Abstract
Background
In critically ill patients with sepsis complicated by type 2 diabetes mellitus (T2DM), renal dysfunction is prevalent; however, the association between dynamic serum creatinine trajectory patterns and short-term mortality risk remains inadequately characterized.
Methods
This study was conducted using data from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. The target population comprised adult intensive care unit (ICU) patients with sepsis complicated by T2DM. Group-Based Trajectory Modeling (GBTM) was employed to categorize the patterns of serum creatinine changes within 72 hours following ICU admission. Multivariable Cox proportional hazards models and Kaplan-Meier (K-M) survival curves were constructed to assess the association between trajectory groups and 28-day all-cause mortality, while restricted cubic spline (RCS) curves were utilized to examine the trend of short-term mortality risk in relation to serum creatinine levels. Furthermore, subgroup analyses and mediation effect analysis of continuous renal replacement therapy (CRRT) were performed.
Results
A total of 1909 patients were included in the study, with 439 deaths occurring within 28 days, corresponding to a 28-day all-cause mortality rate of approximately 23.0%. Three distinct serum creatinine trajectory patterns were identified: low-stable, moderate-increasing, and persistently high. Compared with the low-stable trajectory group, patients in the moderate-increasing and persistently high groups exhibited significantly elevated 28-day mortality risk, which remained statistically significant following multivariable adjustment, with adjustments for age, gender, temperature, platelet count, albumin, blood urea nitrogen, pH, renal disease, severe liver disease, mechanical ventilation, 24-hour urine output, and the Charlson Comorbidity Index (model 2: HR = 1.50, 95% CI: 1.16-1.93, p = 0.02; model 3: HR = 1.46, 95% CI: 1.04-2.05, p = 0.03). RCS analysis revealed a significant nonlinear association between serum creatinine levels and mortality risk. The subgroup analysis results revealed significant interaction effects across gender and CRRT subgroups. Mediation analysis demonstrated that CRRT may exert a partial mediating effect on the association between creatinine elevation and mortality risk.
Conclusion
The early trajectory of serum creatinine changes following ICU admission is independently associated with 28-day mortality in critically ill patients with sepsis complicated by T2DM. Sustained higher serum creatinine levels are significantly associated with an increased risk of mortality.
Keywords: 28-day mortality, intensive care unit, MIMIC-IV, sepsis complicated by type 2 diabetes mellitus, serum creatinine
1. Introduction
Sepsis is among the leading causes of mortality in intensive care unit (ICU) patients and is associated with high incidence and case-fatality rates (1). Patients with type 2 diabetes mellitus (T2DM) are particularly vulnerable to sepsis-associated acute kidney injury (S-AKI) (2). This vulnerability may be related to preexisting metabolic dysregulation, impaired immune responses, and reduced renal reserve. Consequently, this population has a higher risk of in-hospital mortality and long-term adverse renal outcomes (3).
Serum creatinine remains a widely used biomarker for assessing kidney function. In S-AKI, serum creatinine often changes dynamically during the early course of critical illness (4). Compared with a single creatinine value, dynamic changes in creatinine may better capture the progression of renal dysfunction. These changes include the magnitude, rate, and duration of creatinine elevation, which have been associated with short-term mortality (5). Although the Kidney Disease: Improving Global Outcomes (KDIGO) criteria are widely used for AKI staging, they rely mainly on threshold-based changes and may not fully capture heterogeneous renal function trajectories in critically ill patients. Therefore, a trajectory-based approach may provide additional prognostic information beyond conventional static measures.
Current clinical evaluation of S-AKI relies largely on peak serum creatinine values or KDIGO staging at a single time point (6). This approach may overlook important information contained in early serial creatinine changes during ICU admission. Patients with similar peak creatinine levels may have different temporal patterns, such as stable, improving, or persistently high trajectories, which may correspond to different risks of adverse outcomes (7). Previous studies have described temporal AKI patterns in sepsis, such as transient, persistent, or recurrent AKI (4). However, most of these studies focused on the general sepsis population.
Importantly, evidence remains limited regarding whether early serum creatinine trajectories can identify distinct mortality-risk subgroups specifically among critically ill patients with both sepsis and T2DM. This subgroup may have unique metabolic and renal vulnerability, but it has not been adequately examined using rigorous trajectory modeling. In addition, continuous renal replacement therapy (CRRT) is commonly used in severe S-AKI. CRRT may reflect both therapeutic intervention and disease severity. However, its role in the association between creatinine trajectories and mortality remains insufficiently understood, particularly in observational critical care data (8).
Leveraging the MIMIC-IV database, this study extracted serial serum creatinine measurements from patients with sepsis and T2DM during the first 72 hours after ICU admission. Group-based trajectory modeling (GBTM) was used to identify distinct creatinine trajectory patterns (1, 9, 10). We then examined whether these trajectory groups were associated with 28-day all-cause mortality. Multivariable Cox proportional hazards models were constructed to evaluate the association between creatinine trajectories and mortality. Restricted cubic splines were used to assess the non-linear association between early creatinine levels and mortality risk. Subgroup analyses were performed to explore whether the associations were consistent across clinically relevant subgroups. Finally, an exploratory mediation analysis was conducted to assess whether CRRT may partly account for the association between creatinine trajectories and mortality, while recognizing that CRRT may also represent disease severity and treatment selection.
This study aims to identify distinct subtypes of early ICU serum creatinine trajectories among patients with sepsis and T2DM using GBTM; assess the independent association between each identified trajectory subtype and 28-day all-cause mortality; and examine whether continuous CRRT moderates the association between creatinine trajectory and mortality—specifically, through a formal mediation analysis. Collectively, these objectives seek to advance dynamic renal function monitoring and refine risk stratification for mortality in this high-risk patient population.
2. Materials and methods
2.1. Data source
This study was carried out using data extracted from the MIMIC-IV 3.1 database. The first author, Yang Shu, was granted full access to the database and tasked with extracting data related to diabetic patients with sepsis (Credential ID: 62274870). Additionally, this study strictly adheres to the STROBE guidelines.
2.2. Participants
In this study, cases of sepsis and T2DM were identified using the International Classification of Diseases, 9/10th Revision, Clinical Modification (ICD-9/10-CM) coding systems. The detailed list of codes is provided in Supplementary Table 1. Given that this study was conducted using a retrospective database, sepsis was defined based on documented diagnostic codes rather than a prospective bedside Sepsis-3 assessment. While ICD-based definitions are widely employed in MIMIC database studies, this approach may not fully encompass all clinical components specified in the Sepsis-3 definition, such as suspected infection and acute organ dysfunction. The definition of sepsis complicated with T2DM was defined as the concurrent fulfillment of diagnostic criteria for both sepsis and T2DM during the same hospitalization period. Eligibility for study inclusion required simultaneous satisfaction of the following three criteria: (i) First admission to the ICU; (ii) Age ≥ 18 years; (iii) At least one serum creatinine measurement obtained daily within the 3 days prior to ICU admission. Following rigorous screening, a total of 1, 909 patients meeting all inclusion criteria were enrolled in the final analytical cohort (see Figure 1). The requirement of at least one daily serum creatinine measurement prior to ICU admission was designed to ensure the adequacy of trajectory estimation. However, this criterion may preferentially enroll patients who received more intensive clinical monitoring or presented with greater disease severity.
Figure 1.
Flow chart of study population inclusion and exclusion.
2.3. Data extraction
This study employed Navicat Premium in conjunction with Structured Query Language (SQL) for data extraction. The extracted data encompassed demographic characteristics, vital signs, laboratory indicators, disease severity scores, comorbidities and clinical interventions. The primary endpoint was 28-day all-cause mortality following ICU admission. Furthermore, this study incorporated CRRT as a potential mediating variable to further investigate its mechanistic role in the prognostic pathway of patients with sepsis complicated by diabetes mellitus.
2.4. Covariate selection
Covariates were selected using a combination of the Boruta algorithm and clinical relevance (Supplementary Figure 1). Variable selection was conducted by integrating the Boruta algorithm, insights from prior literature, and expert clinical judgment. Variables deemed clinically relevant to sepsis severity, renal dysfunction, and mortality were retained in the multivariable models even if their importance ranking in the Boruta analysis was relatively low. This hybrid strategy was adopted to mitigate the risk of excluding clinically meaningful confounders while minimizing overfitting. Variables confirmed as important by Boruta and considered clinically meaningful were included in the final multivariable model, including age, sex, temperature, platelet count, albumin, BUN, pH, chronic kidney disease, severe liver disease, mechanical ventilation, 24-hour urine output, and Charlson Comorbidity Index.
2.5. GBTM
GBTM was employed to identify distinct longitudinal patterns of serum creatinine levels during the early ICU stay. Latent class mixed models were fitted using the lcmm package in R. Creatinine trajectories were modeled via polynomial functions incorporating linear and quadratic time terms, and models with varying numbers of trajectory groups were evaluated (11). To assess the adequacy of the group-based trajectory model, several recommended diagnostic metrics were examined (Supplementary Table 2): posterior probability of group membership (average posterior probability, AvePP), odds of correct classification (OCC), estimated group probability (Prob), and the proportion of individuals assigned to each group (Group rate). A well-fitted model is generally characterized by an AvePP > 0.70 and an OCC > 5 (11). In the present study, the AvePP values for the three trajectory groups were 0.957, 0.945, and 0.969, respectively, indicating excellent classification accuracy. The corresponding OCC values were all substantially higher than the recommended threshold, confirming the reliability of individual assignment to their most likely trajectory group. Furthermore, the estimated group probabilities were highly consistent with the observed proportions of participants assigned to each trajectory group, which further supports the adequacy and stability of the final three-group trajectory model (Figure 2). More importantly, we compared candidate group-based trajectory models specifying two, three, and four distinct trajectories. The corresponding Bayesian Information Criterion (BIC) values were 15, 576.12, 13, 088.87, and 11, 760.22, respectively—indicating progressively better absolute model fit with increasing complexity. However, the 4-group solution introduced a small, clinically marginal subgroup comprising only 10.6% of the cohort and yielded minimal incremental insight beyond the 3-group structure. In contrast, the 3-group model achieved excellent classification performance (average posterior probabilities: 0.945–0.969; odds of correct classification: 30.2–109.8) while ensuring robust subgroup representation and clear clinical differentiation among trajectories. Given its superior balance of statistical adequacy, parsimony, classification reliability, and substantive interpretability, the 3-group model was selected as the final analytic framework (Supplementary Table 3).
Figure 2.
Distinct early serum creatinine trajectory patterns identified by group-based trajectory modeling.
2.6. Statistical analysis
Baseline characteristics were summarized using descriptive statistics: categorical variables were presented as frequencies (n, %), with intergroup comparisons conducted via the chi-square test or Fisher’s exact test; continuous variables that followed a normal distribution were expressed as mean ± standard deviation, and intergroup differences were assessed using the independent samples t-test; for non-normally distributed continuous variables, values were reported as median, with intergroup comparisons performed using the Wilcoxon rank-sum test. A multivariate Cox proportional hazards regression model was employed to assess the association between serum creatinine trajectory groups and 28-day mortality, the proportional hazards assumption was assessed using Schoenfeld residuals; Kaplan-Meier (K-M) survival curves were generated to compare survival disparities across different trajectory groups; RCS curves were utilized to characterize the dose-response relationship between serum creatinine levels and clinical outcomes, while potential interactions were explored via forest plots; mediation analysis was performed to examine the mediating effect of CRRT. To assess the robustness of the identified creatinine trajectory patterns, a sensitivity analysis was conducted using serum creatinine measurements collected within the first 7 days following ICU admission. Group-based trajectory modeling was then repeated with identical modeling strategies and selection criteria to ensure methodological consistency. Variables with over 40% missing data were excluded from the analysis. For the remaining variables, missing values were addressed via multiple imputation by chained equations (MICE). Five imputed datasets were generated and pooled in accordance with Rubin’s rules. All statistical analyses were performed using R software (version 4.1.2) and Free Statistics software (version 1.9). A two-tailed P value < 0.05 was considered statistically significant.
3. Results
3.1. Baseline characteristics of study individuals
A total of 1, 909 patients were enrolled in this study, with 803 cases (42.1%) assigned to Trajectory 1, 683 cases (35.6%) to Trajectory 2, and 423 cases (22.2%) to Trajectory 3 (Table 1). Among these groups, Trajectory 3 exhibited the highest proportion of male patients. Patients in the higher trajectory groups were more likely to present with comorbid kidney disease and oliguria. Notably, the proportion of patients undergoing CRRT and mechanical ventilation was significantly higher in Trajectory 3, which may indicate a more severe clinical status. Specifically, among patients receiving CRRT, the intergroup differences in proportions were statistically significant. Patients in the lower trajectory groups had lower levels of WBC count, anion gap, BUN, potassium, INR, and PT. As the creatinine trajectory transitioned from the stable group (Trajectory 1) to the increasing group (Trajectory 3), patients showed a heavier disease burden and greater treatment requirements, further suggesting that distinct creatinine trajectories may reflect varying degrees of clinical severity.
Table 1.
Baseline characteristics of patients according to creatinine trajectory groups.
| Variables | Total (n = 1909) |
Trajectory 1 (n = 803) |
Trajectory 2 (n = 683) |
Trajectory 3 (n = 423) |
p |
|---|---|---|---|---|---|
| Demographics | |||||
| Age(year) | 68.0 (59.0, 77.0) | 66.0 (56.5, 77.0) | 71.0 (62.0, 79.0) | 67.0 (58.0, 74.5) | < 0.001 |
| Gender, n (%) | < 0.001 | ||||
| Female | 754 (39.5) | 392 (48.8) | 242 (35.4) | 120 (28.4) | |
| Male | 1155 (60.5) | 411 (51.2) | 441 (64.6) | 303 (71.6) | |
| Marita status, n (%) | 0.059 | ||||
| Married | 792 (41.5) | 325 (40.5) | 294 (43) | 173 (40.9) | |
| Single | 536 (28.1) | 215 (26.8) | 181 (26.5) | 140 (33.1) | |
| Windowed | 239 (12.5) | 105 (13.1) | 94 (13.8) | 40 (9.5) | |
| Divorced | 342 (17.9) | 158 (19.7) | 114 (16.7) | 70 (16.5) | |
| Insurance, n (%) | 0.117 | ||||
| Medicare | 1240 (65.0) | 500 (62.3) | 472 (69.1) | 268 (63.4) | |
| Private | 344 (18.0) | 155 (19.3) | 105 (15.4) | 84 (19.9) | |
| Medicaid | 271 (14.2) | 127 (15.8) | 87 (12.7) | 57 (13.5) | |
| Other | 54 (2.8) | 21 (2.6) | 19 (2.8) | 14 (3.3) | |
| Race, n (%) | 0.021 | ||||
| White | 1143 (59.9) | 490 (61) | 417 (61.1) | 236 (55.8) | |
| Black | 269 (14.1) | 97 (12.1) | 91 (13.3) | 81 (19.1) | |
| Asian | 71 (3.7) | 35 (4.4) | 28 (4.1) | 8 (1.9) | |
| Hispanic/Latino | 73 (3.8) | 34 (4.2) | 25 (3.7) | 14 (3.3) | |
| Other/Unknown | 353 (18.5) | 147 (18.3) | 122 (17.9) | 84 (19.9) | |
| Vital signs | |||||
| 24-hour urine output (ml/kg/h) | 0.7 (0.3, 2.2) | 1.2 (0.6, 3.2) | 0.6 (0.3, 1.7) | 0.3 (0.2, 0.8) | < 0.001 |
| Heart rate (beats/minute) | 107.0 (93.0, 123.0) | 110.0 (97.0, 124.0) | 106.0 (92.0, 122.0) | 103.0 (88.0, 122.0) | < 0.001 |
| Systolic blood pressure (mmHg) | 86.0 (77.0, 95.0) | 89.0 (80.0, 97.0) | 85.0 (76.0, 94.0) | 84.0 (73.0, 92.0) | < 0.001 |
| Diastolic blood pressure (mmHg) | 44.0 (38.0, 50.0) | 45.0 (39.0, 52.0) | 43.0 (37.0, 49.0) | 42.0 (36.0, 48.0) | < 0.001 |
| Mean arterial pressure (mmHg) | 56.0 (50.0, 63.0) | 58.0 (52.0, 64.0) | 56.0 (49.0, 62.0) | 55.0 (49.0, 60.0) | < 0.001 |
| Respiratory rate (breath/minute) | 29.0 (25.0, 33.0) | 29.0 (25.0, 34.0) | 29.0 (25.0, 33.0) | 28.0 (24.0, 33.0) | 0.07 |
| Temperature (°C) | 36.5 (36.3, 36.7) | 36.6 (36.3, 36.8) | 36.5 (36.2, 36.7) | 36.4 (36.0, 36.7) | < 0.001 |
| SpO2 (%) | 92.0 (89.0, 94.0) | 92.0 (90.0, 95.0) | 92.0 (89.0, 94.0) | 92.0 (89.0, 94.0) | 0.097 |
| Laboratory indicators | |||||
| Glucose (g/dL) | 172.5 (136.0, 219.6) | 173.0 (136.7, 219.2) | 174.1 (141.4, 221.1) | 163.5 (124.2, 216.9) | 0.012 |
| Hematocrit (%) | 29.3 (25.1, 33.9) | 29.9 (25.6, 34.3) | 29.0 (25.2, 34.0) | 28.5 (24.3, 32.6) | < 0.001 |
| Hemoglobin (g/dL) | 9.5 (8.1, 11.0) | 9.8 (8.3, 11.3) | 9.3 (8.0, 11.0) | 9.1 (7.7, 10.6) | < 0.001 |
| Platelet (10(9)/L) | 173.0 (114.0, 243.0) | 181.0 (122.0, 251.0) | 168.0 (109.0, 236.5) | 165.0 (105.5, 237.0) | 0.002 |
| WBC (109/L) | 15.8 (10.9, 21.9) | 14.8 (10.6, 20.5) | 16.4 (11.1, 22.8) | 16.7 (11.6, 22.4) | < 0.001 |
| Albumin(g/dL) | 2.9 (2.6, 3.3) | 2.9 (2.6, 3.3) | 2.9 (2.6, 3.3) | 2.9 (2.5, 3.2) | 0.074 |
| Anion gap (mmol/L) | 18.0 (15.0, 21.0) | 16.0 (14.0, 19.0) | 18.0 (16.0, 22.0) | 21.0 (18.0, 25.0) | < 0.001 |
| Bicarbonate (mmol/L) | 19.0 (16.0, 22.0) | 20.0 (18.0, 23.0) | 19.0 (15.0, 22.0) | 18.0 (14.0, 21.0) | < 0.001 |
| BUN (mg/dL) | 37.0 (23.0, 58.0) | 22.0 (16.0, 32.0) | 43.0 (31.0, 58.0) | 67.0 (49.0, 91.5) | < 0.001 |
| Calcium (mg/dL) | 7.9 (7.4, 8.4) | 7.8 (7.3, 8.3) | 7.9 (7.4, 8.4) | 7.9 (7.3, 8.5) | 0.348 |
| Chloride (mmol/L) | 105.0 (101.0, 109.0) | 105.0 (102.0, 109.0) | 105.0 (101.5, 110.0) | 102.0 (97.0, 107.0) | < 0.001 |
| Sodium (mmol/L) | 139.0 (136.0, 142.0) | 139.0 (136.0, 142.0) | 140.0 (137.0, 143.0) | 138.0 (135.0, 141.0) | < 0.001 |
| Potassium (mmol/L) | 4.6 (4.1, 5.2) | 4.3 (4.0, 4.8) | 4.7 (4.3, 5.3) | 5.0 (4.5, 5.7) | < 0.001 |
| INR | 1.5 (1.3, 1.9) | 1.4 (1.2, 1.7) | 1.5 (1.3, 2.0) | 1.5 (1.3, 2.1) | < 0.001 |
| INR (seconds) | 15.9 (13.8, 20.5) | 15.4 (13.6, 18.8) | 16.5 (14.0, 21.6) | 16.2 (14.1, 22.8) | < 0.001 |
| PTT (seconds) | 34.5 (29.3, 48.4) | 32.9 (28.2, 45.0) | 34.9 (29.4, 49.5) | 37.9 (31.1, 53.5) | < 0.001 |
| ALT (IU/L) | 36.0 (20.0, 85.0) | 35.0 (20.0, 75.0) | 34.0 (20.0, 81.0) | 38.0 (20.0, 115.8) | 0.062 |
| ALP (IU/L) | 111.0 (78.0, 168.0) | 109.0 (78.0, 165.0) | 105.0 (74.5, 158.4) | 119.0 (84.5, 187.6) | < 0.001 |
| AST (IU/L) | 57.0 (30.0, 129.3) | 53.0 (28.0, 108.1) | 58.1 (32.0, 129.4) | 73.0 (33.0, 196.0) | < 0.001 |
| Total Bilirubin (mg/dL) | 0.9 (0.5, 1.9) | 0.9 (0.5, 1.6) | 0.9 (0.5, 2.0) | 1.0 (0.5, 2.3) | 0.137 |
| Lactate (mmol/L) | 2.4 (1.7, 3.6) | 2.2 (1.7, 3.1) | 2.7 (1.8, 4.0) | 2.4 (1.7, 3.8) | < 0.001 |
| PH | 7.3 (7.3, 7.4) | 7.3 (7.3, 7.4) | 7.3 (7.2, 7.4) | 7.3 (7.2, 7.4) | < 0.001 |
| Severity scores | |||||
| OASIS | 36.0 (29.0, 44.0) | 34.0 (28.0, 41.0) | 38.0 (31.0, 45.0) | 40.0 (32.0, 48.0) | < 0.001 |
| Charlson Comorbidity Index | 8.0 (6.0, 10.0) | 6.0 (5.0, 8.0) | 8.0 (6.0, 10.0) | 9.0 (7.0, 10.0) | < 0.001 |
| APSIII | 62.0 (48.0, 83.0) | 50.0 (40.0, 66.0) | 67.0 (53.0, 88.0) | 76.0 (61.0, 100.5) | < 0.001 |
| SAPSII | 43.0 (35.0, 53.0) | 37.0 (30.0, 46.0) | 46.0 (38.0, 55.5) | 51.0 (42.0, 61.0) | < 0.001 |
| Comorbidities | |||||
| Myocardial infarct, n (%) | 431 (22.6) | 131 (16.3) | 181 (26.5) | 119 (28.1) | < 0.001 |
| Congestive heart failure, n (%) | 807 (42.3) | 246 (30.6) | 331 (48.5) | 230 (54.4) | < 0.001 |
| Peripheral vascular disease, n (%) | 249 (13.0) | 81 (10.1) | 100 (14.6) | 68 (16.1) | 0.004 |
| Cerebrovascular disease, n (%) | 249 (13.0) | 102 (12.7) | 80 (11.7) | 67 (15.8) | 0.131 |
| Chronic pulmonary disease, n (%) | 500 (26.2) | 219 (27.3) | 185 (27.1) | 96 (22.7) | 0.179 |
| Mild liver disease, n (%) | 324 (17.0) | 117 (14.6) | 110 (16.1) | 97 (22.9) | < 0.001 |
| Renal disease, n (%) | 742 (38.9) | 91 (11.3) | 331 (48.5) | 320 (75.7) | < 0.001 |
| Malignant cancer, n (%) | 281 (14.7) | 140 (17.4) | 88 (12.9) | 53 (12.5) | 0.017 |
| Severe liver disease, n (%) | 151 (7.9) | 50 (6.2) | 58 (8.5) | 43 (10.2) | 0.041 |
| Metastatic solid tumor, n (%) | 129 (6.8) | 69 (8.6) | 39 (5.7) | 21 (5) | 0.022 |
| Treatment | |||||
| Aspirin, n (%) | 444 (23.3) | 151 (18.8) | 190 (27.8) | 103 (24.3) | < 0.001 |
| Vasoactive, n (%) | 1000 (52.4) | 356 (44.3) | 389 (57) | 255 (60.3) | < 0.001 |
| Mechanical ventilation, n (%) | 231 (12.1) | 81 (10.1) | 85 (12.4) | 65 (15.4) | 0.025 |
| CRRT, n (%) | 279 (14.6) | 28 (3.5) | 86 (12.6) | 165 (39) | < 0.001 |
SpO2, Saturation of peripheral oxygen; WBC, White Blood Cell; BUN, Blood Urea Nitrogen; INR, International Normalized Ratio; APSIII, Acute physiology score III; SAPSII, Simplified acute physiology score II; OASIS, Oxford acute severity of illness score.
Bold text indicates category headings used for organizational purposes only and does not denote statistical significance.
3.2. Association between serum creatinine and 28-day mortality
Cubic spline analysis (Supplementary Figure 2) revealed a significant nonlinear association between the average serum creatinine level during the first 3 days of ICU admission and 28-day mortality (P for non−linearity: 0.003). Using a reference value of approximately 1.4 mg/dL, the adjusted mortality risk increased progressively with elevated creatinine levels, particularly within the range of 1.5 to 3.0 mg/dL. This finding indicates that even moderately elevated serum creatinine levels may be associated with a significantly increased risk of short-term mortality. At higher creatinine levels, the risk curve plateaued, accompanied by a wider confidence interval, which may reflect reduced sample size and clinical heterogeneity. In the Cox proportional hazards regression model, distinct creatinine trajectory groups were significantly associated with 28-day mortality (Table 2). In the unadjusted model (Model 1), relative to the Trajectory 1 group, patients in the Trajectory 2 and Trajectory 3 groups exhibited a significantly elevated 28-day mortality risk (HR = 1.81, 95% CI: 1.44–2.29, P < 0.001;HR= 2.29, 95% CI: 1.78–2.94, P < 0.001). This indicates that patients with higher or persistently abnormal creatinine trajectories have a significantly increased short-term mortality risk compared to those with low and stable trajectories. Notably, the risk of death increases substantially with higher levels of creatinine trajectory changes, suggesting that dynamic changes in kidney function may help identify high-risk patients. In the further adjusted model (Model 2), the aforementioned associations remained statistically significant (Trajectory 2: HR = 1.53, 95% CI: 1.20–1.95, P = 0.001;Trajectory 3: HR = 1.50, 95% CI: 1.10–2.04, P = 0.011). In the fully adjusted model (Model 3), the association between creatinine trajectory groups and 28-day mortality remained robust (Trajectory 2: HR = 1.50, 95% CI: 1.16–1.93, P = 0.002; Trajectory 3: HR = 1.46, 95% CI: 1.04–2.05, P = 0.030). Even after adjusting for relevant confounding factors, higher creatinine trajectories remained associated with an approximately 50% increase in short-term mortality risk, indicating that dynamic changes in creatinine may hold independent prognostic value. Multivariable-adjusted results revealed that patients with higher creatinine change trajectories (Trajectories 2 and 3) exhibited a significantly elevated short-term mortality risk relative to those with low and stable trajectories (Trajectory 1). These findings further support the proposition that dynamic creatinine trajectories may be more conducive to early risk stratification and identification of high-risk patients compared to single-point creatinine measurements.
Table 2.
Multivariable Cox proportional hazards models for 28-day mortality according to creatinine trajectory groups.
| Variables | Model 1 | Model 2 | Model 3 | |||
|---|---|---|---|---|---|---|
| HR(95% CI) | P-value | HR(95% CI) | P-value | HR(95% CI) | P-value | |
| 28-day mortality | ||||||
| Creatinine trajectories | ||||||
| Trajectory 1 | 1 (Reference) | 1 (Reference) | 1 (Reference) | |||
| Trajectory 2 | 1.81 (1.44~2.29) | <0.001 | 1.53 (1.2~1.95) | 0.001 | 1.5 (1.16~1.93) | 0.002 |
| Trajectory 3 | 2.29 (1.78~2.94) | <0.001 | 1.5 (1.1~2.04) | 0.011 | 1.46 (1.04~2.05) | 0.03 |
| P for trend | <0.001 | 0.006 | 0.022 | |||
Model 1: No Adjusted.
Model 2: Adjusted for age、gender、temperature、platelets、albumin、BUN and PH.
Model 3: Adjusted for Model 2 Plus renal disease、severe liver disease、mechanical ventilation、24-hour urine output and charlson comorbidity index.
3.3. Cumulative mortality curve analysis
Kaplan-Meier curves grouped by trajectory patterns (Figure 3) revealed statistically significant differences in 28-day survival rates across the groups (p < 0.0001). Specifically, the low-creatinine trajectory group exhibited the highest survival rate, while the high-creatinine trajectory group had the lowest survival rate. These findings indicate a consistent prognostic disparity between trajectory patterns and survival outcomes.
Figure 3.
Kaplan–Meier survival curves for 28-day mortality according to creatinine trajectory groups.
3.4. Subgroup analysis
In the subgroup analysis (Figure 4), the association between trajectory grouping and 28-day mortality remained consistent across most subgroups: relative to Trajectory 1, Trajectory 2 and Trajectory 3 generally exhibited a higher risk of mortality, with a more distinct increasing trend observed in subgroups including those without pre-existing kidney disease, those not receiving mechanical ventilation, and those with higher urine output. Interaction tests revealed no significant effect modification by age, urine output stratification, pre-existing kidney disease, or mechanical ventilation. In contrast, significant interactions were identified for gender (P = 0.012) and receipt of CRRT (P = 0.034): among patients not receiving CRRT, the mortality risk of Trajectory 2 and Trajectory 3 was significantly higher than that of Trajectory 1; whereas among patients receiving CRRT, the risk difference between Trajectory 2/3 and Trajectory 1 was not statistically significant. However, given the small sample size of Trajectory 1 in the CRRT subgroup and the potential indication of higher disease severity by CRRT, this interaction result warrants further investigation.
Figure 4.
Forest plot of subgroup analyses evaluating the association between creatinine trajectory patterns and 28-day mortality.
3.5. Mediation analysis
Mediation analysis (Supplementary Table 4) demonstrated a significant overall association between the average creatinine level on the first day of ICU admission and 28-day mortality (total effect: OR = 1.10, 95% CI 1.05–1.16, P < 0.001). Elevated creatinine was found to significantly increase the likelihood of CRRT utilization (a path: OR = 1.47, 95% CI 1.38–1.57, P < 0.001), and CRRT was significantly associated with 28-day mortality (b path: OR = 3.44, 95% CI 2.59–4.58, P < 0.001). Following the inclusion of CRRT in the model, the direct effect of creatinine on 28-day mortality was no longer statistically significant (c’ path: OR = 1.01, 95% CI 0.96–1.08, P = 0.638), indicating that CRRT may act as a partial mediator in the pathway linking creatinine trajectories to mortality; however, it may also serve as a marker of disease severity and reflect treatment decisions driven by clinical judgment.
3.6. Sensitivity analysis
In the sensitivity analysis leveraging serum creatinine trajectories collected during the first 7 days following ICU admission, analogous three-group trajectory patterns were identified. The model exhibited excellent classification performance, with AvePP ranging from 0.976 to 0.986 and OCC values spanning 74.387 to 189.429 (Supplementary Table 5). The identified trajectory patterns were generally consistent with those observed in the primary 72-hour analysis, thereby supporting the robustness of the trajectory classification framework (Supplementary Figure 3).
4. Discussion
In this study, we used GBTM for modeling longitudinal serum creatinine profiles among ICU patients who have both sepsis and T2DM, identifying three clinically meaningful trajectory patterns in the current sample of interest, namely; stable low-levels (42%) and medium high level fluctuating group (36%) groups, and a persistently high-level deteriorating group. The patients who were categorized into the persistently high level trajectory group had a significantly higher 28-day all-cause mortality(HR = 2.29, 95% CI: 1.78 - 2.94), but persisted even when fully adjusting for confounders (HR = 1.46, 95% CI: 1.04 - 2.05). A potential interaction between CRRT and creatinine trajectories was observed for mortality outcomes (interaction P = 0.034). However, given the potential for confounding by indication and small subgroup sample sizes, this finding should be interpreted as exploratory. CRRT may reflect both treatment selection and the severity of underlying disease, rather than a definitive effect modifier.
Peak creatinine is less sensitive than creatinine trajectories in capturing the temporal course and dynamic nature of renal dysfunction. In this study, different creatinine trajectories may indicate heterogeneous patterns of early renal dysfunction rather than directly proving specific biological mechanisms. A persistently high or worsening creatinine trajectory may be associated with more severe renal impairment, impaired renal reserve, and systemic inflammatory burden. In patients with sepsis and T2DM, several mechanisms may plausibly contribute to this association. Sepsis-induced systemic inflammation can aggravate endothelial dysfunction, while T2DM-related glomerular hyperfiltration, tubulointerstitial fibrosis, and reduced renal reserve may increase susceptibility to septic kidney injury (12–14). As previously reported, baseline creatinine levels in patients with T2DM may be related to renal functional reserve: low creatinine may be associated with sarcopenia, whereas increased creatinine often indicates a substantial reduction in glomerular filtration rate (15). Under septic conditions, these patients may therefore be more vulnerable to rapid deterioration in renal function (4). Furthermore, T2DM-associated metabolic dysregulation, including hyperglycemia and insulin resistance, may amplify sepsis-related oxidative stress and mitochondrial injury, thereby contributing to renal tubular epithelial cell dysfunction (16). Consistent with this possibility, patients in the persistently high creatinine trajectory group showed higher inflammatory marker levels, such as WBC, in our cohort. However, these findings should be interpreted as associations rather than direct evidence of microvascular injury or metabolic dysregulation (17–19).
The majority of previous AKI research has used either the highest creatinine value, or a change in creatinine over two days to define stages according to the KDIGO classification (20), but these static measures are not able to reflect the heterogeneity of renal function evolution (12). Re-defining AKI as a dynamic syndrome by using trajectory modeling, these results are consistent with several other studies: Takkavatakarn et al. divided sepsis-related AKI patients into eight creatinine profile clusters and found that the cluster labeled “severe AKI but with mild but sustained recovery” was at increased risk for acute kidney disease (AKD) (4). In addition, trajectory analysis for creatinine after cardiac surgery also validated “persistent moderate-to high level” as an independent predictor of late mortality (21). The novelty of this study lies in its focus on the high-risk sepsis-T2DM subpopulation, demonstrating that trajectory-based classification enables more granular risk stratification against the backdrop of pre-existing diabetic nephropathy. Notably, GBTM has also been applied to analyze dynamic changes in the UACR in T2DM patients, revealing associations between distinct trajectories and cardiovascular events—further validating the utility of trajectory analysis in managing diabetic complications (22).
In the stratified analysis for CRRT, patients with a high creatinine trajectory had a higher risk of death among those receiving CRRT than among those not receiving CRRT. This finding suggests that CRRT use may identify a subgroup of patients with more severe renal dysfunction and critical illness. Although CRRT may partly lie on the pathway between renal dysfunction and mortality, it is also strongly influenced by clinical decision-making and disease severity. Day 1 mean creatinine was associated with higher odds of initiating CRRT (OR = 1.47; 95% CI: 1.38–1.57), and CRRT was also associated with mortality (OR = 3.44; 95% CI: 2.59–4.58). The mediation analysis was therefore considered exploratory. Although the indirect pathway through CRRT was observed, this should not be interpreted as definitive causal evidence, given the observational design and the potential for confounding by indication. The decision to initiate CRRT is based on clinical conditions such as refractory metabolic derangements, electrolyte imbalance, fluid overload, and hemodynamic instability, all of which are themselves associated with poor prognosis (23). Thus, CRRT may simultaneously represent a therapeutic intervention, a marker of disease severity, and a potential intermediate factor. Differences in the timing, modality, and intensity of CRRT may also influence outcomes independently of creatinine trajectory. In sex-stratified analyses, higher creatinine trajectories were also associated with mortality among women. However, this subgroup finding should be interpreted cautiously because the analysis was exploratory and may have been affected by residual confounding and limited subgroup size. Potential biological explanations, such as sex-related differences in renal vascular function, endothelial vulnerability, or hormonal regulation, remain speculative and were not directly tested in the present study (24–26). Further studies are needed to validate whether sex modifies the relationship between creatinine trajectories and outcomes in patients with sepsis and T2DM.
Our study suggest that real-time tracking of creatinine curves during the first three days in the ICU can help to identify patient clusters at risk for mortality, who were diagnosed with sepsis and type two diabetes mellitus, thus creating an important therapeutic window for tailored renoprotection interventions - such as optimized fluid management and avoidance of nephrotoxic agents (9, 27). Further integration of trajectory-based classification into clinical decision support systems holds promise for enhancing precision monitoring of this patient population.
5. Limitations
This study has several limitations that warrant acknowledgment. First, as a retrospective single-center study leveraging the MIMIC-IV database, residual confounding cannot be fully eliminated despite rigorous multivariable adjustment. Second, information related to CRRT was extracted from electronic health records and coding systems, which may introduce information bias and indication-related confounding. Third, serum creatinine levels may be confounded by non-steady-state physiological conditions, including muscle wasting, fluid balance disturbances, and altered creatinine synthesis. Consequently, creatinine trajectories may not fully capture true renal function. Future studies incorporating complementary biomarkers such as cystatin C could enhance the accuracy of renal function assessment (28). Notably, group-based trajectory modeling is a data-driven approach that hinges on modeling assumptions, including the number of trajectory groups and polynomial specifications. Alternative model selections may yield divergent classification results, and a certain degree of misclassification remains unavoidable. Thus, the identified trajectory groups should be interpreted as statistical constructs rather than definitive biological phenotypes. Furthermore, Sepsis was identified via ICD-9/10-CM diagnostic codes rather than prospective Sepsis-3 clinical adjudication, which may have introduced classification bias. In addition, the requirement for repeated creatinine measurements prior to ICU admission may have preferentially enrolled patients with more intensive monitoring or greater disease severity, potentially introducing selection bias. Finally, given that this study was conducted using a single intensive care unit database, the external generalizability of the findings to other populations and healthcare systems may be limited. External validation in multicenter prospective cohorts is therefore imperative. Multi-center prospective studies should be conducted in future for validation of the trajectory model, and further exploration of how to integrate multi-omics markers (e.g., metabolites, cytokines) to improve risk prediction (29, 30).
6. Conclusion
In critically ill patients with sepsis complicated by T2DM, early changes in serum creatinine are significantly associated with 28-day mortality. Patients with persistently elevated creatinine levels exhibit a higher risk of death compared to those with stable creatinine levels. The requirement for continuous renal replacement therapy may partially account for the association between early renal dysfunction and mortality.
Funding Statement
The author(s) declared that financial support was received for this work and/or its publication. This research was funded by Wuxi Science and Technology Development Fund Project (Y20232030) and Scientific Research of Wuxi Municipal Health Commission (Youth Program, Q202336).
Footnotes
Edited by: Tommaso Bucci, University of Perugia, Italy
Reviewed by: Enrico Tartaglia, University of Modena and Reggio Emilia, Italy
Anes Jogunčić, Public Health Institute of Canton Sarajevo, Bosnia and Herzegovina
Data availability statement
The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding authors.
Ethics statement
In the MIMIC-IV database, patients’ private information is encrypted. According to national legislation and institutional requirements, this study does not require ethical approval or informed consent.
Author contributions
SY: Conceptualization, Writing – original draft. LM: Conceptualization, Writing – review & editing. JZ: Writing – review & editing. YG: Writing – review & editing. CW: Writing – review & editing. XZ: Writing – review & editing. MZ: Funding acquisition, Writing – review & editing.
Conflict of interest
The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Generative AI statement
The author(s) declared that generative AI was not used in the creation of this manuscript.
Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
Supplementary material
The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fendo.2026.1822280/full#supplementary-material.
References
- 1. Wu K, Yu Y, Li L, Wu J, Xu J, Zheng B, et al. Association between the serum albumin-to-creatinine ratio and 28-day all-cause mortality in sepsis: a retrospective cohort study. Front Med (Lausanne). (2025) 12:1540647. doi: 10.3389/fmed.2025.1540647 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Xin Q, Xie T, Chen R, Wang H, Zhang X, Wang S, et al. Predictive nomogram model for major adverse kidney events within 30 days in sepsis patients with type 2 diabetes mellitus. Front Endocrinol (Lausanne). (2022) 13:1024500. doi: 10.3389/fendo.2022.1024500 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Han WW, Fang JJ. Impact of diabetes on clinical characteristics and prognosis in sepsis: a retrospective study. J Inflammation Res. (2025) 18:13823–34. doi: 10.2147/jir.S548570 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Takkavatakarn K, Oh W, Chan L, Hofer I, Shawwa K, Kraft M, et al. Machine learning derived serum creatinine trajectories in acute kidney injury in critically ill patients with sepsis. Crit Care. (2024) 28:156. doi: 10.1186/s13054-024-04935-x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Cai Q, Ye X, He Y, Liu Q, Huang J, Chen G, et al. Association between the lactate to creatinine ratio and hospital mortality in patients with pediatric sepsis: a cohort study. Pediatr Res. (2026). doi: 10.1038/s41390-025-04751-1 [DOI] [PubMed] [Google Scholar]
- 6. Zhuang C, Hu R, Li K, Liu Z, Bai S, Zhang S, et al. Machine learning prediction models for mortality risk in sepsis-associated acute kidney injury: evaluating early versus late CRRT initiation. Front Med (Lausanne). (2024) 11:1483710. doi: 10.3389/fmed.2024.1483710 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Azzopardi G, Mekhaimar A, Haines RW, Stephens TJ, Puthucheary Z, Prowle JR. Clinician assessment of kidney function from plasma creatinine values during critical illness: a scenario-based international multi-professional survey. J Crit Care. (2025) 85:154926. doi: 10.1016/j.jcrc.2024.154926 [DOI] [PubMed] [Google Scholar]
- 8. Wang Z, Zhang L, Xu F, Han D, Lyu J. The association between continuous renal replacement therapy as treatment for sepsis-associated acute kidney injury and trend of lactate trajectory as risk factor of 28-day mortality in intensive care units. BMC Emerg Med. (2022) 22:32. doi: 10.1186/s12873-022-00589-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Ying J, Zhou H, Zhang Y, Zhu S, Zhang G, Huang J. Association of different serum creatinine trajectories with 28-day mortality in patients with acute kidney injury on chronic kidney disease: based on the MIMIC-IV database. Eur J Med Res. (2025) 31:101. doi: 10.1186/s40001-025-03632-x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Hu Z, Song C, Zhang J. Elevated serum albumin-to-creatinine ratio as a protective factor on clinical outcomes among critically ill patients with sepsis: a retrospective study. Front Med (Lausanne). (2024) 11:1436533. doi: 10.3389/fmed.2024.1436533 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Mirza SS, Wolters FJ, Swanson SA, Koudstaal PJ, Hofman A, Tiemeier H, et al. 10-year trajectories of depressive symptoms and risk of dementia: a population-based study. Lancet Psychiatry. (2016) 3:628–35. doi: 10.1016/s2215-0366(16)00097-3 [DOI] [PubMed] [Google Scholar]
- 12. Wan J, Zou Y, Kuang M, Xiong S, He W, Zhu Y, et al. Predictive value of dynamic creatinine cumulative exposure and trajectory classification onmortality in severe acute pancreatitis: a multicenter retrospective cohort study. Int J Med Inform. (2026) 206:106182. doi: 10.1016/j.ijmedinf.2025.106182 [DOI] [PubMed] [Google Scholar]
- 13. Zhao P, Yan J, Pan B, Liu J, Fu S, Cheng J, et al. Association between the risk of non-alcoholic fatty liver disease in patients with type 2 diabetes and chronic kidney disease. Diabetes Metab Syndr Obes. (2022) 15:1141–51. doi: 10.2147/dmso.S356497 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Shin JY. Low serum creatinine to cystatin C ratio is independently associated with sarcopenia and high carotid plaque score in patients with type 2 diabetes. Nutr Metab Cardiovasc Dis. (2022) 32:1454–62. doi: 10.1016/j.numecd.2022.02.005 [DOI] [PubMed] [Google Scholar]
- 15. Song DK, Hong YS, Sung YA, Lee H. Association of serum creatinine levels and risk of type 2 diabetes mellitus in Korea: a case control study. BMC Endocr Disord. (2022) 22:4. doi: 10.1186/s12902-021-00915-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Hoshikawa T, Okamoto N, Natsuyama T, Fujii R, Ikenouchi A, Honma Y, et al. Associations of serum cytokines, growth factors, and high-sensitivity C-reactive protein levels in patients with major depression with and without type 2 diabetes mellitus: an explanatory investigation. Neuropsychiatr Dis Treat. (2022) 18:173–86. doi: 10.2147/ndt.S350121 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Vachliotis ID, Polyzos SA. Osteoprotegerin/receptor activator of nuclear factor-kappa B ligand/receptor activator of nuclear factor-kappa B axis in obesity, type 2 diabetes mellitus, and nonalcoholic fatty liver disease. Curr Obes Rep. (2023) 12:147–62. doi: 10.1007/s13679-023-00505-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Li S, Li N, Li L, Zhan J. Sex difference in the association between serum versican and albuminuria in patients with type 2 diabetes mellitus. Diabetes Metab Syndr Obes. (2023) 16:3631–39. doi: 10.2147/dmso.S434287 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Zomorodian SA, Shafiee M, Karimi Z, Masjedi F, Roshanshad A. Assessment of the relationship between 25-hydroxyvitamin D and albuminuria in type 2 diabetes mellitus. BMC Endocr Disord. (2022) 22:171. doi: 10.1186/s12902-022-01088-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Zhang Y, Peng W, Huang J, Zhang W, Jiang P, He Y, et al. Causal associations of iron status with the renal function and diabetic nephropathy in patients with diabetes mellitus: a two-sample mendelian randomization study. J Diabetes Res. (2025) 2025:6658794. doi: 10.1155/jdr/6658794 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Cho JS, Choi M, Shim JK, Park JH, Shin HJ, Choi HW, et al. Association of serum creatinine trajectories with 1-year mortality after valvular heart surgery: a retrospective cohort study. Int J Surg. (2024) 110:7097–105. doi: 10.1097/js9.0000000000001933 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Masrouri S, Chevli PA, Bancks MP, Mirzai S, Lima JAC, Bluemke DA, et al. Longitudinal trajectories of albuminuria and risk of subclinical and clinical heart failure, atrial fibrillation, and coronary heart disease: the MESA study. Eur J Prev Cardiol. (2025). doi: 10.1093/eurjpc/zwaf699 [DOI] [PubMed] [Google Scholar]
- 23. Lim MW, Szücs A, Ong D, Van Der Lubbe SCC. Incidence of hyperkalaemia and major creatinine elevation after renin-angiotensin system inhibitor initiation in a Singapore primary care cohort. BMC Prim Care. (2026) 27:82. doi: 10.1186/s12875-026-03195-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. van Eeghen SA, Nokoff NJ, Vosters TG, Oosterom-Eijmael MJP, Cherney DZI, van Valkengoed IGM, et al. Unraveling sex differences in kidney health and CKD: a review of the effect of sex hormones. Clin J Am Soc Nephrol. (2024) 20:301–10. doi: 10.2215/cjn.0000000642 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Lee HJ, Min L, Gao J, Matta S, Drel V, Saliba A, et al. Female protection against diabetic kidney disease is regulated by kidney-specific AMPK activity. Diabetes. (2024) 73:1167–77. doi: 10.2337/db23-0807 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Clotet-Freixas S, Zaslaver O, Kotlyar M, Pastrello C, Quaile AT, McEvoy CM, et al. Sex differences in kidney metabolism may reflect sex-dependent outcomes in human diabetic kidney disease. Sci Transl Med. (2024) 16:eabm2090. doi: 10.1126/scitranslmed.abm2090 [DOI] [PubMed] [Google Scholar]
- 27. Ghosh E, Eshelman L, Lanius S, Schwager E, Pasupathy KS, Barreto EF, et al. Estimation of baseline serum creatinine with machine learning. Am J Nephrol. (2021) 52:753–62. doi: 10.1159/000518902 [DOI] [PubMed] [Google Scholar]
- 28. Eisinger F, Neumann M, Wörn M, Fritsche A, Heyne N, Peter A, et al. Comparison of GFR estimation in patients with diabetes mellitus using the EKFC and CKD-EPI equations. J Nephrol. (2025) 38:707–16. doi: 10.1007/s40620-024-02202-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Okamoto N, Hoshikawa T, Ikenouchi A, Natsuyama T, Fujii R, Igata R, et al. Comparison of serum metabolomics pathways and patterns between patients with major depressive disorder with and without type 2 diabetes mellitus: an exploratory study. J Integr Neurosci. (2023) 22:13. doi: 10.31083/j.jin2201013 [DOI] [PubMed] [Google Scholar]
- 30. Xiu L, Yao XA, Jiang T. Correlation between 25-hydroxyvitamin D level and cardiac diastolic dysfunction in Chinese adults with early-onset type 2 diabetes mellitus: a cross-sectional study. Diabetes Metab Syndr Obes. (2021) 14:1823–31. doi: 10.2147/dmso.S299422 [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 original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding authors.