Abstract
Background
Sepsis is a life-threatening complication in patients with orthopedic trauma, associated with significant mortality. Dysregulated glucose metabolism is a hallmark of critical illness, but the specific role of glycemic variability (GV), beyond hyperglycemia, in this high-risk population remains inadequately explored.
Methods
This retrospective cohort study utilized 1946 patients from the Beth Israel Deaconess Medical Center between 2008 and 2022. We included adult intensive care unit (ICU) patients with a primary diagnosis of orthopedic trauma and concurrent sepsis. GV was quantified as the coefficient of variation of all blood glucose measurements during the ICU stay. The primary outcome was all-cause mortality at in-hospital, 30-day, 90-day, and 365-day. Restricted cubic spline (RCS) analysis was used to determine the detailed relationship between GV and mortality and the optimal GV threshold for dichotomization. Propensity score matching and multivariable logistic regression were employed to control for confounders.
Results
The study cohort experienced substantial mortality, with rates of 16.0% (in-hospital), 19.8% (30-day), 26.4% (90-day), and 33.9% (365-day). RCS analysis revealed a significant nonlinear, J-shaped relationship between GV and all mortality end points (p < 0.001), characterized by an initial modest increase in risk followed by a sharp escalation beyond a specific threshold. The optimal GV thresholds for predicting in-hospital, 30-day, 90-day, and 365-day mortality were remarkably consistent (0.218, 0.217, 0.205, and 0.210, respectively), leading to a unified average dichotomization threshold of 0.213. After propensity score matching created a well-balanced cohort, high GV remained independently associated with significantly increased risks of mortality across all time points in the fully adjusted model (adjusted odds ratio (OR) for in-hospital mortality: 1.969, 95% confidence interval (CI): 1.379–2.811, p < 0.001; 30-day mortality OR: 1.744, 95% CI: 1.245–2.442, p = 0.001; 90-day mortality OR: 1.735, 95% CI: 1.268–2.375, p < 0.001; 365-day mortality OR: 1.815, 95% CI: 1.352–2.438, p < 0.001). Furthermore, adding GV significantly improved the predictive performance of the Oxford Acute Severity of Illness Score (OASIS) (area under the receiver operating characteristic curve (AUC) increased from 0.684 to 0.699 for in-hospital mortality, p = 0.008) and Charlson Comorbidity Index scores (AUC increased from 0.681 to 0.701, p = 0.001).
Conclusions
In critically ill septic patients with orthopedic trauma, elevated GV is a potent, independent predictor of short- and long-term mortality, exhibiting a J-shaped relationship with mortality risk. GV values above 21.3% are associated with significantly higher mortality risk, and this threshold may serve as a useful prognostic marker for risk stratification. Furthermore, GV provides incremental prognostic value beyond established risk scores, highlighting its potential as a key metric for risk stratification.
Supplementary Information
The online version contains supplementary material available at https://doi.org/10.1186/s10195-026-00940-8.
Keywords: Glycemic variability, Sepsis, Orthopedic trauma, Mortality, Intensive care unit, Propensity score matching
Introduction
Orthopedic trauma, frequently resulting from high-energy mechanisms such as traffic accidents and falls at construction sites, represents a significant and growing global health burden [1, 2]. Global estimates indicate that road traffic injuries alone cause millions of fractures annually, with the incidence of femoral shaft fractures from road traffic collisions reaching between 1 and 2.9 million cases worldwide [1], placing enormous strain on healthcare systems worldwide. Patients with severe orthopedic trauma requiring intensive care unit (ICU) admission are particularly vulnerable. To elaborate, the combined physiological insults of major tissue injury, often necessitating invasive interventions, and the consequent immune dysregulation significantly increase the risk of developing infections [3]. This vulnerability is further magnified when considering glucose metabolism, as patients with orthopedic trauma typically experience a strong systemic inflammatory response, significant metabolic stress, and frequent surgical interventions—all of which may contribute to disruptions in glucose metabolism and increased glycemic variability (GV) [4, 5]. Consequently, when such infections progress to sepsis—a dysregulated host response to infection—the already-compromised metabolic milieu can exacerbate organ dysfunction, leading to drastically worsened patient outcomes [3, 6]. The synergy between the initial traumatic injury and the systemic infection creates a profound and sustained inflammatory and metabolic cascade, leading to unacceptably high morbidity and mortality rates, with some studies reporting in-hospital mortality of 9.5%–19.7% [7, 8]. The management of these critically ill patients, therefore, urgently requires the identification of modifiable risk factors to improve prognostic accuracy and guide targeted interventions [6].
In the critical care setting, glucose metabolic dysregulation is a common manifestation of this systemic stress [9]. Beyond persistent hyperglycemia, which has been extensively studied [10, 11], increasing evidence underscores the importance of GV—defined as the amplitude of blood glucose fluctuations over time. Physiologically, GV is postulated to induce greater oxidative stress [12, 13], endothelial dysfunction [14], and proinflammatory cytokine release than sustained hyperglycemia, ultimately contributing to multi-organ failure. This has been demonstrated in some critical illnesses; for instance, in cohorts of patients with diabetes [15, 16], burned patients [17], and patients with critical cardiovascular diseases [18, 19], higher GV has been consistently and independently associated with increased mortality. However, despite its established theoretical and clinical importance in these populations, the specific impact of GV on the prognosis of septic orthopedic trauma patients—a unique cohort experiencing the dual physical and infectious insults of trauma and sepsis—has not been thoroughly investigated. Consequently, the precise nature and shape of the relationship between GV and short- and long-term mortality in this high-risk group remains largely undefined.
Therefore, this study aimed to comprehensively elucidate the relationship between GV and both short- and long-term all-cause mortality in a large cohort of critically ill adults with sepsis and orthopedic trauma, leveraging a high-fidelity clinical database. We hypothesized that increased GV would be independently associated with a significantly elevated mortality risk, and that this relationship might be nonlinear, potentially exhibiting threshold effects that simple linear models would miss. Furthermore, we posited that GV would provide incremental predictive value over established severity-of-illness scores. Clarifying this complex relationship is of paramount clinical importance, as it could offer a novel, modifiable physiological target for therapeutic strategies and substantially enhance risk stratification for this critically ill patient group.
Study population and methods
Study design and population
This was a retrospective observational cohort study conducted using 1946 patients at the Beth Israel Deaconess Medical Center between 2008 and 2022, available from the Medical Information Mart for Intensive Care IV (MIMIC-IV)database (version 3.1) [20]. The MIMIC-IV is a large, publicly available database comprising deidentified health-related data of patients admitted to the intensive care units (ICUs). Access was granted following completion of the required Protecting Human Research Participants training and examination. We included all adult patients (> 18 years) meeting the following criteria: 1) a diagnosis of orthopedic trauma, identified using International Classification of Diseases (ICD)−9 and ICD-10 codes; and 2) a concurrent diagnosis of sepsis, as defined by the Sepsis-3 criteria. The specific ICD codes for fracture trauma used for inclusion were ICD-9: 800–829 and ICD-10: S12, S22, and S32. In addition, the exclusion criteria were applied sequentially as follows: 1) patients not diagnosed with fracture trauma; 2) those who were discharged from the ICU or died within 12 h of admission, to ensure sufficient data for analysis; 3) those who were not diagnosed with sepsis; and 4) patients with missing GV data, defined as having fewer than two blood glucose measurements during their entire ICU stay, which precludes calculation of a meaningful coefficient of variation. The study flow diagram is presented in Fig. 1. The MIMIC-IV database received approval from the Institutional Review Board of Beth Israel Deaconess Medical Center, which waived the requirement for individual patient informed consent as all data were deidentified prior to public release. Our analysis used only aggregated, anonymized data, posing no risk of privacy breach and fully complying with the Declaration of Helsinki.
Fig. 1.
Study flow diagram. Flowchart illustrating the patient selection process from the MIMIC-IV database. The diagram details the sequential application of inclusion and exclusion criteria, culminating in the final analytical cohort of septic orthopedic trauma patients
Data extraction and clinical characteristics
The extracted variables encompassed demographics (age, sex, weight), comorbidities [Charlson Comorbidity Index, hypertension, congestive heart failure, diabetes, chronic pulmonary disease, and acute kidney injury (AKI)], severity-of-illness scores [Acute Physiology Score III (APS III), Simplified Acute Physiology Score II (SAPS II), sequential organ failure assessment (SOFA) score, Oxford Acute Severity of Illness Score (OASIS), and Glasgow Coma Scale (GCS) score], vital signs upon ICU admission (heart rate and blood pressure), laboratory parameters upon ICU admission [hemoglobin, red blood cell (RBC) count, red cell distribution width (RDW), blood glucose, creatinine, blood urea nitrogen (BUN), and international normalized ratio (INR)], and medication use (antihypertensive drugs, aspirin, corticosteroids, and immunosuppressants).
Definition of GV and outcomes
GV was quantified using the coefficient of variation (CV), a robust and clinically established metric chosen for its capacity to normalize glucose fluctuations relative to an individual’s mean glucose level, thereby facilitating comparisons across patients with differing glycemic baselines [18]. All glucose measurements were obtained via intermittent blood glucose testing, specifically from point-of-care fingerstick devices or arterial/venous blood gas analyzers, as part of routine clinical care; continuous glucose monitoring was not utilized in this dataset. The frequency of glucose sampling was clinically driven, typically ranging from every 1–4 h, with more frequent assessments during periods of insulin infusion, hemodynamic instability, or rapidly changing glucose trends, as per standard ICU protocols. For each patient, this study included all available glucose values recorded throughout the entire ICU stay, with GV defined specifically as the percentage derived from the standard deviation (SD) divided by the mean glucose value (CV = [SD/mean] × 100%) [18, 21]. The primary end point for this analysis was all-cause mortality, which was evaluated at four distinct intervals following ICU admission: during the index hospitalization, and at 30 days, 90 days, and 365 days postadmission.
Restricted cubic spline (RCS) analysis and GV dichotomization
To explore the potential nonlinear relationship between GV (as a continuous variable) and mortality, we performed RCS analysis with four knots at the 5th, 35th, 65th, and 95th percentiles. The nonlinearity was tested using a likelihood ratio test. The RCS analysis was conducted separately for each mortality outcome (in-hospital, 30-day, 90-day, and 365-day). For each outcome, the optimal GV cut-off value was determined by identifying the point on the fitted curve that maximized the Youden index. The final, unified GV threshold for dichotomization was calculated as the mean of these four optimal cut-off values. Subsequently, patients were categorized into two groups: the low-GV group and the high-GV group for subsequent analyses.
Propensity score matching (PSM) analysis
To minimize the potential confounding effects from baseline characteristics, we performed a 1:1 PSM between the low-GV and high-GV groups. The propensity score was estimated using a multivariable logistic regression model, with the high-GV group as the dependent variable. The model included all baseline variables, such as demographics, comorbidities, vital signs, and laboratory parameters. Matching was performed using the nearest-neighbor method without replacement, with a caliper width of 0.15 of the SD of the logit of the propensity score. The balance of covariates before and after matching was assessed using absolute standardized mean difference (SMD), with an SMD < 0.10 indicating a good balance.
Statistical analysis
Categorical variables were presented as frequencies with percentages and were compared using the Chi-square test or Fisher’s exact test, as appropriate. Continuous variables were expressed as mean ± SD and compared using the Student’s t-test or Mann–Whitney U test, respectively. For the initial cohort characterization, patients were stratified into quartiles (Q) on the basis of their GV values (Q1: lowest GV; Q4: highest GV). Inter-group comparisons across these quartiles for continuous variables were performed using one-way analysis of variance (ANOVA) or the Kruskal–Wallis’s test based on distribution normality, while categorical variable comparisons were conducted using the Chi-square test. The association between GV and mortality was evaluated using univariable and multivariable logistic regression models, both before and after PSM. Four models were constructed: model 1 (crude model); model 2 adjusted for age and sex; model 3 further adjusted for comorbidities (Charlson Comorbidity Index, diabetes, congestive heart failure); and model 4 (fully adjusted model) which included all variables in model 3 plus severity-of-illness scores and key laboratory parameters. The results were reported as odds ratios (ORs) with 95% confidence intervals (CIs), treating GV both as a continuous variable (per 1% increase) and as a binary variable (high versus low). To assess the incremental predictive value of GV, we compared the area under the receiver operating characteristic curve (AUC) of established severity scores (OASIS, Charlson Comorbidity Index, APS III, SOFA) with and without the addition of GV. The Delong test was used to compare the differences in AUC. A two-sided p-value < 0.05 was considered statistically significant. All statistical analyses were performed using R programming software (version 4.2.1; R Foundation for Statistical Computing).
Results
Baseline demographics, clinical characteristics, and outcomes of the cohort
The cohort had a mean age of 62.68 years (SD 21.00) and was predominantly male (64.1%) (Table 1). The vast majority were nonsmokers (92.6%) and without a history of alcohol abuse (97.7%). The patients were critically ill, as reflected by high severity-of-illness scores: the mean APS III was 48.71 (SD 20.82), SAPS II was 38.22 (SD 14.60), SOFA score was 6.65 (SD 3.56), and OASIS was 34.78 (SD 7.93). The mean Glasgow Coma Scale score was 10.81 (SD 3.84). Comorbidities were common, with a mean Charlson Comorbidity Index of 3.97 (SD 3.03). Specifically, 35.0% had hypertension, 24.3% had congestive heart failure, 22.3% had diabetes, and 22.0% had chronic pulmonary disease. AKI was highly prevalent, affecting 84.5% of the cohort. The majority of patients (70.1%) required invasive ventilation. Regarding medication use, 67.2% were on antihypertensive drugs, 31.1% were on aspirin, and 17.8% received corticosteroids. Key laboratory findings upon admission included a mean hemoglobin level of 10.98 g/dL (SD 2.27), RBC count of 3.59 × 1012/L (SD 0.76), RDW of 14.76% (SD 2.08), and an admission blood glucose level of 152.60 mg/dL (SD 69.79). Evidence of renal impairment and coagulation abnormality was noted, with mean creatinine at 1.28 mg/dL (SD 1.13), BUN at 23.67 mg/dL (SD 18.50), and INR at 1.41 (SD 0.77). The cohort faced significant mortality, with rates of 16.0% (in-hospital), 19.8% (30-day), 26.4% (90-day), and 33.9% (365-day).
Table 1.
Baseline characteristics and outcomes of the study population stratified by GV quartiles
| Characteristics | Overall | GV | p | |||
|---|---|---|---|---|---|---|
| Q1 (< 12.43%) | Q2 (> 12.43% and < 17.89%) | Q3 (> 17.89% and < 26.44%) | Q4 (> 26.44%) | |||
| n | 1946 | 487 | 486 | 486 | 487 | |
| Gender (male/female, %) | 1247/699 (64.1/35.9) | 321/166 (65.9/34.1) | 317/169 (65.2/34.8) | 320/166 (65.8/34.2) | 289/198 (59.3/40.7) | 0.094 |
| Age (years, mean (SD)) | 62.68 (21.00) | 60.14 (22.52) | 60.16 (22.06) | 64.64 (19.93) | 65.78 (18.69) | < 0.001 |
| Weight (kg, mean (SD)) | 80.50 (22.85) | 80.69 (22.95) | 81.50 (21.03) | 80.64 (24.12) | 79.16 (23.20) | 0.450 |
| Smoker (no/yes, %) | 1802/144 (92.6/7.4) | 450/37 (92.4/7.6) | 449/37 (92.4/7.6) | 446/40 (91.8/8.2) | 457/30 (93.8/6.2) | 0.650 |
| Alcohol abuse (no/yes, %) | 1902/44 (97.7/2.3) | 476/11 (97.7/2.3) | 472/14 (97.1/2.9) | 477/9 (98.1/1.9) | 477/10 (97.9/2.1) | 0.727 |
| AKI (no/yes, %) | 302/1644 (15.5/84.5) | 115/372 (23.6/76.4) | 84/402 (17.3/82.7) | 47/439 (9.7/90.3) | 56/431 (11.5/88.5) | < 0.001 |
| Congestive heart failure (no/yes, %) | 1474/472 (75.7/24.3) | 399/88 (81.9/18.1) | 387/99 (79.6/20.4) | 361/125 (74.3/25.7) | 327/160 (67.1/32.9) | < 0.001 |
| Cerebrovascular disease (no/yes, %) | 1762/184 (90.5/9.5) | 449/38 (92.2/7.8) | 446/40 (91.8/8.2) | 427/59 (87.9/12.1) | 440/47 (90.3/9.7) | 0.089 |
| Dementia (no/yes, %) | 1833/113 (94.2/5.8) | 453/34 (93.0/7.0) | 466/20 (95.9/4.1) | 456/30 (93.8/6.2) | 458/29 (94.0/6.0) | 0.271 |
| Chronic pulmonary disease (no/yes, %) | 1518/428 (78.0/22.0) | 397/90 (81.5/18.5) | 386/100 (79.4/20.6) | 369/117 (75.9/24.1) | 366/121 (75.2/24.8) | 0.055 |
| Hypertension (no/yes, %) | 1264/682 (65.0/35.0) | 320/167 (65.7/34.3) | 329/157 (67.7/32.3) | 297/189 (61.1/38.9) | 318/169 (65.3/34.7) | 0.179 |
| Diabetes (no/yes, %) | 1512/434 (77.7/22.3) | 429/58 (88.1/11.9) | 421/65 (86.6/13.4) | 377/109 (77.6/22.4) | 285/202 (58.5/41.5) | < 0.001 |
| APS III (mean (SD)) | 48.71 (20.82) | 41.94 (17.55) | 44.88 (17.26) | 50.49 (20.89) | 57.51 (23.50) | < 0.001 |
| SAPS II (mean (SD)) | 38.22 (14.60) | 34.18 (13.87) | 35.70 (12.41) | 39.69 (14.52) | 43.32 (15.66) | < 0.001 |
| SOFA (mean (SD)) | 6.65 (3.56) | 5.15 (2.67) | 6.20 (3.05) | 7.30 (3.49) | 7.95 (4.17) | < 0.001 |
| GCS (mean (SD)) | 10.81 (3.84) | 11.69 (3.44) | 10.77 (3.78) | 10.44 (3.93) | 10.33 (4.05) | < 0.001 |
| Charlson Comorbidity Index (mean (SD)) | 3.97 (3.03) | 3.35 (3.02) | 3.45 (2.87) | 4.28 (3.04) | 4.80 (2.96) | < 0.001 |
| OASIS (mean (SD)) | 34.78 (7.93) | 32.94 (7.42) | 34.02 (7.35) | 35.56 (7.99) | 36.60 (8.43) | < 0.001 |
| Invasive ventilation (no/yes, %) | 581/1365 (29.9/70.1) | 197/290 (40.5/59.5) | 126/360 (25.9/74.1) | 128/358 (26.3/73.7) | 130/357 (26.7/73.3) | < 0.001 |
| Immunosuppressant (no/yes, %) | 1907/39 (98.0/2.0) | 483/4 (99.2/0.8) | 482/4 (99.2/0.8) | 479/7 (98.6/1.4) | 463/24 (95.1/4.9) | < 0.001 |
| Aspirin (no/yes, %) | 1340/606 (68.9/31.1) | 364/123 (74.7/25.3) | 356/130 (73.3/26.7) | 326/160 (67.1/32.9) | 294/193 (60.4/39.6) | < 0.001 |
| Antihypertension (no/yes, %) | 639/1307 (32.8/67.2) | 217/270 (44.6/55.4) | 144/342 (29.6/70.4) | 133/353 (27.4/72.6) | 145/342 (29.8/70.2) | < 0.001 |
| Corticosteroids (no/yes, %) | 1599/347 (82.2/17.8) | 433/54 (88.9/11.1) | 414/72 (85.2/14.8) | 381/105 (78.4/21.6) | 371/116 (76.2/23.8) | < 0.001 |
| Heart rate (beats/min, mean (SD)) | 91.61 (21.35) | 90.43 (19.95) | 91.05 (21.01) | 91.37 (21.16) | 93.59 (23.10) | 0.108 |
| Respiratory rate (counts/min, mean (SD)) | 20.20 (5.83) | 19.77 (5.79) | 20.24 (5.87) | 20.17 (5.88) | 20.60 (5.77) | 0.168 |
| Systolic blood pressure (mmHg, mean (SD)) | 126.50 (25.56) | 127.51 (24.46) | 127.04 (24.49) | 125.45 (26.30) | 126.01 (26.93) | 0.579 |
| Diastolic blood pressure (mmHg, mean (SD)) | 70.87 (19.33) | 71.19 (17.81) | 72.16 (19.14) | 70.69 (20.48) | 69.43 (19.76) | 0.168 |
| RDW (%, mean (SD)) | 14.76 (2.08) | 14.45 (1.93) | 14.52 (1.86) | 14.87 (2.15) | 15.18 (2.27) | < 0.001 |
| RBC (× 1012/L, mean (SD)) | 3.59 (0.76) | 3.66 (0.72) | 3.64 (0.75) | 3.56 (0.80) | 3.51 (0.77) | 0.004 |
| Creatinine (mg/dL, mean (SD)) | 1.28 (1.13) | 1.15 (0.89) | 1.10 (0.81) | 1.32 (1.18) | 1.57 (1.46) | < 0.001 |
| BUN (mg/dL, mean (SD)) | 23.67 (18.50) | 20.91 (15.73) | 20.88 (15.72) | 25.16 (21.14) | 27.74 (19.89) | < 0.001 |
| Total calcium (mg/dL, mean (SD)) | 8.19 (0.86) | 8.22 (0.73) | 8.08 (0.78) | 8.24 (0.86) | 8.23 (1.03) | 0.011 |
| INR (mean (SD)) | 1.41 (0.77) | 1.38 (0.98) | 1.34 (0.54) | 1.40 (0.65) | 1.52 (0.82) | 0.002 |
| Glucose (mg/dL, mean (SD)) | 152.60 (69.79) | 127.20 (30.66) | 136.77 (35.13) | 149.22 (52.74) | 197.17 (107.92) | < 0.001 |
| Potassium (mEq/L, mean (SD)) | 4.21 (0.81) | 4.17 (0.71) | 4.14 (0.74) | 4.22 (0.88) | 4.30 (0.90) | 0.010 |
| Sodium (mEq/L, mean (SD)) | 138.30 (5.21) | 138.71 (4.62) | 138.25 (5.32) | 137.98 (5.28) | 138.27 (5.54) | 0.176 |
| Hemoglobin (g/dL, mean (SD)) | 10.98 (2.27) | 11.18 (2.20) | 11.12 (2.19) | 10.89 (2.35) | 10.71 (2.32) | 0.003 |
| In-hospital mortality (no/yes, %) | 1635/311 (84.0/16.0) | 436/51 (89.5/10.5) | 434/52 (89.3/10.7) | 403/83 (82.9/17.1) | 362/125 (74.3/25.7) | < 0.001 |
| 30-day mortality (no/yes, %) | 1560/386 (80.2/19.8) | 417/70 (85.6/14.4) | 415/71 (85.4/14.6) | 384/102 (79.0/21.0) | 344/143 (70.6/29.4) | < 0.001 |
| 90-day mortality (no/yes, %) | 1433/513 (73.6/26.4) | 395/92 (81.1/18.9) | 392/94 (80.7/19.3) | 348/138 (71.6/28.4) | 298/189 (61.2/38.8) | < 0.001 |
| 365-day mortality (no/yes, %) | 1287/659 (66.1/33.9) | 365/122 (74.9/25.1) | 349/137 (71.8/28.2) | 317/169 (65.2/34.8) | 256/231 (52.6/47.4) | < 0.001 |
GV, glycemic variability; SD, standard deviation; AKI, acute kidney injury; APS III, Acute Physiology Score III; SAPS II, Simplified Acute Physiology Score II; SOFA, sequential organ failure assessment; GCS, Glasgow Coma Scale; OASIS, Oxford Acute Severity of Illness Score; RDW, red cell distribution width; RBC, red blood cell; BUN, blood urea nitrogen; INR, international normalized ratio
Comparative baseline characteristics across GV quartiles
When patients were stratified into quartiles (Q1–Q4) of GV, numerous significant differences in baseline characteristics were observed (all p < 0.05) (Table 1), while demographic variables such as gender, weight, smoking status, and alcohol abuse were comparable across all groups. Notably, patients in the highest GV quartile (Q4) were significantly older (65.78 years) than those in the lowest quartiles (Q1: 60.14 years, p < 0.001). A pronounced gradient of increasing illness severity was evident with rising GV, demonstrated by progressively worse mean scores for APS III (Q1: 41.94 to Q4: 57.51), SAPS II (34.18–43.32), SOFA (5.15–7.95), and Charlson Comorbidity Index (3.35–4.80), alongside a declining Glasgow Coma Scale score (11.69–10.33). Comorbidities including diabetes (prevalence from 11.9% in Q1 to 41.5% in Q4), congestive heart failure (18.1%–32.9%), and AKI (76.4%–88.5%) were significantly more prevalent in higher GV quartiles. Similarly, the use of medications such as corticosteroids, aspirin, and immunosuppressants was significantly higher in Q3 and Q4. Laboratory parameters also reflected a more adverse profile with increasing GV, featuring significantly elevated levels of RDW, creatinine, BUN, glucose, and INR, and lower levels of hemoglobin and RBC. Most critically, a strong, graded association was found between GV and both short- and long-term mortality, with all-cause mortality rates rising significantly from Q1 to Q4 for in-hospital (10.5%–25.7%), 30-day (14.4%–29.4%), 90-day (18.9%–38.8%), and 365-day (25.1%–47.4%) outcomes (p < 0.001 for all). Vital signs, including heart rate and blood pressure, showed no significant differences across the quartiles.
RCS analysis of GV for mortality
To elucidate the precise relationship between GV and both short- and long-term all-cause mortality, the RCS analysis was performed. The analysis revealed that GV was significantly associated with in-hospital mortality (p < 0.001), with evidence of a nonlinear relationship (p for nonlinearity < 0.001) (Fig. 2A). The fitted curve demonstrated a characteristic J-shaped pattern: the risk of in-hospital mortality initially exhibited a modest increase with rising GV, followed by a sharp and significant escalation, before eventually plateauing at the highest levels of GV. The optimal threshold for in-hospital mortality, corresponding to the maximum Youden index, was identified at a GV value of 0.218. Similarly, significant nonlinear J-shaped associations were consistently observed for 30-day (Fig. 2B), 90-day (Fig. 2C), and 365-day (Fig. 2D) all-cause mortality (all p for association and nonlinearity < 0.001). The optimal GV thresholds for predicting these outcomes were 0.217, 0.205, and 0.210, respectively. Based on the mean value (0.213) of these four optimal thresholds, GV was subsequently dichotomized into high and low groups for all subsequent analyses.
Fig. 2.
Restricted cubic spline analysis of the relationship between glycemic variability and mortality. Nonlinear association between glycemic variability (GV, as a continuous variable) and the probability of mortality, as analyzed by restricted cubic splines. The solid lines represent the adjusted odds ratios, and the shaded areas indicate the 95% confidence intervals. The red dashed vertical line marks the optimal GV threshold for each risk stratification. A significant J-shaped relationship was observed for all mortality end points: A in-hospital, B 30-day, C 90-day, and D 365-day mortality (p for nonlinearity < 0.001 for all)
Association between GV and mortality before propensity score matching
The univariable and multivariable logistic regression analyses assessing the association between GV and mortality at different time points are detailed in Table 2. To elaborate, in the crude model (model 1), GV was significantly associated with increased risks of in-hospital, 30-day, 90-day, and 365-day mortality, both when treated as a continuous variable (per 1% increase) and as a binary variable (high versus low, all p < 0.001). These significant associations persisted across all progressively adjusted models (models 2–4). In the fully adjusted model (model 4), each 1% increase in GV remained an independent risk factor for all mortality end points, with adjusted ORs ranging from 1.015 to 1.019 (all p < 0.002). Similarly, high GV was consistently associated with a substantially elevated risk of mortality, with adjusted ORs of 1.926 (95% CI: 1.416–2.618) for in-hospital mortality, 1.757 (95% CI: 1.321–2.336) for 30-day mortality, 1.795 (95% CI: 1.374–2.344) for 90-day mortality, and 1.706 (95% CI: 1.321–2.202) for 365-day mortality (all p < 0.001).
Table 2.
Association between GV and mortality at different time points before propensity score matching analysis
| Models | In-hospital mortality | 30-day mortality | 90-day mortality | 365-day mortality | ||||
|---|---|---|---|---|---|---|---|---|
| OR (95% CI) | p | OR (95% CI) | p | OR (95% CI) | p | OR (95% CI) | p | |
| Model 1 | ||||||||
| Continuous GV (%) | 1.023 (1.015–1.030) | < 0.001 | 1.020 (1.013–1.027) | < 0.001 | 1.022 (1.015–1.029) | < 0.001 | 1.022 (1.015–1.029) | < 0.001 |
| Binary GV | < 0.001 | < 0.001 | < 0.001 | < 0.001 | ||||
| Low | Ref. | Ref. | Ref. | Ref. | ||||
| High | 2.465 (1.927–3.153) | 2.118 (1.690–2.653) | 2.200 (1.792–2.702) | 2.099 (1.731–2.545) | ||||
| Model 2 | ||||||||
| Continuous GV (%) | 1.022 (1.014–1.031) | < 0.001 | 1.018 (1.011–1.026) | < 0.001 | 1.020 (1.012–1.028) | < 0.001 | 1.020 (1.012–1.027) | < 0.001 |
| Binary GV | < 0.001 | < 0.001 | < 0.001 | < 0.001 | ||||
| Low | Ref. | Ref. | Ref. | Ref. | ||||
| High | 2.298 (1.761–3.000) | 1.949 (1.518–2.502) | 2.006 (1.587–2.534) | 1.915 (1.531–2.395) | ||||
| Model 3 | ||||||||
| Continuous GV (%) | 1.018 (1.009–1.027) | < 0.001 | 1.014 (1.005–1.022) | < 0.001 | 1.015 (1.007–1.024) | < 0.001 | 1.016 (1.008–1.024) | < 0.001 |
| Binary GV | < 0.001 | < 0.001 | < 0.001 | < 0.001 | ||||
| Low | Ref. | Ref. | Ref. | Ref. | ||||
| High | 1.990 (1.502–2.638) | 1.709 (1.311–2.228) | 1.760 (1.372–2.258) | 1.711 (1.347–2.174) | ||||
| Model 4 | ||||||||
| Continuous GV (%) | 1.019 (1.009–1.029) | < 0.001 | 1.015 (1.006–1.025) | 0.002 | 1.016 (1.007–1.025) | < 0.001 | 1.016 (1.007–1.024) | < 0.001 |
| Binary GV | < 0.001 | < 0.001 | < 0.001 | < 0.001 | ||||
| Low | Ref. | Ref. | Ref. | Ref. | ||||
| High | 1.926 (1.416–2.618) | 1.757 (1.321–2.336) | 1.795 (1.374–2.344) | 1.706 (1.321–2.202) | ||||
Model 1 (crude model): unadjusted analysis
Model 2: adjusted for demographic factors and key comorbidities (gender, admission age, weight, smoker status, alcohol abuse, acute kidney injury, congestive heart failure, cerebrovascular disease, dementia, chronic pulmonary disease, hypertension, and diabetes)
Model 3: model 2 + additional adjustments for initial laboratory values (red blood cell distribution width, red blood cell count, creatinine, blood urea nitrogen, total calcium, international normalized ratio, glucose, potassium, sodium, and hemoglobin)
Model 4: fully adjusted analysis
GV, glycemic variability; OR, odds ratio; CI, confident interval
Comparison of baseline characteristics before and after propensity score matching analysis
Prior to propensity score matching, substantial baseline imbalances were observed between patients with low and high GV. The high GV group was significantly older and had a greater burden of comorbidities, including a higher prevalence of diabetes, congestive heart failure, and AKI (all p < 0.001) (Table 3). They also presented with a higher severity of illness, as indicated by a lower Glasgow Coma Scale score and a higher Charlson Comorbidity Index (all p < 0.001). Significant differences were further noted in laboratory parameters; the high GV group demonstrated elevated markers of impaired renal function, dysglycemia, inflammation, and coagulation, alongside signs of anemia (all p < 0.05). Consequently, the high GV group had significantly elevated crude mortality rates across all follow-up periods, from in-hospital to 365-day mortality (all p < 0.001). These pronounced imbalances across demographic, clinical, and laboratory characteristics underscored the necessity for propensity score matching to mitigate confounding in subsequent analyses.
Table 3.
Baseline characteristics and outcomes of the study population stratified by GV level on the basis of the optimal threshold for all-cause mortality before propensity score matching analysis
| Characteristics | Overall | GV | p | SMD | |
|---|---|---|---|---|---|
| Low (< 21.30%) | High (> 21.30%) | ||||
| n | 1946 | 1216 | 730 | ||
| Gender (male/female, %) | 1247/699 (64.1/35.9) | 813/403 (66.9/33.1) | 434/296 (59.5/40.5) | 0.001 | 0.154 |
| Age (years, mean (SD)) | 62.68 (21.00) | 60.75 (22.06) | 65.90 (18.68) | < 0.001 | 0.252 |
| Weight (kg, mean (SD)) | 80.50 (22.85) | 81.04 (22.20) | 79.59 (23.89) | 0.175 | 0.063 |
| Smoker (no/yes, %) | 1802/144 (92.6/7.4) | 1123/93 (92.4/7.6) | 679/51 (93.0/7.0) | 0.652 | 0.025 |
| Alcohol abuse (no/yes, %) | 1902/44 (97.7/2.3) | 1189/27 (97.8/2.2) | 713/17 (97.7/2.3) | 1.000 | 0.007 |
| AKI (no/yes, %) | 302/1644 (15.5/84.5) | 224/992 (18.4/81.6) | 78/652 (10.7/89.3) | < 0.001 | 0.221 |
| Congestive heart failure (no/yes, %) | 1474/472 (75.7/24.3) | 969/247 (79.7/20.3) | 505/225 (69.2/30.8) | < 0.001 | 0.243 |
| Cerebrovascular disease (no/yes, %) | 1762/184 (90.5/9.5) | 1107/109 (91.0/9.0) | 655/75 (89.7/10.3) | 0.381 | 0.044 |
| Dementia (no/yes, %) | 1833/113 (94.2/5.8) | 1150/66 (94.6/5.4) | 683/47 (93.6/6.4) | 0.411 | 0.043 |
| Chronic pulmonary disease (no/yes, %) | 1518/428 (78.0/22.0) | 975/241 (80.2/19.8) | 543/187 (74.4/25.6) | 0.003 | 0.139 |
| Hypertension (no/yes, %) | 1264/682 (65.0/35.0) | 808/408 (66.4/33.6) | 456/274 (62.5/37.5) | 0.083 | 0.083 |
| Diabetes (no/yes, %) | 1512/434 (77.7/22.3) | 1049/167 (86.3/13.7) | 463/267 (63.4/36.6) | < 0.001 | 0.546 |
| GCS (mean (SD)) | 10.81 (3.84) | 11.05 (3.71) | 10.41 (4.02) | < 0.001 | 0.166 |
| Charlson Comorbidity Index (mean (SD)) | 3.97 (3.03) | 3.53 (3.01) | 4.72 (2.92) | < 0.001 | 0.401 |
| Invasive ventilation (no/yes, %) | 581/1365 (29.9/70.1) | 381/835 (31.3/68.7) | 200/530 (27.4/72.6) | 0.074 | 0.086 |
| Immunosuppressant (no/yes, %) | 1907/39 (98.0/2.0) | 1206/10 (99.2/0.8) | 701/29 (96.0/4.0) | < 0.001 | 0.207 |
| Aspirin (no/yes, %) | 1340/606 (68.9/31.1) | 885/331 (72.8/27.2) | 455/275 (62.3/37.7) | < 0.001 | 0.225 |
| Antihypertension (no/yes, %) | 639/1307 (32.8/67.2) | 435/781 (35.8/64.2) | 204/526 (27.9/72.1) | < 0.001 | 0.169 |
| Corticosteroids (no/yes, %) | 1599/347 (82.2/17.8) | 1040/176 (85.5/14.5) | 559/171 (76.6/23.4) | < 0.001 | 0.230 |
| Heart rate (beats/min, mean (SD)) | 91.61 (21.35) | 90.74 (20.57) | 93.06 (22.54) | 0.020 | 0.108 |
| Respiratory rate (counts/min, mean (SD)) | 20.20 (5.83) | 20.04 (5.84) | 20.46 (5.80) | 0.128 | 0.071 |
| Systolic blood pressure (mmHg, mean (SD)) | 126.50 (25.56) | 126.83 (24.86) | 125.97 (26.69) | 0.471 | 0.033 |
| Diastolic blood pressure (mmHg, mean (SD)) | 70.87 (19.33) | 71.67 (19.16) | 69.52 (19.55) | 0.018 | 0.111 |
| RDW (%, mean (SD)) | 14.76 (2.08) | 14.53 (1.96) | 15.14 (2.22) | < 0.001 | 0.290 |
| RBC (× 1012/L, mean (SD)) | 3.59 (0.76) | 3.65 (0.74) | 3.50 (0.79) | < 0.001 | 0.195 |
| Creatinine (mg/dL, mean (SD)) | 1.28 (1.13) | 1.15 (0.92) | 1.50 (1.39) | < 0.001 | 0.293 |
| BUN (mg/dL, mean (SD)) | 23.67 (18.50) | 21.40 (16.48) | 27.46 (20.93) | < 0.001 | 0.322 |
| Total calcium (mg/dL, mean (SD)) | 8.19 (0.86) | 8.17 (0.78) | 8.24 (0.97) | 0.088 | 0.078 |
| INR (mean (SD)) | 1.41 (0.77) | 1.36 (0.73) | 1.50 (0.82) | < 0.001 | 0.184 |
| Glucose (mg/dL, mean (SD)) | 152.60 (69.79) | 134.50 (35.86) | 182.75 (96.93) | < 0.001 | 0.660 |
| Potassium (mEq/L, mean (SD)) | 4.21 (0.81) | 4.17 (0.76) | 4.26 (0.89) | 0.026 | 0.102 |
| Sodium (mEq/L, mean (SD)) | 138.30 (5.21) | 138.37 (5.01) | 138.18 (5.53) | 0.434 | 0.036 |
| Hemoglobin (g/dL, mean (SD)) | 10.98 (2.27) | 11.15 (2.23) | 10.69 (2.32) | < 0.001 | 0.203 |
| In-hospital mortality (no/yes, %) | 1635/311 (84.0/16.0) | 1079/137 (88.7/11.3) | 556/174 (76.2/23.8) | < 0.001 | – |
| 30-day mortality (no/yes, %) | 1560/386 (80.2/19.8) | 1031/185 (84.8/15.2) | 529/201 (72.5/27.5) | < 0.001 | – |
| 90-day mortality (no/yes, %) | 1433/513 (73.6/26.4) | 967/249 (79.5/20.5) | 466/264 (63.8/36.2) | < 0.001 | – |
| 365-day mortality (no/yes, %) | 1287/659 (66.1/33.9) | 881/335 (72.5/27.5) | 406/324 (55.6/44.4) | < 0.001 | – |
GV, glycemic variability; SD, standard deviation; AKI, acute kidney injury; APS III, Acute Physiology Score III; SAPS II, Simplified Acute Physiology Score II; SOFA, sequential organ failure assessment; GCS, Glasgow Coma Scale; OASIS, Oxford Acute Severity of Illness Score; RDW, red cell distribution width; RBC, red blood cell; BUN, blood urea nitrogen; INR, international normalized ratio; SMD, standardized mean difference
Propensity score matching successfully created a well-balanced cohort of 1096 patients (548 in each GV group). After matching, all measured baseline characteristics—including demographics, comorbidities, vital signs, laboratory parameters, and medication use—showed no significant differences between the low and high GV groups (all p > 0.05) (Table 4), indicating effective mitigation of initial confounding. The distribution of propensity scores visually demonstrated successful matching adequacy (Supplementary Fig. 1 and Supplementary Fig. 2). Notably, while 20 of 31 variables (64.5%) had SMD exceeding 0.10 before matching, all variables achieved SMD below 0.10 after matching (Fig. 3), confirming excellent balance. Despite achieving this balance in baseline features, a significant disparity in mortality outcomes still persisted between the groups. Specifically, the high GV group continued to demonstrate significantly higher rates of in-hospital mortality (p = 0.001), 30-day mortality (p = 0.005), 90-day mortality (p = 0.003), and 365-day mortality (p = 0.001) compared with the low GV group. These findings suggest a strong and independent association between elevated GV and increased mortality risk, supporting the potential causal role of GV in adverse outcomes among patients with septic orthopedic trauma.
Table 4.
Baseline characteristics and outcomes of the study population stratified by GV level on the basis of the optimal threshold for all-cause mortality after propensity score matching analysis
| Characteristics | Overall | GV | p | SMD | |
|---|---|---|---|---|---|
| Low (< 21.30%) | High (> 21.30%) | ||||
| n | 1096 | 548 | 548 | ||
| Gender (male/female, %) | 667/429 (60.9/39.1) | 336/212 (61.3/38.7) | 331/217 (60.4/39.6) | 0.804 | 0.019 |
| Age (years, mean (SD)) | 64.69 (20.18) | 64.42 (20.71) | 64.96 (19.65) | 0.659 | 0.027 |
| Weight (kg, mean (SD)) | 80.15 (23.60) | 80.34 (22.20) | 79.96 (24.95) | 0.789 | 0.016 |
| Smoker (no/yes, %) | 1011/85 (92.2/7.8) | 507/41 (92.5/7.5) | 504/44 (92.0/8.0) | 0.821 | 0.020 |
| Alcohol abuse (no/yes, %) | 1066/30 (97.3/2.7) | 532/16 (97.1/2.9) | 534/14 (97.4/2.6) | 0.853 | 0.022 |
| AKI (no/yes, %) | 148/948 (13.5/86.5) | 79/469 (14.4/85.6) | 69/479 (12.6/87.4) | 0.426 | 0.053 |
| Congestive heart failure (no/yes, %) | 795/301 (72.5/27.5) | 399/149 (72.8/27.2) | 396/152 (72.3/27.7) | 0.892 | 0.012 |
| Cerebrovascular disease (no/yes, %) | 980/116 (89.4/10.6) | 487/61 (88.9/11.1) | 493/55 (90.0/10.0) | 0.623 | 0.036 |
| Dementia (no/yes, %) | 1027/69 (93.7/6.3) | 518/30 (94.5/5.5) | 509/39 (92.9/7.1) | 0.320 | 0.068 |
| Chronic pulmonary disease (no/yes, %) | 828/268 (75.5/24.5) | 421/127 (76.8/23.2) | 407/141 (74.3/25.7) | 0.361 | 0.059 |
| Hypertension (no/yes, %) | 691/405 (63.0/37.0) | 344/204 (62.8/37.2) | 347/201 (63.3/36.7) | 0.900 | 0.011 |
| Diabetes (no/yes, %) | 812/284 (74.1/25.9) | 407/141 (74.3/25.7) | 405/143 (73.9/26.1) | 0.945 | 0.008 |
| GCS (mean (SD)) | 10.43 (3.95) | 10.38 (3.93) | 10.47 (3.97) | 0.712 | 0.022 |
| Charlson Comorbidity Index (mean (SD)) | 4.30 (3.00) | 4.23 (3.07) | 4.38 (2.93) | 0.398 | 0.051 |
| Invasive ventilation (no/yes, %) | 308/788 (28.1/71.9) | 148/400 (27.0/73.0) | 160/388 (29.2/70.8) | 0.460 | 0.049 |
| Immunosuppressant (no/yes, %) | 1075/21 (98.1/1.9) | 538/10 (98.2/1.8) | 537/11 (98.0/2.0) | 1.000 | 0.013 |
| Aspirin (no/yes, %) | 740/356 (67.5/32.5) | 376/172 (68.6/31.4) | 364/184 (66.4/33.6) | 0.478 | 0.047 |
| Antihypertension (no/yes, %) | 323/773 (29.5/70.5) | 165/383 (30.1/69.9) | 158/390 (28.8/71.2) | 0.691 | 0.028 |
| Corticosteroids (no/yes, %) | 872/224 (79.6/20.4) | 444/104 (81.0/19.0) | 428/120 (78.1/21.9) | 0.261 | 0.072 |
| Heart rate (beats/min, mean (SD)) | 92.21 (21.15) | 92.02 (20.22) | 92.41 (22.05) | 0.761 | 0.018 |
| Respiratory rate (counts/min, mean (SD)) | 20.19 (5.88) | 20.19 (6.01) | 20.19 (5.76) | 0.997 | < 0.001 |
| Systolic blood pressure (mmHg, mean (SD)) | 127.11 (26.24) | 127.60 (26.01) | 126.61 (26.49) | 0.536 | 0.037 |
| Diastolic blood pressure (mmHg, mean (SD)) | 70.62 (19.99) | 71.08 (20.17) | 70.17 (19.82) | 0.451 | 0.046 |
| RDW (%, mean (SD)) | 14.96 (2.19) | 14.95 (2.32) | 14.97 (2.05) | 0.912 | 0.007 |
| RBC (× 1012/L, mean (SD)) | 3.54 (0.77) | 3.55 (0.75) | 3.53 (0.78) | 0.730 | 0.021 |
| Creatinine (mg/dL, mean (SD)) | 1.34 (1.17) | 1.34 (1.21) | 1.35 (1.14) | 0.883 | 0.009 |
| BUN (mg/dL, mean (SD)) | 25.22 (19.87) | 24.95 (20.07) | 25.50 (19.68) | 0.645 | 0.028 |
| Total calcium (mg/dL, mean (SD)) | 8.21 (0.90) | 8.21 (0.84) | 8.21 (0.96) | 0.896 | 0.008 |
| INR (mean (SD)) | 1.46 (0.89) | 1.47 (1.00) | 1.45 (0.77) | 0.822 | 0.014 |
| Glucose (mg/dL, mean (SD)) | 148.93 (49.47) | 148.24 (41.27) | 149.62 (56.52) | 0.646 | 0.028 |
| Potassium (mEq/L, mean (SD)) | 4.18 (0.85) | 4.17 (0.81) | 4.19 (0.89) | 0.741 | 0.020 |
| Sodium (mEq/L, mean (SD)) | 138.56 (5.30) | 138.49 (5.12) | 138.64 (5.48) | 0.631 | 0.029 |
| Hemoglobin (g/dL, mean (SD)) | 10.82 (2.30) | 10.84 (2.26) | 10.80 (2.33) | 0.769 | 0.018 |
| In-hospital mortality (no/yes, %) | 893/203 (81.5/18.5) | 469/79 (85.6/14.4) | 424/124 (77.4/22.6) | 0.001 | – |
| 30-day mortality (no/yes, %) | 846/250 (77.2/22.8) | 443/105 (80.8/19.2) | 403/145 (73.5/26.5) | 0.005 | – |
| 90-day mortality (no/yes, %) | 768/328 (70.1/29.9) | 407/141 (74.3/25.7) | 361/187 (65.9/34.1) | 0.003 | – |
| 365-day mortality (no/yes, %) | 686/410 (62.6/37.4) | 370/178 (67.5/32.5) | 316/232 (57.7/42.3) | 0.001 | – |
GV, glycemic variability; SD, standard deviation; AKI, acute kidney injury; APS III, Acute Physiology Score III; SAPS II, Simplified Acute Physiology Score II; SOFA, sequential organ failure assessment; GCS, Glasgow Coma Scale; OASIS, Oxford Acute Severity of Illness Score; RDW, red cell distribution width; RBC, red blood cell; BUN, blood urea nitrogen; INR, international normalized ratio; SMD, standardized mean difference
Fig. 3.
Covariate balance before and after propensity score matching. Plot visualizing the absolute standardized mean differences (SMD) for all covariates before and after propensity score matching. After matching, all variables achieved an SMD below the 0.10 threshold (red dashed vertical line), indicating a well-balanced cohort between the low and high glycemic variability groups
Association between GV and mortality after propensity score matching
Following propensity score matching, which balanced potential confounders between groups with high and low GV, the significant association between GV and mortality remained robust across all time points (Table 5). In the fully adjusted model (model 4), each 1% increase in GV, treated as a continuous variable, was independently associated with a higher risk of in-hospital (OR 1.022, 95% CI: 1.010–1.033, p < 0.001), 30-day (OR 1.018, 95% CI: 1.007–1.029, p < 0.001), 90-day (OR 1.018, 95% CI: 1.008–1.029, p < 0.001), and 365-day (OR 1.016, 95% CI: 1.007–1.026, p = 0.001) mortality. When analyzed as a binary variable, high GV was consistently associated with a significantly elevated risk of mortality, with the strongest effect observed for in-hospital mortality (OR 1.969, 95% CI: 1.379–2.811, p < 0.001). The risk estimates for the other time points were similarly significant, with odds ratios of 1.744 (30-day), 1.735 (90-day), and 1.815 (365-day). Notably, the strength of the association and the statistical significance generally increased with progressive model adjustment from model 1 to the fully adjusted model 4, underscoring the independent effect of GV on mortality risk.
Table 5.
Association between GV and mortality at different time points after propensity score matching analysis
| Models | In-hospital mortality | 30-day mortality | 90-day mortality | 365-day mortality | ||||
|---|---|---|---|---|---|---|---|---|
| OR (95% CI) | P | OR (95% CI) | P | OR (95% CI) | P | OR (95% CI) | P | |
| Model 1 | ||||||||
| Continuous GV (%) | 1.013 (1.004–1.022) | 0.005 | 1.011 (1.002–1.020) | 0.014 | 1.012 (1.003–1.020) | 0.006 | 1.011 (1.003–1.020) | 0.007 |
| Binary GV | 0.001 | 0.004 | 0.002 | 0.001 | ||||
| Low | Ref. | Ref. | Ref. | Ref. | ||||
| High | 1.736 (1.272–2.370) | 1.518 (1.141–2.019) | 1.495 (1.152–1.94) | 1.526 (1.193–1.952) | ||||
| Model 2 | ||||||||
| Continuous GV (%) | 1.016 (1.006–1.026) | 0.002 | 1.013 (1.004–1.023) | 0.007 | 1.014 (1.005–1.024) | 0.002 | 1.014 (1.005–1.023) | 0.002 |
| Binary GV | 0.001 | 0.005 | 0.002 | 0.001 | ||||
| Low | Ref. | Ref. | Ref. | Ref. | ||||
| High | 1.768 (1.278–2.445) | 1.547 (1.142–2.095) | 1.564 (1.176–2.081) | 1.620 (1.234–2.125) | ||||
| Model 3 | ||||||||
| Continuous GV (%) | 1.018 (1.008–1.028) | < 0.001 | 1.015 (1.005–1.025) | 0.003 | 1.016 (1.006–1.025) | 0.001 | 1.015 (1.006–1.025) | 0.001 |
| Binary GV | < 0.001 | 0.003 | 0.002 | < 0.001 | ||||
| Low | Ref. | Ref. | Ref. | Ref. | ||||
| High | 1.815 (1.303–2.528) | 1.590 (1.165–2.169) | 1.605 (1.197–2.153) | 1.674 (1.264–2.217) | ||||
| Model 4 | ||||||||
| Continuous GV (%) | 1.022 (1.010–1.033) | < 0.001 | 1.018 (1.007–1.029) | < 0.001 | 1.018 (1.008–1.029 | < 0.001 | 1.016 (1.007–1.026) | 0.001 |
| Binary GV | < 0.001 | 0.001 | 0.001 | < 0.001 | ||||
| Low | Ref. | Ref. | Ref. | Ref. | ||||
| High | 1.969 (1.379–2.811) | 1.744 (1.250–2.432) | 1.735 (1.270–2.370) | 1.815 (1.348–2.445) | ||||
Model 1 (crude model): unadjusted analysis
Model 2: adjusted for demographic factors and key comorbidities (gender, admission age, weight, smoker status, alcohol abuse, acute kidney injury, congestive heart failure, cerebrovascular disease, dementia, chronic pulmonary disease, hypertension, and diabetes)
Model 3: model 2 + additional adjustments for initial laboratory values (red blood cell distribution width, red blood cell count, creatinine, blood urea nitrogen, total calcium, international normalized ratio, glucose, potassium, sodium, and hemoglobin)
Model 4: fully adjusted analysis
GV, glycemic variability; OR, odds ratio; CI, confident interval
Incremental predictive value of GV for mortality risk stratification
The predictive performance of established severity scores with and without the incorporation of GV was assessed to evaluate its incremental predictive value (Table 6). The addition of GV consistently provided a statistically significant incremental improvement in the discriminatory power for the OASIS (Fig. 4) and Charlson Comorbidity Index scores (Fig. 5) across all mortality time points (all p < 0.05). For instance, the AUC for in-hospital mortality increased from 0.684 to 0.699 for OASIS (Delong test, p = 0.008) and from 0.681 to 0.701 for the Charlson Index (Delong test, p = 0.001). In contrast, the inclusion of GV did not lead to a significant increase in the AUC for the more physiology based APS III (Supplementary Fig. 3) and SOFA (Supplementary Fig. 4) scores at any time point (all p > 0.05), despite their generally higher baseline predictive accuracy. In addition, the SOFA score exhibited the highest predictive value for in-hospital mortality (AUC 0.782), which was not enhanced by GV.
Table 6.
Predictive performance of severity scores with and without GV for in-hospital and postdischarge mortality
| Scores | In-hospital mortality | 30-day mortality | 90-day mortality | 365-day mortality | ||||
|---|---|---|---|---|---|---|---|---|
| AUC (95% CI) | P | AUC (95% CI) | P | AUC (95% CI) | P | AUC (95% CI) | P | |
| OASIS | 0.684 (0.650–0.717) | 0.008 | 0.675 (0.645–0.706) | 0.021 | 0.681 (0.654–0.708) | 0.011 | 0.672 (0.646–0.697) | 0.002 |
| OASIS + GV | 0.699 (0.666–0.731) | 0.687 (0.657–0.716) | 0.693 (0.667–0.719) | 0.687 (0.662–0.711) | ||||
| Charlson Comorbidity Index | 0.681 (0.651–0.711) | 0.001 | 0.693 (0.665–0.721) | 0.006 | 0.737 (0.713–0.761) | 0.003 | 0.765 (0.743–0.786) | 0.006 |
| Charlson + GV | 0.701 (0.672–0.731) | 0.706 (0.679–0.733) | 0.747 (0.723–0.771) | 0.772 (0.750–0.793) | ||||
| APS III | 0.739 (0.710–0.768) | 0.188 | 0.731 (0.704–0.758) | 0.150 | 0.725 (0.700–0.749) | 0.135 | 0.707 (0.683–0.731) | 0.126 |
| APS III + GV | 0.743 (0.714–0.772) | 0.734 (0.708–0.761) | 0.729 (0.705–0.753) | 0.711 (0.688–0.735) | ||||
| SOFA | 0.782 (0.754–0.810) | 0.697 | 0.723 (0.695–0.752) | 0.259 | 0.699 (0.672–0.726) | 0.103 | 0.681 (0.656–0.707) | 0.071 |
| SOFA + GV | 0.783 (0.755–0.811) | 0.726 (0.698–0.755) | 0.705 (0.678–0.731) | 0.688 (0.663–0.713) | ||||
GV, glycemic variability; APS III, Acute Physiology Score III; AUC, area under the receiver operating characteristic curve; CI, confidence interval; OASIS, Oxford Acute Severity of Illness Score; SOFA, sequential organ failure assessment
Fig. 4.
Incremental predictive value of GV for In-hospital mortality when added to the OASIS score. Receiver operating characteristic (ROC) curves comparing the predictive performance for A in-hospital, B 30-day, C 90-day, and D 365-day mortality of the OASIS score alone (red line) versus the OASIS score combined with glycemic variability (GV, blue line). The addition of GV resulted in a statistically significant increase in the area under the curve (AUC), as determined by the Delong test
Fig. 5.
Incremental predictive value of GV for in-hospital mortality when added to the Charlson Comorbidity Index. Receiver operating characteristic (ROC) curves comparing the predictive performance for A in-hospital, B 30-day, C 90-day, and D 365-day mortality of the Charlson Comorbidity Index alone (red line) versus the Charlson Comorbidity Index combined with glycemic variability (GV, blue line). The addition of GV resulted in a statistically significant increase in the area under the curve (AUC), as determined by the Delong test
Discussion
Principal findings
This study found that elevated GV is a strong, dose-dependent, and independent predictor of both short- and long-term all-cause mortality in critically ill septic patients with orthopedic trauma. Our analysis yielded several key insights: (1) a clear gradient of increasing illness severity, comorbidity burden, and mortality rates was observed across ascending GV quartiles, establishing a direct exposure–response relationship; (2) the RCS analysis precisely characterized a significant J-shaped nonlinear association between GV and mortality, identifying a clear threshold (GV = 0.213) beyond which the risk of death escalates sharply, thereby providing an evidence-based cutoff for clinical risk stratification; and (3) most importantly, after rigorously controlling for a comprehensive set of potential confounders through propensity score matching and multivariable adjustment, the association between high GV and increased mortality risk remained robust and statistically significant across all follow-up periods, underscoring the independence of this relationship from other known prognostic factors. Finally, the finding that GV added incremental predictive value to certain severity scores (OASIS, Charlson Comorbidity Index) but not to others (APS III, SOFA) suggests that GV captures a distinct pathophysiological dimension—likely related to metabolic stability—that complements, rather than replaces, existing physiological and comorbidity-based assessments.
GV as an independent prognostic marker across critical illnesses
A growing body of evidence substantiates GV as a significant prognostic indicator across diverse critical care populations. In septic patients, Waeschle et al. [22] demonstrated that greater GV was independently associated with both critical hypoglycemia and elevated mortality, with this relationship being particularly pronounced in patients with septic shock. Similarly, Hoang et al. [23] found GV to be significantly associated with infectious complications in burn patients, while Akirov et al. [24] documented its correlation with prolonged hospitalization and increased mortality in surgical inpatients, with these associations remaining significant after multivariate adjustment. The universality of GV’s prognostic value is further evidenced by studies across cardiovascular diseases [18, 19], cerebrovascular accidents [25, 26], and oncological critical care [27], where GV consistently emerges as an independent predictor of adverse outcomes. Particularly noteworthy is the study by Oliveira et al. [27] which revealed that specific metrics of glucose fluctuation, particularly general amplitude, may serve as valuable prognostic indicators in critically ill cancer patients. Although the present study focuses specifically on patients with septic orthopedic trauma, evidence regarding GV in trauma patients without sepsis provides important context. In a cohort of critically ill trauma patients without sepsis, Liu et al. [28] conducted a large-scale retrospective study of 4009 adult trauma patients and reported that higher GV, quantified as the coefficient of variation, was independently associated with both 30-day mortality and 1-year mortality. Notably, the relationship pattern between GV and mortality exhibits population-specific characteristics. While a study report approximately linear relationships—as shown in critically ill patients with cerebrovascular disease where increasing GV was approximately linearly associated with the risk of cognitive impairment and in-hospital mortality [25]—our investigation revealed a distinctive J-shaped association in patients with septic orthopedic trauma. This nonlinear pattern suggests the existence of a critical threshold beyond which mortality risk escalates dramatically, potentially reflecting the unique physiological stress imposed by the combination of traumatic injury and systemic infection. The identification of this specific threshold at 21.3% provides a clinically actionable target for glycemic management in patients with septic orthopedic trauma. The robustness of our findings is strengthened by the application of propensity score matching, which effectively balanced numerous potential confounders and strengthened the inference of a causal relationship between elevated GV and increased mortality risk. To our knowledge, this represents the first comprehensive analysis specifically examining GV outcomes in patients with septic orthopedic trauma, a population at particularly high risk for metabolic dysregulation. The consistent association between GV and mortality across multiple time points (in-hospital to 365-day) underscores the enduring impact of glycemic fluctuations on long-term survival. Furthermore, the demonstration that GV provides an incremental predictive value beyond established scoring systems positions it as a valuable adjunct for risk stratification in clinical practice, potentially guiding more personalized glycemic management strategies for this complex patient cohort.
Potential mechanisms underlying GV-associated adverse outcomes
The detrimental impact of GV on clinical outcomes is mediated through interconnected biological pathways that converge on the four classical pathophysiological cascades central to polytrauma: systemic inflammation, coagulopathy, endothelial dysfunction with increased permeability, and post-traumatic immunosuppression [29–31]. First, regarding inflammation and oxidative stress, glucose oscillations trigger greater oxidative damage than sustained hyperglycemia, as evidenced by elevated biomarkers of lipid peroxidation and cellular membrane destabilization [12, 13]. This pro-oxidant state propagates systemically, affecting endothelial cells and various critical tissues [14], and may perpetuate proinflammatory pathways even after glycemic normalization. Second, concerning coagulopathy and microvascular thrombosis, rapid glucose variations enhance monocyte–endothelial adhesion and generate oxidative stress that drives vasoconstriction and microvascular thrombosis [32], thereby exacerbating the trauma-induced coagulopathy commonly seen in polytrauma patients. Third, regarding endothelial permeability and vascular dysfunction, GV induces endothelial injury through oxidative stress-mediated mechanisms, compromising the vascular barrier and contributing to capillary leak syndrome—a hallmark of both severe trauma and sepsis [32]. Fourth, concerning immunosuppression and infection susceptibility, emerging evidence suggests that GV may impair immune cell function and increase vulnerability to secondary infections, although the precise mechanisms remain under investigation [33]. In the polytrauma patient, these four cascades are not isolated but rather self-perpetuating; for example, GV-induced endothelial injury amplifies inflammation, which in turn promotes coagulopathy and further immunosuppression, creating a vicious cycle that worsens clinical outcomes [29]. Collectively, these mechanisms—encompassing oxidative stress, coagulopathy, endothelial dysfunction, and immune dysregulation—provide a compelling biological rationale for GV’s association with adverse outcomes in critically ill polytrauma patients. This multidimensional understanding underscores the imperative for therapeutic strategies that address not only absolute glucose levels but also glycemic stability in clinical practice.
Clinical implications and future directions
The findings from this study carry significant implications for clinical practice and future research. Despite consensus on the importance of glycemic control in critically ill patients, an appropriate target has not been established and likely varies across patient populations [32]. Our findings have several direct clinical implications. To begin with, the identification of a specific GV threshold (21.3%) provides a tangible prognostic marker for risk stratification in patients with septic orthopedic trauma, complementing traditional metrics such as mean glucose levels. Next, given GV’s incremental predictive value over established scoring systems (OASIS and Charlson Comorbidity Index), we suggest that GV monitoring could be integrated into standard critical care assessment for this population. Finally, these results demonstrate that at similar mean blood glucose values, GV carries significant prognostic implications, reinforcing that GV represents an important dimension of metabolic control beyond absolute glucose levels.
Regarding future directions, given the observational nature of this study, prospective research is needed to determine whether actively maintaining GV below the 21.3% threshold through targeted glucose management strategies can improve clinical outcomes. In addition, multi-center trials are required to validate our findings and to establish optimal GV-targeted protocols that balance glycemic stability with hypoglycemia prevention. Additionally, future studies should explore the potential role of continuous glucose monitoring and closed-loop systems in achieving stable glycemic control, as well as whether GV targets should be personalized on the basis of individual patient characteristics [32].
Limitations
This study has several limitations that warrant consideration. First, its retrospective and observational design inherently precludes the establishment of causality, as unmeasured or residual confounding factors may persist despite our robust use of propensity score matching and multivariable adjustment. Second, as a single-center study utilizing data from a tertiary care academic medical center in the USA, the generalizability of our findings to other healthcare settings, regions, or patient populations with different demographic characteristics may be compromised. Third, the definition of GV was based on all glucose measurements taken during the ICU stay, which could introduce bias as the frequency and timing of measurements were not standardized and may have been influenced by the patient’s clinical condition. Fourth, while we adjusted for a wide array of clinically relevant variables, we lacked data on specific nutritional support, details of glycemic management protocols, and postdischarge care, all of which could influence both GV and long-term outcomes. Thus, future prospective, multi-center studies with standardized glucose monitoring protocols are needed to validate these findings.
Conclusions
In this large retrospective cohort study of critically ill septic patients with orthopedic trauma, increased GV is an independent risk factor for short- and long-term mortality, exhibiting a J-shaped relationship. This study identifies a specific GV threshold (21.3%) beyond which mortality risk escalates substantially, and GV provides incremental prognostic value over established risk scores. Given the observational design, prospective studies are needed to determine whether targeting GV below this threshold improves clinical outcomes.
Supplementary information
Acknowledgements
Not applicable.
Author contributions
All authors contributed to the study design, conducted the data collection and analyses, and drafted the paper. All authors have read and approved the manuscript.
Funding
This study was supported by Beijing Natural Science Foundation (No. 7254420), National Natural Science Foundation of China (No. 82502988), China Postdoctoral Science Foundation General Funding (No. 2025M774436) and Special Funding (No. 2025T181188), the Second Batch of Medical and Health Talents under Hainan Province’s “South China Sea New Star” Project, and Clinical Center of Hainan Province.
Data availability
The data that support the findings of this study are derived from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. MIMIC-IV is a publicly available, deidentified clinical database containing comprehensive health-related data from patients admitted to intensive care units at the Beth Israel Deaconess Medical Center. Data will be available under reasonable request from the corresponding author.
Declarations
Ethics approval and consent to participate
This study adhered strictly to the protocols for using the MIMIC-IV database. The database itself received approval from the Institutional Review Board of Beth Israel Deaconess Medical Center, which waived the requirement for individual patient informed consent because all data were deidentified prior to public release. Authors completed the mandatory “Protecting Human Research Participants” training and passed the examination to obtain data access credentials. Our analysis utilized only aggregated, anonymized data, thereby posing no risk of privacy breach or harm to individual patients, in full compliance with the ethical principles of the Declaration of Helsinki.
Consent for publication
Not applicable.
Competing interests
The authors declare that they have no competing interests.
Footnotes
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Contributor Information
Tao Han, Email: hantao_16@163.com.
Feng Lin, Email: lyf1011@aliyun.com.
Feng Li, Email: Lifeng73666@163.com.
Mingxing Lei, Email: leimingxing2@sina.com, Email: leimingxing@301hospital.com.cn.
References
- 1.Agarwal-Harding KJ, Meara JG, Greenberg SL, Hagander LE, Zurakowski D, Dyer GS (2015) Estimating the global incidence of femoral fracture from road traffic collisions: a literature review. J Bone Joint Surg Am 97(6):e31 [DOI] [PubMed] [Google Scholar]
- 2.Chang FR, Huang HL, Schwebel DC, Chan AHS, Hu GQ (2020) Global road traffic injury statistics: challenges, mechanisms and solutions. Chin J Traumatol 23(4):216–218 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Dvorak JE, Lasinski AM, Romeo NM, Hirschfeld A, Claridge JA (2024) Fracture related infection and sepsis in orthopedic trauma: a review. Surgery 176(2):535–540 [DOI] [PubMed] [Google Scholar]
- 4.Paul SN, Nessel I, Puthucheary Z, Henson SM (2026) Sepsis and the immunometabolic inflammatory response. npj Metab Health Dis 4(1):4 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Willmann K, Moita LF (2024) Physiologic disruption and metabolic reprogramming in infection and sepsis. Cell Metab 36(5):927–946 [DOI] [PubMed] [Google Scholar]
- 6.Eguia E, Cobb AN, Baker MS, Joyce C, Gilbert E, Gonzalez R, Afshar M, Churpek MM (2019) Risk factors for infection and evaluation of Sepsis-3 in patients with trauma. Am J Surg 218(5):851–857 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Eguia E, Bunn C, Kulshrestha S, Markossian T, Durazo-Arvizu R, Baker MS, Gonzalez R, Behzadi F, Churpek M, Joyce C et al (2020) Trends, cost, and mortality from sepsis after trauma in the United States: an evaluation of the national inpatient sample of hospitalizations, 2012–2016. Crit Care Med 48(9):1296–1303 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Long Z, Tan S, Sun B, Qin Y, Wang S, Han Z, Han T, Lin F, Lei M (2025) Predicting in-hospital mortality in critical orthopedic trauma patients with sepsis using machine learning models. Shock 63(6):815–825 [DOI] [PubMed] [Google Scholar]
- 9.Imamura M, Clowes GH Jr., Blackburn GL, O’Donnell TF Jr., Trerice M, Bhimjee Y, Ryan NT (1975) Liver metabolism and glucogenesis in trauma and sepsis. Surgery 77(6):868–880 [PubMed] [Google Scholar]
- 10.Şimşek T, Şimşek HU, Cantürk NZ (2014) Response to trauma and metabolic changes: posttraumatic metabolism. Ulus Cerrahi Derg 30(3):153–159 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Douglas RG, Shaw JH (1989) Metabolic response to sepsis and trauma. Br J Surg 76(2):115–122 [DOI] [PubMed] [Google Scholar]
- 12.Papachristoforou E, Lambadiari V, Maratou E, Makrilakis K (2020) Association of glycemic indices (hyperglycemia, glucose variability, and hypoglycemia) with oxidative stress and diabetic complications. J Diabetes Res. 10.1155/2020/7489795 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Rodrigues R, de Medeiros LA, Cunha LM, Garrote-Filho MDS, Bernardino Neto M, Jorge PT, Resende ES, Penha-Silva N (2018) Correlations of the glycemic variability with oxidative stress and erythrocytes membrane stability in patients with type 1 diabetes under intensive treatment. Diabetes Res Clin Pract 144:153–160 [DOI] [PubMed] [Google Scholar]
- 14.Huang Y, Yue L, Qiu J, Gao M, Liu S, Wang J (2022) Endothelial dysfunction and platelet hyperactivation in diabetic complications induced by glycemic variability. Horm Metab Res 54(7):419–428 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Lee S, Zhou J, Wong WT, Liu T, Wu WKK, Wong ICK, Zhang Q, Tse G (2021) Glycemic and lipid variability for predicting complications and mortality in diabetes mellitus using machine learning. BMC Endocr Disord 21(1):94 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Shi C, Liu S, Yu HF, Han B (2020) Glycemic variability and all-cause mortality in patients with diabetes receiving hemodialysis: a prospective cohort study. J Diabetes Complicat 34(4):107549 [DOI] [PubMed] [Google Scholar]
- 17.Pisarchik AN, Pochepen ON, Pisarchyk LA (2012) Increasing blood glucose variability is a precursor of sepsis and mortality in burned patients. PLoS ONE 7(10):e46582 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Wang F, Guo Y, Tang Y, Zhao S, Xuan K, Mao Z, Lu R, Hou R, Zhu X (2025) 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(1):199 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.He HM, Zheng SW, Xie YY, Wang Z, Jiao SQ, Yang FR, Li XX, Li J, Sun YH (2024) Simultaneous assessment of stress hyperglycemia ratio and glycemic variability to predict mortality in patients with coronary artery disease: a retrospective cohort study from the MIMIC-IV database. Cardiovasc Diabetol 23(1):61 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Johnson AEW, Bulgarelli L, Shen L, Gayles A, Shammout A, Horng S, Pollard TJ, Hao S, Moody B, Gow B et al (2023) MIMIC-IV, a freely accessible electronic health record dataset. Sci Data 10(1):1 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Ceriello A, Monnier L, Owens D (2019) Glycaemic variability in diabetes: clinical and therapeutic implications. Lancet Diabetes Endocrinol 7(3):221–230 [DOI] [PubMed] [Google Scholar]
- 22.Waeschle RM, Moerer O, Hilgers R, Herrmann P, Neumann P, Quintel M (2008) The impact of the severity of sepsis on the risk of hypoglycaemia and glycaemic variability. Crit Care 12(5):R129 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Hoang K, Ly A, Hill D (2024) Effect of glycemic variability on infectious outcomes in critically ill burn patients. Burns 50(6):1555–1561 [DOI] [PubMed] [Google Scholar]
- 24.Akirov A, Shochat T, Dotan I, Diker-Cohen T, Gorshtein A, Shimon I (2019) Glycemic variability and mortality in patients hospitalized in general surgery wards. Surgery 166(2):184–192 [DOI] [PubMed] [Google Scholar]
- 25.Cai W, Li Y, Guo K, Wu X, Chen C, Lin X (2023) 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 22(1):315 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Lin J, Cai C, Xie Y, Yi L (2022) Acute glycemic variability and mortality of patients with acute stroke: a meta-analysis. Diabetol Metab Syndr 14(1):69 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Oliveira AP, Castro MDS, Lima DVM (2023) Glycemic variability and mortality in oncologic intensive care units. Rev Bras Enferm 76(4):e20220812 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Liu B, Zhang J, Song C, Miao J, Li X, Wang J, Lv R, Wang X, Liu X, Jia G (2025) Association of glycemic variability with short- and long-term mortality in critically ill patients with trauma. Sci Rep 16(1):2588 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Watt C, Sanchez-Rangel E, Hwang JJ (2020) Glycemic variability and CNS inflammation: reviewing the connection. Nutrients. 10.3390/nu12123906 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Risso A, Mercuri F, Quagliaro L, Damante G, Ceriello A (2001) Intermittent high glucose enhances apoptosis in human umbilical vein endothelial cells in culture. Am J Physiol Endocrinol Metab 281(5):E924–E930 [DOI] [PubMed] [Google Scholar]
- 31.Otto NM, Schindler R, Lun A, Boenisch O, Frei U, Oppert M (2008) Hyperosmotic stress enhances cytokine production and decreases phagocytosis in vitro. Crit Care 12(4):R107 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Hsu CW (2012) Glycemic control in critically ill patients. World J Crit Care Med 1(1):31–39 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Krinsley J, Preiser JC (2010) Intensive insulin therapy to control hyperglycemia in the critically ill: a look back at the evidence shapes the challenges ahead. Crit Care 14(6):330 [DOI] [PMC free article] [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
The data that support the findings of this study are derived from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. MIMIC-IV is a publicly available, deidentified clinical database containing comprehensive health-related data from patients admitted to intensive care units at the Beth Israel Deaconess Medical Center. Data will be available under reasonable request from the corresponding author.