Abstract
BACKGROUND:
Sepsis-associated acute respiratory distress syndrome (SA-ARDS) is frequently accompanied by dysregulated glucose metabolism. The stress-hyperglycemia ratio (SHR) and glycemic variability (GV) have emerged as valuable tools for assessing acute dysglycemia. However, their ability to predict disease severity and mortality in patients with SA-ARDS remains unclear.
METHODS:
This retrospective study included 3,243 SA-ARDS patients from the MIMIC-IV database. Patients were stratified into four phenotypes based on the cohort median SHR (1.09) and GV (58.42%). Ordinal logistic regression was used to evaluate ARDS severity, while binary logistic and Cox proportional hazards models were used to assess mortality. Predictive performance was assessed using the area under the receiver operating characteristic curve (AUC). In addition, five machine learning models with SHapley Additive exPlanations (SHAP) analysis were used to identify clinically relevant risk thresholds. External validation using the eICU database was limited to GV-related findings because of the unavailability of hemoglobin A1c (HbA1c) data.
RESULTS:
Compared with the low-GV/low-SHR phenotype, the high-GV/high-SHR phenotype had the strongest association with increased ARDS severity (common odds ratio [cOR] 4.37, 95% confidence intervals [95% CI]: 3.60-5.30, P<0.001) and markedly elevated 28-day mortality (hazard ratio [HR] 12.4, 95% CI: 7.95-19.30, P<0.001). However, compared with conventional clinical scores, the combined assessment of the SHR and GV had superior performance in mortality prediction (all AUC=0.748). Furthermore, machine learning and SHAP analyses revealed a non-linear increase in mortality risk when GV exceeded approximately 39.1%.
CONCLUSION:
Combined assessment of the SHR and GV may provide prognostic information for patients with SA-ARDS. A high-GV/high-SHR phenotype may indicate a high risk of poor prognosis.
Keywords: Sepsis, Acute respiratory distress syndrome, Stress-hyperglycemia ratio, Glycemic variability, Machine learning
INTRODUCTION
Sepsis is defined as life-threatening organ dysfunction caused by a dysregulated host response to infection.[1-2] The lung is particularly susceptible to this systemic injury, and sepsis remains the leading cause of acute respiratory distress syndrome (ARDS). Compared with ARDS of other etiologies, sepsis-associated ARDS (SA-ARDS) is characterized by higher mortality and worse clinical outcomes.[3-6] Therefore, early identification of high-risk patients with SA-ARDS is of clinical importance.[7]
Endothelial injury and alveolar-capillary barrier disruption are central features of SA-ARDS and may be influenced by systemic metabolic disturbances. [8] Stress hyperglycemia, although initially adaptive, becomes maladaptive when excessive, contributing to inflammation and oxidative injury.[9] The stress-hyperglycemia ratio (SHR), defined as the ratio of admission glucose to estimated mean glucose, has emerged as a valuable tool for quantifying stress hyperglycemia.[10]
In parallel, glycemic variability (GV) reflects the temporal instability or oscillation of glucose concentration and is associated with oxidative stress, endothelial glycocalyx injury, and impaired vascular integrity.[11,12] These mechanisms may directly contribute to pulmonary microvascular injury in patients with SA-ARDS.
Despite their complementary biological roles, SHR and GV have largely been studied independently, often in critical care populations.[13-16] Their combined prognostic value has not been fully explored. Given the known heterogeneity of ARDS[17] and the potential non-linear relationship between glucose-related metrics and outcomes, we further incorporated machine learning with SHapley Additive exPlanations (SHAP) analysis[18,19] to identify clinically relevant thresholds. The objectives of this study were to evaluate the combined association of the SHR and GV with disease severity and mortality in patients with SA-ARDS and to assess whether these metabolic indicators could improve risk stratification.
METHODS
Data source and ethical approval
This retrospective cohort study utilized data from the Medical Information Mart for Intensive Care IV (MIMIC-IV, version 3.0) and the eICU Collaborative Research Database (eICU-CRD). Owing to insufficient hemoglobin A1c (HbA1c) data in the eICU-CRD, this external cohort was exclusively used to validate GV-associated outcomes. Ethical approval was obtained with a waiver of informed consent because of the use of de-identified data. The study adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines.[20]
Inclusion and exclusion criteria
Patients were included if they (1) were ≥18 years old, (2) met the Sepsis-3.0 criteria, or (3) had ARDS diagnosed according to the Berlin definition. The exclusion criteria were as follows: (1) ICU stay <24 h; (2) fewer than three glucose measurements or missing HbA1c; or (3) repeated ICU admissions (only the first admission was retained). To prevent carry-over effects and correlated observations, only the first ICU admission per patient was retained.
SA-ARDS was defined as the concurrent presence of sepsis and ARDS during the same ICU stay.
To facilitate disease-specific comparisons, an auxiliary cohort of sepsis patients without ARDS was constructed from the MIMIC-IV database using identical inclusion criteria. This cohort was only used for supplementary analyses. For external validation, the same criteria were applied to the eICU database to identify adult SA-ARDS patients with sufficient glucose measurements for GV calculation. The study flowchart is presented in Figure 1.
Figure 1. The study flowchart. ICU: intensive care unit; HbA1c: hemoglobin A1c; GV: glycemic variability; SHR: stress-hyperglycemia ratio; ARDS: acute respiratory distress syndrome; MIMIC-IV: Medical Information Mart for Intensive Care IV.
Data processing
Data extraction was conducted using Structured Query Language (SQL). The extracted variables included demographic characteristics, vital signs, laboratory measurements, comorbidities, treatments, severity scores, and clinical outcomes, primarily within the first 24 h after ICU admission. The missing data are summarized in supplementary Table 1.
Variables with less than 20% missingness were imputed using multiple imputation by chained equations (MICE), with 10 imputed datasets with 10 iterations each. The imputation model incorporated the SHR, GV, SHR-GV phenotype, baseline covariates, treatment variables, severity scores, and outcomes. Predictive mean matching, logistic regression, and multinomial logistic regression were applied for continuous, binary, and unordered categorical variables, respectively. Variables with more than 20% missing data were excluded from the primary adjustment model. Continuous variables were winsorized at the 1st and 99th percentiles to minimize the influence of extreme values.
Exposure and outcome definitions
The primary exposure variables were the SHR and GV. SHR was calculated as the ratio of admission blood glucose (mg/dL) to estimated average glucose (eAG), where eAG was derived as follows:
| eAG=28.7×HbA1c(%)−46.7 |
GV was quantified using the coefficient of variation (CV) of all glucose measurements obtained during the ICU stay:
| CV=(mean glucose/standard deviation)×100 |
To enhance bedside interpretability and ensure balanced group sizes, patients were categorized into four phenotypes based on the cohort median values of SHR (1.09) and GV (58.42%): low GV/low SHR (reference), low GV/high SHR, high GV/low SHR, and high GV/high SHR. These cutoffs were selected to facilitate clinical interpretation and achieve balanced comparisons rather than to reflect fixed biological thresholds. To mitigate potential information loss due to dichotomization, complementary analyses were conducted to verify the robustness of the findings.
The primary outcomes included SA-ARDS severity and all-cause mortality. ARDS severity classification was based on the lowest PaO2/FiO2 ratio recorded during hospitalization. The mortality outcomes included in-hospital mortality and 28-day, 90-day, and 1-year all-cause mortality.
Statistical analysis
Baseline characteristics were compared using analysis of variance (ANOVA), the Kruskal-Wallis test, or the Chi-square test, as appropriate. Ordinal logistic regression was used to evaluate ARDS severity, binary logistic regression was used for in-hospital mortality, and Cox proportional hazards models were applied for time-to-event outcomes (28-day, 90-day, and 1-year mortality).
Three hierarchical models were constructed: (1) an unadjusted model; (2) a model adjusted for demographic characteristics and vital signs; and (3) a fully adjusted model incorporating body mass index (BMI), white blood cell count (WBC), hemoglobin, platelet count, congestive heart failure, chronic lung disease, hypertension, diabetes mellitus, insulin use, and Sequential Organ Failure Assessment (SOFA) score.
Model diagnostics included the generalized variance inflation factor (GVIF), Brant test, and Schoenfeld residual test. Receiver operating characteristic (ROC) curve analysis was performed for a descriptive evaluation of discriminative ability. A formal DeLong test was not conducted because the SHR-GV phenotype was categorized into four groups. Pre-specified subgroup and sensitivity analyses are detailed in the supplementary materials. All the statistical analyses were performed using R version 4.4.2.
Machine learning analysis
Feature selection for severe SA-ARDS and 28-day mortality was performed using the Boruta algorithm. The MIMIC-IV cohort was randomly divided into training and internal validation sets at a 7:3 ratio. The training set included multiple machine learning algorithms: random forest, extreme gradient boosting (XGBoost), support vector machine (SVM), least absolute shrinkage and selection operator (LASSO) regression, and light gradient boosting machine (LightGBM). The hyperparameters were optimized using a grid search combined with five-fold cross-validation, with the average area under the curve (AUC) as the evaluation metric. The final model performance was assessed exclusively in the validation set. SHapley Additive exPlanations (SHAP) analysis was applied to identify key predictors and determine clinically relevant thresholds.
RESULTS
Baseline characteristics
A total of 3,243 SA-ARDS patients were included in the primary analysis and were stratified into four phenotypes based on the cohort medians for SHR and GV (Table 1 and supplementary Table 2). The high-GV/high-SHR phenotype (n=964) had the most severe clinical profile and highest mortality (28%), whereas the low-GV/low-SHR phenotype had the lowest mortality (2.8%). When the SHR and GV were analyzed separately, both variables demonstrated consistent trends: higher SHR or GV was associated with increased disease severity and poorer clinical outcomes (supplementary Tables 3 and 4).
Table 1.
Baseline characteristics of SA-ARDS patients by SHR-GV phenotype
| Characteristic | Overall (n=3,243) |
Low GV/low SHR (n=963) |
Low GV/high SHR (n=658) |
High GV/low SHR (n=658) |
High GV/high SHR (n=964) |
P-value |
|---|---|---|---|---|---|---|
| Demographics | ||||||
| Age, years | 66.1±13.5 | 68.6±11.5 | 66.8±12.2 | 64.5±14.6 | 64.0±15.0 | <0.001 |
| Female, n (%) | 1,110 (34.2) | 306 (31.8) | 195 (29.6) | 230 (35.0) | 379 (39.3) | <0.001 |
| BMI, kg/m² | 29.5 (25.8-34.4) | 30.2 (26.5-35.1) | 29.8 (26.6-34.3) | 29.0 (25.2-34.9) | 28.5 (24.7-33.5) | <0.001 |
| Respiratory status | ||||||
| PaO2/FiO2 ratio, mmHg* |
117.5 (75.0-182.0) | 156.0 (103.0-208.0) | 142.0 (92.0-202.0) | 100.0 (68.0-153.3) | 88.9 (62.4-138.2) | <0.001 |
| Laboratory findings | ||||||
| HbA1c, % | 5.9 (5.5-6.7) | 6.3 (5.8-7.4) | 5.5 (5.3-5.9) | 6.2 (5.7-7.8) | 5.7 (5.3-6.3) | <0.001 |
| Admission glucose, mg/dL |
136.0 (122.4-166.4) | 127.4 (119.6-138.2) | 136.0 (127.4-150.7) | 122.0 (106.3-152.0) | 168.8 (140.7-209.0) | <0.001 |
| Lactate, mmol/L | 2.0 (1.5-2.8) | 1.9 (1.5-2.4) | 2.0 (1.6-2.8) | 1.7 (1.2-2.4) | 2.3 (1.6-3.9) | <0.001 |
| Severity scores | ||||||
| SOFA score | 3.0 (2.0-5.0) | 3.0 (2.0-5.0) | 4.0 (3.0-5.0) | 3.0 (2.0-5.0) | 4.0 (2.0-6.0) | <0.001 |
| SAPS II score | 40.0±13.7 | 37.3±12.2 | 37.3±13.0 | 41.0±14.1 | 43.9±14.4 | <0.001 |
| Comorbidity and treatment, n (%) | ||||||
| Diabetes | 1,312 (40,5) | 481 (49.9) | 155 (23.6) | 337 (51.2) | 339 (35.2) | <0.001 |
| Insulin use | 2,509 (77.4) | 896 (93.0) | 595 (90.4) | 417 (63.4) | 601 (62.3) | <0.001 |
| Outcome, n (%) | ||||||
| In-hospital mortality | 474 (14.6) | 27 (2.8) | 49 (7.4) | 132 (20.1) | 266 (27.6) | <0.001 |
Data are presented as the mean±SD, median (IQR), or n (%), as appropriate. P-values indicate comparisons across the four SHR-GV phenotypes. The full baseline characteristics are provided in Supplementary Table S1. BMI: body mass index; GV: glycemic variability; HbA1c: hemoglobin A1c; PaO2/FiO2: ratio of arterial oxygen partial pressure to fractional inspired oxygen; SA-ARDS: sepsis-associated acute respiratory distress syndrome; SAPS II: Simplified Acute Physiology score II; SHR: stress-hyperglycemia ratio; SOFA: Sequential Organ Failure Assessment. *The PaO2/FiO2 ratio represents the minimum value recorded during the hospital stay.
ARDS severity
Adjusted probabilities derived from the fully adjusted ordinal logistic regression model (Model 3) demonstrated a clear gradient in SA-ARDS severity across the SHR-GV phenotypes (Figure 2A). Increased SHR and GV were associated with increased disease severity. In the high-GV/high-SHR group, the adjusted probability of severe ARDS reached 58.5%, whereas the probability of mild ARDS decreased to 8.6%. In contrast, the low-GV/low-SHR phenotype exhibited a 25.5% probability of severe ARDS and a 28.5% probability of mild ARDS. The high-GV/low-SHR group also demonstrated a shift toward more severe disease, with an adjusted probability of severe ARDS of 51.2%.
Figure 2. The combined effects of SHR and GV on predicting ARDS severity. A: adjusted predicted probabilities for mild, moderate, and severe ARDS across the SHR-GV phenotypes; B: adjusted common odds ratios (cOR) representing the shift effect toward higher ARDS severity categories. ARDS: acute respiratory distress syndrome; GV: glycemic variability; SHR: stress-hyperglycemia ratio; OR: odds ratio; 95% CI: 95% confidence interval.
Multivariable ordinal logistic regression further quantified these associations. Multicollinearity was minimal in the fully adjusted model (all variance inflation factors [VIFs)]≤1.14; supplementary Table 5). The Brant test indicated that the SHR-GV phenotype met the proportional odds assumption, although several adjustment covariates showed deviations (supplementary Table 6). According to the fully adjusted model (Figure 2B; supplementary Table 7), the high-GV/high-SHR phenotype was most strongly associated with ARDS severity (common odds ratio [cOR] 4.37, 95% confidence intervals [95% CI] 3.60-5.30; P<0.001), followed by the high-GV/low-SHR group (cOR 3.22, 95% CI 2.62-3.96; P<0.001). The low-GV/high-SHR group had a weaker but statistically significant association (cOR 1.21, 95% CI 1.01-1.47; P=0.041). These findings suggest that compared with SHR alone, GV was more strongly associated with disease severity, whereas the combination of both markers resulted in the identification of the highest-risk patients.
Mortality
Multivariable logistic and Cox regression analyses revealed that the high-GV/high-SHR phenotype had the strongest association with mortality among all the phenotypes (Figure 3A; supplementary Tables 8 and 9). According to the fully adjusted Cox model, compared with the low-GV/low-SHR phenotype, the high-GV/high-SHR phenotype had significantly increased 28-day mortality (hazard ratio [HR] 12.4, 95% CI 7.95-19.30; P<0.001) and 90-day mortality (HR 10.0, 95% CI 7.05-14.30; P<0.001). The corresponding adjusted odds ratio for in-hospital mortality was 14.8 (95% CI 9.74-23.50; P<0.001).
Figure 3. Prognostic value of combined SHR and GV phenotypes in SA-ARDS patients. A: forest plot showing multivariable-adjusted odds ratio (OR) for in-hospital mortality and hazard ratios (HR) for 28-day and 90-day mortality; B: Kaplan-Meier survival curves illustrating one-year survival probability, stratified by the four SHR-GV phenotypes. The P-value was calculated using the log-rank test. SA-ARDS: sepsis-associated acute respiratory distress syndrome; GV: glycemic variability; SHR: stress-hyperglycemia ratio.
Schoenfeld residual testing confirmed that the SHR-GV phenotype satisfied the proportional hazards assumption, supporting the validity of the hazard ratio estimates. Although deviations were observed in several covariates (e.g., malignancy and renal disease), these were unlikely to affect the primary conclusions (supplementary Table 10).
In the ancillary sepsis cohort without ARDS, the magnitude of the association was attenuated, with a 28-day mortality HR of 3.36 (95% CI 1.88-6.03) for the high-GV/high-SHR phenotype (supplementary Tables 11 and 12).
Predictive performance
ROC curve analysis demonstrated that conventional clinical scores (supplementary Figure 1), particularly the Logistic Organ Dysfunction System (LODS) (AUC=0.790), outperformed metabolic phenotypes in predicting severe ARDS. However, compared with conventional clinical scores, the combined assessment of SHR and GV had superior performance in mortality prediction (AUC 0.748 vs. <0.680).
Machine learning
Boruta and SHAP analyses revealed that the GV and SHR were important predictors of severe SA-ARDS and 28-day mortality (supplementary Figure 2). Among the evaluated models, the LightGBM achieved the best predictive performance.
SHAP analysis revealed a clinically relevant GV threshold of approximately 39.1%, above which mortality risk increased significantly. Decision curve analysis further supported the added clinical utility of combining the SHR and GV with age and the SOFA score (supplementary Figure 3).
Sensitivity analysis
The results remained robust across multiple analytical approaches, including alternative models, subgroup analyses, and external validation (for GV). Logistic regression analyses for mortality yielded findings consistent with those of the primary Cox models (supplementary Table 8). Analyses using the SHR and GV as continuous variables confirmed their independent associations with ARDS severity and mortality (supplementary Table 13).
Subgroup analyses revealed consistent associations between high-GV phenotypes and severe ARDS and in-hospital mortality across strata defined by age, SOFA score, diabetes status, and chronic pulmonary disease (supplementary Figure 4).
In the external eICU cohort (n=4,487), high GV was associated with increased odds of severe ARDS (adjusted OR 1.63, 95% CI 1.39-1.91; P<0.001) and higher 28-day (adjusted HR 1.30, 95% CI 1.10-1.55) and 90-day mortality (adjusted HR 1.31, 95% CI 1.11-1.54) (supplementary Tables 14-16). Owing to the absence of HbA1c data, external validation was limited to GV rather than the full SHR-GV phenotype.
DISCUSSION
This study demonstrated that combined metabolic dysregulation, characterized by elevated SHR and GV, is strongly associated with adverse outcomes in patients with SA-ARDS. Our findings are consistent with previous literature indicating that increased GV is associated with short-term mortality[14,21] and that elevated SHR is linked to adverse clinical outcomes.[13,15,22] This study extends prior research in several important ways.
First, it focuses on SA-ARDS, a condition characterized by endothelial injury and disruption of the alveolar-capillary barrier. Second, when evaluated separately, compared with the SHR, the GV was more strongly associated with disease severity and mortality. Third, the combined SHR-GV phenotype provided incremental prognostic value beyond that of conventional clinical scoring systems.
SHR and GV may reflect different but complementary aspects of metabolic disorders. SHR reflects acute stress-induced hyperglycemia relative to glucose status, whereas GV reflects instability of glucose control. Rapid fluctuations in blood glucose are associated with oxidative stress, endothelial injury, and glycocalyx shedding, all of which may lead to alveolar-capillary barrier injury in SA-ARDS patients. [23-24] The SHAP-derived GV threshold (~39.1%) may help identify patients who require closer glucose monitoring, nutritional optimization, and careful insulin titration. However, this threshold should be interpreted as a risk indicator rather than a therapeutic target. Prospective interventional studies are needed to determine whether reducing GV can improve clinical outcomes.
Several limitations should be considered. First, the retrospective design limits causal inference, and the findings should be interpreted as associative. Second, despite comprehensive adjustment and sensitivity analyses, residual confounding cannot be excluded. Variables such as glucose monitoring frequency, nutritional support, infection characteristics, and ventilatory strategies were not fully captured and may influence both GV and outcomes. Third, the study population included patients from U.S. ICUs, which may limit generalizability. Fourth, owing to the absence of HbA1c data in the eICU database, external validation of the full SHR-GV phenotype was not feasible. Finally, intermittent glucose monitoring may underestimate true glycemic fluctuations, potentially attenuating the observed associations.
CONCLUSION
Combined assessment of the SHR and GV may provide prognostic information for patients with SA-ARDS. A high-SHR/high-GV phenotype indicates a high risk of poor prognosis.
Funding: This study was supported by the National Natural Science Foundation of China (82472233), the Special Project of the National Natural Science Foundation of China (82241035), the Jiangxi Province Double Thousand Talent Plan (jxsq2023201040), the National Natural Science Foundation of China (82502624, to SJ Bai), and the Natural Science Foundation of Jiangxi Province (20252BAC200111, to SJ Bai).
Ethical approval: This study utilized data from the Medical Information Mart for Intensive Care (MIMIC)-IV database and the eICU Collaborative Research Database (eICU-CRD). The study was approved by the Institutional Review Boards of the Massachusetts Institute of Technology (MIT) and Beth Israel Deaconess Medical Center (BIDMC), and the requirement for informed consent was waived due to the de-identified nature of the data.
Conflicts of interests: The authors declare that they have no competing interests.
Contributors: HC, YHL, and SJB contributed equally to this work and shared the co-first authorship. HC, YHL, SJB, FL, and KJQ conceived and designed the study. HC, YHL, and SJB acquired the data and performed the statistical analysis. YL, TW, LC, XYY, YF, JGW, and YQT participated in data interpretation and validation. YHL, WQT, MLH, and WZ drafted the manuscript and prepared the figures. FL and KJQ supervised the project, critically revised the manuscript for important intellectual content, and acquired funding. All authors read and approved the final manuscript.
All the supplementary files in this paper are available at http://wjem.com.cn.
Contributor Information
Fen Liu, Email: ndyfy01300@ncu.edu.cn.
Kejian Qian, Email: ndyfy00754@ncu.edu.cn.
REFERENCES
- 1. Singer M, Deutschman CS, Seymour CW, Shankar-Hari M, Annane D, Bauer M, et al. . The Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3). JAMA. 2016; 315(8):801-10. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Huang M, Yang X, Li C, Chen H, Pan Y, Chen L, et al. . Sepsis-induced PICS: immunometabolic mechanisms, emerging biomarkers, and precision management. TransMed. 2026; 1(1):100009. [Google Scholar]
- 3. Englert JA, Bobba C, Baron RM. . Integrating molecular pathogenesis and clinical translation in sepsis-induced acute respiratory distress syndrome. JCI Insight. 2019; 4(2):e124061. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Bellani G, Laffey JG, Pham T, Fan E, Brochard L, Esteban A, et al. . Epidemiology, patterns of care, and mortality for patients with acute respiratory distress syndrome in intensive care units in 50 countries. JAMA. 2016; 315(8):788-800. [DOI] [PubMed] [Google Scholar]
- 5. Sheu CC, Gong MN, Zhai R, Chen F, Bajwa EK, Clardy PF, et al. . Clinical characteristics and outcomes of sepsis-related vs. non-sepsis-related ARDS. Chest. 2010; 138(3):559-67. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Mikkelsen ME, Shah CV, Meyer NJ, Gaieski DF, Lyon S, Miltiades AN, et al. . The epidemiology of acute respiratory distress syndrome in patients presenting to the emergency department with severe sepsis. Shock. 2013; 40(5):375-81. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Hudson LD, Milberg JA, Anardi D, Maunder RJ. . Clinical risks for development of the acute respiratory distress syndrome. Am J Respir Crit Care Med. 1995; 151(2 Pt 1):293-301. [DOI] [PubMed] [Google Scholar]
- 8. Matthay MA, Zemans RL, Zimmerman GA, Arabi YM, Beitler JR, Mercat A, et al. . Acute respiratory distress syndrome. Nat Rev Dis Primers. 2019; 5(1):18. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Marik PE, Bellomo R. . Stress hyperglycemia: an essential survival response! Crit Care. 2013; 17(2):305. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Roberts GW, Quinn SJ, Valentine N, Alhawassi T, O’Dea H, Stranks SN, et al. . Relative hyperglycemia, a marker of critical illness: introducing the stress hyperglycemia ratio. J Clin Endocrinol Metab. 2015; 100(12):4490-7. [DOI] [PubMed] [Google Scholar]
- 11. Wang R, Han Q, Fan J, Xu Z, Liu W, Liu D, et al. . Sepsis-induced endothelial barrier dysfunction: mechanisms, pathology, and therapeutic advances. Research (Wash D C). 2025; 8:0997. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Alphonsus CS, Rodseth RN. . The endothelial glycocalyx: a review of the vascular barrier. Anesthesia. 2014; 69(7):777-84. [DOI] [PubMed] [Google Scholar]
- 13. Yan F, Chen X, Quan X, Wang L, Wei X, Zhu J. . Association between the stress hyperglycemia ratio and 28-day all-cause mortality in critically ill patients with sepsis: a retrospective cohort study and predictive model establishment based on machine learning. Cardiovasc Diabetol. 2024; 23(1):163. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Lu Z, Tao G, Sun X, Zhang Y, Jiang M, Liu Y, et al. . Association of blood glucose level and glycemic variability with mortality in sepsis patients during ICU hospitalization. Front Public Health. 2022; 10:857368. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Feng S, Zou R, Wang Y, Huang Y, Zhou Q, Huang Q, et al. . Stress hyperglycemia ratio as a mortality predictor in nondiabetic septic patients: a retrospective cohort analysis. BMC Infect Dis. 2025; 25(1):752. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. He HM, Zheng SW, Xie YY, Wang Z, Jiao SQ, Yang FR, et al. . Simultaneous assessment of stress hyperglycemia ratio and glycemic variability to predict mortality in patients with coronary artery disease: a retrospective cohort study from the MIMIC-IV database. Cardiovasc Diabetol. 2024; 23(1):61. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Lin M, Xu F, Deng Y, Wei Y, Shi F, Xie Y, et al. . Large-scale proteomic profiling identifies distinct inflammatory phenotypes in acute respiratory distress syndrome (ARDS): a multicenter, prospective cohort study. Eur Respir J. 2025;2500933. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Lundberg SM, Erion G, Chen H, DeGrave A, Prutkin JM, Nair B, et al. . From local explanations to global understanding with explainable AI for trees. Nat Mach Intell. 2020; 2(1):56-67. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Liu Q, Zhang Y, Sun J, Wang K, Wang Y, Wang Y, et al. . Early identification of high-risk patients admitted to emergency departments using vital signs and machine learning. World J Emerg Med. 2025; 16(2):113. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP, et al. . The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. Lancet. 2007; 370(9596):1453-7. [DOI] [PubMed] [Google Scholar]
- 21. Hryciw BN, Ghossein J, Rochwerg B, Meggison H, Fernando SM, Kyeremanteng K, et al. . Glycemic variability as a prognostic factor for mortality in patients with critical illness: a systematic review and meta-analysis. Crit Care Explor. 2024; 6(1):e1025. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Xia D, Luo X, Zhu Y, Zhu J, Xie Y. . Assessment of stress hyperglycemia ratio to predict mortality in critically ill patients with sepsis: a retrospective cohort study from the MIMIC-IV database. Front Endocrinol (Lausanne). 2025; 16:1496696. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Nieuwdorp M, van Haeften TW, Gouverneur MCLG, Mooij HL, van Lieshout MHP, Levi M, et al. . Loss of endothelial glycocalyx during acute hyperglycemia coincides with endothelial dysfunction and coagulation activation in vivo. Diabetes. 2006; 55(2):480-6. [DOI] [PubMed] [Google Scholar]
- 24. Schmidt EP, Yang Y, Janssen WJ, Gandjeva A, Perez MJ, Barthel L, et al. . The pulmonary endothelial glycocalyx regulates neutrophil adhesion and lung injury during experimental sepsis. Nat Med. 2012; 18(8):1217-23. [DOI] [PMC free article] [PubMed] [Google Scholar]