Abstract
Background
Gastric surgery is a critical intervention for conditions, such as gastric cancer, obesity, and peptic ulcer disease. Despite advances in surgical techniques and perioperative care, postoperative complications, including elevated mortality, remain a major concern. Glycemic variability (GV), calculated as the coefficient of variation of blood glucose levels during the ICU stay, has emerged as a potential predictor of adverse outcomes in critically ill patients. This study aimed to investigate the association between GV and postoperative mortality in patients undergoing gastric surgery.
Methods
Data were obtained from the MIMIC-IV database, which contains anonymized health records of ICU patients admitted to Beth Israel Deaconess Medical Center. The cohort included adult patients admitted to the ICU following gastric surgery. GV was assessed using the coefficient of variation of all recorded blood glucose measurements during the ICU stay. The primary outcome was 30-day all-cause in-hospital mortality; the secondary outcome was 90-day mortality. Associations between GV and outcomes were analyzed using Cox proportional hazards models and Kaplan–Meier survival analysis. In addition, machine learning models were developed to evaluate the predictive value of GV.
Results
A total of 1099 patients were included. Higher GV was significantly associated with increased 30-day and 90-day mortality (HR 1.15, 95% CI 1.09–1.21; and HR 1.14, 95% CI 1.09–1.20, respectively). Threshold analysis identified inflection points at GV = 20.24 for 30-day mortality and GV = 33.96 for 90-day mortality. The stacking ensemble model incorporating GV achieved strong predictive performance, with an area under the receiver operating characteristic curve (AUC) of 0.83.
Conclusions
GV is a significant and independent predictor of postoperative mortality in gastric surgery patients. The observed threshold effects suggest that maintaining GV below critical levels may improve early outcomes. These findings highlight the prognostic value of GV and support its potential as a target for postoperative management. Further prospective, multicenter studies are warranted to validate these results and guide clinical practice.
Supplementary Information
The online version contains supplementary material available at 10.1186/s40001-025-03501-7.
Keywords: Glycemic variability, Mortality, Gastric surgery, Threshold effect, Critical care
Introduction
Annually, more than 300 million surgical procedures are conducted worldwide, and this number is steadily increasing, with gastric surgeries representing a notable share of the total [1, 2]. These procedures, including gastrectomies, gastric bypasses, and gastrointestinal resections, are essential for treating conditions, such as gastric cancer, obesity, and peptic ulcers. Although there have been significant improvements in surgical methods and postoperative management, including the introduction of Enhanced Recovery after Surgery protocols [3], the management of postoperative complications remains a major challenge for gastric surgery patients.
Glycemic variability (GV), defined as the ratio between the standard deviation and the mean blood glucose level measured during the entire ICU stay, has been identified as a potential predictor of poor outcomes in critically ill patients [4–6]. Unlike average blood glucose levels, GV quantifies the extent of glucose fluctuations, and previous research indicates that this fluctuation may more accurately capture an enhanced risk of cardiovascular incidents, organ failure, and higher mortality rates [4, 5]. In gastric surgery patients, GV is particularly concerning due to the common occurrence of postoperative metabolic instability, including insulin resistance and impaired glucose regulation [7, 8]. Earlier research suggests that the physiological mechanisms linking GV to mortality may involve diminished insulin sensitivity, increased oxidative stress, and widespread inflammation, although the exact nature of this association remains unclear [9–11]. These elements may play a role in causing endothelial dysfunction, reducing organ perfusion, and leading to the failure of multiple organs [12, 13]. While the importance of GV in critically ill patients is widely recognized, its specific role in predicting mortality in gastric surgery patients remains unclear.
This study aims to investigate the relationship between GV and postoperative outcomes in patients undergoing gastric surgery. In addition, we seek to develop a GV-based predictive model for postoperative mortality, with the goal of providing clinicians with a practical tool to identify high-risk patients early. By clarifying the prognostic significance of GV in this patient population, our findings may contribute to improved risk stratification and more effective postoperative management strategies.
Materials and methods
Research design
In this research, we employed the MIMIC-IV database (version 3.1), a publicly accessible data set that includes de-identified health records of ICU patients who were admitted to the Beth Israel Deaconess Medical Center from 2008 to 2022. This data set includes comprehensive clinical data, encompassing demographics, vital signs, laboratory findings, procedures, and medications. Data extraction was carried out by one author (Ge ID: 13547277), who completed the necessary data use training.
This study involved adult individuals who were admitted to the ICU for the first time following gastric surgery, including gastrotomy, gastrectomy, gastric bypass, dissection, resection, and ablation. Inclusion criteria were adults aged 18 years or older and first-time ICU admission following gastric surgery. Exclusion criteria included patients under 18 years, those with an ICU stay of less than 6 h, and individuals who had fewer than three blood glucose readings recorded during their ICU stay. The cohort was divided into three tertiles based on GV (Fig. S1 Supplementary Material).
Data collection
Clinical data were extracted using structured queries in PostgreSQL. Only information from the first 24 h of ICU admission was included. Demographic variables collected included age, sex, race, and body weight. Vital signs consisted of body temperature, respiratory rate, heart rate, systolic blood pressure (SBP), diastolic blood pressure (DBP), and peripheral oxygen saturation (SpO₂). Severity of illness was assessed using multiple scoring systems: Charlson Comorbidity Index (CCI), Glasgow Coma Scale (GCS), Oxford Acute Severity of Illness Score (OASIS), and Sequential Organ Failure Assessment (SOFA). Documented comorbidities included congestive heart failure, cerebrovascular disease, liver disease, malignancy, diabetes mellitus, hypertension, acute kidney injury (AKI), sepsis, and delirium. Laboratory variables included complete blood count parameters—white blood cells (WBC), red blood cells (RBC), platelets, red cell distribution width (RDW)—as well as metabolic and coagulation markers: pH, base excess (BE), electrolytes (serum sodium, potassium, calcium), anion gap, blood urea nitrogen (BUN), creatinine, prothrombin time (PT), partial thromboplastin time (PTT), international normalized ratio (INR), and urine output. Therapeutic interventions included administration of epinephrine, norepinephrine, insulin, statins, mechanical ventilation (MV), and continuous renal replacement therapy (CRRT). Recorded outcomes included ICU and hospital length of stay, ICU mortality, hospital mortality, and 30-day and 90-day all-cause mortality.
Calculated variables and clinical outcomes
GV was evaluated by calculating the coefficient of variation, which is the ratio of the standard deviation to the mean of all blood glucose measurements taken during the patient’s ICU stay [4]. The primary outcome of this study was in-hospital all-cause mortality occurring within 30 days of admission. The secondary outcome focused on all-cause mortality within 90 days of hospitalization.
Statistical methods
Descriptive statistics were conducted for the full study cohort. For continuous variables that followed a normal distribution, the data were expressed as the mean along with standard deviations. In contrast, variables that did not conform to a normal distribution were summarized using medians and interquartile ranges. The chi-square test was applied to compare categorical variables. The Student’s t test was utilized for continuous outcomes when the data met the normality assumptions; if these conditions were not met, the Mann–Whitney U test was employed instead. Variables with missing data exceeding 20% were excluded from the analysis. For variables with missing data below 20%, multiple imputation techniques were used to address the gaps in the data set (Table S1 Supplementary Material).
Examination of the relationship between GV and results
To explore the connection between GV and patient outcomes, Kaplan–Meier survival analyses were performed, with participants grouped into tertiles according to their GV levels. The log-rank test was utilized to compare the survival distributions across the different groups. The relationship between GV and the risk of outcomes was assessed using Cox regression models, which provided hazard ratios (HRs) along with their respective 95% confidence intervals (CIs). Three sequential models were applied: Model 1 included no covariate adjustment; Model 2 incorporated adjustments for age, sex, race, and weight; and Model 3 further accounted for SBP, comorbidities, such as congestive heart failure, liver disease, malignancy, and diabetes, in addition to WBC, serum potassium, INR, insulin administration, and MV.
To assess potential non-linear trends, restricted cubic spline (RCS) functions with three knots were utilized. Piecewise regression was employed to detect threshold effects, identifying turning points in the GV–outcome association. Interaction analyses were carried out to explore whether demographic, clinical, or therapeutic factors modified the observed effects. Furthermore, mediation analyses were performed to determine whether BUN and urine output served as mediators in the pathway linking GV to adverse clinical outcomes.
Prediction model methodology
To create a predictive model for 30-day all-cause in-hospital mortality, the data set was divided randomly into two groups: a training set comprising 70% of the data and a validation set consisting of the remaining 30%. Feature selection for the training cohort was performed using two techniques: the Least Absolute Shrinkage and Selection Operator (LASSO) and the Boruta algorithm. Only those variables selected by both methods were included in the final model development.
The chosen features were employed to train seven different machine learning classifiers: elastic net (ENet), random forest (RF), k-nearest neighbors (KNN), support vector machine (SVM), extreme gradient boosting (XGBoost), ridge regression (Ridge), and multilayer perceptron (MLP). During model training, hyperparameters for each algorithm were optimized using Bayesian search strategies to identify the best-performing parameter sets. The individual models were then integrated into a stacking ensemble to enhance predictive robustness.
The performance of the model was assessed on the validation set through various metrics, including the area under the receiver operating characteristic curve (AUC–ROC), calibration plots, and decision curve analysis. SHapley Additive exPlanations (SHAP) were utilized to assess the importance of each predictor, helping to interpret the final model that exhibited the best predictive performance.
All data analyses were performed using the R programming language, version 4.4.2. A p value less than 0.05 was deemed to indicate statistical significance.
Results
A total of 1099 adult patients who underwent gastric surgery and were admitted to the ICU for the first time were included in the final cohort. The cohort was divided into three tertiles based on GV: Tertile 1 (GV ≤ 8.77%), Tertile 2 (GV 8.78–18.60%), and Tertile 3 (GV ≥ 18.61%). The 30-day hospital mortality rates for Tertile 1, Tertile 2, and Tertile 3 were 5%, 18%, and 30%, respectively. Similarly, the 90-day hospital mortality rates for Tertile 1, Tertile 2, and Tertile 3 were 11%, 24%, and 37%, respectively.
Baseline characteristics
Table 1 shows the baseline characteristics of 1099 postoperative gastric surgery patients across GV tertiles. Significant differences were observed in age, race, weight, SBP, respiratory rate, and temperature (all p < 0.05). Clinical scores including SOFA, OASIS, and CCI were higher in Tertile 3. Comorbidities with statistical differences included congestive heart failure, liver disease, diabetes, AKI, sepsis, and delirium (all p < 0.05). Laboratory tests with notable variation included creatinine, PT, PTT, potassium, anion gap, BUN, INR, WBC, urine output, pH, and BE (all p < 0.05). Treatment differences were found in the use of epinephrine, norepinephrine, insulin, CRRT, and MV (all p < 0.05). Outcomes including LOS hospital, hospital mortality, LOS ICU, ICU mortality, 30-day, and 90-day hospital mortality were significantly different across GV tertiles (all p < 0.05).
Table 1.
Baseline characteristics of postoperative gastric surgery patients
| Characteristic | Overall N = 1099 | Tertile 1 N = 363 | Tertile 2 N = 373 | Tertile 3 N = 363 | p value1 |
|---|---|---|---|---|---|
| Demographics | |||||
| Age (year) | 65.81 (55.85, 76.66) | 65.13 (53.43, 75.79) | 65.01 (55.04, 76.62) | 67.41 (59.16, 77.40) | 0.046 |
| Gender, n (%) | 0.846 | ||||
| Female | 455 (41%) | 152 (42%) | 150 (40%) | 153 (42%) | |
| Male | 644 (59%) | 211 (58%) | 223 (60%) | 210 (58%) | |
| Race, n (%) | 0.010 | ||||
| Other | 332 (30%) | 96 (26%) | 105 (28%) | 131 (36%) | |
| White | 767 (70%) | 267 (74%) | 268 (72%) | 232 (64%) | |
| Weight (Kg) | 75.20 (64.10, 92.00) | 75.30 (65.10, 92.60) | 78.00 (65.40, 93.60) | 73.40 (62.40, 89.60) | 0.011 |
| Vital signs | |||||
| Heart rate (bmp) | 88.03 (76.92, 100.44) | 86.96 (75.85, 98.48) | 89.50 (76.65, 99.71) | 87.88 (77.74, 103.79) | 0.155 |
| SBP (mmHg) | 115.07 (106.03, 127.46) | 117.88 (107.52, 129.38) | 115.07 (105.50, 127.70) | 113.48 (104.91, 125.00) | 0.007 |
| DBP (mmHg) | 61.52 (55.18, 68.86) | 63.14 (55.58, 70.20) | 61.33 (55.18, 69.26) | 61.08 (54.65, 67.71) | 0.064 |
| Respiratory rate (bmp) | 18.42 (16.19, 21.28) | 17.71 (15.90, 20.65) | 18.26 (16.21, 21.44) | 19.25 (16.45, 22.27) | < 0.001 |
| Temperature (℃) | 36.86 (36.63, 37.16) | 36.85 (36.64, 37.11) | 36.90 (36.68, 37.20) | 36.82 (36.57, 37.13) | 0.046 |
| Spo2 (%) | 97.61 (96.16, 98.86) | 97.41 (95.92, 98.84) | 97.64 (96.29, 98.86) | 97.73 (96.25, 98.92) | 0.234 |
| Clinical scores | |||||
| SOFA | 1.00 (0.00, 3.00) | 1.00 (0.00, 2.00) | 1.00 (0.00, 3.00) | 2.00 (0.00, 4.00) | < 0.001 |
| GCS | 15.00 (15.00, 15.00) | 15.00 (15.00, 15.00) | 15.00 (15.00, 15.00) | 15.00 (15.00, 15.00) | 0.088 |
| OASIS | 33.00 (27.00, 40.00) | 29.00 (24.00, 36.00) | 34.00 (28.00, 40.00) | 36.00 (30.00, 43.00) | < 0.001 |
| CCI | 5.00 (3.00, 8.00) | 5.00 (3.00, 7.00) | 5.00 (3.00, 8.00) | 6.00 (4.00, 8.00) | < 0.001 |
| Comorbidities (%) | |||||
| Congestive heart failure, n (%) | 201 (18%) | 51 (14%) | 66 (18%) | 84 (23%) | 0.006 |
| Cerebrovascular disease, n (%) | 107 (10%) | 27 (7%) | 34 (9%) | 46 (13%) | 0.052 |
| Liver disease, n (%) | 208 (19%) | 48 (13%) | 82 (22%) | 78 (21%) | 0.003 |
| Malignant cancer, n (%) | 355 (32%) | 117 (32%) | 118 (32%) | 120 (33%) | 0.918 |
| Diabetes, n (%) | 282 (26%) | 68 (19%) | 79 (21%) | 135 (37%) | < 0.001 |
| Hypertension, n (%) | 660 (60%) | 204 (56%) | 226 (61%) | 230 (63%) | 0.139 |
| AKI, n (%) | 804 (73%) | 208 (57%) | 298 (80%) | 298 (82%) | < 0.001 |
| Sepsis, n (%) | 571 (52%) | 126 (35%) | 215 (58%) | 230 (63%) | < 0.001 |
| Delirium, n (%) | 348 (32%) | 58 (16%) | 142 (38%) | 148 (41%) | < 0.001 |
| Laboratory test | |||||
| RBC (109/L) | 3.29 (2.89, 3.75) | 3.32 (2.89, 3.77) | 3.32 (2.95, 3.77) | 3.23 (2.83, 3.72) | 0.062 |
| WBC (109/L) | 10.70 (7.88, 13.93) | 10.00 (7.40, 13.57) | 11.00 (8.10, 13.87) | 10.96 (8.15, 14.73) | 0.027 |
| Platelet (109/L) | 190.00 (137.00, 255.13) | 195.25 (145.00, 260.33) | 194.67 (133.33, 256.00) | 179.75 (128.71, 250.00) | 0.144 |
| RDW (%) | 15.20 (13.93, 16.83) | 15.09 (13.95, 16.65) | 15.20 (13.83, 16.88) | 15.38 (14.00, 17.35) | 0.364 |
| GV (%) | 18.60 (11.09, 28.71) | 8.77 (5.32, 11.02) | 18.60 (16.36, 21.74) | 33.89 (28.79, 44.77) | < 0.001 |
| PH | 7.37 (7.33, 7.41) | 7.37 (7.34, 7.40) | 7.37 (7.33, 7.41) | 7.35 (7.30, 7.40) | < 0.001 |
| BE | − 1.00 (− 3.91, 1.00) | − 0.67 (− 2.50, 1.25) | − 1.00 (− 4.00, 1.00) | − 2.00 (− 5.33, 0.17) | < 0.001 |
| Sodium (mmol/L) | 138.78 (136.57, 141.14) | 139.00 (137.00, 140.80) | 138.67 (136.33, 141.00) | 139.00 (136.33, 141.83) | 0.497 |
| Potassium (mmol/L) | 4.11 (3.85, 4.43) | 4.08 (3.82, 4.33) | 4.12 (3.87, 4.43) | 4.16 (3.85, 4.52) | 0.013 |
| Calcium (mg/dL) | 8.20 (7.81, 8.60) | 8.20 (7.84, 8.57) | 8.20 (7.82, 8.60) | 8.18 (7.78, 8.63) | 0.969 |
| Anion gap (m Eq/L) | 12.57 (10.83, 14.67) | 12.00 (10.60, 13.67) | 12.50 (10.80, 14.50) | 13.33 (11.25, 16.00) | < 0.001 |
| BUN (mg/dL) | 19.67 (13.00, 32.20) | 17.33 (11.67, 26.00) | 19.33 (12.67, 30.11) | 23.80 (15.14, 41.33) | < 0.001 |
| Creatinine (mg/dL) | 0.90 (0.69, 1.30) | 0.80 (0.67, 1.07) | 0.89 (0.70, 1.26) | 1.05 (0.73, 1.85) | < 0.001 |
| INR | 1.27 (1.15, 1.50) | 1.23 (1.12, 1.37) | 1.28 (1.15, 1.50) | 1.30 (1.18, 1.60) | < 0.001 |
| PT (S) | 14.00 (12.70, 16.20) | 13.65 (12.50, 15.23) | 14.15 (12.83, 16.28) | 14.33 (12.85, 17.45) | < 0.001 |
| PTT (S) | 30.70 (27.30, 37.39) | 29.85 (27.07, 36.00) | 30.60 (27.48, 37.08) | 31.90 (27.50, 40.47) | 0.016 |
| Urine output (mL) | 1360.00 (865.00, 2010.00) | 1615.00 (1072.00, 2450.00) | 1315.00 (810.00, 1925.00) | 1200.00 (645.00, 1729.00) | < 0.001 |
| Treatments | |||||
| Epinephrine, n (%) | 33 (3%) | 3 (1%) | 8 (2%) | 22 (6%) | < 0.001 |
| Norepinephrine, n (%) | 242 (22%) | 32 (9%) | 91 (24%) | 119 (33%) | < 0.001 |
| Insulin, n (%) | 627 (57%) | 145 (40%) | 230 (62%) | 252 (69%) | < 0.001 |
| Statin, n (%) | 213 (19%) | 58 (16%) | 72 (19%) | 83 (23%) | 0.064 |
| CRRT, n (%) | 51 (5%) | 0 (0%) | 20 (5%) | 31 (9%) | < 0.001 |
| MV, n (%) | 948 (86%) | 289 (80%) | 331 (89%) | 328 (90%) | < 0.001 |
| Events | |||||
| Los hospital (day) | 11.27 (7.33, 21.28) | 9.31 (5.89, 16.27) | 12.24 (7.93, 23.09) | 12.83 (7.73, 23.00) | < 0.001 |
| Hospital Mortality, n (%) | 163 (15%) | 12 (3%) | 57 (15%) | 94 (26%) | < 0.001 |
| Los ICU (day) | 2.68 (1.75, 4.95) | 1.93 (1.28, 2.95) | 3.21 (2.00, 5.91) | 3.10 (1.94, 6.55) | < 0.001 |
| ICU Mortality, n (%) | 99 (9%) | 8 (2%) | 29 (8%) | 62 (17%) | < 0.001 |
| 30-day hospital mortality, n (%) | 195 (18%) | 19 (5%) | 67 (18%) | 109 (30%) | < 0.001 |
| 90-day hospital mortality, n (%) | 266 (24%) | 41 (11%) | 89 (24%) | 136 (37%) | < 0.001 |
GV Tertiles, Tertile1 (0–14.03), Tertile (14.03–24.75), Tertile (24.75–99.94)
GV: Glucose variability, SOFA: Sequential organ failure assessment, GCS: Glasgow Coma Scale, CCI: Charlson Comorbidity Index, SpO2: Oxygen saturation, SBP: Systolic blood pressure, DBP: Diastolic blood pressure, AKI: Acute kidney injury, WBC: White blood cell count, RBC: Red blood cell count, Platelet: Platelet count, INR: International normalized ratio, MV: Mechanical Ventilation, CRRT: Continuous renal replacement therapy
Relationship of GV with 30-day and 90-day overall mortality
The variables incorporated into the multivariable Cox regression model were selected based on the outcomes from the univariable Cox analysis (Table S2 Supplementary Material), the Boruta algorithm (Fig. 5), and expert clinical input. The results from the fully adjusted Cox proportional hazards model (Model 3) are displayed in Table 2, which examines the association between GV and all-cause in-hospital mortality at both 30 and 90 days. When GV was analyzed as a continuous variable, each 10-unit rise in GV was found to be independently linked to an increased risk of both 30-day mortality (HR 1.15, 95% CI 1.09–1.21; p < 0.001) and 90-day mortality (HR 1.14, 95% CI 1.09–1.20; p < 0.001).
Fig. 5.
Feature selection for 30-day mortality prediction model. a Variable importance ranking based on the Boruta algorithm; green bars represent confirmed important features, red bars indicate rejected ones, b LASSO coefficient profiles of candidate predictors as a function of log (lambda), c fivefold cross-validation for LASSO logistic regression identifying the optimal value of the regularization parameter (lambda), d Venn diagram showing the intersection of features selected by both LASSO and Boruta methods; 15 features were identified by both approaches. GV: Glucose variability, SOFA: Sequential organ failure assessment, GCS: Glasgow Coma Scale, CCI: Charlson Comorbidity Index, SpO2: Oxygen saturation, SBP: Systolic blood pressure, DBP: Diastolic blood pressure, AKI: Acute kidney injury, WBC: White blood cell count, RBC: Red blood cell count, Platelet: Platelet count, INR: International normalized ratio, MV: Mechanical Ventilation, CRRT: Continuous renal replacement therapy
Table 2.
Relationship between GV and mortality in postoperative gastric surgery patients
| Variables | Model1 | Model2 | Model3 | |||
|---|---|---|---|---|---|---|
| HR (95%CI) | P | HR (95%CI) | P | HR (95%CI) | P | |
| 30-day hospital mortality | ||||||
| GV per 10 units | 1.20 (1.14–1.25) | < 0.001 | 1.19 (1.14–1.25) | < 0.001 | 1.15 (1.09–1.21) | < 0.001 |
| GV Tertiles | ||||||
| Tertile 1 | 1.00 (Reference) | 1.00 (Reference) | 1.00 (Reference) | |||
| Tertile 2 | 3.65 (2.19–6.08) | < 0.001 | 3.59 (2.16–5.98) | < 0.001 | 3.14 (1.87–5.28) | < 0.001 |
| Tertile 3 | 6.65 (4.09–10.83) | < 0.001 | 6.15 (3.77–10.02) | < 0.001 | 5.13 (3.08–8.57) | < 0.001 |
| P for trend | < 0.001 | < 0.001 | < 0.001 | |||
| 90-day hospital mortality | ||||||
| GV per 10 units | 1.18 (1.13–1.23) | < 0.001 | 1.18 (1.13–1.23) | < 0.001 | 1.14 (1.09–1.20) | < 0.001 |
| GV Tertiles | ||||||
| Tertile 1 | 1.00 (Reference) | 1.00 (Reference) | 1.00 (Reference) | |||
| Tertile 2 | 2.30 (1.59–3.33) | < 0.001 | 2.25 (1.55–3.26) | < 0.001 | 2.02 (1.38–2.96) | < 0.001 |
| Tertile 3 | 4.04 (2.85–5.73) | < 0.001 | 3.69 (2.60–5.24) | < 0.001 | 3.16 (2.18–4.59) | < 0.001 |
| P for trend | < 0.001 | < 0.001 | < 0.001 | |||
Model 1: Crude
Model 2: Adjust, Age, Gender, Race, Weight
Model 3: Adjust, Age, Gender, Race, Weight, SBP, Congestive heart failure, Liver disease, Malignant cancer, Diabetes, WBC, Potassium, INR, Insulin, MV
GV Tertiles, Tertile1 (0–14.03), Tertile (14.03–24.75), Tertile (24.75–99.94)
GV: Glucose variability, SBP: Systolic blood pressure, WBC: White blood cell count, RBC: Red blood cell count, INR: International normalized ratio, MV: Mechanical Ventilation
When GV was categorized into tertiles, a clear dose–response relationship with mortality was observed. For 30-day mortality, the adjusted HRs for Tertile 2 and Tertile 3 were 3.14 (95% CI 1.87–5.28; p < 0.001) and 5.13 (95% CI 3.08–8.57; p < 0.001), respectively, when compared to Tertile 1, with a trend p value of < 0.001. For 90-day mortality, the HRs for Tertiles 2 and 3 were 2.02 (95% CI 1.38–2.96; p < 0.001) and 3.16 (95% CI 2.18–4.59; p < 0.001), respectively, compared to Tertile 1, with a trend p value of < 0.001. Kaplan–Meier survival analysis indicated that Tertile 3 exhibited the highest mortality rates for both 30-day and 90-day outcomes, with the difference being statistically significant (log-rank p < 0.001) (Fig. 1).
Fig. 1.
Kaplan–Meier survival curves for 30-day and 90-day mortality stratified by GV tertiles. a GV and 30-day mortality. b GV and 90-day mortality
Nonlinear effects of GV on 30- and 90-day in-hospital mortality
In the fully adjusted model, the use of RCS analysis revealed a significant nonlinear association between GV and in-hospital mortality at both 30-day and 90-day intervals. For 30-day mortality, the overall association was statistically significant (P-overall < 0.001), with compelling evidence of nonlinearity (P-non-linear < 0.001). A similar pattern was observed for 90-day mortality (P-overall < 0.001; P-non-linear < 0.001) (Fig. 2).
Fig. 2.
RCS analysis of GV and mortality in postoperative gastric surgery patients: (a) 30-day mortality with GV; (b) 90-day mortality with GV
Table 3 presents the analysis of the threshold effect between GV and mortality rates during hospitalization. For 30-day mortality, an inflection point was identified at GV = 20.24. Below this threshold, GV was strongly associated with mortality (HR 4.01, 95% CI 2.07–7.75; p < 0.001), but not above (HR 1.06, 95% CI 0.99–1.13; p = 0.093). For 90-day mortality, the inflection point was 33.96, with a significant association below (HR 1.61, 95% CI 1.34–1.94; p < 0.001) and none above (HR 0.98, 95% CI 0.89–1.09; p = 0.761). Likelihood ratio tests supported the two-piecewise model over the linear model (both p < 0.001).
Table 3.
Threshold effect analysis of on mortality risk
| Outcome | HR (95% CI) | P |
|---|---|---|
| 30-day hospital mortality | ||
| GV per 10 units | ||
| Model 1 Fitting model by standard linear regression | 1.15 (1.09–1.21) | < 0.001 |
| Model 2 Fitting model by two-piecewise linear regression | ||
| Inflection point | 20.24 | |
| < 20.24 | 4.01 (2.07–7.75) | < 0.001 |
| ≥ 20.24 | 1.06 (0.99–1.13) | 0.093 |
| P for likelihood test | < 0.001 | |
| 90-day hospital mortality | ||
| GV per 10 units | ||
| Model 1 Fitting model by standard linear regression | 1.14 (1.09–1.20) | < 0.001 |
| Model 2 Fitting model by two-piecewise linear regression | ||
| Inflection point | 33.96 | |
| < 33.96 | 1.61 (1.34–1.94) | < 0.001 |
| ≥ 33.96 | 0.98 (0.89–1.09) | 0.761 |
| P for likelihood test | < 0.001 |
Subgroup and interaction analysis
Interaction analyses were conducted based on the fully adjusted model to assess whether the association between GV and mortality differed across subgroups defined by age, congestive heart failure, diabetes, hypertension, sepsis, delirium, insulin use, and statin use. No meaningful interactions were detected for either 30-day or 90-day in-hospital mortality (all p values for interaction > 0.05) (Fig. 3).
Fig. 3.
Subgroup analysis of the association between GV and mortality in postoperative gastric surgery patients. a GV and 28-day all-cause mortality. b GV and 90-day all-cause mortality
Mediation analysis of renal function in the association between GV and mortality
Mediation analysis was conducted to explore the indirect effects of GV on mortality through renal function, represented by BUN and urine output. For 30-day in-hospital mortality, BUN and urine output accounted for 5.99% and 8.54% of the total effect of GV, respectively. For 90-day mortality, the proportions mediated by BUN and urine output were 6.11% and 9.51%, respectively (Fig. 4).
Fig. 4.
Mediation analysis of the association between GV and mortality via renal function. a Mediation of the association between GV and 30-day mortality through BUN. b Mediation of the association between GV and 90-day mortality through BUN. c Mediation of the association between GV and 30-day mortality through urine output. d Mediation of the association between GV and 90-day mortality through urine output
Feature selection for 30-day mortality prediction
For the purpose of model development, the data set was randomly split into two cohorts: a training group consisting of 768 participants and a validation group with 331 participants. The 30-day in-hospital mortality rate was identical across both cohorts, with each group exhibiting a rate of 18% (136 deaths in the training group and 59 in the validation group). A comparison of baseline characteristics revealed no significant differences across any of the variables, suggesting that the cohorts were well-balanced before model training (Table S3 Supplementary Material).
Feature selection was performed on the training cohort using two complementary approaches: LASSO regression and the Boruta algorithm. Variables identified by both methods were retained for model construction. The final set of selected features included SBP, respiratory rate, SOFA, GCS, OASIS, CCI, norepinephrine use, sepsis, RDW, serum sodium, anion gap, INR, PT, PTT, and GV (Fig. 5).
Model development and performance evaluation
Based on the selected features, seven machine learning models were trained to predict 30-day in-hospital mortality. Bayesian optimization was used during training to determine optimal hyperparameters. The models included Ridge, ENet, KNN, RF, XGBoost, MLP, and SVM, which were subsequently combined into a stacking ensemble to enhance predictive performance.
In the validation cohort, the stacking model achieved the highest AUC at 0.8311 (95% CI 0.7845–0.8977). The individual model AUCs were as follows: ridge 0.8283 (95% CI 0.7796–0.8970), ENet 0.8280 (95% CI 0.7809–0.8951), XGBoost 0.8224 (95% CI 0.7800–0.8849), SVM 0.8174 (95% CI 0.7688–0.8860), MLP 0.8164 (95% CI 0.7685–0.8843), RF 0.8067 (95% CI 0.7575–0.8760), and KNN 0.7960 (95% CI 0.7461–0.8658) (Fig. 6).
Fig. 6.
a ROC curves of different prediction models for 30-day mortality in the test set, b SHAP analysis for interpretation of the stacking model predicting 30-day mortality. ROC: Receiver Operating Characteristic, Ridge: Ridge regression, ENet: Elastic Net, KNN: K-Nearest Neighbors, RF: Random Forest, XGBoost: Extreme Gradient Boosting, SVM: Support Vector Machine, MLP: Multilayer Perceptron, Stacking: Stacking ensemble model
Beyond AUC, the stacking model also showed favorable performance in both calibration and decision curve analyses (Fig. S2, Supplementary Material). Taken together, these results support the stacking model as the overall best-performing approach for mortality prediction.
Model interpretation based on SHAP values
SHAP analysis of the stacking model revealed that GV was the most influential predictor of 30-day in-hospital mortality. Higher GV values were strongly associated with increased mortality risk. Other top-ranking features included respiratory rate, PTT, PT, and INR, but their contributions were lower than that of GV. These results underscore the central role of GV in mortality prediction within the model (Fig. 6).
Web-based clinical decision support tool
We deployed both the full and simplified versions of the stacking model as web-based clinical decision support tools to enhance clinical applicability. The full model, incorporating 15 variables, achieved an AUC of 0.83 (https://docterge.shinyapps.io/Gastric/). Considering that some coagulation-related indicators (INR, PT, and PTT) may not be rapidly available in emergency settings, we developed a simplified version by excluding these variables. The simplified model, comprising 12 variables, maintained good discrimination with an AUC of 0.80 (Fig. S3, Supplementary Material), showing only a modest reduction in performance compared with the full model. The simplified model is available at https://doctorge.shinyapps.io/SimplifiedGastric/.
Discussion
This study is the first to thoroughly investigate the prognostic value of GV in critically ill patients who have undergone gastric surgery. Increased GV was independently associated with a greater risk of in-hospital mortality at both 30 and 90 days, with consistent results across continuous, categorical, and nonlinear analysis methods. Threshold analysis identified inflection points at GV = 20.24 and GV = 33.96, below which mortality risk increased significantly. Mediation analysis indicated partial effects through renal function. A prediction model incorporating GV demonstrated strong performance, with the stacking ensemble achieving the highest discrimination. SHAP analysis confirmed GV as the most influential predictor of mortality.
This research supports and builds upon prior studies regarding the prognostic value of GV in patients who are critically ill. In contrast to the average blood glucose levels assessed by hemoglobin A1c, GV measures the short-term variations in blood glucose, encompassing both the highest and lowest levels. This variability may provide a more precise indication of cardiovascular risk [14]. Studies in type 1 diabetes patients have also shown that larger glucose fluctuations are linked to slower processing speeds and decreased accuracy in cognitive tasks, highlighting GV’s role in cognitive vulnerability and its importance in optimizing cognitive function and overall health in T1D patients [15]. In patients who are critically ill, including those suffering from pneumonia, elevated GV has been linked to higher 28-day mortality rates and extended durations of ICU admission. In particular, patients with high GV experienced a mortality rate of 37.5%, while those with low GV had a rate of 25.4% [16]. Our research supports these findings, demonstrating that elevated GV is linked to a higher mortality rate in patients who have undergone gastric surgery. These findings highlight the critical need for early management of GV in the ICU to enhance patient outcomes.
Additional research has indicated that GV is linked to long-term conditions, such as coronary artery disease, metabolic syndrome, and type 2 diabetes, each of which contributes to a higher risk of cardiovascular events [17]. Importantly, several studies have highlighted a connection between GV and oxidative stress, as well as DNA damage, indicating that these factors may serve as a potential pathway by which GV affects long-term health outcomes [17]. Furthermore, elevated GV has been linked to a heightened risk of ventricular arrhythmias and in-hospital mortality among ICU patients. Specifically, each additional unit increase in cardiovascular risk corresponds to a 21% greater likelihood of arrhythmias and a 30% increased risk of mortality [18]. These results support our findings, highlighting the significant impact of GV on both immediate and long-term outcomes in critically ill patients. In individuals with diabetes, GV has been recognized as an independent risk factor for major adverse cardiovascular events, with high GV associated with a 61% greater risk [19]. This further highlight that fluctuations in glucose levels, rather than just average blood glucose levels, significantly contribute to cardiovascular risk. Elevated GV has been associated with significant cognitive decline and higher mortality rates in critically ill patients suffering from cerebrovascular conditions [20], underscoring the importance of more effective GV management in these high-risk groups. These studies collectively emphasize that GV, as a modifiable risk factor, warrants closer monitoring and management to optimize patient outcomes and reduce mortality risk.
GV plays a pivotal role in determining mortality outcomes among patients who have undergone gastric surgery, especially those who have experienced substantial gastrointestinal resections or modifications. Gastric surgery induces structural and functional changes to the gastrointestinal tract, profoundly impacting glucose metabolism and increasing the risk of glucose fluctuations, which include both hyperglycemia and hypoglycemia [21, 22]. These fluctuations compromise the patient’s physiological stability, especially in the early postoperative period. The pathophysiology of GV in this context is multifactorial. Surgical trauma and the stress response contribute to insulin resistance, impairing glucose regulation and resulting in hyperglycemia [23, 24]. Conversely, in the postoperative phase, rapid nutrient absorption via intravenous nutrition pathways may lead to excessive insulin secretion, causing hypoglycemia [25]. This imbalance between insulin production and glucose uptake exacerbates glucose fluctuations. Furthermore, gastric surgeries, such as gastrectomy or gastric bypass, alter the gut microbiome and gastrointestinal motility, affecting the secretion of hormones, such as GLP-1 and GIP, which are key in glucose regulation [26]. These hormonal changes, alongside disruptions to the gut–brain axis, destabilize blood glucose control, further exacerbating GV [27]. The surgical stress response, characterized by elevated catecholamines and cortisol, further impairs insulin sensitivity, exacerbating metabolic dysregulation and GV [28]. Clinically, high GV contributes to endothelial dysfunction, impairing vascular tone and organ perfusion, which in turn worsens postoperative conditions, such as sepsis, acute kidney injury, and cardiovascular instability [29, 30]. The complex interplay of insulin resistance, hormonal changes, and metabolic stress following gastric surgery creates a cycle of glucose instability that directly impacts organ function and increases mortality risk. Proper management of GV, especially through stringent glucose regulation, is an essential therapeutic approach for enhancing the outcomes of patients following gastric surgery.
A key finding in this study is the threshold effect observed between GV and mortality. Our findings demonstrated a non-linear correlation between GV and both 30-day and 90-day in-hospital mortality, with specific thresholds that influence the intensity of this relationship. Specifically, for 30-day mortality, a threshold at GV = 20.24 was identified, below which GV had a strong, statistically significant effect on mortality. However, beyond this threshold, the effect weakened, suggesting that GV beyond a certain level may no longer significantly contribute to early mortality. Similarly, for 90-day mortality, the threshold was at GV = 33.96, with the association diminishing above this point. This pattern does not contradict our continuous analyses, which consistently showed that higher GV is associated with higher mortality overall. The threshold analysis refines this finding by revealing that mortality risk increases more steeply in the lower GV range (below the threshold), whereas above the threshold, the risk curve plateaus and other clinical factors—such as organ dysfunction, sepsis, and prolonged inflammation—may become the dominant drivers of mortality. These findings indicate that the impact of GV on mortality is not linear but exhibits a threshold effect. Below these inflection points, increasing GV correlates strongly with higher mortality. However, once GV surpasses these thresholds, other clinical factors, such as organ dysfunction, sepsis, and prolonged inflammation, may become more influential in determining mortality outcomes. Clinically, our data indicate that maintaining GV at approximately ≤ 20% may be considered a pragmatic early postoperative target, supported by close glucose monitoring, timely insulin titration, nutritional optimization, and avoidance of iatrogenic hypoglycemia. In contrast, once GV exceeds 33.96%, further increases in GV were not significantly associated with higher 90-day mortality in our cohort; therefore, management should prioritize concurrent risk drivers—such as early recognition and source control of sepsis, hemodynamic stabilization, renal/respiratory support, and targeted anti-inflammatory interventions—alongside reasonable glycemic control. This threshold pattern emphasizes that glycemic control alone may be insufficient, necessitating a comprehensive, individualized, and multi-domain approach.
This research emphasizes the significant role of GV in predicting mortality outcomes among patients who have undergone gastric surgery. Gastric surgeries often lead to significant physiological changes, including alterations in gastric anatomy, impaired insulin secretion, and disruptions in nutrient absorption. These factors contribute to metabolic instability, including insulin resistance and dysregulated glucose metabolism, which in turn increases GV. The identification of specific threshold points for GV reveals that GV has a stronger association with mortality when it falls below these levels. Managing GV below these thresholds could, therefore, significantly improve patient outcomes, particularly in the early postoperative period. The clinical implications of these findings suggest that interventions targeting GV stabilization should be prioritized for postoperative gastric surgery patients, especially those with high risk of mortality. Pharmacological strategies, such as the use of insulin sensitizers or drugs modulating the inflammatory response, could be considered to control both GV and underlying metabolic disturbances. In addition, non-pharmacologic measures such as intensive glucose monitoring, dietary adjustments, and early mobilization may help stabilize GV. By addressing both GV and related metabolic disturbances, clinicians can improve outcomes and reduce mortality risk in this high-risk patient population.
There are several limitations to this study. First, although we adjusted for insulin use, we did not account for the specific insulin dosage or different administration routes, both of which may influence GV and potentially confound the relationship between GV and postoperative outcomes. Second, GV was assessed over the entire ICU stay using the coefficient of variation based on blood glucose measurements, but without a fixed time window for measurements. Variations in the timing and frequency of glucose monitoring across patients may have introduced measurement bias, and this approach may not capture glucose fluctuations outside the ICU or during other recovery phases. Finally, the study is based on data from a single-center ICU, limiting the generalizability of the findings to other healthcare settings with different patient populations or management protocols. Future studies should include large-scale, multi-center prospective trials to confirm these results and further investigate the potential of GV as a prognostic indicator in patients who have undergone gastric surgery.
Conclusion
This study suggests that GV serve as a crucial predictor of in-hospital mortality, significantly impacting both 30-day and 90-day outcomes for patients who have undergone gastric surgery. The threshold effect observed, with GV thresholds of 20.24 and 33.96 for 30-day and 90-day mortality, respectively, highlights the critical role of GV in postoperative outcomes. In addition, the stacking ensemble model developed in this study showed strong predictive performance for mortality, underscoring GV’s potential as an important prognostic tool. These results imply that GV could be an important indicator for recognizing patients at high risk, helping to inform and improve postoperative care strategies. Future research involving multi-center, prospective trials is essential to confirm these findings and to deepen our understanding of the role of GV in enhancing postoperative care.
Supplementary Information
Acknowledgements
Not applicable.
Author contributions
YG: data curation, formal analysis, methodology, and writing—original draft. GW: formal analysis and writing—review and editing. YH and BL: conceptualization and supervision. YZ: conceptualization, supervision, and writing—review and editing. All authors read and approved the final draft.
Funding
None.
Data availability
All data and material were available at https://mimic.mit.edu/.
Declarations
Ethics approval and consent to participate
The MIMIC-IV database adheres to the principles of the Helsinki Declaration and has been approved by the Institutional Review Board (IRB) of Beth Israel Deaconess Medical Center (2001P-001699/14). The IRB evaluated the data collection process and the creation of the research resource, authorized the data-sharing initiative, and exempted the need for informed consent.
Consent for publications
Not applicable.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Yuanshuo Ge and Guangdong Wang have contributed to the manuscript equally.
References
- 1.Weiser TG, Haynes AB, Molina G, Lipsitz SR, Esquivel MM, Uribe-Leitz T, et al. Estimate of the global volume of surgery in 2012: an assessment supporting improved health outcomes. Lancet. 2015;385(2):11. [DOI] [PubMed] [Google Scholar]
- 2.Weiser TG, Haynes AB, Molina G, Lipsitz SR, Esquivel MM, Uribe-Leitz T, et al. Size and distribution of the global volume of surgery in 2012. Bull World Health Organ. 2016;94(3):201–209f. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Lobo DN, Joshi GP, Kehlet H. Challenges in enhanced recovery after surgery (ERAS) research. Br J Anaesth. 2024;133(4):717–21. [DOI] [PubMed] [Google Scholar]
- 4.Chen Y, Yang Z, Liu Y, Gue Y, Zhong Z, Chen T, et al. Prognostic value of glycaemic variability for mortality in critically ill atrial fibrillation patients and mortality prediction model using machine learning. Cardiovasc Diabetol. 2024;23(1):426. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.He HM, Xie YY, Wang Z, Li J, Zheng SW, Li XX, et al. Associations of variability in blood glucose and systolic blood pressure with mortality in patients with coronary artery disease: a retrospective cohort study from the MIMIC-IV database. Diabetes Res Clin Pract. 2024;209:111595. [DOI] [PubMed] [Google Scholar]
- 6.Chao WC, Tseng CH, Wu CL, Shih SJ, Yi CY, Chan MC. Higher glycemic variability within the first day of ICU admission is associated with increased 30-day mortality in ICU patients with sepsis. Ann Intensive Care. 2020;10(1):17. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Nilsen I, Sundbom M, Osterberg J, Laurenius A, Andersson A, Haenni A. Glycemic variability and hypoglycemia before and after Roux-en-Y gastric bypass and sleeve gastrectomy—a cohort study of females without diabetes. Surg Obes Relat Dis. 2024;20(1):10–6. 10.1016/j.soard.2023.07.008. [DOI] [PubMed] [Google Scholar]
- 8.Ilesanmi I, Tharakan G, Alexiadou K, Behary P, Alessimii H, Bovill-Taylor C, et al. Roux-en-Y gastric bypass increases glycemic variability and time in hypoglycemia in patients with obesity and prediabetes or type 2 diabetes: a prospective cohort study. Diabetes Care. 2021;44(2):614–7. [DOI] [PubMed] [Google Scholar]
- 9.Alfieri V, Myasoedova VA, Vinci MC, Rondinelli M, Songia P, Massaiu I, et al. The role of glycemic variability in cardiovascular disorders. Int J Mol Sci. 2021;22(16):8393. 10.3390/ijms22168393. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Zhou Z, Sun B, Huang S, Zhu C, Bian M. Glycemic variability: adverse clinical outcomes and how to improve it? Cardiovasc Diabetol. 2020;19(1):102. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Klimontov VV, Semenova JF. Glucose variability in subjects with normal glucose tolerance: relations with body composition, insulin secretion and sensitivity. Diabetes Metab Syndr. 2022;16(1):102387. 10.1016/j.dsx.2022.102387. [DOI] [PubMed] [Google Scholar]
- 12.Maruhashi T, Higashi Y. Pathophysiological association between diabetes mellitus and endothelial dysfunction. Antioxidants. 2021;10(8):1306. 10.3390/antiox10081306. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Yang Z, Le TD, Simovic MO, Liu B, Fraker TL, Cancio TS, et al. Traumatized triad of complementopathy, endotheliopathy, and coagulopathy-impact on clinical outcomes in severe polytrauma patients. Front Immunol. 2022;13:991048. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Belli M, Bellia A, Sergi D, Barone L, Lauro D, Barillà F. Glucose variability: a new risk factor for cardiovascular disease. Acta Diabetol. 2023;60(10):1291–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Hawks ZW, Beck ED, Jung L, Fonseca LM, Sliwinski MJ, Weinstock RS, et al. Dynamic associations between glucose and ecological momentary cognition in Type 1 diabetes. NPJ Digit Med. 2024;7(1):59. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Kim SH, Kim JY, Kim ES, Park IR, Ha EY, Chung SM, et al. Early glycaemic variability increases 28-day mortality and prolongs intensive care unit stay in critically ill patients with pneumonia. Ann Med. 2022;54(1):2736–43. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Bragagna L, Polak C, Schütz L, Maqboul L, Klammer C, Feldbauer R, et al. Effect of repeated bolus and continuous glucose infusion on DNA damage and oxidative stress biomarkers in healthy male volunteers. Int J Mol Sci. 2023;24(17):13608. 10.3390/ijms241713608. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.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. 2023;22(1):134. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Lim S, Chung SH, Kim JH, Kim YH, Kim EJ, Joo HJ. Effects of metabolic parameters’ variability on cardiovascular outcomes in diabetic patients. Cardiovasc Diabetol. 2023;22(1):114. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Cai W, Li Y, Guo K, Wu X, Chen C, Lin X. Association of glycemic variability with death and severe consciousness disturbance among critically ill patients with cerebrovascular disease: analysis of the MIMIC-IV database. Cardiovasc Diabetol. 2023;22(1):315. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Stentebjerg LL, Madsen LR, Støving RK, Andersen LLT, Vinter CA, Juhl CB, et al. Roux-en-Y gastric bypass increases glycemic excursions during pregnancy and postpartum: a prospective cohort study. Diabetes Care. 2023;46(3):502–10. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Jin SM. Asymptomatic hypoglycemia after metabolic surgery: new insights from perioperative continuous glucose monitoring. Diabetes Metab J. 2022;46(5):675–6. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Cao H, Zhang K, Cao M, Zhang X. A commentary on ‘Preoperative carbohydrate loading with individualized supplemental insulin in diabetic patients undergoing gastrointestinal surgery: a randomized trial.’ Int J Surg. 2023;109(10):3201–2. 10.1097/JS9.0000000000000570. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Chae S, Choi J, Lim S, Choi HJ, Park J, Hong SH, et al. Stress burden related to postreperfusion syndrome may aggravate hyperglycemia with insulin resistance during living donor liver transplantation: a propensity score-matching analysis. PLoS ONE. 2020;15(12):e0243873. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Rohli KE, Boyer CK, Blom SE, Stephens SB. Nutrient regulation of pancreatic islet β-cell secretory capacity and insulin production. Biomolecules. 2022. 10.3390/biom12020335. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Yadav J, Liang T, Qin T, Nathan N, Schwenger KJP, Pickel L, et al. Gut microbiome modified by bariatric surgery improves insulin sensitivity and correlates with increased brown fat activity and energy expenditure. Cell Rep Med. 2023;4(5):101051. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Martinou E, Stefanova I, Iosif E, Angelidi AM. Neurohormonal changes in the gut-brain axis and underlying neuroendocrine mechanisms following bariatric surgery. Int J Mol Sci. 2022;23(6):3339. 10.3390/ijms23063339. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Kivimäki M, Bartolomucci A, Kawachi I. The multiple roles of life stress in metabolic disorders. Nat Rev Endocrinol. 2023;19(1):10–27. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Akhter MS, Goodwin JE. Endothelial dysfunction in cardiorenal conditions: implications of endothelial glucocorticoid receptor-Wnt signaling. Int J Mol Sci. 2023;24(18):14261. 10.3390/ijms241814261. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Xu S, Ilyas I, Little PJ, Li H, Kamato D, Zheng X, et al. Endothelial dysfunction in atherosclerotic cardiovascular diseases and beyond: from mechanism to pharmacotherapies. Pharmacol Rev. 2021;73(3):924–67. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
All data and material were available at https://mimic.mit.edu/.