Association of Glycemic Variability, Stress Hyperglycemia Ratio, and Hemoglobin Glycation Index With 28-Day Mortality in Sepsis: A Multicenter Retrospective Study With Mediation and Machine-Learning Analyses

Association of Glycemic Variability, Stress Hyperglycemia Ratio, and Hemoglobin Glycation Index With 28-Day Mortality in Sepsis: A Multicenter Retrospective Study With Mediation and Machine-Learning Analyses | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Association of Glycemic Variability, Stress Hyperglycemia Ratio, and Hemoglobin Glycation Index With 28-Day Mortality in Sepsis: A Multicenter Retrospective Study With Mediation and Machine-Learning Analyses kangxing wang, Huaiyu Xiong, Yukun Zhu, Yongfang Zhou, Yan Kang This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8247079/v1 This work is licensed under a CC BY 4.0 License Status: Under Revision Version 1 posted 14 You are reading this latest preprint version Abstract Background Glycemic dysregulation is a hallmark of sepsis, but the comparative prognostic value of different glycemic metrics remains unclear. We aimed to systematically compare the associations of Glycemic Variability (GV), Stress Hyperglycemia Ratio (SHR), and Hemoglobin Glycation Index (HGI) with mortality in sepsis and to develop a machine learning-based prediction model for mortality in sepsis. Methods This multicenter retrospective study included 7,260 adult patients with sepsis from the MIMIC-IV and NWICU databases. GV was calculated as the coefficient of variation of blood glucose; SHR and HGI were computed using established formulas. The primary outcome was 28-day all-cause mortality. Associations between glycemic indices and mortality were assessed using multivariable Cox regression, restricted cubic splines, mediation analysis, and subgroup analyses. An XGBoost model incorporating glycemic and clinical variables was developed, and variable importance was evaluated using SHAP (SHapley Additive exPlanations) values. Results After full adjustment, patients in the highest quintile of GV (HR = 1.62, 95% CI 1.33–1.97) and SHR (HR = 1.51, 95% CI 1.25–1.82) had significantly higher 28-day mortality, whereas those in the highest HGI quintile had lower mortality (HR = 0.74, 95% CI 0.61–0.91). Lactate partially mediated these associations. The XGBoost model demonstrated excellent performance (AUC = 0.798), and SHAP analysis identified GV, SHR, and HGI among the top predictors of mortality. These findings remained robust in sensitivity analysis using Random Forest imputation. Conclusion GV, SHR, and HGI are independent and complementary prognostic markers in sepsis. A multidimensional evaluation of glycemic dysregulation enhances risk stratification, and integrating these indices into machine learning models substantially improves predictive accuracy. Health sciences/Biomarkers Health sciences/Diseases Health sciences/Endocrinology Health sciences/Medical research Health sciences/Risk factors Sepsis Glycemic Variability Stress Hyperglycemia Ratio Hemoglobin Glycation Index Machine Learning Mortality Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Introduction Sepsis, a syndrome of life-threatening organ dysfunction precipitated by a dysregulated host response to infection, represents a formidable global health burden and remains a leading cause of mortality in intensive care units (ICUs) [ 1 ]. Central to the pathophysiology of sepsis are profound metabolic derangements, of which dysglycemia—encompassing hyperglycemia, hypoglycemia, and increased glycemic lability—is a near-universal and particularly pernicious feature [ 2 ]. This disruption of glycemic homeostasis is driven by a complex interplay between systemic inflammation, counter-regulatory hormone excess, and severe insulin resistance, creating a vicious cycle that exacerbates organ injury. A robust body of evidence has firmly established that the severity of dysglycemia is independently associated with increased mortality and organ failure in sepsis, thereby highlighting the critical importance of sophisticated glycemic monitoring and management in this vulnerable population [ 3 , 4 ]. To address the complexity of dysglycemia in sepsis, the field has shifted beyond single-point glucose measurements toward a multidimensional assessment paradigm aimed at capturing the distinct pathophysiological dimensions of glycemic dyshomeostasis. Glycemic Variability (GV), for instance, quantifies the dynamic instability of glucose levels. This lability has been shown to independently drive adverse clinical outcomes through mechanisms such as exacerbated oxidative stress and endothelial injury [ 5 , 6 ]. Complementing this dynamic assessment, the Stress Hyperglycemia Ratio (SHR) isolates the intensity of the pure "stress" component of acute hyperglycemia by normalizing it against the patient's chronic glycemic background, a metric validated for its prognostic significance across diverse critically ill populations [ 7 , 8 ]. Transcending these short-term and acute-stress dimensions, the Hemoglobin Glycation Index (HGI) offers a deeper perspective by revealing the intrinsic discordance between an individual's long-term (via HbA1c) and short-term (via fasting glucose) glycemic control. This incongruity, potentially reflecting idiosyncratic traits such as non-enzymatic glycation rates or red blood cell lifespan, has emerged as a novel biomarker with promise for discerning distinct risk phenotypes [ 9 , 10 ]. While Glycemic Variability (GV), the Stress Hyperglycemia Ratio (SHR), and the Hemoglobin Glycation Index (HGI) each offer a unique perspective on the glycemic dysregulation in sepsis—reflecting short-term lability, acute stress, and chronic adaptation, respectively—they have predominantly been investigated in isolation. To date, no study has comprehensively compared their prognostic utility head-to-head within the same large, multi-center cohort of patients with sepsis. Furthermore, the relative importance and potential interplay of these factors within advanced predictive models that integrate multidimensional clinical data remain unexplored, limiting a comprehensive understanding of which glycemic profiles confer the highest risk. Therefore, this study aimed to systematically evaluate and compare the independent prognostic value of GV, SHR, and HGI for 28-day mortality in patients with sepsis, and to determine their roles in a machine learning-based predictive model. Methods Data source: The data utilized in this study were derived from two large, publicly available critical care databases. The first dataset was the Medical Information Mart for Intensive Care IV (MIMIC-IV, v3.1), an electronic health record dataset comprising patients admitted to Beth Israel Deaconess Medical Center in Boston, MA, from 2008 to 2022 [ 11 ]. The second dataset was the Northwestern ICU (NWICU, v0.1.0) database, which contains data from multiple hospitals within the Northwestern Medicine network in Chicago, IL, from 2020 to 2022 [ 12 ]. The NWICU database is structurally harmonized with MIMIC-IV, facilitating the integration of data from both sources for this multi-center retrospective analysis. The Institutional Review Boards of both originating institutions approved the data collection and sharing initiatives, granting a waiver of informed consent. An author of this study (Certification number: 13024213) obtained certified access to use these databases. Study population: We included adult patients (aged ≥ 18 years) admitted to the ICU for the first time with a diagnosis of sepsis from both the MIMIC-IV and NWICU databases. Sepsis was defined according to the Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3) criteria, requiring a suspected infection and a sequential organ failure assessment (SOFA) score increase of ≥ 2 points [ 13 ]. Initially, a total of 48,175 patients with sepsis were identified, comprising 41,295 from MIMIC-IV and 6,880 from NWICU. Patients were subsequently excluded based on the following criteria: (1) ICU length of stay less than 24 hours (n = 5,186); (2) not the first ICU admission (n = 4,020); (3) fewer than three blood glucose measurements recorded during the ICU stay (n = 4,149); and (4) missing data for hemoglobin A1c or fasting blood glucose required for calculating exposure variables (n = 27,560). After applying these criteria, a final analytical sample of 7,260 patients was included for analysis (Fig. 1 ). Data extraction and definitions: We used Structured Query Language (SQL) with PostgreSQL to extract all relevant data from the MIMIC-IV and NWICU databases. Data for baseline characteristics, severity scores, and laboratory results required for calculating SHR and HGI were extracted from the first 24 hours of ICU admission. Blood glucose measurements for the calculation of GV were collected throughout the entire ICU stay. The extracted data included demographic information (age, gender, race, BMI), severity of illness scores (SOFA, OASIS, Charlson Comorbidity Index), vital signs (mean arterial pressure, heart rate, respiratory rate, temperature), laboratory results (hemoglobin (Hb), white blood cell count (WBC), platelet count (Plt), lactate (Lac), total bilirubin (Tbil), aspartate aminotransferase (AST), alanine aminotransferase (ALT), and creatinine (Cre)), and details on interventions such as mechanical ventilation, renal replacement therapy, and vasopressor use. Comorbidities, including hypertension, diabetes, COPD, heart failure, and stroke, were identified using International Classification of Diseases, Ninth and Tenth Revision (ICD-9/10) codes. The primary outcome was 28-day all-cause mortality, ascertained from hospital records and the Social Security Death Index. The primary exposure variables were defined as follows. Glycemic variability (GV) was calculated as the coefficient of variation (standard deviation/mean × 100%) of all blood glucose measurements during the entire ICU stay [ 14 ]. The stress hyperglycemia ratio (SHR) was calculated using the formula: admission glucose (mg/dL) / (28.7 × HbA1c [%] – 46.7) [ 7 ]. The hemoglobin glycation index (HGI) was calculated as the difference between the measured HbA1c and the predicted HbA1c. To account for potential differences between the two source databases, the predicted HbA1c was derived from separate linear regression models fitted within each cohort based on the patient's first fasting blood glucose (FBG), following a previously established methodology [ 15 ]. The resulting predictive equations were predicted HbA1c (%) = 0.0078 × FBG (mg/dL) + 5.2165 for the MIMIC-IV cohort, and predicted HbA1c (%) = 0.0085 × FBG (mg/dL) + 5.2137 for the NWICU cohort. The correlations between HGI and HbA1c within each cohort are shown in Supplementary Figure S1 . Study endpoint: The primary outcome of interest was 28-day all-cause mortality. This was defined as death from any cause occurring within 28 days following the date and time of ICU admission. The vital status of each patient at day 28 was determined using electronic health records for in-hospital deaths and supplemented by data from the Social Security Death Index for out-of-hospital deaths. Statistical analysis: For our primary analyses, Glycemic Variability (GV), Stress Hyperglycemia Ratio (SHR), and Hemoglobin Glycation Index (HGI) were categorized into quintiles based on their distributions in the study cohort. In the baseline data analysis, continuous variables were expressed as mean (standard deviation) or median (interquartile range) and compared using t-tests or Mann-Whitney U tests as appropriate. Categorical variables were reported as frequencies (percentages) and compared using the chi-square or Fisher's exact test. For survival analysis, Kaplan-Meier curves were plotted to illustrate 28-day survival trends across the quintiles of GV, SHR, and HGI, with differences assessed by the log-rank test. We then conducted Cox proportional hazards regression analyses to further elucidate the relationship between each glycemic metric and 28-day all-cause mortality. To account for potential confounders, we applied multivariate Cox regression with hierarchical adjustments. Model 1 provided unadjusted estimates. Model 2 was partially adjusted for age, gender, and race. Model 3 was fully adjusted for all covariates listed in Model 2 plus BMI, severity of illness scores (SOFA, OASIS, Charlson Comorbidity Index) [ 16 – 18 ], comorbidities, vital signs, laboratory results, interventions and database source. In addition, restricted cubic spline (RCS) plots with three knots were used to assess the linear or nonlinear relationships between the continuous values of GV, SHR, and HGI and clinical outcomes. To determine whether these associations differed among various populations, we performed subgroup analyses based on key clinical characteristics and calculated p-values for interaction. The results of the subgroup analyses were visualized using forest plots. A mediation analysis was then performed to investigate the extent to which lactate levels mediated the effects of the glycemic metrics on 28-day mortality. Finally, to evaluate the predictive importance of our glycemic metrics and build a precise prediction model, we designed a systematic machine learning pipeline. First, we performed feature selection by taking the union of variables identified by both the Least Absolute Shrinkage and Selection Operator (LASSO) regression and the Boruta algorithm to account for both linear and non-linear associations. Based on this selected feature set, we then developed and compared twelve distinct machine learning models: Extreme Gradient Boosting (XGBoost), Bayesian Additive Regression Trees (BART), Gradient Boosting Machine (GBM), Random Forest, Light Gradient Boosting Machine (LightGBM), Logistic Regression, Gradient Boosting with Linear Models (GLMBoost), Neural Network (Nnet), Support Vector Machine (SVM), Naïve Bayes, Recursive Partitioning and Regression Trees (Rpart), and K-Nearest Neighbors (K-NN). The predictive performance of each model was evaluated on the validation set using the area under the receiver operating characteristic curve (AUC) [ 8 , 19 , 20 ]. Covariates with a missing data rate of less than 40% were included, and any remaining missing values were handled using multiple imputation by chained equations (MICE). To verify the robustness of our findings regarding the missing data handling, we performed a sensitivity analysis by repeating the multivariable Cox regression models using the Random Forest algorithm for imputation. All statistical analyses were performed using R software (version 4.4.2), and a two-sided p-value of less than 0.05 was considered statistically significant. Results Baseline characteristics: A total of 7,260 patients with sepsis from the combined databases were included, with baseline characteristics stratified by 28-day mortality presented in Table 1. Compared to survivors, non-survivors were significantly older and had a lower body mass index (BMI). They presented with a greater severity of illness upon admission, as evidenced by significantly higher SOFA, OASIS, and Charlson Comorbidity Index scores. In terms of comorbidities, non-survivors had a higher prevalence of COPD and stroke, alongside more unstable vital signs, including elevated heart and respiratory rates. Laboratory findings revealed that non-survivors had significantly increased levels of lactate, creatinine, and white blood cells. Regarding glucose metabolism, although there were no significant differences in the prevalence of diabetes or baseline HbA1c levels between the groups, non-survivors exhibited significantly higher admission fasting blood glucose (FBG) levels. Crucially, non-survivors displayed significantly higher glucose variability (GV) and stress hyperglycemia ratio (SHR), but a lower hemoglobin glycation index (HGI) (all p < 0.001). regarding interventions, non-survivors were more likely to receive renal replacement therapy (RRT) but, notably, were less frequently treated with insulin or vasopressors compared to survivors. Survival analysis: To assess the independent prognostic value of GV, SHR, and HGI for 28-day mortality, we constructed multivariable Cox proportional hazards regression models (Table 2). The analysis revealed that both Glucose Variability (GV) and the Stress Hyperglycemia Ratio (SHR) were positively and independently associated with mortality. After full adjustment for confounders (Model 3), the highest quintiles of GV (Q5 vs Q1: HR = 1.62, 95% CI 1.33–1.97) and SHR (Q5 vs Q1: HR = 1.51, 95% CI 1.25–1.82) remained strong predictors of death. In contrast, the Hemoglobin Glycation Index (HGI) demonstrated a significant protective effect, where higher levels were associated with improved survival compared to the lowest quintile. This inverse association was robust across all adjustments; specifically, in the fully adjusted model, the highest HGI quintile (Q5) remained significantly associated with reduced mortality risk compared to the lowest quintile (HR = 0.74, 95% CI 0.61–0.91). These findings indicate that high GV, high SHR, and low HGI are all independent markers of poor prognosis in patients with sepsis. Association with outcomes: To further explore the dose-response relationship between GV, SHR, HGI, and 28-day mortality, we conducted adjusted restricted cubic spline (RCS) analysis (Fig. 2 A-C). The results revealed significant non-linear, positive associations for both Glucose Variability (GV) and the Stress Hyperglycemia Ratio (SHR) with mortality risk (P for non-linearity < 0.05 and 1). In contrast, the Hemoglobin Glycation Index (HGI) exhibited a linear, inverse association with mortality (P for non-linearity > 0.05), indicating that lower HGI levels—particularly below the threshold of -0.4—were correlated with a progressively higher risk of death. These findings were strongly corroborated by the Kaplan-Meier survival analysis (Fig. 2 D-F). The Log-rank tests were highly significant for all three variables (all P < 0.01), demonstrating distinct survival probabilities across the quintiles. Patients in the highest quintile (Q5) of GV and SHR had the lowest survival rates. Conversely, for HGI, those in the lowest quintile (Q1) experienced the worst survival outcomes, perfectly aligning with the trends observed in the RCS analysis. Subgroup analyses: To assess the consistency of the associations between GV, SHR, HGI, and mortality across different patient populations, we performed stratified analyses and tested for interactions based on key clinical variables, including age, gender, SOFA score, comorbidities, and interventions (Fig. 3 ). Overall, the adverse effects of high Glucose Variability (GV) and high Stress Hyperglycemia Ratio (SHR), as well as the protective effect of a high Hemoglobin Glycation Index (HGI), were largely consistent across most subgroups. However, interaction analyses revealed several significant points of heterogeneity. For GV, a significant interaction was detected with age (P for interaction = 0.031) and vasopressor use (P for interaction = 0.002), suggesting its risk effect might be more pronounced in younger patients. For SHR, its association with mortality was significantly modified by heart failure (P = 0.036), mechanical ventilation (P = 0.001), insulin use (P = 0.001), and vasopressor use (P = 0.001), indicating that the detrimental impact of high SHR was substantially magnified in patients without heart failure and in those requiring critical care interventions. For HGI, its protective effect was modified by diabetes (borderline interaction, P = 0.055) and vasopressor use (P = 0.008), implying that the benefit of a high HGI may be attenuated in patients with diabetes. No other significant interactions were observed for the remaining subgroups. Mediation analysis: Given that elevated lactate is a key biomarker of poor prognosis in sepsis, we performed a mediation analysis to explore whether lactate mediates the associations between GV, SHR, HGI, and 28-day mortality, with the results visualized in path diagrams (Fig. 4 ) and detailed in Supplementary Table S1 . After full adjustment for potential confounders (Model 3), lactate was identified as a significant mediator for all three glycemic indices. Specifically, lactate mediated 12.8% (95% CI: 5.6%, 23.7%) of the total effect of Glucose Variability (GV) on mortality, 19.5% (95% CI: 9.4%, 36.2%) of the total effect of the Stress Hyperglycemia Ratio (SHR), and 7.9% (95% CI: 2.9%, 41.7%) of the total effect of the Hemoglobin Glycation Index (HGI), with all mediation effects being statistically significant (p < 0.05). Notably, the average direct effects (ADE) of GV, SHR, and HGI remained significant after accounting for lactate's mediating role across all models. This indicates that lactate functions as a partial mediator, suggesting that these glycemic indices influence patient outcomes not only through the lactate-mediated pathway but also via other independent pathophysiological mechanisms. Machine learning analysis: To comprehensively evaluate the predictive importance of numerous clinical variables, including GV, SHR, and HGI, and to build a precise mortality prediction model, we designed a systematic feature selection and modeling pipeline. We first employed both LASSO regression (Fig. 5 A-B) and the Boruta algorithm (Fig. 5 C) to screen all potential predictors (see Supplementary Tables S2 and S3 for detailed results), taking the union of the selected features to account for both potential linear and non-linear associations and form the final feature set. The cohort was then divided into training (70%) and validation (30%) sets using stratified sampling based on the database source. Based on the selected features, multiple machine learning models were developed and compared (see Supplementary Table S4 for detailed performance metrics). ROC analysis demonstrated that the XGBoost model achieved the best predictive performance, with an AUC of 0.798 on the validation set (Fig. 6 ). Notably, our three glycemic management indices of interest—GV, SHR, and HGI—were all included in the final predictive model through this comprehensive selection process, highlighting their central role in sepsis prognostication. To interpret the top-performing XGBoost model, we utilized SHAP analysis. The SHAP feature importance plot further corroborated our findings: while traditional markers like age and hemoglobin remained significant, GV, SHR, and HGI were ranked among the most important predictors, even outperforming some conventional clinical variables. The SHAP summary plot further revealed that higher GV and SHR values drive the model to predict a higher risk of death. In contrast, higher HGI values drive a lower risk prediction, which is perfectly consistent with our survival analysis results. Sensitivity analyses: To verify the robustness of our findings regarding the handling of missing data, we performed a sensitivity analysis by comparing different imputation methods. While the primary analysis utilized Multiple Imputation by Chained Equations (MICE), we repeated the multivariable Cox regression models using Random Forest imputation. The results obtained from the Random Forest dataset were highly consistent with the primary analysis (Supplementary Table S5). specifically, in the fully adjusted model (Model 3), high GV (Q5 vs Q1: HR = 1.63, 95% CI 1.34–1.98) and high SHR (Q5 vs Q1: HR = 1.50, 95% CI 1.24–1.81) remained significant independent predictors of mortality. Similarly, the protective association of HGI was preserved, with the highest quintile showing a significantly reduced risk compared to the lowest (HR = 0.74, 95% CI 0.61–0.90). These findings confirm that our results are robust and not driven by the specific method of missing data imputation. Discussion In this large, bi-centric retrospective analysis, we conducted the first systematic, head-to-head comparison of three key glycemic indices—Glycemic Variability (GV), the Stress Hyperglycemia Ratio (SHR), and the Hemoglobin Glycation Index (HGI)—to evaluate their prognostic value in critically ill patients with sepsis. We found that high GV, high SHR, and low HGI were each independently and strongly associated with increased 28-day mortality. Using advanced, interpretable machine learning approaches, we further confirmed that these three indices were among the most influential predictors of death, even when integrated with a wide array of clinical variables. Together, they represent distinct yet complementary dimensions of dysglycemia—short-term instability (GV), acute stress response (SHR), and chronic glycemic adaptation (HGI)—forming a multidimensional glycemic profile that more accurately reflects the metabolic complexity of sepsis. Glycemic Variability and Stress Hyperglycemia: Dynamic and Acute Markers of Severity Our findings reinforce prior evidence linking both glucose fluctuation and stress hyperglycemia with adverse outcomes in critical illness [ 6 , 8 , 21 , 22 ]. In this large, multi-center cohort, patients in the highest quintiles of GV and SHR exhibited significantly higher mortality, independent of illness severity, comorbidities, and therapeutic interventions. These results emphasize that dysglycemia in sepsis cannot be captured by static glucose levels alone; instead, the magnitude and instability of glucose excursions, along with the degree of stress-induced metabolic response, are crucial determinants of clinical outcome. Mechanistically, these two indices likely reflect distinct aspects of metabolic derangement. Glycemic Variability quantifies fluctuations in glucose over time—episodes of hyper- and hypoglycemia—that trigger oxidative stress, mitochondrial injury, and endothelial dysfunction, thereby amplifying systemic inflammation and multi-organ failure [ 21 – 24 ]. SHR, in contrast, reflects the disproportionate elevation of admission glucose relative to chronic glycemic background, capturing the intensity of the neuroendocrine stress response. Elevated SHR signals heightened sympathetic activation, increased cortisol and catecholamine release, and severe insulin resistance—hallmarks of the metabolic storm accompanying septic shock [ 25 ]. Our mediation analysis supports this interpretation: lactate partially mediated the associations between both GV and SHR and mortality (12.8% and 19.5% of total effects, respectively). Elevated lactate reflects tissue hypoperfusion and metabolic stress, suggesting that these glycemic perturbations contribute to adverse outcomes through impaired cellular energetics and augmented anaerobic metabolism. Furthermore, subgroup analyses revealed that the detrimental effects of high SHR were especially pronounced among patients requiring mechanical ventilation or vasopressor support, highlighting its role as a marker of extreme physiological stress. By contrast, GV’s adverse effects were more evident in younger patients, suggesting that glucose fluctuations may exert independent cytotoxic effects even in individuals with greater physiological reserve. Hemoglobin Glycation Index: A Paradoxical but Insightful Marker The most novel finding of this study is the inverse association between HGI and mortality, where a low HGI predicted poor survival, while a high HGI was relatively protective. This pattern, consistent with reports in other critically ill populations such as coronary artery disease and post–transcatheter aortic valve replacement [ 9 , 26 ], suggests that HGI reflects intrinsic biological differences in glycemic response beyond average glucose control. We propose two potential, non-mutually exclusive explanations. The first, the “Physiological Resilience” hypothesis, posits that a high HGI identifies individuals who maintain greater tolerance to hyperglycemia-related oxidative and inflammatory stress, possibly through altered red blood cell turnover or glycation kinetics that blunt hyperglycemia-induced injury [ 27 ]. The second, the “Acute Stress Amplifier” hypothesis, interprets a low HGI as a marker of disproportionate acute stress hyperglycemia. In sepsis, massive catecholamine and cortisol surges can markedly elevate fasting glucose independent of prior glycemic history. Thus, a low HGI—where measured HbA1c is markedly lower than fasting glucose–predicted HbA1c—likely signifies an exaggerated stress response and severe metabolic disarray. In this sense, a low HGI does not denote good long-term glycemic control, but rather excessive acute hyperglycemia relative to the individual’s baseline, identifying patients with profound catabolic drive and cytokine-mediated injury [ 28 ]. The partial mediation of HGI’s mortality association through lactate supports this hypothesis, implying that low HGI reflects an underlying metabolic collapse characterized by both hyperglycemia and tissue hypoxia. Notably, the strength of the HGI–mortality association varied slightly between the MIMIC-IV and NWICU cohorts, possibly due to differences in illness severity or glucose monitoring practices. Nevertheless, the consistent directionality of the association across cohorts underscores HGI’s robustness as a potential metabolic stress biomarker in sepsis. Machine Learning and the Multidimensional Glycemic Profile Our machine learning analysis provided complementary, data-driven evidence of the prognostic importance of these glycemic indices. Among twelve tested algorithms, the XGBoost model achieved the highest discrimination (AUC = 0.798), outperforming conventional clinical scores such as SOFA and OASIS. Importantly, SHAP (SHapley Additive Explanations) analysis provided interpretability by ranking variable contributions: GV, SHR, and HGI were among the most influential predictors of mortality, surpassing several traditional laboratory and physiological markers. The directional effects—higher GV and SHR increasing mortality risk, higher HGI reducing it—mirrored our regression analyses, confirming the internal coherence of these findings. Together, these results suggest that integrating multiple glycemic dimensions provides a more comprehensive assessment of metabolic dysfunction than reliance on any single index. This approach captures the interplay between chronic adaptation, acute stress, and dynamic instability—three interconnected elements of glucose metabolism that collectively define the “glycemic signature” of sepsis. Clinical Implications These findings have several important clinical implications. First, incorporating GV, SHR, and HGI into routine ICU monitoring could markedly improve risk stratification in septic patients. These indices are easily derived from routinely collected data—inuous glucose measurements, admission glucose, fasting glucose, and HbA1c—and could feasibly be integrated into automated electronic health record systems. Second, recognizing the distinct biological roles of these indices could inform individualized management strategies. Patients with extreme GV may benefit from interventions to minimize glycemic fluctuations, whereas those with high SHR or low HGI may warrant earlier metabolic resuscitation or targeted modulation of the stress response. Finally, the integration of these metrics into real-time decision-support tools—potentially powered by machine learning—may enable early detection of high-risk phenotypes and personalized therapeutic adjustment. Strengths and Limitations This study has several notable strengths. It leverages two large, harmonized ICU databases, enhancing generalizability across diverse patient populations. It is also the first to compare GV, SHR, and HGI simultaneously within a unified analytic framework, combining conventional regression, mediation, and interpretable machine learning to ensure robust conclusions. However, several limitations warrant acknowledgment. The retrospective design limits causal inference, and despite extensive adjustments, residual confounding cannot be excluded. The accuracy of GV depends on glucose monitoring frequency, which may vary across institutions. Moreover, HGI assumes a stable linear relationship between fasting glucose and HbA1c—an assumption potentially altered during acute illness due to changes in erythrocyte lifespan and protein turnover. Finally, our findings, while internally validated, require prospective and external validation before clinical implementation. Conclusion In conclusion, Glycemic Variability, the Stress Hyperglycemia Ratio, and the Hemoglobin Glycation Index are all powerful and independent prognostic markers in patients with sepsis, each capturing a unique and complementary dimension of glycemic dysregulation. High GV and high SHR are associated with increased mortality, while a high HGI is paradoxically protective. The integration of these three indices into machine learning models significantly enhances the accuracy of risk prediction. Our findings strongly support a multi-dimensional approach to glycemic assessment for superior risk stratification in critically ill patients with sepsis. Abbreviations ACME Average Causal Mediation Effect ADE Average Direct Effect ALT Alanine Aminotransferase AST Aspartate Aminotransferase AUC Area Under the Curve BMI Body Mass Index CI Confidence Interval COPD Chronic Obstructive Pulmonary Disease Cre Creatinine FBG Fasting Blood Glucose GV Glycemic Variability Hb Hemoglobin HbA1c Hemoglobin A1c HF Heart Failure HGI Hemoglobin Glycation Index HR Hazard Ratio ICU Intensive Care Unit Lac Lactate LASSO Least Absolute Shrinkage and Selection Operator MAP Mean Arterial Pressure MIMIC-IV Medical Information Mart for Intensive Care IV NWICU Northwestern Intensive Care Unit OASIS Oxford Acute Severity of Illness Score Plt Platelet Count RCS Restricted Cubic Spline ROC Receiver Operating Characteristic RRT Renal Replacement Therapy SHAP SHapley Additive exPlanations SHR Stress Hyperglycemia Ratio SOFA Sequential Organ Failure Assessment Tbil Total Bilirubin Vaso Vasopressor Use Vent Mechanical Ventilation WBC White Blood Cell Count XGBoost Extreme Gradient Boosting Declarations Ethics approval and consent to participate: The study was conducted in accordance with the Declaration of Helsinki. Ethical approval and individual patient consent were waived for all three databases (MIMIC-IV, MIMIC-III-CareVue, and eICU-CRD) because they contain de-identified health information that is publicly available for research purposes. The use of the MIMIC-IV and MIMIC-III databases for this study was specifically approved by the Massachusetts Institute of Technology Institutional Review Board. The authorized researcher (certification number 13024213) completed the required data user training, which granted access approval for all three publicly available cohorts, including eICU-CRD. Consent for publication: Not applicable. Availability of data and materials The datasets utilized in this study were obtained from three large-scale, publicly available critical care databases: the Medical Information Mart for Intensive Care IV (MIMIC-IV, version 3.1), the MIMIC-III CareVue subset (v1.4), and the eICU Collaborative Research Database (eICU-CRD, version 2.0). Data access was granted and data were extracted by an authorized researcher (Kangxing Wang, certification number: 13024213) after obtaining necessary approvals and completing ethical training. The datasets are publicly accessible for research purposes via PhysioNet. Competing Interests: The authors declare no competing financial interests. Fundings: The study was approved by National Key R&D Program of China（2022YFC2504500）and 1.3.5 project for disciplines of excellence, West China Hospital, Sichuan University（ZYGD23012） Author Contributions WKX, XHY, and YYK contributed equally to this work as co-first authors. WKX was the primary contributor, responsible for the study conception, essential data acquisition (coding and extraction), performing the main statistical analysis, and drafting the manuscript. XHY refined the methodology, provided technical support for the analysis, and assisted with data management. YYK assisted with the statistical modeling, interpretation of results, and critical revision of the manuscript. KY and ZYF served as co-corresponding authors, secured funding, and supervised the study. Both KY and ZYF critically revised the manuscript for intellectual content. All authors read and approved the final manuscript. Acknowledgments: Not applicable References Meyer, N. J. & Prescott, H. C. Sepsis and septic shock. Hardin CC, editor. N Engl J Med. ;391:2133–46. (2024). https://doi.org/10.1056/NEJMra2403213 Hotamisligil, G. S. Inflammation and metabolic disorders. Nature 444 , 860–867. https://doi.org/10.1038/nature05485 (2006). Krinsley, J. S. et al. Diabetic status and the relation of the three domains of glycemic control tomortality in critically ill patients: an international multicenter cohort study. Crit. Care . 17 , R37. https://doi.org/10.1186/cc12547 (2013). The NICE-SUGAR Study Investigators. Hypoglycemia and risk of death in critically ill patients. N Engl. J. Med. 367 , 1108–1118. https://doi.org/10.1056/NEJMoa1204942 (2012). Service, F. J. Glucose variability. Diabetes 62 , 1398–1404. https://doi.org/10.2337/db12-1396 (2013). Ali, N. A. et al. Glucose variability and mortality in patients with sepsis. Crit. Care Med. 36 , 2316–2321. https://doi.org/10.1097/CCM.0b013e3181810378 (2008). Roberts, G. W. et al. Relative hyperglycemia, a marker of critical illness: introducing the stress hyperglycemia ratio. J. Clin. Endocrinol. Metab. 100 , 4490–4497. https://doi.org/10.1210/jc.2015-2660 (2015). Yan, F. et al. 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. Engl. 23 , 163. https://doi.org/10.1186/s12933-024-02265-4 (2024). Wei, X. et al. Risk analysis of the association between different hemoglobin glycation index and poor prognosis in critical patients with coronary heart disease-a study based on the MIMIC-IV database. Cardiovasc. Diabetol. 23 , 113. https://doi.org/10.1186/s12933-024-02206-1 (2024). Hempe, J. M. et al. The hemoglobin glycation index identifies subpopulations with harms or benefits from intensive treatment in the ACCORD trial. Diabetes Care . 38 , 1067–1074. https://doi.org/10.2337/dc14-1844 (2015). Johnson, A. E. W. et al. MIMIC-IV, a freely accessible electronic health record dataset. Sci. Data . 10 , 1. https://doi.org/10.1038/s41597-022-01899-x (2023). Moukheiber, D. et al. Northwestern ICU (NWICU) database [Internet]. PhysioNet; [cited 2025 Sep 20]. https://doi.org/10.13026/S84W-1829 Singer, M. et al. The third international consensus definitions for sepsis and septic shock (sepsis-3). JAMA 315 , 801. https://doi.org/10.1001/jama.2016.0287 (2016). Chun, K-H. et al. In-hospital glycemic variability and all-cause mortality among patients hospitalized for acute heart failure. Cardiovasc. Diabetol. 21 , 291. https://doi.org/10.1186/s12933-022-01720-4 (2022). Hempe, J. M., Gomez, R., McCarter, R. J. & Chalew, S. A. High and low hemoglobin glycation phenotypes in type 1 diabetes. J. Diabetes Complications . 16 , 313–320. https://doi.org/10.1016/S1056-8727(01)00227-6 (2002). Charlson, M. E., Pompei, P., Ales, K. L. & MacKenzie, C. R. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J. Chronic Dis. 40 , 373–383. https://doi.org/10.1016/0021-9681(87)90171-8 (1987). Johnson, A. E. W., Kramer, A. A. & Clifford, G. D. A new severity of illness scale using a subset of acute physiology and chronic health evaluation data elements shows comparable predictive accuracy*. Crit. Care Med. 41 , 1711–1718. https://doi.org/10.1097/CCM.0b013e31828a24fe (2013). The, S. O. F. A. sepsis-related organ failure assessment) score to describe organ dysfunction/failure. Intensive Care Med. 22 , 707–710 (1996). Wang, F. et al. Combined assessment of stress hyperglycemia ratio and glycemic variability to predict all-cause mortality in critically ill patients with atherosclerotic cardiovascular diseases across different glucose metabolic states: an observational cohort study with machine learning. Cardiovasc. Diabetol. 24 , 199. https://doi.org/10.1186/s12933-025-02762-0 (2025). Bomrah, S. et al. A scoping review of machine learning for sepsis prediction- feature engineering strategies and model performance: a step towards explainability. Crit. Care . 28 , 180. https://doi.org/10.1186/s13054-024-04948-6 (2024). Qi, L. et al. Association of glycemic variability and prognosis in patients with traumatic brain injury: a retrospective study from the MIMIC-IV database. Diabetes Res. Clin. Pract. 217 , 111869. https://doi.org/10.1016/j.diabres.2024.111869 (2024). Lucci, C. et al. Hypoglycemia and glycemic variability in acute myocardial infarction: the lesser-known aspects of glycemic control. Cardiovasc. Diabetol. 24 , 309. https://doi.org/10.1186/s12933-025-02862-x (2025). Zhou, Z., Zhang, H., Gu, Y., Zhang, K. & Ouyang, C. Relationship between glycemic variability and the incidence of postoperative atrial fibrillation following cardiac surgery: a retrospective study from MIMIC-IV database. Diabetes Res. Clin. Pract. Irel. 219 , 111978. https://doi.org/10.1016/j.diabres.2024.111978 (2024). Su, Y., Fan, W., Liu, Y. & Hong, K. Glycemic variability and in-hospital death of critically ill patients and the role of ventricular arrhythmias. Cardiovasc. Diabetol. Engl. 22 , 134. https://doi.org/10.1186/s12933-023-01861-0 (2023). Marik, P. E. & Bellomo, R. Stress hyperglycemia: an essential survival response! Crit. Care . 17 , 305 (2013). Yu, Q. et al. Impact of glycemic control metrics on short- and long-term mortality in transcatheter aortic valve replacement patients: a retrospective cohort study from the MIMIC-IV database. Cardiovasc. Diabetol. Engl. 24 , 135. https://doi.org/10.1186/s12933-025-02684-x (2025). Piagnerelli, M., Boudjeltia, K. Z., Vanhaeverbeek, M. & Vincent, J-L. Red blood cell rheology in sepsis. Intensive Care Med. 29 , 1052–1061. https://doi.org/10.1007/s00134-003-1783-2 (2003). Rawal, G., Kumar, R., Yadav, S. & Singh, A. Anemia in intensive care: a review of current concepts. J. Crit. Care Med. 2 , 109–114. https://doi.org/10.1515/jccm-2016-0017 (2016). Tables Table 1 and 2 are available in the Supplementary Files section. Additional Declarations No competing interests reported. Supplementary Files Table1.xlsx Table2.xlsx Supplementary.docx GraphicalAbstract.png Cite Share Download PDF Status: Under Revision Version 1 posted Editorial decision: Revision requested 07 Jan, 2026 Reviews received at journal 26 Dec, 2025 Reviews received at journal 24 Dec, 2025 Reviewers agreed at journal 16 Dec, 2025 Reviews received at journal 14 Dec, 2025 Reviewers agreed at journal 14 Dec, 2025 Reviewers agreed at journal 11 Dec, 2025 Reviewers agreed at journal 11 Dec, 2025 Reviewers agreed at journal 10 Dec, 2025 Reviewers invited by journal 10 Dec, 2025 Editor invited by journal 04 Dec, 2025 Editor assigned by journal 02 Dec, 2025 Submission checks completed at journal 02 Dec, 2025 First submitted to journal 01 Dec, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8247079","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":560619787,"identity":"55a33bef-a693-487e-841b-cf36cafba059","order_by":0,"name":"kangxing wang","email":"","orcid":"","institution":"West China Hospital of Sichuan University","correspondingAuthor":false,"prefix":"","firstName":"kangxing","middleName":"","lastName":"wang","suffix":""},{"id":560619791,"identity":"04e6b596-bd4f-4276-ac63-a8e2019a1157","order_by":1,"name":"Huaiyu Xiong","email":"","orcid":"","institution":"West China Hospital of Sichuan University","correspondingAuthor":false,"prefix":"","firstName":"Huaiyu","middleName":"","lastName":"Xiong","suffix":""},{"id":560619794,"identity":"34054788-f63e-4d76-a0b7-0be7fc32ee3b","order_by":2,"name":"Yukun Zhu","email":"","orcid":"","institution":"West China Hospital of Sichuan University","correspondingAuthor":false,"prefix":"","firstName":"Yukun","middleName":"","lastName":"Zhu","suffix":""},{"id":560619796,"identity":"cc8cc0a1-e9e0-471b-8a43-228e08798421","order_by":3,"name":"Yongfang Zhou","email":"","orcid":"","institution":"West China Hospital of Sichuan University","correspondingAuthor":false,"prefix":"","firstName":"Yongfang","middleName":"","lastName":"Zhou","suffix":""},{"id":560619798,"identity":"6dd9d3fc-b24b-4890-b7e7-e820f353f341","order_by":4,"name":"Yan Kang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA0klEQVRIiWNgGAWjYBACCSgtB6HYSNBiTLqWxAaitUi2nz38mqfmTvr89jMGDB/KDjPwz27Ar0WaJy/NmufYs9wNZ3IMGGecO8wgcecAfi1yDDlmxjxsh3M3SPAYMPO2HWYwkEggoIX/DVDLv8Pp8jOAWv4So0VaIsf4MdDwBIYbQC2MxGiRnPHGjHFu32HDDWfSCg72nEvnkbhBQIvE+RzjD2++HZaXbz+88cGPMms5/hkEtAABmxQPlHUAiHnwqIQD5o8/iFE2CkbBKBgFIxcAAHf0QIqxyCiqAAAAAElFTkSuQmCC","orcid":"","institution":"West China Hospital of Sichuan University","correspondingAuthor":true,"prefix":"","firstName":"Yan","middleName":"","lastName":"Kang","suffix":""}],"badges":[],"createdAt":"2025-12-01 07:08:42","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8247079/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8247079/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":98376003,"identity":"94433d62-ac7a-4de8-8305-40eb252fb54e","added_by":"auto","created_at":"2025-12-17 07:04:05","extension":"png","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":140465,"visible":true,"origin":"","legend":"","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/49fe77a28002bd25f60b6623.png"},{"id":98440827,"identity":"75b0f8dc-149c-43cb-9ce9-231525f354a0","added_by":"auto","created_at":"2025-12-17 17:04:25","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":73149,"visible":true,"origin":"","legend":"","description":"","filename":"article.docx","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/d05040ac2f3387a4a6fec19e.docx"},{"id":98376008,"identity":"444462f5-02d7-45e7-8df6-550cfb40f15c","added_by":"auto","created_at":"2025-12-17 07:04:05","extension":"png","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1415402,"visible":true,"origin":"","legend":"","description":"","filename":"Figure2.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/a464750500e089bc8706a901.png"},{"id":98376007,"identity":"47ae4e71-d4fb-468a-bf4b-362f4dd392fe","added_by":"auto","created_at":"2025-12-17 07:04:05","extension":"xlsx","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":12915,"visible":true,"origin":"","legend":"","description":"","filename":"Table1.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/c8c00e7eb499f21786c68fbe.xlsx"},{"id":98439200,"identity":"e6720bf1-29e0-4605-b6a2-43328fb7e46c","added_by":"auto","created_at":"2025-12-17 17:01:27","extension":"png","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3975783,"visible":true,"origin":"","legend":"","description":"","filename":"Figure3.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/d1f52c77a2a23b4bea358783.png"},{"id":98439879,"identity":"f5f448e9-3755-46c8-bb71-a904b1d153b4","added_by":"auto","created_at":"2025-12-17 17:03:02","extension":"xlsx","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":10465,"visible":true,"origin":"","legend":"","description":"","filename":"Table2.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/a3115aafaf25f797c24aaf78.xlsx"},{"id":98441210,"identity":"f7ba5ec5-1367-409b-aef6-e6945f23166f","added_by":"auto","created_at":"2025-12-17 17:05:04","extension":"png","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":454984,"visible":true,"origin":"","legend":"","description":"","filename":"Figure4.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/fab0df6ee060b1a694498be6.png"},{"id":98376021,"identity":"14193a5d-f574-4c56-9898-6d0d599c537d","added_by":"auto","created_at":"2025-12-17 07:04:05","extension":"png","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1710280,"visible":true,"origin":"","legend":"","description":"","filename":"Figure5.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/04d83ba870bc513dea126766.png"},{"id":98376019,"identity":"1d299167-ead7-4bfd-8db0-3700e2860d8a","added_by":"auto","created_at":"2025-12-17 07:04:05","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1920012,"visible":true,"origin":"","legend":"","description":"","filename":"Figure6.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/768ccc61b3be27ac37cbd900.png"},{"id":98438842,"identity":"fe38ca07-cef5-450e-8e56-d58d4165eb8e","added_by":"auto","created_at":"2025-12-17 17:00:09","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5190465,"visible":true,"origin":"","legend":"","description":"","filename":"GraphicalAbstract.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/6987f8e41e630fc368d1e970.png"},{"id":98440866,"identity":"a77506d4-cf1b-499c-9444-b20d46287437","added_by":"auto","created_at":"2025-12-17 17:04:31","extension":"json","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":7904,"visible":true,"origin":"","legend":"","description":"","filename":"ec3c1c9078d04dfd8303f54a5e1cd034.json","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/895795631969101f1452d5eb.json"},{"id":98440588,"identity":"3ebc4fec-73ca-487f-b0ed-d20c97523c88","added_by":"auto","created_at":"2025-12-17 17:04:04","extension":"docx","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":821437,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementary.docx","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/7c6e0bbd09ee34afb9fd3286.docx"},{"id":98441276,"identity":"921d7e71-14e0-4578-9c53-b4f876749bf5","added_by":"auto","created_at":"2025-12-17 17:05:07","extension":"xml","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":101246,"visible":true,"origin":"","legend":"","description":"","filename":"ec3c1c9078d04dfd8303f54a5e1cd0341enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/b6f4b44d0201588579e0030c.xml"},{"id":98440022,"identity":"c15d57da-21c2-40e1-b049-5d8c9136cc9f","added_by":"auto","created_at":"2025-12-17 17:03:14","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":140465,"visible":true,"origin":"","legend":"","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/f52776837b6ea44300a806d4.png"},{"id":98376025,"identity":"f485136b-231d-4137-8527-cd9eed34a626","added_by":"auto","created_at":"2025-12-17 07:04:05","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1415402,"visible":true,"origin":"","legend":"","description":"","filename":"Figure2.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/2b6d2e1f34fd68c326a7f7f6.png"},{"id":98376035,"identity":"df5341c5-05fd-4b35-9c44-175e54c10339","added_by":"auto","created_at":"2025-12-17 07:04:05","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3975783,"visible":true,"origin":"","legend":"","description":"","filename":"Figure3.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/7bcaebc22546b8757c1dd4ff.png"},{"id":98376032,"identity":"ce64e829-b4a4-4e94-8853-102235836c37","added_by":"auto","created_at":"2025-12-17 07:04:05","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":454984,"visible":true,"origin":"","legend":"","description":"","filename":"Figure4.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/13bd70ade1128377a3ae718a.png"},{"id":98376028,"identity":"70e562ad-637e-4cec-8542-8af17ebf0a19","added_by":"auto","created_at":"2025-12-17 07:04:05","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1710280,"visible":true,"origin":"","legend":"","description":"","filename":"Figure5.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/a2b0ed94b18e834f13353804.png"},{"id":98439827,"identity":"c1d32833-5b37-4e85-bcef-f3eabbf57907","added_by":"auto","created_at":"2025-12-17 17:02:59","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1920012,"visible":true,"origin":"","legend":"","description":"","filename":"Figure6.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/30add42d20b748e5115f0fb5.png"},{"id":98376034,"identity":"33748e05-66f4-49cd-a00c-fe33adc98159","added_by":"auto","created_at":"2025-12-17 07:04:05","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5190465,"visible":true,"origin":"","legend":"","description":"","filename":"GraphicalAbstract.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/50e559db178ec4249dda0221.png"},{"id":98376023,"identity":"908f7baa-be3c-46cf-a882-d42c6188476c","added_by":"auto","created_at":"2025-12-17 07:04:05","extension":"png","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":98895,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFigure1.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/aa7a799d081c9843127b350a.png"},{"id":98441176,"identity":"9c0a5731-e3cb-49ce-ac7b-2223545a1822","added_by":"auto","created_at":"2025-12-17 17:05:01","extension":"png","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":553186,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFigure2.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/a2af0029cdf45a413a9cdaf4.png"},{"id":98376040,"identity":"7354e204-8016-4ab6-a003-74193c49d82d","added_by":"auto","created_at":"2025-12-17 07:04:06","extension":"png","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":726892,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFigure3.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/a2e6da8b692297d1307e0f51.png"},{"id":98440608,"identity":"1ffa38db-6ef9-443e-bca9-328881af820c","added_by":"auto","created_at":"2025-12-17 17:04:06","extension":"png","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":160508,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFigure4.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/ff43c54b2d3b721bf1533fe0.png"},{"id":98376031,"identity":"0ba5980a-45e6-4fd5-97d1-35eb8947d20c","added_by":"auto","created_at":"2025-12-17 07:04:05","extension":"png","order_by":24,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":452022,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFigure5.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/450ea3f425791a3aa5ede273.png"},{"id":98440636,"identity":"6b6d0baf-4995-4c75-9233-f918923140ea","added_by":"auto","created_at":"2025-12-17 17:04:07","extension":"png","order_by":25,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":488648,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFigure6.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/d11188d6ab51ca96a8d834a4.png"},{"id":98439877,"identity":"544358e8-c47c-49c2-affa-77937167f627","added_by":"auto","created_at":"2025-12-17 17:03:02","extension":"png","order_by":26,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1580488,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineGraphicalAbstract.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/24a3c787ddeadf93ec0c2e12.png"},{"id":98376036,"identity":"4e7d0cec-10c1-4826-ba57-a0ba13379d93","added_by":"auto","created_at":"2025-12-17 07:04:05","extension":"xml","order_by":27,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":100686,"visible":true,"origin":"","legend":"","description":"","filename":"ec3c1c9078d04dfd8303f54a5e1cd0341structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/5a61a260e07165bf5eba5bda.xml"},{"id":98376038,"identity":"c2040d29-768b-4d6b-a3b0-9a10b5203165","added_by":"auto","created_at":"2025-12-17 07:04:06","extension":"html","order_by":28,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":114264,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/d785e650e68ea6d1502b1cfa.html"},{"id":98376002,"identity":"34866bdd-ac6b-45a6-9e70-0c2b2ffe3b99","added_by":"auto","created_at":"2025-12-17 07:04:05","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":140465,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart of patient selection.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAbbreviations: \u003c/em\u003eFBG, fasting blood glucose; ICU, intensive care unit; MIMIC-IV, Medical Information Mart for Intensive Care IV; n, number of patients; NWICU, Northwestern Intensive Care Unit.\u003c/p\u003e","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/4c696fdea8bf9ef4fcdc48dd.png"},{"id":98376005,"identity":"0982522f-7ef6-4d21-8bf7-6ba9aa7c044a","added_by":"auto","created_at":"2025-12-17 07:04:05","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":1415402,"visible":true,"origin":"","legend":"\u003cp\u003eDose-response relationship and Kaplan-Meier survival curves for the association of GV, SHR, and HGI with 28-day all-cause mortality.\u003c/p\u003e\n\u003cp\u003e(A-C) Restricted cubic spline (RCS) plots showing the adjusted hazard ratios (HRs) for 28-day mortality according to the continuous values of GV, SHR, and HGI. The solid red line represents the HR, and the shaded area represents the 95% confidence interval. The horizontal dashed line is set at an HR of 1.0. The histograms show the distribution of each glycemic metric in the study population.\u003c/p\u003e\n\u003cp\u003e(D-F) Kaplan-Meier survival curves for 28-day mortality, stratified by quintiles of GV, SHR, and HGI. P-values were calculated using the log-rank test. The tables below the curves indicate the number of patients at risk at different time points.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAbbreviations:\u003c/em\u003eGV, Glycemic Variability; HGI, Hemoglobin Glycation Index; HR, hazard ratio; RCS, restricted cubic spline; SHR, Stress Hyperglycemia Ratio.\u003c/p\u003e","description":"","filename":"Figure2.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/d338619c38c4a84fa468566f.png"},{"id":98376011,"identity":"4427be10-4080-484d-b1f7-c83daa8c4053","added_by":"auto","created_at":"2025-12-17 07:04:05","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":3975783,"visible":true,"origin":"","legend":"\u003cp\u003eSubgroup analyses for the association between the highest versus lowest quintiles of GV, SHR, and HGI and 28-day all-cause mortality.\u003c/p\u003e\n\u003cp\u003eThe forest plots show the hazard ratios (HRs) and 95% confidence intervals (CIs) for the association of the highest quintile (Q5) of each glycemic metric with 28-day mortality compared to the lowest quintile (Q1) across various subgroups. The analyses were adjusted for all covariates included in Model 3, except for the stratifying variable itself. The \u003cem\u003eP\u003c/em\u003e for interaction was calculated to assess for effect modification by the subgroup variable.(A) Subgroup analysis for Glycemic Variability (GV).(B) Subgroup analysis for Stress Hyperglycemia Ratio (SHR).(C) Subgroup analysis for Hemoglobin Glycation Index (HGI).\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAbbreviations: \u003c/em\u003eCI, confidence interval; GV, Glycemic Variability; HGI, Hemoglobin Glycation Index; HF, heart failure; HR, hazard ratio; Q, quintile; SHR, Stress Hyperglycemia Ratio; SOFA, Sequential Organ Failure Assessment; Vent, mechanical ventilation; Vaso, vasopressor use.\u003c/p\u003e","description":"","filename":"Figure3.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/acfc0b441de6cf73ae6ce569.png"},{"id":98376018,"identity":"a792adf7-4a53-4d1d-a4b1-63510483a0a9","added_by":"auto","created_at":"2025-12-17 07:04:05","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":454984,"visible":true,"origin":"","legend":"\u003cp\u003eMediation analysis of lactate in the association between glycemic metrics and 28-day mortality.\u003c/p\u003e\n\u003cp\u003eThe path diagrams illustrate the mediating effect of lactate on the relationship between each glycemic metric and 28-day mortality under different adjustment models. The analyses were performed for (A-C) Glycemic Variability (GV), (D-F) Stress Hyperglycemia Ratio (SHR), and (G-I) Hemoglobin Glycation Index (HGI). The models correspond to the hierarchical adjustments used in the Cox regression analysis: Model 1 (A, D, G) is unadjusted; Model 2 (B, E, H) is partially adjusted for age, gender, and race; and Model 3 (C, F, I) is the fully adjusted model.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAbbreviations: \u003c/em\u003eACME, average causal mediation effect (indirect effect); ADE, average direct effect; GV, Glycemic Variability; HGI, Hemoglobin Glycation Index; SHR, Stress Hyperglycemia Ratio.\u003c/p\u003e","description":"","filename":"Figure4.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/31e53b7a41dadae64184210e.png"},{"id":98376020,"identity":"229888cd-025f-4d6d-bb89-2912615cc0f9","added_by":"auto","created_at":"2025-12-17 07:04:05","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":1710280,"visible":true,"origin":"","legend":"\u003cp\u003eFeature selection process using LASSO regression and the Boruta algorithm.\u003c/p\u003e\n\u003cp\u003ePanels (A) and (B) show the variable selection process using the LASSO model. Panels (C) and (D) illustrate the feature importance ranking and selection by the Boruta algorithm. The final set of predictors for model development was formed by taking the union of features identified as important by both methods.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAbbreviations: \u003c/em\u003eLASSO, Least Absolute Shrinkage and Selection Operator.\u003c/p\u003e","description":"","filename":"Figure5.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/795a1e534e21af165a009d67.png"},{"id":98376014,"identity":"ee8fdf91-68cd-4386-9ce6-f4c6025ad349","added_by":"auto","created_at":"2025-12-17 07:04:05","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":1920012,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance and interpretation of the machine learning models.\u003c/p\u003e\n\u003cp\u003e(A) Receiver Operating Characteristic (ROC) curves for various machine learning models compared to the SOFA and OASIS scores on the validation set. The Area Under the Curve (AUC) for each model is provided in the legend. (B) SHAP feature importance plot for the best-performing XGBoost model. Features are ranked by the mean absolute SHAP value. (C) SHAP summary plot (beeswarm plot). Each dot represents a patient for a given feature. The color indicates the feature's value (high or low), and the position on the x-axis indicates the SHAP value's impact on the model's prediction (positive values increase the predicted risk of death). (D) SHAP force plot for a single representative patient prediction, illustrating how individual features contribute to push the model's prediction from the base value to the final output.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eAbbreviations: \u003c/em\u003eAUC, Area Under the Curve; OASIS, Oxford Acute Severity of Illness Score; SHAP, SHapley Additive exPlanations; SOFA, Sequential Organ Failure Assessment.\u003c/p\u003e","description":"","filename":"Figure6.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/b2735cea1e53459e6230aad9.png"},{"id":98445852,"identity":"0c09b6e1-412c-47b7-a5ae-4068fd61714e","added_by":"auto","created_at":"2025-12-17 17:22:01","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":10346082,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/e0bce613-4e46-4206-b71e-227c3b14dea4.pdf"},{"id":98440003,"identity":"6fe5b792-933e-4586-beb9-b1d98389a228","added_by":"auto","created_at":"2025-12-17 17:03:12","extension":"xlsx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":12915,"visible":true,"origin":"","legend":"","description":"","filename":"Table1.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/6e654dd56405d763f7e30385.xlsx"},{"id":98376010,"identity":"b9c23d64-1355-422d-8c50-f7f3bf459f70","added_by":"auto","created_at":"2025-12-17 07:04:05","extension":"xlsx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":10465,"visible":true,"origin":"","legend":"","description":"","filename":"Table2.xlsx","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/aec14cb559388f511c818b67.xlsx"},{"id":98439756,"identity":"e2aad569-4f81-49a1-a65a-378a45b7ccda","added_by":"auto","created_at":"2025-12-17 17:02:53","extension":"docx","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":821437,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementary.docx","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/a403d92e321b2c80f666c04f.docx"},{"id":98376022,"identity":"3f2998bc-9f2c-4529-836a-deaffa88f1e9","added_by":"auto","created_at":"2025-12-17 07:04:05","extension":"png","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":5190465,"visible":true,"origin":"","legend":"","description":"","filename":"GraphicalAbstract.png","url":"https://assets-eu.researchsquare.com/files/rs-8247079/v1/4abca1b1e5f2c37cc3d92d5c.png"}],"financialInterests":"No competing interests reported.","formattedTitle":"Association of Glycemic Variability, Stress Hyperglycemia Ratio, and Hemoglobin Glycation Index With 28-Day Mortality in Sepsis: A Multicenter Retrospective Study With Mediation and Machine-Learning Analyses","fulltext":[{"header":"Introduction","content":"\u003cp\u003eSepsis, a syndrome of life-threatening organ dysfunction precipitated by a dysregulated host response to infection, represents a formidable global health burden and remains a leading cause of mortality in intensive care units (ICUs) [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Central to the pathophysiology of sepsis are profound metabolic derangements, of which dysglycemia\u0026mdash;encompassing hyperglycemia, hypoglycemia, and increased glycemic lability\u0026mdash;is a near-universal and particularly pernicious feature [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. This disruption of glycemic homeostasis is driven by a complex interplay between systemic inflammation, counter-regulatory hormone excess, and severe insulin resistance, creating a vicious cycle that exacerbates organ injury. A robust body of evidence has firmly established that the severity of dysglycemia is independently associated with increased mortality and organ failure in sepsis, thereby highlighting the critical importance of sophisticated glycemic monitoring and management in this vulnerable population [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eTo address the complexity of dysglycemia in sepsis, the field has shifted beyond single-point glucose measurements toward a multidimensional assessment paradigm aimed at capturing the distinct pathophysiological dimensions of glycemic dyshomeostasis. Glycemic Variability (GV), for instance, quantifies the dynamic instability of glucose levels. This lability has been shown to independently drive adverse clinical outcomes through mechanisms such as exacerbated oxidative stress and endothelial injury [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Complementing this dynamic assessment, the Stress Hyperglycemia Ratio (SHR) isolates the intensity of the pure \"stress\" component of acute hyperglycemia by normalizing it against the patient's chronic glycemic background, a metric validated for its prognostic significance across diverse critically ill populations [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Transcending these short-term and acute-stress dimensions, the Hemoglobin Glycation Index (HGI) offers a deeper perspective by revealing the intrinsic discordance between an individual's long-term (via HbA1c) and short-term (via fasting glucose) glycemic control. This incongruity, potentially reflecting idiosyncratic traits such as non-enzymatic glycation rates or red blood cell lifespan, has emerged as a novel biomarker with promise for discerning distinct risk phenotypes [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eWhile Glycemic Variability (GV), the Stress Hyperglycemia Ratio (SHR), and the Hemoglobin Glycation Index (HGI) each offer a unique perspective on the glycemic dysregulation in sepsis\u0026mdash;reflecting short-term lability, acute stress, and chronic adaptation, respectively\u0026mdash;they have predominantly been investigated in isolation. To date, no study has comprehensively compared their prognostic utility head-to-head within the same large, multi-center cohort of patients with sepsis. Furthermore, the relative importance and potential interplay of these factors within advanced predictive models that integrate multidimensional clinical data remain unexplored, limiting a comprehensive understanding of which glycemic profiles confer the highest risk. Therefore, this study aimed to systematically evaluate and compare the independent prognostic value of GV, SHR, and HGI for 28-day mortality in patients with sepsis, and to determine their roles in a machine learning-based predictive model.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eData source:\u003c/h2\u003e \u003cp\u003eThe data utilized in this study were derived from two large, publicly available critical care databases. The first dataset was the Medical Information Mart for Intensive Care IV (MIMIC-IV, v3.1), an electronic health record dataset comprising patients admitted to Beth Israel Deaconess Medical Center in Boston, MA, from 2008 to 2022 [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. The second dataset was the Northwestern ICU (NWICU, v0.1.0) database, which contains data from multiple hospitals within the Northwestern Medicine network in Chicago, IL, from 2020 to 2022 [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. The NWICU database is structurally harmonized with MIMIC-IV, facilitating the integration of data from both sources for this multi-center retrospective analysis. The Institutional Review Boards of both originating institutions approved the data collection and sharing initiatives, granting a waiver of informed consent. An author of this study (Certification number: 13024213) obtained certified access to use these databases.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eStudy population:\u003c/h3\u003e\n\u003cp\u003eWe included adult patients (aged\u0026thinsp;\u0026ge;\u0026thinsp;18 years) admitted to the ICU for the first time with a diagnosis of sepsis from both the MIMIC-IV and NWICU databases. Sepsis was defined according to the Third International Consensus Definitions for Sepsis and Septic Shock (Sepsis-3) criteria, requiring a suspected infection and a sequential organ failure assessment (SOFA) score increase of \u0026ge;\u0026thinsp;2 points [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eInitially, a total of 48,175 patients with sepsis were identified, comprising 41,295 from MIMIC-IV and 6,880 from NWICU. Patients were subsequently excluded based on the following criteria: (1) ICU length of stay less than 24 hours (n\u0026thinsp;=\u0026thinsp;5,186); (2) not the first ICU admission (n\u0026thinsp;=\u0026thinsp;4,020); (3) fewer than three blood glucose measurements recorded during the ICU stay (n\u0026thinsp;=\u0026thinsp;4,149); and (4) missing data for hemoglobin A1c or fasting blood glucose required for calculating exposure variables (n\u0026thinsp;=\u0026thinsp;27,560). After applying these criteria, a final analytical sample of 7,260 patients was included for analysis (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\n\u003ch3\u003eData extraction and definitions:\u003c/h3\u003e\n\u003cp\u003eWe used Structured Query Language (SQL) with PostgreSQL to extract all relevant data from the MIMIC-IV and NWICU databases. Data for baseline characteristics, severity scores, and laboratory results required for calculating SHR and HGI were extracted from the first 24 hours of ICU admission. Blood glucose measurements for the calculation of GV were collected throughout the entire ICU stay. The extracted data included demographic information (age, gender, race, BMI), severity of illness scores (SOFA, OASIS, Charlson Comorbidity Index), vital signs (mean arterial pressure, heart rate, respiratory rate, temperature), laboratory results (hemoglobin (Hb), white blood cell count (WBC), platelet count (Plt), lactate (Lac), total bilirubin (Tbil), aspartate aminotransferase (AST), alanine aminotransferase (ALT), and creatinine (Cre)), and details on interventions such as mechanical ventilation, renal replacement therapy, and vasopressor use. Comorbidities, including hypertension, diabetes, COPD, heart failure, and stroke, were identified using International Classification of Diseases, Ninth and Tenth Revision (ICD-9/10) codes. The primary outcome was 28-day all-cause mortality, ascertained from hospital records and the Social Security Death Index.\u003c/p\u003e \u003cp\u003eThe primary exposure variables were defined as follows. Glycemic variability (GV) was calculated as the coefficient of variation (standard deviation/mean \u0026times; 100%) of all blood glucose measurements during the entire ICU stay [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. The stress hyperglycemia ratio (SHR) was calculated using the formula: admission glucose (mg/dL) / (28.7 \u0026times; HbA1c [%] \u0026ndash; 46.7) [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. The hemoglobin glycation index (HGI) was calculated as the difference between the measured HbA1c and the predicted HbA1c. To account for potential differences between the two source databases, the predicted HbA1c was derived from separate linear regression models fitted within each cohort based on the patient's first fasting blood glucose (FBG), following a previously established methodology [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]. The resulting predictive equations were predicted HbA1c (%)\u0026thinsp;=\u0026thinsp;0.0078 \u0026times; FBG (mg/dL)\u0026thinsp;+\u0026thinsp;5.2165 for the MIMIC-IV cohort, and predicted HbA1c (%)\u0026thinsp;=\u0026thinsp;0.0085 \u0026times; FBG (mg/dL)\u0026thinsp;+\u0026thinsp;5.2137 for the NWICU cohort. The correlations between HGI and HbA1c within each cohort are shown in Supplementary Figure \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e.\u003c/p\u003e\n\u003ch3\u003eStudy endpoint:\u003c/h3\u003e\n\u003cp\u003eThe primary outcome of interest was 28-day all-cause mortality. This was defined as death from any cause occurring within 28 days following the date and time of ICU admission. The vital status of each patient at day 28 was determined using electronic health records for in-hospital deaths and supplemented by data from the Social Security Death Index for out-of-hospital deaths.\u003c/p\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003eStatistical analysis:\u003c/h2\u003e \u003cp\u003eFor our primary analyses, Glycemic Variability (GV), Stress Hyperglycemia Ratio (SHR), and Hemoglobin Glycation Index (HGI) were categorized into quintiles based on their distributions in the study cohort. In the baseline data analysis, continuous variables were expressed as mean (standard deviation) or median (interquartile range) and compared using t-tests or Mann-Whitney U tests as appropriate. Categorical variables were reported as frequencies (percentages) and compared using the chi-square or Fisher's exact test. For survival analysis, Kaplan-Meier curves were plotted to illustrate 28-day survival trends across the quintiles of GV, SHR, and HGI, with differences assessed by the log-rank test. We then conducted Cox proportional hazards regression analyses to further elucidate the relationship between each glycemic metric and 28-day all-cause mortality. To account for potential confounders, we applied multivariate Cox regression with hierarchical adjustments. Model 1 provided unadjusted estimates. Model 2 was partially adjusted for age, gender, and race. Model 3 was fully adjusted for all covariates listed in Model 2 plus BMI, severity of illness scores (SOFA, OASIS, Charlson Comorbidity Index) [\u003cspan additionalcitationids=\"CR17\" citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e], comorbidities, vital signs, laboratory results, interventions and database source. In addition, restricted cubic spline (RCS) plots with three knots were used to assess the linear or nonlinear relationships between the continuous values of GV, SHR, and HGI and clinical outcomes. To determine whether these associations differed among various populations, we performed subgroup analyses based on key clinical characteristics and calculated p-values for interaction. The results of the subgroup analyses were visualized using forest plots. A mediation analysis was then performed to investigate the extent to which lactate levels mediated the effects of the glycemic metrics on 28-day mortality. Finally, to evaluate the predictive importance of our glycemic metrics and build a precise prediction model, we designed a systematic machine learning pipeline. First, we performed feature selection by taking the union of variables identified by both the Least Absolute Shrinkage and Selection Operator (LASSO) regression and the Boruta algorithm to account for both linear and non-linear associations. Based on this selected feature set, we then developed and compared twelve distinct machine learning models: Extreme Gradient Boosting (XGBoost), Bayesian Additive Regression Trees (BART), Gradient Boosting Machine (GBM), Random Forest, Light Gradient Boosting Machine (LightGBM), Logistic Regression, Gradient Boosting with Linear Models (GLMBoost), Neural Network (Nnet), Support Vector Machine (SVM), Na\u0026iuml;ve Bayes, Recursive Partitioning and Regression Trees (Rpart), and K-Nearest Neighbors (K-NN). The predictive performance of each model was evaluated on the validation set using the area under the receiver operating characteristic curve (AUC) [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Covariates with a missing data rate of less than 40% were included, and any remaining missing values were handled using multiple imputation by chained equations (MICE). To verify the robustness of our findings regarding the missing data handling, we performed a sensitivity analysis by repeating the multivariable Cox regression models using the Random Forest algorithm for imputation. All statistical analyses were performed using R software (version 4.4.2), and a two-sided p-value of less than 0.05 was considered statistically significant.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003eBaseline characteristics:\u003c/h2\u003e \u003cp\u003eA total of 7,260 patients with sepsis from the combined databases were included, with baseline characteristics stratified by 28-day mortality presented in Table\u0026nbsp;1. Compared to survivors, non-survivors were significantly older and had a lower body mass index (BMI). They presented with a greater severity of illness upon admission, as evidenced by significantly higher SOFA, OASIS, and Charlson Comorbidity Index scores. In terms of comorbidities, non-survivors had a higher prevalence of COPD and stroke, alongside more unstable vital signs, including elevated heart and respiratory rates. Laboratory findings revealed that non-survivors had significantly increased levels of lactate, creatinine, and white blood cells. Regarding glucose metabolism, although there were no significant differences in the prevalence of diabetes or baseline HbA1c levels between the groups, non-survivors exhibited significantly higher admission fasting blood glucose (FBG) levels. Crucially, non-survivors displayed significantly higher glucose variability (GV) and stress hyperglycemia ratio (SHR), but a lower hemoglobin glycation index (HGI) (all p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). regarding interventions, non-survivors were more likely to receive renal replacement therapy (RRT) but, notably, were less frequently treated with insulin or vasopressors compared to survivors.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eSurvival analysis:\u003c/h3\u003e\n\u003cp\u003eTo assess the independent prognostic value of GV, SHR, and HGI for 28-day mortality, we constructed multivariable Cox proportional hazards regression models (Table\u0026nbsp;2). The analysis revealed that both Glucose Variability (GV) and the Stress Hyperglycemia Ratio (SHR) were positively and independently associated with mortality. After full adjustment for confounders (Model 3), the highest quintiles of GV (Q5 vs Q1: HR\u0026thinsp;=\u0026thinsp;1.62, 95% CI 1.33\u0026ndash;1.97) and SHR (Q5 vs Q1: HR\u0026thinsp;=\u0026thinsp;1.51, 95% CI 1.25\u0026ndash;1.82) remained strong predictors of death. In contrast, the Hemoglobin Glycation Index (HGI) demonstrated a significant protective effect, where higher levels were associated with improved survival compared to the lowest quintile. This inverse association was robust across all adjustments; specifically, in the fully adjusted model, the highest HGI quintile (Q5) remained significantly associated with reduced mortality risk compared to the lowest quintile (HR\u0026thinsp;=\u0026thinsp;0.74, 95% CI 0.61\u0026ndash;0.91). These findings indicate that high GV, high SHR, and low HGI are all independent markers of poor prognosis in patients with sepsis.\u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eAssociation with outcomes:\u003c/h2\u003e \u003cp\u003eTo further explore the dose-response relationship between GV, SHR, HGI, and 28-day mortality, we conducted adjusted restricted cubic spline (RCS) analysis (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA-C). The results revealed significant non-linear, positive associations for both Glucose Variability (GV) and the Stress Hyperglycemia Ratio (SHR) with mortality risk (P for non-linearity\u0026thinsp;\u0026lt;\u0026thinsp;0.05 and \u0026lt;\u0026thinsp;0.01, respectively). Specifically, the risk of death increased noticeably after GV exceeded a threshold of 23 and after SHR exceeded 1.08 (where HR\u0026thinsp;\u0026gt;\u0026thinsp;1). In contrast, the Hemoglobin Glycation Index (HGI) exhibited a linear, inverse association with mortality (P for non-linearity\u0026thinsp;\u0026gt;\u0026thinsp;0.05), indicating that lower HGI levels\u0026mdash;particularly below the threshold of -0.4\u0026mdash;were correlated with a progressively higher risk of death. These findings were strongly corroborated by the Kaplan-Meier survival analysis (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eD-F). The Log-rank tests were highly significant for all three variables (all P\u0026thinsp;\u0026lt;\u0026thinsp;0.01), demonstrating distinct survival probabilities across the quintiles. Patients in the highest quintile (Q5) of GV and SHR had the lowest survival rates. Conversely, for HGI, those in the lowest quintile (Q1) experienced the worst survival outcomes, perfectly aligning with the trends observed in the RCS analysis.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eSubgroup analyses:\u003c/h2\u003e \u003cp\u003eTo assess the consistency of the associations between GV, SHR, HGI, and mortality across different patient populations, we performed stratified analyses and tested for interactions based on key clinical variables, including age, gender, SOFA score, comorbidities, and interventions (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Overall, the adverse effects of high Glucose Variability (GV) and high Stress Hyperglycemia Ratio (SHR), as well as the protective effect of a high Hemoglobin Glycation Index (HGI), were largely consistent across most subgroups. However, interaction analyses revealed several significant points of heterogeneity. For GV, a significant interaction was detected with age (P for interaction\u0026thinsp;=\u0026thinsp;0.031) and vasopressor use (P for interaction\u0026thinsp;=\u0026thinsp;0.002), suggesting its risk effect might be more pronounced in younger patients. For SHR, its association with mortality was significantly modified by heart failure (P\u0026thinsp;=\u0026thinsp;0.036), mechanical ventilation (P\u0026thinsp;=\u0026thinsp;0.001), insulin use (P\u0026thinsp;=\u0026thinsp;0.001), and vasopressor use (P\u0026thinsp;=\u0026thinsp;0.001), indicating that the detrimental impact of high SHR was substantially magnified in patients without heart failure and in those requiring critical care interventions. For HGI, its protective effect was modified by diabetes (borderline interaction, P\u0026thinsp;=\u0026thinsp;0.055) and vasopressor use (P\u0026thinsp;=\u0026thinsp;0.008), implying that the benefit of a high HGI may be attenuated in patients with diabetes. No other significant interactions were observed for the remaining subgroups.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eMediation analysis:\u003c/h2\u003e \u003cp\u003eGiven that elevated lactate is a key biomarker of poor prognosis in sepsis, we performed a mediation analysis to explore whether lactate mediates the associations between GV, SHR, HGI, and 28-day mortality, with the results visualized in path diagrams (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e) and detailed in Supplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e. After full adjustment for potential confounders (Model 3), lactate was identified as a significant mediator for all three glycemic indices. Specifically, lactate mediated 12.8% (95% CI: 5.6%, 23.7%) of the total effect of Glucose Variability (GV) on mortality, 19.5% (95% CI: 9.4%, 36.2%) of the total effect of the Stress Hyperglycemia Ratio (SHR), and 7.9% (95% CI: 2.9%, 41.7%) of the total effect of the Hemoglobin Glycation Index (HGI), with all mediation effects being statistically significant (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05). Notably, the average direct effects (ADE) of GV, SHR, and HGI remained significant after accounting for lactate's mediating role across all models. This indicates that lactate functions as a partial mediator, suggesting that these glycemic indices influence patient outcomes not only through the lactate-mediated pathway but also via other independent pathophysiological mechanisms.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eMachine learning analysis:\u003c/h2\u003e \u003cp\u003eTo comprehensively evaluate the predictive importance of numerous clinical variables, including GV, SHR, and HGI, and to build a precise mortality prediction model, we designed a systematic feature selection and modeling pipeline. We first employed both LASSO regression (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eA-B) and the Boruta algorithm (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eC) to screen all potential predictors (see Supplementary Tables S2 and S3 for detailed results), taking the union of the selected features to account for both potential linear and non-linear associations and form the final feature set. The cohort was then divided into training (70%) and validation (30%) sets using stratified sampling based on the database source. Based on the selected features, multiple machine learning models were developed and compared (see Supplementary Table S4 for detailed performance metrics). ROC analysis demonstrated that the XGBoost model achieved the best predictive performance, with an AUC of 0.798 on the validation set (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e). Notably, our three glycemic management indices of interest\u0026mdash;GV, SHR, and HGI\u0026mdash;were all included in the final predictive model through this comprehensive selection process, highlighting their central role in sepsis prognostication. To interpret the top-performing XGBoost model, we utilized SHAP analysis. The SHAP feature importance plot further corroborated our findings: while traditional markers like age and hemoglobin remained significant, GV, SHR, and HGI were ranked among the most important predictors, even outperforming some conventional clinical variables. The SHAP summary plot further revealed that higher GV and SHR values drive the model to predict a higher risk of death. In contrast, higher HGI values drive a lower risk prediction, which is perfectly consistent with our survival analysis results.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eSensitivity analyses:\u003c/h2\u003e \u003cp\u003eTo verify the robustness of our findings regarding the handling of missing data, we performed a sensitivity analysis by comparing different imputation methods. While the primary analysis utilized Multiple Imputation by Chained Equations (MICE), we repeated the multivariable Cox regression models using Random Forest imputation. The results obtained from the Random Forest dataset were highly consistent with the primary analysis (Supplementary Table S5). specifically, in the fully adjusted model (Model 3), high GV (Q5 vs Q1: HR\u0026thinsp;=\u0026thinsp;1.63, 95% CI 1.34\u0026ndash;1.98) and high SHR (Q5 vs Q1: HR\u0026thinsp;=\u0026thinsp;1.50, 95% CI 1.24\u0026ndash;1.81) remained significant independent predictors of mortality. Similarly, the protective association of HGI was preserved, with the highest quintile showing a significantly reduced risk compared to the lowest (HR\u0026thinsp;=\u0026thinsp;0.74, 95% CI 0.61\u0026ndash;0.90). These findings confirm that our results are robust and not driven by the specific method of missing data imputation.\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eIn this large, bi-centric retrospective analysis, we conducted the first systematic, head-to-head comparison of three key glycemic indices\u0026mdash;Glycemic Variability (GV), the Stress Hyperglycemia Ratio (SHR), and the Hemoglobin Glycation Index (HGI)\u0026mdash;to evaluate their prognostic value in critically ill patients with sepsis. We found that high GV, high SHR, and low HGI were each independently and strongly associated with increased 28-day mortality. Using advanced, interpretable machine learning approaches, we further confirmed that these three indices were among the most influential predictors of death, even when integrated with a wide array of clinical variables. Together, they represent distinct yet complementary dimensions of dysglycemia\u0026mdash;short-term instability (GV), acute stress response (SHR), and chronic glycemic adaptation (HGI)\u0026mdash;forming a multidimensional glycemic profile that more accurately reflects the metabolic complexity of sepsis.\u003c/p\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eGlycemic Variability and Stress Hyperglycemia: Dynamic and Acute Markers of Severity\u003c/h2\u003e \u003cp\u003eOur findings reinforce prior evidence linking both glucose fluctuation and stress hyperglycemia with adverse outcomes in critical illness [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. In this large, multi-center cohort, patients in the highest quintiles of GV and SHR exhibited significantly higher mortality, independent of illness severity, comorbidities, and therapeutic interventions. These results emphasize that dysglycemia in sepsis cannot be captured by static glucose levels alone; instead, the magnitude and instability of glucose excursions, along with the degree of stress-induced metabolic response, are crucial determinants of clinical outcome.\u003c/p\u003e \u003cp\u003eMechanistically, these two indices likely reflect distinct aspects of metabolic derangement. Glycemic Variability quantifies fluctuations in glucose over time\u0026mdash;episodes of hyper- and hypoglycemia\u0026mdash;that trigger oxidative stress, mitochondrial injury, and endothelial dysfunction, thereby amplifying systemic inflammation and multi-organ failure [\u003cspan additionalcitationids=\"CR22 CR23\" citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. SHR, in contrast, reflects the disproportionate elevation of admission glucose relative to chronic glycemic background, capturing the intensity of the neuroendocrine stress response. Elevated SHR signals heightened sympathetic activation, increased cortisol and catecholamine release, and severe insulin resistance\u0026mdash;hallmarks of the metabolic storm accompanying septic shock [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eOur mediation analysis supports this interpretation: lactate partially mediated the associations between both GV and SHR and mortality (12.8% and 19.5% of total effects, respectively). Elevated lactate reflects tissue hypoperfusion and metabolic stress, suggesting that these glycemic perturbations contribute to adverse outcomes through impaired cellular energetics and augmented anaerobic metabolism. Furthermore, subgroup analyses revealed that the detrimental effects of high SHR were especially pronounced among patients requiring mechanical ventilation or vasopressor support, highlighting its role as a marker of extreme physiological stress. By contrast, GV\u0026rsquo;s adverse effects were more evident in younger patients, suggesting that glucose fluctuations may exert independent cytotoxic effects even in individuals with greater physiological reserve.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003eHemoglobin Glycation Index: A Paradoxical but Insightful Marker\u003c/h2\u003e \u003cp\u003eThe most novel finding of this study is the inverse association between HGI and mortality, where a low HGI predicted poor survival, while a high HGI was relatively protective. This pattern, consistent with reports in other critically ill populations such as coronary artery disease and post\u0026ndash;transcatheter aortic valve replacement [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e], suggests that HGI reflects intrinsic biological differences in glycemic response beyond average glucose control.\u003c/p\u003e \u003cp\u003eWe propose two potential, non-mutually exclusive explanations. The first, the \u0026ldquo;Physiological Resilience\u0026rdquo; hypothesis, posits that a high HGI identifies individuals who maintain greater tolerance to hyperglycemia-related oxidative and inflammatory stress, possibly through altered red blood cell turnover or glycation kinetics that blunt hyperglycemia-induced injury [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. The second, the \u0026ldquo;Acute Stress Amplifier\u0026rdquo; hypothesis, interprets a low HGI as a marker of disproportionate acute stress hyperglycemia. In sepsis, massive catecholamine and cortisol surges can markedly elevate fasting glucose independent of prior glycemic history. Thus, a low HGI\u0026mdash;where measured HbA1c is markedly lower than fasting glucose\u0026ndash;predicted HbA1c\u0026mdash;likely signifies an exaggerated stress response and severe metabolic disarray. In this sense, a low HGI does not denote good long-term glycemic control, but rather excessive acute hyperglycemia relative to the individual\u0026rsquo;s baseline, identifying patients with profound catabolic drive and cytokine-mediated injury [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe partial mediation of HGI\u0026rsquo;s mortality association through lactate supports this hypothesis, implying that low HGI reflects an underlying metabolic collapse characterized by both hyperglycemia and tissue hypoxia. Notably, the strength of the HGI\u0026ndash;mortality association varied slightly between the MIMIC-IV and NWICU cohorts, possibly due to differences in illness severity or glucose monitoring practices. Nevertheless, the consistent directionality of the association across cohorts underscores HGI\u0026rsquo;s robustness as a potential metabolic stress biomarker in sepsis.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003eMachine Learning and the Multidimensional Glycemic Profile\u003c/h2\u003e \u003cp\u003eOur machine learning analysis provided complementary, data-driven evidence of the prognostic importance of these glycemic indices. Among twelve tested algorithms, the XGBoost model achieved the highest discrimination (AUC\u0026thinsp;=\u0026thinsp;0.798), outperforming conventional clinical scores such as SOFA and OASIS. Importantly, SHAP (SHapley Additive Explanations) analysis provided interpretability by ranking variable contributions: GV, SHR, and HGI were among the most influential predictors of mortality, surpassing several traditional laboratory and physiological markers. The directional effects\u0026mdash;higher GV and SHR increasing mortality risk, higher HGI reducing it\u0026mdash;mirrored our regression analyses, confirming the internal coherence of these findings.\u003c/p\u003e \u003cp\u003eTogether, these results suggest that integrating multiple glycemic dimensions provides a more comprehensive assessment of metabolic dysfunction than reliance on any single index. This approach captures the interplay between chronic adaptation, acute stress, and dynamic instability\u0026mdash;three interconnected elements of glucose metabolism that collectively define the \u0026ldquo;glycemic signature\u0026rdquo; of sepsis.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003eClinical Implications\u003c/h2\u003e \u003cp\u003eThese findings have several important clinical implications. First, incorporating GV, SHR, and HGI into routine ICU monitoring could markedly improve risk stratification in septic patients. These indices are easily derived from routinely collected data\u0026mdash;inuous glucose measurements, admission glucose, fasting glucose, and HbA1c\u0026mdash;and could feasibly be integrated into automated electronic health record systems. Second, recognizing the distinct biological roles of these indices could inform individualized management strategies. Patients with extreme GV may benefit from interventions to minimize glycemic fluctuations, whereas those with high SHR or low HGI may warrant earlier metabolic resuscitation or targeted modulation of the stress response. Finally, the integration of these metrics into real-time decision-support tools\u0026mdash;potentially powered by machine learning\u0026mdash;may enable early detection of high-risk phenotypes and personalized therapeutic adjustment.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section2\"\u003e \u003ch2\u003eStrengths and Limitations\u003c/h2\u003e \u003cp\u003eThis study has several notable strengths. It leverages two large, harmonized ICU databases, enhancing generalizability across diverse patient populations. It is also the first to compare GV, SHR, and HGI simultaneously within a unified analytic framework, combining conventional regression, mediation, and interpretable machine learning to ensure robust conclusions. However, several limitations warrant acknowledgment. The retrospective design limits causal inference, and despite extensive adjustments, residual confounding cannot be excluded. The accuracy of GV depends on glucose monitoring frequency, which may vary across institutions. Moreover, HGI assumes a stable linear relationship between fasting glucose and HbA1c\u0026mdash;an assumption potentially altered during acute illness due to changes in erythrocyte lifespan and protein turnover. Finally, our findings, while internally validated, require prospective and external validation before clinical implementation.\u003c/p\u003e \u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003eIn conclusion, Glycemic Variability, the Stress Hyperglycemia Ratio, and the Hemoglobin Glycation Index are all powerful and independent prognostic markers in patients with sepsis, each capturing a unique and complementary dimension of glycemic dysregulation. High GV and high SHR are associated with increased mortality, while a high HGI is paradoxically protective. The integration of these three indices into machine learning models significantly enhances the accuracy of risk prediction. Our findings strongly support a multi-dimensional approach to glycemic assessment for superior risk stratification in critically ill patients with sepsis.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eACME\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eAverage Causal Mediation Effect\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eADE\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eAverage Direct Effect\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eALT\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eAlanine Aminotransferase\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eAST\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eAspartate Aminotransferase\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eAUC\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eArea Under the Curve\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eBMI\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eBody Mass Index\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eCI\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eConfidence Interval\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eCOPD\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eChronic Obstructive Pulmonary Disease\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eCre\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eCreatinine\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eFBG\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eFasting Blood Glucose\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eGV\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eGlycemic Variability\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eHb\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eHemoglobin\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eHbA1c\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eHemoglobin A1c\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eHF\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eHeart Failure\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eHGI\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eHemoglobin Glycation Index\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eHR\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eHazard Ratio\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eICU\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eIntensive Care Unit\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eLac\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eLactate\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eLASSO\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eLeast Absolute Shrinkage and Selection Operator\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eMAP\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMean Arterial Pressure\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eMIMIC-IV\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMedical Information Mart for Intensive Care IV\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eNWICU\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eNorthwestern Intensive Care Unit\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eOASIS\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eOxford Acute Severity of Illness Score\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003ePlt\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003ePlatelet Count\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eRCS\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eRestricted Cubic Spline\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eROC\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eReceiver Operating Characteristic\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eRRT\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eRenal Replacement Therapy\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eSHAP\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eSHapley Additive exPlanations\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eSHR\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eStress Hyperglycemia Ratio\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eSOFA\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eSequential Organ Failure Assessment\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eTbil\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eTotal Bilirubin\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eVaso\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eVasopressor Use\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eVent\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eMechanical Ventilation\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eWBC\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eWhite Blood Cell Count\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003e\u003cb\u003eXGBoost\u003c/b\u003e\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eExtreme Gradient Boosting\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study was conducted in accordance with the Declaration of Helsinki. Ethical approval and individual patient consent were waived for all three databases (MIMIC-IV, MIMIC-III-CareVue, and eICU-CRD) because they contain de-identified health information that is publicly available for research purposes. The use of the MIMIC-IV and MIMIC-III databases for this study was specifically approved by the Massachusetts Institute of Technology Institutional Review Board. The authorized researcher (certification number 13024213) completed the required data user training, which granted access approval for all three publicly available cohorts, including eICU-CRD.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets utilized in this study were obtained from three large-scale, publicly available critical care databases: the Medical Information Mart for Intensive Care IV (MIMIC-IV, version 3.1), the MIMIC-III CareVue subset (v1.4), and the eICU Collaborative Research Database (eICU-CRD, version 2.0). Data access was granted and data were extracted by an authorized researcher (Kangxing Wang, certification number: 13024213) after obtaining necessary approvals and completing ethical training. The datasets are publicly accessible for research purposes via PhysioNet.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting Interests:\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing financial interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFundings:\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study was approved by\u0026nbsp;National Key R\u0026amp;D Program of China（2022YFC2504500）and 1.3.5 project for disciplines of excellence, West China Hospital, Sichuan University（ZYGD23012）\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWKX, XHY, and YYK contributed equally to this work as co-first authors. WKX was the primary contributor, responsible for the study conception, essential data acquisition (coding and extraction), performing the main statistical analysis, and drafting the manuscript. XHY refined the methodology, provided technical support for the analysis, and assisted with data management. YYK assisted with the statistical modeling, interpretation of results, and critical revision of the manuscript. KY and ZYF served as co-corresponding authors, secured funding, and supervised the study. Both KY and ZYF critically revised the manuscript for intellectual content. All authors read and approved the final manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments:\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eMeyer, N. J. \u0026amp; Prescott, H. C. Sepsis and septic shock. Hardin CC, editor. N Engl J Med. ;391:2133\u0026ndash;46. (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1056/NEJMra2403213\u003c/span\u003e\u003cspan address=\"10.1056/NEJMra2403213\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHotamisligil, G. S. Inflammation and metabolic disorders. \u003cem\u003eNature\u003c/em\u003e \u003cb\u003e444\u003c/b\u003e, 860\u0026ndash;867. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/nature05485\u003c/span\u003e\u003cspan address=\"10.1038/nature05485\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2006).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKrinsley, J. S. et al. Diabetic status and the relation of the three domains of glycemic control tomortality in critically ill patients: an international multicenter cohort study. \u003cem\u003eCrit. Care\u003c/em\u003e. \u003cb\u003e17\u003c/b\u003e, R37. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/cc12547\u003c/span\u003e\u003cspan address=\"10.1186/cc12547\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eThe NICE-SUGAR Study Investigators. Hypoglycemia and risk of death in critically ill patients. \u003cem\u003eN Engl. J. Med.\u003c/em\u003e \u003cb\u003e367\u003c/b\u003e, 1108\u0026ndash;1118. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1056/NEJMoa1204942\u003c/span\u003e\u003cspan address=\"10.1056/NEJMoa1204942\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2012).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eService, F. J. Glucose variability. \u003cem\u003eDiabetes\u003c/em\u003e \u003cb\u003e62\u003c/b\u003e, 1398\u0026ndash;1404. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2337/db12-1396\u003c/span\u003e\u003cspan address=\"10.2337/db12-1396\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAli, N. A. et al. Glucose variability and mortality in patients with sepsis. \u003cem\u003eCrit. Care Med.\u003c/em\u003e \u003cb\u003e36\u003c/b\u003e, 2316\u0026ndash;2321. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1097/CCM.0b013e3181810378\u003c/span\u003e\u003cspan address=\"10.1097/CCM.0b013e3181810378\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2008).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRoberts, G. W. et al. Relative hyperglycemia, a marker of critical illness: introducing the stress hyperglycemia ratio. \u003cem\u003eJ. Clin. Endocrinol. Metab.\u003c/em\u003e \u003cb\u003e100\u003c/b\u003e, 4490\u0026ndash;4497. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1210/jc.2015-2660\u003c/span\u003e\u003cspan address=\"10.1210/jc.2015-2660\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYan, F. et al. 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. \u003cem\u003eCardiovasc. Diabetol. Engl.\u003c/em\u003e \u003cb\u003e23\u003c/b\u003e, 163. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12933-024-02265-4\u003c/span\u003e\u003cspan address=\"10.1186/s12933-024-02265-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWei, X. et al. Risk analysis of the association between different hemoglobin glycation index and poor prognosis in critical patients with coronary heart disease-a study based on the MIMIC-IV database. \u003cem\u003eCardiovasc. Diabetol.\u003c/em\u003e \u003cb\u003e23\u003c/b\u003e, 113. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12933-024-02206-1\u003c/span\u003e\u003cspan address=\"10.1186/s12933-024-02206-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHempe, J. M. et al. The hemoglobin glycation index identifies subpopulations with harms or benefits from intensive treatment in the ACCORD trial. \u003cem\u003eDiabetes Care\u003c/em\u003e. \u003cb\u003e38\u003c/b\u003e, 1067\u0026ndash;1074. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.2337/dc14-1844\u003c/span\u003e\u003cspan address=\"10.2337/dc14-1844\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJohnson, A. E. W. et al. MIMIC-IV, a freely accessible electronic health record dataset. \u003cem\u003eSci. Data\u003c/em\u003e. \u003cb\u003e10\u003c/b\u003e, 1. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41597-022-01899-x\u003c/span\u003e\u003cspan address=\"10.1038/s41597-022-01899-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMoukheiber, D. et al. Northwestern ICU (NWICU) database [Internet]. PhysioNet; [cited 2025 Sep 20]. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.13026/S84W-1829\u003c/span\u003e\u003cspan address=\"10.13026/S84W-1829\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSinger, M. et al. The third international consensus definitions for sepsis and septic shock (sepsis-3). \u003cem\u003eJAMA\u003c/em\u003e \u003cb\u003e315\u003c/b\u003e, 801. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1001/jama.2016.0287\u003c/span\u003e\u003cspan address=\"10.1001/jama.2016.0287\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChun, K-H. et al. In-hospital glycemic variability and all-cause mortality among patients hospitalized for acute heart failure. \u003cem\u003eCardiovasc. Diabetol.\u003c/em\u003e \u003cb\u003e21\u003c/b\u003e, 291. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12933-022-01720-4\u003c/span\u003e\u003cspan address=\"10.1186/s12933-022-01720-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHempe, J. M., Gomez, R., McCarter, R. J. \u0026amp; Chalew, S. A. High and low hemoglobin glycation phenotypes in type 1 diabetes. \u003cem\u003eJ. Diabetes Complications\u003c/em\u003e. \u003cb\u003e16\u003c/b\u003e, 313\u0026ndash;320. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/S1056-8727(01)00227-6\u003c/span\u003e\u003cspan address=\"10.1016/S1056-8727(01)00227-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2002).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCharlson, M. E., Pompei, P., Ales, K. L. \u0026amp; MacKenzie, C. R. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. \u003cem\u003eJ. Chronic Dis.\u003c/em\u003e \u003cb\u003e40\u003c/b\u003e, 373\u0026ndash;383. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/0021-9681(87)90171-8\u003c/span\u003e\u003cspan address=\"10.1016/0021-9681(87)90171-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (1987).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJohnson, A. E. W., Kramer, A. A. \u0026amp; Clifford, G. D. A new severity of illness scale using a subset of acute physiology and chronic health evaluation data elements shows comparable predictive accuracy*. \u003cem\u003eCrit. Care Med.\u003c/em\u003e \u003cb\u003e41\u003c/b\u003e, 1711\u0026ndash;1718. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1097/CCM.0b013e31828a24fe\u003c/span\u003e\u003cspan address=\"10.1097/CCM.0b013e31828a24fe\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eThe, S. O. F. A. sepsis-related organ failure assessment) score to describe organ dysfunction/failure. \u003cem\u003eIntensive Care Med.\u003c/em\u003e \u003cb\u003e22\u003c/b\u003e, 707\u0026ndash;710 (1996).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang, F. et al. Combined assessment of stress hyperglycemia ratio and glycemic variability to predict all-cause mortality in critically ill patients with atherosclerotic cardiovascular diseases across different glucose metabolic states: an observational cohort study with machine learning. \u003cem\u003eCardiovasc. Diabetol.\u003c/em\u003e \u003cb\u003e24\u003c/b\u003e, 199. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12933-025-02762-0\u003c/span\u003e\u003cspan address=\"10.1186/s12933-025-02762-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBomrah, S. et al. A scoping review of machine learning for sepsis prediction- feature engineering strategies and model performance: a step towards explainability. \u003cem\u003eCrit. Care\u003c/em\u003e. \u003cb\u003e28\u003c/b\u003e, 180. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s13054-024-04948-6\u003c/span\u003e\u003cspan address=\"10.1186/s13054-024-04948-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eQi, L. et al. Association of glycemic variability and prognosis in patients with traumatic brain injury: a retrospective study from the MIMIC-IV database. \u003cem\u003eDiabetes Res. Clin. Pract.\u003c/em\u003e \u003cb\u003e217\u003c/b\u003e, 111869. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.diabres.2024.111869\u003c/span\u003e\u003cspan address=\"10.1016/j.diabres.2024.111869\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLucci, C. et al. Hypoglycemia and glycemic variability in acute myocardial infarction: the lesser-known aspects of glycemic control. \u003cem\u003eCardiovasc. Diabetol.\u003c/em\u003e \u003cb\u003e24\u003c/b\u003e, 309. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12933-025-02862-x\u003c/span\u003e\u003cspan address=\"10.1186/s12933-025-02862-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhou, Z., Zhang, H., Gu, Y., Zhang, K. \u0026amp; Ouyang, C. Relationship between glycemic variability and the incidence of postoperative atrial fibrillation following cardiac surgery: a retrospective study from MIMIC-IV database. \u003cem\u003eDiabetes Res. Clin. Pract. Irel.\u003c/em\u003e \u003cb\u003e219\u003c/b\u003e, 111978. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.diabres.2024.111978\u003c/span\u003e\u003cspan address=\"10.1016/j.diabres.2024.111978\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSu, Y., Fan, W., Liu, Y. \u0026amp; Hong, K. Glycemic variability and in-hospital death of critically ill patients and the role of ventricular arrhythmias. \u003cem\u003eCardiovasc. Diabetol. Engl.\u003c/em\u003e \u003cb\u003e22\u003c/b\u003e, 134. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12933-023-01861-0\u003c/span\u003e\u003cspan address=\"10.1186/s12933-023-01861-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMarik, P. E. \u0026amp; Bellomo, R. Stress hyperglycemia: an essential survival response! \u003cem\u003eCrit. Care\u003c/em\u003e. \u003cb\u003e17\u003c/b\u003e, 305 (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYu, Q. et al. Impact of glycemic control metrics on short- and long-term mortality in transcatheter aortic valve replacement patients: a retrospective cohort study from the MIMIC-IV database. \u003cem\u003eCardiovasc. Diabetol. Engl.\u003c/em\u003e \u003cb\u003e24\u003c/b\u003e, 135. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12933-025-02684-x\u003c/span\u003e\u003cspan address=\"10.1186/s12933-025-02684-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePiagnerelli, M., Boudjeltia, K. Z., Vanhaeverbeek, M. \u0026amp; Vincent, J-L. Red blood cell rheology in sepsis. \u003cem\u003eIntensive Care Med.\u003c/em\u003e \u003cb\u003e29\u003c/b\u003e, 1052\u0026ndash;1061. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s00134-003-1783-2\u003c/span\u003e\u003cspan address=\"10.1007/s00134-003-1783-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2003).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRawal, G., Kumar, R., Yadav, S. \u0026amp; Singh, A. Anemia in intensive care: a review of current concepts. \u003cem\u003eJ. Crit. Care Med.\u003c/em\u003e \u003cb\u003e2\u003c/b\u003e, 109\u0026ndash;114. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1515/jccm-2016-0017\u003c/span\u003e\u003cspan address=\"10.1515/jccm-2016-0017\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2016).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTable 1 and 2 are available in the Supplementary Files section.\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Sepsis, Glycemic Variability, Stress Hyperglycemia Ratio, Hemoglobin Glycation Index, Machine Learning, Mortality","lastPublishedDoi":"10.21203/rs.3.rs-8247079/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8247079/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eGlycemic dysregulation is a hallmark of sepsis, but the comparative prognostic value of different glycemic metrics remains unclear. We aimed to systematically compare the associations of Glycemic Variability (GV), Stress Hyperglycemia Ratio (SHR), and Hemoglobin Glycation Index (HGI) with mortality in sepsis and to develop a machine learning-based prediction model for mortality in sepsis.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eThis multicenter retrospective study included 7,260 adult patients with sepsis from the MIMIC-IV and NWICU databases. GV was calculated as the coefficient of variation of blood glucose; SHR and HGI were computed using established formulas. The primary outcome was 28-day all-cause mortality. Associations between glycemic indices and mortality were assessed using multivariable Cox regression, restricted cubic splines, mediation analysis, and subgroup analyses. An XGBoost model incorporating glycemic and clinical variables was developed, and variable importance was evaluated using SHAP (SHapley Additive exPlanations) values.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eAfter full adjustment, patients in the highest quintile of GV (HR\u0026thinsp;=\u0026thinsp;1.62, 95% CI 1.33\u0026ndash;1.97) and SHR (HR\u0026thinsp;=\u0026thinsp;1.51, 95% CI 1.25\u0026ndash;1.82) had significantly higher 28-day mortality, whereas those in the highest HGI quintile had lower mortality (HR\u0026thinsp;=\u0026thinsp;0.74, 95% CI 0.61\u0026ndash;0.91). Lactate partially mediated these associations. The XGBoost model demonstrated excellent performance (AUC\u0026thinsp;=\u0026thinsp;0.798), and SHAP analysis identified GV, SHR, and HGI among the top predictors of mortality. These findings remained robust in sensitivity analysis using Random Forest imputation.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eGV, SHR, and HGI are independent and complementary prognostic markers in sepsis. A multidimensional evaluation of glycemic dysregulation enhances risk stratification, and integrating these indices into machine learning models substantially improves predictive accuracy.\u003c/p\u003e","manuscriptTitle":"Association of Glycemic Variability, Stress Hyperglycemia Ratio, and Hemoglobin Glycation Index With 28-Day Mortality in Sepsis: A Multicenter Retrospective Study With Mediation and Machine-Learning Analyses","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-12-17 07:04:00","doi":"10.21203/rs.3.rs-8247079/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-01-07T10:27:17+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-12-26T14:43:02+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-12-24T15:44:33+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"26130119595727925733172582530980979503","date":"2025-12-16T09:17:28+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-12-15T02:04:46+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"198999225641609905740990674370712183725","date":"2025-12-14T06:25:04+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"6976489611181774141525319085383157302","date":"2025-12-11T15:32:34+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"336120108016648859248370266287669750381","date":"2025-12-11T08:58:37+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"168054373633644127093081417380302576650","date":"2025-12-10T13:39:06+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-12-10T13:19:21+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-12-04T11:30:28+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-12-02T13:05:17+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-12-02T13:02:04+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2025-12-01T06:51:04+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"f6b59d3f-7591-4c03-8cb3-107ceff11720","owner":[],"postedDate":"December 17th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"in-revision","subjectAreas":[{"id":59682501,"name":"Health sciences/Biomarkers"},{"id":59682502,"name":"Health sciences/Diseases"},{"id":59682503,"name":"Health sciences/Endocrinology"},{"id":59682504,"name":"Health sciences/Medical research"},{"id":59682505,"name":"Health sciences/Risk factors"}],"tags":[],"updatedAt":"2026-01-07T10:38:09+00:00","versionOfRecord":[],"versionCreatedAt":"2025-12-17 07:04:00","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8247079","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8247079","identity":"rs-8247079","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}