Trajectory of early creatinine change is associated with the prognosis of sepsis patients

Abstract

Background

Numerous studies have investigated the creatinine in patients with sepsis, and many meaningful results have been obtained. However, few studies have measured the creatinine multiple times and established trajectory models for patients with sepsis. Thus, we explored the influence of 72-h creatinine trajectory in patients with sepsis through latent class trajectory models.

Methods

Data from patients with sepsis were extracted from the Medical Information Mart for Intensive Care-IV database for inclusion in a retrospective cohort study. Three creatinine values within 72 h after hospitalization were extracted, and the interval between each creatinine value was 24 h. The creatinine measured three times after admission was analyzed using latent class trajectory models to construct a trajectory model. The outcome was 28-day mortality. The survival probability of different trajectory groups was investigated using Kaplan-Meier (K-M) analysis, and the relationship between different creatinine trajectories and 28-day mortality risk was investigated using Cox proportional-hazards regression model.

Results

This study included 8663 patients with sepsis. Five different creatinine trajectories were identified based on model-fit criteria. Class 1 started at a low level and remained stable at that low level; Class 2 showed a gradual upward trend; Class 3 stabilized at a high level; Class 4 exhibited a trend of initially decreasing and then increasing; Class 5 presented an inverted U-shape. The K-M analysis indicated that patients in class 2 had the lowest probability of survival. Univariate and multivariate Cox regression analysis indicated that, with class 1 as a reference, patients in class 2 had the highest 28-day mortality risk (P < 0.001). Subgroup analysis indicated that a nominal interaction occurred between CRRT group and creatinine trajectory in the 28-day mortality (P < 0.05).

Conclusion

Trajectory of early creatinine change is associated with the prognosis of sepsis patients, patients with persistently elevated creatinine in the first three days had a higher mortality risk.

Clinical trial

Non-applicable

Supplementary Information

The online version contains supplementary material available at 10.1186/s12879-026-12655-6.

Introduction

Sepsis is a life-threatening condition caused by dysregulated host response to infection, leading to systemic inflammation, tissue damage and organ failure [1]. It is also one of the leading causes of intensive care unit (ICU) admission and mortality [2, 3]. Advances in modern medical technology have led to gradual improvements in intensive care unit (ICU) treatment. However, the sepsis-related mortality rate remains up at to 40%, resulting in significant social burden and healthcare costs [4, 5]. Therefore, the identification of death-related indicators in sepsis patients is important for timely intervention and improved prognosis of sepsis.

Serum creatinine, an indicator of renal function, was linked to oxidative stress, endothelial function, and inflammation [6, 7], played a pivotal role in diagnosing sepsis-associated acute kidney injury [8], levels of serum creatinine upon admission could aid in predicting the onset of adverse events in sepsis patients [9]. However, the relationship between trajectories of repeated measures of creatinine collected from daily clinical practice and the prognosis of septic patients is not yet clear.

Latent class trajectory models (LCTM) is a popular longitudinal data modeling method used to identify groups with different trends over time as well as explore the characteristics of the trends and trajectories in various subgroups [2, 10, 11]. The aim of this study is to identify Trajectory of early creatinine change by leveraging the public electronic health record using the latent class trajectory models (LCTM) [12], which integrates a latent class mixed model with a survival model, and to explore the relationship between creatinine and an unfavorable prognosis.

Methods

Data source

This study is a retrospective, observational study based on data from the Medical Information Mart for Intensive Care IV (MIMIC-IV) v2.2 database [13]. The MIMIC-IV database contains thorough, de-identified data from over 50,000 admissions from 2008 to 2019 at the Beth Israel Deaconess Medical Center in Boston, Massachusetts. The data include demographics, clinical measurements, interventions, history, clinical laboratory testing, and medical data. This study used data from MIMIC-IV 2.2.

Because the study was conducted entirely using a third-party, anonymous, publicly available database, the requirement for individual patient consent was waived. Moreover, because the MIMIC project has been approved by the institutional review board (IRB) of Beth Israel Deaconess Medical Center, the study was exempt from the requirement for IRB approval from our organization. We completed the courses required for qualification to use the database and gained access to it (Record ID: 62595271).

Study population

We extracted patients diagnosed with sepsis, patients with sepsis were identified with the sepsis 3.0 criteria. Patients were excluded if they met any of the following criteria: (1) age < 18 years; (2) Non-first ICU admission; (3) length of ICU stay < 72 h; (4) Patients was diagnosed with chronic kidney disease; and (5) Creatinine is not measured at 24 h,48 h,72 h. For patients who were admitted to the ICU more than once, only the first ICU stay was included in the analysis. Ultimately, a total of 8663 ICU septic patients from MIMIC-IV were included in our analysis.

Variables measurement and outcomes

After using the database’s identifiers (stay_id) to determine the study population, we utilized Structured Query Language (SQL) to acquire all the information from the MIMIC-IV database. Longitudinal creatinine measured continuously over a 72 h period after ICU admission were collected from the databases. Three creatinine values within 72 h after hospitalization were extracted, and the interval between each creatinine value was 24 h.We extracted variables including demographic information, laboratory tests, vital signs, disease severity scores, comorbidities and interventions during ICU stay. These variables included: age, sex (male, female), race (white, others), heart rate (b.p.m.), systolic blood pressure (SBP, mmHg), diastolic blood pressure (DBP, mmHg), respiratory rate (times per minute), temperature (°C), arterial partial pressure of oxygen (PaO₂, mmHg), arterial partial pressure carbon dioxide (PaCO₂, mmHg), pH, white blood cells (10⁹/L), red blood cell (10¹²/L), platelet (10⁹/L), haemoglobin (g/dL), prothrombin time (seconds), activated partial thromboplastin time(seconds), sodium (mmol/L), potassium (mmol/L), blood urea nitrogen (mg/dL), SOFA (score), Simplified Acute Physiology Score II (SAPS II, score), Glasgow Coma Scale (GCS, score), congestive heart failure(yes, no), cerebrovascular disease(yes, no), chronic pulmonary disease(yes, no), diabetes(yes, no), renal-disease(yes, no), use of vasopressor (yes, no), use of mechanical ventilation (yes, no), use of CRRT (yes, no). To minimize bias due to loss of data, patients with incomplete data were excluded from the final cohort.The outcome of this study was 28-day mortality and was identified by the difference between mortality time and ICU admission.

LCTM

In epidemiology, latent class trajectory models (LCTM) is usually used to describe exposure during the life course, which is a method to identify different groups with similar underlying trajectories in longitudinal data with the purpose of simplifying heterogeneous populations into homogeneous patterns or categories [12]. Longitudinal studies can be difficult to summarize due to the large amount of data provided by long-term studies. LCTM allows researchers to summarize data from multiple time points in an unbiased manner to identify patterns as this approach does not require prior knowledge of the number or direction of existing tracks in a given population. Therefore, LCTM is a useful tool to summarize data to identify high-risk groups, which can then be targeted for intervention or prevention strategies [14, 15].

We used the LCTM to classify 72 h creatinine trajectories in patients with sepsis. Indicators reflecting the goodness of fit of the LCTM include log-likelihood, entropy, and information criteria such as the Akaike information criterion (AIC), Bayesian information criterion (BIC) [16]. The optimal number of classes was determined using the lowest information criteria, best goodness of fit, highest log-likelihood, and entropy.

In order to ensure the accuracy of the statistical analysis, we required the case number in each category to be at least 1% of the total cases. Moreover, a variety of posterior probability analyses were used to evaluate the model fitting effect. Posterior probability values ranged from 0% to 100%, with values closer to 100% indicating more accurate classification. To facilitate interpretability, we assigned labels to the trajectories on the basis of their modelled graphic patterns.

Statistical analysis

Data were extracted from the MIMIC-IV database using Navicat Premium 16 with Structured Query Language in PostgreSQL (version 12.0) [17]. Continuous variables are presented using medians and quartiles, and categorical variables are presented using frequencies and percentages. Survival probability curves were drawn using the Kaplan-Meier (K-M) analysis, and log-rank tests were used to compare survival probabilities between the trajectory categories. Cox proportional-hazards regression was used to identify differences in mortality risk between patients in different trajectory categories, and results were expressed as hazard ratios (HRs) and 95% confidence intervals (95% CIs). In Model I, only classes were analyzed and no other variables were adjusted. In Model II, age, race, gender, were included in the Cox proportional-hazards regression model and analyzed using forward partial likelihood estimation. In Model III, age, race, gender, heart rate, sbp, resprate, temperature, WBC, RBC, hemoglobin, PO2,PH, Na+,K+,PT, APTT, BUN, SOFA, SAPSII, congestive heart failure, cerebrovascular disease, vasopressors, menchanical ventilation, CRRT were included in the Cox proportional-hazards regression model and analyzed using forward partial likelihood estimation. Cox regression analysis was performed using the interaction test to determine whether there was an interaction between the above-mentioned subgroups and the effect of trajectories on 28-day mortality. Probability values of P < 0.05 were considered significant, and all statistical analyses were performed using R software (version 4.2.0).

Results

Among 29,929 patients with sepsis in MIMIC-IV Database, 21,266 patients were excluded and 8663 patients were included. Figure 1 illustrates the patient exclusion process.

Fig. 1.

Flowchart illustrating the process of patient selection. MIMIC-IV Medical Information Mart for Intensive Care IV

Identification of subpopulations using the LCTM

Table 1 shows the fitted statistical results of the LCTM models. The AIC, BIC, values ontinuously decreased from the one- to seven-class models; however, the decreasing trend slowed down for the six- and seven-class models. In this model, the sample size of the least populated trajectories in class-6 and class-7 is less than 1%, which does not meet the predetermined criteria. Further, considering the clinical interpretability, we eventually selected the five-class model, which has maximum log-likelihood of -8780.715, AIC of 17615.431 and BIC of 17806.23.

Table 1.

Statistics for choosing the best number of classes

Nmber of classes	Log likelihood	AIC	BIC	Entropy	%class1	%class2	%class3	%class4	%class5	%class6	%class7
1	-20582.9	41179.81	41229.27	1	100
2	-12171.66	24367.328	24452.13	0.831	79.60291	20.39709	-	-	-	-	-
3	-10690.52	21415.042	21535.18	0.896	79.72989	12.82466	7.445458	-	-	-	-
4	-9726.933	19497.865	19653.34	0.778	64.55039	21.29747	8.495902	5.656239	-	-	-
5	-8780.715	17615.431	17806.23	0.776	54.98095	11.50872	27.60014	1.269768	4.640425	-	-
6	-8180.525	16425.05	16651.19	0.786	51.19474	29.36627	7.318481	9.361653	0.969641	1.789219	-
7	-7819.651	15713.302	15974.77	0.789	42.49105	34.1683	7.041441	8.080342	0.969641	2.089345	5.159875

AIC Akaike information criterion, BIC Bayesian information criteria

The creatinine change trajectory of the five-class model is shown in Fig. 2.The proportion of samples in class 1 is 55%, in which the creatinine changed steadily with no obvious increasing or decreasing trend, and the mean value was about 0.4 mg/dl. Class 2 accounted for 11.5%, in which the creatinine gradually increased from 2.2 mg/dl to 2.7 mg/dl, and the change trend was generally stable. Class 3 accounted for 27.6%, in which the creatinine change trend was stable, and the mean value was about 1.2 mg/dl. Class 4 accounted for 1.3%, exhibited an initial decrease followed by an increase. Class 5 accounted for 4.6%, in an inverted U-shape.

Fig. 2.

Class-spectific mean predicted trajectory. Five trajectories of the Creatinine based on LCTM. The dotted line represents the 95% confidence interval for each trajectory. The percentages in the parentheses indicate the percentages of patients each class accounts for. LCTM latent class trajectory models

The baseline characteristics of patients stratified based on the five trajectory classes are listed in Table 2. The study included 8663 eligible patients, with males predominating (56%). The median age of the patients was 64 years.

Table 2.

Descriptive characteristics of overall participants and by creatinine trajectory

Variables	Overall	Class1	Class2	Class3	Class4	Class5	P-value
Number	8663	4763	997	2391	110	402
Age	64.11 ± 16.63	62.37 ± 17.01	64.18 ± 15.36	68.51 ± 15.71	58.90 ± 16.19	59.85 ± 15.32	< 0.001
Gender							< 0.001
Male	4857 (56.07%)	2338 (49.09%)	618 (61.99%)	1553 (64.95%)	67 (60.91%)	281 (69.90%)
Female	3806 (43.93%)	2425 (50.91%)	379 (38.01%)	838 (35.05%)	43 (39.09%)	121 (30.10%)
Race							0.005
Other races	2994 (34.56%)	1646 (34.56%)	354 (35.51%)	786 (32.87%)	37 (33.64%)	171 (42.54%)
White	5669 (65.44%)	3117 (65.44%)	643 (64.49%)	1605 (67.13%)	73 (66.36%)	231 (57.46%)
Heart rate (beats/min) (h)	92.78 ± 21.20	92.00 ± 21.06	96.61 ± 21.48	91.91 ± 20.98	94.77 ± 20.55	97.09 ± 22.16	< 0.001
SBP (mmHg)	122.90 ± 24.95	125.63 ± 24.66	117.90 ± 25.36	121.12 ± 24.75	114.39 ± 21.34	115.85 ± 24.86	< 0.001
DBP (mmHg)	68.62 ± 18.39	70.55 ± 17.75	65.92 ± 19.51	66.54 ± 18.62	63.81 ± 18.55	66.07 ± 18.91	< 0.001
Respiration rate (beats/min)	20.22 ± 6.39	19.87 ± 6.17	21.43 ± 6.62	20.22 ± 6.62	21.49 ± 6.52	21.08 ± 6.47	< 0.001
Temperature (℃) (T)	36.72 ± 0.99	36.77 ± 0.93	36.70 ± 1.05	36.65 ± 1.03	36.60 ± 1.14	36.61 ± 1.13	< 0.001
PO2(KPa)	161.63 ± 125.36	166.81 ± 124.74	131.60 ± 109.57	167.01 ± 132.74	136.77 ± 111.93	149.56 ± 115.61	< 0.001
PCO2(KPa)	43.06 ± 13.67	43.37 ± 13.77	42.11 ± 13.74	43.11 ± 13.63	41.49 ± 12.36	41.85 ± 12.69	0.018
PH	7.36 ± 0.10	7.38 ± 0.09	7.31 ± 0.12	7.35 ± 0.11	7.30 ± 0.12	7.30 ± 0.12	< 0.001
WBC (× 10^9/l)	13.58 ± 11.48	13.01 ± 11.19	15.59 ± 15.08	13.71 ± 9.12	14.39 ± 7.41	14.36 ± 16.39	< 0.001
RBC (× 10^12/l)	3.63 ± 0.81	3.68 ± 0.77	3.51 ± 0.85	3.59 ± 0.82	3.51 ± 0.83	3.53 ± 0.91	< 0.001
PLT (K/µl)	208.76 ± 117.98	219.59 ± 119.83	192.28 ± 121.67	197.82 ± 108.78	208.73 ± 117.81	186.50 ± 125.68	< 0.001
Hemoglobin(g/l)	10.89 ± 2.35	11.05 ± 2.27	10.55 ± 2.47	10.74 ± 2.36	10.65 ± 2.52	10.72 ± 2.63	< 0.001
PT(s)	16.84 ± 9.69	15.55 ± 8.08	20.34 ± 13.33	17.27 ± 9.40	19.86 ± 15.01	20.12 ± 12.55	< 0.001
APTT(s)	38.65 ± 24.08	36.35 ± 22.83	43.09 ± 25.89	40.48 ± 24.89	37.88 ± 13.63	44.25 ± 27.66	< 0.001
Na (mmol/l)	138.57 ± 5.69	138.63 ± 5.29	138.30 ± 7.23	138.72 ± 5.43	137.87 ± 6.93	137.93 ± 6.92	0.027
K (mmol/l)	4.15 ± 0.75	4.01 ± 0.64	4.41 ± 0.91	4.26 ± 0.75	4.74 ± 1.20	4.49 ± 0.87	< 0.001
Bun(mg/dl)	24.30 ± 19.56	16.37 ± 8.64	45.10 ± 29.61	27.18 ± 14.94	57.73 ± 39.26	40.36 ± 31.87	< 0.001
SOFA score	2.14 ± 2.41	1.54 ± 1.86	3.47 ± 3.18	2.43 ± 2.35	4.02 ± 3.55	3.75 ± 3.21	< 0.001
SAPSII score	40.64 ± 13.93	36.43 ± 12.22	49.80 ± 14.15	42.96 ± 13.26	47.13 ± 14.73	52.28 ± 14.58	< 0.001
GCS score	14.28 ± 2.15	14.30 ± 2.06	14.26 ± 2.20	14.26 ± 2.27	14.09 ± 2.29	14.29 ± 2.40	0.817
Congestive heart failure, n (%)							< 0.001
Yes	2293 (26.47%)	978 (20.53%)	306 (30.69%)	875 (36.60%)	24 (21.82%)	110 (27.36%)
No	6370 (73.53%)	3785 (79.47%)	691 (69.31%)	1516 (63.40%)	86 (78.18%)	292 (72.64%)
Cerebrovascular disease, n (%)							< 0.001
Yes	1681 (19.40%)	1079 (22.65%)	123 (12.34%)	423 (17.69%)	13 (11.82%)	43 (10.70%)
No	6982 (80.60%)	3684 (77.35%)	874 (87.66%)	1968 (82.31%)	97 (88.18%)	359 (89.30%)
Chronic pulmonary disease, n (%)							0.575
Yes	2351 (27.14%)	1295 (27.19%)	260 (26.08%)	668 (27.94%)	25 (22.73%)	103 (25.62%)
No	6312 (72.86%)	3468 (72.81%)	737 (73.92%)	1723 (72.06%)	85 (77.27%)	299 (74.38%)
Diabetes, n (%)							< 0.001
Yes	2118 (24.45%)	931 (19.55%)	322 (32.30%)	732 (30.61%)	27 (24.55%)	106 (26.37%)
No	6545 (75.55%)	3832 (80.45%)	675 (67.70%)	1659 (69.39%)	83 (75.45%)	296 (73.63%)
Renal_disease, n (%)							< 0.001
Yes	131 (1.51%)	12 (0.25%)	63 (6.32%)	31 (1.30%)	7 (6.36%)	18 (4.48%)
No	8532 (98.49%)	4751 (99.75%)	934 (93.68%)	2360 (98.70%)	103 (93.64%)	384 (95.52%)
Vasopressors, n (%)							< 0.001
Yes	5292 (61.09%)	2508 (52.66%)	775 (77.73%)	1610 (67.34%)	81 (73.64%)	318 (79.10%)
No	3371 (38.91%)	2255 (47.34%)	222 (22.27%)	781 (32.66%)	29 (26.36%)	84 (20.90%)
Menchanical ventilation, n (%)							0.368
Yes	8129 (93.84%)	4450 (93.43%)	941 (94.38%)	2258 (94.44%)	101 (91.82%)	379 (94.28%)
No	534 (6.16%)	313 (6.57%)	56 (5.62%)	133 (5.56%)	9 (8.18%)	23 (5.72%)
CRRT, n (%)							< 0.001
Yes	612 (7.06%)	43 (0.90%)	246 (24.67%)	100 (4.18%)	47 (42.73%)	176 (43.78%)
No	8051 (92.94%)	4720 (99.10%)	751 (75.33%)	2291 (95.82%)	63 (57.27%)	226 (56.22%)

The survival probabilities of each class plotted using the K-M analysis are shown in Fig. 3. Patients in class 2 had the highest mortality rate, and the results of the log-rank test indicated that the survival probability differed significantly between the classes (P < 0.001).

Fig. 3.

Survival probability curve during hospitalization of patients with different trajectories of creatinine

The results of the Cox proportional-hazards regression analysis are listed in Table 3. In the univariate analysis (Model I), with class 1 as a reference, the HR (95% CI) values of classes 2 to 5 were 2.25 (1.98–2.57), 1.44(1.29–1.60), 1.97(1.39–2.79), 1.78 (1.46–2.18), respectively. In the multivariate analysis (Model II), with class 1 as a reference, the HR (95% CI) values of classes 2 to 5 were 2.29(2.01–2.61), 1.36(1.22–1.52), 2.10 (1.48–2.97), 1.85(1.51–2.26), respectively. In the multivariate analysis (Model III), with class 1 as a reference, the HR (95% CI) values of classes 2 to 5 were 1.26(1.07–1.49), 1.13(1.01–1.27), 1.06(0.73–1.54), 0.91(0.72–1.16), respectively. The three models suggest that when the creatinine change trend of patients with sepsis conforms to class 2, the 28-day mortality risk is the highest (P < 0.05).

Table 3.

Results of Cox proportional hazard models

Class	Model1		Model2		Model3
	HR(95% CI)	P value	HR(95% CI)	P value	HR(95% CI)	P value
Class1	Reference		Reference		Reference
Class2	2.25 [1.98, 2.57]	< 0.001	2.29 [2.01, 2.61]	<0.001	1.26 [1.07, 1.49]	0.006
Class3	1.44 [1.29, 1.60]	< 0.001	1.36 [1.22, 1.52]	<0.001	1.13 [1.01, 1.27]	0.041
Class4	1.97 [1.39, 2.79]	< 0.001	2.10 [1.48, 2.97]	<0.001	1.06 [0.73, 1.54]	0.77
Class5	1.78 [1.46, 2.18]	< 0.001	1.85 [1.51, 2.26]	<0.001	0.91 [0.72, 1.16]	0.45

In Model I, only classes were analyzed and no other variables were adjusted. In Model II, age, race, gender, were included in the Cox proportional-hazards regression model and analyzed using forward partial likelihood estimation. In Model III, age, race, gender, heart rate, sbp, resp rate, temperature, WBC, RBC, hemoglobin, PO2,PH, Na+,K+,PT, APTT, BUN, SOFA, SAPSII, congestive heart failure, cerebrovascular disease, vasoactive, ventilator, CRRTwere included in the Cox proportional-hazards regression model and analyzed using forward partial likelihood estimation.

The study also performed subgroup analyses for the following variables: age, race, gender, congestive heart failure, cerebrovascular disease, vasopressors, menchanical ventilation, CRRT. The results indicated that in the subgroups of race, congestive heart failure, cerebrovascular disease, vasopressors and menchanical ventilation, the above-five variables and creatinine trajectories did not interact with 28-day mortality (P > 0.05). A nominal interaction occurred between age, gender, CRRT and creatinine trajectory in 28-day mortality (P < 0.05). Figure 4 presents the subgroup analysis results.

Fig. 4.

28-day mortality. Prespecified subgroup analyses

Discussion

Sepsis is one of the most common causes of death in the ICU. The latest cross-sectional survey on the epidemiologic characteristics of ICUs in China showed that the incidence of sepsis in ICUs was 23.9%, and the 90-day mortality rate was 35.5% [18]. Approximately 1.5 million people in the United States suffer from sepsis each year, and the healthcare costs of sepsis care amount to $24 billion, with an increasing trend each year [19]. Early monitoring of the patient’s condition and timely intervention are important for the prognosis of patients with sepsis.

The prevalence of acute kidney injury is high among patients with sepsis, and creatinine, a crucial indicator for assessing kidney function, plays a pivotal role in diagnosing sepsis-associated acute kidney injury [8]. A multicenter observational study indicated that elevated creatinine levels were closely related to poor prognosis in patients with sepsis [20]. Vanmassenhove et al. [21]suggested that changes in serum creatinine within the first 24 h of admission were associated with mortality, with an increase of 0.3 mg/mL in creatinine demonstrating predictive value for sepsis-related death. Therefore, serum creatinine, as a biomarker of AKI, had the potential to reflect the severity of sepsis and aided in the identification of high-risk patients. Sepsis-related Organ Failure Assessment (SOFA) [22], Acute Physiology and Chronic Health Evaluation II (APACHE-II) [23] are now the most widely used approaches for assessing sepsis severity, and creatinine is involved in both scoring systems, highlighting its importance in assessing the severity of the disease.

Our study aimed to investigate how early changes in creatinine levels influence outcomes in patients with sepsis. We chose to exclude patients with chronic kidney disease (CKD) for the following reasons: Baseline creatinine levels in CKD patients often exhibit significant variability [24], particularly in stages 4 and 5. This variability compromises the discriminatory power and stability of trajectory analysis models. Furthermore, fluctuations in CKD patients’ creatinine levels do not necessarily correlate with the severity of sepsis-induced acute kidney injury (AKI), undermining the reliability of our findings. Previous studies evaluating the prognostic value of creatinine in sepsis and other conditions similarly excluded chronic kidney disease patients from their samples [25, 26]. Based on these considerations, we decided to exclude patients with chronic kidney disease from our analysis.

However, although existing studies claimed to use continuously measured creatinine data, the longitudinal trajectory of creatinine has not been analyzed in existing studies. Several studies have demonstrated the impact of dynamic changes in indicators (trajectories) on the outcome of patients A multi-cohort, retrospective, observational study indicated that the rapid decline in platelet counts is a critical prognostic factor for septic patients in an intensive care unit, and serial platelet measures are associated with prognosis [27]. Yang et al. suggested that lymphocyte trajectories is associated with the prognosis of sepsis patients [28]. Another retrospective cohort study suggested the influence of systolic blood pressure trajectory on in-hospital mortality in patients with sepsis [29]. This study is the first to identify creatinine trajectories in septic patients and explore the relationship between dynamics in creatinine and 28-day mortality risk.

Serum creatinine trajectories reflect the dynamic changes in creatinine levels of patients over time. In this study, serum creatinine levels of sepsis patients admitted to ICU for 72 h were categorized into five trajectories. Early creatinine trajectories can reflect several types of AKI: Class 1 starts at a low level and remains stable at that low level; Class 2 shows a gradual upward trend; Class 3 stabilizes at a high level; Class 4 exhibits a trend of initially decreasing and then increasing; Class 5 presents an inverted U-shape. The results showed that patients with sepsis had the highest risk of in-hospital death in the persistently rising trajectory group compared with the normal level stable trajectory group. Previous studies suggested that an increase in serum creatinine of 0.3 mg/mL within the first 24 h of admission demonstrating predictive value for sepsis-related death [21], Our study extends that finding further. This observation aligns with our clinical experience and suggests that patients experiencing persistent deterioration of renal function demonstrate an elevated risk of mortality.

We also observed that in the subgroup analysis, patients who did not receive CRRT had the highest mortality in the class 2, indicating that the primary outcome was stable. The lower CRRT rate in class 2 compared to class 4 and class 5 may be due to the initial creatinine levels. Patients in class 4 and class 5 tend to have higher initial creatinine levels and more unstable conditions, which might result in a higher likelihood of receiving CRRT. Among patients who did not undergo CRRT treatment, class 2 had the highest mortality rate. However, in patients who received CRRT, the mortality rate in class 2 decreased.This suggests that the decision to administer CRRT can have a significant impact on patient mortality. Based on our research findings, it is suggested that appropriately expanding the criteria for CRRT treatment in AKI patients and starting CRRT earlier may improve patient prognosis.

This study also has the following limitations. (1) As a retrospective cohort study, it was only able to establish association rather than causation. Given the observational nature of our study, other unmeasured confounders may have influenced the results. (2) Serum creatinine levels were measured within the first 72 h of ICU admission. These initial values may not fully reflect changes in renal function that may occur during ICU hospitalization. (3) Determining whether sepsis is the primary reason for ICU admission is challenging. However, it should be noted that ICU admissions are often driven by critical illness associated with multiple diseases, including sepsis. And the MIMIC database cannot definitively identify the site of infection in sepsis patients. (4) This study used data from the US MIMIC database, which may limit the generalizability of its findings to other populations. (5) Our findings do not apply to patients with CKD complicated by sepsis.

Conclusion

The creatinine trajectory within 72 h of admission in patients with sepsis has an impact on 28-day mortality. Patients with persistently elevated creatinine in the first three days had a higher mortality risk than those with stable creatinine around 0.4 mg/dl. The trend of creatinine in patients should therefore be closely observed, and corresponding treatment should be provided as soon as possible to improve their prognoses.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

ICU

Intensive care unit

LCTM

Latent class trajectory models

MIMIC-IV

Medical Information Mart for Intensive Care IV

IRB

Institutional review board

SOL

Structured Query Language

AIC

Akaike information criterion

BIC

Bayesian information criterion

K-M

Kaplan-Meier

HR

Hazard ratios

CI

Confidence intervals

WBC

White blood cell count

RBC

Red blood cell count

Na+

Sodium

K+

Potassium

PT

Prothrombin yime

APTT

Activated partial thromboplastin time

BUN

Blood urea nitrogen

SOFA

Sequential organ failure assessment

SAPSII

Simplified Acute Physiologic Score-II

AKI

Acute kidney injury

APACHE II

Acute physiology and chronic health evaluation II

CRRT

Continuous renal replacement therapy

Author contributions

Dan Liu, Zhonglin Zhang andAihua Qin contributed equally to this work. They were responsible for the conception, design, and execution of the study. Xiaofeng Li, Jian Zhao, Huihua Jiang and Yi Zhao responsible for data collection and analysis. Yuan Zhuo Chen and Hu Peng supervised the project, provided critical revisions, and guided the writing of the manuscript. All authors read and approved the final version of the manuscript.

Funding

This work was supported by the National Natural Science Foundation of China (No.82472196).

Data availability

Data of this study were obtained from the MIMIC-IV database, which was a publicly available dataset, which were exhibited in Materials and Methods.

Declarations

Ethics approval and consent to participate

The MIMIC-IV database was approved by the Institutional Review Boards of Massachusetts Institute of Technology (Cambridge, MA) and Beth Israel Deaconess Medical Center (Boston, MA). Informed consent was obtained for the original data collection. The requirement for individual patient consent was waived as the project did not impact clinical care and all protected health information was de-identified. This study adheres to the ethical principles outlined in the Declaration of Helsinki. The researchers have completed the ethics training required by the CITI Program and obtained authorization to access the database(Record ID: 62595271).

Consent for publication

Non-applicable.

Competing interests

The authors declare no competing interests.