Prognostic Value of Albumin-Bilirubin Score in Acute Pancreatitis: Insights From the MIMIC-IV Database

Introduction

Acute pancreatitis (AP) is a common gastrointestinal disorder. The pathogenesis of AP is complex, including local and systemic inflammation triggered by various factors [1–3]. While mild AP can typically be managed with conservative treatments such as fasting, gastrointestinal decompression, and intravenous fluids, about 15% to 20% of patients develop complications, including pancreatic necrosis, systemic inflammatory response syndrome, and multi-organ failure [4–6]. Despite advances in medical treatments, the mortality rate in patients with severe AP remains high. Hence, accurate and timely assessment of disease severity is essential for guiding treatment strategies and improving patient outcomes. Existing assessment methods such as the Ranson and Sequential Organ Failure Assessment scores are time-consuming and complex [7–10]. Therefore, it is necessary to explore simpler and faster tools for assessing the prognosis of AP [10].

The albumin-bilirubin (ALBI) score, originally developed to evaluate liver function in patients with hepatocellular carcinoma [11], integrates 2 readily available laboratory parameters, serum albumin and total bilirubin. It has demonstrated distinct advantages in prognostic assessment for individuals with hepatocellular carcinoma. It is also gradually being applied in the research on inflammatory diseases such as sepsis [12–15]. Previous research based on the Medical Information Mart for Intensive Care (MIMIC)-III database has investigated the association of the ALBI score with in-hospital death in individuals with AP. Nonetheless, due to a relatively small sample size (284 individuals) and the inherent limitations of the database (covering the period 2001–2008), the generalizability of its findings is limited [14]. Conversely, the MIMIC-IV database (2008–2022), which covers a wider population, may help analyze the potential role of the ALBI score in assessing short- and long-term prognosis for individuals with AP [16–18]. Recent research has explored the total bilirubin-to-albumin ratio (TBAR) based on 477 patients from 2012 to 2019 (MIMIC-IV V2.2) [19]. In contrast, our study used MIMIC-IV 3.0 (2008–2022), with a larger intensive care unit (ICU) cohort, to explore the role of the ALBI score, a more widely validated tool [11], in predicting short- and long-term mortality in critically ill patients with AP.

The objective of this research was to utilize the MIMIC-IV database to explore the association of the ALBI score with short- and long-term outcomes in patients with AP and to assess its clinical utility as a predictive tool for managing severe AP.

Material and Methods

DATA SOURCE:

Data for this study were obtained from the MIMIC-IV 3.0 database, a single-center resource developed by the Laboratory for Computational Physiology at the Massachusetts Institute of Technology. As this study used de-identified data and was in compliance with the BIDMC Institutional Review Board (IRB) waiver, no additional institutional IRB approval was required. The database encompassed comprehensive clinical information on over 70 000 ICU patients from 2008 to 2022, including demographic characteristics, vital signs, laboratory findings, pharmacological treatments, diagnoses, and clinical outcomes [17,18]. We obtained the necessary access to utilize the dataset (record ID: 13443082) and were responsible for data extraction.

Inclusion criteria were as follows: patients diagnosed with AP according to the International Classification of Diseases, Ninth Revision (ICD-9) code 577.0 and Tenth Revision (ICD-10) codes K85-K85.92. Exclusion criteria were as follows: (1) individuals who were not admitted to the ICU; (2) cases in which either albumin or bilirubin laboratory values were missing; (3) individuals with no data on 30-day or 1-year survival outcome. Details are provided in Figure 1.

DATA EXTRACTION:

Data on demographics, laboratory indicators, comorbidities, outcome measures, medication administration, and surgical procedures were retrieved from the MIMIC-IV database, using Structured Query Language (SQL). Demographic information consisted of sex, age, and race. Vital signs consisted of heart rate, respiratory rate, oxygen saturation, temperature, partial pressure of oxygen (pO₂), and arterial blood pressure monitoring. For vital sign data, the initial measurements at admission were used. Other laboratory biomarkers included bilirubin, albumin, creatinine, blood urea nitrogen (BUN), sodium, potassium, calcium, chloride, glucose, alanine transaminase (ALT), alkaline phosphatase (ALP), aspartate transaminase (AST), amylase, lipase, white blood cell count (WBC), hemoglobin, hematocrit, platelet count, pH value, and international normalized ratio (INR).

Obesity was defined as a body mass index of 30 or higher. Comorbidities and medical history were identified based on ICD-9 and ICD-10 codes, including acute heart failure, acute respiratory failure, atrial fibrillation, and obesity. Medications included carbapenems, fluoroquinolones, and proton-pump inhibitors (PPIs). The surgical and procedural data encompassed cholecystectomy, endoscopic retrograde cholangiopancreatography (ERCP), and mechanical ventilation.

For the laboratory indicators that were measured multiple times after admission, the initial measurements at admission were used for analysis. Data on mechanical ventilation were extracted within the first 24 hours of ICU admission to distinguish more critically ill patients. Other information on medications and procedures was collected during hospitalization, focusing solely on usage status.

OUTCOME MEASURES:

The primary outcome was all-cause death within 30 days, and the secondary outcome was all-cause death within 1 year.

CALCULATION OF ALBI SCORE:

The ALBI value was derived using the following formula: ALBI=(0.66×log10 bilirubin [μmol/L])-(0.085×albumin [g/L]) [20]. Patients were divided into 3 groups according to previously established thresholds of ALBI [20]: T1 (ALBI <−2.60), T2 (−2.60 ≤ALBI ≤−1.39), and T3 (ALBI >−1.39).

HANDLING MISSING DATA AND OUTLIERS:

For outcome and exposure variables, cases with missing data were excluded. For other covariates, variables with more than 20% missing data were excluded, while those with less than 20% missing data were imputed using the mice package in R. For abnormal values, 2.5% and 97.5% percentile Winsorization was applied to replace values below the 2.5th percentile and above the 97.5th percentile. Variables with high missing rates, such as pO2 (42.19%), body temperature (83.72%), and arterial blood pressure monitoring (54.38%), were excluded from subsequent analyses (Table 1). All data cleaning and imputation processes were completed before formal statistical analyses, such as Cox regression.

STATISTICAL ANALYSIS:

The Shapiro-Wilk test was used to evaluate the normality of continuous variables. Mean±standard deviation (SD) was used to describe normally distributed continuous variables, while median (interquartile range [IQR]) was used for non-normally distributed continuous variables. Intergroup comparison was performed using either the t test (for normally distributed continuous variables) or nonparametric tests (for non-normally distributed continuous variables), respectively. Categorical variables were assessed using counts and proportions (%), and differences between groups were assessed using either the Pearson chi-square (χ²) test or Fisher exact test, as appropriate.

To avoid multicollinearity, variables with a variance inflation factor of more than 5 were removed. Based on the actual processing results, hemoglobin and ALT were removed from subsequent analyses. Univariate Cox regression analysis was performed to identify potential confounders associated with outcomes, which were used to adjust models. The association of the ALBI score with mortality within 30 days and 1 year was assessed using Cox regression. The ALBI score was analyzed both as a continuous measure and as a categorical variable within the 3 groups. The hazard ratios (HRs) and corresponding 95% CIs were computed. Several potential confounders were then adjusted to explore the effect of the ALBI score on survival prognosis, and a tendency test was performed. The lowest ALBI score (T1) was selected as the reference group. Specifically, model 1 was unadjusted; in contrast, model 2 was adjusted for demographics such as age, sex, and race. Model 3 was further adjusted for variables with statistical significance in univariate Cox regression. Univariate Cox regression analysis is a commonly used preliminary screening method to identify variables that are significantly related to outcomes, providing a reference for subsequent multivariate analysis. In the Cox proportional hazards model analysis, we used the Schoenfeld residual test to verify the proportional hazards assumption. The Schoenfeld residual test is a common method to test whether the proportional hazards assumption in the model holds.

Kaplan-Meier survival curves were generated to visualize survival outcomes across different groups stratified by ALBI score, and the log-rank test was used to compare the survival distributions across these different groups. Furthermore, the potential nonlinear relationship between ALBI score and the risk of all-cause mortality was explored using restricted cubic splines, with 5 knots placed at the data percentiles. The placement/reference points were determined using the default parameters of the cph function in the rms package. Subgroup analyses by age, sex, and comorbidities or treatments identified through univariate analysis were conducted to further explore the association of the ALBI score with 30-day mortality. Interaction analyses were conducted using Cox regression models with interaction terms to evaluate how the prognostic performance of the ALBI score varied across subgroups. The same covariates as in the full model were adjusted for; however, subgroup-specific variables were excluded in the corresponding subgroup analyses to avoid redundant adjustments. Age was treated as a continuous variable in covariate correction and baseline description. However, in subgroup and interaction analysis, age was stratified into 2 groups at the cutoff of 60 years, based on its clinical relevance and common practice in previous literature. The significance of the interactions was tested using the chi-square test.

Data extraction and preprocessing were performed using pgAdmin4 and PostgreSQL. All statistical and survival analyses were performed using R software (V4.2.0). P<0.05 indicated statistical significance.

Results

BASELINE CHARACTERISTICS OF SAMPLES:

A total of 1953 patients with AP admitted to the ICU were included in this study, with 56.6% (1105) male and 43.4% (848) female. The median age was 60.79 years (IQR, 48.40–73.61 years). In the highest ALBI group (T3), the 30-day and 1-year mortality rates were 26.6% and 36.1%, respectively, which were higher than those in the T2 (15.6% and 28.2%) and T1 groups (11.2% and 22.2%) (P<0.001). Compared with other groups, the T3 group exhibited markedly elevated heart rate and respiratory rate and bilirubin, ALT, ALP, AST, and INR levels, as well as increased carbapenem, mechanical ventilation, PPI, and ERCP use. Conversely, levels of albumin, hemoglobin, hematocrit, and calcium and the rate of obesity were markedly decreased (P<0.05) in the T3 group, as shown in Table 2.

ADJUSTMENT OF COVARIATES:

Univariate Cox regression analysis was conducted to identify confounders that may affect the occurrence of outcomes. The results for the following variables were significant (P<0.05): respiratory rate, oxygen saturation, creatinine, BUN, potassium, ALP, hematocrit, platelet count, pH value, INR, the presence of atrial fibrillation, obesity, the use of carbapenems and fluoroquinolones, history of cholecystectomy, and the need for mechanical ventilation. These variables were used to adjust models, as shown in Table 3.

Model 3 included 19 covariates (sex, age, race, and variables selected). The 30-day mortality rate was 309 of 1953 patients, and the 1-year mortality rate was 533 of 1953 patients. This design ensured the model’s stability and effectively supported the observed results. Sex, age, and race were considered potential confounding factors because they were significantly associated with the outcomes and could influence mortality risk. Biochemical indicators such as creatinine, BUN, and potassium reflected patients’ renal function, electrolyte balance, and other physiological states. These factors may be significantly associated with mortality risk and, therefore, need to be adjusted for bias in the study. Carbapenem, fluoroquinolone, and mechanical ventilation use were later-stage clinical interventions, reflecting the treatment methods received by patients.

ASSOCIATION OF ALBI WITH 30-DAY AND 1-YEAR MORTALITY:

Table 4 illustrates the association of the ALBI score with 30-day and 1-year mortality rates. A higher ALBI score, as a continuous variable, was markedly associated with an elevated 30-day mortality rate in model 1 (HR, 1.62; 95% CI, 1.42–1.85; P<0.001), model 2 (HR, 1.61; 95% CI, 1.40–1.86; P<0.001), and model 3 (HR, 1.37; 95% CI, 1.16–1.61; P<0.001). In addition, the ALBI score, when treated as a categorical variable, exhibited a significant association with the risk of 30-day mortality. Notably, the T3 group had a significantly elevated 30-day mortality rate compared with the T1 group (HR, 1.72; 95% CI, 1.20–2.45; P=0.003). A tendency test further confirmed a significant association of the ALBI score with 30-day mortality rates (P<0.05).

The Schoenfeld residual test results indicated that the proportional hazards hypothesis held (P>0.05), and the model was suitable for data analysis. The association of the ALBI score with 1-year all-cause mortality was also evaluated. The findings were consistent with those of 30-day mortality, as summarized in Table 4. Sensitivity analyses comparing models with and without Winsorization yielded consistent results, confirming the robustness of our primary findings.

KAPLAN-MEIER CURVE ANALYSIS OF ALBI AND PATIENT SURVIVAL:

Kaplan-Meier survival curves demonstrated significant differences in 30-day and 1-year mortality rates across the 3 ALBI groups. Patients with higher ALBI scores exhibited lower 30-day and 1-year survival rates, compared with the T1 group (P for log-rank test <0.001) (Figure 2).

NONLINEAR ASSOCIATION BETWEEN ALBI SCORE AND MORTALITY:

Restricted cubic splines analysis revealed a statistically significant nonlinear association of ALBI score with 30-day mortality (P for nonlinear=0.035), as illustrated in Figure 3A. A turning point was found at an ALBI score of −2.3, and the overall model showed statistical significance (P for overall <0.001). This threshold of ALBI −2.3 may be a clinically meaningful threshold, beyond which the risk of mortality significantly increased. Consistent findings were obtained for 1-year mortality (Figure 3B).

SUBGROUP ANALYSIS:

Subgroup analyses were performed based on sex, age (≥60 years), race, atrial fibrillation, obesity, the use of fluoroquinolones or ciprofloxacin, and the need for mechanical ventilation. These analyses were exploratory in nature, and no multiplicity control was applied.

As shown in Figure 4, among younger patients (<60 years), female patients, Black patients, White patients, those without atrial fibrillation or obesity, those not receiving fluoroquinolones or ciprofloxacin, and those not receiving mechanical ventilation, the ALBI score showed a significant association with 30-day mortality rates (P<0.05). In older patients (≥60 years), male patients, Asian populations, other races, patients with concurrent atrial fibrillation, patients with obesity, those receiving fluoroquinolones or ciprofloxacin, or those undergoing mechanical ventilation within the first 24 hours of ICU admission, the ALBI score was not significantly associated with 30-day mortality rates (P≥0.05).

Notably, the use of fluoroquinolones (Pinteraction=0.032), ciprofloxacin (Pinteraction=0.048), and mechanical ventilation within the first 24 hours of ICU admission (Pinteraction=0.029) had significant interaction effects in the association between the ALBI score and 30-day mortality rates. Specifically, the association of the ALBI score with 30-day mortality was stronger in individuals not receiving fluoroquinolones (HR, 1.85) or ciprofloxacin (HR, 2.05) than in those receiving fluoroquinolones (P≥0.05). Similarly, the association between ALBI score and 30-day mortality rate was stronger in individuals not undergoing mechanical ventilation within the first 24 hours of ICU admission (HR, 2.21) than in those undergoing mechanical ventilation (P≥0.05).

Discussion

This study examined the association between the ALBI score and all-cause mortality in patients with AP, using the MIMIC-IV database. Overall, a higher ALBI score was positively associated with the risk of short-term (30-day) and long-term (1-year) mortality. The association between ALBI score and mortality rates was significant in patients not receiving carbapenems, fluoroquinolones, or mechanical ventilation (mechanical ventilation) within the first 24 hours of ICU admission. In contrast, this association was not significant in patients who received these treatments. Moreover, when the ALBI score exceeded −2.3, the risk of mortality increased sharply, highlighting its potential as a critical marker for early risk stratification in ICU patients.

Pathophysiologically, systemic inflammatory response, systemic organ dysfunction, and potential organ failure are significant contributors to adverse outcomes in individuals with AP [21,22]. Abnormal changes in albumin levels are typically associated with persistent inflammation and unresolved malnutrition [23], while elevated bilirubin levels may indicate biliary obstruction or tumor progression [24,25]. Therefore, the ALBI score may serve as an indirect indicator of inflammatory load and organ injury. It is a dependable tool for predicting outcomes in various diseases, such as hepatocellular carcinoma and heart failure [26–28]. Shi et al have also observed markedly elevated ALBI scores in patients with AP who died during their hospital stay, compared with survivors [14]. Bilirubin also possesses antioxidant and immunomodulatory properties [29–31], which provide a biological context for understanding the prognostic value of the ALBI score in AP.

Subgroup analysis revealed that the association between ALBI score and 30-day mortality was more significant in patients not receiving fluoroquinolones or ciprofloxacin or not requiring mechanical ventilation within the first 24 hours of ICU admission. These patients are likely less critically ill, and thus the ALBI score may reflect their mortality risk. In contrast, the prognostic effect of ALBI may be diluted by the competing risk of other acute clinical factors. Our study also suggested a clinically significant turning point at an ALBI score of −2.3, beyond which the risk of mortality increased substantially.

Our results indicated that the ALBI score could serve as a straightforward and valuable tool for predicting the prognosis of patients with AP. Clinicians can use the ALBI score to promptly identify patients at high risk, thus developing tailored interventions. For instance, for patients with an ALBI score greater than −2.3, a more aggressive therapeutic strategy should be considered, such as enhanced anti-inflammatory therapy, optimized fluid management, and early nutritional support. The ALBI score is very simple and thus is particularly practical, especially in resource-constrained healthcare settings.

This study has certain limitations. First, the dataset used in this research was obtained from a single-center MIMIC-IV database, which may introduce selection bias. Also, the accuracy of data may be limited by the recording of patient admission details and treatment protocols. Second, due to limitations within the database, detailed information on specific causes of death was unavailable. Thus, it is infeasible to analyze the impact of the ALBI score on AP-specific mortality. Ultimately, this study focused exclusively on patients in the ICU. Therefore, certain confounding variables that could impact the results, such as patients’ dietary habits and long-term medication, may have been neglected.

Future prospective, multi-center investigations are warranted to corroborate our results and extend the clinical utility of the ALBI score in AP.

Conclusions

The ALBI score is significantly associated with short- and long-term mortality in patients with AP. Elevated ALBI scores may be applied as a strong predictor of an increased risk of death. Our findings extend the results of previous studies on ALBI indices and provide further evidence of the clinical utility of the ALBI score in AP. However, given the study’s limitations, further large-scale, multi-center prospective studies are warranted to further corroborate our findings.