, where the association was non-significant in HS and lacked a monotonic dose-response trend. Such divergence may be attributed to the severe early neurological deterioration and high frequency of invasive interventions characteristic of HS overshadowing the incremental metabolic signal captured by SHR.^7,34

The predictive robustness of SHR across both diabetic and non-diabetic cohorts further underscores its clinical utility. Traditionally, the prognostic value of stress hyperglycemia has been a subject of debate, with some studies suggesting its relevance is confined to non-diabetic populations.^19,35 However, our findings align with the emerging consensus that “relative” hyperglycemia, as captured by SHR, retains its prognostic significance by accounting for chronic glycemic backgrounds. Methodologically, by integrating HbA1c to normalize acute glucose elevations, SHR avoids the “ceiling effect” often encountered with absolute glucose levels in diabetic patients, whose baseline tolerance to hyperglycemia is higher.^12,13 Biologically, while elevated SHR in non-diabetic patients may primarily mirror the intensity of the systemic inflammatory response and stroke severity, in diabetic individuals, it likely signifies an acute metabolic decompensation superimposed on pre-existing endothelial and microvascular dysfunction.^35,36 Our strategy of employing SHR as a continuous variable further enhances its sensitivity in detecting these non-linear associations, confirming that the magnitude of acute glycemic stress—relative to the patient’s physiological baseline—is a universal determinant of poor post-stroke outcomes.

Despite the robust associations observed, several limitations of this study should be acknowledged. First, the retrospective observational design precludes causal inference, despite comprehensive multivariable adjustments. Second, as data were derived from a single tertiary-care center in the United States (MIMIC-IV), generalizability to other populations may be limited. Third, sepsis is identified based on EHR data and may be subject to misclassification or underreporting. Fourth, SHR was calculated using admission glucose and a single HbA1c measurement, which may not reflect dynamic glycemic changes or long-term glycemic control, and could be affected by conditions such as anemia. Additionally, our analysis involved multiple clinical outcomes and various exposure parameterizations (modeled both continuously and by quartiles), which increases the potential for Type I error. Consequently, our results should be interpreted as hypothesis-generating, requiring confirmation in independent, pre-specified validation studies. Future studies should aim to validate these findings in prospective, multicenter cohorts with standardized infection definitions and dynamic glucose monitoring. It is also essential to explore whether integrating SHR into clinical decision-making—either as part of a risk prediction model or as a treatment guide—can improve outcomes. Interventional trials targeting SIH, particularly in high-risk non-diabetic stroke patients, may offer new insights into optimizing glycemic control strategies in critical care settings.

Conclusion

Elevated stress hyperglycemia ratio is independently associated with higher risks of sepsis and short-to long-term mortality among critically ill patients with stroke, with consistent associations observed irrespective of diabetes status. In contrast, no statistically significant association between SHR and sepsis was identified in the hemorrhagic stroke subgroup.

ORCID iDs

Shuai Yuan https://orcid.org/0000-0003-4567-2233

Jun Lu https://orcid.org/0000-0003-4403-7406

Ethical Considerations

The data used in this study were extracted from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database, which is a publicly available critical care database developed by the MIT Laboratory for Computational Physiology. Access to the database is granted to credentialed users who complete the required data use training. The MIMIC-IV database is available at: https://physionet.org/content/mimiciv/2.2/. The authors confirm that all data used were de-identified and comply with HIPAA standards. No identifiable private information was accessed or included. All data handling and analysis procedures adhered to the PhysioNet Credentialed Health Data Use Agreement. No additional datasets were generated or analyzed in this study.

Consent to Participate

The study used a publicly available de-identified database; no individual patient consent was required.

Consent for Publication

No identifiable personal data, images, or videos are included in this manuscript.

Data Availability Statement

The data used in this study are available from the MIMIC-IV database (https://physionet.org/content/mimiciv/1.0/), which is publicly accessible under a data use agreement.