Association between renal mean perfusion pressure and prognosis in patients with sepsis-associated acute kidney injury: insights from the MIMIC IV database

Abstract

Objective

To investigate the association between renal mean perfusion pressure (MPP) and prognosis in sepsis-associated acute kidney injury (SA-AKI).

Methods

Data were extracted from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. Group-based trajectory modeling (GBTM) was applied to identify dynamic MPP patterns, while restricted cubic spline (RCS) curves were utilized to confirm the non-linear relationship between MPP and mortality. Cox regression analysis assessed the risk of mortality across different MPP levels, adjusting for potential confounders. Subgroup analyses and sensitivity analyses were conducted to ensure the robustness of the findings.

Results

A total of 2318 patients with SA-AKI were stratified into five MPP trajectories by GBTM. Patients in Traj-1 and Traj-2, characterized by consistently low MPP (<60 mmHg), demonstrated markedly higher 90-d mortality (62.86% and 26.98%). RCS curves revealed a non-linear inverse relationship between MPP and 90-d mortality, identifying 60 mmHg as the optimal threshold. Patients with MPP ≤ 60 mmHg exhibited significantly elevated 90-d mortality compared to those with MPP > 60 mmHg (29.81% vs. 20.88%). Cox regression analysis established Traj-1 and Traj-2 as independent risk factors for increased mortality relative to Traj-3 (60–70 mmHg), with hazard ratios (HRs) of 4.67 (95%-CI 3.28–6.67) and 1.45 (95%-CI 1.20–1.76). MPP > 60 mmHg was significantly associated with reduced 90-d mortality (HR 0.65, 95%-CI 0.55–0.77). Subgroup and PSM analyses supported these findings.

Conclusions

Dynamic MPP trajectory serves as a valuable prognostic biomarker for SA-AKI. Early monitoring of MPP trends offers critical insights into renal perfusion management, potentially improving outcomes in SA-AKI.

1. Introduction

Acute kidney injury (AKI) is a common complication among intensive care unit (ICU) patients and remains a critical focus in nephrology and critical care research. AKI in critically ill individuals is associated with high mortality rates and significantly increases the risk of chronic kidney disease (CKD) progression following hospital discharge, contributing to substantial medical and societal burdens [1,2]. Among critically ill patients, sepsis and septic shock is a leading cause of AKI [3]. Reports estimate the incidence of sepsis-associated AKI (SA-AKI) in ICU populations to range from 15% to 20%, with mortality rates surpassing 50% [3,4], presenting a severe challenge to global public health systems.

The pathogenesis of SA-AKI remains incompletely understood, with hemodynamic abnormalities and tissue hypoperfusion identified as critical mechanisms. Traditional approaches to hemodynamic management in SA-AKI have emphasized optimizing blood pressure to enhance renal perfusion pressure, aiming to preserve renal perfusion and maintain kidney function [5,6]. Current guidelines from the Surviving Sepsis Campaign (SSC) recommend targeting mean arterial pressure (MAP) during fluid resuscitation to ensure adequate tissue perfusion and improve patient prognosis [7]. Several studies have demonstrated a close association between MAP levels and the development of SA-AKI [8–10]. However, emerging evidence suggests that renal venous outflow restriction or congestion, characterized by increased afterload on the kidneys, plays a central role in limiting renal perfusion [11]. Notably, elevated central venous pressure (CVP), representing renal perfusion afterload, has been shown to independently predict SA-AKI risk, overshadowing MAP as a determinant [12]. To address these findings, mean perfusion pressure (MPP), defined as the difference between MAP and CVP, has been introduced as a novel biomarker for individualized assessment of tissue perfusion pressure [13,14]. Over the past decade, extensive research has highlighted the predictive value of MPP in diverse AKI cohorts [12,13,15,16]. Despite this progress, the relationship between MPP levels, their dynamic patterns, and the prognosis of patients with SA-AKI remains unclear.

In light of these gaps, this study aimed to evaluate the clinical significance of both single-point MPP measurements and their dynamic trajectories in predicting outcomes for patients with SA-AKI. These findings provide new perspectives on early intervention and tissue perfusion management strategies for this high-risk patient population.

2. Materials and methods

2.1. Study design

This study conducts a single-center retrospective analysis using data from the publicly accessible Medical Information Mart for Intensive Care IV (MIMIC-IV) database version 2.2 [17]. The MIMIC-IV dataset, which is anonymized and institutional review board (IRB)-approved by the Massachusetts Institute of Technology (MIT), does not require informed consent. As no additional data collection was undertaken, further ethical statements are unnecessary for this study. The MIMIC-IV database includes over 70,000 ICU admission records from Beth Israel Deaconess Medical Center in the United States, spanning 2008–2019. Data extraction was performed by Yipeng Fang, certified by the Collaborative Institutional Training Program, with authorized access to the database. Structured query language (SQL) software was used for data extraction. The study adheres to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines to ensure rigorous reporting standards [18].

2.2. Study population

Patient selection began by identifying those meeting the Sepsis 3.0 diagnostic criteria, which require a Sequential Organ Failure Assessment (SOFA) score ≥ 2 points based on a confirmed or suspected infection [19]. For patients with multiple hospitalizations, only the first admission was included. Data from patients younger than 18 years were excluded. From the identified sepsis cohort, the target population was further refined based on the SA-AKI criteria established by the 28th Acute Disease Quality Initiative workgroup [20]. Specifically, patients who developed AKI within 7 d following a sepsis diagnosis were classified as having SA-AKI. The diagnosis of AKI followed the Kidney Disease Improving Global Outcomes (KDIGO) creatinine criteria, defined as either a ≥ 50% increase in baseline creatinine within 7 d or an increase of ≥0.3 mg/dL within 48 h [21]. Patients with AKI prior to sepsis onset were excluded. Additionally, individuals with missing MPP data, including MAP or CVP measurements within 24 h of SA-AKI diagnosis, were excluded from the analysis.

Trajectory modeling offers a statistical framework for depicting the various trends in the evolution of MPPs, providing a detailed view of MPP fluctuations over time. This approach yields novel insights into the dynamic changes in MPP and their potential role in prognostic assessments for patients with SA-AKI. Using group-based trajectory modeling (GBTM), patients were grouped based on the dynamic changes in MPP within the first 24 h of SA-AKI diagnosis. MPP was determined as the difference between MAP and CVP [22]. This technique identifies characteristic patterns of MPP variation, clustering patients according to the trends in their MPP values. Several models were explored with trajectory numbers ranging from two to six (Table S1). The optimal trajectory model was selected based on the goodness of fit, assessed using the Akaike information criterion (AIC) and Bayesian information criterion (BIC), alongside clinical interpretability. Patients were assigned to trajectory groups based on the highest posterior probability, with an average posterior probability (AvePP) exceeding 0.7 indicating robust and reliable classification [23]. The distribution of MPP across different trajectories was visualized using kernel density plots. Additionally, restricted cubic spline (RCS) curves were employed to assess the nonlinear relationship between MPP and 90-d mortality risk. We established RCS with four knots, positioning them at the MPP values that correspond to the 5%, 35%, 65%, and 95% percentage, which were 48, 58, 65, and 79 mmHg, respectively. Using the null effect line (odds ratio [OR] = 1) as a reference, the optimal MPP threshold was identified to categorize patients into high-MPP and low-MPP groups.

2.3. Exposure and endpoint

The primary exposure variable in this study was the MPP value within 24 h of SA-AKI diagnosis. MAP and CVP data were collected at 3-h intervals, and the average MAP and CVP values within each interval were used to calculate the MPP, minimizing the influence of outliers. The MPP values from each time interval were utilized to construct the trajectory model. Additionally, the initial, maximum, minimum, and mean MPP values within 24 h were calculated.

The primary outcome was 90-d mortality, reflecting patient survival status within 90 d following SA-AKI diagnosis. Secondary outcomes included the use of continuous renal replacement therapy (CRRT), 28-d mortality, in-hospital mortality, ICU mortality, length of hospital stay, and ICU length of stay (LOS).

2.4. Baseline information collection

Baseline information included demographic characteristics (age, sex, body weight, and race), multimorbidities (coronary heart disease [CHD], acute or chronic heart failure [HF], hypertension [HT], diabetes mellitus [DM], liver disease, and CKD), vital signs (heart rate [HR], respiratory rate [RR], temperature, and oxygen saturation [SpO₂]), laboratory parameters (white blood cell [WBC] count, hemoglobin, platelets, international normalized ratio [INR], prothrombin time [PT], activated partial thromboplastin time [APTT], sodium, potassium, chloride, anion gap (AG), and blood urea nitrogen [BUN]), risk assessment scores (acute physiology score III [APS III] and SOFA), and special interventions (mechanical ventilation [MV], vasoactive drugs, and sedative drug exposure). All information about vital signs and laboratory parameters was obtained within the first 24 h after SA-AKI diagnosis. The window of opportunity for undergoing special examinations is restricted to the period between ICU admission and the development of SA-AKI. Details about multimorbidities definition are shown in Table S2. Serum creatinine levels were recorded within the first 24 h after sepsis diagnosis, along with the highest creatinine values within 7 d following sepsis diagnosis.

2.5. Statistical analysis

The Shapiro–Wilk test was employed to assess the normality of continuous variables (shown in Table S3). For variables conforming to a Gaussian distribution, results were reported as mean ± standard deviation (SD), with intergroup differences analyzed using Student’s t-test or one-way ANOVA. For non-Gaussian distributions, data were expressed as medians and interquartile ranges (IQRs) and further presented as M (Q1, Q3). Intergroup differences for non-Gaussian variables were evaluated using the Mann–Whitney U test or the Kruskal–Wallis test. Categorical variables were summarized as counts (percentages) and analyzed using the chi-squared test.

Data cleaning followed standardized protocols. Continuous variables that did not conform to a Gaussian distribution were checked for outliers, defined as values exceeding Q3 + 1.5IQR or below Q1-1.5IQR, which were subsequently treated as missing data. Missing values were imputed using the median when the missing proportion was below 10% (see in Table S4). For missing proportions between 10% and 30%, regression-based imputation was applied. Parameters with missing rates exceeding 30% were excluded from the analysis. Samples with missing MAP or CVP data were also excluded from the final dataset.

Kaplan–Meier (K–M) curves were constructed to depict cumulative survival over time across different trajectories and categories, with intergroup differences assessed using the Log-rank and Wilcoxon tests. Cox regression analysis was performed to evaluate the effects of trajectory groups and MPP categories on clinical outcomes, using Traj-3 and MPP ≤ 60 mmHg as reference groups. To reduce bias from population heterogeneity and disease severity, demographic data, special interventions, and SOFA scores were incorporated into adjustment models. Adjusted Model 1 included demographic data and multimorbidity, while Adjusted Model 2 further accounted for special interventions and SOFA scores. Variance inflation factor (VIF) analysis was conducted to assess collinearity in the models, with variables exhibiting significant collinearity (VIF > 10 or VIF < 0.1) converted into binary variables based on median values. The predictive value of MPP for 90-d mortality was evaluated using receiver operating characteristic (ROC) curves, with the area under the curve (AUC), sensitivity, and specificity calculated based on the maximum Youden index.

To ensure the robustness of the findings, subgroup and sensitivity analyses were conducted. Subgroup analyses stratified patients based on demographic factors and comorbidities, with interaction terms included in the models to evaluate the presence of significant interactive effects. For these analyses, age and weight were stratified by their median values. Sensitivity analysis was performed using the propensity score matching (PSM) method to balance baseline characteristics between groups [24]. A 1:1 nested matching model was applied, with a caliper value of 0.05 and no replacement. Matching variables included demographic data, comorbidities, special interventions, and SOFA scores.

All statistical analyses were performed using Stata and R software. A two-sided p value < 0.05 was considered statistically significant for intergroup comparisons.

3. Results

3.1. Study cohort

Figure 1 provides a detailed flowchart outlining the patient selection process for the study. Initially, 73,181 ICU admission records were retrieved from the MIMIC-IV database. Following the application of inclusion and exclusion criteria, a total of 2318 patients diagnosed with SA-AKI were included in the final analysis. Of these, 581 patients succumbed within 90 d, while 1737 survived. Baseline characteristics of patients in the survival and non-survival group are shown in Table S5. Lower percentage of patients in survival group receiving CRRT intervention (8.29% vs. 23.58%, p < 0.001). Survivals had higher levels of initial [62(57,67) mmHg vs. 59(53,66) mmHg], maximum [63(58,70) mmHg vs. 61(55,69) mmHg], minimum [60(55,66) mmHg vs. 56(50,63) mmHg], and mean [62(57,68) mmHg vs. 59(53,66) mmHg] MPP value (all p < 0.001).

Figure 1.

Flowchart of patient selection in the study.

3.2. Characterization and clinical outcomes of MPP trajectories

Trajectory models for MPP were sequentially constructed with two to five trajectories. The model with five trajectories demonstrated the lowest BIC and AIC values, with each trajectory achieving an AvePP ≥ 0.70, signifying high classification accuracy (Table S1). The longitudinal trends of MPP for each trajectory revealed distinct patterns (Figure 2(A)). Traj-1 (N = 70) represented the lowest MPP trajectory, characterized by persistently low MPP values (40–50 mmHg). Traj-2 (N = 1201) exhibited slightly higher but still relatively low MPP levels (50–60 mmHg). Traj-3 (N = 802) demonstrated MPP values fluctuating around 60–70 mmHg. Traj-4 (N = 185) showed MPP levels around 70–80 mmHg, while Traj-5 (N = 60) represented the highest MPP trajectory, with values starting above 80 mmHg and gradually increasing. Patients in Traj-3, Traj-4, and Traj-5 displayed better renal perfusion, with MPP consistently exceeding 60 mmHg, compared to Traj-1 and Traj-2. Among the 2318 patients included, 5.4% were classified as Traj-1, 45.5% as Traj-2, 34.9% as Traj-3, 11.3% as Traj-4, and 2.9% as Traj-5. Most patients were grouped into Traj-2 and Traj-3. As visualized in the kernel density graph (Figure 2(B)), Traj-5 exhibited the highest mean MPP value within the first 24 h post-SA-AKI diagnosis, followed sequentially by Traj-4, Traj-3, Traj-2, and Traj-1.

Figure 2.

MPP Trajectories and 90-d mortality in patients with SA-AKI. (A) MPP trajectory models. (B) Kernel density curve representing the distribution of MPP among patients in different trajectories. (C) Kaplan–Meier survival curve demonstrating significant differences in survival among patients in different trajectories.

Baseline characteristics and outcomes across the trajectories are summarized in Table 1. Patients in Traj-1 and Traj-2 were older and had a higher prevalence of HF and DM compared to those in Traj-3, Traj-4, and Traj-5. Traj-1 showed elevated levels of MCV, RDW, INR, PT, APTT, potassium, AG, lactate, BUN, the initial value of creatinine and the maximum value of creatinine. Traj-1 also had the decreased levels of hemoglobin. Conversely, Traj-5 presented higher glucose levels. Regarding interventions, vasoactive drug use was more prevalent in Traj-1 and Traj-2, although no significant differences were observed in MV usage among trajectories. In terms of disease severity, Traj-1 exhibited the highest SOFA and APSIII scores, while Traj-3, Traj-4, and Traj-5 showed relatively lower scores. Traj-1 also had the highest proportion of patients requiring CRRT and the highest mortality rates across all categories, including 28-d, 90-d, in-hospital, and ICU mortality. Traj-2 followed similar trends but with slightly lower rates. Traj-3 had the lowest mortality rates, while Traj-4 showed the lowest CRRT requirement. Hospital stays were the longest in Traj-5, whereas ICU stays were the longest in Traj-4. K–M survival curves (Figure 2(C)) revealed significantly lower survival rates for Traj-1 and Traj-2 over time compared to other trajectories, with Traj-1 showing the poorest outcomes. In contrast, survival curves for Traj-3, Traj-4, and Traj-5 were not significantly different. Both the Log-rank and Wilcoxon tests confirmed significant survival differences among the trajectories (both p < 0.001).

Table 1.

Characteristics and clinical outcomes of patients with SA-AKI in different trajectory groups.

	Traj-1	Traj-2	Traj-3	Traj-4	Traj-5	p Value
Age, years, M (Q1, Q3)	72.7(63.8,81.1)	72.3(62.1,80.7)	69.5(59.9,78.2)	64.1(52.9,74.4)	60.9(52.4,71.0)	<0.001
Male, n (%)	40(57.14)	725(60.37)	503(62.72)	119(64.32)	35(58.33)	0.633
Race-White, n (%)	46(65.71)	846(70.44)	524(65.34)	121(65.41)	30(50.00)	0.004
Body weight, Kg, M (Q1, Q3)	81.6(68.9,92.5)	82.6(69.7,98.3)	80.4(67.0,95.0)	78.7(65.6,92.1)	81.5(69.2,96.8)	0.007
CHD, n (%)	28(40.00)	629(52.37)	375(46.76)	51(27.57)	14(23.33)	<0.001
HF, n (%)	25(35.71)	452(37.64)	237(29.55)	49(26.49)	13(21.67)	<0.001
Liver disease, n (%)	28(40.00)	178(14.82)	128(15.96)	41(22.16)	26(43.33)	<0.001
CKD, n (%)	21(30.00)	313(26.06)	171(21.32)	39(21.08)	13(21.67)	0.079
HT, n (%)	29(41.43)	546(45.46)	394(49.13)	77(41.62)	31(51.67)	0.208
DM, n (%)	26(37.14)	450(37.47)	263(32.79)	55(29.73)	15(25.00)	0.041
HR, bpm, M (Q1, Q3)	91(77,108)	84(76,97)	84(77,96)	87(77,100)	86(68,106)	0.350
RR, cpm, M (Q1, Q3)	20(14,25)	17(14,22)	16(14,21)	19(16,23)	18(15,22)	0.002
Temperature, M (Q1, Q3)	36.8(36.1,37.0)	36.7(36.2,36.9)	36.7(36.1,37.0)	36.7(36.1,37.2)	36.9(36.6,37.4)	0.005
SpO2, %, M (Q1, Q3)	98(94,100)	99(97,100)	100(97,100)	100(97,100)	99(97,100)	0.002
WBC, K/UL, M (Q1, Q3)	12.9(10.3,19.3)	13.3(9.4,18.3)	12.7(9.1,17.0)	12.2(7.7,16.8)	11.2(7.7,15.5)	<0.001
Hemoglobin, mg/dL, M (Q1, Q3)	9.6(83,11.0)	10.1(8.9,11.2)	10.4(8.2,11.6)	10.4(9.1,12.1)	11.0(9.7,12.3)	<0.001
Platelet, K/UL, M (Q1, Q3)	140(92,213)	149(113,207)	149(114,202)	151(107,224)	157(107,224)	0.717
MCV, dL, M (Q1, Q3)	93(89,99)	91(88,95)	90(87,94)	90(87,94)	90(86,96)	0.010
RDW, %, M (Q1, Q3)	15.4(14.5,16.6)	14.6(13.7,15.9)	14.3(13.4,15.6)	14.6(13.8,16.0)	15.0(13.7,16.4)	<0.001
INR, M (Q1, Q3)	1.6(1.3,2.3)	1.3(1.2,1.5)	1.3(1.2,1.5)	1.3(1.2,1.5)	1.3(1.2,1.5)	<0.001
PT, sec, M (Q1, Q3)	17.6(14.5,24.8)	14.6(13.3,16.9)	14.4(13.0,16.3)	14.4(13.1,16.2)	14.5(13.0,16.1)	<0.001
APTT, sec, M (Q1, Q3)	39.6(33.6,51.4)	34.0(29.6,42.6)	32.8(28.3,40.5)	33.1(28.2,43.0)	32.0(27.1,35.9)	<0.001
Na, mEq/L, M (Q1, Q3)	140(134,143)	139(136,141)	139(137,141)	139(137,142)	140(137,143)	0.106
K, mEq/L, M (Q1, Q3)	4.5(4.1,5.0)	4.3(3.9,4.8)	4.3(3.9,4.7)	4.2(3.8,4.7)	4.2(3.8,4.6)	0.027
Cl, mEq/L, M (Q1, Q3)	106(101,110)	107(104,111)	108(104,111)	106(103,110)	105(103,108)	0.002
AG, mEq/L, M (Q1, Q3)	17(13,21)	13(11,17)	13(11,16)	14(12,17)	15(13,18)	<0.001
Glucose, mg/dL, M (Q1, Q3)	141(110,205)	140(113,175)	139(114,170)	143(117,213)	170(132,234)	0.002
Lactate, mmol/L, M (Q1, Q3)	3.3(2.0,5.1)	2.2(1.5,3.4)	2.1(1.5,3.0)	2.0(1.4,3.2)	1.9(1.5,2.6)	<0.001
BUN, mg/dL, M (Q1, Q3)	31(21,46)	22(16,33)	19(14,28)	23(16,33)	20(16,33)	<0.001
#Creatinine, mg/dL, M (Q1, Q3)	1.9(1.3,3.1)	1.1(0.8,1.6)	1.0(0.8,1.5)	1.1(0.9,1.7)	1.2(0.9,1.7)	<0.001
$Creatinine, mg/dL, M (Q1, Q3)	2.8(1.9,3.9)	1.7(1.3,2.6)	1.5(1.1,2.3)	1.7(1.1,2.9)	1.7(1.2,2.6)	<0.001
Mechanical ventilation use, n (%)	59(84.29)	975(81.18)	651(81.17)	616(87.03)	46(76.67)	0.269
Vasoactive drugs use, n (%)	51(72.86)	722(60.12)	364(45.39)	87(47.03)	15(25.00)	<0.001
Sedative drug use, n (%)	52(74.29)	1040(86.59)	724(90.27)	156(84.32)	48(80.00)	<0.001
SOFA score, M (Q1, Q3)	13(10,16)	9(6,12)	8(5,11)	8(6,11)	7(5,11)	<0.001
APSIII score, M (Q1, Q3)	93(69,116)	61(41,88)	54(34,82)	62(44,84)	55(40,71)	<0.001
CRRT for SA-AKI, n (%)	23(32.86)	155(12.91)	80(9.98)	17(9.19)	6(10.00)	<0.001
28-d mortality,n (%)	42(60.00)	283(23.56)	132(16.46)	34(18.38)	10(16.67)	<0.001
90-d mortality, n (%)	44(62.86)	324(26.98)	158(19.70)	42(22.70)	13(21.67)	<0.001
Hospital mortality, n (%)	40(57.14)	277(23.06)	120(14.96)	34(18.38)	11(18.33)	<0.001
ICU mortality, n (%)	38(54.29)	251(20.90)	110(13.72)	27(14.59)	7(11.67)	<0.001
Hospital LOS, day, M (Q1, Q3)	7.7(1.4,18.0)	9.4(6.1,15.4)	8.9(5.6,15.8)	11.7(6.6,19.9)	12.5(7.1,18.3)	<0.001
ICU LOS, day, M (Q1, Q3)	4.0(1.4,10.0)	4.8(2.9,8.9)	4.1(2.3,7.9)	5.1(2.9,10.0)	3.6(2.1,10.4)	0.003

^#The initial value and ^$ the maximum value of creatinine.

CHD: coronary heart disease; HF: acute or chronic heart failure; CKD: chronic kidney disease; HT: hypertension; DM: diabetes mellitus; HR: heart rate; RR: respiratory rate; WBC: white blood cell; MCV: mean corpuscular volume; RDW: red blood cell distribution width; INR: international normalized ratio; PT: prothrombin time; APTT: activated partial thromboplastin time; AG: anion gap; BUN: blood urea nitrogen; SOFA: Sequential organ failure assessment; APS: acute physiology score; CRRT: continuous renal replacement therapy; SA-AKI: sepsis-associated acute kidney injury; ICU: intensive care unit; LOS: length of stay

3.3. Non-linear association between MPP and risk of mortality

For each patient, the initial, mean, maximum, and minimum MPP values within 24 h post-SA-AKI diagnosis were calculated. The RCS curve (Figure 3(A)) visualized the nonlinear negative relationship between MPP and 90-d mortality risk. As MPP increased, the risk of mortality declined nonlinearly (p for non-linearity < 0.001), with the reduction being more pronounced at low MPP levels. However, the decline plateaued at higher MPP levels. The RCS curve identified 60 mmHg as the optimal threshold, below which MPP was an independent risk factor for increased mortality. These results are consistent with trajectory analysis, where Traj-1 and Traj-2, representing MPP levels below 60 mmHg, were associated with higher mortality risks. This underscores the importance of maintaining adequate MPP levels in patients with SA-AKI.

Figure 3.

Nonlinear relationship between MPP values and 90-d mortality risk. (A) RCS curves illustrate a nonlinear negative association between initial, mean, minimum, and maximum MPP values and 90-d mortality risk. Lower MPP levels are associated with a higher OR (or) for mortality, with 60 mmHg identified as the optimal cutoff for average MPP. (B) Kaplan–Meier survival curves depict significant survival differences between patients with low (MPP ≤ 60 mmHg) and high (MPP > 60 mmHg) MPP levels. (C) ROC curves evaluate the clinical predictive value of various MPP metrics for 90-d mortality in patients with SA-AKI, identifying the average MPP as the most predictive measure with an AUC of 0.581.

Based on the threshold of 60 mmHg, patients were stratified into low-MPP (MPP ≤ 60 mmHg, N = 1087) and high-MPP (MPP > 60 mmHg, N = 1231) groups. Baseline characteristics and outcomes for these groups are summarized in Table 2. Patients in low-MPP group had significantly higher rate of mortality and usage of CRRT intervention than those in high-MPP group (all p < 0.001). Low-MPP group also needed longer ICU-LOS (p = 0.023), but not significant difference was found in hospital-LOS (p = 0.421). The K–M survival curve (Figure 3(B)) indicated that survival rates in the low-MPP group were significantly lower than in the high-MPP group over time. Both Log-rank and Wilcoxon tests confirmed statistically significant differences between the groups (p < 0.001 for both). The ROC curve analysis (Figure 3(C)) further assessed the predictive value of MPP for 90-d mortality in patients with SA-AKI. The AUC ranged from 0.545 to 0.581, indicating mild predictive utility. Among all calculated MPP values, the mean MPP demonstrated the highest predictive value, with an AUC of 0.581.

Table 2.

Characteristics and clinical outcomes of patients in low-MPP and high-MPP groups.

	Low-MPP	High-MPP	p Value
Age, years, M (Q1, Q3)	72.5(62.0,80.7)	68.5(59.2,78.0)	<0.001
Male, n (%)	647(59.52)	775(62.96)	0.090
Race-White, n (%)	764(70.29)	803(65.23)	0.009
Body weight, kg, M (Q1, Q3)	82.4(69.7,98.0)	80.7(67.3,95.2)	0.004
CVD, n (%)	551(50.69)	546(44.35)	0.002
HF, n (%)	409(37.63)	367(29.81)	<0.001
Liver disease, n (%)	184(16.93)	217(17.63)	0.656
CKD, n (%)	293(26.95)	264(21.45)	0.002
HT, n (%)	478(43.97)	599(48.66)	0.024
DM, n (%)	409(37.63)	400(32.49)	0.010
HR, bpm, M (Q1, Q3)	84(76,98)	84(77,96)	0.402
RR, cpm, M (Q1, Q3)	18(14,22)	17(14,22)	0.479
Temperature, M (Q1, Q3)	36.7(36.3,37.0)	36.6(36.1,37.0)	0.173
SpO2, %, M (Q1, Q3)	99(96,100)	100(97,100)	0.025
WBC, K/UL, M (Q1, Q3)	13.3(9.4,18.5)	12.6(9.1,17.0)	0.004
Hemoglobin, mg/dL, M (Q1, Q3)	10.0(8.9,11.2)	10.4(9.2,11.7)	<0.001
Platelet, K/UL, M (Q1, Q3)	149(113,208)	149(112,207)	0.915
MCV, dL, M (Q1, Q3)	91(88,95)	90(87,94)	0.005
RDW, %, M (Q1, Q3)	14.7(13.8,16.1)	14.5(13.5,15.6)	<0.001
INR, M (Q1, Q3)	1.3(1.2,1.6)	1.3(1.2,1.5)	<0.001
PT, sec, M (Q1, Q3)	14.7(13.4,17.5)	14.4(13.0,16.2)	<0.001
APTT, sec, M (Q1, Q3)	34.2(29.7,43.4)	33.0(28.5,40.8)	<0.001
Na, mEq/L, M (Q1, Q3)	139(136,141)	139(137,141)	0.068
K, mEq/L, M (Q1, Q3)	4.3(3.9,4.8)	4.3(3.9,4.7)	0.029
Cl, mEq/L, M (Q1, Q3)	107(104,111)	107(104,111)	>0.999
AG, mEq/L, M (Q1, Q3)	14(11,17)	13(11,16)	0.043
Glucose, mg/dL, M (Q1, Q3)	140(112,176)	140(115,177)	0.380
Lactate, mmol/L, M (Q1, Q3)	2.3(1.6,3.6)	2.1(1.5,3.0)	<0.001
BUN, mg/dL, M (Q1, Q3)	22(16,35)	20(15,29)	<0.001
#Creatinine, mg/dL, M (Q1, Q3)	1.2(0.9,1.8)	1.0(0.8,1.5)	<0.001
$Creatinine, mg/dL, M (Q1, Q3)	1.8(1.3,2.8)	1.5(1.1,2.4)	<0.001
Mechanical ventilation use, n (%)	885(81.42)	1,007(81.80)	0.810
Vasopressor use, n (%)	675(62.01)	565(45.90)	<0.001
Sedative drug use, n (%)	925(85.10)	1,095(88.95)	0.006
SOFA score, M (Q1, Q3)	9(6,12)	8(5,11)	<0.001
APSIII score, M (Q1, Q3)	65(43,92)	56(36,82)	<0.001
CRRT for SA-AKI, n (%)	161(14.81)	120(9.75)	<0.001
28-d mortality, n (%)	291(26.77)	210(17.06)	<0.001
90-d mortality, n (%)	324(29.81)	257(20.88)	<0.001
Hospital mortality, n (%)	281(25.85)	201(16.33)	<0.001
ICU mortality, n (%)	256(23.55)	177(14.38)	<0.001
Hospital LOS, day, M (Q1, Q3)	9.4(6.0,15.5)	9.5(5.9,16.6)	0.421
ICU LOS, day, M (Q1, Q3)	4.9(2.7,9.0)	4.2(2.3,8.6)	0.023

^#The initial value and ^$ the maximum value of creatinine.

CHD: coronary heart disease; HF: acute or chronic heart failure; CKD: chronic kidney disease; HT: hypertension; DM: diabetes mellitus; HR: heart rate; RR: respiratory rate; WBC: white blood cell; MCV: mean corpuscular volume; RDW: red blood cell distribution width; INR: international normalized ratio; PT: prothrombin time; APTT: activated partial thromboplastin time; AG: anion gap; BUN: blood urea nitrogen; SOFA: Sequential organ failure assessment; APS: acute physiology score; CRRT: continuous renal replacement therapy; SA-AKI: sepsis-associated acute kidney injury; ICU: intensive care unit; LOS: length of stay

3.4. Cox regression analysis

In all models, Traj-1 and Traj-2 were identified as independent risk factors for 90-d mortality in patients with SA-AKI, using Traj-3 as the reference group (Table 3). Patients in Traj-1 demonstrated a substantially higher risk of mortality, with HRs ranging from 2.98 to 4.67, corresponding to a 198–367% increased risk (all p < 0.001). Similarly, Traj-2 was associated with a 28–47% increased risk of mortality (HR 1.28–1.47, all p < 0.05). No significant associations were observed between either Traj-4 or Traj-5 and mortality risk (all p > 0.05). Additionally, patients in the high-MPP group (MPP > 60 mmHg) had a lower risk of 90-d mortality compared to those in the low-MPP group (MPP ≤ 60 mmHg), with HRs ranging from 0.65 to 0.76 (all p < 0.001).

Table 3.

Cox regression analysis for patients in different trajectories and categories.

	Unadjusted model		Adjusted model-1		Adjusted model-2
Trajectory model	HR (95% CI)	p Value	HR (95% CI)	p Value	HR (95% CI)	p Value
Traj-1	4.67(3.28–6.67)	<0.001	4.34(3.02–6.13)	<0.001	2.98(2.06–4.29)	<0.001
Traj-2	1.45(1.20–1.76)	<0.001	1.47(1.21–1.79)	<0.001	1.28(1.06–1.56)	0.012
Traj-3	Reference		Reference		Reference
Traj-4	1.16(0.82–1.64)	0.395	1.04(0.74–1.47)	0.818	1.02(0.72–1.44)	0.918
Traj-5	1.04(0.58–1.86)	0.908	0.91(0.50–1.63)	0.743	1.00(0.55–1.81)	0.993
Categories
MPP ≤ 60 mmHg	Reference		Reference		Reference
MPP > 60 mmHg	0.65(0.55–0.77)	<0.001	0.65(0.55–0.77)	<0.001	0.76(0.64–0.90)	<0.001

Adjusted Model 1: Adjustments were made for demographic characteristics and comorbidities.

Adjusted Model 2: Building on Adjusted Model 1, further adjustments were incorporated for specific interventions and SOFA scores.

3.5. Subgroup analysis

Figure 4 presents a forest plot summarizing the results of subgroup analyses. The findings align with those observed in the overall cohort, confirming that MPP > 60 mmHg serves as a protective factor (HR < 1) compared to MPP ≤ 60 mmHg. In the unadjusted model, a significant interaction was identified between liver disease and MPP (p for interaction = 0.029), indicating that the protective effect of high MPP was more pronounced in patients with liver disease (HR 0.47, 95% CI 0.35–0.67) compared to those without liver disease (HR 0.71, 95% CI 0.59–0.87). However, in the adjusted model, this interaction was no longer significant (P for liver disease × MPP interaction = 0.552). Although there was no significant interaction found in the adjusted models (all p for interaction > 0.05), markedly difference was found in race (p for interaction = 0.062) and HT subgroups (p for interaction = 0.153). High MPP demonstrated a significant protective effect in non-White patients (HR 0.60, 95% CI 0.46–0.79) and normotensive patients (HR 0.68, 95% CI 0.55–0.85), but not in White patients (HR 0.90, 95% CI 0.72–1.12) and hypertensive populations (HR 0.93, 95% CI 0.70–1.24).

Figure 4.

Effect size of different MPP categories on 90-d mortality in prespecified and exploratory subgroups. (A) Unadjusted model. (B) Adjusted model. Mean MPP ≤ 60 mmHg was used as the reference, and hazard ratios (HRs) were adjusted according to Model 2 from Table 3.

3.6. Sensitivity analysis

Baseline characteristics between low-MPP and high-MPP groups were balanced using PSM. After matching, no statistically significant differences were observed between the groups in baseline parameters, except for the SOFA score (p = 0.001, Table 4). PSM also reduced the standardized mean difference (SMD) of SOFA scores between the groups (Figure S1). In the PSM cohort, the high-MPP group exhibited a significantly lower 90-d mortality rate compared to the low-MPP group (23.06% vs. 28.38%, p = 0.010). Additionally, the high-MPP group had a reduced percentage of CRRT intervention (10.64% vs. 13.75%, p = 0.044). No significant differences were observed in LOS between the groups (p > 0.05).

Table 4.

Comparison between low-MPP patients and high-MPP patients in the PSM cohort.

	Low-MPP	High-MPP	p Value
Age, years, M (Q1, Q3)	71.1(60.7,80.2)	71.1(61.4,79.8)	0.493
Male, n (%)	554(61.42)	547(60.64)	0.735
Race-White, n (%)	613(67.96)	634(70.29)	0.285
Body weight, Kg, M (Q1, Q3)	81.0(69.0,97.0)	83.0(69.0,98.6)	0.258
CVD, n (%)	443(49.11)	453(50.22)	0.638
HF, n (%)	312(34.59)	318(35.25)	0.767
Liver disease, n (%)	148(16.41)	139(15.41)	0.562
CKD, n (%)	216(23.61)	230(25.50)	0.352
HT, n (%)	417(46.23)	416(46.12)	0.962
DM, n (%)	318(35.25)	314(34.81)	0.844
Mechanical ventilation use, n (%)	742(82.26)	731(81.04)	0.503
Vasopressor use, n (%)	501(55.54)	522(57.87)	0.318
Sedative drug use, n (%)	797(88.36)	781(86.59)	0.255
SOFA score, M (Q1, Q3)	9(6,12)	8(6,11)	0.001
CRRT for SA-AKI, n (%)	124(13.75)	96(10.64)	0.044
28-d mortality, n (%)	231(25.61)	170(18.85)	0.001
90-d mortality, n (%)	256(28.38)	208(23.06)	0.010
Hospital mortality, n (%)	223(24.72)	163(18.07)	0.001
ICU mortality, n (%)	205(22.73)	144(15.96)	<0.001
Hospital LOS, day, M (Q1, Q3)	9.3(6.0,15.1)	9.7(5.9,16.7)	0.175
ICU LOS, day, M (Q1, Q3)	4.7(2.6,8.6)	4.4(2.4,8.9)	0.859

CHD: coronary heart disease; HF: acute or chronic heart failure; CKD: chronic kidney disease; HT: hypertension; DM: diabetes mellitus; SOFA: Sequential organ failure assessment; CRRT: continuous renal replacement therapy; SA-AKI: sepsis-associated acute kidney injury; ICU: intensive care unit; LOS: length of stay

Patients were categorized into low-MPP (MPP ≤ 60 mmHg) and high-MPP (MPP > 60 mmHg) groups based on a cutoff value 60 mmHg. PSM was performed using a 1:1 nested matching design without replacement, with a caliper value of 0.05 to ensure optimal balance between groups.

4. Discussion

This study investigated the association between early MPP levels and short-term mortality risk in patients with SA-AKI. The analysis examined not only the predictive value of single-point MPP measurements but also the impact of dynamic changes in MPP on patient outcomes. Trajectory modeling revealed that patients in Traj-1 and Traj-2, whose MPP consistently remained below 60 mmHg, exhibited significantly higher mortality rates and required a greater proportion of CRRT. RCS curves demonstrated a nonlinear negative correlation between MPP and 90-d mortality, identifying 60 mmHg as the optimal cutoff value. Patients with MPP ≤ 60 mmHg had significantly higher mortality rates compared to those with MPP > 60 mmHg, consistent with trajectory model findings. After adjusting for potential confounders in Cox regression analysis, Traj-1 and Traj-2 remained independent risk factors for mortality in SA-AKI, while an MPP > 60 mmHg was associated with a reduced mortality risk. Subgroup analysis and PSM sensitivity analysis corroborated these results, consistently identifying MPP > 60 mmHg as a protective factor. These results suggest that maintaining renal perfusion pressure levels, specifically with an MPP > 60 mmHg during the early stages of SA-AKI, may improve clinical outcomes. This study provides critical insights into early intervention and therapeutic strategies for managing SA-AKI.

Tissue hypoperfusion is a major contributor to multiple organ dysfunction in sepsis, driven by factors such as capillary leakage, vasodilation, and sepsis-induced cardiac dysfunction [25,26]. Hypotension leads to inadequate perfusion, initiating a cascade of organ failures. Maintaining optimal volume status to preserve organ perfusion is therefore pivotal in sepsis management. The kidney, which receives approximately 20–25% of systemic blood flow, is particularly vulnerable to sepsis-induced damage. Impaired renal hemodynamic autoregulation has been identified as a key factor affecting kidney function. Under normal circumstances, the kidney can rely on autoregulation mechanism to maintain stable perfusion within a certain range of arterial blood pressure. However, disruptions caused by hypovolemia, inflammatory responses, and toxic effects can impair autoregulation, leading to fluctuations in glomerular filtration rate (GFR) and contributing to the onset of AKI [27–29]. Hypotensive episodes exacerbate this issue by significantly reducing effective renal blood flow, causing microcirculatory dysfunction, decreased GFR, and impaired tubular reabsorption, rapidly progressing to AKI. Reports estimate that 42–48% of sepsis cases are complicated by SA-AKI [30]. Thus, improving renal perfusion through hemodynamic management is critical for preventing SA-AKI and enhancing patient outcomes [31]. However, clinical tools for directly assessing renal blood flow and perfusion pressure remain limited, highlighting the need for further research and innovation in this area.

MAP has traditionally served as a key indicator of tissue perfusion. Previous studies have established that persistently low MAP levels, particularly below 60–65 mmHg, are strongly associated with adverse outcomes in critically ill patients [32,33]. Current guidelines from the SSC recommend maintaining MAP above 65 mmHg in patients with septic shock receiving vasopressors, as this threshold is considered effective in reducing the risk of mortality and organ failure [34]. The sepsis and mean arterial pressure (SEPSISPAM) trial examined the impact of MAP-targeted resuscitation on septic shock outcomes [35]. This study found no significant differences in 28- or 90-d mortality or severe adverse events between patients maintained at a low MAP target (65–70 mmHg) and those at a high MAP target (80–85 mmHg) for 5 d. However, the high MAP target was associated with reduced rates of serum creatinine doubling and decreased need for CRRT, albeit at the expense of an increased risk of atrial fibrillation [35]. Despite its widespread use, recent evidence indicates that MAP alone may not sufficiently capture tissue and organ perfusion levels [15]. Renal afterload factors, particularly CVP, must also be considered. Elevated CVP has been shown to impair renal venous outflow, cause renal congestion, and diminish renal perfusion [11]. A meta-analysis identified a significant association between elevated CVP and increased mortality (or, 1.65; 95% confidence interval [CI], 1.19–2.29) as well as a heightened risk of AKI (or, 2.09; 95% CI, 1.39–3.14) in critically ill patients [36]. Similarly, the vasopressin in septic shock trial (VASST) demonstrated that patients with septic shock exhibiting CVP > 12 mmHg had the highest mortality rates, whereas those with CVP < 8 mmHg had the lowest risk of death [37]. These observations underscore the importance of reducing CVP to maintain renal perfusion, highlighting it as a critical strategy for improving outcomes in septic patients.

In recent years, MPP has gained recognition as a more reliable clinical indicator than MAP and CVP for assessing renal perfusion and predicting the occurrence of AKI. Low MPP levels have proven to be significant predictors of AKI in critically ill patients, including those with sepsis or undergoing cardiac surgery [12,13,16]. Ostermann et al. reported that in critically ill ICU patients, an MPP ≤ 59 mmHg serves as an independent risk factor for the progression of AKI to stage III [15]. Similarly, a study by Ling Li et al. involving 5,867 patients with sepsis, found that higher MPP levels were associated with a reduced risk of AKI (OR 0.98, 95% CI 0.97–0.99) [38]. Using ROC curves, the study identified an optimal MPP cutoff, indicating that patients with an MPP > 63 mmHg had lower mortality rates. In this study, RCS curves and trajectory models identified an optimal MPP cutoff of 60 mmHg, consistent with the 59 and 63 mmHg thresholds suggested by previous research. The findings underscore the importance of avoiding low MPP levels to improve outcomes in patients with SA-AKI, such as reducing mortality and the need for CRRT intervention. However, further increasing MPP beyond 70 or 80 mmHg did not yield additional clinical benefits. Excessively elevated perfusion pressures may not improve outcomes and can even lead to adverse effects. For example, the SEPSISPAM trial demonstrated that increasing MAP to 80–85 mmHg did not improve patient prognosis but raised the risk of atrial fibrillation in patients with sepsis [35]. Additionally, a study using a mechanically ventilated pig model of septic shock revealed that targeting a MAP of 75–85 mmHg resulted in excessive fluid overload and vasopressor requirements during resuscitation without suppressing the inflammatory response [5]. These observations suggest that overly high perfusion pressures offer limited benefits and may induce ischemic injuries due to excessive vasoconstriction. Notably, patients with HT may require higher perfusion pressure levels for favorable outcomes compared to those without HT. Previous research indicated that hypertensive patients with SA-AKI benefit from higher MAP ranges (70–80 mmHg) compared to normotensive patients (65–73 mmHg) to maintain renal perfusion [,39]. In this study, we also examined the predictive value of MPP in both hypertensive and normotensive populations, revealing no statistical differences between these groups. However, our positive findings were present only in the normotensive population, suggesting that the demand for MPP in hypertensive patients may differ from that in normotensive individuals, which remains an intriguing question for further investigation. A similar pattern was observed in the racial subgroup analysis. We found positive results in non-White populations, while this conclusion was not supported in the White population. The reasons for this discrepancy are unknown. Previous research has indicated that Black and Latino populations in the United States experience a significantly higher prevalence of HT than their White counterparts [40]. Additionally, racial disparities also affect HT management, as both Latino and Black individuals tend to have lower adherence rates to antihypertensive medications compared to Whites [41]. This lack of adherence could stem from a combination of factors, including limited awareness of HT and unequal access to healthcare resources for these group [42,43]. These findings remind us that racial differences may impact the relationship between MPP and the prognosis of patients with SA-AKI, highlighting the need for further research in this area.

The findings of this study have important implications for guiding the early treatment of patients with SA-AKI. In clinical practice, optimal fluid management for sepsis and septic shock typically involves a phased approach: transitioning from initial liberal positive fluid infusion to a steady-state fluid balance, followed by the controlled removal of excess fluids that may contribute to organ overload. While prior research has emphasized the role of MAP in guiding fluid resuscitation and enhancing renal perfusion [35,44], MAP does not always accurately reflect organ perfusion. CVP provides an additional indicator of volume status, with the SSC guidelines recommending CVP levels of 8–12 mmHg to ensure effective fluid resuscitation [7]. However, positive fluid balance can impair organ function, including renal function, and is associated with increased mortality [45]. Mullens et al. observed that patients who developed AKI exhibited higher baseline and post-treatment CVP levels [46], while another study noted that SA-AKI is particularly sensitive to fluid accumulation compared to AKI caused by other factors [47]. Moreover, several observational studies have linked volume overload in critically ill patients with AKI to increased mortality and delayed renal recovery [48,49]. These findings underscore the critical role of fluid management in AKI. The latest SSC guidelines advocate for using comprehensive and dynamic indicators, rather than static measures like MAP and CVP, to assess fluid status and guide resuscitation in patients with sepsis [50]. In this study, MPP – calculated as the difference between MAP and CVP – was utilized as a dynamic and comprehensive marker to reflect the relationship between renal perfusion and clinical outcomes in patients with SA-AKI. RCS analysis demonstrated that an MPP > 60 mmHg at a single time point was associated with reduced mortality in patients with SA-AKI. Furthermore, trajectory modeling revealed that maintaining a dynamic MPP above 60 mmHg is crucial for reducing short-term mortality in these patients. However, excessively high dynamic MPP levels did not confer additional clinical benefits, highlighting the importance of avoiding overly aggressive perfusion pressure targets. These findings suggest that MPP may serve as a valuable indicator for fluid management and prognosis improvement in patients with SA-AKI. It should be noted, however, that ROC curve analysis revealed that MPP alone has a limited ability to predict outcomes. This limitation is partly due to the non-linear relationship between MPP and mortality risk, as well as the complex pathogenesis of SA-AKI. Thus, while maintaining higher MPP levels may improve patient outcomes, relying solely on MPP as a single indicator is insufficient. A more comprehensive evaluation incorporating additional parameters is needed to guide treatment effectively. Numerous newly reported methods have proven effective in assessing renal perfusion, garnering attention, particularly ultrasonography and positron emission tomography (PET). Recent studies propose that bedside Doppler ultrasonography for intrarenal venous flow (IRVF) assessment may offer a real-time, noninvasive method for evaluating renal venous flow and guiding fluid resuscitation in critically ill patients [51–53]. Exploratory research conducted by Bibeau-Delisle A et al. introduced a renal assessment method based on noninvasive 82-rubidium PET imaging, revealing that PET-detected renal blood flow and renal vascular resistance can effectively reflect renal function. PET-measured RBF and RVR correlate with renal function markers and differ significantly by renal function status [54]. Future research should integrate multimodal detection methods, including tools like IRVF and rubidium PET imaging, to identify optimal blood pressure targets for renal perfusion and improve patient outcomes.

Several limitations of this study must be acknowledged. First, as a retrospective analysis, the potential influence of confounding factors cannot be entirely ruled out. Although Cox regression and PSM were employed to mitigate their effects, unmeasured or unanticipated confounders may still introduce bias into the results. Second, the generalizability of the findings is constrained by the single-center design and the relatively small sample size, which necessitate cautious interpretation. Validation through prospective, large-scale, and high-quality clinical studies will be essential in future research. Third, this study focused primarily on short-term outcome indicators, including mortality, LOS, and CRRT usage. Due to the inherent limitations of the database, data on long-term outcomes, such as renal function recovery and the progression of chronic kidney disease, were unavailable. Fourth, different kinds of vasoactive drugs have specific effects on hemodynamics and renal perfusion pressure [31], but these differences were overlooked in this study and may lead to potential bias, despite adjusting for the use of vasoactive drugs as a confounding factor. Future research should emphasize long-term prognosis to provide a more comprehensive understanding of SA-AKI.

5. Conclusion

The study demonstrated that dynamic MPP levels during the early stages of SA-AKI are closely associated with short-term mortality, with maintaining MPP > 60 mmHg identified as a protective factor. However, excessively high MPP levels did not confer additional survival benefits. These findings underscore the importance of early renal perfusion management and highlight the need for individualized treatment strategies based on dynamic MPP monitoring to improve the prognosis of patients with SA-AKI.

Funding Statement

This research was funded by grants from three sources: Clinical Research Program of Tianjin Medical University General Hospital (Grant No. 22ZYYLCCG05), China International Medical Foundation’s 2021-N-15 project (Grant No. 2021-N-15-8), and Tianjin Health Research Project (Grant No. TJWJ2024XK006). Keliang Xie was the primary recipient of these financial supports.

Ethics approval and consent to participate

The MIMIC-IV dataset, which is anonymized and IRB-approved by the MIT, does not require informed consent. As no additional data collection was undertaken, further ethical statements are unnecessary for this study.

Consent for publication

Not applicable.

Authors’ contributions

Yipeng Fang: Writing – original draft, Conceptualization, Data curation, Formal Analysis, Methodology, Project administration, Software, Validation, Visualization. Aizhen Dou: Writing – original draft, Data curation, Formal Analysis, Methodology. Hui Xie: Writing – review & editing, Conceptualization, Data curation, Formal Analysis, Methodology. Weiwei Zhu, Yanchao Su, Caifeng Li: Writing – review & editing, Conceptualization. Yingjin Zhang: Writing – review & editing. Yunfei Zhang: Writing – review & editing, Formal Analysis, Visualization. Keliang Xie and Ying Gao: Writing – review & editing, Conceptualization, Funding acquisition, Project administration, Supervision. All contributing authors have carefully examined and approved the final version of the manuscript, ensuring the integrity and taking responsibility for all aspects of the research presented.

Disclosure statement

The authors declare that they have no competing interests.

Data availability statement

The datasets analyzed during this study are available in the MIMIC-IV version 2.2 (accessible at https://physionet.org/content/mimiciv/2.2/). All raw data can be obtained through the corresponding author, Keliang Xie, and first author Yipeng Fang, upon reasonable request.