Impact of early albumin therapy on 28-day survival in ICU patients with septic shock concurrent hypoalbuminemia: a retrospective analysis of combined clinical datasets

Abstract

BACKGROUND To evaluate the association between early albumin infusion and 28-day mortality in intensive care unit (ICU) patients with septic shock and concurrent hypoalbuminemia.

METHODS This study analyzed data from the MIMIC-IV database and admission records from the university-affiliated teaching hospital ICU. Patients with septic shock and concurrent hypoalbuminemia were categorized into early albumin infusion (within 72 h of ICU admission) and late albumin infusion (after 72 h of ICU admission), regardless of dosage or colloid osmotic pressure. The primary outcome was the 28-day mortality rate.

RESULTS In the MIMIC-IV cohort, 1,216 patients received an albumin infusion, including 546 with early infusion and 670 with late infusion (odds ratios [ORs]=1.556, 95% confidence intervals [95% CIs]: 1.193-2.030). Early albumin infusion was associated with significantly higher 28-day mortality compared to late or no infusion. Subgroup analysis showed a survival benefit in patients with severe hypoalbuminaemia, but worse outcomes in those with mild-to-moderate hypoalbuminaemia. These findings were partially validated in an independent cohort from Sun Yat-sen Memorial Hospital (n=1,233), where similar trends were observed. Early infusion was associated with increased mortality, and albumin use in mild-to-moderate hypoalbuminaemia was associated with worse survival.

CONCLUSION Early albumin infusion for patients with septic shock did not demonstrate a clear clinical benefit regarding 28-day mortality. Late infusion, however, appeared to be associated with improved survival outcomes. Among patients with septic shock and severe hypoalbuminemia, albumin infusion was associated with higher 28-day survival. In contrast, for those with mild-to-moderate hypoalbuminemia, albumin infusion did not offer a significant survival advantage.

INTRODUCTION

Septic shock is a severe and often fatal condition encountered in intensive care units (ICUs) and typically results from infection, hemorrhage, or other critical insults. It is characterized by acute circulatory dysfunction, tissue hypoperfusion, and cellular hypoxia.^[1-3] Despite recent advances in the recognition and management of sepsis, septic shock frequently progresses rapidly to multiple organ dysfunction syndrome (MODS) and remains associated with a mortality rate exceeding 40%.^[4-6] Hypoalbuminemia is common in patients with septic shock and may result from impaired hepatic synthesis, increased capillary leakage secondary to systemic inflammation, heightened catabolism, or nutritional deficiencies.^[7,8] Recent studies have shown that both lower admission albumin levels and persistently low levels during the first ICU week are linked to worse outcomes, whereas higher or rising trajectories are associated with improved survival.^[9-11] Collectively, the current evidence indicates that hypoalbuminemia not only reflects disease severity but also functions as an independent predictor of poor prognosis, highlighting the potential clinical value of its early recognition and correction.^[12,13]

Albumin infusion is often considered in septic shock patients with hypoalbuminaemia.^[14-16] In this population, potential benefits include restoring plasma colloid osmotic pressure to counteract capillary leak, stabilizing hemodynamics, and reducing cumulative fluid balance.^[17] Observational studies have also suggested that early supplementation in hypoalbuminaemic sepsis may shorten vasopressor duration and enhance organ function recovery.^[14,18] Albumin also exerts anti-inflammatory effects and plays a role in the transport of drugs and fatty acids, as well as in acid-base buffering.^[19,20] However, the inappropriate use of albumin may result in adverse effects, such as fluid overload, pulmonary edema, and cardiac dysfunction.^[21,22] Post hoc analyses of the SAFE trial found no mortality benefit—and a higher incidence of pulmonary edema in some subgroups—when albumin was given without consideration of baseline albumin levels.^[23] The application of albumin therapy in septic shock has been a subject of ongoing debate.^[24-27] Some studies suggest that early albumin infusion could improve 28-day survival among patients with shock upon ICU admission; however, others, including large-scale randomized controlled trials such as the SAFE study, have reported no significant differences in major outcomes compared with crystalloid infusion.^[27,28] Additionally, some studies have indicated the potential harm or lack of benefit associated with albumin infusion in critically ill patients.^[29,30] These discrepancies highlight the ongoing uncertainty regarding the optimal indications and timing for albumin administration in patients with septic shock.^[17,24,25]

The MIMIC-IV database provides comprehensive clinical records, including diagnoses and therapeutic interventions, offering a rich source of information for clinical research and hypothesis generation. However, owing to potential population differences between Western cohorts and other regions, findings derived solely from the MIMIC-IV may lack generalizability. To address these limitations and enhance the robustness of the results, this study also incorporates real-world ICU data from Sun Yat-sen Memorial Hospital, a large university-affiliated teaching hospital in China. By combining both datasets, we aimed to investigate the impact of different albumin infusion strategies on 28-day mortality in ICU patients with septic shock across diverse patient populations.

METHODS

Study participants

This study utilized data from two primary sources: the MIMIC-IV database and admission records from the Department of Critical Care Medicine at Sun Yat-sen Memorial Hospital, Sun Yat-sen University. The MIMIC-IV database (version 3.0), developed and maintained by the MIT Laboratory for Computational Physiology, is a large, publicly available critical care database containing deidentified clinical data from patients admitted to the emergency department or intensive care units at Beth Israel Deaconess Medical Center between 2008 and 2022 (https://physionet.org/content/mimiciv/).^[31,32] Owing to the deidentified nature of the data, additional ethical approval and informed consent were not needed. Access to the MIMIC-IV database was granted after the principal investigator (FRL) completed the Collaborative Institutional Training Initiative (CITI) course and passed the required exams on “Conflicts of Interest” and “Data or Specimens Only Research” (Pass ID: 51867208). To enhance the generalizability of the findings and account for potential population differences, a complementary dataset was obtained from Sun Yat-sen Memorial Hospital. This hospital’s electronic medical records system provided detailed ICU patient data, ensuring data completeness and reliability. The use of this real-world dataset was approved by the institutional ethics committee of Sun Yat-sen Memorial Hospital, Sun Yat-sen University (Ethics Approval ID: LX-HG-2024020).

Population

Our study included patients with septic shock from the MIMIC-IV database. We used Navicat (version 9.6.18) to construct a data management platform and extract the necessary data for our research. Enrolled patients were defined as critically ill individuals with septic shock, as confirmed by comorbidities for ICU admission. Septic shock was defined according to the Sepsis-3 criteria and supplemented by corresponding International Classification of Diseases (ICD-9 and ICD-10) codes to ensure completeness across different study periods. The study population was selected based on ICD-9 and ICD-10 codes: R6521 (severe sepsis with septic shock), 78552 (septic shock), T8112XA (post-procedural septic shock, initial encounter), and 99802 (post-operative shock, septic).

The exclusion criteria were as follows: (1) not first-time hospital admission, or not first-time ICU admission; (2) no human serum albumin (HSA) levels recorded on first day of ICU admission; (3) non-hypoproteinemic; (4) patients with non-septic shock.

A total of 3,401 patients from the MIMIC-IV database met the inclusion and exclusion criteria and were included in the final analysis. Additionally, an independent validation cohort was constructed using critically ill patients with septic shock admitted to the ICU of Sun Yat-sen Memorial Hospital, Sun Yat-sen University, between January 2021 and December 2024. The same inclusion and exclusion criteria were applied to ensure consistency and comparability between cohorts. In the end, 1,233 patients were enrolled in this study.

Data extraction

We extracted patient information from the first day in the ICU, including age, gender, height, weight, race, laboratory indicators (i.e., red blood cells, white blood cells, platelets, haemoglobin, total bilirubin, serum glucose, serum creatinine, blood urea nitrogen, and lactate), vital signs (i.e., body temperature, heart rate, blood pressure, and mean arterial pressure [MAP], and respiratory rate), total urine volume during the first 24 h in the ICU, comorbidities (i.e., congestive heart failure, cerebrovascular disease, chronic pulmonary disease, diabetes, liver disease, and hypertension), mechanical ventilation status, albumin infusion status, and scores from various quantification systems from the first 24 h in the ICU (Logistic Organ Dysfunction System [LODS], Oxford Acute Illness Severity Score [OASIS], Glasgow Coma Scale [GCS], Simplified Acute Physiology Score II [SAPS-II], Charlson Comorbidity Index. The follow-up data included mortality and days until death. Missing data were assessed using Little’s missing completely at random (MCAR) multivariate test (P<0.05) and were handled using a multiple imputation method to minimize bias. Data extraction was performed using PostgreSQL software (version 16.2.1) and Navicat Premium software (version 17.0.4) with Structured Query Language (SQL). All the variables were the first values measured on the first day of the patient’s admission to the ICU.

Similarly, for the cohort from Sun Yat-sen Memorial Hospital, data were extracted from the hospital’s electronic medical record system using identical variable definitions and time windows to ensure consistency and comparability with the MIMIC-IV cohort. Data collection followed standardized protocols for ICU data extraction and variable harmonization.

Infusion time definition

The time of albumin infusion was defined as the time interval between the first infusion of albumin and ICU admission. The first 72 h of septic shock represent a physiologically distinct phase characterized by rapid hemodynamic fluctuations, endothelial dysfunction, and marked capillary leakage, during which time the serum albumin concentration and fluid responsiveness differ substantially from those of later stages.^[33,34] Consistent with this early critical phase, previous studies have suggested that interventions initiated within the first 72 h of ICU admission—including early mobilization—may lead to different clinical outcomes and complication rates compared with those initiated afterward.^[35,36] Additional clinical investigations have also used the initial 72-hour period as a meaningful threshold for differentiating early versus late therapeutic interventions in critically ill patients.^[37] Patients who received albumin both before and after the 72-hour mark were categorized on the basis of the timing of their first dose.^[38] Patients were therefore classified into two groups, early albumin infusion (infusion within 72 h of ICU admission) and late albumin infusion (infusion more than 72 h after ICU admission), to evaluate the relationship between the timing of albumin infusion and 28-day mortality.

Variables and definitions

Albumin levels were assessed at the time of patient enrollment. Hypoalbuminemia was classified into two categories based on serum albumin concentration^[39]: an albumin concentration of 20 to 35 g/L was defined as mild to moderate hypoalbuminemia, and a concentration less than 20 g/L was defined as severe hypoalbuminemia.

Clinical outcomes

Patient survival data were obtained from the long-term follow-up records provided by the MIMIC-IV database. Patient survival data from Sun Yat-sen Memorial Hospital of Sun Yat-sen University were obtained through manual follow-up. The primary endpoint was 28-day all-cause mortality, defined as death occurring within 28 calendar days of ICU admission (index day), regardless of location (ICU or hospital ward).

Statistical analysis

Continuous variables were assessed for normality using the Kolmogorov-Smirnov test. Variables with a normal distribution are presented as the mean ± standard deviation and were compared between groups using independent-sample t-tests. Non-normally distributed variables are expressed as medians (interquartile range, IQR) and were compared using the Wilcoxon rank-sum test. Categorical variables are presented as frequencies (percentages) and were analyzed using the χ² test. To identify independent risk factors for 28-day mortality, a binomial logistic regression analysis was performed. Variables with a P-value<0.05 in univariate analysis were entered into the multivariate logistic regression model. Odds ratios (ORs) with 95% confidence intervals (95% CIs) were calculated to evaluate the association between albumin infusion (and its timing) and mortality risk. Survival analysis was conducted using Kaplan-Meier survival curves, with group comparisons assessed using the log-rank test.

To minimize potential confounding, propensity score matching (PSM) was performed separately for the MIMIC-IV cohort and the Sun Yat-sen Memorial Hospital cohort. Covariate selection for the propensity score model followed widely accepted principles of causal inference, with variables selected a priori based on their clinical relevance to both albumin infusion decisions and mortality risk, while posttreatment mediators were avoided.^[40,41] For the MIMIC-IV cohort, PSM was conducted based on the following clinical covariates: age and sex (baseline demographic factors), SAPS II score (overall severity of illness), MAP (an indicator of hemodynamic stability relevant to shock), baseline serum albumin level (the primary therapeutic target), and the presence of obesity (a factor affecting metabolic status and clinical outcomes).^[42] Propensity scores were calculated using a multivariate logistic regression model,^[43] and a 1:1 nearest-neighbor matching approach was applied using a caliper width of 0.01 without replacement. For the Sun Yat-sen Memorial Hospital cohort, matching was performed using age, sex, SAPS II score, MAP, and baseline serum albumin level, consistent with the locally available structured data and clinical relevance. After PSM, covariate balance was assessed using standardized mean differences (SMDs), with an SMD<0.1 indicating adequate balance. Baseline characteristics before and after matching are displayed in supplementary Figure 1, showing satisfactory balance across all the covariates.^[44] Subgroup analyses stratified by baseline albumin levels were performed to further explore the effect of albumin infusion strategies on survival outcomes across different hypoalbuminemia categories. Sensitivity analyses were performed to evaluate the robustness of the findings, including multivariable logistic regression conducted after PSM and calculation of E values for the association between early albumin infusion and 28-day mortality to assess the potential influence of unmeasured confounding.^[45,46] Variables with more than 20% missing data were excluded from the subsequent analysis. For variables with less than 20% missing data, missing values were addressed using the multiple imputation method. All the statistical analyses were performed using R software (version 3.5.1) and MedCalc software (version 19.6.1). A two-tailed P-value < 0.05 was considered to indicate statistical significance.

RESULTS

Baseline characteristics

A total of 3,401 adult patients with septic shock and hypoalbuminemia who met the inclusion and exclusion criteria were identified from the MIMIC-IV database. The cohort selection flowchart is presented in Figure 1. Among them, 1,216 patients received an albumin infusion, while 2,185 patients received a non-infusion. In terms of disease severity, the albumin infusion group had higher SAPS II scores compared to the non-albumin infusion group (50.0 [40.0,61.0] vs. 46.0 [36.0,58.5], P<0.05). After PSM, the baseline characteristics of the two groups are summarized in supplementary Table 1. No statistically significant differences were observed between the matched groups in age, sex, obesity status, MAP, serum albumin levels, or SAPS II scores, indicating effective matching.

Figure 1. Flowchart of cohort selection. Initially, adult patients diagnosed with shock between 2008 and 2022 were identified from the MIMIC-IV database. Of these, 18,326 patients were excluded based on predefined criteria, resulting in a final cohort of 3,401 patients for analysis. Additionally, following the same criteria, 1,233 consecutive critically ill patients admitted to the intensive care unit (ICU) of Sun Yat-sen Memorial Hospital between January 2021 and December 2024 were enrolled. HSA: human serum albumin.

A total of 1,233 adult patients with septic shock who met the inclusion and exclusion criteria were identified from the cohort at Sun Yat-sen Memorial Hospital, Sun Yat-sen University. The cohort selection flowchart is presented in Figure 1. Among them, 395 patients received an albumin infusion, while 838 patients did not receive an albumin infusion. In terms of disease severity, compared with the non-albumin infusion group, the albumin infusion group had higher SAPS II scores (16.0 [11.0,22.0] vs. 13.0 [6.0,21.0], P<0.05)and lower serum albumin levels (28 [25,31] mg/dL vs. 29 [25,32] mg/dL, P<0.05). After PSM, the baseline characteristics of the two groups were well balanced, with 391 patients in each group. No statistically significant differences were observed in age, sex distribution, co-morbidities, mean blood pressure (MBP), serum albumin levels, or SAPS II scores between the matched groups, indicating effective covariate balance (Table 1).

Table 1.

Comparison of baseline characteristics between the albumin infusion group and the non-albumin infusion group (Sun Yat-sen Memorial Hospital, Sun Yat-sen University)

Characteristics	Before PSM				After PSM
	Albumin infusion (n=395)	Non-albumin infusion (n=838)	P-value		Albumin infusion (n=391)	Non-albumin infusion (n=391)	P-value
Population
Age, years	67.0 (57.0, 76.0)	64.0 (53.0, 73.0)	<0.010		67.0 (57.0, 75.0)	67.0 (56.0, 76.0)	0.921
Gender, male, n (%)	255 (64.6)	534 (63.7)	0.776		253 (64.7)	252 (64.5)	0.940
Co-morbidities, n (%)
Cardiovascular diseases	277 (70.1)	545 (65.0)	0.077		273 (69.8)	275 (70.3)	0.876
Respiratory diseases	270 (68.4)	556 (66.3)	0.485		268 (68.5)	267 (68.3)	0.939
Gastrointestinal diseases	227 (57.5)	480 (57.3)	0.950		225 (57.5)	236 (60.4)	0.424
Neurological diseases	66 (16.7)	106 (12.6)	0.055		65 (16.6)	51 (13.0)	0.159
Rheumatic disease	3 (0.8)	3 (0.4)	0.344		3 (0.8)	2 (0.5)	0.654
Endocrine diseases	232 (58.7)	468 (55.8)	0.340		230 (58.8)	225 (57.5)	0.717
Scoring system
SAPS-II score	16.0 (11.0, 22.0)	13.0 (6.0, 21.0)	<0.010		16.0 (11.0, 22.0)	15.0 (10.0, 22.0)	0.531
Vital signs and laboratory indicators
MBP, mmHg	89.7 (77.3, 103.3)	88.3 (74.3, 99.3)	0.071		89.0 (77.3, 103.0)	90.0 (77.3, 103.3)	0.466
Albumin, mg/L	28 (25, 31)	29 (25, 32)	<0.010		28 (25, 31)	28 (25, 32)	0.232

Continuous variables were assessed for normality using the Kolmogorov-Smirnov test. Variables not conforming to a normal distribution were expressed as median (IQR), with comparisons made using the Wilcoxon rank-sum test. Categorical variables were expressed as frequency (percentage), and comparisons were performed using the χ² test, SAPS-II: simplified Acute Physiology Score II; MBP: mean blood pressure; PSM: propensity score matching.

Correlation analysis

Multivariate analysis of albumin infusion vs. non-albumin infusion and 28-day mortality

Multivariate logistic regression analysis (Table 2) demonstrated that age, liver disease, cerebrovascular disease, and the SAPS II score were independent risk factors for 28-day mortality (P<0.05). Comorbidities such as heart failure, myocardial infarction, and renal disease were not significantly associated with mortality (P>0.05). After PSM, age, liver disease, and the SAPS II score remained independent predictors of 28-day mortality. Albumin infusion was not significantly associated with 28-day survival (adjusted OR=0.979; 95% CI: 0.811-1.182; P=0.826).

Table 2.

Univariate and multivariate Logistic regression analyses of 28-day survival outcomes of patients with septic shock after infusion of human albumin (MIMIC-IV database)

Characteristic	Before PSM					After PSM
	Univariate		Multivariate			Univariate			Multivariate
	OR (95% CI)	P	OR (95% CI)	P		OR (95% CI)	P		OR (95% CI)	P
Age	1.022 (1.018-1.027)	<0.01	1.016 (1.010-1.022)	<0.01		1.029 (1.023-1.035)	<0.01		1.020 (1.013-1.026)	<0.01
Gender	1.036 (0.903-1.189)	0.612				0.984 (0.835-1.160)	0.847
Obesity	0.857 (0.699-1.050)	0.136				0.945 (0.750-1.192)	0.635
SAPS-II score	1.061 (1.056-1.067)	<0.01	1.056 (1.050-1.062)	<0.01		1.055 (1.048-1.061)	<0.01		1.048 (1.042-1.055)	<0.01
Myocardial infarct	1.296 (1.084-1.549)	<0.01	1.079 (0.884-1.318)	0.455		1.255 (1.003-1.570)	0.047
Congestive heart failure	1.144 (0.985-1.329)	0.078				1.236 (1.027-1.487)	0.025
Chronic pulmonary disease	1.005 (0.856-1.179)	0.955				0.993 (0.817-1.206)	0.942
Liver disease	1.916 (1.640-2.239)	<0.01	2.032 (1.682-2.454)	<0.01		1.775 (1.471-2.094)	<0.01		2.018 (1.635-2.490)	<0.01
Renal disease	1.415 (1.207-1.659)	<0.01	1.061 (0.887-1.270)	0.516		1.483 (1.218-1.806)	<0.01		1.063 (0.854-1.322)	0.585
Cerebrovascular disease	1.252 (1.005-1.559)	0.045	1.277 (1.002-1.627)	0.048		1.196 (0.919-1.557)	0.183
Diabetes	1.006 (0.868-1.166)	0.932				1.001 (0.837-1.197)	0.994
Rheumatic disease	0.933 (0.667-1.305)	0.686				1.097 (0.738-1.630)	0.648
Peptic ulcer disease	1.036 (0.760-1.414)	0.821				1.067 (0.753-1.512)	0.716
Albumin infusion	1.439 (1.248-1.660)	<0.01	1.178 (0.995-1.394)	0.057		1.110 (0.943-1.307)	0.211		0.979 (0.811-1.182)	0.826

OR: odds ratio; CI: confidence interval; SAPS-II: simplified Acute Physiology Score II.

Multivariate analysis of early vs. late albumin infusion and 28-day mortality

As shown in Table 3, age, liver disease, infusion time, and the SAPS II score were identified as independent risk factors for 28-day mortality (P<0.05). Other comorbidities, including chronic pulmonary disease, heart failure, and myocardial infarction, were not significantly associated with mortality (P>0.05). After PSM, age, liver disease, and the SAPS II score remained independent predictors of 28-day mortality. In addition, early albumin infusion was independently associated with increased 28-day mortality compared with late infusion (adjusted OR=1.556; 95% CI: 1.193-2.030; P<0.01). Using the effect estimate obtained from the matched cohort (OR=1.56, 95% CI: 1.193-2.030), the calculated E-value was 2.49 for the point estimate and 1.67 for its lower bound. These E-values indicate that only a relatively strong unmeasured confounder could nullify the observed association.

Table 3.

Univariate and multivariate Logistic regression analyses of early infusion of human albumin for 28-day survival outcomes in patients with septic shock (MIMIC-IV database)

Characteristic	Before PSM						After PSM
	Univariate			Multivariate			Univariate			Multivariate
	OR (95% CI)	P		OR (95% CI)	P		OR (95% CI)	P		OR (95% CI)	P
Age	1.022 (1.014-1.030)	<0.01		1.024 (1.015-1.034)	<0.01		1.019 (1.010-1.028)	<0.01		1.025 (1.015-1.036)	<0.01
Gender	0.853 (0.680-1.070)	0.169					0.834 (0.646-1.075)	0.161
Obesity	0.905 (0.661-1.240)	0.534					0.864 (0.602-1.240)	0.427
SAPS-II score	1.038 (1.029-1.046)	<0.01		1.031 (1.023-1.040)	<0.01		1.039 (1.029-1.048)	<0.01		1.034 (1.024-1.044)	<0.01
Myocardial infarct	0.964 (0.693-1.340)	0.826					0.816 (0.556-1.196)	0.296
Congestive heart failure	0.980 (0.754-1.275)	0.883					1.002 (0.743-1.350)	0.992
Chronic pulmonary disease	1.032 (0.789-1.349)	0.820					1.071 (0.794-1.446)	0.653
Liver disease	2.017 (1.602-2.539)	<0.01		2.659 (2.045-3.457)	<0.01		2.018 (1.558-2.616)	<0.01		2.895 (2.164-3.873)	<0.01
Renal disease	1.310 (0.986-1.740)	0.063					1.049 (0.763-1.442)	0.770
Cerebrovascular disease	1.058 (0.717-1.559)	0.777					0.948 (0.610-1.472)	0.811
Diabetes	0.874 (0.680-1.123)	0.292					0.769 (0.580-1.019)	0.068
Rheumatic disease	1.410 (0.808-2.462)	0.227					1.207 (0.655-2.226)	0.546
Peptic ulcer disease	0.979 (0.640-1.498)	0.923					1.191 (0.730-1.945)	0.484
Infusion time	1.618 (1.288-2.032)	<0.01		1.536 (1.204-1.960)	<0.01		1.522 (1.179-1.964)	<0.01		1.556 (1.193-2.030)	<0.01

¹ OR: odds ratio; CI: confidence interval; SAPS-II: simplified Acute Physiology Score II.

All regression analyses in this section were conducted based on the MIMIC-IV database. Owing to the lack of comparable external datasets with detailed infusion records and standardized clinical variables, no additional cohort analysis was performed.

Survival analysis

Effect of albumin infusion on 28-day survival in septic shock patients

Kaplan-Meier survival analysis of the MIMIC-IV cohort demonstrated that before PSM, 28-day mortality was significantly higher in the albumin infusion group than in the non-infusion group, with a hazard ratio (HR) of 1.229 (95% CI: 1.099-1.375; P<0.01) (Figure 2A). However, after PSM, no significant difference in 28-day survival was observed between the two groups, as indicated by the log-rank test (P=0.737), with an HR of 0.979 (95% CI: 0.865-1.108) (Figure 2B). These findings suggest that albumin infusion was not associated with improved 28-day survival in patients with septic shock.

Figure 2. The cumulative 28-day survival rates of septic shock patients categorized by albumin infusion status are shown before and after propensity score matching (PSM). Panels A (before PSM) and B (after PSM) present Kaplan-Meier survival curves from the MIMIC-IV cohort, while panels C (before PSM) and D (after PSM) show corresponding results from the Sun Yat-sen Memorial Hospital cohort. In each dataset, survival curves are stratified by albumin infusion status (albumin infusion and non-albumin infusion).

To validate these findings, a parallel analysis was conducted in an independent internal cohort from Sun Yat-sen Memorial Hospital. In this retrospective cohort, a similar trend was observed. Before PSM, patients who received an albumin infusion showed a numerically higher 28-day mortality compared to those who did not, with an HR of 1.510 (95% CI: 1.199-1.902; P<0.01) (Figure 2C). After PSM, the Kaplan-Meier curves between the two groups remained similar, although the difference was statistically significant difference persisted (HR: 1.367, 95% CI: 1.065-1.755; P=0.014) (Figure 2D). These results are broadly consistent with those of the MIMIC-IV cohort and further support the notion that albumin infusion may not confer a survival benefit in real-world septic shock patients.

Effect of early albumin infusion on 28-day survival in septic shock patients

Kaplan-Meier survival analysis in the MIMIC-IV cohort revealed significant differences in 28-day survival among patients who received no albumin, early albumin infusion, or late albumin infusion (P<0.05) (Figure 3A). The log-rank test further confirmed that these differences were statistically significant (P<0.05). Compared with the non-infusion group, the early infusion group was associated with a significantly higher hazard of death (HR=1.598; 95% CI: 1.367-1.869; P<0.05). Early infusion was associated with a higher hazard compared to late infusion (HR=1.645; 95% CI: 1.373-1.972; P<0.05). These results indicate that early albumin infusion may be associated with worse 28-day survival in patients with septic shock.

Figure 3. Kaplan-Meier survival curves illustrating the 28-day cumulative survival rates of septic shock patients stratified by the albumin infusion time: no infusion, early infusion, and late infusion. Panel A presents the results from the MIMIC-IV cohort, and panel B shows the corresponding results from the Sun Yat-sen Memorial Hospital cohort.

To validate these results, a parallel analysis was conducted in an independent internal cohort from Sun Yat-sen Memorial Hospital. Consistently, Kaplan-Meier survival analysis revealed significant differences in 28-day survival among septic shock patients in the non-infusion, early infusion, and late infusion groups (P<0.01; Figure 3B). Compared with the non-infusion group, early albumin infusion significantly increased the risk of mortality (HR=1.534, 95% CI: 1.194-1.972; P<0.01). However, there was no significant difference in mortality between the late infusion and non-infusion groups (HR=0.882, 95% CI: 0.580-1.342; P>0.05). Additionally, the hazard ratio between the early and late infusion groups was not significantly different (HR=1.353; 95% CI: 0.853-2.147; P>0.05). These results suggest that early albumin infusion may adversely impact 28-day survival in septic shock patients.

Subgroup analysis

Subgroup analysis of 28-day survival by baseline serum albumin levels in septic shock patients

In the MIMIC-IV cohort, the association between albumin infusion and 28-day mortality across subgroups stratified by serum albumin levels is shown in Figure 4A. In patients with severe hypoalbuminemia (serum albumin concentration <20 g/L), Kaplan-Meier analysis demonstrated that a significantly higher 28-day survival rate in the albumin infusion group compared to the non-infusion group (P<0.01; Figure 4A). The log-rank test confirmed this difference (P<0.01), with HR 0.553 (95% CI: 0.410-0.745; P<0.01), indicating that albumin infusion was associated with higher 28-day survival in this subgroup. In contrast, among patients with mild to moderate hypoalbuminemia (serum albumin concentration 20-35 g/L), Kaplan-Meier analysis revealed a significant difference in 28-day survival between the albumin and nonalbumin groups (Figure 4B). The log-rank test confirmed these results, with an HR of 1.295 (95% CI: 1.145-1.464; P<0.01), suggesting that the use of an albumin infusion was independently associated with decreased 28-day survival in this subgroup.

Figure 4. Kaplan-Meier survival curves showing 28-day cumulative survival rates in septic shock patients stratified by baseline serum albumin levels. Panels A and B present subgroup analyses from the MIMIC-IV cohort: panel A shows patients with severe hypoalbuminaemia (serum albumin concentration < 20 g/L), and panel B shows those with mild-to-moderate hypoalbuminaemia (20-35 g/L), comparing albumin infusion versus non-infusion. Panels C and D display corresponding subgroup analyses from the Sun Yat-sen Memorial Hospital cohort for the same albumin strata.

These findings were partially validated in the internal cohort from Sun Yat-sen Memorial Hospital. Among patients with severe hypoalbuminemia Kaplan-Meier survival analysis indicated no statistically significant difference in 28-day survival between the albumin infusion group and non-infusion groups (P>0.05; Figure 4C). In contrast, among patients with mild to moderate hypoalbuminemia, Kaplan-Meier analysis demonstrated significantly different survival outcomes between the two groups (P<0.01; Figure 4D). Albumin infusion in this subgroup was associated with an increased hazard of death (HR=1.499; 95% CI: 1.178-1.907; P<0.01), suggesting that albumin infusion may be associated with poorer survival among patients with mild to moderate hypoalbuminemia.

DISCUSSION

This study constitutes the first systematic investigation of individualized albumin infusion strategies in patients with septic shock, aiming to evaluate their impact on clinical outcomes. Utilizing data from the MIMIC-IV database and a university-affiliated hospital ICU, we analyzed the demographic and clinical characteristics of ICU patients with septic shock, with a particular focus on differences in albumin infusion timing and their association with 28-day mortality. Our findings indicate that variations in albumin infusion strategies significantly affect clinical outcomes, underscoring the need for tailored infusion regimens based on individual patient characteristics. We observed significant differences in baseline characteristics between patients who received albumin and those who did not. Specifically, compared with the non-infusion group, the albumin infusion group exhibited higher illness severity, as indicated by significantly elevated SAPS-II, OASIS, and LODS scores, with a median SAPS-II score of 50.0 compared to 46.0 in the non-infusion group. This observation implies that clinicians tend to administer albumin to more critically ill patients, possibly to improve outcomes, though the effectiveness of this clinical practice requires further validation through prospective studies. This discrepancy likely reflects more conservative fluid management strategies in patients with preexisting cardiovascular or neurologic conditions, as clinicians may delay albumin infusion to prevent potential complications such as fluid overload or vascular leakage. Additionally, early infusion was associated with marginally higher SAPS-II scores, suggesting greater clinical severity in the early infusion group.^[47-51] While this difference was modest, it may have influenced clinicians’ decisions regarding earlier initiation of albumin infusion to rapidly achieve hemodynamic stabilization.^[52,53] In real-world settings, such differences in baseline characteristics are common. To address potential confounding factors, we applied PSM, which effectively balanced key variables such as age, sex, baseline serum albumin levels, and severity scores between groups. This adjustment enhances the validity of our conclusions regarding the associations between the albumin infusion strategy and patient outcomes. This study specifically focused on the timing and clinical context of albumin infusion rather than dose or concentration, as these parameters vary widely in real-world ICU practice and are often protocol-driven in randomized trials. While previous studies such as ALBIOS achieved physiologic effects through standardized 20% albumin dosing to maintain serum levels ≥30 g/L, our analysis aimed to explore how the timing of albumin initiation relates to outcomes under routine care conditions.^[27] The higher mortality in the early infusion group may partly reflect confounding by indication, as clinicians tend to administer albumin earlier to patients with greater circulatory compromise. After PSM, illness severity was largely balanced, and subgroup and external validation analyses supported the robustness of our findings. Rather than suggesting harm, these results likely reflect treatment selection and potential benefit among patients with severe hypoalbuminemia, consistent with the ALBIOS trial, which showed improved hemodynamics but no overall survival benefit.

The severity of hypoalbuminemia may be a more meaningful determinant of outcome than infusion timing is. Patients with albumin levels <20 g/L derived measurable benefits from albumin therapy, consistent with evidence linking profound hypoalbuminemia to endothelial and microcirculatory injury.^[54] In our cohort, this pattern was further supported by subgroup analyses showing that albumin infusion improved 28-day survival specifically in patients with severe hypoalbuminemia, whereas no significant benefit was observed among those with mild or moderate hypoalbuminemia. These results indicate that the biological response to albumin replacement depends more on the extent of hypoalbuminemia and its correction than on the precise timing of administration. Previous studies have established that the severity of hypoalbuminemia is closely associated with poor clinical prognosis.^[55-57] About 30%-50% of critically ill patients have serum albumin concentrations less than 35 g/L, which is correlated with increased mortality.^[58-60] When serum albumin levels fall below 20 g/L, the risk of mortality increases significantly.^[61,62] Our findings support the notion that timely correction of profound hypoalbuminemia may attenuate disease progression and improve survival. These beneficial effects may be mediated through albumin’s multiple physiological functions. In addition to maintaining oncotic pressure and intravascular volume, albumin possesses anti-inflammatory and antioxidant properties and plays a role in modulating nitric oxide metabolism.^[63-66] Moreover, albumin binds to various endogenous and exogenous substances, thereby influencing drug pharmacokinetics.^[67-70] Under hypoalbuminemic conditions, elevated free drug fractions may increase toxicity, particularly for medications with high protein-binding capacity, such as warfarin, digoxin, and certain NSAIDs or antibiotics.^[71,72] Thus, the correction of hypoalbuminemia may not only stabilize hemodynamic parameters but also optimize pharmacologic interventions, thereby contributing to improved clinical outcomes. Beyond these systemic effects, albumin also supports endothelial glycocalyx preservation and microvascular integrity, reducing capillary leakage and oxidative stress. Experimental and clinical evidence has indicated that albumin replacement can restore microcirculatory perfusion and mitigate endothelial injury in sepsis patients, suggesting biological plausibility for the observed benefits among patients with severe hypoalbuminemia.^[73] However, our analysis was limited to baseline albumin measurements, as serial data were not consistently available. Dynamic albumin trajectories (Δalbumin within 24-72 h) could better reflect treatment response and physiologic correction, and future prospective studies should incorporate such measures to capture these time-dependent effects more accurately.

Conversely, we did not observe statistically significant survival benefits of albumin infusion in patients with septic shock overall. Notably, early albumin infusion was associated with worse clinical outcomes compared to late infusion. These findings are consistent with those of previous large-scale trials, such as those comparing colloids and crystalloids for volume resuscitation, which reported no significant mortality benefit associated with albumin infusion.^[74] Moreover, although albumin may reduce vasopressor requirements, its impact on survival remains controversial.^[27] In the context of septic shock, early albumin infusion may be potentially counterproductive. The early phase of sepsis is characterized by severe systemic inflammation, damage to the vascular endothelium, and elevated capillary permeability.^[75-77] Under these conditions, infused albumin may rapidly extravasate, leading to interstitial edema and elevated cardiac preload, thereby increasing the risk of pulmonary edema and subsequent organ dysfunction.^[21,78,79] Our findings suggest that albumin may be more effective when it is infused at a later stage after partial recovery of endothelial integrity and capillary permeability, thereby facilitating improved intravascular retention and efficacy. The potential mechanism underlying the improved outcomes from late infusion may be related to the partial restoration of vascular endothelial function following the peak inflammatory response in sepsis. During this period, the colloid osmotic effects of albumin can more effectively maintain intravascular volume stability and reduce the risk of interstitial fluid retention. This finding aligns with existing hypotheses suggesting that colloid administration in the later stages of sepsis may be safer and more effective.^[80,81] Given the high cost and limited availability of albumin, determining the optimal timing for its infusion is of considerable clinical and economic importance. Our study highlights the necessity of individualized treatment strategies based on illness severity and pathophysiologic stage rather than a uniform approach to early hypoalbuminemia correction.

This study has several limitations. This study focused primarily on the role and timing of albumin administration in correcting hypoalbuminemia rather than on the full spectrum of sepsis management. Information on concurrent treatments—such as crystalloids, corticosteroids, and renal replacement therapy—was incomplete and could not be systematically adjusted for, which may have contributed to residual confounding.

Although our study benefits from its large sample size and access to high-quality data, several limitations must be acknowledged. This study combined data from the MIMIC-IV (2008-2022) and a single-center ICU cohort (2021-2024) to increase generalizability. However, temporal heterogeneity may exist, as the MIMIC-IV reflects the Sepsis-2 and early Sepsis-3 eras,^[6,82] whereas the institutional cohort was defined entirely by Sepsis-3 criteria. Although both capture infection-associated circulatory failure, evolving sepsis management—such as earlier diagnosis, refined hemodynamic targets, and more restrictive fluid resuscitation—may bias the estimated association toward the null rather than exaggerate it. Future studies should incorporate standardized diagnostic frameworks and temporal correction to better account for these secular trends. Adjustment for calendar year and treatment center was not feasible because of anonymized time fields in the MIMIC-IV. Future multicenter datasets with temporal identifiers could allow adjustment for secular trends and evolving sepsis management protocols.

The retrospective design and dependence on electronic health record data may have introduced unmeasured confounding, even after adjustment using PSM. Despite effectively controlling baseline characteristics through PSM, residual confounding factors remain, including dynamic changes in SAPS-II scores, fluid balance, and the use of other vasoactive or immunomodulatory therapies, which were not measured. The dichotomization of albumin infusion timing (<72 h vs. ≥72 h) followed established critical care frameworks but may introduce immortal time bias, as patients in the late group must have survived the first 72 h to receive albumin. In our cohort, early mortality was rare, and survival curves overlapped before 72 h, suggesting minimal practical impact. Although PSM reduced the baseline imbalance, time-dependent factors could not be fully adjusted. Future studies using target trial emulation or marginal structural models are warranted to address these dynamic effects. Future studies should consider prospective designs or other robust methodologies to further validate these findings and comprehensively address potential residual confounding. Another limitation is that the dose, concentration, and cumulative exposure to albumin could not be fully standardized because of variability in clinical practice and incomplete dosing records in the database. However, the initial weight-adjusted albumin dose did not differ significantly between the early and late infusion groups (0.31 g/kg in both groups before and after matching; supplementary Table 2). Nonetheless, dose-response relationships should be evaluated in future prospective studies to rule out potential confounding effects arising from minor dose difference.^[83-84]

CONCLUSION

In conclusion, early albumin infusion in patients with septic shock does not provide a clear survival benefit. In contrast, late infusion may be associated with improved short-term outcomes. Survival differences were observed mainly among patients with severe hypoalbuminemia, in whom early albumin infusion was associated with higher survival rates. In contrast, patients with mild-to-moderate hypoalbuminemia showed no clear difference among different albumin infusion strategies. These findings suggest that albumin administration may be most relevant for patients with profound hypoalbuminemia. A careful evaluation of baseline serum albumin levels and the timing of infusion is crucial when considering albumin therapy in patients with septic shock.

Funding: This research was sponsored by the National Natural Science Foundation of China (82372207) and Guangzhou Key Laboratory of Neuromodulation and Regenerative Medicine Rehabilitation (2024A03J0699).

Ethical approval: This study received ethics approval from an affiliate of the Massachusetts Institute of Technology (11628356). This research was also approved by the ethics committee of Sun Yat-sen Memorial Hospital, Sun Yat-sen University (LX-HG-2024020). Written informed consent was not needed in accordance with national legislation and institutional guidelines.

Conflicts of interest: The authors do not have a financial interest or relationship to disclose regarding this paper.

Contributors: FRL and TH contributed equally to the study. FRL and TH contributed to the conception and design of the study and participated in refining the research questions. FRL conducted data collection, curated the dataset, and performed the statistical analysis with JZZ, who also contributed to data preprocessing and quality control. FRL drafted the initial manuscript. SQL and ZFY analyzed and interpreted the results, contributed to the discussion of the findings, and assisted in manuscript revision. All the authors participated in the literature review, critically revised the manuscript for important intellectual content, approved the final version for submission, and agreed to be accountable for all the aspects of the work. ZFY is the guarantor of the work and takes responsibility for the integrity of the work as a whole, from inception to the published article.

All the supplementary files in this paper are available at http://wjem.com.cn.