Impact of Colloid Infusion on Outcomes in Patients with Septic Shock

Background: Although colloid solution has been widely used in practice, its impact on mortality in patients with septic shock remains unknown. We evaluated the association of colloid infusion with outcomes in septic shock patients. Methods: Medical Information Mart for Intensive Care (MIMIC)-III was used to identify patients with septic shock. Propensity score matching (PSM) was employed to balance the baseline differences. Cox proportional hazards model, Wilcoxon rank-sum test, and logistic regression were utilized to determine the associations of colloid infusion with mortality, length of stay, and recovery of kidney function, respectively. Results: A total of 4,553 septic shock patients were studied, including 1,158 with colloid infusion, and 3,395 without colloid infusion. After PSM, 1,012 pairs of patients were matched. Signicant benets in the mortality rate were observed in the colloid group compared with the non-colloid group, with the 28-day mortality [hazard ratio (HR) 0.62; 95% condence interval [CI], 0.52-0.73; P < 0.001], and the 90-day mortality [HR 0.76; 95% CI 0.65-0.88; P < 0.001]. Colloid infusion was not associated with the renal function recovery [HR 1.06; 95% CI 0.87–1.29; P = 0.547] in either population. Nevertheless, subgroup analysis revealed that colloid infusion did not affect the 28-day mortality in people with sepsis of AKI stage 1. In addition, the use of dextran did not decrease the 28-day mortality (HR 1.41; 95% CI 0.19-10.59; P = 0.736). Conclusion: In septic shock patients, colloid infusion (albumin or hydroxyethyl starch) improved short-term survival, but had no clear effect on the recovery of renal function. Abbreviations: IQR interquartile range, ICU intensive care unit, HR hazard ratio, CI condence interval a Cox proportional hazard models were used to assess the impact of colloid infusion on mortality outcomes adjusting for confounders with a P-value < 0.05 in univariate analysis. b Recovery of renal function was dened as being discharged from ICU with serum creatinine below 1.5 times the baseline value and normal urine output (> 0.5 ml/kg/h). Impact of colloid infusion on the recovery of renal function was assessed using the logistic regression model adjusting for age, SOFA score, SAPSII score, and RRT use. c Wilcoxon rank sum test was used to assess the association between colloid infusion and length of stay.


Introduction
Septic shock syndrome resulting from systemic in ammation and excessive host immune responses to infection is a top cause of death in patients in the hospitals, with 40-50% mortality [1,2]. For decades, natural or synthetic colloids have been given to patients with shock to maintain su cient colloid-osmotic pressure (COP) and the volume of vessels [3][4][5]. However, a recent large meta-analysis indicated that the administration of colloid made no difference to mortality in critically ill patients, as compared with crystalloid solutions [6]. Meanwhile, some studies reported contradicting results [7][8][9][10].
hydroxyethyl starch (HES) or dextran] solution [11]. Colloids have larger molecules, cost more, and may promote faster expansion in the vessel volume, but it may result in coagulation dysfunction, allergic reactions, or renal failure [6]. Whether colloid infusion can impact the clinical outcomes in the early resuscitation of septic shock is not clear [9,12,13]. Currently, no clear colloid infusion strategies for patients with sepsis and septic shock are recommended in the national guidelines [14]. Therefore, we carried out a retrospective observational study to determine the impact of colloids (albumin, HES, or dextran) on outcomes in septic shock patients.

Data Source
The data utilized in this retrospective study was from MIMIC-III, which is an openly accessible US-based critical care repository [15]. The MIMIC III database includes clinical information on patients hospitalized from 2001 to 2012 in the adult ICUs of Beth Israel Deaconess Medical Center. It is also approved by the Massachusetts Institute of Technology Institutional Review Boards. Patients were selected using the PostgreSQL 9.6 software from the latest version (MIMIC-III v1.4), which was released on the 2nd of September 2016. After successfully completing the course of the Collaborative Institutional Training Initiative Program (Record ID 35897056), we were allowed to utilize the stored clinical data from the MIMIC-III repository.

Participants and de nition
PgAdmin (version 4.1, Bedford, USA) was used to mine data from the MIMIC III data bank. The study inclusion criteria included: (1) Patients that were diagnosed with sepsis. According to the Sepsis-3 de nition [16], sepsis was de ned as patients with suspected or veri ed infection, plus the SOFA scores ≥2 during the rst day after ICU admission [16]. Infection was diagnosed according to the method established by Angus in the MIMIC-III database; (2) All sepsis patients who were supported with vasopressor within 24 hours after ICU admission were de ned as patients with septic shock. For a patient with more than one ICU admissions, only the rst admission was included. According to the colloid infusion status within 48 hours after ICU admission, the participants were separated into two groups: A Colloid group (intervention) and a Non-colloid group (control).
Baseline data were obtained from the MIMICIII. The rst laboratory data since ICU admission was recorded. Sequential organ failure assessment (SOFA) score, Simpli ed Acute Physiology Score II (SAPSII), and Glasgow coma score (GCS) were calculated as described in previous studies [17,18]. AKI was de ned according to the Kidney Disease: Improving Global Outcomes (KDIGO) criteria [19]. The use of crystalloid and colloid solution during the rst 48 h of ICU admission was included in this analysis.
No more than 25% missing values were recorded in all variables (Supplementary Table S1).
Missing values were lled by single imputation or linear regression as appropriate.

Endpoints
The primary end point was 28-day mortality. The secondary end points are 90-day mortality, recovery of renal function, length of stay (LOS) in hospital, and LOS in ICU. Recovery of renal function was de ned if the urine output on discharge is normal (> 0.5 ml/kg/h for 24 h) and a return to a creatinine level of 150% as of the baseline on ICU discharge.

Statistics analysis
Continuous variables are expressed as median [interquartile range (IQR)] due to their non-normal distribution. The differences between groups were determined by the Mann-Whitney U test. Categorical variables are shown as frequencies and percentages. The comparisons were performed by the χ 2 test or Fisher's exact test as appropriate.
To balance the baseline differences, propensity-score matching (PSM) was conducted with a caliper width of 0.2 logits of the standard difference. Patients were divided using 1:1 nearest neighbor matching, so that each person in the Colloid group was matched with those in the Non-colloid group. The standardized mean difference (SMD) was used to assess the effectiveness of PSM to decrease the baseline differences [20]( Supplementary Figure S1). The Cox regression model was performed to assess the relationship between colloid infusion and mortality after adjustment for confounding variables with p < 0.05 in univariate analysis (Supplementary Table S2). The logistic regression model was used to assess the impact of colloid infusion on the recovery of renal function after adjusting for age, SOFA score, SAPSII score, and RRT use. LOS in hospital and LOS in ICU were compared by Wilcoxon rank-sum test between two groups. Survival analyses were censored at day 28 and day 90 after PSM.
Various subgroups were classi ed by different age, lactate, AKI stage, and cardiovascular disease. The association between the daily dose or type of colloid and 28-day mortality was also assessed.
Multivariate analysis by Cox regression was used in subgroup analyses after adjusting for potential confounders, which were performed after PSM.
Statistical analysis was carried out using software Stata 15.1 and R 4.0.0 in the Windows operative system. Statistical signi cance was determined when the p value is less than 0.05.

Basic characteristics
A total of 46,520 patients with the rst ICU admission were extracted from the MIMIC-III database. A total of 12,884 patients tted the de nition of sepsis-3.0 within 48 h after ICU admission. A total of 4,553 patients were examined in the analysis based on the exclusion criteria. Of the study cohort, 1,158 patients were administrated with colloid infusion in the rst 48 h in the ICU, while the remaining 3,395 patients did not receive colloid infusion (Fig. 1).
The baseline characteristics of the two groups before PSM are presented in Table 1. The mean age was slightly lower (68 vs. 70), and the weight, SOFA score, SAPSII score was higher in the Colloid group at admission, in comparison with the Non-colloid group. Patients with AKI, hypertension, coagulopathy, or obesity were more likely to be given a colloid solution. The levels of bilirubin, PT, and PH were lower, while the levels of platelet, glucose, hemoglobin, and lactate were statistically higher in the Non-colloid set compared to the Colloid set. The uses of RRT and mechanical ventilation were much more common in the colloid group. The volume of daily crystalloid infusion was more and the urine output was less in the Colloid group.

Relationship between colloid infusion and outcomes
After PSM, 1012 patients who did not receive colloid infusion were matched with 1012 patients who received colloid infusion. The imbalance between the two groups was signi cantly reduced after PSM, and SMDs of all variables were less than 10% (Table 2 and Figure S1). Abbreviations: SOFA sequential organ failure assessment, SAPSII simpli ed acute physiology score II, GCS Glasgow coma score, MAP mean arterial pressure, ARDS acute respiratory distress syndrome, RRT renal replacement therapy, PT prothrombin time, WBC white blood cell, SMD standardized mean difference.
a The initial value during the rst 24h after ICU admission.
b The values were calculated during the rst 24h after ICU admission.
In the pre-matched cohort, colloid infusion was related to improved mortality at 28 days (HR 0.65; 95% CI 0.57-0.76; P < 0.001), and at 90 days (HR 0.78; 95% CI 0.69-0.88; P < 0.001), following adjustment of the confounders with P < 0.05 in univariate analysis (Table 3 and Supplementary Table S2). The impact of colloid infusion on the renal function recovery was assessed by the logistic regression model, and the result showed that colloid infusion was associated with delayed recovery of renal function (HR 0.83; 95% CI 0.71-0.96; P = 0.015). In addition, we found that colloid infusion was associated with extended LOS in ICU and hospital (Table 2). Abbreviations: IQR interquartile range, ICU intensive care unit, HR hazard ratio, CI con dence interval a Cox proportional hazard models were used to assess the impact of colloid infusion on mortality outcomes adjusting for confounders with a P-value < 0.05 in univariate analysis.
b Recovery of renal function was de ned as being discharged from ICU with serum creatinine below 1.5 times the baseline value and normal urine output (> 0.5 ml/kg/h). Impact of colloid infusion on the recovery of renal function was assessed using the logistic regression model adjusting for age, SOFA score, SAPSII score, and RRT use.
c Wilcoxon rank sum test was used to assess the association between colloid infusion and length of stay.
Outcomes Abbreviations: IQR interquartile range, ICU intensive care unit, HR hazard ratio, CI con dence interval a Cox proportional hazard models were used to assess the impact of colloid infusion on mortality outcomes adjusting for confounders with a P-value < 0.05 in univariate analysis.
b Recovery of renal function was de ned as being discharged from ICU with serum creatinine below 1.5 times the baseline value and normal urine output (> 0.5 ml/kg/h). Impact of colloid infusion on the recovery of renal function was assessed using the logistic regression model adjusting for age, SOFA score, SAPSII score, and RRT use.
c Wilcoxon rank sum test was used to assess the association between colloid infusion and length of stay.
In the post-matched cohort, similar to the results before PSM, colloid infusion was associated with improved 28-day survival (HR 0.62; 95% CI 0.52-0.73; P < 0.001), and with reduced 90-day mortality (HR 0.76; 95% CI 0.65-0.88; P < 0.001) (Table 3). Nevertheless, the recovery of renal function was not statistically different between the Colloid group and the Non-colloid group (HR 1.06; 95% CI 0.87-1.29; P = 0.547) ( Table 3). The colloid infusion was also associated with longer LOS in ICU and hospital (Table 2). Kaplan-Meier's survival estimates of patients according to colloid infusion at ICU admission are shown in Fig. 2. The colloid infusion remained associated with improved the 28-day (P < 0.0001) and the 90-day (P = 0.0021) mortality.
Taking the dose and type of colloid into consideration, we found that colloid infusion did reduce the 28day mortality, independent of dose (  (Table S4).

Subgroup analysis
The relationship between colloid infusion and 28-day mortality in subgroups is shown in Fig. 3. Similarly, the study did manifest signi cant bene cial effects of colloid infusion on mortality at 28 days in patients with septic shock, regardless of age and lactate concentration. The bene cial effect of colloid infusion on mortality was also observed in people with cardiovascular disease or AKI. Colloid infusion may reduce the risk of death in patients with AKI stage 2 (HR 0.42; 95% CI 0.3-0.59; p < 0.001) and stage 3 (HR 0.63; 95% CI 0.49-0.81; p < 0.001) but not in patients with AKI stage 1 (HR 1.28; 95% CI 0.69-2.37; p = 0.437) (Fig. 3).

Discussion
Our results suggested that colloid infusion within 48 h after ICU admission lowered short-term mortality in people with septic shock, but it may lead to longer LOS in ICU and in hospital. The results of our study also showed that no signi cant correlation was found between colloid infusion and the recovery of renal function. However, subgroup analysis showed that no signi cant bene cial effect was observed in patients with AKI stage 1 and treated by colloid infusion. The use of albumin or HES was associated with decreased 28-day mortality, but no apparent survival advantage was observed in the dextran group.
Whether using colloid infusion to resuscitate patients with septic shock will improve patient-centered outcomes, and for whom bene ts will outweigh the risks remain debated. Other studies have reported the effectiveness of albumin in critically ill or sepsis patients [6,8,11,21,22], but only a few studies focused on the use of colloid in patients suffering from septic shock [4,12]. The prospective sequential analysis [4] showed that hydroxyethyl starch and gelatin may impair renal function. But the study did not assess the impact of colloid infusion on mortality. Our ndings may appear not to agree with the results of the trial from Pietro Caironi [12], which suggested that the use of albumin within the rst 28 days of treatment did not improve the survival rate compared with crystalloids alone, over a 90-day follow-up. However, Caironi's study had con rmed the physiological bene t of albumin administration, such as larger uid diffusion in the intravascular space. Besides, albumin may mediate peripheral vasodilation during sepsis as a scavenging agent for nitric oxide [23,24]. A meta-analysis found that albumin infusion contributed to decreased 90-day mortality in people with severe sepsis and septic shock [9].
Unfortunately, its impacts on renal function and stage of AKI were not conclusive in the study. Further studies are needed to evaluate the effects of colloid infusion on renal function and prognosis in different subgroups of AKI.
Our study found that colloid infusion had no effect on renal function recovery in septic shock people.
This nding may be different from the previous report by Bayer et al. [4]. However, it should be noted that septic shock itself can affect the recovery of renal function as well. In addition, our subgroup analysis showed that colloid infusion did not improve the 28-day survival in sepsis patients with AKI stage 1.
Therefore, it is suggested that colloid infusion is not necessary for the early stage of septic shock as the renal function of less severe patients is likely to be recovered by crystalloid solution infusion, antiinfection, and other related treatments.
However, there are still several limitations to our study. Firstly, the current research is based on a clinical electronic medical record that contains some missing values. To improve statistical power, multiple imputations were used to decrease the risk of deviation due to missing values. Secondly, we did not report the range of albumin concentrations in this study. The effect of the concentrations of albumin on mortality is unclear. Finally, due to the characteristics of a retrospective research, the relationship between the colloid infusion and mortality can only be interpreted indirectly, which may only provide preliminary evidence for further investigation.

Conclusion
In people with septic shock, the infusion of albumin or HES decreased short-term mortality, but dextran had no apparent survival advantage. In addition, colloid infusion had no effect on renal function recovery in patients with septic shock.
Abbreviations MIMIC Medical Information Mart for Intensive Care, PSM Propensity score matching, IQR interquartile range, ICU intensive care unit, HR hazard ratio, CI con dence interval, COP colloid-osmotic pressure, FFP fresh frozen plasma, HES hydroxyethyl starch, KDIGO Kidney Disease: Improving Global Outcomes, SOFA sequential organ failure assessment, LOS length of stay, SAPSII simpli ed acute physiology score II, GCS Glasgow coma score, MAP mean arterial pressure, ARDS acute respiratory distress syndrome, RRT renal replacement therapy, PT prothrombin time, WBC white blood cell, SMD standardized mean difference.

Declarations
Ethics approval and consent to participate MIMIC III database used in this study was approved by MIT's Institutional Review Board (IRB) and does not include protected health information. Thus, requirement for individual patient consent was waived because the study did not impact clinical care, and all protected health information was deidenti ed.

Consent for publication
Not applicable.

Availability of data and materials
The datasets used in the present study are available from the rst author and corresponding authors on reasonable request.

Competing interests
The authors declare that they have no competing interests.

Funding
The work was partly funded by the National Natural Science Foundation of China (81671960, 81974285).
Authors' contributions CG collected and analyzed data, and co-wrote the manuscript. QP and YJ was helpful for statistical analysis and interpretation of results. ZL and WL prepared the gures and tables. LZ and YA designed and supervised the study, and YA revised the manuscript. All authors have read and reviewed the nal manuscript.