The Infusion of Human Serum Albumin had no Benet on Acute Pancreatitis Therapy: An Analysis of Patients in Two Observational Cohorts

Objective: Human serum albumin (HSA) infusion is a common administration in acute pancreatitis patients in the Intensive Care Unit (ICU), but its actual association with patients' outcomes has not been conrmed. The study was aimed to determine whether the in-hospital prognosis of ICU patients with acute pancreatitis could benet from albumin infusion. Methods: 950 acute pancreatitis patients diagnosed in 2008-2019 were extracted from the MIMIC-IV database as our primary study cohort. The primary outcome was in-hospital mortality. We also performed an external validation with a cohort of 104 acute pancreatitis patients after PSM matching from the eICU database. Results: In MIMIC-IV, 228 acute pancreatitis patients received HSA infusion (Alb group) during their hospitalization, while 722 patients did not (non-Alb group). Patients in the Alb group presented a poorer survival curve than the non-Alb group, while this difference disappeared after PSM or IPTW matching (log-rank test: PSM: p = 0.660, IPTW: p = 0.760). After including covariates, no association was found between albumin infusion and patients' in-hospital mortality before and after matching (original cohort: HR: 1.00, 95% CI: 0.66–1.52, p = 0.998). HSA infusion also did not benet patients' 28-day or ICU mortality, while it signicantly prolonged their duration in hospital and ICU. In addition, the initial serum albumin levels, infections or the amount of total albumin infusion did not affect the conclusion. Finally, in the eICU cohort, albumin infusion was still not a benecial prognostic factor on patients' in-hospital mortality ( p = 0.087). Conclusion: Intravenous albumin infusion could not benet acute pancreatitis patients' in-hospital prognosis RRT mechanical ventilation, SOFA Sequential Organ Failure Assessment, SAPS II Simplied Acute Physiology Score II, CHF congestive heart failure, COPD chronic obstructive pulmonary disease, MAP mean arterial pressure, pO 2 partial pressure of oxygen, pCO 2 partial pressure of carbon dioxide, WBC white blood cell, Bun blood urea nitrogen


Introduction
As the most common gastrointestinal disease requiring emergency hospitalization, acute pancreatitis has an annual incidence of 34 cases per 100,000 in high-income countries [1,2]. And its incidence is generally considered to be positively correlated with the national sociodemographic index (SDI) [3]. As recent guidelines indicated, gallstones (45%) and alcohol abuse (20%) remain the critical factors in the pathogenesis of acute pancreatitis, which also contributes to the imbalance of incidence between different regions [1,4]. Acute pancreatitis is characterized by complex and variable symptoms and various prognoses between individuals. Mild cases showed only pancreatic edema, which was often self-limiting and had a good prognosis. In contrast, severe cases (20%) might result in pancreatic necrosis, peritonitis, shock, and systemic multiple organ failure, with 20%-40% mortality [1,5].
The hypoperfusion and intestinal bacterial translocation accompanying acute pancreatitis systemic in ammatory reactions could lead to irretrievably serious consequences [1,6]. Therefore, current guidelines clearly state that adequate uid resuscitation and nutritional support are essential strategies in the initial treatment of acute pancreatitis [6]. The infusion rate of uid resuscitation is recommended at 5-10 ml/kg per hour until the patient's vital signs meet the resuscitation criteria, including heart rate, mean arterial pressure, and urinary output [7]. However, few studies had investigated the type of resuscitation uid, though several guidelines recommended Ringer's lactate solution compared with normal saline [8][9][10].
Human serum albumin (HSA) has been widely used for volume expansion and correcting hypoalbuminemia in critical care for nearly 70 years worldwide [11]. However, clinical evidence for the recommendation of albumin infusion for uid resuscitation in critical ills remains weak [12], and its value in improving hypoalbuminemia is also controversial [13,14]. Recently, more and more studies have focused on the actual survival bene t value on speci c patient groups due to many inappropriate applications of albumin infusion in clinical practice and its reported negative in uence on patients' mortality [11,15]. Albumin administration is common among clinical practice in patients with acute pancreatitis, as in other critical ills. However, few studies have analyzed whether it has a bene cial impact on acute pancreatitis patients' outcomes.
Therefore, our study was aimed to determine the effect of human serum albumin infusion on multiple inhospital outcomes among patients diagnosed with acute pancreatitis through MIMIC-IV (v1.0), a large, retrospective, recently presented, single-centre critical care database with a variety of high-quality clinical data from hospital monitoring systems [16]. Our study cohort enrolled multifaceted clinical variables of acute pancreatitis patients with ICU (Intensive Care Units) admission to con rm the analysis result. Moreover, some of the potential factors that might affect results, as patients' initial serum albumin level during the rst 24 hours of ICU admission, the total infusion dose of albumin for each patient and bacterial culture results of patients' body uids, were all analyzed in our study. We also analyzed another acute pancreatitis cohort from the eICU database to increase the robustness of our investigation.

Data source description
Our data were extracted from MIMIC and eICU databases. The Medical Information Mart for Intensive Care (MIMIC) program is an extensive, single-centre and freely accessible clinical database hosted by the Laboratory for Computational Physiology at the Massachusetts Institute of Technology (MIT) [17,18].

Page 4/19
Adults patients diagnosed with acute pancreatitis from both databases were enrolled in the study. There were 3,753 acute pancreatitis patients in MIMIC-IV totally, and we screened 950 of them only with their rst admission to ICU. And the sample screening process is shown in Figure 1. MIMIC-IV had patients diagnosed in ICD-10 codes (International Classi cation of Diseases code, version 10), which was  different from MIMIC-III[16, 20,21], and the number of patients in each diagnosed title is provided in Table S1. All 563 patients with acute pancreatitis from eICU met the extraction criteria of MIMIC-IV.

Variable extraction and Outcomes
Patients were grouped based on whether they received intravenous albumin infusion during the hospitalization. The total dose (g) of infused albumin for each patient in MIMIC-IV was also recorded.
Other covariates within the rst 24 h after ICU admission included the following: age, gender, weight, admission period, Sequential Organ Failure Assessment (SOFA) score, Simpli ed Acute Physiology Score II (SAPS II) score, application of renal replacement therapy (RRT), mechanical ventilation (MV). The patient's vital signs were also extracted, including heart rate, mean arterial pressure (MAP), respiratory rate and temperature (°C). We also enrolled pH, partial pressure of oxygen (pO2), partial pressure of carbon dioxide (pCO2), lactate level, hemoglobin, platelet counts, white blood cell (WBC) count, serum albumin level, blood urea nitrogen (Bun) and creatinine as laboratory tests in the rst 24 h. The comorbidities we extracted as covariates included congestive heart failure (CHF), chronic obstructive pulmonary disease (COPD), liver disease, renal disease and malignancy. The primary outcome of our study was in-hospital mortality. Secondary outcomes included 28-day mortality, ICU mortality, hospital length of stay (days) and ICU length of stay (days).

Statistical analysis
Continuous variables were described by medians with interquartile ranges (IQRs) and compared by t-test or Wilcoxon rank-sum test between groups. And we used total number and percentage to present categorical variables and compared the proportions using χ2 or Fisher exact tests. As for survival analysis, we used patients' hospital length of stay (days) as follow-up time, hospital mortality as the primary endpoint. Kaplan-Meier (K-M) survival curve analysis was used to generate curves and the logrank test to determine statistical differences among groups. Multivariate Cox regression models were performed to determine whether albumin infusion affects patients' outcomes after the inclusion of varying clinical factors. Moreover, we also applied multivariate linear regression models to analyze hospital outcomes of continuous variables. The selection of covariates including in multivariate analysis considered the data loss rate and its clinical impact on prognosis at the same time. Finally, pH, pO2 and pCO2 were excluded for their highest rate of missing data, while all other covariates were enrolled in multivariate analysis. The amount and percentage of missing data for each covariate in MIMIC-IV are shown in Table S2. At the same time, multiple imputations were applied to mitigate the estimation bias caused by missing data and assuming that data were missing randomly in both MIMIC-IV and eICU. The absence rates of covariates included in multivariate analysis in eICU were all less than 20%.
Imbalanced covariates between treatment and control groups might make the results of multivariate analysis less accurate. Propensity score matching (PSM) and propensity score-based inverse probability of treatment weighting (IPTW) methods were applied to minimize the covariate differences between groups [22,23]. We matched patients in the treatment group to the control group as 1:1 nearest neighbour by estimating the patients' propensity scores for albumin infusion measurement in PSM [24]. Moreover, we created two virtual cohorts by weighting each patient through IPTW, which showed a similar distribution of covariates and different administration exposure [25]. We also calculated the standardized mean differences (SMD) before and after matching to test the effects of PSM and IPTW. Due to insu cient covariates, small sample size and signi cant differences in the number of patients between groups of the eICU cohort, We used PSM matched eICU cohort as an external validation of patients from MIMIC-IV. The baseline covariates and SMDs of patients from eICU are presented in Table S3 after being PSM matched.
All our patients' data from the database were extracted in SQL (Structured Query Language), and all statistical analyses were performed by Rstudio software (v3.6.3). Two-sided p < 0.05 was considered statistically signi cant.  (Fig. 2a). The Alb group tended to have worse in-hospital survival status compared with the non-Alb group. Then the independence of albumin infusion as a prognostic factor in patients' hospital mortality was analyzed using multivariate Cox regression models. However, we found no association (HR: 1.00, 95% CI: 0.66-1.52, p = 0.998) between albumin infusion and hospital mortality of acute pancreatitis patients admitted in ICU (Fig. 3).
Furthermore, to mitigate the estimation bias caused by imbalanced covariates between different treatment groups, we performed PSM and IPTW methods. The imbalance of covariates between groups was signi cantly reduced after both matches (Additional le 1 : Table S4). In these two matched study cohorts, differences between groups of the K-M survival curves disappeared through log-rank tests (Fig. 2b, PSM: p = 0.660; Fig. 2c, IPTW: p = 0.760). Moreover, the multivariate Cox regression model showed similar results as the original cohort after being matched (Fig. 3). To increase the robustness of this study, we also performed multivariate analyses in original and matched cohorts of patients without missing data, and the results remained similar. (Additional le 1 : Table S5-6).

Secondary outcomes
Albumin infusion was also found no association with 28-day mortality and ICU mortality by multivariate  (Fig. 4).

Subgroup and sensitivity analyses
To investigate the effect of albumin infusion on patients' outcome with different initial serum albumin levels, patients in the MIMIC-IV cohort were divided into four subgroups by rst measured serum albumin level after ICU admission: < 2.5 g/dL group, 2.5-3.0 g/dL group, 3.0-3.5 g/dL group and ≥ 3.5 g/dL group.
K-M survival analyses of each subgroup are shown as Figure S1, while multivariate Cox regressions were also performed, as shown in Figure 5. Interestingly, albumin infusion negatively in uenced patients with 3.0-3.5 g/dL serum albumin (HR: 8.17, 95% CI: 2.01-33.14, p = 0.003, n = 250), while no association with hospital mortality in other subgroups had been found (Fig. 5). After being matched by different methods, we repeated the analysis and got similar results (Additional le 1 : Table S7-10). It seemed that HSA infusion could not bene t patients' outcomes. This result was independent of the patients' initial serum albumin level.
Bacteraemia and secondary infection are common in acute pancreatitis patients and are essential factors leading to more severe disease [26]. To determine whether patients subgroup with de nite infection would bene t from albumin infusion, we performed a subgroup analysis of patients with positive blood or peritoneal uid culture. In the MIMIC-IV cohort, 161 patients with acute pancreatitis had positive blood or peritoneal uid culture during hospitalization. We performed K-M survival analyses and multivariate Cox regressions, and albumin infusion was still not associated with in-hospital mortality in acute pancreatitis patients with de nite infection (Additional le 1 : Figure S2, Table S11).
Additionally, we also investigated the effect of different total doses of albumin infusion on hospital outcomes as sensitivity analyses. According to clinical practice, patients with albumin infusion were divided into two subgroups (infusion dose < 100 g; ≥ 100 g) and compared against the non-Alb group, respectively. According to our results, albumin administration had no in uence on the primary outcome of patients with acute pancreatitis in either infusion volume subgroup (Additional le 1 : Figure S3, Table S12-13).
External validation with propensity score-matched eICU cohort We also validated our results with acute pancreatitis patients extracted from the eICU database. After being PSM matched, 104 acute pancreatitis patients with ICU admission were enrolled (52 in the Alb group and 52 in the non-Alb group). Baseline characteristics of these patients after PSM are presented in Table S3, as mentioned. Though there was a survival difference between treatment groups in K-M survival curves analysis (p = 0.037), albumin infusion still did not in uence multiple in-hospital outcomes of acute pancreatitis patients after multivariate analyses (Additional le 1 : Figure S4-5).

Discussion
In general, our research is the rst clinical investigation concentrating on the role of albumin infusion in the hospital outcomes of patients diagnosed with acute pancreatitis since 2008. Through a retrospective cohort of 950 contemporary acute pancreatitis patients from the MIMIC-IV database, our research showed that the infusion of human serum albumin (HSA) was not associated with hospital or ICU mortality in acute pancreatitis patients with ICU admission and signi cantly prolonged their hospitalization and ICU duration. This result was independent of whether the patient had a positive bacterial culture result. In subsequent subgroup analyses, albumin infusion still did not affect the patients' prognosis with signi cant hypoalbuminemia (< 2.5 g/dL), while even tended to have an adverse effect on group of patients with near-normal (3.0-3.5 g/dL) initial serum albumin levels. A cohort from the eICU database partially supported these results above after being PSM matched (104 patients).
Acute pancreatitis is characterized by local, systemic in ammatory and immune responses, leading to organ failure even death in severe cases. Subsequent uid extravasation in the third space is one of the critical reasons for the severity of the disease [1,6,27]. In current guidelines, adequate uid resuscitation was considered an essential step of the initial treatment in severe acute pancreatitis patients [1]. As indicated, the goal-directed therapy advised a resuscitation rate of 5-10 mL/kg/h to avoid the potentially detrimental in uence that improper uid replacement might cause [28]. While as for the type of uid, study evidence with high con dence is still scarce. Only several RCTs with small sample size concluded that Ringer's lactate solution had an uncon rmed bene t in reducing the chance of SIRS (systemic in ammatory response syndrome) and C-relative protein concentrations compared with normal salina [7][8][9][10]29].
Though the value of colloids was not con rmed in the therapy of acute pancreatitis [27], it has been demonstrated that HSA required less uid than crystalloid solutions to provide effective uid resuscitation [11]. On the other hand, some studies had also shown that hypoalbuminemia negatively in uenced acute pancreatitis patients' prognosis signi cantly [14,30]. For the reasons above, the doctors were accustomed to apply the albumin infusion to increase colloidal osmotic pressure and improve hypoalbuminemia in clinical practice. However, the actual association between albumin infusion and acute pancreatitis patients' prognosis has not been con rmed by clinical studies so far.
To con rm the in uence of HSA infusion on acute pancreatitis, we designed the research. Our primary study cohort was extracted from MIMIC-IV (v1.0), published on March 16, 2021. MIMIC-IV was a newlyupdated version of MIMIC-III, which had been improved on numerous aspects [16]. In MIMIC-IV, the patients' data from 2008 to 2019 could better re ect the current diagnosis and treatment of diseases and provide better suggestions for the current clinical practice. Our research also used data of acute pancreatitis patients from the eICU database. The eICU Collaborative Research Database (v2.0) contained clinical data of patients with ICU admission from 208 hospitals in 2014 and 2015 [19]. In our study process, we applied the MIMIC-IV cohort as our primary analysis group, while the eICU cohort after PSM matching was applied as an external veri cation.
In our study, patients in the albumin infusion group showed a more severe disease state than the other group, which could be indicated by discrepant parameters such as higher SOFA, SAPS II scores, lower serum albumin level, and lower mean arterial pressure. This phenomenon was consistent with the clinical decision strategy often made by doctors previously analyzed. The Alb group showed a poorer prognosis in K-M survival analyses, and this survival difference disappeared after balancing covariates between treatment groups by PSM or IPTW methods. Furthermore, after including covariates from multiple clinical aspects of each patient, multivariate Cox regressions still showed no correlation between albumin infusion and patients' in-hospital prognosis before and after matching. In addition, through multivariate linear regressions, we found that intravenous albumin infusion was associated with a longer hospital length of stay and ICU duration. It seemed that albumin infusion could not bene t the prognosis of acute pancreatitis patients. This is consistent with previous studies [31][32][33][34][35][36][37][38]. Since the meta-analysis study of increased mortality rates in patients who received albumin solutions was rst reported in 1998 [15], more and more well-controlled RCTs have concentrated on the actual bene t of human serum albumin in speci c patients groups. One of the most in uential prospective studies, published in 2014, was a multicentre trial of 1,818 patients with severe sepsis, which concluded that the addition of albumin did not improve the 28-or 90-day mortality compared with crystalloids alone [31]. Another trial, including 193 cirrhotic patients with infection other than SBP (spontaneous bacterial peritonitis) in 2015, also negated the bene t of albumin infusion in overall patient survival and improvement of renal failure [32]. The plausibility of our results was strongly supported by many prospective studies that had concluded similar opinions with other speci c patients groups in recent years [33][34][35][36][37][38]. We also performed an external validation with a cohort of 104 well-matched acute pancreatitis patients from the eICU database to strengthen our conclusion. Similar to the results of the MIMIC-IV cohort, albumin infusion continued to have no bene cial effect on primary and secondary outcomes in patients with acute pancreatitis through K-M survival analysis and multivariate Cox regressions.
To further support our results, we also conducted well-developed subgroup and sensitivity analyses. Patient's initial serum albumin level was likely to in uence the study results from clinical practice, and we performed subgroup analyses for patients with different rst measured serum albumin level after ICU admission. According to our results, even among acute pancreatitis patients with obvious hypoalbuminemia (< 2.5 g/dL), albumin infusion still had no statistically signi cant advantage on patients' outcomes before and after matching. This nding was consistent with a large meta-analysis study published in recent years that there was no evidence albumin infusion improved prognosis in critically ill patients with baseline hypoalbuminemia [36]. Moreover, compared with other subgroups, albumin infusion had a negative effect on the prognosis of patients whose initial serum albumin level was 3.0-3.5 g/dL. This phenomenon might be explained by the body's self-compensation mechanism, which suggested that albumin infusion might be even more discouraged in acute pancreatitis patients where albumin levels are near normal. It has been proved that early bacteraemia and secondary pancreatic or peripancreatic necrosis might result in sepsis with a poor prognosis in acute pancreatitis patients [26]. Thus, we performed a subgroup analysis of patients with positive blood or peritoneal uid bacterial cultures, and the results were robust with the primary analysis. As mentioned above, the role of albumin infusion in sepsis patients has been extensively studied in recent years. The current consensus was that the bene t of albumin in improving the prognosis of patients with sepsis relative to crystalloid remained unclear [39], which was consistent with our results. Our study also considered the possible impact of the total albumin infusion dose on the results as the sensitive analysis. According to the results, albumin infusion was not a bene cial factor, regardless of the total dose.
The initial management of acute pancreatitis included not only adequate uid resuscitation but also effective nutritional support [1]. Current guidelines mainly recommended enteral feeding because of its bene cial role in nourishing the intestinal barrier, preventing bacterial translocation and reducing the probability of SIRS when compared with conventional parenteral nutrition [1,6,40,41]. On the other hand, serum albumin level was also an important indicator to evaluate the nutritional status of patients [6]. In combination with our ndings, it was not di cult to conclude, due to the irrelevance of intravenous albumin infusion to patient prognosis, our study might emphasize the importance and necessity of enteral nutritional support for patients with acute pancreatitis from another perspective.
There were still several limitations in our research. First, the estimation bias was unavoidable as a retrospective study due to complex confounding factors in actual clinical treatment that could not be considered, though we had already signi cantly reduced the bias by several ways of adjustment and welldeveloped subgroup analysis. Large-scale, well-controlled RCTs are still desperately required to reach a more convincing conclusion. Secondly, due to the limitations of the MIMIC-IV and eICU database, the clinical indicators re ecting the possible bene ts of drug administration were still insu cient, resulting in the possibility to neglect the potential bene cial effects of albumin infusion for patients with acute pancreatitis. For example, previous studies had reported signi cant hemodynamic advantages of albumin infusion in patients with sepsis, although improvement in patients outcomes was also not observed [31]. In addition, the MIMIC-IV database was still unable to obtain data on patients' out-ofhospital survival status due to inadequate follow-up time [16], which might result in our study ignoring the possible positive in uence of albumin infusion on patients' long-term survival.

Conclusion
In conclusion, intravenous albumin infusion could not bene t acute pancreatitis patients' in-hospital prognosis and possibly prolong the hospital or ICU duration. This conclusion remained robust in patients subgroups with signi cant hypoalbuminemia (< 2.5 g/dL), positive bacterial cultures in blood or peritoneal uid and different total albumin infusion doses.

Declarations
Ethics approval and consent to participate This study was approved by the Ethics Committee of the First A liated Hospital of Xi'an Jiaotong University. Informed patient consent was not required to access and use MIMIC-IV and eICU data.

Consent for publication
Not applicable.
Availability of data and materials The datasets generated and analyzed during the current study are available in the MIMIC-IV and eICU databases, https://physionet.org/content/mimiciv/1.0/ and https://physionet.org/content/eicu-crd/2.0/ .

Competing interests
There is no potential con icts of interest related to personal, nancial, professional, or relationship, etc.  Kaplan-Meire survival curve analysis between treatment groups before and after matching from the MIMIC-IV database. a Survival curves of hospital mortality in acute pancreatitis patients between treatment groups before matching from the MIMIC-IV database. b Survival curves of hospital mortality in acute pancreatitis patients between treatment groups after PSM matching from the MIMIC-IV database. c Survival curves of hospital mortality in acute pancreatitis patients between treatment groups after IPTW matching from the MIMIC-IV database  Effect of albumin infusion on secondary outcomes in acute pancreatitis patients from the MIMIC-IV database before and after matching through multivariate analyses. a Effect of albumin infusion on 28day mortality and ICU mortality in acute pancreatitis patients from the MIMIC-IV database before and after matching through multivariate Cox regressions. b Effect of albumin infusion on hospital and ICU length of stays in acute pancreatitis patients from the MIMIC-IV database before and after matching through multivariate linear regressions Figure 5